blob: 993061f83067a9a65c5a58908908f5948423f11a [file] [log] [blame]
Chris Mason6cbd5572007-06-12 09:07:21 -04001/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
Chris Mason8f18cf12008-04-25 16:53:30 -040019#include <linux/kernel.h>
Chris Mason065631f2008-02-20 12:07:25 -050020#include <linux/bio.h>
Chris Mason39279cc2007-06-12 06:35:45 -040021#include <linux/buffer_head.h>
Sage Weilf2eb0a22008-05-02 14:43:14 -040022#include <linux/file.h>
Chris Mason39279cc2007-06-12 06:35:45 -040023#include <linux/fs.h>
24#include <linux/pagemap.h>
25#include <linux/highmem.h>
26#include <linux/time.h>
27#include <linux/init.h>
28#include <linux/string.h>
Chris Mason39279cc2007-06-12 06:35:45 -040029#include <linux/backing-dev.h>
30#include <linux/mpage.h>
31#include <linux/swap.h>
32#include <linux/writeback.h>
Chris Mason39279cc2007-06-12 06:35:45 -040033#include <linux/compat.h>
Chris Mason9ebefb182007-06-15 13:50:00 -040034#include <linux/bit_spinlock.h>
Josef Bacik5103e942007-11-16 11:45:54 -050035#include <linux/xattr.h>
Josef Bacik33268ea2008-07-24 12:16:36 -040036#include <linux/posix_acl.h>
Yan Zhengd899e052008-10-30 14:25:28 -040037#include <linux/falloc.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090038#include <linux/slab.h>
David Sterba7a36dde2011-05-06 15:33:15 +020039#include <linux/ratelimit.h>
Josef Bacik22c44fe2011-11-30 10:45:38 -050040#include <linux/mount.h>
Filipe Brandenburger55e301f2013-01-29 06:04:50 +000041#include <linux/btrfs.h>
David Woodhouse53b381b2013-01-29 18:40:14 -050042#include <linux/blkdev.h>
Josef Bacikf23b5a52013-06-19 10:16:26 -040043#include <linux/posix_acl_xattr.h>
Christoph Hellwige2e40f22015-02-22 08:58:50 -080044#include <linux/uio.h>
Josef Bacik69fe2d72017-10-19 14:15:57 -040045#include <linux/magic.h>
Chris Mason39279cc2007-06-12 06:35:45 -040046#include "ctree.h"
47#include "disk-io.h"
48#include "transaction.h"
49#include "btrfs_inode.h"
Chris Mason39279cc2007-06-12 06:35:45 -040050#include "print-tree.h"
Chris Masone6dcd2d2008-07-17 12:53:50 -040051#include "ordered-data.h"
Christoph Hellwig95819c02008-08-28 06:21:17 -040052#include "xattr.h"
Chris Masone02119d2008-09-05 16:13:11 -040053#include "tree-log.h"
Jan Schmidt4a54c8c2011-07-22 15:41:52 +020054#include "volumes.h"
Chris Masonc8b97812008-10-29 14:49:59 -040055#include "compression.h"
Chris Masonb4ce94d2009-02-04 09:25:08 -050056#include "locking.h"
Josef Bacikdc89e982011-01-28 17:05:48 -050057#include "free-space-cache.h"
Li Zefan581bb052011-04-20 10:06:11 +080058#include "inode-map.h"
Liu Bo38c227d2013-01-29 03:18:40 +000059#include "backref.h"
Josef Bacikf23b5a52013-06-19 10:16:26 -040060#include "hash.h"
Filipe David Borba Manana63541922014-01-07 11:47:46 +000061#include "props.h"
Dongsheng Yang31193212014-12-12 16:44:35 +080062#include "qgroup.h"
Wang Xiaoguangdda32452016-07-11 11:05:29 +080063#include "dedupe.h"
Chris Mason39279cc2007-06-12 06:35:45 -040064
65struct btrfs_iget_args {
Chris Mason90d3e592014-01-09 17:28:00 -080066 struct btrfs_key *location;
Chris Mason39279cc2007-06-12 06:35:45 -040067 struct btrfs_root *root;
68};
69
Filipe Mananaf28a4922015-12-08 19:23:20 +000070struct btrfs_dio_data {
Filipe Mananaf28a4922015-12-08 19:23:20 +000071 u64 reserve;
72 u64 unsubmitted_oe_range_start;
73 u64 unsubmitted_oe_range_end;
Liu Bo4aaedfb2016-12-14 22:36:05 -080074 int overwrite;
Filipe Mananaf28a4922015-12-08 19:23:20 +000075};
76
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -070077static const struct inode_operations btrfs_dir_inode_operations;
78static const struct inode_operations btrfs_symlink_inode_operations;
79static const struct inode_operations btrfs_dir_ro_inode_operations;
80static const struct inode_operations btrfs_special_inode_operations;
81static const struct inode_operations btrfs_file_inode_operations;
Alexey Dobriyan7f094102009-09-21 17:01:10 -070082static const struct address_space_operations btrfs_aops;
83static const struct address_space_operations btrfs_symlink_aops;
Alexey Dobriyan828c0952009-10-01 15:43:56 -070084static const struct file_operations btrfs_dir_file_operations;
David Sterba20e55062015-11-19 11:42:28 +010085static const struct extent_io_ops btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -040086
87static struct kmem_cache *btrfs_inode_cachep;
88struct kmem_cache *btrfs_trans_handle_cachep;
Chris Mason39279cc2007-06-12 06:35:45 -040089struct kmem_cache *btrfs_path_cachep;
Josef Bacikdc89e982011-01-28 17:05:48 -050090struct kmem_cache *btrfs_free_space_cachep;
Chris Mason39279cc2007-06-12 06:35:45 -040091
92#define S_SHIFT 12
David Sterba4d4ab6d2015-11-19 11:42:31 +010093static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
Chris Mason39279cc2007-06-12 06:35:45 -040094 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
95 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
96 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
97 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
98 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
99 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
100 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
101};
102
Eric Sandeen3972f262013-01-12 02:57:22 +0000103static int btrfs_setsize(struct inode *inode, struct iattr *attr);
Josef Bacika41ad392011-01-31 15:30:16 -0500104static int btrfs_truncate(struct inode *inode);
Josef Bacik5fd02042012-05-02 14:00:54 -0400105static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
Chris Mason771ed682008-11-06 22:02:51 -0500106static noinline int cow_file_range(struct inode *inode,
107 struct page *locked_page,
Wang Xiaoguangdda32452016-07-11 11:05:29 +0800108 u64 start, u64 end, u64 delalloc_end,
109 int *page_started, unsigned long *nr_written,
110 int unlock, struct btrfs_dedupe_hash *hash);
Liu Bo6f9994d2017-01-31 07:50:22 -0800111static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
112 u64 orig_start, u64 block_start,
113 u64 block_len, u64 orig_block_len,
114 u64 ram_bytes, int compress_type,
115 int type);
Josef Bacik7b128762008-07-24 12:17:14 -0400116
Qu Wenruo524272602017-03-08 10:25:52 +0800117static void __endio_write_update_ordered(struct inode *inode,
118 const u64 offset, const u64 bytes,
119 const bool uptodate);
120
121/*
122 * Cleanup all submitted ordered extents in specified range to handle errors
123 * from the fill_dellaloc() callback.
124 *
125 * NOTE: caller must ensure that when an error happens, it can not call
126 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
127 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
128 * to be released, which we want to happen only when finishing the ordered
129 * extent (btrfs_finish_ordered_io()). Also note that the caller of the
130 * fill_delalloc() callback already does proper cleanup for the first page of
131 * the range, that is, it invokes the callback writepage_end_io_hook() for the
132 * range of the first page.
133 */
134static inline void btrfs_cleanup_ordered_extents(struct inode *inode,
135 const u64 offset,
136 const u64 bytes)
137{
Naohiro Aota63d71450c2017-09-01 17:58:47 +0900138 unsigned long index = offset >> PAGE_SHIFT;
139 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
140 struct page *page;
141
142 while (index <= end_index) {
143 page = find_get_page(inode->i_mapping, index);
144 index++;
145 if (!page)
146 continue;
147 ClearPagePrivate2(page);
148 put_page(page);
149 }
Qu Wenruo524272602017-03-08 10:25:52 +0800150 return __endio_write_update_ordered(inode, offset + PAGE_SIZE,
151 bytes - PAGE_SIZE, false);
152}
153
Eric Sandeen48a3b632013-04-25 20:41:01 +0000154static int btrfs_dirty_inode(struct inode *inode);
Chris Mason39279cc2007-06-12 06:35:45 -0400155
Josef Bacik6a3891c2015-03-16 17:38:52 -0400156#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
157void btrfs_test_inode_set_ops(struct inode *inode)
158{
159 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
160}
161#endif
162
Yan, Zhengf34f57a2009-11-12 09:35:27 +0000163static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
Eric Paris2a7dba32011-02-01 11:05:39 -0500164 struct inode *inode, struct inode *dir,
165 const struct qstr *qstr)
Jim Owens0279b4c2009-02-04 09:29:13 -0500166{
167 int err;
168
Yan, Zhengf34f57a2009-11-12 09:35:27 +0000169 err = btrfs_init_acl(trans, inode, dir);
Jim Owens0279b4c2009-02-04 09:29:13 -0500170 if (!err)
Eric Paris2a7dba32011-02-01 11:05:39 -0500171 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
Jim Owens0279b4c2009-02-04 09:29:13 -0500172 return err;
173}
174
Chris Masond352ac62008-09-29 15:18:18 -0400175/*
Chris Masonc8b97812008-10-29 14:49:59 -0400176 * this does all the hard work for inserting an inline extent into
177 * the btree. The caller should have done a btrfs_drop_extents so that
178 * no overlapping inline items exist in the btree
179 */
Chris Mason40f76582014-05-21 13:35:51 -0700180static int insert_inline_extent(struct btrfs_trans_handle *trans,
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000181 struct btrfs_path *path, int extent_inserted,
Chris Masonc8b97812008-10-29 14:49:59 -0400182 struct btrfs_root *root, struct inode *inode,
183 u64 start, size_t size, size_t compressed_size,
Li Zefanfe3f5662011-03-28 08:30:38 +0000184 int compress_type,
Chris Masonc8b97812008-10-29 14:49:59 -0400185 struct page **compressed_pages)
186{
Chris Masonc8b97812008-10-29 14:49:59 -0400187 struct extent_buffer *leaf;
188 struct page *page = NULL;
189 char *kaddr;
190 unsigned long ptr;
191 struct btrfs_file_extent_item *ei;
Chris Masonc8b97812008-10-29 14:49:59 -0400192 int ret;
193 size_t cur_size = size;
Chris Masonc8b97812008-10-29 14:49:59 -0400194 unsigned long offset;
Chris Masonc8b97812008-10-29 14:49:59 -0400195
Li Zefanfe3f5662011-03-28 08:30:38 +0000196 if (compressed_size && compressed_pages)
Chris Masonc8b97812008-10-29 14:49:59 -0400197 cur_size = compressed_size;
Chris Masonc8b97812008-10-29 14:49:59 -0400198
Chris Masonc8b97812008-10-29 14:49:59 -0400199 inode_add_bytes(inode, size);
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000200
201 if (!extent_inserted) {
202 struct btrfs_key key;
203 size_t datasize;
204
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +0200205 key.objectid = btrfs_ino(BTRFS_I(inode));
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000206 key.offset = start;
David Sterba962a2982014-06-04 18:41:45 +0200207 key.type = BTRFS_EXTENT_DATA_KEY;
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000208
209 datasize = btrfs_file_extent_calc_inline_size(cur_size);
210 path->leave_spinning = 1;
211 ret = btrfs_insert_empty_item(trans, root, path, &key,
212 datasize);
David Sterba79b4f4c2017-06-15 19:09:51 +0200213 if (ret)
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000214 goto fail;
Chris Masonc8b97812008-10-29 14:49:59 -0400215 }
216 leaf = path->nodes[0];
217 ei = btrfs_item_ptr(leaf, path->slots[0],
218 struct btrfs_file_extent_item);
219 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
220 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
221 btrfs_set_file_extent_encryption(leaf, ei, 0);
222 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
223 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
224 ptr = btrfs_file_extent_inline_start(ei);
225
Li Zefan261507a02010-12-17 14:21:50 +0800226 if (compress_type != BTRFS_COMPRESS_NONE) {
Chris Masonc8b97812008-10-29 14:49:59 -0400227 struct page *cpage;
228 int i = 0;
Chris Masond3977122009-01-05 21:25:51 -0500229 while (compressed_size > 0) {
Chris Masonc8b97812008-10-29 14:49:59 -0400230 cpage = compressed_pages[i];
Chris Mason5b050f02008-11-11 09:34:41 -0500231 cur_size = min_t(unsigned long, compressed_size,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300232 PAGE_SIZE);
Chris Masonc8b97812008-10-29 14:49:59 -0400233
Cong Wang7ac687d2011-11-25 23:14:28 +0800234 kaddr = kmap_atomic(cpage);
Chris Masonc8b97812008-10-29 14:49:59 -0400235 write_extent_buffer(leaf, kaddr, ptr, cur_size);
Cong Wang7ac687d2011-11-25 23:14:28 +0800236 kunmap_atomic(kaddr);
Chris Masonc8b97812008-10-29 14:49:59 -0400237
238 i++;
239 ptr += cur_size;
240 compressed_size -= cur_size;
241 }
242 btrfs_set_file_extent_compression(leaf, ei,
Li Zefan261507a02010-12-17 14:21:50 +0800243 compress_type);
Chris Masonc8b97812008-10-29 14:49:59 -0400244 } else {
245 page = find_get_page(inode->i_mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300246 start >> PAGE_SHIFT);
Chris Masonc8b97812008-10-29 14:49:59 -0400247 btrfs_set_file_extent_compression(leaf, ei, 0);
Cong Wang7ac687d2011-11-25 23:14:28 +0800248 kaddr = kmap_atomic(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300249 offset = start & (PAGE_SIZE - 1);
Chris Masonc8b97812008-10-29 14:49:59 -0400250 write_extent_buffer(leaf, kaddr + offset, ptr, size);
Cong Wang7ac687d2011-11-25 23:14:28 +0800251 kunmap_atomic(kaddr);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300252 put_page(page);
Chris Masonc8b97812008-10-29 14:49:59 -0400253 }
254 btrfs_mark_buffer_dirty(leaf);
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000255 btrfs_release_path(path);
Chris Masonc8b97812008-10-29 14:49:59 -0400256
Yan, Zhengc2167752009-11-12 09:34:21 +0000257 /*
258 * we're an inline extent, so nobody can
259 * extend the file past i_size without locking
260 * a page we already have locked.
261 *
262 * We must do any isize and inode updates
263 * before we unlock the pages. Otherwise we
264 * could end up racing with unlink.
265 */
Chris Masonc8b97812008-10-29 14:49:59 -0400266 BTRFS_I(inode)->disk_i_size = inode->i_size;
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100267 ret = btrfs_update_inode(trans, root, inode);
Yan, Zhengc2167752009-11-12 09:34:21 +0000268
Chris Masonc8b97812008-10-29 14:49:59 -0400269fail:
David Sterba79b4f4c2017-06-15 19:09:51 +0200270 return ret;
Chris Masonc8b97812008-10-29 14:49:59 -0400271}
272
273
274/*
275 * conditionally insert an inline extent into the file. This
276 * does the checks required to make sure the data is small enough
277 * to fit as an inline extent.
278 */
Josef Bacik00361582013-08-14 14:02:47 -0400279static noinline int cow_file_range_inline(struct btrfs_root *root,
280 struct inode *inode, u64 start,
281 u64 end, size_t compressed_size,
282 int compress_type,
283 struct page **compressed_pages)
Chris Masonc8b97812008-10-29 14:49:59 -0400284{
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400285 struct btrfs_fs_info *fs_info = root->fs_info;
Josef Bacik00361582013-08-14 14:02:47 -0400286 struct btrfs_trans_handle *trans;
Chris Masonc8b97812008-10-29 14:49:59 -0400287 u64 isize = i_size_read(inode);
288 u64 actual_end = min(end + 1, isize);
289 u64 inline_len = actual_end - start;
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400290 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
Chris Masonc8b97812008-10-29 14:49:59 -0400291 u64 data_len = inline_len;
292 int ret;
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000293 struct btrfs_path *path;
294 int extent_inserted = 0;
295 u32 extent_item_size;
Chris Masonc8b97812008-10-29 14:49:59 -0400296
297 if (compressed_size)
298 data_len = compressed_size;
299
300 if (start > 0 ||
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400301 actual_end > fs_info->sectorsize ||
302 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
Chris Masonc8b97812008-10-29 14:49:59 -0400303 (!compressed_size &&
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400304 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
Chris Masonc8b97812008-10-29 14:49:59 -0400305 end + 1 < isize ||
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400306 data_len > fs_info->max_inline) {
Chris Masonc8b97812008-10-29 14:49:59 -0400307 return 1;
308 }
309
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000310 path = btrfs_alloc_path();
311 if (!path)
312 return -ENOMEM;
313
Josef Bacik00361582013-08-14 14:02:47 -0400314 trans = btrfs_join_transaction(root);
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000315 if (IS_ERR(trans)) {
316 btrfs_free_path(path);
Josef Bacik00361582013-08-14 14:02:47 -0400317 return PTR_ERR(trans);
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000318 }
Josef Bacik69fe2d72017-10-19 14:15:57 -0400319 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
Josef Bacik00361582013-08-14 14:02:47 -0400320
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000321 if (compressed_size && compressed_pages)
322 extent_item_size = btrfs_file_extent_calc_inline_size(
323 compressed_size);
324 else
325 extent_item_size = btrfs_file_extent_calc_inline_size(
326 inline_len);
327
328 ret = __btrfs_drop_extents(trans, root, inode, path,
329 start, aligned_end, NULL,
330 1, 1, extent_item_size, &extent_inserted);
Josef Bacik00361582013-08-14 14:02:47 -0400331 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -0400332 btrfs_abort_transaction(trans, ret);
Josef Bacik00361582013-08-14 14:02:47 -0400333 goto out;
334 }
Chris Masonc8b97812008-10-29 14:49:59 -0400335
336 if (isize > actual_end)
337 inline_len = min_t(u64, isize, actual_end);
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000338 ret = insert_inline_extent(trans, path, extent_inserted,
339 root, inode, start,
Chris Masonc8b97812008-10-29 14:49:59 -0400340 inline_len, compressed_size,
Li Zefanfe3f5662011-03-28 08:30:38 +0000341 compress_type, compressed_pages);
Josef Bacik2adcac12012-05-23 16:10:14 -0400342 if (ret && ret != -ENOSPC) {
Jeff Mahoney66642832016-06-10 18:19:25 -0400343 btrfs_abort_transaction(trans, ret);
Josef Bacik00361582013-08-14 14:02:47 -0400344 goto out;
Josef Bacik2adcac12012-05-23 16:10:14 -0400345 } else if (ret == -ENOSPC) {
Josef Bacik00361582013-08-14 14:02:47 -0400346 ret = 1;
347 goto out;
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100348 }
Josef Bacik2adcac12012-05-23 16:10:14 -0400349
Josef Bacikbdc20e62013-02-28 13:23:38 -0500350 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
Nikolay Borisovdcdbc052017-02-20 13:50:45 +0200351 btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
Josef Bacik00361582013-08-14 14:02:47 -0400352out:
Qu Wenruo94ed9382015-09-08 17:25:56 +0800353 /*
354 * Don't forget to free the reserved space, as for inlined extent
355 * it won't count as data extent, free them directly here.
356 * And at reserve time, it's always aligned to page size, so
357 * just free one page here.
358 */
Qu Wenruobc42bda2017-02-27 15:10:39 +0800359 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000360 btrfs_free_path(path);
Jeff Mahoney3a45bb22016-09-09 21:39:03 -0400361 btrfs_end_transaction(trans);
Josef Bacik00361582013-08-14 14:02:47 -0400362 return ret;
Chris Masonc8b97812008-10-29 14:49:59 -0400363}
364
Chris Mason771ed682008-11-06 22:02:51 -0500365struct async_extent {
366 u64 start;
367 u64 ram_size;
368 u64 compressed_size;
369 struct page **pages;
370 unsigned long nr_pages;
Li Zefan261507a02010-12-17 14:21:50 +0800371 int compress_type;
Chris Mason771ed682008-11-06 22:02:51 -0500372 struct list_head list;
373};
374
375struct async_cow {
376 struct inode *inode;
377 struct btrfs_root *root;
378 struct page *locked_page;
379 u64 start;
380 u64 end;
Liu Bof82b7352017-10-23 23:18:16 -0600381 unsigned int write_flags;
Chris Mason771ed682008-11-06 22:02:51 -0500382 struct list_head extents;
383 struct btrfs_work work;
384};
385
386static noinline int add_async_extent(struct async_cow *cow,
387 u64 start, u64 ram_size,
388 u64 compressed_size,
389 struct page **pages,
Li Zefan261507a02010-12-17 14:21:50 +0800390 unsigned long nr_pages,
391 int compress_type)
Chris Mason771ed682008-11-06 22:02:51 -0500392{
393 struct async_extent *async_extent;
394
395 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100396 BUG_ON(!async_extent); /* -ENOMEM */
Chris Mason771ed682008-11-06 22:02:51 -0500397 async_extent->start = start;
398 async_extent->ram_size = ram_size;
399 async_extent->compressed_size = compressed_size;
400 async_extent->pages = pages;
401 async_extent->nr_pages = nr_pages;
Li Zefan261507a02010-12-17 14:21:50 +0800402 async_extent->compress_type = compress_type;
Chris Mason771ed682008-11-06 22:02:51 -0500403 list_add_tail(&async_extent->list, &cow->extents);
404 return 0;
405}
406
Timofey Titovetsc2fcdcd2017-07-17 16:52:58 +0300407static inline int inode_need_compress(struct inode *inode, u64 start, u64 end)
Wang Shilongf79707b2014-07-17 11:44:09 +0800408{
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400409 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Wang Shilongf79707b2014-07-17 11:44:09 +0800410
411 /* force compress */
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400412 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
Wang Shilongf79707b2014-07-17 11:44:09 +0800413 return 1;
David Sterbaeec63c62017-07-17 19:41:31 +0200414 /* defrag ioctl */
415 if (BTRFS_I(inode)->defrag_compress)
416 return 1;
Wang Shilongf79707b2014-07-17 11:44:09 +0800417 /* bad compression ratios */
418 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
419 return 0;
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400420 if (btrfs_test_opt(fs_info, COMPRESS) ||
Wang Shilongf79707b2014-07-17 11:44:09 +0800421 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
David Sterbab52aa8c2017-07-17 19:17:20 +0200422 BTRFS_I(inode)->prop_compress)
Timofey Titovetsc2fcdcd2017-07-17 16:52:58 +0300423 return btrfs_compress_heuristic(inode, start, end);
Wang Shilongf79707b2014-07-17 11:44:09 +0800424 return 0;
425}
426
Nikolay Borisov6158e1c2017-02-20 13:50:43 +0200427static inline void inode_should_defrag(struct btrfs_inode *inode,
Anand Jain26d30f82016-12-19 19:09:06 +0800428 u64 start, u64 end, u64 num_bytes, u64 small_write)
429{
430 /* If this is a small write inside eof, kick off a defrag */
431 if (num_bytes < small_write &&
Nikolay Borisov6158e1c2017-02-20 13:50:43 +0200432 (start > 0 || end + 1 < inode->disk_i_size))
Anand Jain26d30f82016-12-19 19:09:06 +0800433 btrfs_add_inode_defrag(NULL, inode);
434}
435
Chris Masonc8b97812008-10-29 14:49:59 -0400436/*
Chris Mason771ed682008-11-06 22:02:51 -0500437 * we create compressed extents in two phases. The first
438 * phase compresses a range of pages that have already been
439 * locked (both pages and state bits are locked).
Chris Masonc8b97812008-10-29 14:49:59 -0400440 *
Chris Mason771ed682008-11-06 22:02:51 -0500441 * This is done inside an ordered work queue, and the compression
442 * is spread across many cpus. The actual IO submission is step
443 * two, and the ordered work queue takes care of making sure that
444 * happens in the same order things were put onto the queue by
445 * writepages and friends.
Chris Masonc8b97812008-10-29 14:49:59 -0400446 *
Chris Mason771ed682008-11-06 22:02:51 -0500447 * If this code finds it can't get good compression, it puts an
448 * entry onto the work queue to write the uncompressed bytes. This
449 * makes sure that both compressed inodes and uncompressed inodes
Artem Bityutskiyb2570312012-07-25 18:12:06 +0300450 * are written in the same order that the flusher thread sent them
451 * down.
Chris Masond352ac62008-09-29 15:18:18 -0400452 */
Filipe Mananac44f6492014-10-09 21:15:44 +0100453static noinline void compress_file_range(struct inode *inode,
Chris Mason771ed682008-11-06 22:02:51 -0500454 struct page *locked_page,
455 u64 start, u64 end,
456 struct async_cow *async_cow,
457 int *num_added)
Chris Masonb888db22007-08-27 16:49:44 -0400458{
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400459 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Chris Masonb888db22007-08-27 16:49:44 -0400460 struct btrfs_root *root = BTRFS_I(inode)->root;
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400461 u64 blocksize = fs_info->sectorsize;
Chris Masonc8b97812008-10-29 14:49:59 -0400462 u64 actual_end;
Chris Mason42dc7ba2008-12-15 11:44:56 -0500463 u64 isize = i_size_read(inode);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400464 int ret = 0;
Chris Masonc8b97812008-10-29 14:49:59 -0400465 struct page **pages = NULL;
466 unsigned long nr_pages;
Chris Masonc8b97812008-10-29 14:49:59 -0400467 unsigned long total_compressed = 0;
468 unsigned long total_in = 0;
Chris Masonc8b97812008-10-29 14:49:59 -0400469 int i;
470 int will_compress;
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400471 int compress_type = fs_info->compress_type;
Chris Mason4adaa612013-03-26 13:07:00 -0400472 int redirty = 0;
Chris Masonb888db22007-08-27 16:49:44 -0400473
Nikolay Borisov6158e1c2017-02-20 13:50:43 +0200474 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
475 SZ_16K);
Chris Mason4cb53002011-05-24 15:35:30 -0400476
Chris Mason42dc7ba2008-12-15 11:44:56 -0500477 actual_end = min_t(u64, isize, end + 1);
Chris Masonc8b97812008-10-29 14:49:59 -0400478again:
479 will_compress = 0;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300480 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
David Sterba069eac72017-02-14 19:36:54 +0100481 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
482 nr_pages = min_t(unsigned long, nr_pages,
483 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
Chris Masonc8b97812008-10-29 14:49:59 -0400484
Chris Masonf03d9301f2009-02-04 09:31:06 -0500485 /*
486 * we don't want to send crud past the end of i_size through
487 * compression, that's just a waste of CPU time. So, if the
488 * end of the file is before the start of our current
489 * requested range of bytes, we bail out to the uncompressed
490 * cleanup code that can deal with all of this.
491 *
492 * It isn't really the fastest way to fix things, but this is a
493 * very uncommon corner.
494 */
495 if (actual_end <= start)
496 goto cleanup_and_bail_uncompressed;
497
Chris Masonc8b97812008-10-29 14:49:59 -0400498 total_compressed = actual_end - start;
499
Shilong Wang4bcbb332014-10-07 18:44:35 -0400500 /*
501 * skip compression for a small file range(<=blocksize) that
Nicholas D Steeves01327612016-05-19 21:18:45 -0400502 * isn't an inline extent, since it doesn't save disk space at all.
Shilong Wang4bcbb332014-10-07 18:44:35 -0400503 */
504 if (total_compressed <= blocksize &&
505 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
506 goto cleanup_and_bail_uncompressed;
507
David Sterba069eac72017-02-14 19:36:54 +0100508 total_compressed = min_t(unsigned long, total_compressed,
509 BTRFS_MAX_UNCOMPRESSED);
Chris Masonc8b97812008-10-29 14:49:59 -0400510 total_in = 0;
511 ret = 0;
Chris Masondb945352007-10-15 16:15:53 -0400512
Chris Mason771ed682008-11-06 22:02:51 -0500513 /*
514 * we do compression for mount -o compress and when the
515 * inode has not been flagged as nocompress. This flag can
516 * change at any time if we discover bad compression ratios.
Chris Masonc8b97812008-10-29 14:49:59 -0400517 */
Timofey Titovetsc2fcdcd2017-07-17 16:52:58 +0300518 if (inode_need_compress(inode, start, end)) {
Chris Masonc8b97812008-10-29 14:49:59 -0400519 WARN_ON(pages);
David Sterba31e818f2015-02-20 18:00:26 +0100520 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
Li Zefan560f7d72011-09-08 10:22:01 +0800521 if (!pages) {
522 /* just bail out to the uncompressed code */
523 goto cont;
524 }
Chris Mason179e29e2007-11-01 11:28:41 -0400525
David Sterbaeec63c62017-07-17 19:41:31 +0200526 if (BTRFS_I(inode)->defrag_compress)
527 compress_type = BTRFS_I(inode)->defrag_compress;
528 else if (BTRFS_I(inode)->prop_compress)
David Sterbab52aa8c2017-07-17 19:17:20 +0200529 compress_type = BTRFS_I(inode)->prop_compress;
Li Zefan261507a02010-12-17 14:21:50 +0800530
Chris Mason4adaa612013-03-26 13:07:00 -0400531 /*
532 * we need to call clear_page_dirty_for_io on each
533 * page in the range. Otherwise applications with the file
534 * mmap'd can wander in and change the page contents while
535 * we are compressing them.
536 *
537 * If the compression fails for any reason, we set the pages
538 * dirty again later on.
539 */
540 extent_range_clear_dirty_for_io(inode, start, end);
541 redirty = 1;
David Sterbaf51d2b52017-09-15 17:36:57 +0200542
543 /* Compression level is applied here and only here */
544 ret = btrfs_compress_pages(
545 compress_type | (fs_info->compress_level << 4),
Li Zefan261507a02010-12-17 14:21:50 +0800546 inode->i_mapping, start,
David Sterba38c31462017-02-14 19:04:07 +0100547 pages,
David Sterba4d3a8002017-02-14 19:04:07 +0100548 &nr_pages,
Li Zefan261507a02010-12-17 14:21:50 +0800549 &total_in,
David Sterbae5d74902017-02-14 19:45:05 +0100550 &total_compressed);
Chris Masonc8b97812008-10-29 14:49:59 -0400551
552 if (!ret) {
553 unsigned long offset = total_compressed &
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300554 (PAGE_SIZE - 1);
David Sterba4d3a8002017-02-14 19:04:07 +0100555 struct page *page = pages[nr_pages - 1];
Chris Masonc8b97812008-10-29 14:49:59 -0400556 char *kaddr;
557
558 /* zero the tail end of the last page, we might be
559 * sending it down to disk
560 */
561 if (offset) {
Cong Wang7ac687d2011-11-25 23:14:28 +0800562 kaddr = kmap_atomic(page);
Chris Masonc8b97812008-10-29 14:49:59 -0400563 memset(kaddr + offset, 0,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300564 PAGE_SIZE - offset);
Cong Wang7ac687d2011-11-25 23:14:28 +0800565 kunmap_atomic(kaddr);
Chris Masonc8b97812008-10-29 14:49:59 -0400566 }
567 will_compress = 1;
568 }
569 }
Li Zefan560f7d72011-09-08 10:22:01 +0800570cont:
Chris Masonc8b97812008-10-29 14:49:59 -0400571 if (start == 0) {
572 /* lets try to make an inline extent */
Timofey Titovets6018ba02017-09-15 01:57:26 +0300573 if (ret || total_in < actual_end) {
Chris Masonc8b97812008-10-29 14:49:59 -0400574 /* we didn't compress the entire range, try
Chris Mason771ed682008-11-06 22:02:51 -0500575 * to make an uncompressed inline extent.
Chris Masonc8b97812008-10-29 14:49:59 -0400576 */
Josef Bacik00361582013-08-14 14:02:47 -0400577 ret = cow_file_range_inline(root, inode, start, end,
Anand Jainf74670f2016-12-06 12:43:07 +0800578 0, BTRFS_COMPRESS_NONE, NULL);
Chris Masonc8b97812008-10-29 14:49:59 -0400579 } else {
Chris Mason771ed682008-11-06 22:02:51 -0500580 /* try making a compressed inline extent */
Josef Bacik00361582013-08-14 14:02:47 -0400581 ret = cow_file_range_inline(root, inode, start, end,
Li Zefanfe3f5662011-03-28 08:30:38 +0000582 total_compressed,
583 compress_type, pages);
Chris Masonc8b97812008-10-29 14:49:59 -0400584 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100585 if (ret <= 0) {
Josef Bacik151a41b2013-07-29 13:22:24 -0400586 unsigned long clear_flags = EXTENT_DELALLOC |
Josef Bacik8b62f872017-10-19 14:15:55 -0400587 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
588 EXTENT_DO_ACCOUNTING;
Filipe Mananae6eb4312014-10-10 10:45:12 +0100589 unsigned long page_error_op;
590
Filipe Mananae6eb4312014-10-10 10:45:12 +0100591 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
Josef Bacik151a41b2013-07-29 13:22:24 -0400592
Chris Mason771ed682008-11-06 22:02:51 -0500593 /*
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100594 * inline extent creation worked or returned error,
595 * we don't need to create any more async work items.
596 * Unlock and free up our temp pages.
Josef Bacik8b62f872017-10-19 14:15:55 -0400597 *
598 * We use DO_ACCOUNTING here because we need the
599 * delalloc_release_metadata to be done _after_ we drop
600 * our outstanding extent for clearing delalloc for this
601 * range.
Chris Mason771ed682008-11-06 22:02:51 -0500602 */
Qu Wenruoba8b04c2016-07-19 16:50:36 +0800603 extent_clear_unlock_delalloc(inode, start, end, end,
604 NULL, clear_flags,
605 PAGE_UNLOCK |
Josef Bacikc2790a22013-07-29 11:20:47 -0400606 PAGE_CLEAR_DIRTY |
607 PAGE_SET_WRITEBACK |
Filipe Mananae6eb4312014-10-10 10:45:12 +0100608 page_error_op |
Josef Bacikc2790a22013-07-29 11:20:47 -0400609 PAGE_END_WRITEBACK);
Chris Masonc8b97812008-10-29 14:49:59 -0400610 goto free_pages_out;
611 }
612 }
613
614 if (will_compress) {
615 /*
616 * we aren't doing an inline extent round the compressed size
617 * up to a block size boundary so the allocator does sane
618 * things
619 */
Qu Wenruofda28322013-02-26 08:10:22 +0000620 total_compressed = ALIGN(total_compressed, blocksize);
Chris Masonc8b97812008-10-29 14:49:59 -0400621
622 /*
623 * one last check to make sure the compression is really a
Timofey Titovets170607e2017-06-06 14:41:15 +0300624 * win, compare the page count read with the blocks on disk,
625 * compression must free at least one sector size
Chris Masonc8b97812008-10-29 14:49:59 -0400626 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300627 total_in = ALIGN(total_in, PAGE_SIZE);
Timofey Titovets170607e2017-06-06 14:41:15 +0300628 if (total_compressed + blocksize <= total_in) {
Ashish Samantc8bb0c82016-03-25 19:01:33 -0700629 *num_added += 1;
630
631 /*
632 * The async work queues will take care of doing actual
633 * allocation on disk for these compressed pages, and
634 * will submit them to the elevator.
635 */
Timofey Titovets11708622017-10-03 18:06:01 +0300636 add_async_extent(async_cow, start, total_in,
David Sterba4d3a8002017-02-14 19:04:07 +0100637 total_compressed, pages, nr_pages,
Ashish Samantc8bb0c82016-03-25 19:01:33 -0700638 compress_type);
639
Timofey Titovets11708622017-10-03 18:06:01 +0300640 if (start + total_in < end) {
641 start += total_in;
Ashish Samantc8bb0c82016-03-25 19:01:33 -0700642 pages = NULL;
643 cond_resched();
644 goto again;
645 }
646 return;
Chris Masonc8b97812008-10-29 14:49:59 -0400647 }
648 }
Ashish Samantc8bb0c82016-03-25 19:01:33 -0700649 if (pages) {
Chris Masonc8b97812008-10-29 14:49:59 -0400650 /*
651 * the compression code ran but failed to make things smaller,
652 * free any pages it allocated and our page pointer array
653 */
David Sterba4d3a8002017-02-14 19:04:07 +0100654 for (i = 0; i < nr_pages; i++) {
Chris Mason70b99e62008-10-31 12:46:39 -0400655 WARN_ON(pages[i]->mapping);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300656 put_page(pages[i]);
Chris Masonc8b97812008-10-29 14:49:59 -0400657 }
658 kfree(pages);
659 pages = NULL;
660 total_compressed = 0;
David Sterba4d3a8002017-02-14 19:04:07 +0100661 nr_pages = 0;
Chris Masonc8b97812008-10-29 14:49:59 -0400662
663 /* flag the file so we don't compress in the future */
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400664 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
David Sterbab52aa8c2017-07-17 19:17:20 +0200665 !(BTRFS_I(inode)->prop_compress)) {
Chris Masona555f812010-01-28 16:18:15 -0500666 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
Chris Mason1e701a32010-03-11 09:42:04 -0500667 }
Chris Masonc8b97812008-10-29 14:49:59 -0400668 }
Chris Masonf03d9301f2009-02-04 09:31:06 -0500669cleanup_and_bail_uncompressed:
Ashish Samantc8bb0c82016-03-25 19:01:33 -0700670 /*
671 * No compression, but we still need to write the pages in the file
672 * we've been given so far. redirty the locked page if it corresponds
673 * to our extent and set things up for the async work queue to run
674 * cow_file_range to do the normal delalloc dance.
675 */
676 if (page_offset(locked_page) >= start &&
677 page_offset(locked_page) <= end)
678 __set_page_dirty_nobuffers(locked_page);
679 /* unlocked later on in the async handlers */
680
681 if (redirty)
682 extent_range_redirty_for_io(inode, start, end);
683 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
684 BTRFS_COMPRESS_NONE);
685 *num_added += 1;
Chris Mason771ed682008-11-06 22:02:51 -0500686
Filipe Mananac44f6492014-10-09 21:15:44 +0100687 return;
Chris Mason771ed682008-11-06 22:02:51 -0500688
689free_pages_out:
David Sterba4d3a8002017-02-14 19:04:07 +0100690 for (i = 0; i < nr_pages; i++) {
Chris Mason771ed682008-11-06 22:02:51 -0500691 WARN_ON(pages[i]->mapping);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300692 put_page(pages[i]);
Chris Mason771ed682008-11-06 22:02:51 -0500693 }
Chris Masond3977122009-01-05 21:25:51 -0500694 kfree(pages);
Chris Mason771ed682008-11-06 22:02:51 -0500695}
Chris Mason771ed682008-11-06 22:02:51 -0500696
Filipe Manana40ae8372014-10-06 22:14:24 +0100697static void free_async_extent_pages(struct async_extent *async_extent)
698{
699 int i;
700
701 if (!async_extent->pages)
702 return;
703
704 for (i = 0; i < async_extent->nr_pages; i++) {
705 WARN_ON(async_extent->pages[i]->mapping);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300706 put_page(async_extent->pages[i]);
Filipe Manana40ae8372014-10-06 22:14:24 +0100707 }
708 kfree(async_extent->pages);
709 async_extent->nr_pages = 0;
710 async_extent->pages = NULL;
Chris Mason771ed682008-11-06 22:02:51 -0500711}
712
713/*
714 * phase two of compressed writeback. This is the ordered portion
715 * of the code, which only gets called in the order the work was
716 * queued. We walk all the async extents created by compress_file_range
717 * and send them down to the disk.
718 */
Filipe Mananadec8f172014-10-06 22:14:26 +0100719static noinline void submit_compressed_extents(struct inode *inode,
Chris Mason771ed682008-11-06 22:02:51 -0500720 struct async_cow *async_cow)
721{
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400722 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Chris Mason771ed682008-11-06 22:02:51 -0500723 struct async_extent *async_extent;
724 u64 alloc_hint = 0;
Chris Mason771ed682008-11-06 22:02:51 -0500725 struct btrfs_key ins;
726 struct extent_map *em;
727 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason771ed682008-11-06 22:02:51 -0500728 struct extent_io_tree *io_tree;
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500729 int ret = 0;
Chris Mason771ed682008-11-06 22:02:51 -0500730
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500731again:
Chris Masond3977122009-01-05 21:25:51 -0500732 while (!list_empty(&async_cow->extents)) {
Chris Mason771ed682008-11-06 22:02:51 -0500733 async_extent = list_entry(async_cow->extents.next,
734 struct async_extent, list);
735 list_del(&async_extent->list);
736
737 io_tree = &BTRFS_I(inode)->io_tree;
738
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500739retry:
Chris Mason771ed682008-11-06 22:02:51 -0500740 /* did the compression code fall back to uncompressed IO? */
741 if (!async_extent->pages) {
742 int page_started = 0;
743 unsigned long nr_written = 0;
744
745 lock_extent(io_tree, async_extent->start,
Josef Bacik2ac55d42010-02-03 19:33:23 +0000746 async_extent->start +
Jeff Mahoneyd0082372012-03-01 14:57:19 +0100747 async_extent->ram_size - 1);
Chris Mason771ed682008-11-06 22:02:51 -0500748
749 /* allocate blocks */
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500750 ret = cow_file_range(inode, async_cow->locked_page,
751 async_extent->start,
752 async_extent->start +
753 async_extent->ram_size - 1,
Wang Xiaoguangdda32452016-07-11 11:05:29 +0800754 async_extent->start +
755 async_extent->ram_size - 1,
756 &page_started, &nr_written, 0,
757 NULL);
Chris Mason771ed682008-11-06 22:02:51 -0500758
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100759 /* JDM XXX */
760
Chris Mason771ed682008-11-06 22:02:51 -0500761 /*
762 * if page_started, cow_file_range inserted an
763 * inline extent and took care of all the unlocking
764 * and IO for us. Otherwise, we need to submit
765 * all those pages down to the drive.
766 */
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500767 if (!page_started && !ret)
Chris Mason771ed682008-11-06 22:02:51 -0500768 extent_write_locked_range(io_tree,
769 inode, async_extent->start,
Chris Masond3977122009-01-05 21:25:51 -0500770 async_extent->start +
Chris Mason771ed682008-11-06 22:02:51 -0500771 async_extent->ram_size - 1,
772 btrfs_get_extent,
773 WB_SYNC_ALL);
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500774 else if (ret)
775 unlock_page(async_cow->locked_page);
Chris Mason771ed682008-11-06 22:02:51 -0500776 kfree(async_extent);
777 cond_resched();
778 continue;
779 }
780
781 lock_extent(io_tree, async_extent->start,
Jeff Mahoneyd0082372012-03-01 14:57:19 +0100782 async_extent->start + async_extent->ram_size - 1);
Chris Mason771ed682008-11-06 22:02:51 -0500783
Wang Xiaoguang18513092016-07-25 15:51:40 +0800784 ret = btrfs_reserve_extent(root, async_extent->ram_size,
Chris Mason771ed682008-11-06 22:02:51 -0500785 async_extent->compressed_size,
786 async_extent->compressed_size,
Miao Xiee570fd22014-06-19 10:42:50 +0800787 0, alloc_hint, &ins, 1, 1);
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500788 if (ret) {
Filipe Manana40ae8372014-10-06 22:14:24 +0100789 free_async_extent_pages(async_extent);
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500790
Josef Bacikfdf8e2e2013-06-14 16:58:23 -0400791 if (ret == -ENOSPC) {
792 unlock_extent(io_tree, async_extent->start,
793 async_extent->start +
794 async_extent->ram_size - 1);
Liu Boce620032014-07-24 22:48:05 +0800795
796 /*
797 * we need to redirty the pages if we decide to
798 * fallback to uncompressed IO, otherwise we
799 * will not submit these pages down to lower
800 * layers.
801 */
802 extent_range_redirty_for_io(inode,
803 async_extent->start,
804 async_extent->start +
805 async_extent->ram_size - 1);
806
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100807 goto retry;
Josef Bacikfdf8e2e2013-06-14 16:58:23 -0400808 }
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500809 goto out_free;
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500810 }
Yan, Zhengc2167752009-11-12 09:34:21 +0000811 /*
812 * here we're doing allocation and writeback of the
813 * compressed pages
814 */
Liu Bo6f9994d2017-01-31 07:50:22 -0800815 em = create_io_em(inode, async_extent->start,
816 async_extent->ram_size, /* len */
817 async_extent->start, /* orig_start */
818 ins.objectid, /* block_start */
819 ins.offset, /* block_len */
820 ins.offset, /* orig_block_len */
821 async_extent->ram_size, /* ram_bytes */
822 async_extent->compress_type,
823 BTRFS_ORDERED_COMPRESSED);
824 if (IS_ERR(em))
825 /* ret value is not necessary due to void function */
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500826 goto out_free_reserve;
Liu Bo6f9994d2017-01-31 07:50:22 -0800827 free_extent_map(em);
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500828
Li Zefan261507a02010-12-17 14:21:50 +0800829 ret = btrfs_add_ordered_extent_compress(inode,
830 async_extent->start,
831 ins.objectid,
832 async_extent->ram_size,
833 ins.offset,
834 BTRFS_ORDERED_COMPRESSED,
835 async_extent->compress_type);
Filipe Mananad9f85962014-08-25 10:43:00 +0100836 if (ret) {
Nikolay Borisovdcdbc052017-02-20 13:50:45 +0200837 btrfs_drop_extent_cache(BTRFS_I(inode),
838 async_extent->start,
Filipe Mananad9f85962014-08-25 10:43:00 +0100839 async_extent->start +
840 async_extent->ram_size - 1, 0);
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500841 goto out_free_reserve;
Filipe Mananad9f85962014-08-25 10:43:00 +0100842 }
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400843 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
Chris Mason771ed682008-11-06 22:02:51 -0500844
Chris Mason771ed682008-11-06 22:02:51 -0500845 /*
846 * clear dirty, set writeback and unlock the pages.
847 */
Josef Bacikc2790a22013-07-29 11:20:47 -0400848 extent_clear_unlock_delalloc(inode, async_extent->start,
Chris Masona791e352009-10-08 11:27:10 -0400849 async_extent->start +
850 async_extent->ram_size - 1,
Qu Wenruoba8b04c2016-07-19 16:50:36 +0800851 async_extent->start +
852 async_extent->ram_size - 1,
Josef Bacik151a41b2013-07-29 13:22:24 -0400853 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
854 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
Josef Bacikc2790a22013-07-29 11:20:47 -0400855 PAGE_SET_WRITEBACK);
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200856 if (btrfs_submit_compressed_write(inode,
Chris Masond3977122009-01-05 21:25:51 -0500857 async_extent->start,
858 async_extent->ram_size,
859 ins.objectid,
860 ins.offset, async_extent->pages,
Liu Bof82b7352017-10-23 23:18:16 -0600861 async_extent->nr_pages,
862 async_cow->write_flags)) {
Filipe Mananafce2a4e2014-10-06 22:14:23 +0100863 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
864 struct page *p = async_extent->pages[0];
865 const u64 start = async_extent->start;
866 const u64 end = start + async_extent->ram_size - 1;
867
868 p->mapping = inode->i_mapping;
869 tree->ops->writepage_end_io_hook(p, start, end,
870 NULL, 0);
871 p->mapping = NULL;
Qu Wenruoba8b04c2016-07-19 16:50:36 +0800872 extent_clear_unlock_delalloc(inode, start, end, end,
873 NULL, 0,
Filipe Mananafce2a4e2014-10-06 22:14:23 +0100874 PAGE_END_WRITEBACK |
875 PAGE_SET_ERROR);
Filipe Manana40ae8372014-10-06 22:14:24 +0100876 free_async_extent_pages(async_extent);
Filipe Mananafce2a4e2014-10-06 22:14:23 +0100877 }
Chris Mason771ed682008-11-06 22:02:51 -0500878 alloc_hint = ins.objectid + ins.offset;
879 kfree(async_extent);
880 cond_resched();
881 }
Filipe Mananadec8f172014-10-06 22:14:26 +0100882 return;
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500883out_free_reserve:
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400884 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400885 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100886out_free:
Josef Bacikc2790a22013-07-29 11:20:47 -0400887 extent_clear_unlock_delalloc(inode, async_extent->start,
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500888 async_extent->start +
889 async_extent->ram_size - 1,
Qu Wenruoba8b04c2016-07-19 16:50:36 +0800890 async_extent->start +
891 async_extent->ram_size - 1,
Josef Bacikc2790a22013-07-29 11:20:47 -0400892 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
Filipe Mananaa7e3b972017-04-03 10:45:46 +0100893 EXTENT_DELALLOC_NEW |
Josef Bacik151a41b2013-07-29 13:22:24 -0400894 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
895 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
Filipe Manana704de492014-10-06 22:14:22 +0100896 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
897 PAGE_SET_ERROR);
Filipe Manana40ae8372014-10-06 22:14:24 +0100898 free_async_extent_pages(async_extent);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100899 kfree(async_extent);
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500900 goto again;
Chris Mason771ed682008-11-06 22:02:51 -0500901}
902
Josef Bacik4b46fce2010-05-23 11:00:55 -0400903static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
904 u64 num_bytes)
905{
906 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
907 struct extent_map *em;
908 u64 alloc_hint = 0;
909
910 read_lock(&em_tree->lock);
911 em = search_extent_mapping(em_tree, start, num_bytes);
912 if (em) {
913 /*
914 * if block start isn't an actual block number then find the
915 * first block in this inode and use that as a hint. If that
916 * block is also bogus then just don't worry about it.
917 */
918 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
919 free_extent_map(em);
920 em = search_extent_mapping(em_tree, 0, 0);
921 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
922 alloc_hint = em->block_start;
923 if (em)
924 free_extent_map(em);
925 } else {
926 alloc_hint = em->block_start;
927 free_extent_map(em);
928 }
929 }
930 read_unlock(&em_tree->lock);
931
932 return alloc_hint;
933}
934
Chris Mason771ed682008-11-06 22:02:51 -0500935/*
936 * when extent_io.c finds a delayed allocation range in the file,
937 * the call backs end up in this code. The basic idea is to
938 * allocate extents on disk for the range, and create ordered data structs
939 * in ram to track those extents.
940 *
941 * locked_page is the page that writepage had locked already. We use
942 * it to make sure we don't do extra locks or unlocks.
943 *
944 * *page_started is set to one if we unlock locked_page and do everything
945 * required to start IO on it. It may be clean and already done with
946 * IO when we return.
947 */
Josef Bacik00361582013-08-14 14:02:47 -0400948static noinline int cow_file_range(struct inode *inode,
949 struct page *locked_page,
Wang Xiaoguangdda32452016-07-11 11:05:29 +0800950 u64 start, u64 end, u64 delalloc_end,
951 int *page_started, unsigned long *nr_written,
952 int unlock, struct btrfs_dedupe_hash *hash)
Chris Mason771ed682008-11-06 22:02:51 -0500953{
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400954 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Josef Bacik00361582013-08-14 14:02:47 -0400955 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason771ed682008-11-06 22:02:51 -0500956 u64 alloc_hint = 0;
957 u64 num_bytes;
958 unsigned long ram_size;
959 u64 disk_num_bytes;
Filipe Mananaa315e682017-03-06 23:04:20 +0000960 u64 cur_alloc_size = 0;
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400961 u64 blocksize = fs_info->sectorsize;
Chris Mason771ed682008-11-06 22:02:51 -0500962 struct btrfs_key ins;
963 struct extent_map *em;
Filipe Mananaa315e682017-03-06 23:04:20 +0000964 unsigned clear_bits;
965 unsigned long page_ops;
966 bool extent_reserved = false;
Chris Mason771ed682008-11-06 22:02:51 -0500967 int ret = 0;
968
Nikolay Borisov70ddc552017-02-20 13:50:35 +0200969 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
Josef Bacik02ecd2c2013-10-25 16:19:08 -0400970 WARN_ON_ONCE(1);
Josef Bacik29bce2f2014-02-07 12:21:23 -0500971 ret = -EINVAL;
972 goto out_unlock;
Josef Bacik02ecd2c2013-10-25 16:19:08 -0400973 }
Chris Mason771ed682008-11-06 22:02:51 -0500974
Qu Wenruofda28322013-02-26 08:10:22 +0000975 num_bytes = ALIGN(end - start + 1, blocksize);
Chris Mason771ed682008-11-06 22:02:51 -0500976 num_bytes = max(blocksize, num_bytes);
977 disk_num_bytes = num_bytes;
Chris Mason771ed682008-11-06 22:02:51 -0500978
Nikolay Borisov6158e1c2017-02-20 13:50:43 +0200979 inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
Chris Mason4cb53002011-05-24 15:35:30 -0400980
Chris Mason771ed682008-11-06 22:02:51 -0500981 if (start == 0) {
982 /* lets try to make an inline extent */
Anand Jainf74670f2016-12-06 12:43:07 +0800983 ret = cow_file_range_inline(root, inode, start, end, 0,
984 BTRFS_COMPRESS_NONE, NULL);
Chris Mason771ed682008-11-06 22:02:51 -0500985 if (ret == 0) {
Josef Bacik8b62f872017-10-19 14:15:55 -0400986 /*
987 * We use DO_ACCOUNTING here because we need the
988 * delalloc_release_metadata to be run _after_ we drop
989 * our outstanding extent for clearing delalloc for this
990 * range.
991 */
Qu Wenruoba8b04c2016-07-19 16:50:36 +0800992 extent_clear_unlock_delalloc(inode, start, end,
993 delalloc_end, NULL,
Josef Bacikc2790a22013-07-29 11:20:47 -0400994 EXTENT_LOCKED | EXTENT_DELALLOC |
Josef Bacik8b62f872017-10-19 14:15:55 -0400995 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
996 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
Josef Bacikc2790a22013-07-29 11:20:47 -0400997 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
998 PAGE_END_WRITEBACK);
Chris Mason771ed682008-11-06 22:02:51 -0500999 *nr_written = *nr_written +
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001000 (end - start + PAGE_SIZE) / PAGE_SIZE;
Chris Mason771ed682008-11-06 22:02:51 -05001001 *page_started = 1;
Chris Mason771ed682008-11-06 22:02:51 -05001002 goto out;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001003 } else if (ret < 0) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001004 goto out_unlock;
Chris Mason771ed682008-11-06 22:02:51 -05001005 }
1006 }
Chris Masonc8b97812008-10-29 14:49:59 -04001007
1008 BUG_ON(disk_num_bytes >
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001009 btrfs_super_total_bytes(fs_info->super_copy));
Chris Masonc8b97812008-10-29 14:49:59 -04001010
Josef Bacik4b46fce2010-05-23 11:00:55 -04001011 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
Nikolay Borisovdcdbc052017-02-20 13:50:45 +02001012 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1013 start + num_bytes - 1, 0);
Chris Mason3b951512008-04-17 11:29:12 -04001014
Chris Masond3977122009-01-05 21:25:51 -05001015 while (disk_num_bytes > 0) {
Josef Bacik287a0ab2010-03-19 18:07:23 +00001016 cur_alloc_size = disk_num_bytes;
Wang Xiaoguang18513092016-07-25 15:51:40 +08001017 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001018 fs_info->sectorsize, 0, alloc_hint,
Miao Xiee570fd22014-06-19 10:42:50 +08001019 &ins, 1, 1);
Josef Bacik00361582013-08-14 14:02:47 -04001020 if (ret < 0)
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001021 goto out_unlock;
Filipe Mananaa315e682017-03-06 23:04:20 +00001022 cur_alloc_size = ins.offset;
1023 extent_reserved = true;
Chris Masond3977122009-01-05 21:25:51 -05001024
Chris Mason771ed682008-11-06 22:02:51 -05001025 ram_size = ins.offset;
Liu Bo6f9994d2017-01-31 07:50:22 -08001026 em = create_io_em(inode, start, ins.offset, /* len */
1027 start, /* orig_start */
1028 ins.objectid, /* block_start */
1029 ins.offset, /* block_len */
1030 ins.offset, /* orig_block_len */
1031 ram_size, /* ram_bytes */
1032 BTRFS_COMPRESS_NONE, /* compress_type */
Liu Bo1af4a0a2017-02-13 15:35:09 -08001033 BTRFS_ORDERED_REGULAR /* type */);
Liu Bo6f9994d2017-01-31 07:50:22 -08001034 if (IS_ERR(em))
Liu Boace68ba2013-04-22 10:53:47 +00001035 goto out_reserve;
Liu Bo6f9994d2017-01-31 07:50:22 -08001036 free_extent_map(em);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001037
Chris Masone6dcd2d2008-07-17 12:53:50 -04001038 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
Chris Mason771ed682008-11-06 22:02:51 -05001039 ram_size, cur_alloc_size, 0);
Liu Boace68ba2013-04-22 10:53:47 +00001040 if (ret)
Filipe Mananad9f85962014-08-25 10:43:00 +01001041 goto out_drop_extent_cache;
Chris Masonc8b97812008-10-29 14:49:59 -04001042
Yan Zheng17d217f2008-12-12 10:03:38 -05001043 if (root->root_key.objectid ==
1044 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1045 ret = btrfs_reloc_clone_csums(inode, start,
1046 cur_alloc_size);
Qu Wenruo4dbd80f2017-03-08 10:25:51 +08001047 /*
1048 * Only drop cache here, and process as normal.
1049 *
1050 * We must not allow extent_clear_unlock_delalloc()
1051 * at out_unlock label to free meta of this ordered
1052 * extent, as its meta should be freed by
1053 * btrfs_finish_ordered_io().
1054 *
1055 * So we must continue until @start is increased to
1056 * skip current ordered extent.
1057 */
Josef Bacik00361582013-08-14 14:02:47 -04001058 if (ret)
Qu Wenruo4dbd80f2017-03-08 10:25:51 +08001059 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1060 start + ram_size - 1, 0);
Yan Zheng17d217f2008-12-12 10:03:38 -05001061 }
1062
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001063 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
Filipe Manana9cfa3e32016-04-26 15:39:32 +01001064
Chris Masonc8b97812008-10-29 14:49:59 -04001065 /* we're not doing compressed IO, don't unlock the first
1066 * page (which the caller expects to stay locked), don't
1067 * clear any dirty bits and don't set any writeback bits
Chris Mason8b62b722009-09-02 16:53:46 -04001068 *
1069 * Do set the Private2 bit so we know this page was properly
1070 * setup for writepage
Chris Masonc8b97812008-10-29 14:49:59 -04001071 */
Filipe Mananaa315e682017-03-06 23:04:20 +00001072 page_ops = unlock ? PAGE_UNLOCK : 0;
1073 page_ops |= PAGE_SET_PRIVATE2;
Chris Masona791e352009-10-08 11:27:10 -04001074
Josef Bacikc2790a22013-07-29 11:20:47 -04001075 extent_clear_unlock_delalloc(inode, start,
Qu Wenruoba8b04c2016-07-19 16:50:36 +08001076 start + ram_size - 1,
1077 delalloc_end, locked_page,
Josef Bacikc2790a22013-07-29 11:20:47 -04001078 EXTENT_LOCKED | EXTENT_DELALLOC,
Filipe Mananaa315e682017-03-06 23:04:20 +00001079 page_ops);
Qu Wenruo4dbd80f2017-03-08 10:25:51 +08001080 if (disk_num_bytes < cur_alloc_size)
1081 disk_num_bytes = 0;
1082 else
1083 disk_num_bytes -= cur_alloc_size;
Chris Masonc59f8952007-12-17 20:14:04 -05001084 num_bytes -= cur_alloc_size;
1085 alloc_hint = ins.objectid + ins.offset;
1086 start += cur_alloc_size;
Filipe Mananaa315e682017-03-06 23:04:20 +00001087 extent_reserved = false;
Qu Wenruo4dbd80f2017-03-08 10:25:51 +08001088
1089 /*
1090 * btrfs_reloc_clone_csums() error, since start is increased
1091 * extent_clear_unlock_delalloc() at out_unlock label won't
1092 * free metadata of current ordered extent, we're OK to exit.
1093 */
1094 if (ret)
1095 goto out_unlock;
Chris Masonb888db22007-08-27 16:49:44 -04001096 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001097out:
Chris Masonbe20aa92007-12-17 20:14:01 -05001098 return ret;
Miao Xieb7d5b0a2012-11-01 07:32:18 +00001099
Filipe Mananad9f85962014-08-25 10:43:00 +01001100out_drop_extent_cache:
Nikolay Borisovdcdbc052017-02-20 13:50:45 +02001101 btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
Liu Boace68ba2013-04-22 10:53:47 +00001102out_reserve:
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001103 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04001104 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001105out_unlock:
Filipe Mananaa7e3b972017-04-03 10:45:46 +01001106 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1107 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
Filipe Mananaa315e682017-03-06 23:04:20 +00001108 page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1109 PAGE_END_WRITEBACK;
1110 /*
1111 * If we reserved an extent for our delalloc range (or a subrange) and
1112 * failed to create the respective ordered extent, then it means that
1113 * when we reserved the extent we decremented the extent's size from
1114 * the data space_info's bytes_may_use counter and incremented the
1115 * space_info's bytes_reserved counter by the same amount. We must make
1116 * sure extent_clear_unlock_delalloc() does not try to decrement again
1117 * the data space_info's bytes_may_use counter, therefore we do not pass
1118 * it the flag EXTENT_CLEAR_DATA_RESV.
1119 */
1120 if (extent_reserved) {
1121 extent_clear_unlock_delalloc(inode, start,
1122 start + cur_alloc_size,
1123 start + cur_alloc_size,
1124 locked_page,
1125 clear_bits,
1126 page_ops);
1127 start += cur_alloc_size;
1128 if (start >= end)
1129 goto out;
1130 }
Qu Wenruoba8b04c2016-07-19 16:50:36 +08001131 extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1132 locked_page,
Filipe Mananaa315e682017-03-06 23:04:20 +00001133 clear_bits | EXTENT_CLEAR_DATA_RESV,
1134 page_ops);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001135 goto out;
Chris Mason771ed682008-11-06 22:02:51 -05001136}
Chris Masonc8b97812008-10-29 14:49:59 -04001137
Chris Mason771ed682008-11-06 22:02:51 -05001138/*
1139 * work queue call back to started compression on a file and pages
1140 */
1141static noinline void async_cow_start(struct btrfs_work *work)
1142{
1143 struct async_cow *async_cow;
1144 int num_added = 0;
1145 async_cow = container_of(work, struct async_cow, work);
1146
1147 compress_file_range(async_cow->inode, async_cow->locked_page,
1148 async_cow->start, async_cow->end, async_cow,
1149 &num_added);
Josef Bacik8180ef82012-06-08 15:16:12 -04001150 if (num_added == 0) {
Josef Bacikcb77fcd2012-06-15 12:19:48 -06001151 btrfs_add_delayed_iput(async_cow->inode);
Chris Mason771ed682008-11-06 22:02:51 -05001152 async_cow->inode = NULL;
Josef Bacik8180ef82012-06-08 15:16:12 -04001153 }
Chris Mason771ed682008-11-06 22:02:51 -05001154}
1155
1156/*
1157 * work queue call back to submit previously compressed pages
1158 */
1159static noinline void async_cow_submit(struct btrfs_work *work)
1160{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001161 struct btrfs_fs_info *fs_info;
Chris Mason771ed682008-11-06 22:02:51 -05001162 struct async_cow *async_cow;
1163 struct btrfs_root *root;
1164 unsigned long nr_pages;
1165
1166 async_cow = container_of(work, struct async_cow, work);
1167
1168 root = async_cow->root;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001169 fs_info = root->fs_info;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001170 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1171 PAGE_SHIFT;
Chris Mason771ed682008-11-06 22:02:51 -05001172
David Sterbaee863952015-02-16 19:41:40 +01001173 /*
1174 * atomic_sub_return implies a barrier for waitqueue_active
1175 */
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001176 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
Byongho Leeee221842015-12-15 01:42:10 +09001177 5 * SZ_1M &&
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001178 waitqueue_active(&fs_info->async_submit_wait))
1179 wake_up(&fs_info->async_submit_wait);
Chris Mason771ed682008-11-06 22:02:51 -05001180
Chris Masond3977122009-01-05 21:25:51 -05001181 if (async_cow->inode)
Chris Mason771ed682008-11-06 22:02:51 -05001182 submit_compressed_extents(async_cow->inode, async_cow);
Chris Mason771ed682008-11-06 22:02:51 -05001183}
Chris Masonc8b97812008-10-29 14:49:59 -04001184
Chris Mason771ed682008-11-06 22:02:51 -05001185static noinline void async_cow_free(struct btrfs_work *work)
1186{
1187 struct async_cow *async_cow;
1188 async_cow = container_of(work, struct async_cow, work);
Josef Bacik8180ef82012-06-08 15:16:12 -04001189 if (async_cow->inode)
Josef Bacikcb77fcd2012-06-15 12:19:48 -06001190 btrfs_add_delayed_iput(async_cow->inode);
Chris Mason771ed682008-11-06 22:02:51 -05001191 kfree(async_cow);
1192}
1193
1194static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1195 u64 start, u64 end, int *page_started,
Liu Bof82b7352017-10-23 23:18:16 -06001196 unsigned long *nr_written,
1197 unsigned int write_flags)
Chris Mason771ed682008-11-06 22:02:51 -05001198{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001199 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Chris Mason771ed682008-11-06 22:02:51 -05001200 struct async_cow *async_cow;
1201 struct btrfs_root *root = BTRFS_I(inode)->root;
1202 unsigned long nr_pages;
1203 u64 cur_end;
Chris Mason771ed682008-11-06 22:02:51 -05001204
Chris Masona3429ab2009-10-08 12:30:20 -04001205 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1206 1, 0, NULL, GFP_NOFS);
Chris Masond3977122009-01-05 21:25:51 -05001207 while (start < end) {
Chris Mason771ed682008-11-06 22:02:51 -05001208 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001209 BUG_ON(!async_cow); /* -ENOMEM */
Josef Bacik8180ef82012-06-08 15:16:12 -04001210 async_cow->inode = igrab(inode);
Chris Mason771ed682008-11-06 22:02:51 -05001211 async_cow->root = root;
1212 async_cow->locked_page = locked_page;
1213 async_cow->start = start;
Liu Bof82b7352017-10-23 23:18:16 -06001214 async_cow->write_flags = write_flags;
Chris Mason771ed682008-11-06 22:02:51 -05001215
Wang Shilongf79707b2014-07-17 11:44:09 +08001216 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001217 !btrfs_test_opt(fs_info, FORCE_COMPRESS))
Chris Mason771ed682008-11-06 22:02:51 -05001218 cur_end = end;
1219 else
Byongho Leeee221842015-12-15 01:42:10 +09001220 cur_end = min(end, start + SZ_512K - 1);
Chris Mason771ed682008-11-06 22:02:51 -05001221
1222 async_cow->end = cur_end;
1223 INIT_LIST_HEAD(&async_cow->extents);
1224
Liu Bo9e0af232014-08-15 23:36:53 +08001225 btrfs_init_work(&async_cow->work,
1226 btrfs_delalloc_helper,
1227 async_cow_start, async_cow_submit,
1228 async_cow_free);
Chris Mason771ed682008-11-06 22:02:51 -05001229
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001230 nr_pages = (cur_end - start + PAGE_SIZE) >>
1231 PAGE_SHIFT;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001232 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
Chris Mason771ed682008-11-06 22:02:51 -05001233
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001234 btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
Chris Mason771ed682008-11-06 22:02:51 -05001235
Chris Mason771ed682008-11-06 22:02:51 -05001236 *nr_written += nr_pages;
1237 start = cur_end + 1;
1238 }
1239 *page_started = 1;
1240 return 0;
Chris Masonbe20aa92007-12-17 20:14:01 -05001241}
1242
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04001243static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
Yan Zheng17d217f2008-12-12 10:03:38 -05001244 u64 bytenr, u64 num_bytes)
1245{
1246 int ret;
1247 struct btrfs_ordered_sum *sums;
1248 LIST_HEAD(list);
1249
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001250 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
Arne Jansena2de7332011-03-08 14:14:00 +01001251 bytenr + num_bytes - 1, &list, 0);
Yan Zheng17d217f2008-12-12 10:03:38 -05001252 if (ret == 0 && list_empty(&list))
1253 return 0;
1254
1255 while (!list_empty(&list)) {
1256 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1257 list_del(&sums->list);
1258 kfree(sums);
1259 }
1260 return 1;
1261}
1262
Chris Masond352ac62008-09-29 15:18:18 -04001263/*
1264 * when nowcow writeback call back. This checks for snapshots or COW copies
1265 * of the extents that exist in the file, and COWs the file as required.
1266 *
1267 * If no cow copies or snapshots exist, we write directly to the existing
1268 * blocks on disk
1269 */
Chris Mason7f366cf2009-03-12 20:12:45 -04001270static noinline int run_delalloc_nocow(struct inode *inode,
1271 struct page *locked_page,
Chris Mason771ed682008-11-06 22:02:51 -05001272 u64 start, u64 end, int *page_started, int force,
1273 unsigned long *nr_written)
Chris Masonbe20aa92007-12-17 20:14:01 -05001274{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001275 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Chris Masonbe20aa92007-12-17 20:14:01 -05001276 struct btrfs_root *root = BTRFS_I(inode)->root;
1277 struct extent_buffer *leaf;
Chris Masonbe20aa92007-12-17 20:14:01 -05001278 struct btrfs_path *path;
Yan Zheng80ff3852008-10-30 14:20:02 -04001279 struct btrfs_file_extent_item *fi;
Chris Masonbe20aa92007-12-17 20:14:01 -05001280 struct btrfs_key found_key;
Liu Bo6f9994d2017-01-31 07:50:22 -08001281 struct extent_map *em;
Yan Zheng80ff3852008-10-30 14:20:02 -04001282 u64 cow_start;
1283 u64 cur_offset;
1284 u64 extent_end;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001285 u64 extent_offset;
Yan Zheng80ff3852008-10-30 14:20:02 -04001286 u64 disk_bytenr;
1287 u64 num_bytes;
Josef Bacikb4939682012-12-03 10:31:19 -05001288 u64 disk_num_bytes;
Josef Bacikcc95bef2013-04-04 14:31:27 -04001289 u64 ram_bytes;
Yan Zheng80ff3852008-10-30 14:20:02 -04001290 int extent_type;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001291 int ret, err;
Yan Zhengd899e052008-10-30 14:25:28 -04001292 int type;
Yan Zheng80ff3852008-10-30 14:20:02 -04001293 int nocow;
1294 int check_prev = 1;
Li Zefan82d59022011-04-20 10:33:24 +08001295 bool nolock;
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02001296 u64 ino = btrfs_ino(BTRFS_I(inode));
Chris Masonbe20aa92007-12-17 20:14:01 -05001297
1298 path = btrfs_alloc_path();
Josef Bacik17ca04a2012-05-31 15:58:55 -04001299 if (!path) {
Qu Wenruoba8b04c2016-07-19 16:50:36 +08001300 extent_clear_unlock_delalloc(inode, start, end, end,
1301 locked_page,
Josef Bacikc2790a22013-07-29 11:20:47 -04001302 EXTENT_LOCKED | EXTENT_DELALLOC |
Josef Bacik151a41b2013-07-29 13:22:24 -04001303 EXTENT_DO_ACCOUNTING |
1304 EXTENT_DEFRAG, PAGE_UNLOCK |
Josef Bacikc2790a22013-07-29 11:20:47 -04001305 PAGE_CLEAR_DIRTY |
1306 PAGE_SET_WRITEBACK |
1307 PAGE_END_WRITEBACK);
Mark Fashehd8926bb2011-07-13 10:38:47 -07001308 return -ENOMEM;
Josef Bacik17ca04a2012-05-31 15:58:55 -04001309 }
Li Zefan82d59022011-04-20 10:33:24 +08001310
Nikolay Borisov70ddc552017-02-20 13:50:35 +02001311 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
Li Zefan82d59022011-04-20 10:33:24 +08001312
Yan Zheng80ff3852008-10-30 14:20:02 -04001313 cow_start = (u64)-1;
1314 cur_offset = start;
1315 while (1) {
Liu Boe4c3b2d2017-01-30 12:25:28 -08001316 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
Yan Zheng80ff3852008-10-30 14:20:02 -04001317 cur_offset, 0);
Josef Bacikd788a342013-10-25 16:55:08 -04001318 if (ret < 0)
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001319 goto error;
Yan Zheng80ff3852008-10-30 14:20:02 -04001320 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1321 leaf = path->nodes[0];
1322 btrfs_item_key_to_cpu(leaf, &found_key,
1323 path->slots[0] - 1);
Li Zefan33345d012011-04-20 10:31:50 +08001324 if (found_key.objectid == ino &&
Yan Zheng80ff3852008-10-30 14:20:02 -04001325 found_key.type == BTRFS_EXTENT_DATA_KEY)
1326 path->slots[0]--;
1327 }
1328 check_prev = 0;
1329next_slot:
1330 leaf = path->nodes[0];
1331 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1332 ret = btrfs_next_leaf(root, path);
Josef Bacikd788a342013-10-25 16:55:08 -04001333 if (ret < 0)
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001334 goto error;
Yan Zheng80ff3852008-10-30 14:20:02 -04001335 if (ret > 0)
1336 break;
1337 leaf = path->nodes[0];
1338 }
Chris Masonbe20aa92007-12-17 20:14:01 -05001339
Yan Zheng80ff3852008-10-30 14:20:02 -04001340 nocow = 0;
1341 disk_bytenr = 0;
Yan Zheng17d217f2008-12-12 10:03:38 -05001342 num_bytes = 0;
Yan Zheng80ff3852008-10-30 14:20:02 -04001343 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
Chris Masonbe20aa92007-12-17 20:14:01 -05001344
Filipe Manana1d512cb2015-11-09 00:33:58 +00001345 if (found_key.objectid > ino)
1346 break;
1347 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1348 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1349 path->slots[0]++;
1350 goto next_slot;
1351 }
1352 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
Yan Zheng80ff3852008-10-30 14:20:02 -04001353 found_key.offset > end)
1354 break;
Chris Masonbe20aa92007-12-17 20:14:01 -05001355
Yan Zheng80ff3852008-10-30 14:20:02 -04001356 if (found_key.offset > cur_offset) {
1357 extent_end = found_key.offset;
Chris Masone9061e22009-10-09 09:57:45 -04001358 extent_type = 0;
Yan Zheng80ff3852008-10-30 14:20:02 -04001359 goto out_check;
1360 }
Chris Masonc31f8832008-01-08 15:46:31 -05001361
Yan Zheng80ff3852008-10-30 14:20:02 -04001362 fi = btrfs_item_ptr(leaf, path->slots[0],
1363 struct btrfs_file_extent_item);
1364 extent_type = btrfs_file_extent_type(leaf, fi);
Chris Masonbe20aa92007-12-17 20:14:01 -05001365
Josef Bacikcc95bef2013-04-04 14:31:27 -04001366 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
Yan Zhengd899e052008-10-30 14:25:28 -04001367 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1368 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng80ff3852008-10-30 14:20:02 -04001369 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001370 extent_offset = btrfs_file_extent_offset(leaf, fi);
Yan Zheng80ff3852008-10-30 14:20:02 -04001371 extent_end = found_key.offset +
1372 btrfs_file_extent_num_bytes(leaf, fi);
Josef Bacikb4939682012-12-03 10:31:19 -05001373 disk_num_bytes =
1374 btrfs_file_extent_disk_num_bytes(leaf, fi);
Yan Zheng80ff3852008-10-30 14:20:02 -04001375 if (extent_end <= start) {
1376 path->slots[0]++;
1377 goto next_slot;
1378 }
Yan Zheng17d217f2008-12-12 10:03:38 -05001379 if (disk_bytenr == 0)
1380 goto out_check;
Yan Zheng80ff3852008-10-30 14:20:02 -04001381 if (btrfs_file_extent_compression(leaf, fi) ||
1382 btrfs_file_extent_encryption(leaf, fi) ||
1383 btrfs_file_extent_other_encoding(leaf, fi))
1384 goto out_check;
Yan Zhengd899e052008-10-30 14:25:28 -04001385 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1386 goto out_check;
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04001387 if (btrfs_extent_readonly(fs_info, disk_bytenr))
Yan Zheng80ff3852008-10-30 14:20:02 -04001388 goto out_check;
Liu Boe4c3b2d2017-01-30 12:25:28 -08001389 if (btrfs_cross_ref_exist(root, ino,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001390 found_key.offset -
1391 extent_offset, disk_bytenr))
Yan Zheng17d217f2008-12-12 10:03:38 -05001392 goto out_check;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001393 disk_bytenr += extent_offset;
Yan Zheng17d217f2008-12-12 10:03:38 -05001394 disk_bytenr += cur_offset - found_key.offset;
1395 num_bytes = min(end + 1, extent_end) - cur_offset;
1396 /*
Wang Shilonge9894fd2014-03-27 11:12:25 +08001397 * if there are pending snapshots for this root,
1398 * we fall into common COW way.
1399 */
1400 if (!nolock) {
David Sterbaea14b57f2017-06-22 02:19:11 +02001401 err = btrfs_start_write_no_snapshotting(root);
Wang Shilonge9894fd2014-03-27 11:12:25 +08001402 if (!err)
1403 goto out_check;
1404 }
1405 /*
Yan Zheng17d217f2008-12-12 10:03:38 -05001406 * force cow if csum exists in the range.
1407 * this ensure that csum for a given extent are
1408 * either valid or do not exist.
1409 */
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04001410 if (csum_exist_in_range(fs_info, disk_bytenr,
Robbie Ko91e1f562016-10-07 10:01:29 +08001411 num_bytes)) {
1412 if (!nolock)
David Sterbaea14b57f2017-06-22 02:19:11 +02001413 btrfs_end_write_no_snapshotting(root);
Yan Zheng17d217f2008-12-12 10:03:38 -05001414 goto out_check;
Robbie Ko91e1f562016-10-07 10:01:29 +08001415 }
1416 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) {
1417 if (!nolock)
David Sterbaea14b57f2017-06-22 02:19:11 +02001418 btrfs_end_write_no_snapshotting(root);
Filipe Mananaf78c4362016-05-09 13:15:41 +01001419 goto out_check;
Robbie Ko91e1f562016-10-07 10:01:29 +08001420 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001421 nocow = 1;
1422 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1423 extent_end = found_key.offset +
Chris Mason514ac8a2014-01-03 21:07:00 -08001424 btrfs_file_extent_inline_len(leaf,
1425 path->slots[0], fi);
Jeff Mahoneyda170662016-06-15 09:22:56 -04001426 extent_end = ALIGN(extent_end,
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001427 fs_info->sectorsize);
Yan Zheng80ff3852008-10-30 14:20:02 -04001428 } else {
1429 BUG_ON(1);
1430 }
1431out_check:
1432 if (extent_end <= start) {
1433 path->slots[0]++;
Wang Shilonge9894fd2014-03-27 11:12:25 +08001434 if (!nolock && nocow)
David Sterbaea14b57f2017-06-22 02:19:11 +02001435 btrfs_end_write_no_snapshotting(root);
Filipe Mananaf78c4362016-05-09 13:15:41 +01001436 if (nocow)
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001437 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
Yan Zheng80ff3852008-10-30 14:20:02 -04001438 goto next_slot;
1439 }
1440 if (!nocow) {
1441 if (cow_start == (u64)-1)
1442 cow_start = cur_offset;
1443 cur_offset = extent_end;
1444 if (cur_offset > end)
1445 break;
1446 path->slots[0]++;
1447 goto next_slot;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001448 }
Chris Masonbe20aa92007-12-17 20:14:01 -05001449
David Sterbab3b4aa72011-04-21 01:20:15 +02001450 btrfs_release_path(path);
Yan Zheng80ff3852008-10-30 14:20:02 -04001451 if (cow_start != (u64)-1) {
Josef Bacik00361582013-08-14 14:02:47 -04001452 ret = cow_file_range(inode, locked_page,
1453 cow_start, found_key.offset - 1,
Wang Xiaoguangdda32452016-07-11 11:05:29 +08001454 end, page_started, nr_written, 1,
1455 NULL);
Wang Shilonge9894fd2014-03-27 11:12:25 +08001456 if (ret) {
1457 if (!nolock && nocow)
David Sterbaea14b57f2017-06-22 02:19:11 +02001458 btrfs_end_write_no_snapshotting(root);
Filipe Mananaf78c4362016-05-09 13:15:41 +01001459 if (nocow)
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001460 btrfs_dec_nocow_writers(fs_info,
Filipe Mananaf78c4362016-05-09 13:15:41 +01001461 disk_bytenr);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001462 goto error;
Wang Shilonge9894fd2014-03-27 11:12:25 +08001463 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001464 cow_start = (u64)-1;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001465 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001466
Yan Zhengd899e052008-10-30 14:25:28 -04001467 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
Liu Bo6f9994d2017-01-31 07:50:22 -08001468 u64 orig_start = found_key.offset - extent_offset;
1469
1470 em = create_io_em(inode, cur_offset, num_bytes,
1471 orig_start,
1472 disk_bytenr, /* block_start */
1473 num_bytes, /* block_len */
1474 disk_num_bytes, /* orig_block_len */
1475 ram_bytes, BTRFS_COMPRESS_NONE,
1476 BTRFS_ORDERED_PREALLOC);
1477 if (IS_ERR(em)) {
1478 if (!nolock && nocow)
David Sterbaea14b57f2017-06-22 02:19:11 +02001479 btrfs_end_write_no_snapshotting(root);
Liu Bo6f9994d2017-01-31 07:50:22 -08001480 if (nocow)
1481 btrfs_dec_nocow_writers(fs_info,
1482 disk_bytenr);
1483 ret = PTR_ERR(em);
1484 goto error;
Yan Zhengd899e052008-10-30 14:25:28 -04001485 }
Liu Bo6f9994d2017-01-31 07:50:22 -08001486 free_extent_map(em);
1487 }
1488
1489 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zhengd899e052008-10-30 14:25:28 -04001490 type = BTRFS_ORDERED_PREALLOC;
1491 } else {
1492 type = BTRFS_ORDERED_NOCOW;
1493 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001494
1495 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
Yan Zhengd899e052008-10-30 14:25:28 -04001496 num_bytes, num_bytes, type);
Filipe Mananaf78c4362016-05-09 13:15:41 +01001497 if (nocow)
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001498 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001499 BUG_ON(ret); /* -ENOMEM */
Chris Mason771ed682008-11-06 22:02:51 -05001500
Yan, Zhengefa56462010-05-16 10:49:59 -04001501 if (root->root_key.objectid ==
Qu Wenruo4dbd80f2017-03-08 10:25:51 +08001502 BTRFS_DATA_RELOC_TREE_OBJECTID)
1503 /*
1504 * Error handled later, as we must prevent
1505 * extent_clear_unlock_delalloc() in error handler
1506 * from freeing metadata of created ordered extent.
1507 */
Yan, Zhengefa56462010-05-16 10:49:59 -04001508 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1509 num_bytes);
Yan, Zhengefa56462010-05-16 10:49:59 -04001510
Josef Bacikc2790a22013-07-29 11:20:47 -04001511 extent_clear_unlock_delalloc(inode, cur_offset,
Qu Wenruoba8b04c2016-07-19 16:50:36 +08001512 cur_offset + num_bytes - 1, end,
Josef Bacikc2790a22013-07-29 11:20:47 -04001513 locked_page, EXTENT_LOCKED |
Wang Xiaoguang18513092016-07-25 15:51:40 +08001514 EXTENT_DELALLOC |
1515 EXTENT_CLEAR_DATA_RESV,
1516 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1517
Wang Shilonge9894fd2014-03-27 11:12:25 +08001518 if (!nolock && nocow)
David Sterbaea14b57f2017-06-22 02:19:11 +02001519 btrfs_end_write_no_snapshotting(root);
Yan Zheng80ff3852008-10-30 14:20:02 -04001520 cur_offset = extent_end;
Qu Wenruo4dbd80f2017-03-08 10:25:51 +08001521
1522 /*
1523 * btrfs_reloc_clone_csums() error, now we're OK to call error
1524 * handler, as metadata for created ordered extent will only
1525 * be freed by btrfs_finish_ordered_io().
1526 */
1527 if (ret)
1528 goto error;
Yan Zheng80ff3852008-10-30 14:20:02 -04001529 if (cur_offset > end)
1530 break;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001531 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001532 btrfs_release_path(path);
Yan Zheng80ff3852008-10-30 14:20:02 -04001533
Josef Bacik17ca04a2012-05-31 15:58:55 -04001534 if (cur_offset <= end && cow_start == (u64)-1) {
Yan Zheng80ff3852008-10-30 14:20:02 -04001535 cow_start = cur_offset;
Josef Bacik17ca04a2012-05-31 15:58:55 -04001536 cur_offset = end;
1537 }
1538
Yan Zheng80ff3852008-10-30 14:20:02 -04001539 if (cow_start != (u64)-1) {
Wang Xiaoguangdda32452016-07-11 11:05:29 +08001540 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1541 page_started, nr_written, 1, NULL);
Josef Bacikd788a342013-10-25 16:55:08 -04001542 if (ret)
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001543 goto error;
Yan Zheng80ff3852008-10-30 14:20:02 -04001544 }
1545
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001546error:
Josef Bacik17ca04a2012-05-31 15:58:55 -04001547 if (ret && cur_offset < end)
Qu Wenruoba8b04c2016-07-19 16:50:36 +08001548 extent_clear_unlock_delalloc(inode, cur_offset, end, end,
Josef Bacikc2790a22013-07-29 11:20:47 -04001549 locked_page, EXTENT_LOCKED |
Josef Bacik151a41b2013-07-29 13:22:24 -04001550 EXTENT_DELALLOC | EXTENT_DEFRAG |
1551 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1552 PAGE_CLEAR_DIRTY |
Josef Bacikc2790a22013-07-29 11:20:47 -04001553 PAGE_SET_WRITEBACK |
1554 PAGE_END_WRITEBACK);
Yan Zheng7ea394f2008-08-05 13:05:02 -04001555 btrfs_free_path(path);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001556 return ret;
Chris Masonbe20aa92007-12-17 20:14:01 -05001557}
1558
Wang Shilong47059d92014-07-03 18:22:07 +08001559static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1560{
1561
1562 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1563 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1564 return 0;
1565
1566 /*
1567 * @defrag_bytes is a hint value, no spinlock held here,
1568 * if is not zero, it means the file is defragging.
1569 * Force cow if given extent needs to be defragged.
1570 */
1571 if (BTRFS_I(inode)->defrag_bytes &&
1572 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1573 EXTENT_DEFRAG, 0, NULL))
1574 return 1;
1575
1576 return 0;
1577}
1578
Chris Masond352ac62008-09-29 15:18:18 -04001579/*
1580 * extent_io.c call back to do delayed allocation processing
1581 */
Josef Bacikc6100a42017-05-05 11:57:13 -04001582static int run_delalloc_range(void *private_data, struct page *locked_page,
Chris Mason771ed682008-11-06 22:02:51 -05001583 u64 start, u64 end, int *page_started,
Liu Bof82b7352017-10-23 23:18:16 -06001584 unsigned long *nr_written,
1585 struct writeback_control *wbc)
Chris Masonbe20aa92007-12-17 20:14:01 -05001586{
Josef Bacikc6100a42017-05-05 11:57:13 -04001587 struct inode *inode = private_data;
Chris Masonbe20aa92007-12-17 20:14:01 -05001588 int ret;
Wang Shilong47059d92014-07-03 18:22:07 +08001589 int force_cow = need_force_cow(inode, start, end);
Liu Bof82b7352017-10-23 23:18:16 -06001590 unsigned int write_flags = wbc_to_write_flags(wbc);
Chris Masona2135012008-06-25 16:01:30 -04001591
Wang Shilong47059d92014-07-03 18:22:07 +08001592 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
Chris Masonc8b97812008-10-29 14:49:59 -04001593 ret = run_delalloc_nocow(inode, locked_page, start, end,
Chris Masond3977122009-01-05 21:25:51 -05001594 page_started, 1, nr_written);
Wang Shilong47059d92014-07-03 18:22:07 +08001595 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
Yan Zhengd899e052008-10-30 14:25:28 -04001596 ret = run_delalloc_nocow(inode, locked_page, start, end,
Chris Masond3977122009-01-05 21:25:51 -05001597 page_started, 0, nr_written);
Timofey Titovetsc2fcdcd2017-07-17 16:52:58 +03001598 } else if (!inode_need_compress(inode, start, end)) {
Wang Xiaoguangdda32452016-07-11 11:05:29 +08001599 ret = cow_file_range(inode, locked_page, start, end, end,
1600 page_started, nr_written, 1, NULL);
Josef Bacik7ddf5a42012-06-08 15:26:47 -04001601 } else {
1602 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1603 &BTRFS_I(inode)->runtime_flags);
Chris Mason771ed682008-11-06 22:02:51 -05001604 ret = cow_file_range_async(inode, locked_page, start, end,
Liu Bof82b7352017-10-23 23:18:16 -06001605 page_started, nr_written,
1606 write_flags);
Josef Bacik7ddf5a42012-06-08 15:26:47 -04001607 }
Qu Wenruo524272602017-03-08 10:25:52 +08001608 if (ret)
1609 btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
Chris Masonb888db22007-08-27 16:49:44 -04001610 return ret;
1611}
1612
Josef Bacikc6100a42017-05-05 11:57:13 -04001613static void btrfs_split_extent_hook(void *private_data,
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001614 struct extent_state *orig, u64 split)
Josef Bacik9ed74f22009-09-11 16:12:44 -04001615{
Josef Bacikc6100a42017-05-05 11:57:13 -04001616 struct inode *inode = private_data;
Josef Bacikdcab6a32015-02-11 15:08:59 -05001617 u64 size;
1618
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001619 /* not delalloc, ignore it */
Josef Bacik9ed74f22009-09-11 16:12:44 -04001620 if (!(orig->state & EXTENT_DELALLOC))
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001621 return;
Josef Bacik9ed74f22009-09-11 16:12:44 -04001622
Josef Bacikdcab6a32015-02-11 15:08:59 -05001623 size = orig->end - orig->start + 1;
1624 if (size > BTRFS_MAX_EXTENT_SIZE) {
David Sterba823bb202017-01-04 11:09:51 +01001625 u32 num_extents;
Josef Bacikdcab6a32015-02-11 15:08:59 -05001626 u64 new_size;
1627
1628 /*
Josef Bacikba117212015-03-13 15:01:24 -04001629 * See the explanation in btrfs_merge_extent_hook, the same
1630 * applies here, just in reverse.
Josef Bacikdcab6a32015-02-11 15:08:59 -05001631 */
1632 new_size = orig->end - split + 1;
David Sterba823bb202017-01-04 11:09:51 +01001633 num_extents = count_max_extents(new_size);
Josef Bacikba117212015-03-13 15:01:24 -04001634 new_size = split - orig->start;
David Sterba823bb202017-01-04 11:09:51 +01001635 num_extents += count_max_extents(new_size);
1636 if (count_max_extents(size) >= num_extents)
Josef Bacikdcab6a32015-02-11 15:08:59 -05001637 return;
1638 }
1639
Josef Bacik9e0baf62011-07-15 15:16:44 +00001640 spin_lock(&BTRFS_I(inode)->lock);
Josef Bacik8b62f872017-10-19 14:15:55 -04001641 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
Josef Bacik9e0baf62011-07-15 15:16:44 +00001642 spin_unlock(&BTRFS_I(inode)->lock);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001643}
1644
1645/*
1646 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1647 * extents so we can keep track of new extents that are just merged onto old
1648 * extents, such as when we are doing sequential writes, so we can properly
1649 * account for the metadata space we'll need.
1650 */
Josef Bacikc6100a42017-05-05 11:57:13 -04001651static void btrfs_merge_extent_hook(void *private_data,
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001652 struct extent_state *new,
1653 struct extent_state *other)
Josef Bacik9ed74f22009-09-11 16:12:44 -04001654{
Josef Bacikc6100a42017-05-05 11:57:13 -04001655 struct inode *inode = private_data;
Josef Bacikdcab6a32015-02-11 15:08:59 -05001656 u64 new_size, old_size;
David Sterba823bb202017-01-04 11:09:51 +01001657 u32 num_extents;
Josef Bacikdcab6a32015-02-11 15:08:59 -05001658
Josef Bacik9ed74f22009-09-11 16:12:44 -04001659 /* not delalloc, ignore it */
1660 if (!(other->state & EXTENT_DELALLOC))
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001661 return;
Josef Bacik9ed74f22009-09-11 16:12:44 -04001662
Josef Bacik8461a3d2015-03-13 15:12:08 -04001663 if (new->start > other->start)
1664 new_size = new->end - other->start + 1;
1665 else
1666 new_size = other->end - new->start + 1;
Josef Bacikdcab6a32015-02-11 15:08:59 -05001667
1668 /* we're not bigger than the max, unreserve the space and go */
1669 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1670 spin_lock(&BTRFS_I(inode)->lock);
Josef Bacik8b62f872017-10-19 14:15:55 -04001671 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
Josef Bacikdcab6a32015-02-11 15:08:59 -05001672 spin_unlock(&BTRFS_I(inode)->lock);
1673 return;
1674 }
1675
1676 /*
Josef Bacikba117212015-03-13 15:01:24 -04001677 * We have to add up either side to figure out how many extents were
1678 * accounted for before we merged into one big extent. If the number of
1679 * extents we accounted for is <= the amount we need for the new range
1680 * then we can return, otherwise drop. Think of it like this
1681 *
1682 * [ 4k][MAX_SIZE]
1683 *
1684 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1685 * need 2 outstanding extents, on one side we have 1 and the other side
1686 * we have 1 so they are == and we can return. But in this case
1687 *
1688 * [MAX_SIZE+4k][MAX_SIZE+4k]
1689 *
1690 * Each range on their own accounts for 2 extents, but merged together
1691 * they are only 3 extents worth of accounting, so we need to drop in
1692 * this case.
Josef Bacikdcab6a32015-02-11 15:08:59 -05001693 */
Josef Bacikba117212015-03-13 15:01:24 -04001694 old_size = other->end - other->start + 1;
David Sterba823bb202017-01-04 11:09:51 +01001695 num_extents = count_max_extents(old_size);
Josef Bacikba117212015-03-13 15:01:24 -04001696 old_size = new->end - new->start + 1;
David Sterba823bb202017-01-04 11:09:51 +01001697 num_extents += count_max_extents(old_size);
1698 if (count_max_extents(new_size) >= num_extents)
Josef Bacikdcab6a32015-02-11 15:08:59 -05001699 return;
1700
Josef Bacik9e0baf62011-07-15 15:16:44 +00001701 spin_lock(&BTRFS_I(inode)->lock);
Josef Bacik8b62f872017-10-19 14:15:55 -04001702 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
Josef Bacik9e0baf62011-07-15 15:16:44 +00001703 spin_unlock(&BTRFS_I(inode)->lock);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001704}
1705
Miao Xieeb73c1b2013-05-15 07:48:22 +00001706static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1707 struct inode *inode)
1708{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001709 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1710
Miao Xieeb73c1b2013-05-15 07:48:22 +00001711 spin_lock(&root->delalloc_lock);
1712 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1713 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1714 &root->delalloc_inodes);
1715 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1716 &BTRFS_I(inode)->runtime_flags);
1717 root->nr_delalloc_inodes++;
1718 if (root->nr_delalloc_inodes == 1) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001719 spin_lock(&fs_info->delalloc_root_lock);
Miao Xieeb73c1b2013-05-15 07:48:22 +00001720 BUG_ON(!list_empty(&root->delalloc_root));
1721 list_add_tail(&root->delalloc_root,
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001722 &fs_info->delalloc_roots);
1723 spin_unlock(&fs_info->delalloc_root_lock);
Miao Xieeb73c1b2013-05-15 07:48:22 +00001724 }
1725 }
1726 spin_unlock(&root->delalloc_lock);
1727}
1728
1729static void btrfs_del_delalloc_inode(struct btrfs_root *root,
Nikolay Borisov9e3e97f2017-02-20 13:51:07 +02001730 struct btrfs_inode *inode)
Miao Xieeb73c1b2013-05-15 07:48:22 +00001731{
Nikolay Borisov9e3e97f2017-02-20 13:51:07 +02001732 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001733
Miao Xieeb73c1b2013-05-15 07:48:22 +00001734 spin_lock(&root->delalloc_lock);
Nikolay Borisov9e3e97f2017-02-20 13:51:07 +02001735 if (!list_empty(&inode->delalloc_inodes)) {
1736 list_del_init(&inode->delalloc_inodes);
Miao Xieeb73c1b2013-05-15 07:48:22 +00001737 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
Nikolay Borisov9e3e97f2017-02-20 13:51:07 +02001738 &inode->runtime_flags);
Miao Xieeb73c1b2013-05-15 07:48:22 +00001739 root->nr_delalloc_inodes--;
1740 if (!root->nr_delalloc_inodes) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001741 spin_lock(&fs_info->delalloc_root_lock);
Miao Xieeb73c1b2013-05-15 07:48:22 +00001742 BUG_ON(list_empty(&root->delalloc_root));
1743 list_del_init(&root->delalloc_root);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001744 spin_unlock(&fs_info->delalloc_root_lock);
Miao Xieeb73c1b2013-05-15 07:48:22 +00001745 }
1746 }
1747 spin_unlock(&root->delalloc_lock);
1748}
1749
Chris Masond352ac62008-09-29 15:18:18 -04001750/*
1751 * extent_io.c set_bit_hook, used to track delayed allocation
1752 * bytes in this file, and to maintain the list of inodes that
1753 * have pending delalloc work to be done.
1754 */
Josef Bacikc6100a42017-05-05 11:57:13 -04001755static void btrfs_set_bit_hook(void *private_data,
David Sterba9ee49a042015-01-14 19:52:13 +01001756 struct extent_state *state, unsigned *bits)
Chris Mason291d6732008-01-29 15:55:23 -05001757{
Josef Bacikc6100a42017-05-05 11:57:13 -04001758 struct inode *inode = private_data;
Josef Bacik9ed74f22009-09-11 16:12:44 -04001759
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001760 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1761
Wang Shilong47059d92014-07-03 18:22:07 +08001762 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1763 WARN_ON(1);
Chris Mason75eff682008-12-15 15:54:40 -05001764 /*
1765 * set_bit and clear bit hooks normally require _irqsave/restore
Sergei Trofimovich27160b62011-05-20 20:20:32 +00001766 * but in this case, we are only testing for the DELALLOC
Chris Mason75eff682008-12-15 15:54:40 -05001767 * bit, which is only set or cleared with irqs on
1768 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001769 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
Chris Mason291d6732008-01-29 15:55:23 -05001770 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001771 u64 len = state->end + 1 - state->start;
Josef Bacik8b62f872017-10-19 14:15:55 -04001772 u32 num_extents = count_max_extents(len);
Nikolay Borisov70ddc552017-02-20 13:50:35 +02001773 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
Josef Bacik9ed74f22009-09-11 16:12:44 -04001774
Josef Bacik8b62f872017-10-19 14:15:55 -04001775 spin_lock(&BTRFS_I(inode)->lock);
1776 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
1777 spin_unlock(&BTRFS_I(inode)->lock);
Josef Bacik287a0ab2010-03-19 18:07:23 +00001778
Josef Bacik6a3891c2015-03-16 17:38:52 -04001779 /* For sanity tests */
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001780 if (btrfs_is_testing(fs_info))
Josef Bacik6a3891c2015-03-16 17:38:52 -04001781 return;
1782
Nikolay Borisov104b4e52017-06-20 21:01:20 +03001783 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
1784 fs_info->delalloc_batch);
Miao Xiedf0af1a2013-01-29 10:11:59 +00001785 spin_lock(&BTRFS_I(inode)->lock);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001786 BTRFS_I(inode)->delalloc_bytes += len;
Wang Shilong47059d92014-07-03 18:22:07 +08001787 if (*bits & EXTENT_DEFRAG)
1788 BTRFS_I(inode)->defrag_bytes += len;
Miao Xiedf0af1a2013-01-29 10:11:59 +00001789 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
Miao Xieeb73c1b2013-05-15 07:48:22 +00001790 &BTRFS_I(inode)->runtime_flags))
1791 btrfs_add_delalloc_inodes(root, inode);
Miao Xiedf0af1a2013-01-29 10:11:59 +00001792 spin_unlock(&BTRFS_I(inode)->lock);
Chris Mason291d6732008-01-29 15:55:23 -05001793 }
Filipe Mananaa7e3b972017-04-03 10:45:46 +01001794
1795 if (!(state->state & EXTENT_DELALLOC_NEW) &&
1796 (*bits & EXTENT_DELALLOC_NEW)) {
1797 spin_lock(&BTRFS_I(inode)->lock);
1798 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
1799 state->start;
1800 spin_unlock(&BTRFS_I(inode)->lock);
1801 }
Chris Mason291d6732008-01-29 15:55:23 -05001802}
1803
Chris Masond352ac62008-09-29 15:18:18 -04001804/*
1805 * extent_io.c clear_bit_hook, see set_bit_hook for why
1806 */
Josef Bacikc6100a42017-05-05 11:57:13 -04001807static void btrfs_clear_bit_hook(void *private_data,
David Sterba41074882013-04-29 13:38:46 +00001808 struct extent_state *state,
David Sterba9ee49a042015-01-14 19:52:13 +01001809 unsigned *bits)
Chris Mason291d6732008-01-29 15:55:23 -05001810{
Josef Bacikc6100a42017-05-05 11:57:13 -04001811 struct btrfs_inode *inode = BTRFS_I((struct inode *)private_data);
Nikolay Borisov6fc0ef62017-02-20 13:51:03 +02001812 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
Wang Shilong47059d92014-07-03 18:22:07 +08001813 u64 len = state->end + 1 - state->start;
David Sterba823bb202017-01-04 11:09:51 +01001814 u32 num_extents = count_max_extents(len);
Wang Shilong47059d92014-07-03 18:22:07 +08001815
Filipe Manana4a4b9642017-07-27 19:52:55 +01001816 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
1817 spin_lock(&inode->lock);
Nikolay Borisov6fc0ef62017-02-20 13:51:03 +02001818 inode->defrag_bytes -= len;
Filipe Manana4a4b9642017-07-27 19:52:55 +01001819 spin_unlock(&inode->lock);
1820 }
Wang Shilong47059d92014-07-03 18:22:07 +08001821
Chris Mason75eff682008-12-15 15:54:40 -05001822 /*
1823 * set_bit and clear bit hooks normally require _irqsave/restore
Sergei Trofimovich27160b62011-05-20 20:20:32 +00001824 * but in this case, we are only testing for the DELALLOC
Chris Mason75eff682008-12-15 15:54:40 -05001825 * bit, which is only set or cleared with irqs on
1826 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001827 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
Nikolay Borisov6fc0ef62017-02-20 13:51:03 +02001828 struct btrfs_root *root = inode->root;
Liu Bo83eea1f2012-07-10 05:28:39 -06001829 bool do_list = !btrfs_is_free_space_inode(inode);
Chris Masonbcbfce82008-04-22 13:26:47 -04001830
Josef Bacik8b62f872017-10-19 14:15:55 -04001831 spin_lock(&inode->lock);
1832 btrfs_mod_outstanding_extents(inode, -num_extents);
1833 spin_unlock(&inode->lock);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001834
Josef Bacikb6d08f02013-09-27 14:57:43 -04001835 /*
1836 * We don't reserve metadata space for space cache inodes so we
1837 * don't need to call dellalloc_release_metadata if there is an
1838 * error.
1839 */
Filipe Mananaa315e682017-03-06 23:04:20 +00001840 if (*bits & EXTENT_CLEAR_META_RESV &&
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001841 root != fs_info->tree_root)
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001842 btrfs_delalloc_release_metadata(inode, len);
1843
Josef Bacik6a3891c2015-03-16 17:38:52 -04001844 /* For sanity tests. */
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001845 if (btrfs_is_testing(fs_info))
Josef Bacik6a3891c2015-03-16 17:38:52 -04001846 return;
1847
Filipe Mananaa315e682017-03-06 23:04:20 +00001848 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
1849 do_list && !(state->state & EXTENT_NORESERVE) &&
1850 (*bits & EXTENT_CLEAR_DATA_RESV))
Nikolay Borisov6fc0ef62017-02-20 13:51:03 +02001851 btrfs_free_reserved_data_space_noquota(
1852 &inode->vfs_inode,
Qu Wenruo51773be2015-10-08 18:19:37 +08001853 state->start, len);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001854
Nikolay Borisov104b4e52017-06-20 21:01:20 +03001855 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
1856 fs_info->delalloc_batch);
Nikolay Borisov6fc0ef62017-02-20 13:51:03 +02001857 spin_lock(&inode->lock);
1858 inode->delalloc_bytes -= len;
1859 if (do_list && inode->delalloc_bytes == 0 &&
Miao Xiedf0af1a2013-01-29 10:11:59 +00001860 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
Nikolay Borisov9e3e97f2017-02-20 13:51:07 +02001861 &inode->runtime_flags))
Miao Xieeb73c1b2013-05-15 07:48:22 +00001862 btrfs_del_delalloc_inode(root, inode);
Nikolay Borisov6fc0ef62017-02-20 13:51:03 +02001863 spin_unlock(&inode->lock);
Chris Mason291d6732008-01-29 15:55:23 -05001864 }
Filipe Mananaa7e3b972017-04-03 10:45:46 +01001865
1866 if ((state->state & EXTENT_DELALLOC_NEW) &&
1867 (*bits & EXTENT_DELALLOC_NEW)) {
1868 spin_lock(&inode->lock);
1869 ASSERT(inode->new_delalloc_bytes >= len);
1870 inode->new_delalloc_bytes -= len;
1871 spin_unlock(&inode->lock);
1872 }
Chris Mason291d6732008-01-29 15:55:23 -05001873}
1874
Chris Masond352ac62008-09-29 15:18:18 -04001875/*
1876 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1877 * we don't create bios that span stripes or chunks
Liu Bo6f034ec2016-06-22 18:31:49 -07001878 *
1879 * return 1 if page cannot be merged to bio
1880 * return 0 if page can be merged to bio
1881 * return error otherwise
Chris Masond352ac62008-09-29 15:18:18 -04001882 */
Mike Christie81a75f672016-06-05 14:31:54 -05001883int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
Chris Masonc8b97812008-10-29 14:49:59 -04001884 size_t size, struct bio *bio,
1885 unsigned long bio_flags)
Chris Mason239b14b2008-03-24 15:02:07 -04001886{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001887 struct inode *inode = page->mapping->host;
1888 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Kent Overstreet4f024f32013-10-11 15:44:27 -07001889 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
Chris Mason239b14b2008-03-24 15:02:07 -04001890 u64 length = 0;
1891 u64 map_length;
Chris Mason239b14b2008-03-24 15:02:07 -04001892 int ret;
1893
Chris Mason771ed682008-11-06 22:02:51 -05001894 if (bio_flags & EXTENT_BIO_COMPRESSED)
1895 return 0;
1896
Kent Overstreet4f024f32013-10-11 15:44:27 -07001897 length = bio->bi_iter.bi_size;
Chris Mason239b14b2008-03-24 15:02:07 -04001898 map_length = length;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001899 ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1900 NULL, 0);
Liu Bo6f034ec2016-06-22 18:31:49 -07001901 if (ret < 0)
1902 return ret;
Chris Masond3977122009-01-05 21:25:51 -05001903 if (map_length < length + size)
Chris Mason239b14b2008-03-24 15:02:07 -04001904 return 1;
Jeff Mahoney3444a972011-10-03 23:23:13 -04001905 return 0;
Chris Mason239b14b2008-03-24 15:02:07 -04001906}
1907
Chris Masond352ac62008-09-29 15:18:18 -04001908/*
1909 * in order to insert checksums into the metadata in large chunks,
1910 * we wait until bio submission time. All the pages in the bio are
1911 * checksummed and sums are attached onto the ordered extent record.
1912 *
1913 * At IO completion time the cums attached on the ordered extent record
1914 * are inserted into the btree
1915 */
Linus Torvalds8c27cb32017-07-05 16:41:23 -07001916static blk_status_t __btrfs_submit_bio_start(void *private_data, struct bio *bio,
Mike Christie81a75f672016-06-05 14:31:54 -05001917 int mirror_num, unsigned long bio_flags,
Chris Masoneaf25d92010-05-25 09:48:28 -04001918 u64 bio_offset)
Chris Mason065631f2008-02-20 12:07:25 -05001919{
Josef Bacikc6100a42017-05-05 11:57:13 -04001920 struct inode *inode = private_data;
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02001921 blk_status_t ret = 0;
Chris Masone0156402008-04-16 11:15:20 -04001922
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04001923 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001924 BUG_ON(ret); /* -ENOMEM */
Chris Mason4a69a412008-11-06 22:03:00 -05001925 return 0;
1926}
Chris Masone0156402008-04-16 11:15:20 -04001927
Chris Mason4a69a412008-11-06 22:03:00 -05001928/*
1929 * in order to insert checksums into the metadata in large chunks,
1930 * we wait until bio submission time. All the pages in the bio are
1931 * checksummed and sums are attached onto the ordered extent record.
1932 *
1933 * At IO completion time the cums attached on the ordered extent record
1934 * are inserted into the btree
1935 */
Linus Torvalds8c27cb32017-07-05 16:41:23 -07001936static blk_status_t __btrfs_submit_bio_done(void *private_data, struct bio *bio,
Chris Masoneaf25d92010-05-25 09:48:28 -04001937 int mirror_num, unsigned long bio_flags,
1938 u64 bio_offset)
Chris Mason4a69a412008-11-06 22:03:00 -05001939{
Josef Bacikc6100a42017-05-05 11:57:13 -04001940 struct inode *inode = private_data;
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04001941 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02001942 blk_status_t ret;
Stefan Behrens61891922012-11-05 18:51:52 +01001943
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04001944 ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001945 if (ret) {
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02001946 bio->bi_status = ret;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001947 bio_endio(bio);
1948 }
Stefan Behrens61891922012-11-05 18:51:52 +01001949 return ret;
Chris Mason44b8bd72008-04-16 11:14:51 -04001950}
1951
Chris Masond352ac62008-09-29 15:18:18 -04001952/*
Chris Masoncad321a2008-12-17 14:51:42 -05001953 * extent_io.c submission hook. This does the right thing for csum calculation
1954 * on write, or reading the csums from the tree before a read
Chris Masond352ac62008-09-29 15:18:18 -04001955 */
Linus Torvalds8c27cb32017-07-05 16:41:23 -07001956static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio,
Josef Bacikc6100a42017-05-05 11:57:13 -04001957 int mirror_num, unsigned long bio_flags,
1958 u64 bio_offset)
Chris Mason44b8bd72008-04-16 11:14:51 -04001959{
Josef Bacikc6100a42017-05-05 11:57:13 -04001960 struct inode *inode = private_data;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001961 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Chris Mason44b8bd72008-04-16 11:14:51 -04001962 struct btrfs_root *root = BTRFS_I(inode)->root;
Chandan Rajendra0d51e282015-07-27 15:26:43 +05301963 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02001964 blk_status_t ret = 0;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001965 int skip_sum;
Josef Bacikb812ce22012-11-16 13:56:32 -05001966 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
Chris Mason44b8bd72008-04-16 11:14:51 -04001967
Christoph Hellwig6cbff002009-04-17 10:37:41 +02001968 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
Chris Masoncad321a2008-12-17 14:51:42 -05001969
Nikolay Borisov70ddc552017-02-20 13:50:35 +02001970 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
Chandan Rajendra0d51e282015-07-27 15:26:43 +05301971 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
Jeff Mahoney04173412011-10-03 23:23:12 -04001972
Mike Christie37226b22016-06-05 14:31:52 -05001973 if (bio_op(bio) != REQ_OP_WRITE) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001974 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
Josef Bacik5fd02042012-05-02 14:00:54 -04001975 if (ret)
Stefan Behrens61891922012-11-05 18:51:52 +01001976 goto out;
Josef Bacik5fd02042012-05-02 14:00:54 -04001977
Chris Masond20f7042008-12-08 16:58:54 -05001978 if (bio_flags & EXTENT_BIO_COMPRESSED) {
Stefan Behrens61891922012-11-05 18:51:52 +01001979 ret = btrfs_submit_compressed_read(inode, bio,
1980 mirror_num,
1981 bio_flags);
1982 goto out;
Tsutomu Itohc2db1072011-03-01 06:48:31 +00001983 } else if (!skip_sum) {
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04001984 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
Tsutomu Itohc2db1072011-03-01 06:48:31 +00001985 if (ret)
Stefan Behrens61891922012-11-05 18:51:52 +01001986 goto out;
Tsutomu Itohc2db1072011-03-01 06:48:31 +00001987 }
Chris Mason4d1b5fb2008-08-20 09:44:52 -04001988 goto mapit;
Josef Bacikb812ce22012-11-16 13:56:32 -05001989 } else if (async && !skip_sum) {
Yan Zheng17d217f2008-12-12 10:03:38 -05001990 /* csum items have already been cloned */
1991 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1992 goto mapit;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001993 /* we're doing a write, do the async checksumming */
Josef Bacikc6100a42017-05-05 11:57:13 -04001994 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
1995 bio_offset, inode,
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001996 __btrfs_submit_bio_start,
1997 __btrfs_submit_bio_done);
Stefan Behrens61891922012-11-05 18:51:52 +01001998 goto out;
Josef Bacikb812ce22012-11-16 13:56:32 -05001999 } else if (!skip_sum) {
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002000 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
Josef Bacikb812ce22012-11-16 13:56:32 -05002001 if (ret)
2002 goto out;
Chris Mason19b9bdb2008-10-30 14:23:13 -04002003 }
2004
Chris Mason0b86a832008-03-24 15:01:56 -04002005mapit:
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002006 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
Stefan Behrens61891922012-11-05 18:51:52 +01002007
2008out:
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02002009 if (ret) {
2010 bio->bi_status = ret;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002011 bio_endio(bio);
2012 }
Stefan Behrens61891922012-11-05 18:51:52 +01002013 return ret;
Chris Mason065631f2008-02-20 12:07:25 -05002014}
Chris Mason6885f302008-02-20 16:11:05 -05002015
Chris Masond352ac62008-09-29 15:18:18 -04002016/*
2017 * given a list of ordered sums record them in the inode. This happens
2018 * at IO completion time based on sums calculated at bio submission time.
2019 */
Chris Masonba1da2f2008-07-17 12:54:15 -04002020static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
David Sterbadf9f6282017-02-10 19:35:37 +01002021 struct inode *inode, struct list_head *list)
Chris Masone6dcd2d2008-07-17 12:53:50 -04002022{
Chris Masone6dcd2d2008-07-17 12:53:50 -04002023 struct btrfs_ordered_sum *sum;
2024
Qinghuang Fengc6e30872009-01-21 10:59:08 -05002025 list_for_each_entry(sum, list, list) {
Miao Xie39847c42013-03-28 08:08:20 +00002026 trans->adding_csums = 1;
Chris Masond20f7042008-12-08 16:58:54 -05002027 btrfs_csum_file_blocks(trans,
2028 BTRFS_I(inode)->root->fs_info->csum_root, sum);
Miao Xie39847c42013-03-28 08:08:20 +00002029 trans->adding_csums = 0;
Chris Masone6dcd2d2008-07-17 12:53:50 -04002030 }
2031 return 0;
2032}
2033
Josef Bacik2ac55d42010-02-03 19:33:23 +00002034int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
Filipe Mananae3b8a482017-11-04 00:16:59 +00002035 unsigned int extra_bits,
Qu Wenruoba8b04c2016-07-19 16:50:36 +08002036 struct extent_state **cached_state, int dedupe)
Chris Masonea8c2812008-08-04 23:17:27 -04002037{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002038 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
Chris Masonea8c2812008-08-04 23:17:27 -04002039 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
Filipe Mananae3b8a482017-11-04 00:16:59 +00002040 extra_bits, cached_state);
Chris Masonea8c2812008-08-04 23:17:27 -04002041}
2042
Chris Masond352ac62008-09-29 15:18:18 -04002043/* see btrfs_writepage_start_hook for details on why this is required */
Chris Mason247e7432008-07-17 12:53:51 -04002044struct btrfs_writepage_fixup {
2045 struct page *page;
2046 struct btrfs_work work;
2047};
2048
Christoph Hellwigb2950862008-12-02 09:54:17 -05002049static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
Chris Mason247e7432008-07-17 12:53:51 -04002050{
2051 struct btrfs_writepage_fixup *fixup;
2052 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00002053 struct extent_state *cached_state = NULL;
Qu Wenruo364ecf32017-02-27 15:10:38 +08002054 struct extent_changeset *data_reserved = NULL;
Chris Mason247e7432008-07-17 12:53:51 -04002055 struct page *page;
2056 struct inode *inode;
2057 u64 page_start;
2058 u64 page_end;
Jeff Mahoney87826df2012-02-15 16:23:57 +01002059 int ret;
Chris Mason247e7432008-07-17 12:53:51 -04002060
2061 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2062 page = fixup->page;
Chris Mason4a096752008-07-21 10:29:44 -04002063again:
Chris Mason247e7432008-07-17 12:53:51 -04002064 lock_page(page);
2065 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2066 ClearPageChecked(page);
2067 goto out_page;
2068 }
2069
2070 inode = page->mapping->host;
2071 page_start = page_offset(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002072 page_end = page_offset(page) + PAGE_SIZE - 1;
Chris Mason247e7432008-07-17 12:53:51 -04002073
David Sterbaff13db42015-12-03 14:30:40 +01002074 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01002075 &cached_state);
Chris Mason4a096752008-07-21 10:29:44 -04002076
2077 /* already ordered? We're done */
Chris Mason8b62b722009-09-02 16:53:46 -04002078 if (PagePrivate2(page))
Chris Mason247e7432008-07-17 12:53:51 -04002079 goto out;
Chris Mason4a096752008-07-21 10:29:44 -04002080
Nikolay Borisova776c6f2017-02-20 13:50:49 +02002081 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002082 PAGE_SIZE);
Chris Mason4a096752008-07-21 10:29:44 -04002083 if (ordered) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00002084 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2085 page_end, &cached_state, GFP_NOFS);
Chris Mason4a096752008-07-21 10:29:44 -04002086 unlock_page(page);
2087 btrfs_start_ordered_extent(inode, ordered, 1);
Jeff Mahoney87826df2012-02-15 16:23:57 +01002088 btrfs_put_ordered_extent(ordered);
Chris Mason4a096752008-07-21 10:29:44 -04002089 goto again;
2090 }
Chris Mason247e7432008-07-17 12:53:51 -04002091
Qu Wenruo364ecf32017-02-27 15:10:38 +08002092 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002093 PAGE_SIZE);
Jeff Mahoney87826df2012-02-15 16:23:57 +01002094 if (ret) {
2095 mapping_set_error(page->mapping, ret);
2096 end_extent_writepage(page, ret, page_start, page_end);
2097 ClearPageChecked(page);
2098 goto out;
2099 }
2100
Filipe Mananae3b8a482017-11-04 00:16:59 +00002101 btrfs_set_extent_delalloc(inode, page_start, page_end, 0, &cached_state,
Qu Wenruoba8b04c2016-07-19 16:50:36 +08002102 0);
Chris Mason247e7432008-07-17 12:53:51 -04002103 ClearPageChecked(page);
Jeff Mahoney87826df2012-02-15 16:23:57 +01002104 set_page_dirty(page);
Josef Bacik8b62f872017-10-19 14:15:55 -04002105 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
Chris Mason247e7432008-07-17 12:53:51 -04002106out:
Josef Bacik2ac55d42010-02-03 19:33:23 +00002107 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2108 &cached_state, GFP_NOFS);
Chris Mason247e7432008-07-17 12:53:51 -04002109out_page:
2110 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002111 put_page(page);
Miao Xieb897abe2011-01-26 16:19:22 +08002112 kfree(fixup);
Qu Wenruo364ecf32017-02-27 15:10:38 +08002113 extent_changeset_free(data_reserved);
Chris Mason247e7432008-07-17 12:53:51 -04002114}
2115
2116/*
2117 * There are a few paths in the higher layers of the kernel that directly
2118 * set the page dirty bit without asking the filesystem if it is a
2119 * good idea. This causes problems because we want to make sure COW
2120 * properly happens and the data=ordered rules are followed.
2121 *
Chris Masonc8b97812008-10-29 14:49:59 -04002122 * In our case any range that doesn't have the ORDERED bit set
Chris Mason247e7432008-07-17 12:53:51 -04002123 * hasn't been properly setup for IO. We kick off an async process
2124 * to fix it up. The async helper will wait for ordered extents, set
2125 * the delalloc bit and make it safe to write the page.
2126 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05002127static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
Chris Mason247e7432008-07-17 12:53:51 -04002128{
2129 struct inode *inode = page->mapping->host;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002130 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Chris Mason247e7432008-07-17 12:53:51 -04002131 struct btrfs_writepage_fixup *fixup;
Chris Mason247e7432008-07-17 12:53:51 -04002132
Chris Mason8b62b722009-09-02 16:53:46 -04002133 /* this page is properly in the ordered list */
2134 if (TestClearPagePrivate2(page))
Chris Mason247e7432008-07-17 12:53:51 -04002135 return 0;
2136
2137 if (PageChecked(page))
2138 return -EAGAIN;
2139
2140 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2141 if (!fixup)
2142 return -EAGAIN;
Chris Masonf4219502008-07-22 11:18:09 -04002143
Chris Mason247e7432008-07-17 12:53:51 -04002144 SetPageChecked(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002145 get_page(page);
Liu Bo9e0af232014-08-15 23:36:53 +08002146 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2147 btrfs_writepage_fixup_worker, NULL, NULL);
Chris Mason247e7432008-07-17 12:53:51 -04002148 fixup->page = page;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002149 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
Jeff Mahoney87826df2012-02-15 16:23:57 +01002150 return -EBUSY;
Chris Mason247e7432008-07-17 12:53:51 -04002151}
2152
Yan Zhengd899e052008-10-30 14:25:28 -04002153static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2154 struct inode *inode, u64 file_pos,
2155 u64 disk_bytenr, u64 disk_num_bytes,
2156 u64 num_bytes, u64 ram_bytes,
2157 u8 compression, u8 encryption,
2158 u16 other_encoding, int extent_type)
2159{
2160 struct btrfs_root *root = BTRFS_I(inode)->root;
2161 struct btrfs_file_extent_item *fi;
2162 struct btrfs_path *path;
2163 struct extent_buffer *leaf;
2164 struct btrfs_key ins;
Qu Wenruoa12b8772017-02-27 15:10:37 +08002165 u64 qg_released;
Filipe David Borba Manana1acae572014-01-07 11:42:27 +00002166 int extent_inserted = 0;
Yan Zhengd899e052008-10-30 14:25:28 -04002167 int ret;
2168
2169 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07002170 if (!path)
2171 return -ENOMEM;
Yan Zhengd899e052008-10-30 14:25:28 -04002172
Chris Masona1ed8352009-09-11 12:27:37 -04002173 /*
2174 * we may be replacing one extent in the tree with another.
2175 * The new extent is pinned in the extent map, and we don't want
2176 * to drop it from the cache until it is completely in the btree.
2177 *
2178 * So, tell btrfs_drop_extents to leave this extent in the cache.
2179 * the caller is expected to unpin it and allow it to be merged
2180 * with the others.
2181 */
Filipe David Borba Manana1acae572014-01-07 11:42:27 +00002182 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2183 file_pos + num_bytes, NULL, 0,
2184 1, sizeof(*fi), &extent_inserted);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002185 if (ret)
2186 goto out;
Yan Zhengd899e052008-10-30 14:25:28 -04002187
Filipe David Borba Manana1acae572014-01-07 11:42:27 +00002188 if (!extent_inserted) {
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02002189 ins.objectid = btrfs_ino(BTRFS_I(inode));
Filipe David Borba Manana1acae572014-01-07 11:42:27 +00002190 ins.offset = file_pos;
2191 ins.type = BTRFS_EXTENT_DATA_KEY;
2192
2193 path->leave_spinning = 1;
2194 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2195 sizeof(*fi));
2196 if (ret)
2197 goto out;
2198 }
Yan Zhengd899e052008-10-30 14:25:28 -04002199 leaf = path->nodes[0];
2200 fi = btrfs_item_ptr(leaf, path->slots[0],
2201 struct btrfs_file_extent_item);
2202 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2203 btrfs_set_file_extent_type(leaf, fi, extent_type);
2204 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2205 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2206 btrfs_set_file_extent_offset(leaf, fi, 0);
2207 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2208 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2209 btrfs_set_file_extent_compression(leaf, fi, compression);
2210 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2211 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
Chris Masonb9473432009-03-13 11:00:37 -04002212
Yan Zhengd899e052008-10-30 14:25:28 -04002213 btrfs_mark_buffer_dirty(leaf);
Josef Bacikce195332012-09-25 15:26:16 -04002214 btrfs_release_path(path);
Yan Zhengd899e052008-10-30 14:25:28 -04002215
2216 inode_add_bytes(inode, num_bytes);
Yan Zhengd899e052008-10-30 14:25:28 -04002217
2218 ins.objectid = disk_bytenr;
2219 ins.offset = disk_num_bytes;
2220 ins.type = BTRFS_EXTENT_ITEM_KEY;
Qu Wenruoa12b8772017-02-27 15:10:37 +08002221
Qu Wenruo297d7502015-09-08 17:08:37 +08002222 /*
Qu Wenruo5846a3c2015-10-26 14:11:18 +08002223 * Release the reserved range from inode dirty range map, as it is
2224 * already moved into delayed_ref_head
Qu Wenruo297d7502015-09-08 17:08:37 +08002225 */
Qu Wenruoa12b8772017-02-27 15:10:37 +08002226 ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2227 if (ret < 0)
2228 goto out;
2229 qg_released = ret;
Josef Bacik84f7d8e2017-09-29 15:43:49 -04002230 ret = btrfs_alloc_reserved_file_extent(trans, root,
2231 btrfs_ino(BTRFS_I(inode)),
2232 file_pos, qg_released, &ins);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002233out:
Yan Zhengd899e052008-10-30 14:25:28 -04002234 btrfs_free_path(path);
Chris Masonb9473432009-03-13 11:00:37 -04002235
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002236 return ret;
Yan Zhengd899e052008-10-30 14:25:28 -04002237}
2238
Liu Bo38c227d2013-01-29 03:18:40 +00002239/* snapshot-aware defrag */
2240struct sa_defrag_extent_backref {
2241 struct rb_node node;
2242 struct old_sa_defrag_extent *old;
2243 u64 root_id;
2244 u64 inum;
2245 u64 file_pos;
2246 u64 extent_offset;
2247 u64 num_bytes;
2248 u64 generation;
2249};
2250
2251struct old_sa_defrag_extent {
2252 struct list_head list;
2253 struct new_sa_defrag_extent *new;
2254
2255 u64 extent_offset;
2256 u64 bytenr;
2257 u64 offset;
2258 u64 len;
2259 int count;
2260};
2261
2262struct new_sa_defrag_extent {
2263 struct rb_root root;
2264 struct list_head head;
2265 struct btrfs_path *path;
2266 struct inode *inode;
2267 u64 file_pos;
2268 u64 len;
2269 u64 bytenr;
2270 u64 disk_len;
2271 u8 compress_type;
2272};
2273
2274static int backref_comp(struct sa_defrag_extent_backref *b1,
2275 struct sa_defrag_extent_backref *b2)
2276{
2277 if (b1->root_id < b2->root_id)
2278 return -1;
2279 else if (b1->root_id > b2->root_id)
2280 return 1;
2281
2282 if (b1->inum < b2->inum)
2283 return -1;
2284 else if (b1->inum > b2->inum)
2285 return 1;
2286
2287 if (b1->file_pos < b2->file_pos)
2288 return -1;
2289 else if (b1->file_pos > b2->file_pos)
2290 return 1;
2291
2292 /*
2293 * [------------------------------] ===> (a range of space)
2294 * |<--->| |<---->| =============> (fs/file tree A)
2295 * |<---------------------------->| ===> (fs/file tree B)
2296 *
2297 * A range of space can refer to two file extents in one tree while
2298 * refer to only one file extent in another tree.
2299 *
2300 * So we may process a disk offset more than one time(two extents in A)
2301 * and locate at the same extent(one extent in B), then insert two same
2302 * backrefs(both refer to the extent in B).
2303 */
2304 return 0;
2305}
2306
2307static void backref_insert(struct rb_root *root,
2308 struct sa_defrag_extent_backref *backref)
2309{
2310 struct rb_node **p = &root->rb_node;
2311 struct rb_node *parent = NULL;
2312 struct sa_defrag_extent_backref *entry;
2313 int ret;
2314
2315 while (*p) {
2316 parent = *p;
2317 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2318
2319 ret = backref_comp(backref, entry);
2320 if (ret < 0)
2321 p = &(*p)->rb_left;
2322 else
2323 p = &(*p)->rb_right;
2324 }
2325
2326 rb_link_node(&backref->node, parent, p);
2327 rb_insert_color(&backref->node, root);
2328}
2329
2330/*
2331 * Note the backref might has changed, and in this case we just return 0.
2332 */
2333static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2334 void *ctx)
2335{
2336 struct btrfs_file_extent_item *extent;
Liu Bo38c227d2013-01-29 03:18:40 +00002337 struct old_sa_defrag_extent *old = ctx;
2338 struct new_sa_defrag_extent *new = old->new;
2339 struct btrfs_path *path = new->path;
2340 struct btrfs_key key;
2341 struct btrfs_root *root;
2342 struct sa_defrag_extent_backref *backref;
2343 struct extent_buffer *leaf;
2344 struct inode *inode = new->inode;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002345 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Liu Bo38c227d2013-01-29 03:18:40 +00002346 int slot;
2347 int ret;
2348 u64 extent_offset;
2349 u64 num_bytes;
2350
2351 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02002352 inum == btrfs_ino(BTRFS_I(inode)))
Liu Bo38c227d2013-01-29 03:18:40 +00002353 return 0;
2354
2355 key.objectid = root_id;
2356 key.type = BTRFS_ROOT_ITEM_KEY;
2357 key.offset = (u64)-1;
2358
Liu Bo38c227d2013-01-29 03:18:40 +00002359 root = btrfs_read_fs_root_no_name(fs_info, &key);
2360 if (IS_ERR(root)) {
2361 if (PTR_ERR(root) == -ENOENT)
2362 return 0;
2363 WARN_ON(1);
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04002364 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
Liu Bo38c227d2013-01-29 03:18:40 +00002365 inum, offset, root_id);
2366 return PTR_ERR(root);
2367 }
2368
2369 key.objectid = inum;
2370 key.type = BTRFS_EXTENT_DATA_KEY;
2371 if (offset > (u64)-1 << 32)
2372 key.offset = 0;
2373 else
2374 key.offset = offset;
2375
2376 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05302377 if (WARN_ON(ret < 0))
Liu Bo38c227d2013-01-29 03:18:40 +00002378 return ret;
Josef Bacik50f13192013-07-22 12:50:37 -04002379 ret = 0;
Liu Bo38c227d2013-01-29 03:18:40 +00002380
2381 while (1) {
2382 cond_resched();
2383
2384 leaf = path->nodes[0];
2385 slot = path->slots[0];
2386
2387 if (slot >= btrfs_header_nritems(leaf)) {
2388 ret = btrfs_next_leaf(root, path);
2389 if (ret < 0) {
2390 goto out;
2391 } else if (ret > 0) {
2392 ret = 0;
2393 goto out;
2394 }
2395 continue;
2396 }
2397
2398 path->slots[0]++;
2399
2400 btrfs_item_key_to_cpu(leaf, &key, slot);
2401
2402 if (key.objectid > inum)
2403 goto out;
2404
2405 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2406 continue;
2407
2408 extent = btrfs_item_ptr(leaf, slot,
2409 struct btrfs_file_extent_item);
2410
2411 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2412 continue;
2413
Liu Boe68afa42013-07-01 22:13:26 +08002414 /*
2415 * 'offset' refers to the exact key.offset,
2416 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2417 * (key.offset - extent_offset).
2418 */
2419 if (key.offset != offset)
Liu Bo38c227d2013-01-29 03:18:40 +00002420 continue;
2421
Liu Boe68afa42013-07-01 22:13:26 +08002422 extent_offset = btrfs_file_extent_offset(leaf, extent);
Liu Bo38c227d2013-01-29 03:18:40 +00002423 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
Liu Boe68afa42013-07-01 22:13:26 +08002424
Liu Bo38c227d2013-01-29 03:18:40 +00002425 if (extent_offset >= old->extent_offset + old->offset +
2426 old->len || extent_offset + num_bytes <=
2427 old->extent_offset + old->offset)
2428 continue;
Liu Bo38c227d2013-01-29 03:18:40 +00002429 break;
2430 }
2431
2432 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2433 if (!backref) {
2434 ret = -ENOENT;
2435 goto out;
2436 }
2437
2438 backref->root_id = root_id;
2439 backref->inum = inum;
Liu Boe68afa42013-07-01 22:13:26 +08002440 backref->file_pos = offset;
Liu Bo38c227d2013-01-29 03:18:40 +00002441 backref->num_bytes = num_bytes;
2442 backref->extent_offset = extent_offset;
2443 backref->generation = btrfs_file_extent_generation(leaf, extent);
2444 backref->old = old;
2445 backref_insert(&new->root, backref);
2446 old->count++;
2447out:
2448 btrfs_release_path(path);
2449 WARN_ON(ret);
2450 return ret;
2451}
2452
2453static noinline bool record_extent_backrefs(struct btrfs_path *path,
2454 struct new_sa_defrag_extent *new)
2455{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002456 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
Liu Bo38c227d2013-01-29 03:18:40 +00002457 struct old_sa_defrag_extent *old, *tmp;
2458 int ret;
2459
2460 new->path = path;
2461
2462 list_for_each_entry_safe(old, tmp, &new->head, list) {
Liu Boe68afa42013-07-01 22:13:26 +08002463 ret = iterate_inodes_from_logical(old->bytenr +
2464 old->extent_offset, fs_info,
Liu Bo38c227d2013-01-29 03:18:40 +00002465 path, record_one_backref,
Zygo Blaxellc995ab32017-09-22 13:58:45 -04002466 old, false);
Josef Bacik4724b102013-11-05 11:11:40 -05002467 if (ret < 0 && ret != -ENOENT)
2468 return false;
Liu Bo38c227d2013-01-29 03:18:40 +00002469
2470 /* no backref to be processed for this extent */
2471 if (!old->count) {
2472 list_del(&old->list);
2473 kfree(old);
2474 }
2475 }
2476
2477 if (list_empty(&new->head))
2478 return false;
2479
2480 return true;
2481}
2482
2483static int relink_is_mergable(struct extent_buffer *leaf,
2484 struct btrfs_file_extent_item *fi,
Liu Bo116e0022013-08-02 16:30:40 +08002485 struct new_sa_defrag_extent *new)
Liu Bo38c227d2013-01-29 03:18:40 +00002486{
Liu Bo116e0022013-08-02 16:30:40 +08002487 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
Liu Bo38c227d2013-01-29 03:18:40 +00002488 return 0;
2489
2490 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2491 return 0;
2492
Liu Bo116e0022013-08-02 16:30:40 +08002493 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2494 return 0;
2495
2496 if (btrfs_file_extent_encryption(leaf, fi) ||
Liu Bo38c227d2013-01-29 03:18:40 +00002497 btrfs_file_extent_other_encoding(leaf, fi))
2498 return 0;
2499
2500 return 1;
2501}
2502
2503/*
2504 * Note the backref might has changed, and in this case we just return 0.
2505 */
2506static noinline int relink_extent_backref(struct btrfs_path *path,
2507 struct sa_defrag_extent_backref *prev,
2508 struct sa_defrag_extent_backref *backref)
2509{
2510 struct btrfs_file_extent_item *extent;
2511 struct btrfs_file_extent_item *item;
2512 struct btrfs_ordered_extent *ordered;
2513 struct btrfs_trans_handle *trans;
Liu Bo38c227d2013-01-29 03:18:40 +00002514 struct btrfs_root *root;
2515 struct btrfs_key key;
2516 struct extent_buffer *leaf;
2517 struct old_sa_defrag_extent *old = backref->old;
2518 struct new_sa_defrag_extent *new = old->new;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002519 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
Liu Bo38c227d2013-01-29 03:18:40 +00002520 struct inode *inode;
2521 struct extent_state *cached = NULL;
2522 int ret = 0;
2523 u64 start;
2524 u64 len;
2525 u64 lock_start;
2526 u64 lock_end;
2527 bool merge = false;
2528 int index;
2529
2530 if (prev && prev->root_id == backref->root_id &&
2531 prev->inum == backref->inum &&
2532 prev->file_pos + prev->num_bytes == backref->file_pos)
2533 merge = true;
2534
2535 /* step 1: get root */
2536 key.objectid = backref->root_id;
2537 key.type = BTRFS_ROOT_ITEM_KEY;
2538 key.offset = (u64)-1;
2539
Liu Bo38c227d2013-01-29 03:18:40 +00002540 index = srcu_read_lock(&fs_info->subvol_srcu);
2541
2542 root = btrfs_read_fs_root_no_name(fs_info, &key);
2543 if (IS_ERR(root)) {
2544 srcu_read_unlock(&fs_info->subvol_srcu, index);
2545 if (PTR_ERR(root) == -ENOENT)
2546 return 0;
2547 return PTR_ERR(root);
2548 }
Liu Bo38c227d2013-01-29 03:18:40 +00002549
Wang Shilongbcbba5e2014-02-08 23:46:35 +08002550 if (btrfs_root_readonly(root)) {
2551 srcu_read_unlock(&fs_info->subvol_srcu, index);
2552 return 0;
2553 }
2554
Liu Bo38c227d2013-01-29 03:18:40 +00002555 /* step 2: get inode */
2556 key.objectid = backref->inum;
2557 key.type = BTRFS_INODE_ITEM_KEY;
2558 key.offset = 0;
2559
2560 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2561 if (IS_ERR(inode)) {
2562 srcu_read_unlock(&fs_info->subvol_srcu, index);
2563 return 0;
2564 }
2565
2566 srcu_read_unlock(&fs_info->subvol_srcu, index);
2567
2568 /* step 3: relink backref */
2569 lock_start = backref->file_pos;
2570 lock_end = backref->file_pos + backref->num_bytes - 1;
2571 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
David Sterbaff13db42015-12-03 14:30:40 +01002572 &cached);
Liu Bo38c227d2013-01-29 03:18:40 +00002573
2574 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2575 if (ordered) {
2576 btrfs_put_ordered_extent(ordered);
2577 goto out_unlock;
2578 }
2579
2580 trans = btrfs_join_transaction(root);
2581 if (IS_ERR(trans)) {
2582 ret = PTR_ERR(trans);
2583 goto out_unlock;
2584 }
2585
2586 key.objectid = backref->inum;
2587 key.type = BTRFS_EXTENT_DATA_KEY;
2588 key.offset = backref->file_pos;
2589
2590 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2591 if (ret < 0) {
2592 goto out_free_path;
2593 } else if (ret > 0) {
2594 ret = 0;
2595 goto out_free_path;
2596 }
2597
2598 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2599 struct btrfs_file_extent_item);
2600
2601 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2602 backref->generation)
2603 goto out_free_path;
2604
2605 btrfs_release_path(path);
2606
2607 start = backref->file_pos;
2608 if (backref->extent_offset < old->extent_offset + old->offset)
2609 start += old->extent_offset + old->offset -
2610 backref->extent_offset;
2611
2612 len = min(backref->extent_offset + backref->num_bytes,
2613 old->extent_offset + old->offset + old->len);
2614 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2615
2616 ret = btrfs_drop_extents(trans, root, inode, start,
2617 start + len, 1);
2618 if (ret)
2619 goto out_free_path;
2620again:
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02002621 key.objectid = btrfs_ino(BTRFS_I(inode));
Liu Bo38c227d2013-01-29 03:18:40 +00002622 key.type = BTRFS_EXTENT_DATA_KEY;
2623 key.offset = start;
2624
Liu Boa09a0a72013-03-11 09:20:58 +00002625 path->leave_spinning = 1;
Liu Bo38c227d2013-01-29 03:18:40 +00002626 if (merge) {
2627 struct btrfs_file_extent_item *fi;
2628 u64 extent_len;
2629 struct btrfs_key found_key;
2630
Gui Hecheng3c9665d2014-01-23 13:41:09 +08002631 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
Liu Bo38c227d2013-01-29 03:18:40 +00002632 if (ret < 0)
2633 goto out_free_path;
2634
2635 path->slots[0]--;
2636 leaf = path->nodes[0];
2637 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2638
2639 fi = btrfs_item_ptr(leaf, path->slots[0],
2640 struct btrfs_file_extent_item);
2641 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2642
Liu Bo116e0022013-08-02 16:30:40 +08002643 if (extent_len + found_key.offset == start &&
2644 relink_is_mergable(leaf, fi, new)) {
Liu Bo38c227d2013-01-29 03:18:40 +00002645 btrfs_set_file_extent_num_bytes(leaf, fi,
2646 extent_len + len);
2647 btrfs_mark_buffer_dirty(leaf);
2648 inode_add_bytes(inode, len);
2649
2650 ret = 1;
2651 goto out_free_path;
2652 } else {
2653 merge = false;
2654 btrfs_release_path(path);
2655 goto again;
2656 }
2657 }
2658
2659 ret = btrfs_insert_empty_item(trans, root, path, &key,
2660 sizeof(*extent));
2661 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04002662 btrfs_abort_transaction(trans, ret);
Liu Bo38c227d2013-01-29 03:18:40 +00002663 goto out_free_path;
2664 }
2665
2666 leaf = path->nodes[0];
2667 item = btrfs_item_ptr(leaf, path->slots[0],
2668 struct btrfs_file_extent_item);
2669 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2670 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2671 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2672 btrfs_set_file_extent_num_bytes(leaf, item, len);
2673 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2674 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2675 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2676 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2677 btrfs_set_file_extent_encryption(leaf, item, 0);
2678 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2679
2680 btrfs_mark_buffer_dirty(leaf);
2681 inode_add_bytes(inode, len);
Liu Boa09a0a72013-03-11 09:20:58 +00002682 btrfs_release_path(path);
Liu Bo38c227d2013-01-29 03:18:40 +00002683
Josef Bacik84f7d8e2017-09-29 15:43:49 -04002684 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
Liu Bo38c227d2013-01-29 03:18:40 +00002685 new->disk_len, 0,
2686 backref->root_id, backref->inum,
Filipe Mananab06c4bf2015-10-23 07:52:54 +01002687 new->file_pos); /* start - extent_offset */
Liu Bo38c227d2013-01-29 03:18:40 +00002688 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04002689 btrfs_abort_transaction(trans, ret);
Liu Bo38c227d2013-01-29 03:18:40 +00002690 goto out_free_path;
2691 }
2692
2693 ret = 1;
2694out_free_path:
2695 btrfs_release_path(path);
Liu Boa09a0a72013-03-11 09:20:58 +00002696 path->leave_spinning = 0;
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04002697 btrfs_end_transaction(trans);
Liu Bo38c227d2013-01-29 03:18:40 +00002698out_unlock:
2699 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2700 &cached, GFP_NOFS);
2701 iput(inode);
2702 return ret;
2703}
2704
Liu Bo6f519562013-10-29 10:45:05 +08002705static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2706{
2707 struct old_sa_defrag_extent *old, *tmp;
2708
2709 if (!new)
2710 return;
2711
2712 list_for_each_entry_safe(old, tmp, &new->head, list) {
Liu Bo6f519562013-10-29 10:45:05 +08002713 kfree(old);
2714 }
2715 kfree(new);
2716}
2717
Liu Bo38c227d2013-01-29 03:18:40 +00002718static void relink_file_extents(struct new_sa_defrag_extent *new)
2719{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002720 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
Liu Bo38c227d2013-01-29 03:18:40 +00002721 struct btrfs_path *path;
Liu Bo38c227d2013-01-29 03:18:40 +00002722 struct sa_defrag_extent_backref *backref;
2723 struct sa_defrag_extent_backref *prev = NULL;
2724 struct inode *inode;
2725 struct btrfs_root *root;
2726 struct rb_node *node;
2727 int ret;
2728
2729 inode = new->inode;
2730 root = BTRFS_I(inode)->root;
2731
2732 path = btrfs_alloc_path();
2733 if (!path)
2734 return;
2735
2736 if (!record_extent_backrefs(path, new)) {
2737 btrfs_free_path(path);
2738 goto out;
2739 }
2740 btrfs_release_path(path);
2741
2742 while (1) {
2743 node = rb_first(&new->root);
2744 if (!node)
2745 break;
2746 rb_erase(node, &new->root);
2747
2748 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2749
2750 ret = relink_extent_backref(path, prev, backref);
2751 WARN_ON(ret < 0);
2752
2753 kfree(prev);
2754
2755 if (ret == 1)
2756 prev = backref;
2757 else
2758 prev = NULL;
2759 cond_resched();
2760 }
2761 kfree(prev);
2762
2763 btrfs_free_path(path);
Liu Bo38c227d2013-01-29 03:18:40 +00002764out:
Liu Bo6f519562013-10-29 10:45:05 +08002765 free_sa_defrag_extent(new);
2766
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002767 atomic_dec(&fs_info->defrag_running);
2768 wake_up(&fs_info->transaction_wait);
Liu Bo38c227d2013-01-29 03:18:40 +00002769}
2770
2771static struct new_sa_defrag_extent *
2772record_old_file_extents(struct inode *inode,
2773 struct btrfs_ordered_extent *ordered)
2774{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002775 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Liu Bo38c227d2013-01-29 03:18:40 +00002776 struct btrfs_root *root = BTRFS_I(inode)->root;
2777 struct btrfs_path *path;
2778 struct btrfs_key key;
Liu Bo6f519562013-10-29 10:45:05 +08002779 struct old_sa_defrag_extent *old;
Liu Bo38c227d2013-01-29 03:18:40 +00002780 struct new_sa_defrag_extent *new;
2781 int ret;
2782
2783 new = kmalloc(sizeof(*new), GFP_NOFS);
2784 if (!new)
2785 return NULL;
2786
2787 new->inode = inode;
2788 new->file_pos = ordered->file_offset;
2789 new->len = ordered->len;
2790 new->bytenr = ordered->start;
2791 new->disk_len = ordered->disk_len;
2792 new->compress_type = ordered->compress_type;
2793 new->root = RB_ROOT;
2794 INIT_LIST_HEAD(&new->head);
2795
2796 path = btrfs_alloc_path();
2797 if (!path)
2798 goto out_kfree;
2799
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02002800 key.objectid = btrfs_ino(BTRFS_I(inode));
Liu Bo38c227d2013-01-29 03:18:40 +00002801 key.type = BTRFS_EXTENT_DATA_KEY;
2802 key.offset = new->file_pos;
2803
2804 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2805 if (ret < 0)
2806 goto out_free_path;
2807 if (ret > 0 && path->slots[0] > 0)
2808 path->slots[0]--;
2809
2810 /* find out all the old extents for the file range */
2811 while (1) {
2812 struct btrfs_file_extent_item *extent;
2813 struct extent_buffer *l;
2814 int slot;
2815 u64 num_bytes;
2816 u64 offset;
2817 u64 end;
2818 u64 disk_bytenr;
2819 u64 extent_offset;
2820
2821 l = path->nodes[0];
2822 slot = path->slots[0];
2823
2824 if (slot >= btrfs_header_nritems(l)) {
2825 ret = btrfs_next_leaf(root, path);
2826 if (ret < 0)
Liu Bo6f519562013-10-29 10:45:05 +08002827 goto out_free_path;
Liu Bo38c227d2013-01-29 03:18:40 +00002828 else if (ret > 0)
2829 break;
2830 continue;
2831 }
2832
2833 btrfs_item_key_to_cpu(l, &key, slot);
2834
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02002835 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
Liu Bo38c227d2013-01-29 03:18:40 +00002836 break;
2837 if (key.type != BTRFS_EXTENT_DATA_KEY)
2838 break;
2839 if (key.offset >= new->file_pos + new->len)
2840 break;
2841
2842 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2843
2844 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2845 if (key.offset + num_bytes < new->file_pos)
2846 goto next;
2847
2848 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2849 if (!disk_bytenr)
2850 goto next;
2851
2852 extent_offset = btrfs_file_extent_offset(l, extent);
2853
2854 old = kmalloc(sizeof(*old), GFP_NOFS);
2855 if (!old)
Liu Bo6f519562013-10-29 10:45:05 +08002856 goto out_free_path;
Liu Bo38c227d2013-01-29 03:18:40 +00002857
2858 offset = max(new->file_pos, key.offset);
2859 end = min(new->file_pos + new->len, key.offset + num_bytes);
2860
2861 old->bytenr = disk_bytenr;
2862 old->extent_offset = extent_offset;
2863 old->offset = offset - key.offset;
2864 old->len = end - offset;
2865 old->new = new;
2866 old->count = 0;
2867 list_add_tail(&old->list, &new->head);
2868next:
2869 path->slots[0]++;
2870 cond_resched();
2871 }
2872
2873 btrfs_free_path(path);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002874 atomic_inc(&fs_info->defrag_running);
Liu Bo38c227d2013-01-29 03:18:40 +00002875
2876 return new;
2877
Liu Bo38c227d2013-01-29 03:18:40 +00002878out_free_path:
2879 btrfs_free_path(path);
2880out_kfree:
Liu Bo6f519562013-10-29 10:45:05 +08002881 free_sa_defrag_extent(new);
Liu Bo38c227d2013-01-29 03:18:40 +00002882 return NULL;
2883}
2884
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002885static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
Miao Xiee570fd22014-06-19 10:42:50 +08002886 u64 start, u64 len)
2887{
2888 struct btrfs_block_group_cache *cache;
2889
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002890 cache = btrfs_lookup_block_group(fs_info, start);
Miao Xiee570fd22014-06-19 10:42:50 +08002891 ASSERT(cache);
2892
2893 spin_lock(&cache->lock);
2894 cache->delalloc_bytes -= len;
2895 spin_unlock(&cache->lock);
2896
2897 btrfs_put_block_group(cache);
2898}
2899
Chris Masond352ac62008-09-29 15:18:18 -04002900/* as ordered data IO finishes, this gets called so we can finish
2901 * an ordered extent if the range of bytes in the file it covers are
2902 * fully written.
2903 */
Josef Bacik5fd02042012-05-02 14:00:54 -04002904static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
Chris Masone6dcd2d2008-07-17 12:53:50 -04002905{
Josef Bacik5fd02042012-05-02 14:00:54 -04002906 struct inode *inode = ordered_extent->inode;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002907 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Chris Masone6dcd2d2008-07-17 12:53:50 -04002908 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04002909 struct btrfs_trans_handle *trans = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04002910 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Josef Bacik2ac55d42010-02-03 19:33:23 +00002911 struct extent_state *cached_state = NULL;
Liu Bo38c227d2013-01-29 03:18:40 +00002912 struct new_sa_defrag_extent *new = NULL;
Li Zefan261507a02010-12-17 14:21:50 +08002913 int compress_type = 0;
Josef Bacik77cef2e2013-08-29 13:57:21 -04002914 int ret = 0;
2915 u64 logical_len = ordered_extent->len;
Li Zefan82d59022011-04-20 10:33:24 +08002916 bool nolock;
Josef Bacik77cef2e2013-08-29 13:57:21 -04002917 bool truncated = false;
Filipe Mananaa7e3b972017-04-03 10:45:46 +01002918 bool range_locked = false;
2919 bool clear_new_delalloc_bytes = false;
2920
2921 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2922 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2923 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2924 clear_new_delalloc_bytes = true;
Chris Masone6dcd2d2008-07-17 12:53:50 -04002925
Nikolay Borisov70ddc552017-02-20 13:50:35 +02002926 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
Josef Bacik0cb59c92010-07-02 12:14:14 -04002927
Josef Bacik5fd02042012-05-02 14:00:54 -04002928 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2929 ret = -EIO;
2930 goto out;
2931 }
2932
Nikolay Borisov7ab79562017-02-20 13:50:57 +02002933 btrfs_free_io_failure_record(BTRFS_I(inode),
2934 ordered_extent->file_offset,
2935 ordered_extent->file_offset +
2936 ordered_extent->len - 1);
Miao Xief6124962014-09-12 18:44:04 +08002937
Josef Bacik77cef2e2013-08-29 13:57:21 -04002938 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2939 truncated = true;
2940 logical_len = ordered_extent->truncated_len;
2941 /* Truncated the entire extent, don't bother adding */
2942 if (!logical_len)
2943 goto out;
2944 }
2945
Yan, Zhengc2167752009-11-12 09:34:21 +00002946 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002947 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
Qu Wenruo94ed9382015-09-08 17:25:56 +08002948
2949 /*
2950 * For mwrite(mmap + memset to write) case, we still reserve
2951 * space for NOCOW range.
2952 * As NOCOW won't cause a new delayed ref, just free the space
2953 */
Qu Wenruobc42bda2017-02-27 15:10:39 +08002954 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
Qu Wenruo94ed9382015-09-08 17:25:56 +08002955 ordered_extent->len);
Josef Bacik6c760c02012-11-09 10:53:21 -05002956 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2957 if (nolock)
2958 trans = btrfs_join_transaction_nolock(root);
2959 else
2960 trans = btrfs_join_transaction(root);
2961 if (IS_ERR(trans)) {
2962 ret = PTR_ERR(trans);
2963 trans = NULL;
2964 goto out;
Yan, Zhengc2167752009-11-12 09:34:21 +00002965 }
Josef Bacik69fe2d72017-10-19 14:15:57 -04002966 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
Josef Bacik6c760c02012-11-09 10:53:21 -05002967 ret = btrfs_update_inode_fallback(trans, root, inode);
2968 if (ret) /* -ENOMEM or corruption */
Jeff Mahoney66642832016-06-10 18:19:25 -04002969 btrfs_abort_transaction(trans, ret);
Yan, Zhengc2167752009-11-12 09:34:21 +00002970 goto out;
2971 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04002972
Filipe Mananaa7e3b972017-04-03 10:45:46 +01002973 range_locked = true;
Josef Bacik2ac55d42010-02-03 19:33:23 +00002974 lock_extent_bits(io_tree, ordered_extent->file_offset,
2975 ordered_extent->file_offset + ordered_extent->len - 1,
David Sterbaff13db42015-12-03 14:30:40 +01002976 &cached_state);
Chris Masone6dcd2d2008-07-17 12:53:50 -04002977
Liu Bo38c227d2013-01-29 03:18:40 +00002978 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2979 ordered_extent->file_offset + ordered_extent->len - 1,
Liu Bo452e62b2017-05-26 17:44:23 -06002980 EXTENT_DEFRAG, 0, cached_state);
Liu Bo38c227d2013-01-29 03:18:40 +00002981 if (ret) {
2982 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
Josef Bacik8101c8d2014-01-29 16:05:30 -05002983 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
Liu Bo38c227d2013-01-29 03:18:40 +00002984 /* the inode is shared */
2985 new = record_old_file_extents(inode, ordered_extent);
2986
2987 clear_extent_bit(io_tree, ordered_extent->file_offset,
2988 ordered_extent->file_offset + ordered_extent->len - 1,
2989 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2990 }
2991
Josef Bacik0cb59c92010-07-02 12:14:14 -04002992 if (nolock)
Josef Bacik7a7eaa42011-04-13 12:54:33 -04002993 trans = btrfs_join_transaction_nolock(root);
Josef Bacik0cb59c92010-07-02 12:14:14 -04002994 else
Josef Bacik7a7eaa42011-04-13 12:54:33 -04002995 trans = btrfs_join_transaction(root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002996 if (IS_ERR(trans)) {
2997 ret = PTR_ERR(trans);
2998 trans = NULL;
Filipe Mananaa7e3b972017-04-03 10:45:46 +01002999 goto out;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003000 }
Chris Masona79b7d42014-05-22 16:18:52 -07003001
Josef Bacik69fe2d72017-10-19 14:15:57 -04003002 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
Yan, Zhengc2167752009-11-12 09:34:21 +00003003
Chris Masonc8b97812008-10-29 14:49:59 -04003004 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
Li Zefan261507a02010-12-17 14:21:50 +08003005 compress_type = ordered_extent->compress_type;
Yan Zhengd899e052008-10-30 14:25:28 -04003006 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
Li Zefan261507a02010-12-17 14:21:50 +08003007 BUG_ON(compress_type);
Nikolay Borisov7a6d7062017-02-20 13:50:48 +02003008 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
Yan Zhengd899e052008-10-30 14:25:28 -04003009 ordered_extent->file_offset,
3010 ordered_extent->file_offset +
Josef Bacik77cef2e2013-08-29 13:57:21 -04003011 logical_len);
Yan Zhengd899e052008-10-30 14:25:28 -04003012 } else {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003013 BUG_ON(root == fs_info->tree_root);
Yan Zhengd899e052008-10-30 14:25:28 -04003014 ret = insert_reserved_file_extent(trans, inode,
3015 ordered_extent->file_offset,
3016 ordered_extent->start,
3017 ordered_extent->disk_len,
Josef Bacik77cef2e2013-08-29 13:57:21 -04003018 logical_len, logical_len,
Li Zefan261507a02010-12-17 14:21:50 +08003019 compress_type, 0, 0,
Yan Zhengd899e052008-10-30 14:25:28 -04003020 BTRFS_FILE_EXTENT_REG);
Miao Xiee570fd22014-06-19 10:42:50 +08003021 if (!ret)
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04003022 btrfs_release_delalloc_bytes(fs_info,
Miao Xiee570fd22014-06-19 10:42:50 +08003023 ordered_extent->start,
3024 ordered_extent->disk_len);
Yan Zhengd899e052008-10-30 14:25:28 -04003025 }
Josef Bacik5dc562c2012-08-17 13:14:17 -04003026 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
3027 ordered_extent->file_offset, ordered_extent->len,
3028 trans->transid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003029 if (ret < 0) {
Jeff Mahoney66642832016-06-10 18:19:25 -04003030 btrfs_abort_transaction(trans, ret);
Filipe Mananaa7e3b972017-04-03 10:45:46 +01003031 goto out;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003032 }
Josef Bacik2ac55d42010-02-03 19:33:23 +00003033
David Sterbadf9f6282017-02-10 19:35:37 +01003034 add_pending_csums(trans, inode, &ordered_extent->list);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003035
Josef Bacik6c760c02012-11-09 10:53:21 -05003036 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3037 ret = btrfs_update_inode_fallback(trans, root, inode);
3038 if (ret) { /* -ENOMEM or corruption */
Jeff Mahoney66642832016-06-10 18:19:25 -04003039 btrfs_abort_transaction(trans, ret);
Filipe Mananaa7e3b972017-04-03 10:45:46 +01003040 goto out;
Josef Bacik1ef30be2011-04-05 19:25:36 -04003041 }
3042 ret = 0;
Yan, Zhengc2167752009-11-12 09:34:21 +00003043out:
Filipe Mananaa7e3b972017-04-03 10:45:46 +01003044 if (range_locked || clear_new_delalloc_bytes) {
3045 unsigned int clear_bits = 0;
3046
3047 if (range_locked)
3048 clear_bits |= EXTENT_LOCKED;
3049 if (clear_new_delalloc_bytes)
3050 clear_bits |= EXTENT_DELALLOC_NEW;
3051 clear_extent_bit(&BTRFS_I(inode)->io_tree,
3052 ordered_extent->file_offset,
3053 ordered_extent->file_offset +
3054 ordered_extent->len - 1,
3055 clear_bits,
3056 (clear_bits & EXTENT_LOCKED) ? 1 : 0,
3057 0, &cached_state, GFP_NOFS);
3058 }
3059
Miao Xiea698d0752012-09-20 01:51:59 -06003060 if (trans)
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04003061 btrfs_end_transaction(trans);
Josef Bacik0cb59c92010-07-02 12:14:14 -04003062
Josef Bacik77cef2e2013-08-29 13:57:21 -04003063 if (ret || truncated) {
3064 u64 start, end;
3065
3066 if (truncated)
3067 start = ordered_extent->file_offset + logical_len;
3068 else
3069 start = ordered_extent->file_offset;
3070 end = ordered_extent->file_offset + ordered_extent->len - 1;
3071 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
3072
3073 /* Drop the cache for the part of the extent we didn't write. */
Nikolay Borisovdcdbc052017-02-20 13:50:45 +02003074 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
Josef Bacik5fd02042012-05-02 14:00:54 -04003075
Josef Bacik0bec9ef2013-01-31 14:58:00 -05003076 /*
3077 * If the ordered extent had an IOERR or something else went
3078 * wrong we need to return the space for this ordered extent
Josef Bacik77cef2e2013-08-29 13:57:21 -04003079 * back to the allocator. We only free the extent in the
3080 * truncated case if we didn't write out the extent at all.
Josef Bacik0bec9ef2013-01-31 14:58:00 -05003081 */
Josef Bacik77cef2e2013-08-29 13:57:21 -04003082 if ((ret || !logical_len) &&
3083 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
Josef Bacik0bec9ef2013-01-31 14:58:00 -05003084 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04003085 btrfs_free_reserved_extent(fs_info,
3086 ordered_extent->start,
Miao Xiee570fd22014-06-19 10:42:50 +08003087 ordered_extent->disk_len, 1);
Josef Bacik0bec9ef2013-01-31 14:58:00 -05003088 }
3089
3090
Josef Bacik5fd02042012-05-02 14:00:54 -04003091 /*
Liu Bo8bad3c02012-06-18 12:14:23 +08003092 * This needs to be done to make sure anybody waiting knows we are done
3093 * updating everything for this ordered extent.
Josef Bacik5fd02042012-05-02 14:00:54 -04003094 */
3095 btrfs_remove_ordered_extent(inode, ordered_extent);
3096
Liu Bo38c227d2013-01-29 03:18:40 +00003097 /* for snapshot-aware defrag */
Liu Bo6f519562013-10-29 10:45:05 +08003098 if (new) {
3099 if (ret) {
3100 free_sa_defrag_extent(new);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003101 atomic_dec(&fs_info->defrag_running);
Liu Bo6f519562013-10-29 10:45:05 +08003102 } else {
3103 relink_file_extents(new);
3104 }
3105 }
Liu Bo38c227d2013-01-29 03:18:40 +00003106
Chris Masone6dcd2d2008-07-17 12:53:50 -04003107 /* once for us */
3108 btrfs_put_ordered_extent(ordered_extent);
3109 /* once for the tree */
3110 btrfs_put_ordered_extent(ordered_extent);
3111
Josef Bacik5fd02042012-05-02 14:00:54 -04003112 return ret;
3113}
3114
3115static void finish_ordered_fn(struct btrfs_work *work)
3116{
3117 struct btrfs_ordered_extent *ordered_extent;
3118 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3119 btrfs_finish_ordered_io(ordered_extent);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003120}
3121
David Sterbac3988d62017-02-17 15:18:32 +01003122static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
Chris Mason211f90e2008-07-18 11:56:15 -04003123 struct extent_state *state, int uptodate)
3124{
Josef Bacik5fd02042012-05-02 14:00:54 -04003125 struct inode *inode = page->mapping->host;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003126 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Josef Bacik5fd02042012-05-02 14:00:54 -04003127 struct btrfs_ordered_extent *ordered_extent = NULL;
Liu Bo9e0af232014-08-15 23:36:53 +08003128 struct btrfs_workqueue *wq;
3129 btrfs_work_func_t func;
Josef Bacik5fd02042012-05-02 14:00:54 -04003130
liubo1abe9b82011-03-24 11:18:59 +00003131 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3132
Chris Mason8b62b722009-09-02 16:53:46 -04003133 ClearPagePrivate2(page);
Josef Bacik5fd02042012-05-02 14:00:54 -04003134 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3135 end - start + 1, uptodate))
David Sterbac3988d62017-02-17 15:18:32 +01003136 return;
Josef Bacik5fd02042012-05-02 14:00:54 -04003137
Nikolay Borisov70ddc552017-02-20 13:50:35 +02003138 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003139 wq = fs_info->endio_freespace_worker;
Liu Bo9e0af232014-08-15 23:36:53 +08003140 func = btrfs_freespace_write_helper;
3141 } else {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003142 wq = fs_info->endio_write_workers;
Liu Bo9e0af232014-08-15 23:36:53 +08003143 func = btrfs_endio_write_helper;
3144 }
Josef Bacik5fd02042012-05-02 14:00:54 -04003145
Liu Bo9e0af232014-08-15 23:36:53 +08003146 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3147 NULL);
3148 btrfs_queue_work(wq, &ordered_extent->work);
Chris Mason211f90e2008-07-18 11:56:15 -04003149}
3150
Miao Xiedc380ae2014-09-12 18:43:55 +08003151static int __readpage_endio_check(struct inode *inode,
3152 struct btrfs_io_bio *io_bio,
3153 int icsum, struct page *page,
3154 int pgoff, u64 start, size_t len)
3155{
3156 char *kaddr;
3157 u32 csum_expected;
3158 u32 csum = ~(u32)0;
Miao Xiedc380ae2014-09-12 18:43:55 +08003159
3160 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3161
3162 kaddr = kmap_atomic(page);
3163 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
Domagoj Tršan0b5e3da2016-10-27 08:52:33 +01003164 btrfs_csum_final(csum, (u8 *)&csum);
Miao Xiedc380ae2014-09-12 18:43:55 +08003165 if (csum != csum_expected)
3166 goto zeroit;
3167
3168 kunmap_atomic(kaddr);
3169 return 0;
3170zeroit:
Nikolay Borisov0970a222017-02-20 13:50:53 +02003171 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
Qu Wenruo6f6b6432017-02-09 10:45:06 +08003172 io_bio->mirror_num);
Miao Xiedc380ae2014-09-12 18:43:55 +08003173 memset(kaddr + pgoff, 1, len);
3174 flush_dcache_page(page);
3175 kunmap_atomic(kaddr);
Miao Xiedc380ae2014-09-12 18:43:55 +08003176 return -EIO;
3177}
3178
Chris Masond352ac62008-09-29 15:18:18 -04003179/*
Chris Masond352ac62008-09-29 15:18:18 -04003180 * when reads are done, we need to check csums to verify the data is correct
Jan Schmidt4a54c8c2011-07-22 15:41:52 +02003181 * if there's a match, we allow the bio to finish. If not, the code in
3182 * extent_io.c will try to find good copies for us.
Chris Masond352ac62008-09-29 15:18:18 -04003183 */
Miao Xiefacc8a222013-07-25 19:22:34 +08003184static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3185 u64 phy_offset, struct page *page,
3186 u64 start, u64 end, int mirror)
Chris Mason07157aa2007-08-30 08:50:51 -04003187{
Miao Xie4eee4fa2012-12-21 09:17:45 +00003188 size_t offset = start - page_offset(page);
Chris Mason07157aa2007-08-30 08:50:51 -04003189 struct inode *inode = page->mapping->host;
Chris Masond1310b22008-01-24 16:13:08 -05003190 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Chris Masonff79f812007-10-15 16:22:25 -04003191 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masond1310b22008-01-24 16:13:08 -05003192
Chris Masond20f7042008-12-08 16:58:54 -05003193 if (PageChecked(page)) {
3194 ClearPageChecked(page);
Miao Xiedc380ae2014-09-12 18:43:55 +08003195 return 0;
Chris Masond20f7042008-12-08 16:58:54 -05003196 }
Christoph Hellwig6cbff002009-04-17 10:37:41 +02003197
3198 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
Miao Xiedc380ae2014-09-12 18:43:55 +08003199 return 0;
Chris Masond20f7042008-12-08 16:58:54 -05003200
Yan Zheng17d217f2008-12-12 10:03:38 -05003201 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
Chris Mason9655d292009-09-02 15:22:30 -04003202 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
David Sterba91166212016-04-26 23:54:39 +02003203 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
Yan Zheng17d217f2008-12-12 10:03:38 -05003204 return 0;
3205 }
3206
Miao Xiefacc8a222013-07-25 19:22:34 +08003207 phy_offset >>= inode->i_sb->s_blocksize_bits;
Miao Xiedc380ae2014-09-12 18:43:55 +08003208 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3209 start, (size_t)(end - start + 1));
Chris Mason07157aa2007-08-30 08:50:51 -04003210}
Chris Masonb888db22007-08-27 16:49:44 -04003211
Yan, Zheng24bbcf02009-11-12 09:36:34 +00003212void btrfs_add_delayed_iput(struct inode *inode)
3213{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003214 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
David Sterba8089fe62015-11-19 14:15:51 +01003215 struct btrfs_inode *binode = BTRFS_I(inode);
Yan, Zheng24bbcf02009-11-12 09:36:34 +00003216
3217 if (atomic_add_unless(&inode->i_count, -1, 1))
3218 return;
3219
Yan, Zheng24bbcf02009-11-12 09:36:34 +00003220 spin_lock(&fs_info->delayed_iput_lock);
David Sterba8089fe62015-11-19 14:15:51 +01003221 if (binode->delayed_iput_count == 0) {
3222 ASSERT(list_empty(&binode->delayed_iput));
3223 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3224 } else {
3225 binode->delayed_iput_count++;
3226 }
Yan, Zheng24bbcf02009-11-12 09:36:34 +00003227 spin_unlock(&fs_info->delayed_iput_lock);
3228}
3229
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04003230void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
Yan, Zheng24bbcf02009-11-12 09:36:34 +00003231{
Yan, Zheng24bbcf02009-11-12 09:36:34 +00003232
Yan, Zheng24bbcf02009-11-12 09:36:34 +00003233 spin_lock(&fs_info->delayed_iput_lock);
David Sterba8089fe62015-11-19 14:15:51 +01003234 while (!list_empty(&fs_info->delayed_iputs)) {
3235 struct btrfs_inode *inode;
Yan, Zheng24bbcf02009-11-12 09:36:34 +00003236
David Sterba8089fe62015-11-19 14:15:51 +01003237 inode = list_first_entry(&fs_info->delayed_iputs,
3238 struct btrfs_inode, delayed_iput);
3239 if (inode->delayed_iput_count) {
3240 inode->delayed_iput_count--;
3241 list_move_tail(&inode->delayed_iput,
3242 &fs_info->delayed_iputs);
3243 } else {
3244 list_del_init(&inode->delayed_iput);
3245 }
3246 spin_unlock(&fs_info->delayed_iput_lock);
3247 iput(&inode->vfs_inode);
3248 spin_lock(&fs_info->delayed_iput_lock);
Yan, Zheng24bbcf02009-11-12 09:36:34 +00003249 }
David Sterba8089fe62015-11-19 14:15:51 +01003250 spin_unlock(&fs_info->delayed_iput_lock);
Yan, Zheng24bbcf02009-11-12 09:36:34 +00003251}
3252
Yan, Zhengd68fc572010-05-16 10:49:58 -04003253/*
Justin P. Mattock42b2aa82011-11-28 20:31:00 -08003254 * This is called in transaction commit time. If there are no orphan
Yan, Zhengd68fc572010-05-16 10:49:58 -04003255 * files in the subvolume, it removes orphan item and frees block_rsv
3256 * structure.
3257 */
3258void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3259 struct btrfs_root *root)
3260{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003261 struct btrfs_fs_info *fs_info = root->fs_info;
Josef Bacik90290e12011-12-02 15:44:12 -05003262 struct btrfs_block_rsv *block_rsv;
Yan, Zhengd68fc572010-05-16 10:49:58 -04003263 int ret;
3264
Josef Bacik8a35d952012-05-23 14:26:42 -04003265 if (atomic_read(&root->orphan_inodes) ||
Yan, Zhengd68fc572010-05-16 10:49:58 -04003266 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3267 return;
3268
Josef Bacik90290e12011-12-02 15:44:12 -05003269 spin_lock(&root->orphan_lock);
Josef Bacik8a35d952012-05-23 14:26:42 -04003270 if (atomic_read(&root->orphan_inodes)) {
Josef Bacik90290e12011-12-02 15:44:12 -05003271 spin_unlock(&root->orphan_lock);
3272 return;
3273 }
3274
3275 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3276 spin_unlock(&root->orphan_lock);
3277 return;
3278 }
3279
3280 block_rsv = root->orphan_block_rsv;
3281 root->orphan_block_rsv = NULL;
3282 spin_unlock(&root->orphan_lock);
3283
Miao Xie27cdeb72014-04-02 19:51:05 +08003284 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
Yan, Zhengd68fc572010-05-16 10:49:58 -04003285 btrfs_root_refs(&root->root_item) > 0) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003286 ret = btrfs_del_orphan_item(trans, fs_info->tree_root,
Yan, Zhengd68fc572010-05-16 10:49:58 -04003287 root->root_key.objectid);
Josef Bacik4ef31a42013-08-13 14:10:08 -04003288 if (ret)
Jeff Mahoney66642832016-06-10 18:19:25 -04003289 btrfs_abort_transaction(trans, ret);
Josef Bacik4ef31a42013-08-13 14:10:08 -04003290 else
Miao Xie27cdeb72014-04-02 19:51:05 +08003291 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3292 &root->state);
Yan, Zhengd68fc572010-05-16 10:49:58 -04003293 }
3294
Josef Bacik90290e12011-12-02 15:44:12 -05003295 if (block_rsv) {
3296 WARN_ON(block_rsv->size > 0);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04003297 btrfs_free_block_rsv(fs_info, block_rsv);
Yan, Zhengd68fc572010-05-16 10:49:58 -04003298 }
3299}
3300
3301/*
Josef Bacik7b128762008-07-24 12:17:14 -04003302 * This creates an orphan entry for the given inode in case something goes
3303 * wrong in the middle of an unlink/truncate.
Yan, Zhengd68fc572010-05-16 10:49:58 -04003304 *
3305 * NOTE: caller of this function should reserve 5 units of metadata for
3306 * this function.
Josef Bacik7b128762008-07-24 12:17:14 -04003307 */
Nikolay Borisov73f2e542017-02-20 13:50:59 +02003308int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3309 struct btrfs_inode *inode)
Josef Bacik7b128762008-07-24 12:17:14 -04003310{
Nikolay Borisov73f2e542017-02-20 13:50:59 +02003311 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
3312 struct btrfs_root *root = inode->root;
Yan, Zhengd68fc572010-05-16 10:49:58 -04003313 struct btrfs_block_rsv *block_rsv = NULL;
3314 int reserve = 0;
3315 int insert = 0;
3316 int ret;
Josef Bacik7b128762008-07-24 12:17:14 -04003317
Yan, Zhengd68fc572010-05-16 10:49:58 -04003318 if (!root->orphan_block_rsv) {
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04003319 block_rsv = btrfs_alloc_block_rsv(fs_info,
3320 BTRFS_BLOCK_RSV_TEMP);
Tsutomu Itohb5324022011-07-19 07:27:20 +00003321 if (!block_rsv)
3322 return -ENOMEM;
Josef Bacik7b128762008-07-24 12:17:14 -04003323 }
3324
Yan, Zhengd68fc572010-05-16 10:49:58 -04003325 spin_lock(&root->orphan_lock);
3326 if (!root->orphan_block_rsv) {
3327 root->orphan_block_rsv = block_rsv;
3328 } else if (block_rsv) {
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04003329 btrfs_free_block_rsv(fs_info, block_rsv);
Yan, Zhengd68fc572010-05-16 10:49:58 -04003330 block_rsv = NULL;
3331 }
Josef Bacik7b128762008-07-24 12:17:14 -04003332
Josef Bacik8a35d952012-05-23 14:26:42 -04003333 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
Nikolay Borisov73f2e542017-02-20 13:50:59 +02003334 &inode->runtime_flags)) {
Yan, Zhengd68fc572010-05-16 10:49:58 -04003335#if 0
3336 /*
3337 * For proper ENOSPC handling, we should do orphan
3338 * cleanup when mounting. But this introduces backward
3339 * compatibility issue.
3340 */
3341 if (!xchg(&root->orphan_item_inserted, 1))
3342 insert = 2;
3343 else
3344 insert = 1;
3345#endif
3346 insert = 1;
Miao Xie321f0e72012-08-28 22:13:02 -06003347 atomic_inc(&root->orphan_inodes);
Yan, Zhengd68fc572010-05-16 10:49:58 -04003348 }
Josef Bacik7b128762008-07-24 12:17:14 -04003349
Josef Bacik72ac3c02012-05-23 14:13:11 -04003350 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
Nikolay Borisov73f2e542017-02-20 13:50:59 +02003351 &inode->runtime_flags))
Yan, Zhengd68fc572010-05-16 10:49:58 -04003352 reserve = 1;
Yan, Zhengd68fc572010-05-16 10:49:58 -04003353 spin_unlock(&root->orphan_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04003354
Yan, Zhengd68fc572010-05-16 10:49:58 -04003355 /* grab metadata reservation from transaction handle */
3356 if (reserve) {
3357 ret = btrfs_orphan_reserve_metadata(trans, inode);
Josef Bacik3b6571c2016-05-27 13:03:04 -04003358 ASSERT(!ret);
3359 if (ret) {
3360 atomic_dec(&root->orphan_inodes);
3361 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
Nikolay Borisov73f2e542017-02-20 13:50:59 +02003362 &inode->runtime_flags);
Josef Bacik3b6571c2016-05-27 13:03:04 -04003363 if (insert)
3364 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
Nikolay Borisov73f2e542017-02-20 13:50:59 +02003365 &inode->runtime_flags);
Josef Bacik3b6571c2016-05-27 13:03:04 -04003366 return ret;
3367 }
Yan, Zhengd68fc572010-05-16 10:49:58 -04003368 }
3369
3370 /* insert an orphan item to track this unlinked/truncated file */
3371 if (insert >= 1) {
Nikolay Borisov73f2e542017-02-20 13:50:59 +02003372 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
Josef Bacik4ef31a42013-08-13 14:10:08 -04003373 if (ret) {
Filipe David Borba Manana703c88e2013-09-22 21:54:55 +01003374 atomic_dec(&root->orphan_inodes);
Josef Bacik4ef31a42013-08-13 14:10:08 -04003375 if (reserve) {
3376 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
Nikolay Borisov73f2e542017-02-20 13:50:59 +02003377 &inode->runtime_flags);
Josef Bacik4ef31a42013-08-13 14:10:08 -04003378 btrfs_orphan_release_metadata(inode);
3379 }
3380 if (ret != -EEXIST) {
Josef Bacike8e7cff2013-08-21 15:54:00 -04003381 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
Nikolay Borisov73f2e542017-02-20 13:50:59 +02003382 &inode->runtime_flags);
Jeff Mahoney66642832016-06-10 18:19:25 -04003383 btrfs_abort_transaction(trans, ret);
Josef Bacik4ef31a42013-08-13 14:10:08 -04003384 return ret;
3385 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003386 }
3387 ret = 0;
Yan, Zhengd68fc572010-05-16 10:49:58 -04003388 }
3389
3390 /* insert an orphan item to track subvolume contains orphan files */
3391 if (insert >= 2) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003392 ret = btrfs_insert_orphan_item(trans, fs_info->tree_root,
Yan, Zhengd68fc572010-05-16 10:49:58 -04003393 root->root_key.objectid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003394 if (ret && ret != -EEXIST) {
Jeff Mahoney66642832016-06-10 18:19:25 -04003395 btrfs_abort_transaction(trans, ret);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003396 return ret;
3397 }
Yan, Zhengd68fc572010-05-16 10:49:58 -04003398 }
3399 return 0;
Josef Bacik7b128762008-07-24 12:17:14 -04003400}
3401
3402/*
3403 * We have done the truncate/delete so we can go ahead and remove the orphan
3404 * item for this particular inode.
3405 */
Eric Sandeen48a3b632013-04-25 20:41:01 +00003406static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
Nikolay Borisov3d6ae7b2017-02-20 13:50:58 +02003407 struct btrfs_inode *inode)
Josef Bacik7b128762008-07-24 12:17:14 -04003408{
Nikolay Borisov3d6ae7b2017-02-20 13:50:58 +02003409 struct btrfs_root *root = inode->root;
Yan, Zhengd68fc572010-05-16 10:49:58 -04003410 int delete_item = 0;
3411 int release_rsv = 0;
Josef Bacik7b128762008-07-24 12:17:14 -04003412 int ret = 0;
3413
Yan, Zhengd68fc572010-05-16 10:49:58 -04003414 spin_lock(&root->orphan_lock);
Josef Bacik8a35d952012-05-23 14:26:42 -04003415 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
Nikolay Borisov3d6ae7b2017-02-20 13:50:58 +02003416 &inode->runtime_flags))
Yan, Zhengd68fc572010-05-16 10:49:58 -04003417 delete_item = 1;
Josef Bacik7b128762008-07-24 12:17:14 -04003418
Josef Bacik72ac3c02012-05-23 14:13:11 -04003419 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
Nikolay Borisov3d6ae7b2017-02-20 13:50:58 +02003420 &inode->runtime_flags))
Yan, Zhengd68fc572010-05-16 10:49:58 -04003421 release_rsv = 1;
Yan, Zhengd68fc572010-05-16 10:49:58 -04003422 spin_unlock(&root->orphan_lock);
3423
Filipe David Borba Manana703c88e2013-09-22 21:54:55 +01003424 if (delete_item) {
Josef Bacik8a35d952012-05-23 14:26:42 -04003425 atomic_dec(&root->orphan_inodes);
Filipe David Borba Manana703c88e2013-09-22 21:54:55 +01003426 if (trans)
3427 ret = btrfs_del_orphan_item(trans, root,
Nikolay Borisov3d6ae7b2017-02-20 13:50:58 +02003428 btrfs_ino(inode));
Josef Bacik8a35d952012-05-23 14:26:42 -04003429 }
Josef Bacik7b128762008-07-24 12:17:14 -04003430
Filipe David Borba Manana703c88e2013-09-22 21:54:55 +01003431 if (release_rsv)
3432 btrfs_orphan_release_metadata(inode);
3433
Josef Bacik4ef31a42013-08-13 14:10:08 -04003434 return ret;
Josef Bacik7b128762008-07-24 12:17:14 -04003435}
3436
3437/*
3438 * this cleans up any orphans that may be left on the list from the last use
3439 * of this root.
3440 */
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003441int btrfs_orphan_cleanup(struct btrfs_root *root)
Josef Bacik7b128762008-07-24 12:17:14 -04003442{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003443 struct btrfs_fs_info *fs_info = root->fs_info;
Josef Bacik7b128762008-07-24 12:17:14 -04003444 struct btrfs_path *path;
3445 struct extent_buffer *leaf;
Josef Bacik7b128762008-07-24 12:17:14 -04003446 struct btrfs_key key, found_key;
3447 struct btrfs_trans_handle *trans;
3448 struct inode *inode;
Josef Bacik8f6d7f42011-09-26 15:55:20 -04003449 u64 last_objectid = 0;
Josef Bacik7b128762008-07-24 12:17:14 -04003450 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3451
Yan, Zhengd68fc572010-05-16 10:49:58 -04003452 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003453 return 0;
Yan, Zhengc71bf092009-11-12 09:34:40 +00003454
3455 path = btrfs_alloc_path();
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003456 if (!path) {
3457 ret = -ENOMEM;
3458 goto out;
3459 }
David Sterbae4058b52015-11-27 16:31:35 +01003460 path->reada = READA_BACK;
Josef Bacik7b128762008-07-24 12:17:14 -04003461
3462 key.objectid = BTRFS_ORPHAN_OBJECTID;
David Sterba962a2982014-06-04 18:41:45 +02003463 key.type = BTRFS_ORPHAN_ITEM_KEY;
Josef Bacik7b128762008-07-24 12:17:14 -04003464 key.offset = (u64)-1;
3465
Josef Bacik7b128762008-07-24 12:17:14 -04003466 while (1) {
3467 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003468 if (ret < 0)
3469 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04003470
3471 /*
3472 * if ret == 0 means we found what we were searching for, which
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003473 * is weird, but possible, so only screw with path if we didn't
Josef Bacik7b128762008-07-24 12:17:14 -04003474 * find the key and see if we have stuff that matches
3475 */
3476 if (ret > 0) {
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003477 ret = 0;
Josef Bacik7b128762008-07-24 12:17:14 -04003478 if (path->slots[0] == 0)
3479 break;
3480 path->slots[0]--;
3481 }
3482
3483 /* pull out the item */
3484 leaf = path->nodes[0];
Josef Bacik7b128762008-07-24 12:17:14 -04003485 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3486
3487 /* make sure the item matches what we want */
3488 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3489 break;
David Sterba962a2982014-06-04 18:41:45 +02003490 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
Josef Bacik7b128762008-07-24 12:17:14 -04003491 break;
3492
3493 /* release the path since we're done with it */
David Sterbab3b4aa72011-04-21 01:20:15 +02003494 btrfs_release_path(path);
Josef Bacik7b128762008-07-24 12:17:14 -04003495
3496 /*
3497 * this is where we are basically btrfs_lookup, without the
3498 * crossing root thing. we store the inode number in the
3499 * offset of the orphan item.
3500 */
Josef Bacik8f6d7f42011-09-26 15:55:20 -04003501
3502 if (found_key.offset == last_objectid) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003503 btrfs_err(fs_info,
3504 "Error removing orphan entry, stopping orphan cleanup");
Josef Bacik8f6d7f42011-09-26 15:55:20 -04003505 ret = -EINVAL;
3506 goto out;
3507 }
3508
3509 last_objectid = found_key.offset;
3510
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003511 found_key.objectid = found_key.offset;
3512 found_key.type = BTRFS_INODE_ITEM_KEY;
3513 found_key.offset = 0;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003514 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
Rusty Russell8c6ffba2013-07-15 11:20:32 +09303515 ret = PTR_ERR_OR_ZERO(inode);
Filipe Manana67710892016-06-06 11:51:25 +01003516 if (ret && ret != -ENOENT)
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003517 goto out;
Josef Bacika8c9e572011-09-21 16:55:59 -04003518
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003519 if (ret == -ENOENT && root == fs_info->tree_root) {
Arne Jansenf8e9e0b2011-12-14 20:12:02 -05003520 struct btrfs_root *dead_root;
3521 struct btrfs_fs_info *fs_info = root->fs_info;
3522 int is_dead_root = 0;
3523
3524 /*
3525 * this is an orphan in the tree root. Currently these
3526 * could come from 2 sources:
3527 * a) a snapshot deletion in progress
3528 * b) a free space cache inode
3529 * We need to distinguish those two, as the snapshot
3530 * orphan must not get deleted.
3531 * find_dead_roots already ran before us, so if this
3532 * is a snapshot deletion, we should find the root
3533 * in the dead_roots list
3534 */
3535 spin_lock(&fs_info->trans_lock);
3536 list_for_each_entry(dead_root, &fs_info->dead_roots,
3537 root_list) {
3538 if (dead_root->root_key.objectid ==
3539 found_key.objectid) {
3540 is_dead_root = 1;
3541 break;
3542 }
3543 }
3544 spin_unlock(&fs_info->trans_lock);
3545 if (is_dead_root) {
3546 /* prevent this orphan from being found again */
3547 key.offset = found_key.objectid - 1;
3548 continue;
3549 }
3550 }
Josef Bacika8c9e572011-09-21 16:55:59 -04003551 /*
3552 * Inode is already gone but the orphan item is still there,
3553 * kill the orphan item.
3554 */
Filipe Manana67710892016-06-06 11:51:25 +01003555 if (ret == -ENOENT) {
Josef Bacika8c9e572011-09-21 16:55:59 -04003556 trans = btrfs_start_transaction(root, 1);
3557 if (IS_ERR(trans)) {
3558 ret = PTR_ERR(trans);
3559 goto out;
3560 }
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003561 btrfs_debug(fs_info, "auto deleting %Lu",
3562 found_key.objectid);
Josef Bacika8c9e572011-09-21 16:55:59 -04003563 ret = btrfs_del_orphan_item(trans, root,
3564 found_key.objectid);
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04003565 btrfs_end_transaction(trans);
Josef Bacik4ef31a42013-08-13 14:10:08 -04003566 if (ret)
3567 goto out;
Josef Bacika8c9e572011-09-21 16:55:59 -04003568 continue;
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003569 }
Josef Bacik7b128762008-07-24 12:17:14 -04003570
Josef Bacik7b128762008-07-24 12:17:14 -04003571 /*
3572 * add this inode to the orphan list so btrfs_orphan_del does
3573 * the proper thing when we hit it
3574 */
Josef Bacik8a35d952012-05-23 14:26:42 -04003575 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3576 &BTRFS_I(inode)->runtime_flags);
Josef Bacik925396e2013-02-01 15:57:47 -05003577 atomic_inc(&root->orphan_inodes);
Josef Bacik7b128762008-07-24 12:17:14 -04003578
Josef Bacik7b128762008-07-24 12:17:14 -04003579 /* if we have links, this was a truncate, lets do that */
3580 if (inode->i_nlink) {
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05303581 if (WARN_ON(!S_ISREG(inode->i_mode))) {
Josef Bacika41ad392011-01-31 15:30:16 -05003582 iput(inode);
3583 continue;
3584 }
Josef Bacik7b128762008-07-24 12:17:14 -04003585 nr_truncate++;
Josef Bacikf3fe8202013-01-07 17:03:21 -05003586
3587 /* 1 for the orphan item deletion. */
3588 trans = btrfs_start_transaction(root, 1);
3589 if (IS_ERR(trans)) {
Josef Bacikc69b26b2013-06-03 16:51:23 -04003590 iput(inode);
Josef Bacikf3fe8202013-01-07 17:03:21 -05003591 ret = PTR_ERR(trans);
3592 goto out;
3593 }
Nikolay Borisov73f2e542017-02-20 13:50:59 +02003594 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04003595 btrfs_end_transaction(trans);
Josef Bacikc69b26b2013-06-03 16:51:23 -04003596 if (ret) {
3597 iput(inode);
Josef Bacikf3fe8202013-01-07 17:03:21 -05003598 goto out;
Josef Bacikc69b26b2013-06-03 16:51:23 -04003599 }
Josef Bacikf3fe8202013-01-07 17:03:21 -05003600
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003601 ret = btrfs_truncate(inode);
Josef Bacik4a7d0f62013-02-07 16:27:28 -05003602 if (ret)
Nikolay Borisov3d6ae7b2017-02-20 13:50:58 +02003603 btrfs_orphan_del(NULL, BTRFS_I(inode));
Josef Bacik7b128762008-07-24 12:17:14 -04003604 } else {
3605 nr_unlink++;
3606 }
3607
3608 /* this will do delete_inode and everything for us */
3609 iput(inode);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003610 if (ret)
3611 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04003612 }
Miao Xie3254c872011-11-10 20:45:05 -05003613 /* release the path since we're done with it */
3614 btrfs_release_path(path);
3615
Yan, Zhengd68fc572010-05-16 10:49:58 -04003616 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3617
3618 if (root->orphan_block_rsv)
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04003619 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv,
Yan, Zhengd68fc572010-05-16 10:49:58 -04003620 (u64)-1);
3621
Miao Xie27cdeb72014-04-02 19:51:05 +08003622 if (root->orphan_block_rsv ||
3623 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
Josef Bacik7a7eaa42011-04-13 12:54:33 -04003624 trans = btrfs_join_transaction(root);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003625 if (!IS_ERR(trans))
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04003626 btrfs_end_transaction(trans);
Yan, Zhengd68fc572010-05-16 10:49:58 -04003627 }
Josef Bacik7b128762008-07-24 12:17:14 -04003628
3629 if (nr_unlink)
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003630 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
Josef Bacik7b128762008-07-24 12:17:14 -04003631 if (nr_truncate)
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003632 btrfs_debug(fs_info, "truncated %d orphans", nr_truncate);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003633
3634out:
3635 if (ret)
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003636 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003637 btrfs_free_path(path);
3638 return ret;
Josef Bacik7b128762008-07-24 12:17:14 -04003639}
3640
Chris Masond352ac62008-09-29 15:18:18 -04003641/*
Chris Mason46a53cc2009-04-27 11:47:50 -04003642 * very simple check to peek ahead in the leaf looking for xattrs. If we
3643 * don't find any xattrs, we know there can't be any acls.
3644 *
3645 * slot is the slot the inode is in, objectid is the objectid of the inode
3646 */
3647static noinline int acls_after_inode_item(struct extent_buffer *leaf,
Filipe David Borba Manana63541922014-01-07 11:47:46 +00003648 int slot, u64 objectid,
3649 int *first_xattr_slot)
Chris Mason46a53cc2009-04-27 11:47:50 -04003650{
3651 u32 nritems = btrfs_header_nritems(leaf);
3652 struct btrfs_key found_key;
Josef Bacikf23b5a52013-06-19 10:16:26 -04003653 static u64 xattr_access = 0;
3654 static u64 xattr_default = 0;
Chris Mason46a53cc2009-04-27 11:47:50 -04003655 int scanned = 0;
3656
Josef Bacikf23b5a52013-06-19 10:16:26 -04003657 if (!xattr_access) {
Andreas Gruenbacher97d79292015-12-02 14:44:35 +01003658 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3659 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3660 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3661 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
Josef Bacikf23b5a52013-06-19 10:16:26 -04003662 }
3663
Chris Mason46a53cc2009-04-27 11:47:50 -04003664 slot++;
Filipe David Borba Manana63541922014-01-07 11:47:46 +00003665 *first_xattr_slot = -1;
Chris Mason46a53cc2009-04-27 11:47:50 -04003666 while (slot < nritems) {
3667 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3668
3669 /* we found a different objectid, there must not be acls */
3670 if (found_key.objectid != objectid)
3671 return 0;
3672
3673 /* we found an xattr, assume we've got an acl */
Josef Bacikf23b5a52013-06-19 10:16:26 -04003674 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
Filipe David Borba Manana63541922014-01-07 11:47:46 +00003675 if (*first_xattr_slot == -1)
3676 *first_xattr_slot = slot;
Josef Bacikf23b5a52013-06-19 10:16:26 -04003677 if (found_key.offset == xattr_access ||
3678 found_key.offset == xattr_default)
3679 return 1;
3680 }
Chris Mason46a53cc2009-04-27 11:47:50 -04003681
3682 /*
3683 * we found a key greater than an xattr key, there can't
3684 * be any acls later on
3685 */
3686 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3687 return 0;
3688
3689 slot++;
3690 scanned++;
3691
3692 /*
3693 * it goes inode, inode backrefs, xattrs, extents,
3694 * so if there are a ton of hard links to an inode there can
3695 * be a lot of backrefs. Don't waste time searching too hard,
3696 * this is just an optimization
3697 */
3698 if (scanned >= 8)
3699 break;
3700 }
3701 /* we hit the end of the leaf before we found an xattr or
3702 * something larger than an xattr. We have to assume the inode
3703 * has acls
3704 */
Filipe David Borba Manana63541922014-01-07 11:47:46 +00003705 if (*first_xattr_slot == -1)
3706 *first_xattr_slot = slot;
Chris Mason46a53cc2009-04-27 11:47:50 -04003707 return 1;
3708}
3709
3710/*
Chris Masond352ac62008-09-29 15:18:18 -04003711 * read an inode from the btree into the in-memory inode
3712 */
Filipe Manana67710892016-06-06 11:51:25 +01003713static int btrfs_read_locked_inode(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04003714{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003715 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Chris Mason39279cc2007-06-12 06:35:45 -04003716 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04003717 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04003718 struct btrfs_inode_item *inode_item;
3719 struct btrfs_root *root = BTRFS_I(inode)->root;
3720 struct btrfs_key location;
Miao Xie67de1172013-12-26 13:07:06 +08003721 unsigned long ptr;
Chris Mason46a53cc2009-04-27 11:47:50 -04003722 int maybe_acls;
Josef Bacik618e21d2007-07-11 10:18:17 -04003723 u32 rdev;
Chris Mason39279cc2007-06-12 06:35:45 -04003724 int ret;
Miao Xie2f7e33d2011-06-23 07:27:13 +00003725 bool filled = false;
Filipe David Borba Manana63541922014-01-07 11:47:46 +00003726 int first_xattr_slot;
Miao Xie2f7e33d2011-06-23 07:27:13 +00003727
3728 ret = btrfs_fill_inode(inode, &rdev);
3729 if (!ret)
3730 filled = true;
Chris Mason39279cc2007-06-12 06:35:45 -04003731
3732 path = btrfs_alloc_path();
Filipe Manana67710892016-06-06 11:51:25 +01003733 if (!path) {
3734 ret = -ENOMEM;
Mark Fasheh1748f842011-07-12 11:25:31 -07003735 goto make_bad;
Filipe Manana67710892016-06-06 11:51:25 +01003736 }
Mark Fasheh1748f842011-07-12 11:25:31 -07003737
Chris Mason39279cc2007-06-12 06:35:45 -04003738 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
Chris Masondc17ff82008-01-08 15:46:30 -05003739
Chris Mason39279cc2007-06-12 06:35:45 -04003740 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
Filipe Manana67710892016-06-06 11:51:25 +01003741 if (ret) {
3742 if (ret > 0)
3743 ret = -ENOENT;
Chris Mason39279cc2007-06-12 06:35:45 -04003744 goto make_bad;
Filipe Manana67710892016-06-06 11:51:25 +01003745 }
Chris Mason39279cc2007-06-12 06:35:45 -04003746
Chris Mason5f39d392007-10-15 16:14:19 -04003747 leaf = path->nodes[0];
Miao Xie2f7e33d2011-06-23 07:27:13 +00003748
3749 if (filled)
Miao Xie67de1172013-12-26 13:07:06 +08003750 goto cache_index;
Miao Xie2f7e33d2011-06-23 07:27:13 +00003751
Chris Mason5f39d392007-10-15 16:14:19 -04003752 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3753 struct btrfs_inode_item);
Chris Mason5f39d392007-10-15 16:14:19 -04003754 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
Miklos Szeredibfe86842011-10-28 14:13:29 +02003755 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
Eric W. Biederman2f2f43d2012-02-10 11:05:07 -08003756 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3757 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
Nikolay Borisov6ef06d22017-02-20 13:50:34 +02003758 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
Chris Mason5f39d392007-10-15 16:14:19 -04003759
David Sterbaa937b972014-12-12 17:39:12 +01003760 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3761 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
Chris Mason5f39d392007-10-15 16:14:19 -04003762
David Sterbaa937b972014-12-12 17:39:12 +01003763 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3764 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
Chris Mason5f39d392007-10-15 16:14:19 -04003765
David Sterbaa937b972014-12-12 17:39:12 +01003766 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3767 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
Chris Mason5f39d392007-10-15 16:14:19 -04003768
chandan r9cc97d62012-07-04 12:48:07 +05303769 BTRFS_I(inode)->i_otime.tv_sec =
3770 btrfs_timespec_sec(leaf, &inode_item->otime);
3771 BTRFS_I(inode)->i_otime.tv_nsec =
3772 btrfs_timespec_nsec(leaf, &inode_item->otime);
Chris Mason5f39d392007-10-15 16:14:19 -04003773
Yan Zhenga76a3cd2008-10-09 11:46:29 -04003774 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
Chris Masone02119d2008-09-05 16:13:11 -04003775 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
Josef Bacik5dc562c2012-08-17 13:14:17 -04003776 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3777
Josef Bacik0c4d2d92012-04-05 15:03:02 -04003778 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
Chris Masone02119d2008-09-05 16:13:11 -04003779 inode->i_generation = BTRFS_I(inode)->generation;
Josef Bacik618e21d2007-07-11 10:18:17 -04003780 inode->i_rdev = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04003781 rdev = btrfs_inode_rdev(leaf, inode_item);
3782
Josef Bacikaec74772008-07-24 12:12:38 -04003783 BTRFS_I(inode)->index_cnt = (u64)-1;
Yan Zhengd2fb3432008-12-11 16:30:39 -05003784 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
Miao Xie67de1172013-12-26 13:07:06 +08003785
3786cache_index:
Yang Dongsheng6e17d302015-04-09 12:08:43 +08003787 /*
3788 * If we were modified in the current generation and evicted from memory
3789 * and then re-read we need to do a full sync since we don't have any
3790 * idea about which extents were modified before we were evicted from
3791 * cache.
3792 *
3793 * This is required for both inode re-read from disk and delayed inode
3794 * in delayed_nodes_tree.
3795 */
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003796 if (BTRFS_I(inode)->last_trans == fs_info->generation)
Yang Dongsheng6e17d302015-04-09 12:08:43 +08003797 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3798 &BTRFS_I(inode)->runtime_flags);
3799
Filipe Mananabde6c242015-07-24 00:00:19 +01003800 /*
3801 * We don't persist the id of the transaction where an unlink operation
3802 * against the inode was last made. So here we assume the inode might
3803 * have been evicted, and therefore the exact value of last_unlink_trans
3804 * lost, and set it to last_trans to avoid metadata inconsistencies
3805 * between the inode and its parent if the inode is fsync'ed and the log
3806 * replayed. For example, in the scenario:
3807 *
3808 * touch mydir/foo
3809 * ln mydir/foo mydir/bar
3810 * sync
3811 * unlink mydir/bar
3812 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3813 * xfs_io -c fsync mydir/foo
3814 * <power failure>
3815 * mount fs, triggers fsync log replay
3816 *
3817 * We must make sure that when we fsync our inode foo we also log its
3818 * parent inode, otherwise after log replay the parent still has the
3819 * dentry with the "bar" name but our inode foo has a link count of 1
3820 * and doesn't have an inode ref with the name "bar" anymore.
3821 *
3822 * Setting last_unlink_trans to last_trans is a pessimistic approach,
Nicholas D Steeves01327612016-05-19 21:18:45 -04003823 * but it guarantees correctness at the expense of occasional full
Filipe Mananabde6c242015-07-24 00:00:19 +01003824 * transaction commits on fsync if our inode is a directory, or if our
3825 * inode is not a directory, logging its parent unnecessarily.
3826 */
3827 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3828
Miao Xie67de1172013-12-26 13:07:06 +08003829 path->slots[0]++;
3830 if (inode->i_nlink != 1 ||
3831 path->slots[0] >= btrfs_header_nritems(leaf))
3832 goto cache_acl;
3833
3834 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02003835 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
Miao Xie67de1172013-12-26 13:07:06 +08003836 goto cache_acl;
3837
3838 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3839 if (location.type == BTRFS_INODE_REF_KEY) {
3840 struct btrfs_inode_ref *ref;
3841
3842 ref = (struct btrfs_inode_ref *)ptr;
3843 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3844 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3845 struct btrfs_inode_extref *extref;
3846
3847 extref = (struct btrfs_inode_extref *)ptr;
3848 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3849 extref);
3850 }
Miao Xie2f7e33d2011-06-23 07:27:13 +00003851cache_acl:
Chris Mason46a53cc2009-04-27 11:47:50 -04003852 /*
3853 * try to precache a NULL acl entry for files that don't have
3854 * any xattrs or acls
3855 */
Li Zefan33345d012011-04-20 10:31:50 +08003856 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
David Sterbaf85b7372017-01-20 14:54:07 +01003857 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
Filipe David Borba Manana63541922014-01-07 11:47:46 +00003858 if (first_xattr_slot != -1) {
3859 path->slots[0] = first_xattr_slot;
3860 ret = btrfs_load_inode_props(inode, path);
3861 if (ret)
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003862 btrfs_err(fs_info,
David Sterba351fd352014-05-15 16:48:20 +02003863 "error loading props for ino %llu (root %llu): %d",
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02003864 btrfs_ino(BTRFS_I(inode)),
Filipe David Borba Manana63541922014-01-07 11:47:46 +00003865 root->root_key.objectid, ret);
3866 }
3867 btrfs_free_path(path);
3868
Al Viro72c04902009-06-24 16:58:48 -04003869 if (!maybe_acls)
3870 cache_no_acl(inode);
Chris Mason46a53cc2009-04-27 11:47:50 -04003871
Chris Mason39279cc2007-06-12 06:35:45 -04003872 switch (inode->i_mode & S_IFMT) {
Chris Mason39279cc2007-06-12 06:35:45 -04003873 case S_IFREG:
3874 inode->i_mapping->a_ops = &btrfs_aops;
Chris Masond1310b22008-01-24 16:13:08 -05003875 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -04003876 inode->i_fop = &btrfs_file_operations;
3877 inode->i_op = &btrfs_file_inode_operations;
3878 break;
3879 case S_IFDIR:
3880 inode->i_fop = &btrfs_dir_file_operations;
Omar Sandoval67ade052017-01-25 17:06:38 -08003881 inode->i_op = &btrfs_dir_inode_operations;
Chris Mason39279cc2007-06-12 06:35:45 -04003882 break;
3883 case S_IFLNK:
3884 inode->i_op = &btrfs_symlink_inode_operations;
Al Viro21fc61c2015-11-17 01:07:57 -05003885 inode_nohighmem(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04003886 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3887 break;
Josef Bacik618e21d2007-07-11 10:18:17 -04003888 default:
Jim Owens0279b4c2009-02-04 09:29:13 -05003889 inode->i_op = &btrfs_special_inode_operations;
Josef Bacik618e21d2007-07-11 10:18:17 -04003890 init_special_inode(inode, inode->i_mode, rdev);
3891 break;
Chris Mason39279cc2007-06-12 06:35:45 -04003892 }
Christoph Hellwig6cbff002009-04-17 10:37:41 +02003893
3894 btrfs_update_iflags(inode);
Filipe Manana67710892016-06-06 11:51:25 +01003895 return 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003896
3897make_bad:
Chris Mason39279cc2007-06-12 06:35:45 -04003898 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04003899 make_bad_inode(inode);
Filipe Manana67710892016-06-06 11:51:25 +01003900 return ret;
Chris Mason39279cc2007-06-12 06:35:45 -04003901}
3902
Chris Masond352ac62008-09-29 15:18:18 -04003903/*
3904 * given a leaf and an inode, copy the inode fields into the leaf
3905 */
Chris Masone02119d2008-09-05 16:13:11 -04003906static void fill_inode_item(struct btrfs_trans_handle *trans,
3907 struct extent_buffer *leaf,
Chris Mason5f39d392007-10-15 16:14:19 -04003908 struct btrfs_inode_item *item,
Chris Mason39279cc2007-06-12 06:35:45 -04003909 struct inode *inode)
3910{
Liu Bo51fab692012-12-27 09:01:21 +00003911 struct btrfs_map_token token;
Chris Mason5f39d392007-10-15 16:14:19 -04003912
Liu Bo51fab692012-12-27 09:01:21 +00003913 btrfs_init_map_token(&token);
Chris Mason5f39d392007-10-15 16:14:19 -04003914
Liu Bo51fab692012-12-27 09:01:21 +00003915 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3916 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3917 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3918 &token);
3919 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3920 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
Chris Mason5f39d392007-10-15 16:14:19 -04003921
David Sterbaa937b972014-12-12 17:39:12 +01003922 btrfs_set_token_timespec_sec(leaf, &item->atime,
Liu Bo51fab692012-12-27 09:01:21 +00003923 inode->i_atime.tv_sec, &token);
David Sterbaa937b972014-12-12 17:39:12 +01003924 btrfs_set_token_timespec_nsec(leaf, &item->atime,
Liu Bo51fab692012-12-27 09:01:21 +00003925 inode->i_atime.tv_nsec, &token);
Chris Mason5f39d392007-10-15 16:14:19 -04003926
David Sterbaa937b972014-12-12 17:39:12 +01003927 btrfs_set_token_timespec_sec(leaf, &item->mtime,
Liu Bo51fab692012-12-27 09:01:21 +00003928 inode->i_mtime.tv_sec, &token);
David Sterbaa937b972014-12-12 17:39:12 +01003929 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
Liu Bo51fab692012-12-27 09:01:21 +00003930 inode->i_mtime.tv_nsec, &token);
3931
David Sterbaa937b972014-12-12 17:39:12 +01003932 btrfs_set_token_timespec_sec(leaf, &item->ctime,
Liu Bo51fab692012-12-27 09:01:21 +00003933 inode->i_ctime.tv_sec, &token);
David Sterbaa937b972014-12-12 17:39:12 +01003934 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
Liu Bo51fab692012-12-27 09:01:21 +00003935 inode->i_ctime.tv_nsec, &token);
3936
chandan r9cc97d62012-07-04 12:48:07 +05303937 btrfs_set_token_timespec_sec(leaf, &item->otime,
3938 BTRFS_I(inode)->i_otime.tv_sec, &token);
3939 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3940 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3941
Liu Bo51fab692012-12-27 09:01:21 +00003942 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3943 &token);
3944 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3945 &token);
3946 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3947 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3948 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3949 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3950 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
Chris Mason39279cc2007-06-12 06:35:45 -04003951}
3952
Chris Masond352ac62008-09-29 15:18:18 -04003953/*
3954 * copy everything in the in-memory inode into the btree.
3955 */
Chris Mason21151332011-11-10 20:39:08 -05003956static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
Chris Masond3977122009-01-05 21:25:51 -05003957 struct btrfs_root *root, struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04003958{
3959 struct btrfs_inode_item *inode_item;
3960 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04003961 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04003962 int ret;
3963
3964 path = btrfs_alloc_path();
Miao Xie16cdcec2011-04-22 18:12:22 +08003965 if (!path)
3966 return -ENOMEM;
3967
Chris Masonb9473432009-03-13 11:00:37 -04003968 path->leave_spinning = 1;
Miao Xie16cdcec2011-04-22 18:12:22 +08003969 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3970 1);
Chris Mason39279cc2007-06-12 06:35:45 -04003971 if (ret) {
3972 if (ret > 0)
3973 ret = -ENOENT;
3974 goto failed;
3975 }
3976
Chris Mason5f39d392007-10-15 16:14:19 -04003977 leaf = path->nodes[0];
3978 inode_item = btrfs_item_ptr(leaf, path->slots[0],
Miao Xie16cdcec2011-04-22 18:12:22 +08003979 struct btrfs_inode_item);
Chris Mason39279cc2007-06-12 06:35:45 -04003980
Chris Masone02119d2008-09-05 16:13:11 -04003981 fill_inode_item(trans, leaf, inode_item, inode);
Chris Mason5f39d392007-10-15 16:14:19 -04003982 btrfs_mark_buffer_dirty(leaf);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04003983 btrfs_set_inode_last_trans(trans, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04003984 ret = 0;
3985failed:
Chris Mason39279cc2007-06-12 06:35:45 -04003986 btrfs_free_path(path);
3987 return ret;
3988}
3989
Chris Masond352ac62008-09-29 15:18:18 -04003990/*
Chris Mason21151332011-11-10 20:39:08 -05003991 * copy everything in the in-memory inode into the btree.
3992 */
3993noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3994 struct btrfs_root *root, struct inode *inode)
3995{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003996 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason21151332011-11-10 20:39:08 -05003997 int ret;
3998
3999 /*
4000 * If the inode is a free space inode, we can deadlock during commit
4001 * if we put it into the delayed code.
4002 *
4003 * The data relocation inode should also be directly updated
4004 * without delay
4005 */
Nikolay Borisov70ddc552017-02-20 13:50:35 +02004006 if (!btrfs_is_free_space_inode(BTRFS_I(inode))
Josef Bacik1d52c782014-09-18 11:30:44 -04004007 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004008 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
Alexander Block8ea05e32012-07-25 17:35:53 +02004009 btrfs_update_root_times(trans, root);
4010
Chris Mason21151332011-11-10 20:39:08 -05004011 ret = btrfs_delayed_update_inode(trans, root, inode);
4012 if (!ret)
4013 btrfs_set_inode_last_trans(trans, inode);
4014 return ret;
4015 }
4016
4017 return btrfs_update_inode_item(trans, root, inode);
4018}
4019
Josef Bacikbe6aef62012-10-22 15:43:12 -04004020noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4021 struct btrfs_root *root,
4022 struct inode *inode)
Chris Mason21151332011-11-10 20:39:08 -05004023{
4024 int ret;
4025
4026 ret = btrfs_update_inode(trans, root, inode);
4027 if (ret == -ENOSPC)
4028 return btrfs_update_inode_item(trans, root, inode);
4029 return ret;
4030}
4031
4032/*
Chris Masond352ac62008-09-29 15:18:18 -04004033 * unlink helper that gets used here in inode.c and in the tree logging
4034 * recovery code. It remove a link in a directory with a given name, and
4035 * also drops the back refs in the inode to the directory
4036 */
Al Viro92986792011-03-04 17:14:37 +00004037static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4038 struct btrfs_root *root,
Nikolay Borisov4ec59342017-01-18 00:31:44 +02004039 struct btrfs_inode *dir,
4040 struct btrfs_inode *inode,
Al Viro92986792011-03-04 17:14:37 +00004041 const char *name, int name_len)
Chris Mason39279cc2007-06-12 06:35:45 -04004042{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004043 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason39279cc2007-06-12 06:35:45 -04004044 struct btrfs_path *path;
Chris Mason39279cc2007-06-12 06:35:45 -04004045 int ret = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04004046 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04004047 struct btrfs_dir_item *di;
Chris Mason5f39d392007-10-15 16:14:19 -04004048 struct btrfs_key key;
Josef Bacikaec74772008-07-24 12:12:38 -04004049 u64 index;
Li Zefan33345d012011-04-20 10:31:50 +08004050 u64 ino = btrfs_ino(inode);
4051 u64 dir_ino = btrfs_ino(dir);
Chris Mason39279cc2007-06-12 06:35:45 -04004052
4053 path = btrfs_alloc_path();
Chris Mason54aa1f42007-06-22 14:16:25 -04004054 if (!path) {
4055 ret = -ENOMEM;
Tsutomu Itoh554233a2011-02-03 03:16:25 +00004056 goto out;
Chris Mason54aa1f42007-06-22 14:16:25 -04004057 }
4058
Chris Masonb9473432009-03-13 11:00:37 -04004059 path->leave_spinning = 1;
Li Zefan33345d012011-04-20 10:31:50 +08004060 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
Chris Mason39279cc2007-06-12 06:35:45 -04004061 name, name_len, -1);
4062 if (IS_ERR(di)) {
4063 ret = PTR_ERR(di);
4064 goto err;
4065 }
4066 if (!di) {
4067 ret = -ENOENT;
4068 goto err;
4069 }
Chris Mason5f39d392007-10-15 16:14:19 -04004070 leaf = path->nodes[0];
4071 btrfs_dir_item_key_to_cpu(leaf, di, &key);
Chris Mason39279cc2007-06-12 06:35:45 -04004072 ret = btrfs_delete_one_dir_name(trans, root, path, di);
Chris Mason54aa1f42007-06-22 14:16:25 -04004073 if (ret)
4074 goto err;
David Sterbab3b4aa72011-04-21 01:20:15 +02004075 btrfs_release_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04004076
Miao Xie67de1172013-12-26 13:07:06 +08004077 /*
4078 * If we don't have dir index, we have to get it by looking up
4079 * the inode ref, since we get the inode ref, remove it directly,
4080 * it is unnecessary to do delayed deletion.
4081 *
4082 * But if we have dir index, needn't search inode ref to get it.
4083 * Since the inode ref is close to the inode item, it is better
4084 * that we delay to delete it, and just do this deletion when
4085 * we update the inode item.
4086 */
Nikolay Borisov4ec59342017-01-18 00:31:44 +02004087 if (inode->dir_index) {
Miao Xie67de1172013-12-26 13:07:06 +08004088 ret = btrfs_delayed_delete_inode_ref(inode);
4089 if (!ret) {
Nikolay Borisov4ec59342017-01-18 00:31:44 +02004090 index = inode->dir_index;
Miao Xie67de1172013-12-26 13:07:06 +08004091 goto skip_backref;
4092 }
4093 }
4094
Li Zefan33345d012011-04-20 10:31:50 +08004095 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4096 dir_ino, &index);
Josef Bacikaec74772008-07-24 12:12:38 -04004097 if (ret) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004098 btrfs_info(fs_info,
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00004099 "failed to delete reference to %.*s, inode %llu parent %llu",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02004100 name_len, name, ino, dir_ino);
Jeff Mahoney66642832016-06-10 18:19:25 -04004101 btrfs_abort_transaction(trans, ret);
Josef Bacikaec74772008-07-24 12:12:38 -04004102 goto err;
4103 }
Miao Xie67de1172013-12-26 13:07:06 +08004104skip_backref:
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04004105 ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004106 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04004107 btrfs_abort_transaction(trans, ret);
Chris Mason39279cc2007-06-12 06:35:45 -04004108 goto err;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004109 }
Chris Mason39279cc2007-06-12 06:35:45 -04004110
Nikolay Borisov4ec59342017-01-18 00:31:44 +02004111 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4112 dir_ino);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004113 if (ret != 0 && ret != -ENOENT) {
Jeff Mahoney66642832016-06-10 18:19:25 -04004114 btrfs_abort_transaction(trans, ret);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004115 goto err;
4116 }
Chris Masone02119d2008-09-05 16:13:11 -04004117
Nikolay Borisov4ec59342017-01-18 00:31:44 +02004118 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4119 index);
Chris Mason6418c962010-10-30 07:34:24 -04004120 if (ret == -ENOENT)
4121 ret = 0;
Zach Brownd4e39912013-04-02 21:02:16 +00004122 else if (ret)
Jeff Mahoney66642832016-06-10 18:19:25 -04004123 btrfs_abort_transaction(trans, ret);
Chris Mason39279cc2007-06-12 06:35:45 -04004124err:
4125 btrfs_free_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04004126 if (ret)
4127 goto out;
4128
Nikolay Borisov6ef06d22017-02-20 13:50:34 +02004129 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
Nikolay Borisov4ec59342017-01-18 00:31:44 +02004130 inode_inc_iversion(&inode->vfs_inode);
4131 inode_inc_iversion(&dir->vfs_inode);
4132 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4133 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4134 ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
Chris Masone02119d2008-09-05 16:13:11 -04004135out:
Chris Mason39279cc2007-06-12 06:35:45 -04004136 return ret;
4137}
4138
Al Viro92986792011-03-04 17:14:37 +00004139int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4140 struct btrfs_root *root,
Nikolay Borisov4ec59342017-01-18 00:31:44 +02004141 struct btrfs_inode *dir, struct btrfs_inode *inode,
Al Viro92986792011-03-04 17:14:37 +00004142 const char *name, int name_len)
4143{
4144 int ret;
4145 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4146 if (!ret) {
Nikolay Borisov4ec59342017-01-18 00:31:44 +02004147 drop_nlink(&inode->vfs_inode);
4148 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
Al Viro92986792011-03-04 17:14:37 +00004149 }
4150 return ret;
4151}
Yan, Zhenga22285a2010-05-16 10:48:46 -04004152
4153/*
4154 * helper to start transaction for unlink and rmdir.
4155 *
Josef Bacikd52be812013-05-29 14:54:47 -04004156 * unlink and rmdir are special in btrfs, they do not always free space, so
4157 * if we cannot make our reservations the normal way try and see if there is
4158 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4159 * allow the unlink to occur.
Yan, Zhenga22285a2010-05-16 10:48:46 -04004160 */
Josef Bacikd52be812013-05-29 14:54:47 -04004161static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
Yan, Zhenga22285a2010-05-16 10:48:46 -04004162{
Yan, Zhenga22285a2010-05-16 10:48:46 -04004163 struct btrfs_root *root = BTRFS_I(dir)->root;
Yan, Zhenga22285a2010-05-16 10:48:46 -04004164
Josef Bacike70bea52011-10-11 14:18:24 -04004165 /*
4166 * 1 for the possible orphan item
4167 * 1 for the dir item
4168 * 1 for the dir index
4169 * 1 for the inode ref
Josef Bacike70bea52011-10-11 14:18:24 -04004170 * 1 for the inode
4171 */
Filipe Manana8eab77f2015-11-13 23:57:16 +00004172 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
Yan, Zhenga22285a2010-05-16 10:48:46 -04004173}
4174
Chris Mason39279cc2007-06-12 06:35:45 -04004175static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4176{
Yan, Zhenga22285a2010-05-16 10:48:46 -04004177 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04004178 struct btrfs_trans_handle *trans;
David Howells2b0143b2015-03-17 22:25:59 +00004179 struct inode *inode = d_inode(dentry);
Chris Mason39279cc2007-06-12 06:35:45 -04004180 int ret;
4181
Josef Bacikd52be812013-05-29 14:54:47 -04004182 trans = __unlink_start_trans(dir);
Yan, Zhenga22285a2010-05-16 10:48:46 -04004183 if (IS_ERR(trans))
Josef Bacik5df6a9f2009-11-10 21:23:48 -05004184 return PTR_ERR(trans);
Chris Mason5f39d392007-10-15 16:14:19 -04004185
Nikolay Borisov4ec59342017-01-18 00:31:44 +02004186 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4187 0);
Chris Mason12fcfd22009-03-24 10:24:20 -04004188
Nikolay Borisov4ec59342017-01-18 00:31:44 +02004189 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4190 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4191 dentry->d_name.len);
Tsutomu Itohb5324022011-07-19 07:27:20 +00004192 if (ret)
4193 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04004194
Yan, Zhenga22285a2010-05-16 10:48:46 -04004195 if (inode->i_nlink == 0) {
Nikolay Borisov73f2e542017-02-20 13:50:59 +02004196 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
Tsutomu Itohb5324022011-07-19 07:27:20 +00004197 if (ret)
4198 goto out;
Yan, Zhenga22285a2010-05-16 10:48:46 -04004199 }
Josef Bacik7b128762008-07-24 12:17:14 -04004200
Tsutomu Itohb5324022011-07-19 07:27:20 +00004201out:
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04004202 btrfs_end_transaction(trans);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04004203 btrfs_btree_balance_dirty(root->fs_info);
Chris Mason39279cc2007-06-12 06:35:45 -04004204 return ret;
4205}
4206
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004207int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4208 struct btrfs_root *root,
4209 struct inode *dir, u64 objectid,
4210 const char *name, int name_len)
4211{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004212 struct btrfs_fs_info *fs_info = root->fs_info;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004213 struct btrfs_path *path;
4214 struct extent_buffer *leaf;
4215 struct btrfs_dir_item *di;
4216 struct btrfs_key key;
4217 u64 index;
4218 int ret;
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02004219 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004220
4221 path = btrfs_alloc_path();
4222 if (!path)
4223 return -ENOMEM;
4224
Li Zefan33345d012011-04-20 10:31:50 +08004225 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004226 name, name_len, -1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004227 if (IS_ERR_OR_NULL(di)) {
4228 if (!di)
4229 ret = -ENOENT;
4230 else
4231 ret = PTR_ERR(di);
4232 goto out;
4233 }
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004234
4235 leaf = path->nodes[0];
4236 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4237 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4238 ret = btrfs_delete_one_dir_name(trans, root, path, di);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004239 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04004240 btrfs_abort_transaction(trans, ret);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004241 goto out;
4242 }
David Sterbab3b4aa72011-04-21 01:20:15 +02004243 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004244
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004245 ret = btrfs_del_root_ref(trans, fs_info, objectid,
4246 root->root_key.objectid, dir_ino,
4247 &index, name, name_len);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004248 if (ret < 0) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004249 if (ret != -ENOENT) {
Jeff Mahoney66642832016-06-10 18:19:25 -04004250 btrfs_abort_transaction(trans, ret);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004251 goto out;
4252 }
Li Zefan33345d012011-04-20 10:31:50 +08004253 di = btrfs_search_dir_index_item(root, path, dir_ino,
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004254 name, name_len);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004255 if (IS_ERR_OR_NULL(di)) {
4256 if (!di)
4257 ret = -ENOENT;
4258 else
4259 ret = PTR_ERR(di);
Jeff Mahoney66642832016-06-10 18:19:25 -04004260 btrfs_abort_transaction(trans, ret);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004261 goto out;
4262 }
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004263
4264 leaf = path->nodes[0];
4265 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02004266 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004267 index = key.offset;
4268 }
David Sterbab3b4aa72011-04-21 01:20:15 +02004269 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004270
Nikolay Borisove67bbbb2017-01-10 20:35:36 +02004271 ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004272 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04004273 btrfs_abort_transaction(trans, ret);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004274 goto out;
4275 }
Miao Xie16cdcec2011-04-22 18:12:22 +08004276
Nikolay Borisov6ef06d22017-02-20 13:50:34 +02004277 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
Josef Bacik0c4d2d92012-04-05 15:03:02 -04004278 inode_inc_iversion(dir);
Deepa Dinamanic2050a42016-09-14 07:48:06 -07004279 dir->i_mtime = dir->i_ctime = current_time(dir);
Josef Bacik5a24e842012-08-08 10:12:59 -06004280 ret = btrfs_update_inode_fallback(trans, root, dir);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004281 if (ret)
Jeff Mahoney66642832016-06-10 18:19:25 -04004282 btrfs_abort_transaction(trans, ret);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004283out:
Josef Bacik71d7aed2011-06-14 14:24:32 -04004284 btrfs_free_path(path);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004285 return ret;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004286}
4287
Chris Mason39279cc2007-06-12 06:35:45 -04004288static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4289{
David Howells2b0143b2015-03-17 22:25:59 +00004290 struct inode *inode = d_inode(dentry);
Chris Mason1832a6d2007-12-21 16:27:21 -05004291 int err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004292 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04004293 struct btrfs_trans_handle *trans;
Filipe Manana44f714d2016-06-06 16:11:13 +01004294 u64 last_unlink_trans;
Chris Mason39279cc2007-06-12 06:35:45 -04004295
David Sterbab3ae2442012-09-13 16:04:34 -06004296 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
Yan134d4512007-10-25 15:49:25 -04004297 return -ENOTEMPTY;
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02004298 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
David Sterbab3ae2442012-09-13 16:04:34 -06004299 return -EPERM;
Yan134d4512007-10-25 15:49:25 -04004300
Josef Bacikd52be812013-05-29 14:54:47 -04004301 trans = __unlink_start_trans(dir);
Yan, Zhenga22285a2010-05-16 10:48:46 -04004302 if (IS_ERR(trans))
Josef Bacik5df6a9f2009-11-10 21:23:48 -05004303 return PTR_ERR(trans);
Josef Bacik5df6a9f2009-11-10 21:23:48 -05004304
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02004305 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004306 err = btrfs_unlink_subvol(trans, root, dir,
4307 BTRFS_I(inode)->location.objectid,
4308 dentry->d_name.name,
4309 dentry->d_name.len);
4310 goto out;
4311 }
4312
Nikolay Borisov73f2e542017-02-20 13:50:59 +02004313 err = btrfs_orphan_add(trans, BTRFS_I(inode));
Josef Bacik7b128762008-07-24 12:17:14 -04004314 if (err)
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004315 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04004316
Filipe Manana44f714d2016-06-06 16:11:13 +01004317 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4318
Chris Mason39279cc2007-06-12 06:35:45 -04004319 /* now the directory is empty */
Nikolay Borisov4ec59342017-01-18 00:31:44 +02004320 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4321 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4322 dentry->d_name.len);
Filipe Manana44f714d2016-06-06 16:11:13 +01004323 if (!err) {
Nikolay Borisov6ef06d22017-02-20 13:50:34 +02004324 btrfs_i_size_write(BTRFS_I(inode), 0);
Filipe Manana44f714d2016-06-06 16:11:13 +01004325 /*
4326 * Propagate the last_unlink_trans value of the deleted dir to
4327 * its parent directory. This is to prevent an unrecoverable
4328 * log tree in the case we do something like this:
4329 * 1) create dir foo
4330 * 2) create snapshot under dir foo
4331 * 3) delete the snapshot
4332 * 4) rmdir foo
4333 * 5) mkdir foo
4334 * 6) fsync foo or some file inside foo
4335 */
4336 if (last_unlink_trans >= trans->transid)
4337 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4338 }
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004339out:
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04004340 btrfs_end_transaction(trans);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04004341 btrfs_btree_balance_dirty(root->fs_info);
Chris Mason39544012007-12-12 14:38:19 -05004342
Chris Mason39279cc2007-06-12 06:35:45 -04004343 return err;
4344}
4345
Chris Mason28f75a02015-02-04 06:59:29 -08004346static int truncate_space_check(struct btrfs_trans_handle *trans,
4347 struct btrfs_root *root,
4348 u64 bytes_deleted)
4349{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004350 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason28f75a02015-02-04 06:59:29 -08004351 int ret;
4352
Josef Bacikdc95f7b2016-01-13 11:48:06 -05004353 /*
4354 * This is only used to apply pressure to the enospc system, we don't
4355 * intend to use this reservation at all.
4356 */
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04004357 bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004358 bytes_deleted *= fs_info->nodesize;
4359 ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
Chris Mason28f75a02015-02-04 06:59:29 -08004360 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
Josef Bacikdc95f7b2016-01-13 11:48:06 -05004361 if (!ret) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004362 trace_btrfs_space_reservation(fs_info, "transaction",
Josef Bacikdc95f7b2016-01-13 11:48:06 -05004363 trans->transid,
4364 bytes_deleted, 1);
Chris Mason28f75a02015-02-04 06:59:29 -08004365 trans->bytes_reserved += bytes_deleted;
Josef Bacikdc95f7b2016-01-13 11:48:06 -05004366 }
Chris Mason28f75a02015-02-04 06:59:29 -08004367 return ret;
4368
4369}
4370
Josef Bacikddfae632017-10-19 14:16:02 -04004371/*
4372 * Return this if we need to call truncate_block for the last bit of the
4373 * truncate.
4374 */
4375#define NEED_TRUNCATE_BLOCK 1
Filipe Manana0305cd52015-10-16 12:34:25 +01004376
Chris Mason323ac952008-10-01 19:05:46 -04004377/*
Chris Mason39279cc2007-06-12 06:35:45 -04004378 * this can truncate away extent items, csum items and directory items.
4379 * It starts at a high offset and removes keys until it can't find
Chris Masond352ac62008-09-29 15:18:18 -04004380 * any higher than new_size
Chris Mason39279cc2007-06-12 06:35:45 -04004381 *
4382 * csum items that cross the new i_size are truncated to the new size
4383 * as well.
Josef Bacik7b128762008-07-24 12:17:14 -04004384 *
4385 * min_type is the minimum key type to truncate down to. If set to 0, this
4386 * will kill all the items on this inode, including the INODE_ITEM_KEY.
Chris Mason39279cc2007-06-12 06:35:45 -04004387 */
Yan, Zheng80825102009-11-12 09:35:36 +00004388int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4389 struct btrfs_root *root,
4390 struct inode *inode,
4391 u64 new_size, u32 min_type)
Chris Mason39279cc2007-06-12 06:35:45 -04004392{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004393 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason39279cc2007-06-12 06:35:45 -04004394 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04004395 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04004396 struct btrfs_file_extent_item *fi;
Yan, Zheng80825102009-11-12 09:35:36 +00004397 struct btrfs_key key;
4398 struct btrfs_key found_key;
Chris Mason39279cc2007-06-12 06:35:45 -04004399 u64 extent_start = 0;
Chris Masondb945352007-10-15 16:15:53 -04004400 u64 extent_num_bytes = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004401 u64 extent_offset = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004402 u64 item_end = 0;
Filipe Mananac1aa4572015-06-20 18:20:09 +01004403 u64 last_size = new_size;
Yan, Zheng80825102009-11-12 09:35:36 +00004404 u32 found_type = (u8)-1;
Chris Mason39279cc2007-06-12 06:35:45 -04004405 int found_extent;
4406 int del_item;
Chris Mason85e21ba2008-01-29 15:11:36 -05004407 int pending_del_nr = 0;
4408 int pending_del_slot = 0;
Chris Mason179e29e2007-11-01 11:28:41 -04004409 int extent_type = -1;
Yan, Zheng80825102009-11-12 09:35:36 +00004410 int ret;
4411 int err = 0;
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02004412 u64 ino = btrfs_ino(BTRFS_I(inode));
Chris Mason28ed1342014-12-17 09:41:04 -08004413 u64 bytes_deleted = 0;
Thomas Meyer897ca812017-10-07 16:02:21 +02004414 bool be_nice = false;
4415 bool should_throttle = false;
4416 bool should_end = false;
Yan, Zheng80825102009-11-12 09:35:36 +00004417
4418 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
Chris Mason39279cc2007-06-12 06:35:45 -04004419
Chris Mason28ed1342014-12-17 09:41:04 -08004420 /*
4421 * for non-free space inodes and ref cows, we want to back off from
4422 * time to time
4423 */
Nikolay Borisov70ddc552017-02-20 13:50:35 +02004424 if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
Chris Mason28ed1342014-12-17 09:41:04 -08004425 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
Thomas Meyer897ca812017-10-07 16:02:21 +02004426 be_nice = true;
Chris Mason28ed1342014-12-17 09:41:04 -08004427
Mark Fasheh0eb0e192011-07-12 16:44:10 -07004428 path = btrfs_alloc_path();
4429 if (!path)
4430 return -ENOMEM;
David Sterbae4058b52015-11-27 16:31:35 +01004431 path->reada = READA_BACK;
Mark Fasheh0eb0e192011-07-12 16:44:10 -07004432
Josef Bacik5dc562c2012-08-17 13:14:17 -04004433 /*
4434 * We want to drop from the next block forward in case this new size is
4435 * not block aligned since we will be keeping the last block of the
4436 * extent just the way it is.
4437 */
Miao Xie27cdeb72014-04-02 19:51:05 +08004438 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004439 root == fs_info->tree_root)
Nikolay Borisovdcdbc052017-02-20 13:50:45 +02004440 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004441 fs_info->sectorsize),
Jeff Mahoneyda170662016-06-15 09:22:56 -04004442 (u64)-1, 0);
Yan, Zheng80825102009-11-12 09:35:36 +00004443
Miao Xie16cdcec2011-04-22 18:12:22 +08004444 /*
4445 * This function is also used to drop the items in the log tree before
4446 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4447 * it is used to drop the loged items. So we shouldn't kill the delayed
4448 * items.
4449 */
4450 if (min_type == 0 && root == BTRFS_I(inode)->root)
Nikolay Borisov4ccb5c72017-01-10 20:35:38 +02004451 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
Miao Xie16cdcec2011-04-22 18:12:22 +08004452
Li Zefan33345d012011-04-20 10:31:50 +08004453 key.objectid = ino;
Chris Mason39279cc2007-06-12 06:35:45 -04004454 key.offset = (u64)-1;
Chris Mason5f39d392007-10-15 16:14:19 -04004455 key.type = (u8)-1;
4456
Chris Mason85e21ba2008-01-29 15:11:36 -05004457search_again:
Chris Mason28ed1342014-12-17 09:41:04 -08004458 /*
4459 * with a 16K leaf size and 128MB extents, you can actually queue
4460 * up a huge file in a single leaf. Most of the time that
4461 * bytes_deleted is > 0, it will be huge by the time we get here
4462 */
Byongho Leeee221842015-12-15 01:42:10 +09004463 if (be_nice && bytes_deleted > SZ_32M) {
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04004464 if (btrfs_should_end_transaction(trans)) {
Chris Mason28ed1342014-12-17 09:41:04 -08004465 err = -EAGAIN;
4466 goto error;
4467 }
4468 }
4469
4470
Chris Masonb9473432009-03-13 11:00:37 -04004471 path->leave_spinning = 1;
Chris Mason85e21ba2008-01-29 15:11:36 -05004472 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
Yan, Zheng80825102009-11-12 09:35:36 +00004473 if (ret < 0) {
4474 err = ret;
4475 goto out;
4476 }
Chris Masond3977122009-01-05 21:25:51 -05004477
Chris Mason85e21ba2008-01-29 15:11:36 -05004478 if (ret > 0) {
Chris Masone02119d2008-09-05 16:13:11 -04004479 /* there are no items in the tree for us to truncate, we're
4480 * done
4481 */
Yan, Zheng80825102009-11-12 09:35:36 +00004482 if (path->slots[0] == 0)
4483 goto out;
Chris Mason85e21ba2008-01-29 15:11:36 -05004484 path->slots[0]--;
4485 }
4486
Chris Masond3977122009-01-05 21:25:51 -05004487 while (1) {
Chris Mason39279cc2007-06-12 06:35:45 -04004488 fi = NULL;
Chris Mason5f39d392007-10-15 16:14:19 -04004489 leaf = path->nodes[0];
4490 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
David Sterba962a2982014-06-04 18:41:45 +02004491 found_type = found_key.type;
Chris Mason39279cc2007-06-12 06:35:45 -04004492
Li Zefan33345d012011-04-20 10:31:50 +08004493 if (found_key.objectid != ino)
Chris Mason39279cc2007-06-12 06:35:45 -04004494 break;
Chris Mason5f39d392007-10-15 16:14:19 -04004495
Chris Mason85e21ba2008-01-29 15:11:36 -05004496 if (found_type < min_type)
Chris Mason39279cc2007-06-12 06:35:45 -04004497 break;
4498
Chris Mason5f39d392007-10-15 16:14:19 -04004499 item_end = found_key.offset;
Chris Mason39279cc2007-06-12 06:35:45 -04004500 if (found_type == BTRFS_EXTENT_DATA_KEY) {
Chris Mason5f39d392007-10-15 16:14:19 -04004501 fi = btrfs_item_ptr(leaf, path->slots[0],
Chris Mason39279cc2007-06-12 06:35:45 -04004502 struct btrfs_file_extent_item);
Chris Mason179e29e2007-11-01 11:28:41 -04004503 extent_type = btrfs_file_extent_type(leaf, fi);
4504 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
Chris Mason5f39d392007-10-15 16:14:19 -04004505 item_end +=
Chris Masondb945352007-10-15 16:15:53 -04004506 btrfs_file_extent_num_bytes(leaf, fi);
Liu Bo09ed2f12017-03-10 11:09:48 -08004507
4508 trace_btrfs_truncate_show_fi_regular(
4509 BTRFS_I(inode), leaf, fi,
4510 found_key.offset);
Chris Mason179e29e2007-11-01 11:28:41 -04004511 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason179e29e2007-11-01 11:28:41 -04004512 item_end += btrfs_file_extent_inline_len(leaf,
Chris Mason514ac8a2014-01-03 21:07:00 -08004513 path->slots[0], fi);
Liu Bo09ed2f12017-03-10 11:09:48 -08004514
4515 trace_btrfs_truncate_show_fi_inline(
4516 BTRFS_I(inode), leaf, fi, path->slots[0],
4517 found_key.offset);
Chris Mason39279cc2007-06-12 06:35:45 -04004518 }
Yan008630c2007-11-07 13:31:09 -05004519 item_end--;
Chris Mason39279cc2007-06-12 06:35:45 -04004520 }
Yan, Zheng80825102009-11-12 09:35:36 +00004521 if (found_type > min_type) {
Chris Mason39279cc2007-06-12 06:35:45 -04004522 del_item = 1;
Yan, Zheng80825102009-11-12 09:35:36 +00004523 } else {
Filipe Manana76b42ab2017-02-14 16:56:01 +00004524 if (item_end < new_size)
Yan, Zheng80825102009-11-12 09:35:36 +00004525 break;
4526 if (found_key.offset >= new_size)
4527 del_item = 1;
4528 else
4529 del_item = 0;
4530 }
Chris Mason39279cc2007-06-12 06:35:45 -04004531 found_extent = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004532 /* FIXME, shrink the extent if the ref count is only 1 */
Chris Mason179e29e2007-11-01 11:28:41 -04004533 if (found_type != BTRFS_EXTENT_DATA_KEY)
4534 goto delete;
4535
4536 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
Chris Mason39279cc2007-06-12 06:35:45 -04004537 u64 num_dec;
Chris Masondb945352007-10-15 16:15:53 -04004538 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
Miao Xief70a9a6b2012-01-12 19:10:12 -05004539 if (!del_item) {
Chris Masondb945352007-10-15 16:15:53 -04004540 u64 orig_num_bytes =
4541 btrfs_file_extent_num_bytes(leaf, fi);
Qu Wenruofda28322013-02-26 08:10:22 +00004542 extent_num_bytes = ALIGN(new_size -
4543 found_key.offset,
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004544 fs_info->sectorsize);
Chris Masondb945352007-10-15 16:15:53 -04004545 btrfs_set_file_extent_num_bytes(leaf, fi,
4546 extent_num_bytes);
4547 num_dec = (orig_num_bytes -
Chris Mason90692182008-02-08 13:49:28 -05004548 extent_num_bytes);
Miao Xie27cdeb72014-04-02 19:51:05 +08004549 if (test_bit(BTRFS_ROOT_REF_COWS,
4550 &root->state) &&
4551 extent_start != 0)
Yan Zhenga76a3cd2008-10-09 11:46:29 -04004552 inode_sub_bytes(inode, num_dec);
Chris Mason5f39d392007-10-15 16:14:19 -04004553 btrfs_mark_buffer_dirty(leaf);
Chris Mason39279cc2007-06-12 06:35:45 -04004554 } else {
Chris Masondb945352007-10-15 16:15:53 -04004555 extent_num_bytes =
4556 btrfs_file_extent_disk_num_bytes(leaf,
4557 fi);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004558 extent_offset = found_key.offset -
4559 btrfs_file_extent_offset(leaf, fi);
4560
Chris Mason39279cc2007-06-12 06:35:45 -04004561 /* FIXME blocksize != 4096 */
Chris Mason90692182008-02-08 13:49:28 -05004562 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
Chris Mason39279cc2007-06-12 06:35:45 -04004563 if (extent_start != 0) {
4564 found_extent = 1;
Miao Xie27cdeb72014-04-02 19:51:05 +08004565 if (test_bit(BTRFS_ROOT_REF_COWS,
4566 &root->state))
Yan Zhenga76a3cd2008-10-09 11:46:29 -04004567 inode_sub_bytes(inode, num_dec);
Chris Mason39279cc2007-06-12 06:35:45 -04004568 }
4569 }
Chris Mason90692182008-02-08 13:49:28 -05004570 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Masonc8b97812008-10-29 14:49:59 -04004571 /*
4572 * we can't truncate inline items that have had
4573 * special encodings
4574 */
4575 if (!del_item &&
Chris Masonc8b97812008-10-29 14:49:59 -04004576 btrfs_file_extent_encryption(leaf, fi) == 0 &&
Josef Bacikddfae632017-10-19 14:16:02 -04004577 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4578 btrfs_file_extent_compression(leaf, fi) == 0) {
4579 u32 size = (u32)(new_size - found_key.offset);
Chris Mason514ac8a2014-01-03 21:07:00 -08004580
Josef Bacikddfae632017-10-19 14:16:02 -04004581 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4582 size = btrfs_file_extent_calc_inline_size(size);
4583 btrfs_truncate_item(root->fs_info, path, size, 1);
4584 } else if (!del_item) {
Chris Mason514ac8a2014-01-03 21:07:00 -08004585 /*
Josef Bacikddfae632017-10-19 14:16:02 -04004586 * We have to bail so the last_size is set to
4587 * just before this extent.
Chris Mason514ac8a2014-01-03 21:07:00 -08004588 */
Josef Bacikddfae632017-10-19 14:16:02 -04004589 err = NEED_TRUNCATE_BLOCK;
4590 break;
Chris Mason90692182008-02-08 13:49:28 -05004591 }
Josef Bacikddfae632017-10-19 14:16:02 -04004592
4593 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4594 inode_sub_bytes(inode, item_end + 1 - new_size);
Chris Mason39279cc2007-06-12 06:35:45 -04004595 }
Chris Mason179e29e2007-11-01 11:28:41 -04004596delete:
Josef Bacikddfae632017-10-19 14:16:02 -04004597 if (del_item)
4598 last_size = found_key.offset;
4599 else
4600 last_size = new_size;
Chris Mason39279cc2007-06-12 06:35:45 -04004601 if (del_item) {
Chris Mason85e21ba2008-01-29 15:11:36 -05004602 if (!pending_del_nr) {
4603 /* no pending yet, add ourselves */
4604 pending_del_slot = path->slots[0];
4605 pending_del_nr = 1;
4606 } else if (pending_del_nr &&
4607 path->slots[0] + 1 == pending_del_slot) {
4608 /* hop on the pending chunk */
4609 pending_del_nr++;
4610 pending_del_slot = path->slots[0];
4611 } else {
Chris Masond3977122009-01-05 21:25:51 -05004612 BUG();
Chris Mason85e21ba2008-01-29 15:11:36 -05004613 }
Chris Mason39279cc2007-06-12 06:35:45 -04004614 } else {
4615 break;
4616 }
Thomas Meyer897ca812017-10-07 16:02:21 +02004617 should_throttle = false;
Chris Mason28f75a02015-02-04 06:59:29 -08004618
Miao Xie27cdeb72014-04-02 19:51:05 +08004619 if (found_extent &&
4620 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004621 root == fs_info->tree_root)) {
Chris Masonb9473432009-03-13 11:00:37 -04004622 btrfs_set_path_blocking(path);
Chris Mason28ed1342014-12-17 09:41:04 -08004623 bytes_deleted += extent_num_bytes;
Josef Bacik84f7d8e2017-09-29 15:43:49 -04004624 ret = btrfs_free_extent(trans, root, extent_start,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004625 extent_num_bytes, 0,
4626 btrfs_header_owner(leaf),
Filipe Mananab06c4bf2015-10-23 07:52:54 +01004627 ino, extent_offset);
Chris Mason39279cc2007-06-12 06:35:45 -04004628 BUG_ON(ret);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04004629 if (btrfs_should_throttle_delayed_refs(trans, fs_info))
4630 btrfs_async_run_delayed_refs(fs_info,
Wang Xiaoguangdd4b8572016-10-18 15:56:13 +08004631 trans->delayed_ref_updates * 2,
4632 trans->transid, 0);
Chris Mason28f75a02015-02-04 06:59:29 -08004633 if (be_nice) {
4634 if (truncate_space_check(trans, root,
4635 extent_num_bytes)) {
Thomas Meyer897ca812017-10-07 16:02:21 +02004636 should_end = true;
Chris Mason28f75a02015-02-04 06:59:29 -08004637 }
4638 if (btrfs_should_throttle_delayed_refs(trans,
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04004639 fs_info))
Thomas Meyer897ca812017-10-07 16:02:21 +02004640 should_throttle = true;
Chris Mason28f75a02015-02-04 06:59:29 -08004641 }
Chris Mason39279cc2007-06-12 06:35:45 -04004642 }
Chris Mason85e21ba2008-01-29 15:11:36 -05004643
Yan, Zheng80825102009-11-12 09:35:36 +00004644 if (found_type == BTRFS_INODE_ITEM_KEY)
4645 break;
4646
4647 if (path->slots[0] == 0 ||
Josef Bacik12621332015-02-03 07:50:16 -08004648 path->slots[0] != pending_del_slot ||
Chris Mason28f75a02015-02-04 06:59:29 -08004649 should_throttle || should_end) {
Yan, Zheng80825102009-11-12 09:35:36 +00004650 if (pending_del_nr) {
4651 ret = btrfs_del_items(trans, root, path,
4652 pending_del_slot,
4653 pending_del_nr);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004654 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04004655 btrfs_abort_transaction(trans, ret);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004656 goto error;
4657 }
Yan, Zheng80825102009-11-12 09:35:36 +00004658 pending_del_nr = 0;
4659 }
David Sterbab3b4aa72011-04-21 01:20:15 +02004660 btrfs_release_path(path);
Chris Mason28f75a02015-02-04 06:59:29 -08004661 if (should_throttle) {
Josef Bacik12621332015-02-03 07:50:16 -08004662 unsigned long updates = trans->delayed_ref_updates;
4663 if (updates) {
4664 trans->delayed_ref_updates = 0;
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04004665 ret = btrfs_run_delayed_refs(trans,
4666 fs_info,
4667 updates * 2);
Josef Bacik12621332015-02-03 07:50:16 -08004668 if (ret && !err)
4669 err = ret;
4670 }
4671 }
Chris Mason28f75a02015-02-04 06:59:29 -08004672 /*
4673 * if we failed to refill our space rsv, bail out
4674 * and let the transaction restart
4675 */
4676 if (should_end) {
4677 err = -EAGAIN;
4678 goto error;
4679 }
Chris Mason85e21ba2008-01-29 15:11:36 -05004680 goto search_again;
Yan, Zheng80825102009-11-12 09:35:36 +00004681 } else {
4682 path->slots[0]--;
Chris Mason85e21ba2008-01-29 15:11:36 -05004683 }
Chris Mason39279cc2007-06-12 06:35:45 -04004684 }
Yan, Zheng80825102009-11-12 09:35:36 +00004685out:
Chris Mason85e21ba2008-01-29 15:11:36 -05004686 if (pending_del_nr) {
4687 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4688 pending_del_nr);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004689 if (ret)
Jeff Mahoney66642832016-06-10 18:19:25 -04004690 btrfs_abort_transaction(trans, ret);
Chris Mason85e21ba2008-01-29 15:11:36 -05004691 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004692error:
Filipe Manana76b42ab2017-02-14 16:56:01 +00004693 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4694 ASSERT(last_size >= new_size);
4695 if (!err && last_size > new_size)
4696 last_size = new_size;
Josef Bacik7f4f6e02013-08-29 16:43:28 -04004697 btrfs_ordered_update_i_size(inode, last_size, NULL);
Filipe Manana76b42ab2017-02-14 16:56:01 +00004698 }
Chris Mason28ed1342014-12-17 09:41:04 -08004699
Chris Mason39279cc2007-06-12 06:35:45 -04004700 btrfs_free_path(path);
Chris Mason28ed1342014-12-17 09:41:04 -08004701
Byongho Leeee221842015-12-15 01:42:10 +09004702 if (be_nice && bytes_deleted > SZ_32M) {
Chris Mason28ed1342014-12-17 09:41:04 -08004703 unsigned long updates = trans->delayed_ref_updates;
4704 if (updates) {
4705 trans->delayed_ref_updates = 0;
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04004706 ret = btrfs_run_delayed_refs(trans, fs_info,
4707 updates * 2);
Chris Mason28ed1342014-12-17 09:41:04 -08004708 if (ret && !err)
4709 err = ret;
4710 }
4711 }
Yan, Zheng80825102009-11-12 09:35:36 +00004712 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04004713}
4714
Chris Masona52d9a82007-08-27 16:49:44 -04004715/*
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304716 * btrfs_truncate_block - read, zero a chunk and write a block
Josef Bacik2aaa6652012-08-29 14:27:18 -04004717 * @inode - inode that we're zeroing
4718 * @from - the offset to start zeroing
4719 * @len - the length to zero, 0 to zero the entire range respective to the
4720 * offset
4721 * @front - zero up to the offset instead of from the offset on
4722 *
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304723 * This will find the block for the "from" offset and cow the block and zero the
Josef Bacik2aaa6652012-08-29 14:27:18 -04004724 * part we want to zero. This is used with truncate and hole punching.
Chris Masona52d9a82007-08-27 16:49:44 -04004725 */
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304726int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
Josef Bacik2aaa6652012-08-29 14:27:18 -04004727 int front)
Chris Masona52d9a82007-08-27 16:49:44 -04004728{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004729 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Josef Bacik2aaa6652012-08-29 14:27:18 -04004730 struct address_space *mapping = inode->i_mapping;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004731 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4732 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00004733 struct extent_state *cached_state = NULL;
Qu Wenruo364ecf32017-02-27 15:10:38 +08004734 struct extent_changeset *data_reserved = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004735 char *kaddr;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004736 u32 blocksize = fs_info->sectorsize;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004737 pgoff_t index = from >> PAGE_SHIFT;
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304738 unsigned offset = from & (blocksize - 1);
Chris Masona52d9a82007-08-27 16:49:44 -04004739 struct page *page;
Josef Bacik3b16a4e2011-09-21 15:05:58 -04004740 gfp_t mask = btrfs_alloc_write_mask(mapping);
Chris Masona52d9a82007-08-27 16:49:44 -04004741 int ret = 0;
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304742 u64 block_start;
4743 u64 block_end;
Chris Masona52d9a82007-08-27 16:49:44 -04004744
Josef Bacik2aaa6652012-08-29 14:27:18 -04004745 if ((offset & (blocksize - 1)) == 0 &&
4746 (!len || ((len & (blocksize - 1)) == 0)))
Chris Masona52d9a82007-08-27 16:49:44 -04004747 goto out;
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304748
Josef Bacik8b62f872017-10-19 14:15:55 -04004749 block_start = round_down(from, blocksize);
4750 block_end = block_start + blocksize - 1;
4751
Qu Wenruo364ecf32017-02-27 15:10:38 +08004752 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
Josef Bacik8b62f872017-10-19 14:15:55 -04004753 block_start, blocksize);
Josef Bacik5d5e1032009-10-13 16:46:49 -04004754 if (ret)
4755 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04004756
Chris Mason211c17f2008-05-15 09:13:45 -04004757again:
Josef Bacik3b16a4e2011-09-21 15:05:58 -04004758 page = find_or_create_page(mapping, index, mask);
Josef Bacik5d5e1032009-10-13 16:46:49 -04004759 if (!page) {
Qu Wenruobc42bda2017-02-27 15:10:39 +08004760 btrfs_delalloc_release_space(inode, data_reserved,
Josef Bacik8b62f872017-10-19 14:15:55 -04004761 block_start, blocksize);
4762 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize);
Miao Xieac6a2b32012-12-05 10:56:13 +00004763 ret = -ENOMEM;
Chris Masona52d9a82007-08-27 16:49:44 -04004764 goto out;
Josef Bacik5d5e1032009-10-13 16:46:49 -04004765 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04004766
Chris Masona52d9a82007-08-27 16:49:44 -04004767 if (!PageUptodate(page)) {
4768 ret = btrfs_readpage(NULL, page);
4769 lock_page(page);
Chris Mason211c17f2008-05-15 09:13:45 -04004770 if (page->mapping != mapping) {
4771 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004772 put_page(page);
Chris Mason211c17f2008-05-15 09:13:45 -04004773 goto again;
4774 }
Chris Masona52d9a82007-08-27 16:49:44 -04004775 if (!PageUptodate(page)) {
4776 ret = -EIO;
Chris Mason89642222008-07-24 09:41:53 -04004777 goto out_unlock;
Chris Masona52d9a82007-08-27 16:49:44 -04004778 }
4779 }
Chris Mason211c17f2008-05-15 09:13:45 -04004780 wait_on_page_writeback(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004781
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304782 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004783 set_page_extent_mapped(page);
4784
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304785 ordered = btrfs_lookup_ordered_extent(inode, block_start);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004786 if (ordered) {
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304787 unlock_extent_cached(io_tree, block_start, block_end,
Josef Bacik2ac55d42010-02-03 19:33:23 +00004788 &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004789 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004790 put_page(page);
Chris Masoneb84ae02008-07-17 13:53:27 -04004791 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004792 btrfs_put_ordered_extent(ordered);
4793 goto again;
4794 }
4795
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304796 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
Liu Bo9e8a4a82012-09-05 19:10:51 -06004797 EXTENT_DIRTY | EXTENT_DELALLOC |
4798 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
Josef Bacik2ac55d42010-02-03 19:33:23 +00004799 0, 0, &cached_state, GFP_NOFS);
Josef Bacik5d5e1032009-10-13 16:46:49 -04004800
Filipe Mananae3b8a482017-11-04 00:16:59 +00004801 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
Qu Wenruoba8b04c2016-07-19 16:50:36 +08004802 &cached_state, 0);
Josef Bacik9ed74f22009-09-11 16:12:44 -04004803 if (ret) {
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304804 unlock_extent_cached(io_tree, block_start, block_end,
Josef Bacik2ac55d42010-02-03 19:33:23 +00004805 &cached_state, GFP_NOFS);
Josef Bacik9ed74f22009-09-11 16:12:44 -04004806 goto out_unlock;
4807 }
4808
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304809 if (offset != blocksize) {
Josef Bacik2aaa6652012-08-29 14:27:18 -04004810 if (!len)
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304811 len = blocksize - offset;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004812 kaddr = kmap(page);
Josef Bacik2aaa6652012-08-29 14:27:18 -04004813 if (front)
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304814 memset(kaddr + (block_start - page_offset(page)),
4815 0, offset);
Josef Bacik2aaa6652012-08-29 14:27:18 -04004816 else
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304817 memset(kaddr + (block_start - page_offset(page)) + offset,
4818 0, len);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004819 flush_dcache_page(page);
4820 kunmap(page);
4821 }
Chris Mason247e7432008-07-17 12:53:51 -04004822 ClearPageChecked(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004823 set_page_dirty(page);
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304824 unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
Josef Bacik2ac55d42010-02-03 19:33:23 +00004825 GFP_NOFS);
Chris Mason39279cc2007-06-12 06:35:45 -04004826
Chris Mason89642222008-07-24 09:41:53 -04004827out_unlock:
Josef Bacik5d5e1032009-10-13 16:46:49 -04004828 if (ret)
Qu Wenruobc42bda2017-02-27 15:10:39 +08004829 btrfs_delalloc_release_space(inode, data_reserved, block_start,
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304830 blocksize);
Josef Bacik8b62f872017-10-19 14:15:55 -04004831 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize);
Chris Mason39279cc2007-06-12 06:35:45 -04004832 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004833 put_page(page);
Chris Mason39279cc2007-06-12 06:35:45 -04004834out:
Qu Wenruo364ecf32017-02-27 15:10:38 +08004835 extent_changeset_free(data_reserved);
Chris Mason39279cc2007-06-12 06:35:45 -04004836 return ret;
4837}
4838
Josef Bacik16e75492013-10-22 12:18:51 -04004839static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4840 u64 offset, u64 len)
4841{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004842 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Josef Bacik16e75492013-10-22 12:18:51 -04004843 struct btrfs_trans_handle *trans;
4844 int ret;
4845
4846 /*
4847 * Still need to make sure the inode looks like it's been updated so
4848 * that any holes get logged if we fsync.
4849 */
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004850 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4851 BTRFS_I(inode)->last_trans = fs_info->generation;
Josef Bacik16e75492013-10-22 12:18:51 -04004852 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4853 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4854 return 0;
4855 }
4856
4857 /*
4858 * 1 - for the one we're dropping
4859 * 1 - for the one we're adding
4860 * 1 - for updating the inode.
4861 */
4862 trans = btrfs_start_transaction(root, 3);
4863 if (IS_ERR(trans))
4864 return PTR_ERR(trans);
4865
4866 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4867 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04004868 btrfs_abort_transaction(trans, ret);
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04004869 btrfs_end_transaction(trans);
Josef Bacik16e75492013-10-22 12:18:51 -04004870 return ret;
4871 }
4872
David Sterbaf85b7372017-01-20 14:54:07 +01004873 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4874 offset, 0, 0, len, 0, len, 0, 0, 0);
Josef Bacik16e75492013-10-22 12:18:51 -04004875 if (ret)
Jeff Mahoney66642832016-06-10 18:19:25 -04004876 btrfs_abort_transaction(trans, ret);
Josef Bacik16e75492013-10-22 12:18:51 -04004877 else
4878 btrfs_update_inode(trans, root, inode);
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04004879 btrfs_end_transaction(trans);
Josef Bacik16e75492013-10-22 12:18:51 -04004880 return ret;
4881}
4882
Josef Bacik695a0d02011-03-04 15:46:53 -05004883/*
4884 * This function puts in dummy file extents for the area we're creating a hole
4885 * for. So if we are truncating this file to a larger size we need to insert
4886 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4887 * the range between oldsize and size
4888 */
Josef Bacika41ad392011-01-31 15:30:16 -05004889int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
Yan Zheng9036c102008-10-30 14:19:41 -04004890{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004891 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Yan Zheng9036c102008-10-30 14:19:41 -04004892 struct btrfs_root *root = BTRFS_I(inode)->root;
4893 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Yan, Zhenga22285a2010-05-16 10:48:46 -04004894 struct extent_map *em = NULL;
Josef Bacik2ac55d42010-02-03 19:33:23 +00004895 struct extent_state *cached_state = NULL;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004896 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004897 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4898 u64 block_end = ALIGN(size, fs_info->sectorsize);
Yan Zheng9036c102008-10-30 14:19:41 -04004899 u64 last_byte;
4900 u64 cur_offset;
4901 u64 hole_size;
Josef Bacik9ed74f22009-09-11 16:12:44 -04004902 int err = 0;
Yan Zheng9036c102008-10-30 14:19:41 -04004903
Josef Bacika71754f2013-06-17 17:14:39 -04004904 /*
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304905 * If our size started in the middle of a block we need to zero out the
4906 * rest of the block before we expand the i_size, otherwise we could
Josef Bacika71754f2013-06-17 17:14:39 -04004907 * expose stale data.
4908 */
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304909 err = btrfs_truncate_block(inode, oldsize, 0, 0);
Josef Bacika71754f2013-06-17 17:14:39 -04004910 if (err)
4911 return err;
4912
Yan Zheng9036c102008-10-30 14:19:41 -04004913 if (size <= hole_start)
4914 return 0;
4915
Yan Zheng9036c102008-10-30 14:19:41 -04004916 while (1) {
4917 struct btrfs_ordered_extent *ordered;
Miao Xiefa7c1492013-09-26 13:15:27 +08004918
David Sterbaff13db42015-12-03 14:30:40 +01004919 lock_extent_bits(io_tree, hole_start, block_end - 1,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01004920 &cached_state);
Nikolay Borisova776c6f2017-02-20 13:50:49 +02004921 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
Miao Xiefa7c1492013-09-26 13:15:27 +08004922 block_end - hole_start);
Yan Zheng9036c102008-10-30 14:19:41 -04004923 if (!ordered)
4924 break;
Josef Bacik2ac55d42010-02-03 19:33:23 +00004925 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4926 &cached_state, GFP_NOFS);
Miao Xiefa7c1492013-09-26 13:15:27 +08004927 btrfs_start_ordered_extent(inode, ordered, 1);
Yan Zheng9036c102008-10-30 14:19:41 -04004928 btrfs_put_ordered_extent(ordered);
4929 }
4930
Yan Zheng9036c102008-10-30 14:19:41 -04004931 cur_offset = hole_start;
4932 while (1) {
Nikolay Borisovfc4f21b2017-02-20 13:51:06 +02004933 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
Yan Zheng9036c102008-10-30 14:19:41 -04004934 block_end - cur_offset, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004935 if (IS_ERR(em)) {
4936 err = PTR_ERR(em);
Zach Brownf2767952013-01-08 19:37:58 +00004937 em = NULL;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004938 break;
4939 }
Yan Zheng9036c102008-10-30 14:19:41 -04004940 last_byte = min(extent_map_end(em), block_end);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004941 last_byte = ALIGN(last_byte, fs_info->sectorsize);
Yan, Zheng80825102009-11-12 09:35:36 +00004942 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
Josef Bacik5dc562c2012-08-17 13:14:17 -04004943 struct extent_map *hole_em;
Yan Zheng9036c102008-10-30 14:19:41 -04004944 hole_size = last_byte - cur_offset;
Yan, Zheng80825102009-11-12 09:35:36 +00004945
Josef Bacik16e75492013-10-22 12:18:51 -04004946 err = maybe_insert_hole(root, inode, cur_offset,
4947 hole_size);
4948 if (err)
Chris Mason771ed682008-11-06 22:02:51 -05004949 break;
Nikolay Borisovdcdbc052017-02-20 13:50:45 +02004950 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
Josef Bacik5dc562c2012-08-17 13:14:17 -04004951 cur_offset + hole_size - 1, 0);
4952 hole_em = alloc_extent_map();
4953 if (!hole_em) {
4954 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4955 &BTRFS_I(inode)->runtime_flags);
4956 goto next;
4957 }
4958 hole_em->start = cur_offset;
4959 hole_em->len = hole_size;
4960 hole_em->orig_start = cur_offset;
Yan, Zheng80825102009-11-12 09:35:36 +00004961
Josef Bacik5dc562c2012-08-17 13:14:17 -04004962 hole_em->block_start = EXTENT_MAP_HOLE;
4963 hole_em->block_len = 0;
Josef Bacikb4939682012-12-03 10:31:19 -05004964 hole_em->orig_block_len = 0;
Josef Bacikcc95bef2013-04-04 14:31:27 -04004965 hole_em->ram_bytes = hole_size;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004966 hole_em->bdev = fs_info->fs_devices->latest_bdev;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004967 hole_em->compress_type = BTRFS_COMPRESS_NONE;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004968 hole_em->generation = fs_info->generation;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004969
4970 while (1) {
4971 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -04004972 err = add_extent_mapping(em_tree, hole_em, 1);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004973 write_unlock(&em_tree->lock);
4974 if (err != -EEXIST)
4975 break;
Nikolay Borisovdcdbc052017-02-20 13:50:45 +02004976 btrfs_drop_extent_cache(BTRFS_I(inode),
4977 cur_offset,
Josef Bacik5dc562c2012-08-17 13:14:17 -04004978 cur_offset +
4979 hole_size - 1, 0);
4980 }
4981 free_extent_map(hole_em);
Yan Zheng9036c102008-10-30 14:19:41 -04004982 }
Josef Bacik16e75492013-10-22 12:18:51 -04004983next:
Yan Zheng9036c102008-10-30 14:19:41 -04004984 free_extent_map(em);
Yan, Zhenga22285a2010-05-16 10:48:46 -04004985 em = NULL;
Yan Zheng9036c102008-10-30 14:19:41 -04004986 cur_offset = last_byte;
Yan, Zheng80825102009-11-12 09:35:36 +00004987 if (cur_offset >= block_end)
Yan Zheng9036c102008-10-30 14:19:41 -04004988 break;
4989 }
Yan, Zhenga22285a2010-05-16 10:48:46 -04004990 free_extent_map(em);
Josef Bacik2ac55d42010-02-03 19:33:23 +00004991 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4992 GFP_NOFS);
Yan Zheng9036c102008-10-30 14:19:41 -04004993 return err;
4994}
4995
Eric Sandeen3972f262013-01-12 02:57:22 +00004996static int btrfs_setsize(struct inode *inode, struct iattr *attr)
Yan, Zheng80825102009-11-12 09:35:36 +00004997{
Miao Xief4a2f4c2011-12-14 20:12:01 -05004998 struct btrfs_root *root = BTRFS_I(inode)->root;
4999 struct btrfs_trans_handle *trans;
Josef Bacika41ad392011-01-31 15:30:16 -05005000 loff_t oldsize = i_size_read(inode);
Eric Sandeen3972f262013-01-12 02:57:22 +00005001 loff_t newsize = attr->ia_size;
5002 int mask = attr->ia_valid;
Yan, Zheng80825102009-11-12 09:35:36 +00005003 int ret;
5004
Eric Sandeen3972f262013-01-12 02:57:22 +00005005 /*
5006 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5007 * special case where we need to update the times despite not having
5008 * these flags set. For all other operations the VFS set these flags
5009 * explicitly if it wants a timestamp update.
5010 */
Christoph Hellwigdff6efc2013-11-19 07:17:07 -08005011 if (newsize != oldsize) {
5012 inode_inc_iversion(inode);
5013 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5014 inode->i_ctime = inode->i_mtime =
Deepa Dinamanic2050a42016-09-14 07:48:06 -07005015 current_time(inode);
Christoph Hellwigdff6efc2013-11-19 07:17:07 -08005016 }
Eric Sandeen3972f262013-01-12 02:57:22 +00005017
Josef Bacika41ad392011-01-31 15:30:16 -05005018 if (newsize > oldsize) {
Filipe Manana9ea24bb2014-10-29 11:57:59 +00005019 /*
David Sterbaea14b57f2017-06-22 02:19:11 +02005020 * Don't do an expanding truncate while snapshotting is ongoing.
Filipe Manana9ea24bb2014-10-29 11:57:59 +00005021 * This is to ensure the snapshot captures a fully consistent
5022 * state of this file - if the snapshot captures this expanding
5023 * truncation, it must capture all writes that happened before
5024 * this truncation.
5025 */
Zhao Lei0bc19f902016-01-06 18:56:36 +08005026 btrfs_wait_for_snapshot_creation(root);
Josef Bacika41ad392011-01-31 15:30:16 -05005027 ret = btrfs_cont_expand(inode, oldsize, newsize);
Filipe Manana9ea24bb2014-10-29 11:57:59 +00005028 if (ret) {
David Sterbaea14b57f2017-06-22 02:19:11 +02005029 btrfs_end_write_no_snapshotting(root);
Yan, Zheng80825102009-11-12 09:35:36 +00005030 return ret;
Filipe Manana9ea24bb2014-10-29 11:57:59 +00005031 }
Yan, Zheng80825102009-11-12 09:35:36 +00005032
Miao Xief4a2f4c2011-12-14 20:12:01 -05005033 trans = btrfs_start_transaction(root, 1);
Filipe Manana9ea24bb2014-10-29 11:57:59 +00005034 if (IS_ERR(trans)) {
David Sterbaea14b57f2017-06-22 02:19:11 +02005035 btrfs_end_write_no_snapshotting(root);
Miao Xief4a2f4c2011-12-14 20:12:01 -05005036 return PTR_ERR(trans);
Filipe Manana9ea24bb2014-10-29 11:57:59 +00005037 }
Miao Xief4a2f4c2011-12-14 20:12:01 -05005038
5039 i_size_write(inode, newsize);
5040 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
Chandan Rajendra27772b62016-01-21 15:56:03 +05305041 pagecache_isize_extended(inode, oldsize, newsize);
Miao Xief4a2f4c2011-12-14 20:12:01 -05005042 ret = btrfs_update_inode(trans, root, inode);
David Sterbaea14b57f2017-06-22 02:19:11 +02005043 btrfs_end_write_no_snapshotting(root);
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04005044 btrfs_end_transaction(trans);
Josef Bacika41ad392011-01-31 15:30:16 -05005045 } else {
Yan, Zheng80825102009-11-12 09:35:36 +00005046
Josef Bacika41ad392011-01-31 15:30:16 -05005047 /*
5048 * We're truncating a file that used to have good data down to
5049 * zero. Make sure it gets into the ordered flush list so that
5050 * any new writes get down to disk quickly.
5051 */
5052 if (newsize == 0)
Josef Bacik72ac3c02012-05-23 14:13:11 -04005053 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5054 &BTRFS_I(inode)->runtime_flags);
Yan, Zheng80825102009-11-12 09:35:36 +00005055
Josef Bacikf3fe8202013-01-07 17:03:21 -05005056 /*
5057 * 1 for the orphan item we're going to add
5058 * 1 for the orphan item deletion.
5059 */
5060 trans = btrfs_start_transaction(root, 2);
5061 if (IS_ERR(trans))
5062 return PTR_ERR(trans);
5063
5064 /*
5065 * We need to do this in case we fail at _any_ point during the
5066 * actual truncate. Once we do the truncate_setsize we could
5067 * invalidate pages which forces any outstanding ordered io to
5068 * be instantly completed which will give us extents that need
5069 * to be truncated. If we fail to get an orphan inode down we
5070 * could have left over extents that were never meant to live,
Nicholas D Steeves01327612016-05-19 21:18:45 -04005071 * so we need to guarantee from this point on that everything
Josef Bacikf3fe8202013-01-07 17:03:21 -05005072 * will be consistent.
5073 */
Nikolay Borisov73f2e542017-02-20 13:50:59 +02005074 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04005075 btrfs_end_transaction(trans);
Josef Bacikf3fe8202013-01-07 17:03:21 -05005076 if (ret)
5077 return ret;
5078
Josef Bacika41ad392011-01-31 15:30:16 -05005079 /* we don't support swapfiles, so vmtruncate shouldn't fail */
5080 truncate_setsize(inode, newsize);
Miao Xie2e60a512013-02-08 07:01:08 +00005081
5082 /* Disable nonlocked read DIO to avoid the end less truncate */
Nikolay Borisovabcefb12017-02-20 13:51:10 +02005083 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
Miao Xie2e60a512013-02-08 07:01:08 +00005084 inode_dio_wait(inode);
Nikolay Borisov0b581702017-02-20 13:51:11 +02005085 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
Miao Xie2e60a512013-02-08 07:01:08 +00005086
Josef Bacika41ad392011-01-31 15:30:16 -05005087 ret = btrfs_truncate(inode);
Josef Bacik7f4f6e02013-08-29 16:43:28 -04005088 if (ret && inode->i_nlink) {
5089 int err;
5090
Liu Bo19fd2df2016-12-01 13:01:02 -08005091 /* To get a stable disk_i_size */
5092 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5093 if (err) {
Nikolay Borisov3d6ae7b2017-02-20 13:50:58 +02005094 btrfs_orphan_del(NULL, BTRFS_I(inode));
Liu Bo19fd2df2016-12-01 13:01:02 -08005095 return err;
5096 }
5097
Josef Bacik7f4f6e02013-08-29 16:43:28 -04005098 /*
5099 * failed to truncate, disk_i_size is only adjusted down
5100 * as we remove extents, so it should represent the true
5101 * size of the inode, so reset the in memory size and
5102 * delete our orphan entry.
5103 */
5104 trans = btrfs_join_transaction(root);
5105 if (IS_ERR(trans)) {
Nikolay Borisov3d6ae7b2017-02-20 13:50:58 +02005106 btrfs_orphan_del(NULL, BTRFS_I(inode));
Josef Bacik7f4f6e02013-08-29 16:43:28 -04005107 return ret;
5108 }
5109 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
Nikolay Borisov3d6ae7b2017-02-20 13:50:58 +02005110 err = btrfs_orphan_del(trans, BTRFS_I(inode));
Josef Bacik7f4f6e02013-08-29 16:43:28 -04005111 if (err)
Jeff Mahoney66642832016-06-10 18:19:25 -04005112 btrfs_abort_transaction(trans, err);
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04005113 btrfs_end_transaction(trans);
Josef Bacik7f4f6e02013-08-29 16:43:28 -04005114 }
Yan, Zheng80825102009-11-12 09:35:36 +00005115 }
5116
Josef Bacika41ad392011-01-31 15:30:16 -05005117 return ret;
Yan, Zheng80825102009-11-12 09:35:36 +00005118}
5119
Chris Mason39279cc2007-06-12 06:35:45 -04005120static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5121{
David Howells2b0143b2015-03-17 22:25:59 +00005122 struct inode *inode = d_inode(dentry);
Li Zefanb83cc962010-12-20 16:04:08 +08005123 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04005124 int err;
5125
Li Zefanb83cc962010-12-20 16:04:08 +08005126 if (btrfs_root_readonly(root))
5127 return -EROFS;
5128
Jan Kara31051c82016-05-26 16:55:18 +02005129 err = setattr_prepare(dentry, attr);
Chris Mason39279cc2007-06-12 06:35:45 -04005130 if (err)
5131 return err;
5132
Chris Mason5a3f23d2009-03-31 13:27:11 -04005133 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
Eric Sandeen3972f262013-01-12 02:57:22 +00005134 err = btrfs_setsize(inode, attr);
Yan, Zheng80825102009-11-12 09:35:36 +00005135 if (err)
5136 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04005137 }
Yan Zheng9036c102008-10-30 14:19:41 -04005138
Christoph Hellwig10257742010-06-04 11:30:02 +02005139 if (attr->ia_valid) {
5140 setattr_copy(inode, attr);
Josef Bacik0c4d2d92012-04-05 15:03:02 -04005141 inode_inc_iversion(inode);
Josef Bacik22c44fe2011-11-30 10:45:38 -05005142 err = btrfs_dirty_inode(inode);
Josef Bacik33268ea2008-07-24 12:16:36 -04005143
Josef Bacik22c44fe2011-11-30 10:45:38 -05005144 if (!err && attr->ia_valid & ATTR_MODE)
Christoph Hellwig996a7102013-12-20 05:16:43 -08005145 err = posix_acl_chmod(inode, inode->i_mode);
Christoph Hellwig10257742010-06-04 11:30:02 +02005146 }
5147
Chris Mason39279cc2007-06-12 06:35:45 -04005148 return err;
5149}
Chris Mason61295eb2008-01-14 16:24:38 -05005150
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00005151/*
5152 * While truncating the inode pages during eviction, we get the VFS calling
5153 * btrfs_invalidatepage() against each page of the inode. This is slow because
5154 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5155 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5156 * extent_state structures over and over, wasting lots of time.
5157 *
5158 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5159 * those expensive operations on a per page basis and do only the ordered io
5160 * finishing, while we release here the extent_map and extent_state structures,
5161 * without the excessive merging and splitting.
5162 */
5163static void evict_inode_truncate_pages(struct inode *inode)
5164{
5165 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5166 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5167 struct rb_node *node;
5168
5169 ASSERT(inode->i_state & I_FREEING);
Johannes Weiner91b0abe2014-04-03 14:47:49 -07005170 truncate_inode_pages_final(&inode->i_data);
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00005171
5172 write_lock(&map_tree->lock);
5173 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5174 struct extent_map *em;
5175
5176 node = rb_first(&map_tree->map);
5177 em = rb_entry(node, struct extent_map, rb_node);
Wang Shilong180589e2013-12-14 15:27:31 +08005178 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5179 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00005180 remove_extent_mapping(map_tree, em);
5181 free_extent_map(em);
Filipe Manana7064dd52014-08-08 02:47:05 +01005182 if (need_resched()) {
5183 write_unlock(&map_tree->lock);
5184 cond_resched();
5185 write_lock(&map_tree->lock);
5186 }
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00005187 }
5188 write_unlock(&map_tree->lock);
5189
Filipe Manana6ca07092015-05-26 00:55:42 +01005190 /*
5191 * Keep looping until we have no more ranges in the io tree.
5192 * We can have ongoing bios started by readpages (called from readahead)
Filipe Manana9c6429d2015-06-10 12:55:41 +01005193 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5194 * still in progress (unlocked the pages in the bio but did not yet
5195 * unlocked the ranges in the io tree). Therefore this means some
Filipe Manana6ca07092015-05-26 00:55:42 +01005196 * ranges can still be locked and eviction started because before
5197 * submitting those bios, which are executed by a separate task (work
5198 * queue kthread), inode references (inode->i_count) were not taken
5199 * (which would be dropped in the end io callback of each bio).
5200 * Therefore here we effectively end up waiting for those bios and
5201 * anyone else holding locked ranges without having bumped the inode's
5202 * reference count - if we don't do it, when they access the inode's
5203 * io_tree to unlock a range it may be too late, leading to an
5204 * use-after-free issue.
5205 */
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00005206 spin_lock(&io_tree->lock);
5207 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5208 struct extent_state *state;
5209 struct extent_state *cached_state = NULL;
Filipe Manana6ca07092015-05-26 00:55:42 +01005210 u64 start;
5211 u64 end;
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00005212
5213 node = rb_first(&io_tree->state);
5214 state = rb_entry(node, struct extent_state, rb_node);
Filipe Manana6ca07092015-05-26 00:55:42 +01005215 start = state->start;
5216 end = state->end;
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00005217 spin_unlock(&io_tree->lock);
5218
David Sterbaff13db42015-12-03 14:30:40 +01005219 lock_extent_bits(io_tree, start, end, &cached_state);
Qu Wenruob9d0b382015-09-29 10:35:16 +08005220
5221 /*
5222 * If still has DELALLOC flag, the extent didn't reach disk,
5223 * and its reserved space won't be freed by delayed_ref.
5224 * So we need to free its reserved space here.
5225 * (Refer to comment in btrfs_invalidatepage, case 2)
5226 *
5227 * Note, end is the bytenr of last byte, so we need + 1 here.
5228 */
5229 if (state->state & EXTENT_DELALLOC)
Qu Wenruobc42bda2017-02-27 15:10:39 +08005230 btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
Qu Wenruob9d0b382015-09-29 10:35:16 +08005231
Filipe Manana6ca07092015-05-26 00:55:42 +01005232 clear_extent_bit(io_tree, start, end,
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00005233 EXTENT_LOCKED | EXTENT_DIRTY |
5234 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5235 EXTENT_DEFRAG, 1, 1,
5236 &cached_state, GFP_NOFS);
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00005237
Filipe Manana7064dd52014-08-08 02:47:05 +01005238 cond_resched();
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00005239 spin_lock(&io_tree->lock);
5240 }
5241 spin_unlock(&io_tree->lock);
5242}
5243
Al Virobd555972010-06-07 11:35:40 -04005244void btrfs_evict_inode(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04005245{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005246 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Chris Mason39279cc2007-06-12 06:35:45 -04005247 struct btrfs_trans_handle *trans;
5248 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik726c35f2011-09-26 15:46:06 -04005249 struct btrfs_block_rsv *rsv, *global_rsv;
Josef Bacik3bce8762015-02-24 12:35:51 -08005250 int steal_from_global = 0;
Nikolay Borisov3d48d982016-06-29 09:46:41 +03005251 u64 min_size;
Chris Mason39279cc2007-06-12 06:35:45 -04005252 int ret;
5253
liubo1abe9b82011-03-24 11:18:59 +00005254 trace_btrfs_inode_evict(inode);
5255
Nikolay Borisov3d48d982016-06-29 09:46:41 +03005256 if (!root) {
5257 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5258 return;
5259 }
5260
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005261 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
Nikolay Borisov3d48d982016-06-29 09:46:41 +03005262
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00005263 evict_inode_truncate_pages(inode);
5264
Stefan Behrens69e9c6c2013-09-05 16:58:43 +02005265 if (inode->i_nlink &&
5266 ((btrfs_root_refs(&root->root_item) != 0 &&
5267 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
Nikolay Borisov70ddc552017-02-20 13:50:35 +02005268 btrfs_is_free_space_inode(BTRFS_I(inode))))
Al Virobd555972010-06-07 11:35:40 -04005269 goto no_delete;
5270
Chris Mason39279cc2007-06-12 06:35:45 -04005271 if (is_bad_inode(inode)) {
Nikolay Borisov3d6ae7b2017-02-20 13:50:58 +02005272 btrfs_orphan_del(NULL, BTRFS_I(inode));
Chris Mason39279cc2007-06-12 06:35:45 -04005273 goto no_delete;
5274 }
Al Virobd555972010-06-07 11:35:40 -04005275 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
Jeff Mahoneya30e5772015-09-11 21:44:17 -04005276 if (!special_file(inode->i_mode))
5277 btrfs_wait_ordered_range(inode, 0, (u64)-1);
Chris Mason5f39d392007-10-15 16:14:19 -04005278
Nikolay Borisov7ab79562017-02-20 13:50:57 +02005279 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
Miao Xief6124962014-09-12 18:44:04 +08005280
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005281 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
Liu Bo6bf02312012-06-25 21:59:09 -06005282 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
Josef Bacik8a35d952012-05-23 14:26:42 -04005283 &BTRFS_I(inode)->runtime_flags));
Yan, Zhengc71bf092009-11-12 09:34:40 +00005284 goto no_delete;
5285 }
5286
Yan, Zheng76dda932009-09-21 16:00:26 -04005287 if (inode->i_nlink > 0) {
Stefan Behrens69e9c6c2013-09-05 16:58:43 +02005288 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5289 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
Yan, Zheng76dda932009-09-21 16:00:26 -04005290 goto no_delete;
5291 }
5292
Nikolay Borisovaa790212017-01-10 20:35:40 +02005293 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
Miao Xie0e8c36a2012-12-19 06:59:51 +00005294 if (ret) {
Nikolay Borisov3d6ae7b2017-02-20 13:50:58 +02005295 btrfs_orphan_del(NULL, BTRFS_I(inode));
Miao Xie0e8c36a2012-12-19 06:59:51 +00005296 goto no_delete;
5297 }
5298
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04005299 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
Josef Bacik4289a662011-08-05 13:22:24 -04005300 if (!rsv) {
Nikolay Borisov3d6ae7b2017-02-20 13:50:58 +02005301 btrfs_orphan_del(NULL, BTRFS_I(inode));
Josef Bacik4289a662011-08-05 13:22:24 -04005302 goto no_delete;
5303 }
Josef Bacik4a338542011-08-29 11:01:31 -04005304 rsv->size = min_size;
Josef Bacikca7e70f2012-08-27 17:48:15 -04005305 rsv->failfast = 1;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005306 global_rsv = &fs_info->global_block_rsv;
Josef Bacik4289a662011-08-05 13:22:24 -04005307
Nikolay Borisov6ef06d22017-02-20 13:50:34 +02005308 btrfs_i_size_write(BTRFS_I(inode), 0);
Chris Mason5f39d392007-10-15 16:14:19 -04005309
Josef Bacik4289a662011-08-05 13:22:24 -04005310 /*
Miao Xie8407aa42012-09-07 01:43:32 -06005311 * This is a bit simpler than btrfs_truncate since we've already
5312 * reserved our space for our orphan item in the unlink, so we just
5313 * need to reserve some slack space in case we add bytes and update
5314 * inode item when doing the truncate.
Josef Bacik4289a662011-08-05 13:22:24 -04005315 */
Yan, Zheng80825102009-11-12 09:35:36 +00005316 while (1) {
Miao Xie08e007d2012-10-16 11:33:38 +00005317 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5318 BTRFS_RESERVE_FLUSH_LIMIT);
Yan, Zheng80825102009-11-12 09:35:36 +00005319
Josef Bacik726c35f2011-09-26 15:46:06 -04005320 /*
5321 * Try and steal from the global reserve since we will
5322 * likely not use this space anyway, we want to try as
5323 * hard as possible to get this to work.
5324 */
5325 if (ret)
Josef Bacik3bce8762015-02-24 12:35:51 -08005326 steal_from_global++;
5327 else
5328 steal_from_global = 0;
5329 ret = 0;
Josef Bacik726c35f2011-09-26 15:46:06 -04005330
Josef Bacik3bce8762015-02-24 12:35:51 -08005331 /*
5332 * steal_from_global == 0: we reserved stuff, hooray!
5333 * steal_from_global == 1: we didn't reserve stuff, boo!
5334 * steal_from_global == 2: we've committed, still not a lot of
5335 * room but maybe we'll have room in the global reserve this
5336 * time.
5337 * steal_from_global == 3: abandon all hope!
5338 */
5339 if (steal_from_global > 2) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005340 btrfs_warn(fs_info,
5341 "Could not get space for a delete, will truncate on mount %d",
5342 ret);
Nikolay Borisov3d6ae7b2017-02-20 13:50:58 +02005343 btrfs_orphan_del(NULL, BTRFS_I(inode));
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04005344 btrfs_free_block_rsv(fs_info, rsv);
Josef Bacik4289a662011-08-05 13:22:24 -04005345 goto no_delete;
Yan, Zhengd68fc572010-05-16 10:49:58 -04005346 }
5347
Miao Xie0e8c36a2012-12-19 06:59:51 +00005348 trans = btrfs_join_transaction(root);
Josef Bacik4289a662011-08-05 13:22:24 -04005349 if (IS_ERR(trans)) {
Nikolay Borisov3d6ae7b2017-02-20 13:50:58 +02005350 btrfs_orphan_del(NULL, BTRFS_I(inode));
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04005351 btrfs_free_block_rsv(fs_info, rsv);
Josef Bacik4289a662011-08-05 13:22:24 -04005352 goto no_delete;
5353 }
5354
Josef Bacik3bce8762015-02-24 12:35:51 -08005355 /*
Nicholas D Steeves01327612016-05-19 21:18:45 -04005356 * We can't just steal from the global reserve, we need to make
Josef Bacik3bce8762015-02-24 12:35:51 -08005357 * sure there is room to do it, if not we need to commit and try
5358 * again.
5359 */
5360 if (steal_from_global) {
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04005361 if (!btrfs_check_space_for_delayed_refs(trans, fs_info))
Josef Bacik3bce8762015-02-24 12:35:51 -08005362 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
Josef Bacik25d609f2016-03-25 13:25:48 -04005363 min_size, 0);
Josef Bacik3bce8762015-02-24 12:35:51 -08005364 else
5365 ret = -ENOSPC;
5366 }
5367
5368 /*
5369 * Couldn't steal from the global reserve, we have too much
5370 * pending stuff built up, commit the transaction and try it
5371 * again.
5372 */
5373 if (ret) {
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04005374 ret = btrfs_commit_transaction(trans);
Josef Bacik3bce8762015-02-24 12:35:51 -08005375 if (ret) {
Nikolay Borisov3d6ae7b2017-02-20 13:50:58 +02005376 btrfs_orphan_del(NULL, BTRFS_I(inode));
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04005377 btrfs_free_block_rsv(fs_info, rsv);
Josef Bacik3bce8762015-02-24 12:35:51 -08005378 goto no_delete;
5379 }
5380 continue;
5381 } else {
5382 steal_from_global = 0;
5383 }
5384
Josef Bacik4289a662011-08-05 13:22:24 -04005385 trans->block_rsv = rsv;
5386
Yan, Zhengd68fc572010-05-16 10:49:58 -04005387 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
Chris Mason28ed1342014-12-17 09:41:04 -08005388 if (ret != -ENOSPC && ret != -EAGAIN)
Yan, Zheng80825102009-11-12 09:35:36 +00005389 break;
5390
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005391 trans->block_rsv = &fs_info->trans_block_rsv;
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04005392 btrfs_end_transaction(trans);
Yan, Zheng80825102009-11-12 09:35:36 +00005393 trans = NULL;
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04005394 btrfs_btree_balance_dirty(fs_info);
Josef Bacik7b128762008-07-24 12:17:14 -04005395 }
5396
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04005397 btrfs_free_block_rsv(fs_info, rsv);
Josef Bacik4289a662011-08-05 13:22:24 -04005398
Josef Bacik4ef31a42013-08-13 14:10:08 -04005399 /*
5400 * Errors here aren't a big deal, it just means we leave orphan items
5401 * in the tree. They will be cleaned up on the next mount.
5402 */
Yan, Zheng80825102009-11-12 09:35:36 +00005403 if (ret == 0) {
Josef Bacik4289a662011-08-05 13:22:24 -04005404 trans->block_rsv = root->orphan_block_rsv;
Nikolay Borisov3d6ae7b2017-02-20 13:50:58 +02005405 btrfs_orphan_del(trans, BTRFS_I(inode));
Josef Bacik4ef31a42013-08-13 14:10:08 -04005406 } else {
Nikolay Borisov3d6ae7b2017-02-20 13:50:58 +02005407 btrfs_orphan_del(NULL, BTRFS_I(inode));
Yan, Zheng80825102009-11-12 09:35:36 +00005408 }
Chris Mason85e21ba2008-01-29 15:11:36 -05005409
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005410 trans->block_rsv = &fs_info->trans_block_rsv;
5411 if (!(root == fs_info->tree_root ||
Li Zefan581bb052011-04-20 10:06:11 +08005412 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02005413 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
Li Zefan581bb052011-04-20 10:06:11 +08005414
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04005415 btrfs_end_transaction(trans);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04005416 btrfs_btree_balance_dirty(fs_info);
Chris Mason39279cc2007-06-12 06:35:45 -04005417no_delete:
Nikolay Borisovf48d1cf2017-01-10 20:35:39 +02005418 btrfs_remove_delayed_node(BTRFS_I(inode));
Jan Karadbd57682012-05-03 14:48:02 +02005419 clear_inode(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04005420}
5421
5422/*
5423 * this returns the key found in the dir entry in the location pointer.
5424 * If no dir entries were found, location->objectid is 0.
5425 */
5426static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5427 struct btrfs_key *location)
5428{
5429 const char *name = dentry->d_name.name;
5430 int namelen = dentry->d_name.len;
5431 struct btrfs_dir_item *di;
5432 struct btrfs_path *path;
5433 struct btrfs_root *root = BTRFS_I(dir)->root;
Yan0d9f7f32007-10-25 15:48:28 -04005434 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04005435
5436 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07005437 if (!path)
5438 return -ENOMEM;
Chris Mason39544012007-12-12 14:38:19 -05005439
David Sterbaf85b7372017-01-20 14:54:07 +01005440 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5441 name, namelen, 0);
Yan0d9f7f32007-10-25 15:48:28 -04005442 if (IS_ERR(di))
5443 ret = PTR_ERR(di);
Chris Masond3977122009-01-05 21:25:51 -05005444
David Sterbac7040052011-04-19 18:00:01 +02005445 if (IS_ERR_OR_NULL(di))
Chris Mason39544012007-12-12 14:38:19 -05005446 goto out_err;
Chris Masond3977122009-01-05 21:25:51 -05005447
Chris Mason5f39d392007-10-15 16:14:19 -04005448 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
Liu Bo56a0e702017-10-30 11:14:38 -06005449 if (location->type != BTRFS_INODE_ITEM_KEY &&
5450 location->type != BTRFS_ROOT_ITEM_KEY) {
5451 btrfs_warn(root->fs_info,
5452"%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5453 __func__, name, btrfs_ino(BTRFS_I(dir)),
5454 location->objectid, location->type, location->offset);
5455 goto out_err;
5456 }
Chris Mason39279cc2007-06-12 06:35:45 -04005457out:
Chris Mason39279cc2007-06-12 06:35:45 -04005458 btrfs_free_path(path);
5459 return ret;
Chris Mason39544012007-12-12 14:38:19 -05005460out_err:
5461 location->objectid = 0;
5462 goto out;
Chris Mason39279cc2007-06-12 06:35:45 -04005463}
5464
5465/*
5466 * when we hit a tree root in a directory, the btrfs part of the inode
5467 * needs to be changed to reflect the root directory of the tree root. This
5468 * is kind of like crossing a mount point.
5469 */
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04005470static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005471 struct inode *dir,
5472 struct dentry *dentry,
5473 struct btrfs_key *location,
5474 struct btrfs_root **sub_root)
Chris Mason39279cc2007-06-12 06:35:45 -04005475{
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005476 struct btrfs_path *path;
5477 struct btrfs_root *new_root;
5478 struct btrfs_root_ref *ref;
5479 struct extent_buffer *leaf;
David Sterba1d4c08e2015-01-02 19:36:14 +01005480 struct btrfs_key key;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005481 int ret;
5482 int err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04005483
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005484 path = btrfs_alloc_path();
5485 if (!path) {
5486 err = -ENOMEM;
5487 goto out;
5488 }
Chris Mason39279cc2007-06-12 06:35:45 -04005489
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005490 err = -ENOENT;
David Sterba1d4c08e2015-01-02 19:36:14 +01005491 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5492 key.type = BTRFS_ROOT_REF_KEY;
5493 key.offset = location->objectid;
5494
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005495 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005496 if (ret) {
5497 if (ret < 0)
5498 err = ret;
5499 goto out;
5500 }
Chris Mason39279cc2007-06-12 06:35:45 -04005501
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005502 leaf = path->nodes[0];
5503 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02005504 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005505 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5506 goto out;
5507
5508 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5509 (unsigned long)(ref + 1),
5510 dentry->d_name.len);
5511 if (ret)
5512 goto out;
5513
David Sterbab3b4aa72011-04-21 01:20:15 +02005514 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005515
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005516 new_root = btrfs_read_fs_root_no_name(fs_info, location);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005517 if (IS_ERR(new_root)) {
5518 err = PTR_ERR(new_root);
5519 goto out;
5520 }
5521
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005522 *sub_root = new_root;
5523 location->objectid = btrfs_root_dirid(&new_root->root_item);
5524 location->type = BTRFS_INODE_ITEM_KEY;
Chris Mason39279cc2007-06-12 06:35:45 -04005525 location->offset = 0;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005526 err = 0;
5527out:
5528 btrfs_free_path(path);
5529 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04005530}
5531
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005532static void inode_tree_add(struct inode *inode)
5533{
5534 struct btrfs_root *root = BTRFS_I(inode)->root;
5535 struct btrfs_inode *entry;
From: Nick Piggin03e860b2009-08-21 10:09:44 +02005536 struct rb_node **p;
5537 struct rb_node *parent;
Filipe David Borba Mananacef21932013-09-02 12:19:13 +01005538 struct rb_node *new = &BTRFS_I(inode)->rb_node;
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02005539 u64 ino = btrfs_ino(BTRFS_I(inode));
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005540
Al Viro1d3382cb2010-10-23 15:19:20 -04005541 if (inode_unhashed(inode))
Yan, Zheng76dda932009-09-21 16:00:26 -04005542 return;
Miao Xiee1409ce2013-05-15 07:48:16 +00005543 parent = NULL;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005544 spin_lock(&root->inode_lock);
Miao Xiee1409ce2013-05-15 07:48:16 +00005545 p = &root->inode_tree.rb_node;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005546 while (*p) {
5547 parent = *p;
5548 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5549
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02005550 if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
From: Nick Piggin03e860b2009-08-21 10:09:44 +02005551 p = &parent->rb_left;
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02005552 else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
From: Nick Piggin03e860b2009-08-21 10:09:44 +02005553 p = &parent->rb_right;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005554 else {
5555 WARN_ON(!(entry->vfs_inode.i_state &
Al Viroa4ffdde2010-06-02 17:38:30 -04005556 (I_WILL_FREE | I_FREEING)));
Filipe David Borba Mananacef21932013-09-02 12:19:13 +01005557 rb_replace_node(parent, new, &root->inode_tree);
From: Nick Piggin03e860b2009-08-21 10:09:44 +02005558 RB_CLEAR_NODE(parent);
5559 spin_unlock(&root->inode_lock);
Filipe David Borba Mananacef21932013-09-02 12:19:13 +01005560 return;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005561 }
5562 }
Filipe David Borba Mananacef21932013-09-02 12:19:13 +01005563 rb_link_node(new, parent, p);
5564 rb_insert_color(new, &root->inode_tree);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005565 spin_unlock(&root->inode_lock);
5566}
5567
5568static void inode_tree_del(struct inode *inode)
5569{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005570 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005571 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng76dda932009-09-21 16:00:26 -04005572 int empty = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005573
From: Nick Piggin03e860b2009-08-21 10:09:44 +02005574 spin_lock(&root->inode_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005575 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005576 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005577 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
Yan, Zheng76dda932009-09-21 16:00:26 -04005578 empty = RB_EMPTY_ROOT(&root->inode_tree);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005579 }
From: Nick Piggin03e860b2009-08-21 10:09:44 +02005580 spin_unlock(&root->inode_lock);
Yan, Zheng76dda932009-09-21 16:00:26 -04005581
Stefan Behrens69e9c6c2013-09-05 16:58:43 +02005582 if (empty && btrfs_root_refs(&root->root_item) == 0) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005583 synchronize_srcu(&fs_info->subvol_srcu);
Yan, Zheng76dda932009-09-21 16:00:26 -04005584 spin_lock(&root->inode_lock);
5585 empty = RB_EMPTY_ROOT(&root->inode_tree);
5586 spin_unlock(&root->inode_lock);
5587 if (empty)
5588 btrfs_add_dead_root(root);
5589 }
5590}
5591
Jeff Mahoney143bede2012-03-01 14:56:26 +01005592void btrfs_invalidate_inodes(struct btrfs_root *root)
Yan, Zheng76dda932009-09-21 16:00:26 -04005593{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005594 struct btrfs_fs_info *fs_info = root->fs_info;
Yan, Zheng76dda932009-09-21 16:00:26 -04005595 struct rb_node *node;
5596 struct rb_node *prev;
5597 struct btrfs_inode *entry;
5598 struct inode *inode;
5599 u64 objectid = 0;
5600
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005601 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
Liu Bo7813b3d2014-02-10 17:37:25 +08005602 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
Yan, Zheng76dda932009-09-21 16:00:26 -04005603
5604 spin_lock(&root->inode_lock);
5605again:
5606 node = root->inode_tree.rb_node;
5607 prev = NULL;
5608 while (node) {
5609 prev = node;
5610 entry = rb_entry(node, struct btrfs_inode, rb_node);
5611
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02005612 if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
Yan, Zheng76dda932009-09-21 16:00:26 -04005613 node = node->rb_left;
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02005614 else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
Yan, Zheng76dda932009-09-21 16:00:26 -04005615 node = node->rb_right;
5616 else
5617 break;
5618 }
5619 if (!node) {
5620 while (prev) {
5621 entry = rb_entry(prev, struct btrfs_inode, rb_node);
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02005622 if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) {
Yan, Zheng76dda932009-09-21 16:00:26 -04005623 node = prev;
5624 break;
5625 }
5626 prev = rb_next(prev);
5627 }
5628 }
5629 while (node) {
5630 entry = rb_entry(node, struct btrfs_inode, rb_node);
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02005631 objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1;
Yan, Zheng76dda932009-09-21 16:00:26 -04005632 inode = igrab(&entry->vfs_inode);
5633 if (inode) {
5634 spin_unlock(&root->inode_lock);
5635 if (atomic_read(&inode->i_count) > 1)
5636 d_prune_aliases(inode);
5637 /*
Al Viro45321ac2010-06-07 13:43:19 -04005638 * btrfs_drop_inode will have it removed from
Yan, Zheng76dda932009-09-21 16:00:26 -04005639 * the inode cache when its usage count
5640 * hits zero.
5641 */
5642 iput(inode);
5643 cond_resched();
5644 spin_lock(&root->inode_lock);
5645 goto again;
5646 }
5647
5648 if (cond_resched_lock(&root->inode_lock))
5649 goto again;
5650
5651 node = rb_next(node);
5652 }
5653 spin_unlock(&root->inode_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005654}
5655
Chris Masone02119d2008-09-05 16:13:11 -04005656static int btrfs_init_locked_inode(struct inode *inode, void *p)
5657{
5658 struct btrfs_iget_args *args = p;
Chris Mason90d3e592014-01-09 17:28:00 -08005659 inode->i_ino = args->location->objectid;
5660 memcpy(&BTRFS_I(inode)->location, args->location,
5661 sizeof(*args->location));
Chris Masone02119d2008-09-05 16:13:11 -04005662 BTRFS_I(inode)->root = args->root;
Chris Mason39279cc2007-06-12 06:35:45 -04005663 return 0;
5664}
5665
5666static int btrfs_find_actor(struct inode *inode, void *opaque)
5667{
5668 struct btrfs_iget_args *args = opaque;
Chris Mason90d3e592014-01-09 17:28:00 -08005669 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
Chris Masond3977122009-01-05 21:25:51 -05005670 args->root == BTRFS_I(inode)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04005671}
5672
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005673static struct inode *btrfs_iget_locked(struct super_block *s,
Chris Mason90d3e592014-01-09 17:28:00 -08005674 struct btrfs_key *location,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005675 struct btrfs_root *root)
Chris Mason39279cc2007-06-12 06:35:45 -04005676{
5677 struct inode *inode;
5678 struct btrfs_iget_args args;
Chris Mason90d3e592014-01-09 17:28:00 -08005679 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
Filipe David Borba Manana778ba822013-10-06 22:22:33 +01005680
Chris Mason90d3e592014-01-09 17:28:00 -08005681 args.location = location;
Chris Mason39279cc2007-06-12 06:35:45 -04005682 args.root = root;
5683
Filipe David Borba Manana778ba822013-10-06 22:22:33 +01005684 inode = iget5_locked(s, hashval, btrfs_find_actor,
Chris Mason39279cc2007-06-12 06:35:45 -04005685 btrfs_init_locked_inode,
5686 (void *)&args);
5687 return inode;
5688}
5689
Balaji Rao1a54ef82008-07-21 02:01:04 +05305690/* Get an inode object given its location and corresponding root.
5691 * Returns in *is_new if the inode was read from disk
5692 */
5693struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
Josef Bacik73f73412009-12-04 17:38:27 +00005694 struct btrfs_root *root, int *new)
Balaji Rao1a54ef82008-07-21 02:01:04 +05305695{
5696 struct inode *inode;
5697
Chris Mason90d3e592014-01-09 17:28:00 -08005698 inode = btrfs_iget_locked(s, location, root);
Balaji Rao1a54ef82008-07-21 02:01:04 +05305699 if (!inode)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005700 return ERR_PTR(-ENOMEM);
Balaji Rao1a54ef82008-07-21 02:01:04 +05305701
5702 if (inode->i_state & I_NEW) {
Filipe Manana67710892016-06-06 11:51:25 +01005703 int ret;
5704
5705 ret = btrfs_read_locked_inode(inode);
Mark Fasheh1748f842011-07-12 11:25:31 -07005706 if (!is_bad_inode(inode)) {
5707 inode_tree_add(inode);
5708 unlock_new_inode(inode);
5709 if (new)
5710 *new = 1;
5711 } else {
Sergei Trofimoviche0b6d652011-09-11 10:52:24 -04005712 unlock_new_inode(inode);
5713 iput(inode);
Filipe Manana67710892016-06-06 11:51:25 +01005714 ASSERT(ret < 0);
5715 inode = ERR_PTR(ret < 0 ? ret : -ESTALE);
Mark Fasheh1748f842011-07-12 11:25:31 -07005716 }
5717 }
5718
Balaji Rao1a54ef82008-07-21 02:01:04 +05305719 return inode;
5720}
5721
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005722static struct inode *new_simple_dir(struct super_block *s,
5723 struct btrfs_key *key,
5724 struct btrfs_root *root)
5725{
5726 struct inode *inode = new_inode(s);
5727
5728 if (!inode)
5729 return ERR_PTR(-ENOMEM);
5730
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005731 BTRFS_I(inode)->root = root;
5732 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
Josef Bacik72ac3c02012-05-23 14:13:11 -04005733 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005734
5735 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
Li Zefan848cce02012-02-21 17:04:28 +08005736 inode->i_op = &btrfs_dir_ro_inode_operations;
Omar Sandoval1fdf4192017-01-25 17:06:39 -08005737 inode->i_opflags &= ~IOP_XATTR;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005738 inode->i_fop = &simple_dir_operations;
5739 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
Deepa Dinamanic2050a42016-09-14 07:48:06 -07005740 inode->i_mtime = current_time(inode);
chandan r9cc97d62012-07-04 12:48:07 +05305741 inode->i_atime = inode->i_mtime;
5742 inode->i_ctime = inode->i_mtime;
5743 BTRFS_I(inode)->i_otime = inode->i_mtime;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005744
5745 return inode;
5746}
5747
Chris Mason3de45862008-11-17 21:02:50 -05005748struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
Chris Mason39279cc2007-06-12 06:35:45 -04005749{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005750 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
Chris Masond3977122009-01-05 21:25:51 -05005751 struct inode *inode;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005752 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04005753 struct btrfs_root *sub_root = root;
5754 struct btrfs_key location;
Yan, Zheng76dda932009-09-21 16:00:26 -04005755 int index;
Josef Bacikb4aff1f2011-06-28 16:18:59 -04005756 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04005757
5758 if (dentry->d_name.len > BTRFS_NAME_LEN)
5759 return ERR_PTR(-ENAMETOOLONG);
Chris Mason5f39d392007-10-15 16:14:19 -04005760
Jeff Layton39e3c952012-11-28 11:30:53 -05005761 ret = btrfs_inode_by_name(dir, dentry, &location);
Chris Mason39279cc2007-06-12 06:35:45 -04005762 if (ret < 0)
5763 return ERR_PTR(ret);
Chris Mason5f39d392007-10-15 16:14:19 -04005764
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005765 if (location.objectid == 0)
Tsutomu Itoh5662344b32013-12-13 09:51:42 +09005766 return ERR_PTR(-ENOENT);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005767
5768 if (location.type == BTRFS_INODE_ITEM_KEY) {
Josef Bacik73f73412009-12-04 17:38:27 +00005769 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005770 return inode;
Chris Mason39279cc2007-06-12 06:35:45 -04005771 }
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005772
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005773 index = srcu_read_lock(&fs_info->subvol_srcu);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04005774 ret = fixup_tree_root_location(fs_info, dir, dentry,
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005775 &location, &sub_root);
5776 if (ret < 0) {
5777 if (ret != -ENOENT)
5778 inode = ERR_PTR(ret);
5779 else
5780 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5781 } else {
Josef Bacik73f73412009-12-04 17:38:27 +00005782 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04005783 }
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005784 srcu_read_unlock(&fs_info->subvol_srcu, index);
Yan, Zheng76dda932009-09-21 16:00:26 -04005785
Julia Lawall34d19ba2011-01-24 19:55:19 +00005786 if (!IS_ERR(inode) && root != sub_root) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005787 down_read(&fs_info->cleanup_work_sem);
David Howellsbc98a422017-07-17 08:45:34 +01005788 if (!sb_rdonly(inode->i_sb))
Josef Bacik66b4ffd2011-01-31 16:22:42 -05005789 ret = btrfs_orphan_cleanup(sub_root);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005790 up_read(&fs_info->cleanup_work_sem);
Josef Bacik01cd3362013-06-03 21:39:49 -04005791 if (ret) {
5792 iput(inode);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05005793 inode = ERR_PTR(ret);
Josef Bacik01cd3362013-06-03 21:39:49 -04005794 }
Yan, Zhengc71bf092009-11-12 09:34:40 +00005795 }
5796
Chris Mason3de45862008-11-17 21:02:50 -05005797 return inode;
5798}
5799
Nick Pigginfe15ce42011-01-07 17:49:23 +11005800static int btrfs_dentry_delete(const struct dentry *dentry)
Yan, Zheng76dda932009-09-21 16:00:26 -04005801{
5802 struct btrfs_root *root;
David Howells2b0143b2015-03-17 22:25:59 +00005803 struct inode *inode = d_inode(dentry);
Yan, Zheng76dda932009-09-21 16:00:26 -04005804
Li Zefan848cce02012-02-21 17:04:28 +08005805 if (!inode && !IS_ROOT(dentry))
David Howells2b0143b2015-03-17 22:25:59 +00005806 inode = d_inode(dentry->d_parent);
Yan, Zheng76dda932009-09-21 16:00:26 -04005807
Li Zefan848cce02012-02-21 17:04:28 +08005808 if (inode) {
5809 root = BTRFS_I(inode)->root;
Yan, Zhengefefb142009-10-09 09:25:16 -04005810 if (btrfs_root_refs(&root->root_item) == 0)
5811 return 1;
Li Zefan848cce02012-02-21 17:04:28 +08005812
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02005813 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
Li Zefan848cce02012-02-21 17:04:28 +08005814 return 1;
Yan, Zhengefefb142009-10-09 09:25:16 -04005815 }
Yan, Zheng76dda932009-09-21 16:00:26 -04005816 return 0;
5817}
5818
Josef Bacikb4aff1f2011-06-28 16:18:59 -04005819static void btrfs_dentry_release(struct dentry *dentry)
5820{
Daeseok Youn944a4512014-04-14 15:37:02 +09005821 kfree(dentry->d_fsdata);
Josef Bacikb4aff1f2011-06-28 16:18:59 -04005822}
5823
Chris Mason3de45862008-11-17 21:02:50 -05005824static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
Al Viro00cd8dd2012-06-10 17:13:09 -04005825 unsigned int flags)
Chris Mason3de45862008-11-17 21:02:50 -05005826{
Tsutomu Itoh5662344b32013-12-13 09:51:42 +09005827 struct inode *inode;
Josef Bacika66e7cc2011-09-18 10:34:03 -04005828
Tsutomu Itoh5662344b32013-12-13 09:51:42 +09005829 inode = btrfs_lookup_dentry(dir, dentry);
5830 if (IS_ERR(inode)) {
5831 if (PTR_ERR(inode) == -ENOENT)
5832 inode = NULL;
5833 else
5834 return ERR_CAST(inode);
5835 }
5836
Al Viro41d28bc2014-10-12 22:24:21 -04005837 return d_splice_alias(inode, dentry);
Chris Mason39279cc2007-06-12 06:35:45 -04005838}
5839
Miao Xie16cdcec2011-04-22 18:12:22 +08005840unsigned char btrfs_filetype_table[] = {
Chris Mason39279cc2007-06-12 06:35:45 -04005841 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5842};
5843
Josef Bacik23b5ec72017-07-24 15:14:25 -04005844/*
5845 * All this infrastructure exists because dir_emit can fault, and we are holding
5846 * the tree lock when doing readdir. For now just allocate a buffer and copy
5847 * our information into that, and then dir_emit from the buffer. This is
5848 * similar to what NFS does, only we don't keep the buffer around in pagecache
5849 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5850 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5851 * tree lock.
5852 */
5853static int btrfs_opendir(struct inode *inode, struct file *file)
5854{
5855 struct btrfs_file_private *private;
5856
5857 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5858 if (!private)
5859 return -ENOMEM;
5860 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5861 if (!private->filldir_buf) {
5862 kfree(private);
5863 return -ENOMEM;
5864 }
5865 file->private_data = private;
5866 return 0;
5867}
5868
5869struct dir_entry {
5870 u64 ino;
5871 u64 offset;
5872 unsigned type;
5873 int name_len;
5874};
5875
5876static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5877{
5878 while (entries--) {
5879 struct dir_entry *entry = addr;
5880 char *name = (char *)(entry + 1);
5881
5882 ctx->pos = entry->offset;
5883 if (!dir_emit(ctx, name, entry->name_len, entry->ino,
5884 entry->type))
5885 return 1;
5886 addr += sizeof(struct dir_entry) + entry->name_len;
5887 ctx->pos++;
5888 }
5889 return 0;
5890}
5891
Al Viro9cdda8d2013-05-22 16:48:09 -04005892static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
Chris Mason39279cc2007-06-12 06:35:45 -04005893{
Al Viro9cdda8d2013-05-22 16:48:09 -04005894 struct inode *inode = file_inode(file);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04005895 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Chris Mason39279cc2007-06-12 06:35:45 -04005896 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik23b5ec72017-07-24 15:14:25 -04005897 struct btrfs_file_private *private = file->private_data;
Chris Mason39279cc2007-06-12 06:35:45 -04005898 struct btrfs_dir_item *di;
5899 struct btrfs_key key;
Chris Mason5f39d392007-10-15 16:14:19 -04005900 struct btrfs_key found_key;
Chris Mason39279cc2007-06-12 06:35:45 -04005901 struct btrfs_path *path;
Josef Bacik23b5ec72017-07-24 15:14:25 -04005902 void *addr;
Miao Xie16cdcec2011-04-22 18:12:22 +08005903 struct list_head ins_list;
5904 struct list_head del_list;
Chris Mason39279cc2007-06-12 06:35:45 -04005905 int ret;
Chris Mason5f39d392007-10-15 16:14:19 -04005906 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04005907 int slot;
Chris Mason5f39d392007-10-15 16:14:19 -04005908 char *name_ptr;
5909 int name_len;
Josef Bacik23b5ec72017-07-24 15:14:25 -04005910 int entries = 0;
5911 int total_len = 0;
Omar Sandoval02dbfc92016-05-20 13:50:33 -07005912 bool put = false;
Jeff Mahoneyc2951f32016-11-21 15:59:04 +01005913 struct btrfs_key location;
Chris Mason5f39d392007-10-15 16:14:19 -04005914
Al Viro9cdda8d2013-05-22 16:48:09 -04005915 if (!dir_emit_dots(file, ctx))
5916 return 0;
5917
David Woodhouse49593bf2008-08-17 17:08:36 +01005918 path = btrfs_alloc_path();
Miao Xie16cdcec2011-04-22 18:12:22 +08005919 if (!path)
5920 return -ENOMEM;
Chris Masonff5714c2011-05-28 07:00:39 -04005921
Josef Bacik23b5ec72017-07-24 15:14:25 -04005922 addr = private->filldir_buf;
David Sterbae4058b52015-11-27 16:31:35 +01005923 path->reada = READA_FORWARD;
David Woodhouse49593bf2008-08-17 17:08:36 +01005924
Jeff Mahoneyc2951f32016-11-21 15:59:04 +01005925 INIT_LIST_HEAD(&ins_list);
5926 INIT_LIST_HEAD(&del_list);
5927 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
Miao Xie16cdcec2011-04-22 18:12:22 +08005928
Josef Bacik23b5ec72017-07-24 15:14:25 -04005929again:
Jeff Mahoneyc2951f32016-11-21 15:59:04 +01005930 key.type = BTRFS_DIR_INDEX_KEY;
Al Viro9cdda8d2013-05-22 16:48:09 -04005931 key.offset = ctx->pos;
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02005932 key.objectid = btrfs_ino(BTRFS_I(inode));
Chris Mason5f39d392007-10-15 16:14:19 -04005933
Chris Mason39279cc2007-06-12 06:35:45 -04005934 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5935 if (ret < 0)
5936 goto err;
David Woodhouse49593bf2008-08-17 17:08:36 +01005937
5938 while (1) {
Josef Bacik23b5ec72017-07-24 15:14:25 -04005939 struct dir_entry *entry;
5940
Chris Mason5f39d392007-10-15 16:14:19 -04005941 leaf = path->nodes[0];
Chris Mason39279cc2007-06-12 06:35:45 -04005942 slot = path->slots[0];
Li Zefanb9e03af2011-03-23 10:43:58 +08005943 if (slot >= btrfs_header_nritems(leaf)) {
5944 ret = btrfs_next_leaf(root, path);
5945 if (ret < 0)
5946 goto err;
5947 else if (ret > 0)
5948 break;
5949 continue;
Chris Mason39279cc2007-06-12 06:35:45 -04005950 }
Chris Mason3de45862008-11-17 21:02:50 -05005951
Chris Mason5f39d392007-10-15 16:14:19 -04005952 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5953
5954 if (found_key.objectid != key.objectid)
Chris Mason39279cc2007-06-12 06:35:45 -04005955 break;
Jeff Mahoneyc2951f32016-11-21 15:59:04 +01005956 if (found_key.type != BTRFS_DIR_INDEX_KEY)
Chris Mason39279cc2007-06-12 06:35:45 -04005957 break;
Al Viro9cdda8d2013-05-22 16:48:09 -04005958 if (found_key.offset < ctx->pos)
Li Zefanb9e03af2011-03-23 10:43:58 +08005959 goto next;
Jeff Mahoneyc2951f32016-11-21 15:59:04 +01005960 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
Miao Xie16cdcec2011-04-22 18:12:22 +08005961 goto next;
Chris Mason39279cc2007-06-12 06:35:45 -04005962 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
Su Yuee79a3322017-06-06 17:57:01 +08005963 if (verify_dir_item(fs_info, leaf, slot, di))
Jeff Mahoneyc2951f32016-11-21 15:59:04 +01005964 goto next;
David Woodhouse49593bf2008-08-17 17:08:36 +01005965
Jeff Mahoneyc2951f32016-11-21 15:59:04 +01005966 name_len = btrfs_dir_name_len(leaf, di);
Josef Bacik23b5ec72017-07-24 15:14:25 -04005967 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5968 PAGE_SIZE) {
5969 btrfs_release_path(path);
5970 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5971 if (ret)
5972 goto nopos;
5973 addr = private->filldir_buf;
5974 entries = 0;
5975 total_len = 0;
5976 goto again;
Chris Mason39279cc2007-06-12 06:35:45 -04005977 }
Josef Bacik23b5ec72017-07-24 15:14:25 -04005978
5979 entry = addr;
5980 entry->name_len = name_len;
5981 name_ptr = (char *)(entry + 1);
Jeff Mahoneyc2951f32016-11-21 15:59:04 +01005982 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5983 name_len);
Josef Bacik23b5ec72017-07-24 15:14:25 -04005984 entry->type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
Jeff Mahoneyc2951f32016-11-21 15:59:04 +01005985 btrfs_dir_item_key_to_cpu(leaf, di, &location);
Josef Bacik23b5ec72017-07-24 15:14:25 -04005986 entry->ino = location.objectid;
5987 entry->offset = found_key.offset;
5988 entries++;
5989 addr += sizeof(struct dir_entry) + name_len;
5990 total_len += sizeof(struct dir_entry) + name_len;
Li Zefanb9e03af2011-03-23 10:43:58 +08005991next:
5992 path->slots[0]++;
Chris Mason39279cc2007-06-12 06:35:45 -04005993 }
Josef Bacik23b5ec72017-07-24 15:14:25 -04005994 btrfs_release_path(path);
5995
5996 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5997 if (ret)
5998 goto nopos;
David Woodhouse49593bf2008-08-17 17:08:36 +01005999
Jeff Mahoneyd2fbb2b2016-11-05 13:26:35 -04006000 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
Jeff Mahoneyc2951f32016-11-21 15:59:04 +01006001 if (ret)
David Sterbabc4ef752015-11-13 13:44:28 +01006002 goto nopos;
6003
Zach Browndb62efb2013-07-11 16:19:42 -07006004 /*
6005 * Stop new entries from being returned after we return the last
6006 * entry.
6007 *
6008 * New directory entries are assigned a strictly increasing
6009 * offset. This means that new entries created during readdir
6010 * are *guaranteed* to be seen in the future by that readdir.
6011 * This has broken buggy programs which operate on names as
6012 * they're returned by readdir. Until we re-use freed offsets
6013 * we have this hack to stop new entries from being returned
6014 * under the assumption that they'll never reach this huge
6015 * offset.
6016 *
6017 * This is being careful not to overflow 32bit loff_t unless the
6018 * last entry requires it because doing so has broken 32bit apps
6019 * in the past.
6020 */
Jeff Mahoneyc2951f32016-11-21 15:59:04 +01006021 if (ctx->pos >= INT_MAX)
6022 ctx->pos = LLONG_MAX;
6023 else
6024 ctx->pos = INT_MAX;
Chris Mason39279cc2007-06-12 06:35:45 -04006025nopos:
6026 ret = 0;
6027err:
Omar Sandoval02dbfc92016-05-20 13:50:33 -07006028 if (put)
6029 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
Chris Mason39279cc2007-06-12 06:35:45 -04006030 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04006031 return ret;
6032}
6033
Christoph Hellwiga9185b42010-03-05 09:21:37 +01006034int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
Chris Mason39279cc2007-06-12 06:35:45 -04006035{
6036 struct btrfs_root *root = BTRFS_I(inode)->root;
6037 struct btrfs_trans_handle *trans;
6038 int ret = 0;
Josef Bacik0af3d002010-06-21 14:48:16 -04006039 bool nolock = false;
Chris Mason39279cc2007-06-12 06:35:45 -04006040
Josef Bacik72ac3c02012-05-23 14:13:11 -04006041 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
Chris Mason4ca8b412008-08-05 13:30:48 -04006042 return 0;
6043
Nikolay Borisov70ddc552017-02-20 13:50:35 +02006044 if (btrfs_fs_closing(root->fs_info) &&
6045 btrfs_is_free_space_inode(BTRFS_I(inode)))
Li Zefan82d59022011-04-20 10:33:24 +08006046 nolock = true;
Josef Bacik0af3d002010-06-21 14:48:16 -04006047
Christoph Hellwiga9185b42010-03-05 09:21:37 +01006048 if (wbc->sync_mode == WB_SYNC_ALL) {
Josef Bacik0af3d002010-06-21 14:48:16 -04006049 if (nolock)
Josef Bacik7a7eaa42011-04-13 12:54:33 -04006050 trans = btrfs_join_transaction_nolock(root);
Josef Bacik0af3d002010-06-21 14:48:16 -04006051 else
Josef Bacik7a7eaa42011-04-13 12:54:33 -04006052 trans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00006053 if (IS_ERR(trans))
6054 return PTR_ERR(trans);
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04006055 ret = btrfs_commit_transaction(trans);
Chris Mason39279cc2007-06-12 06:35:45 -04006056 }
6057 return ret;
6058}
6059
6060/*
Chris Mason54aa1f42007-06-22 14:16:25 -04006061 * This is somewhat expensive, updating the tree every time the
Chris Mason39279cc2007-06-12 06:35:45 -04006062 * inode changes. But, it is most likely to find the inode in cache.
6063 * FIXME, needs more benchmarking...there are no reasons other than performance
6064 * to keep or drop this code.
6065 */
Eric Sandeen48a3b632013-04-25 20:41:01 +00006066static int btrfs_dirty_inode(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04006067{
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04006068 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Chris Mason39279cc2007-06-12 06:35:45 -04006069 struct btrfs_root *root = BTRFS_I(inode)->root;
6070 struct btrfs_trans_handle *trans;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04006071 int ret;
6072
Josef Bacik72ac3c02012-05-23 14:13:11 -04006073 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
Josef Bacik22c44fe2011-11-30 10:45:38 -05006074 return 0;
Chris Mason39279cc2007-06-12 06:35:45 -04006075
Josef Bacik7a7eaa42011-04-13 12:54:33 -04006076 trans = btrfs_join_transaction(root);
Josef Bacik22c44fe2011-11-30 10:45:38 -05006077 if (IS_ERR(trans))
6078 return PTR_ERR(trans);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04006079
6080 ret = btrfs_update_inode(trans, root, inode);
Chris Mason94b60442010-05-26 11:02:00 -04006081 if (ret && ret == -ENOSPC) {
6082 /* whoops, lets try again with the full transaction */
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04006083 btrfs_end_transaction(trans);
Chris Mason94b60442010-05-26 11:02:00 -04006084 trans = btrfs_start_transaction(root, 1);
Josef Bacik22c44fe2011-11-30 10:45:38 -05006085 if (IS_ERR(trans))
6086 return PTR_ERR(trans);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04006087
Chris Mason94b60442010-05-26 11:02:00 -04006088 ret = btrfs_update_inode(trans, root, inode);
Chris Mason94b60442010-05-26 11:02:00 -04006089 }
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04006090 btrfs_end_transaction(trans);
Miao Xie16cdcec2011-04-22 18:12:22 +08006091 if (BTRFS_I(inode)->delayed_node)
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04006092 btrfs_balance_delayed_items(fs_info);
Josef Bacik22c44fe2011-11-30 10:45:38 -05006093
6094 return ret;
6095}
6096
6097/*
6098 * This is a copy of file_update_time. We need this so we can return error on
6099 * ENOSPC for updating the inode in the case of file write and mmap writes.
6100 */
Josef Bacike41f9412012-03-26 09:46:47 -04006101static int btrfs_update_time(struct inode *inode, struct timespec *now,
6102 int flags)
Josef Bacik22c44fe2011-11-30 10:45:38 -05006103{
Alexander Block2bc5565282012-06-15 09:49:33 +02006104 struct btrfs_root *root = BTRFS_I(inode)->root;
6105
6106 if (btrfs_root_readonly(root))
6107 return -EROFS;
6108
Josef Bacike41f9412012-03-26 09:46:47 -04006109 if (flags & S_VERSION)
Josef Bacik22c44fe2011-11-30 10:45:38 -05006110 inode_inc_iversion(inode);
Josef Bacike41f9412012-03-26 09:46:47 -04006111 if (flags & S_CTIME)
6112 inode->i_ctime = *now;
6113 if (flags & S_MTIME)
6114 inode->i_mtime = *now;
6115 if (flags & S_ATIME)
6116 inode->i_atime = *now;
6117 return btrfs_dirty_inode(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04006118}
6119
Chris Masond352ac62008-09-29 15:18:18 -04006120/*
6121 * find the highest existing sequence number in a directory
6122 * and then set the in-memory index_cnt variable to reflect
6123 * free sequence numbers
6124 */
Nikolay Borisov4c5706552017-02-20 13:50:32 +02006125static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
Josef Bacikaec74772008-07-24 12:12:38 -04006126{
Nikolay Borisov4c5706552017-02-20 13:50:32 +02006127 struct btrfs_root *root = inode->root;
Josef Bacikaec74772008-07-24 12:12:38 -04006128 struct btrfs_key key, found_key;
6129 struct btrfs_path *path;
6130 struct extent_buffer *leaf;
6131 int ret;
6132
Nikolay Borisov4c5706552017-02-20 13:50:32 +02006133 key.objectid = btrfs_ino(inode);
David Sterba962a2982014-06-04 18:41:45 +02006134 key.type = BTRFS_DIR_INDEX_KEY;
Josef Bacikaec74772008-07-24 12:12:38 -04006135 key.offset = (u64)-1;
6136
6137 path = btrfs_alloc_path();
6138 if (!path)
6139 return -ENOMEM;
6140
6141 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6142 if (ret < 0)
6143 goto out;
6144 /* FIXME: we should be able to handle this */
6145 if (ret == 0)
6146 goto out;
6147 ret = 0;
6148
6149 /*
6150 * MAGIC NUMBER EXPLANATION:
6151 * since we search a directory based on f_pos we have to start at 2
6152 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6153 * else has to start at 2
6154 */
6155 if (path->slots[0] == 0) {
Nikolay Borisov4c5706552017-02-20 13:50:32 +02006156 inode->index_cnt = 2;
Josef Bacikaec74772008-07-24 12:12:38 -04006157 goto out;
6158 }
6159
6160 path->slots[0]--;
6161
6162 leaf = path->nodes[0];
6163 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6164
Nikolay Borisov4c5706552017-02-20 13:50:32 +02006165 if (found_key.objectid != btrfs_ino(inode) ||
David Sterba962a2982014-06-04 18:41:45 +02006166 found_key.type != BTRFS_DIR_INDEX_KEY) {
Nikolay Borisov4c5706552017-02-20 13:50:32 +02006167 inode->index_cnt = 2;
Josef Bacikaec74772008-07-24 12:12:38 -04006168 goto out;
6169 }
6170
Nikolay Borisov4c5706552017-02-20 13:50:32 +02006171 inode->index_cnt = found_key.offset + 1;
Josef Bacikaec74772008-07-24 12:12:38 -04006172out:
6173 btrfs_free_path(path);
6174 return ret;
6175}
6176
Chris Masond352ac62008-09-29 15:18:18 -04006177/*
6178 * helper to find a free sequence number in a given directory. This current
6179 * code is very simple, later versions will do smarter things in the btree
6180 */
Nikolay Borisov877574e2017-02-20 13:50:33 +02006181int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
Josef Bacikaec74772008-07-24 12:12:38 -04006182{
6183 int ret = 0;
6184
Nikolay Borisov877574e2017-02-20 13:50:33 +02006185 if (dir->index_cnt == (u64)-1) {
6186 ret = btrfs_inode_delayed_dir_index_count(dir);
Miao Xie16cdcec2011-04-22 18:12:22 +08006187 if (ret) {
6188 ret = btrfs_set_inode_index_count(dir);
6189 if (ret)
6190 return ret;
6191 }
Josef Bacikaec74772008-07-24 12:12:38 -04006192 }
6193
Nikolay Borisov877574e2017-02-20 13:50:33 +02006194 *index = dir->index_cnt;
6195 dir->index_cnt++;
Josef Bacikaec74772008-07-24 12:12:38 -04006196
6197 return ret;
6198}
6199
Chris Masonb0d5d102014-09-08 13:08:51 -07006200static int btrfs_insert_inode_locked(struct inode *inode)
6201{
6202 struct btrfs_iget_args args;
6203 args.location = &BTRFS_I(inode)->location;
6204 args.root = BTRFS_I(inode)->root;
6205
6206 return insert_inode_locked4(inode,
6207 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6208 btrfs_find_actor, &args);
6209}
6210
Anand Jain19aee8d2017-07-18 17:37:05 +08006211/*
6212 * Inherit flags from the parent inode.
6213 *
6214 * Currently only the compression flags and the cow flags are inherited.
6215 */
6216static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6217{
6218 unsigned int flags;
6219
6220 if (!dir)
6221 return;
6222
6223 flags = BTRFS_I(dir)->flags;
6224
6225 if (flags & BTRFS_INODE_NOCOMPRESS) {
6226 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6227 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6228 } else if (flags & BTRFS_INODE_COMPRESS) {
6229 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6230 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6231 }
6232
6233 if (flags & BTRFS_INODE_NODATACOW) {
6234 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6235 if (S_ISREG(inode->i_mode))
6236 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6237 }
6238
6239 btrfs_update_iflags(inode);
6240}
6241
Chris Mason39279cc2007-06-12 06:35:45 -04006242static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6243 struct btrfs_root *root,
Josef Bacikaec74772008-07-24 12:12:38 -04006244 struct inode *dir,
Chris Mason9c583092008-01-29 15:15:18 -05006245 const char *name, int name_len,
Al Viro175a4eb2011-07-26 03:30:54 -04006246 u64 ref_objectid, u64 objectid,
6247 umode_t mode, u64 *index)
Chris Mason39279cc2007-06-12 06:35:45 -04006248{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04006249 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason39279cc2007-06-12 06:35:45 -04006250 struct inode *inode;
Chris Mason5f39d392007-10-15 16:14:19 -04006251 struct btrfs_inode_item *inode_item;
Chris Mason39279cc2007-06-12 06:35:45 -04006252 struct btrfs_key *location;
Chris Mason5f39d392007-10-15 16:14:19 -04006253 struct btrfs_path *path;
Chris Mason9c583092008-01-29 15:15:18 -05006254 struct btrfs_inode_ref *ref;
6255 struct btrfs_key key[2];
6256 u32 sizes[2];
Filipe Mananaef3b9af2014-04-27 20:40:45 +01006257 int nitems = name ? 2 : 1;
Chris Mason9c583092008-01-29 15:15:18 -05006258 unsigned long ptr;
Chris Mason39279cc2007-06-12 06:35:45 -04006259 int ret;
Chris Mason39279cc2007-06-12 06:35:45 -04006260
Chris Mason5f39d392007-10-15 16:14:19 -04006261 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07006262 if (!path)
6263 return ERR_PTR(-ENOMEM);
Chris Mason5f39d392007-10-15 16:14:19 -04006264
Jeff Mahoney0b246af2016-06-22 18:54:23 -04006265 inode = new_inode(fs_info->sb);
Yoshinori Sano8fb27642011-04-09 02:30:07 +00006266 if (!inode) {
6267 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04006268 return ERR_PTR(-ENOMEM);
Yoshinori Sano8fb27642011-04-09 02:30:07 +00006269 }
Chris Mason39279cc2007-06-12 06:35:45 -04006270
Li Zefan581bb052011-04-20 10:06:11 +08006271 /*
Filipe Manana5762b5c2014-08-01 00:10:32 +01006272 * O_TMPFILE, set link count to 0, so that after this point,
6273 * we fill in an inode item with the correct link count.
6274 */
6275 if (!name)
6276 set_nlink(inode, 0);
6277
6278 /*
Li Zefan581bb052011-04-20 10:06:11 +08006279 * we have to initialize this early, so we can reclaim the inode
6280 * number if we fail afterwards in this function.
6281 */
6282 inode->i_ino = objectid;
6283
Filipe Mananaef3b9af2014-04-27 20:40:45 +01006284 if (dir && name) {
liubo1abe9b82011-03-24 11:18:59 +00006285 trace_btrfs_inode_request(dir);
6286
Nikolay Borisov877574e2017-02-20 13:50:33 +02006287 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
Shen Feng09771432009-04-02 16:46:06 -04006288 if (ret) {
Yoshinori Sano8fb27642011-04-09 02:30:07 +00006289 btrfs_free_path(path);
Shen Feng09771432009-04-02 16:46:06 -04006290 iput(inode);
Josef Bacikaec74772008-07-24 12:12:38 -04006291 return ERR_PTR(ret);
Shen Feng09771432009-04-02 16:46:06 -04006292 }
Filipe Mananaef3b9af2014-04-27 20:40:45 +01006293 } else if (dir) {
6294 *index = 0;
Josef Bacikaec74772008-07-24 12:12:38 -04006295 }
6296 /*
6297 * index_cnt is ignored for everything but a dir,
6298 * btrfs_get_inode_index_count has an explanation for the magic
6299 * number
6300 */
6301 BTRFS_I(inode)->index_cnt = 2;
Miao Xie67de1172013-12-26 13:07:06 +08006302 BTRFS_I(inode)->dir_index = *index;
Chris Mason39279cc2007-06-12 06:35:45 -04006303 BTRFS_I(inode)->root = root;
Chris Masone02119d2008-09-05 16:13:11 -04006304 BTRFS_I(inode)->generation = trans->transid;
Josef Bacik76195852010-11-19 02:18:02 +00006305 inode->i_generation = BTRFS_I(inode)->generation;
Chris Masonb888db22007-08-27 16:49:44 -04006306
Josef Bacik5dc562c2012-08-17 13:14:17 -04006307 /*
6308 * We could have gotten an inode number from somebody who was fsynced
6309 * and then removed in this same transaction, so let's just set full
6310 * sync since it will be a full sync anyway and this will blow away the
6311 * old info in the log.
6312 */
6313 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6314
Chris Mason9c583092008-01-29 15:15:18 -05006315 key[0].objectid = objectid;
David Sterba962a2982014-06-04 18:41:45 +02006316 key[0].type = BTRFS_INODE_ITEM_KEY;
Chris Mason9c583092008-01-29 15:15:18 -05006317 key[0].offset = 0;
6318
Chris Mason9c583092008-01-29 15:15:18 -05006319 sizes[0] = sizeof(struct btrfs_inode_item);
Filipe Mananaef3b9af2014-04-27 20:40:45 +01006320
6321 if (name) {
6322 /*
6323 * Start new inodes with an inode_ref. This is slightly more
6324 * efficient for small numbers of hard links since they will
6325 * be packed into one item. Extended refs will kick in if we
6326 * add more hard links than can fit in the ref item.
6327 */
6328 key[1].objectid = objectid;
David Sterba962a2982014-06-04 18:41:45 +02006329 key[1].type = BTRFS_INODE_REF_KEY;
Filipe Mananaef3b9af2014-04-27 20:40:45 +01006330 key[1].offset = ref_objectid;
6331
6332 sizes[1] = name_len + sizeof(*ref);
6333 }
Chris Mason9c583092008-01-29 15:15:18 -05006334
Chris Masonb0d5d102014-09-08 13:08:51 -07006335 location = &BTRFS_I(inode)->location;
6336 location->objectid = objectid;
6337 location->offset = 0;
David Sterba962a2982014-06-04 18:41:45 +02006338 location->type = BTRFS_INODE_ITEM_KEY;
Chris Masonb0d5d102014-09-08 13:08:51 -07006339
6340 ret = btrfs_insert_inode_locked(inode);
6341 if (ret < 0)
6342 goto fail;
6343
Chris Masonb9473432009-03-13 11:00:37 -04006344 path->leave_spinning = 1;
Filipe Mananaef3b9af2014-04-27 20:40:45 +01006345 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
Chris Mason9c583092008-01-29 15:15:18 -05006346 if (ret != 0)
Chris Masonb0d5d102014-09-08 13:08:51 -07006347 goto fail_unlock;
Chris Mason5f39d392007-10-15 16:14:19 -04006348
Dmitry Monakhovecc11fab2010-03-04 17:31:47 +03006349 inode_init_owner(inode, dir, mode);
Yan Zhenga76a3cd2008-10-09 11:46:29 -04006350 inode_set_bytes(inode, 0);
chandan r9cc97d62012-07-04 12:48:07 +05306351
Deepa Dinamanic2050a42016-09-14 07:48:06 -07006352 inode->i_mtime = current_time(inode);
chandan r9cc97d62012-07-04 12:48:07 +05306353 inode->i_atime = inode->i_mtime;
6354 inode->i_ctime = inode->i_mtime;
6355 BTRFS_I(inode)->i_otime = inode->i_mtime;
6356
Chris Mason5f39d392007-10-15 16:14:19 -04006357 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6358 struct btrfs_inode_item);
David Sterbab159fa22016-11-08 18:09:03 +01006359 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
Li Zefan293f7e02012-07-10 00:58:58 -06006360 sizeof(*inode_item));
Chris Masone02119d2008-09-05 16:13:11 -04006361 fill_inode_item(trans, path->nodes[0], inode_item, inode);
Chris Mason9c583092008-01-29 15:15:18 -05006362
Filipe Mananaef3b9af2014-04-27 20:40:45 +01006363 if (name) {
6364 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6365 struct btrfs_inode_ref);
6366 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6367 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6368 ptr = (unsigned long)(ref + 1);
6369 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6370 }
Chris Mason9c583092008-01-29 15:15:18 -05006371
Chris Mason5f39d392007-10-15 16:14:19 -04006372 btrfs_mark_buffer_dirty(path->nodes[0]);
6373 btrfs_free_path(path);
6374
Christoph Hellwig6cbff002009-04-17 10:37:41 +02006375 btrfs_inherit_iflags(inode, dir);
6376
Al Viro569254b2011-07-24 17:08:40 -04006377 if (S_ISREG(mode)) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04006378 if (btrfs_test_opt(fs_info, NODATASUM))
Chris Mason94272162009-07-02 12:26:06 -04006379 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04006380 if (btrfs_test_opt(fs_info, NODATACOW))
Josef Bacikf2bdf9a2013-02-21 15:28:28 -05006381 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6382 BTRFS_INODE_NODATASUM;
Chris Mason94272162009-07-02 12:26:06 -04006383 }
6384
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006385 inode_tree_add(inode);
liubo1abe9b82011-03-24 11:18:59 +00006386
6387 trace_btrfs_inode_new(inode);
Chris Mason1973f0f2011-06-24 13:13:29 -04006388 btrfs_set_inode_last_trans(trans, inode);
liubo1abe9b82011-03-24 11:18:59 +00006389
Alexander Block8ea05e32012-07-25 17:35:53 +02006390 btrfs_update_root_times(trans, root);
6391
Filipe David Borba Manana63541922014-01-07 11:47:46 +00006392 ret = btrfs_inode_inherit_props(trans, inode, dir);
6393 if (ret)
Jeff Mahoney0b246af2016-06-22 18:54:23 -04006394 btrfs_err(fs_info,
Filipe David Borba Manana63541922014-01-07 11:47:46 +00006395 "error inheriting props for ino %llu (root %llu): %d",
David Sterbaf85b7372017-01-20 14:54:07 +01006396 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
Filipe David Borba Manana63541922014-01-07 11:47:46 +00006397
Chris Mason39279cc2007-06-12 06:35:45 -04006398 return inode;
Chris Masonb0d5d102014-09-08 13:08:51 -07006399
6400fail_unlock:
6401 unlock_new_inode(inode);
Chris Mason5f39d392007-10-15 16:14:19 -04006402fail:
Filipe Mananaef3b9af2014-04-27 20:40:45 +01006403 if (dir && name)
Josef Bacikaec74772008-07-24 12:12:38 -04006404 BTRFS_I(dir)->index_cnt--;
Chris Mason5f39d392007-10-15 16:14:19 -04006405 btrfs_free_path(path);
Shen Feng09771432009-04-02 16:46:06 -04006406 iput(inode);
Chris Mason5f39d392007-10-15 16:14:19 -04006407 return ERR_PTR(ret);
Chris Mason39279cc2007-06-12 06:35:45 -04006408}
6409
6410static inline u8 btrfs_inode_type(struct inode *inode)
6411{
6412 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6413}
6414
Chris Masond352ac62008-09-29 15:18:18 -04006415/*
6416 * utility function to add 'inode' into 'parent_inode' with
6417 * a give name and a given sequence number.
6418 * if 'add_backref' is true, also insert a backref from the
6419 * inode to the parent directory.
6420 */
Chris Masone02119d2008-09-05 16:13:11 -04006421int btrfs_add_link(struct btrfs_trans_handle *trans,
Nikolay Borisovdb0a6692017-02-20 13:51:08 +02006422 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
Chris Masone02119d2008-09-05 16:13:11 -04006423 const char *name, int name_len, int add_backref, u64 index)
Chris Mason39279cc2007-06-12 06:35:45 -04006424{
Nikolay Borisovdb0a6692017-02-20 13:51:08 +02006425 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006426 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04006427 struct btrfs_key key;
Nikolay Borisovdb0a6692017-02-20 13:51:08 +02006428 struct btrfs_root *root = parent_inode->root;
6429 u64 ino = btrfs_ino(inode);
6430 u64 parent_ino = btrfs_ino(parent_inode);
Chris Mason5f39d392007-10-15 16:14:19 -04006431
Li Zefan33345d012011-04-20 10:31:50 +08006432 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
Nikolay Borisovdb0a6692017-02-20 13:51:08 +02006433 memcpy(&key, &inode->root->root_key, sizeof(key));
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006434 } else {
Li Zefan33345d012011-04-20 10:31:50 +08006435 key.objectid = ino;
David Sterba962a2982014-06-04 18:41:45 +02006436 key.type = BTRFS_INODE_ITEM_KEY;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006437 key.offset = 0;
6438 }
Chris Mason39279cc2007-06-12 06:35:45 -04006439
Li Zefan33345d012011-04-20 10:31:50 +08006440 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04006441 ret = btrfs_add_root_ref(trans, fs_info, key.objectid,
6442 root->root_key.objectid, parent_ino,
6443 index, name, name_len);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006444 } else if (add_backref) {
Li Zefan33345d012011-04-20 10:31:50 +08006445 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6446 parent_ino, index);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006447 }
6448
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006449 /* Nothing to clean up yet */
6450 if (ret)
6451 return ret;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006452
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006453 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6454 parent_inode, &key,
Nikolay Borisovdb0a6692017-02-20 13:51:08 +02006455 btrfs_inode_type(&inode->vfs_inode), index);
Chris Mason9c520572012-12-17 14:26:57 -05006456 if (ret == -EEXIST || ret == -EOVERFLOW)
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006457 goto fail_dir_item;
6458 else if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04006459 btrfs_abort_transaction(trans, ret);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006460 return ret;
Chris Mason39279cc2007-06-12 06:35:45 -04006461 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006462
Nikolay Borisovdb0a6692017-02-20 13:51:08 +02006463 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006464 name_len * 2);
Nikolay Borisovdb0a6692017-02-20 13:51:08 +02006465 inode_inc_iversion(&parent_inode->vfs_inode);
6466 parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6467 current_time(&parent_inode->vfs_inode);
6468 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006469 if (ret)
Jeff Mahoney66642832016-06-10 18:19:25 -04006470 btrfs_abort_transaction(trans, ret);
Chris Mason39279cc2007-06-12 06:35:45 -04006471 return ret;
Chris Masonfe66a052012-02-20 08:40:56 -05006472
6473fail_dir_item:
6474 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6475 u64 local_index;
6476 int err;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04006477 err = btrfs_del_root_ref(trans, fs_info, key.objectid,
6478 root->root_key.objectid, parent_ino,
6479 &local_index, name, name_len);
Chris Masonfe66a052012-02-20 08:40:56 -05006480
6481 } else if (add_backref) {
6482 u64 local_index;
6483 int err;
6484
6485 err = btrfs_del_inode_ref(trans, root, name, name_len,
6486 ino, parent_ino, &local_index);
6487 }
6488 return ret;
Chris Mason39279cc2007-06-12 06:35:45 -04006489}
6490
6491static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
Nikolay Borisovcef415a2017-02-20 13:51:09 +02006492 struct btrfs_inode *dir, struct dentry *dentry,
6493 struct btrfs_inode *inode, int backref, u64 index)
Chris Mason39279cc2007-06-12 06:35:45 -04006494{
Josef Bacika1b075d2010-11-19 20:36:11 +00006495 int err = btrfs_add_link(trans, dir, inode,
6496 dentry->d_name.name, dentry->d_name.len,
6497 backref, index);
Chris Mason39279cc2007-06-12 06:35:45 -04006498 if (err > 0)
6499 err = -EEXIST;
6500 return err;
6501}
6502
Josef Bacik618e21d2007-07-11 10:18:17 -04006503static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
Al Viro1a67aaf2011-07-26 01:52:52 -04006504 umode_t mode, dev_t rdev)
Josef Bacik618e21d2007-07-11 10:18:17 -04006505{
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04006506 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
Josef Bacik618e21d2007-07-11 10:18:17 -04006507 struct btrfs_trans_handle *trans;
6508 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason1832a6d2007-12-21 16:27:21 -05006509 struct inode *inode = NULL;
Josef Bacik618e21d2007-07-11 10:18:17 -04006510 int err;
6511 int drop_inode = 0;
6512 u64 objectid;
Chris Mason00e4e6b2008-08-05 11:18:09 -04006513 u64 index = 0;
Josef Bacik618e21d2007-07-11 10:18:17 -04006514
Josef Bacik9ed74f22009-09-11 16:12:44 -04006515 /*
6516 * 2 for inode item and ref
6517 * 2 for dir items
6518 * 1 for xattr if selinux is on
6519 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04006520 trans = btrfs_start_transaction(root, 5);
6521 if (IS_ERR(trans))
6522 return PTR_ERR(trans);
Chris Mason1832a6d2007-12-21 16:27:21 -05006523
Li Zefan581bb052011-04-20 10:06:11 +08006524 err = btrfs_find_free_ino(root, &objectid);
6525 if (err)
6526 goto out_unlock;
6527
Josef Bacikaec74772008-07-24 12:12:38 -04006528 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
David Sterbaf85b7372017-01-20 14:54:07 +01006529 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6530 mode, &index);
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04006531 if (IS_ERR(inode)) {
6532 err = PTR_ERR(inode);
Josef Bacik618e21d2007-07-11 10:18:17 -04006533 goto out_unlock;
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04006534 }
Josef Bacik618e21d2007-07-11 10:18:17 -04006535
Casey Schauflerad19db72011-12-15 10:09:07 -05006536 /*
6537 * If the active LSM wants to access the inode during
6538 * d_instantiate it needs these. Smack checks to see
6539 * if the filesystem supports xattrs by looking at the
6540 * ops vector.
6541 */
Casey Schauflerad19db72011-12-15 10:09:07 -05006542 inode->i_op = &btrfs_special_inode_operations;
Chris Masonb0d5d102014-09-08 13:08:51 -07006543 init_special_inode(inode, inode->i_mode, rdev);
6544
6545 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
Josef Bacik618e21d2007-07-11 10:18:17 -04006546 if (err)
Chris Masonb0d5d102014-09-08 13:08:51 -07006547 goto out_unlock_inode;
6548
Nikolay Borisovcef415a2017-02-20 13:51:09 +02006549 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6550 0, index);
Chris Masonb0d5d102014-09-08 13:08:51 -07006551 if (err) {
6552 goto out_unlock_inode;
6553 } else {
Yan1b4ab1b2007-08-29 09:11:44 -04006554 btrfs_update_inode(trans, root, inode);
Chris Masonb0d5d102014-09-08 13:08:51 -07006555 unlock_new_inode(inode);
Al Viro08c422c2011-12-23 07:58:13 -05006556 d_instantiate(dentry, inode);
Josef Bacik618e21d2007-07-11 10:18:17 -04006557 }
Chris Masonb0d5d102014-09-08 13:08:51 -07006558
Josef Bacik618e21d2007-07-11 10:18:17 -04006559out_unlock:
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04006560 btrfs_end_transaction(trans);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04006561 btrfs_balance_delayed_items(fs_info);
6562 btrfs_btree_balance_dirty(fs_info);
Josef Bacik618e21d2007-07-11 10:18:17 -04006563 if (drop_inode) {
6564 inode_dec_link_count(inode);
6565 iput(inode);
6566 }
Josef Bacik618e21d2007-07-11 10:18:17 -04006567 return err;
Chris Masonb0d5d102014-09-08 13:08:51 -07006568
6569out_unlock_inode:
6570 drop_inode = 1;
6571 unlock_new_inode(inode);
6572 goto out_unlock;
6573
Josef Bacik618e21d2007-07-11 10:18:17 -04006574}
6575
Chris Mason39279cc2007-06-12 06:35:45 -04006576static int btrfs_create(struct inode *dir, struct dentry *dentry,
Al Viroebfc3b42012-06-10 18:05:36 -04006577 umode_t mode, bool excl)
Chris Mason39279cc2007-06-12 06:35:45 -04006578{
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04006579 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
Chris Mason39279cc2007-06-12 06:35:45 -04006580 struct btrfs_trans_handle *trans;
6581 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason1832a6d2007-12-21 16:27:21 -05006582 struct inode *inode = NULL;
Filipe Brandenburger43baa572012-11-30 03:40:09 +00006583 int drop_inode_on_err = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -04006584 int err;
Chris Mason39279cc2007-06-12 06:35:45 -04006585 u64 objectid;
Chris Mason00e4e6b2008-08-05 11:18:09 -04006586 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04006587
Josef Bacik9ed74f22009-09-11 16:12:44 -04006588 /*
6589 * 2 for inode item and ref
6590 * 2 for dir items
6591 * 1 for xattr if selinux is on
6592 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04006593 trans = btrfs_start_transaction(root, 5);
6594 if (IS_ERR(trans))
6595 return PTR_ERR(trans);
Josef Bacik9ed74f22009-09-11 16:12:44 -04006596
Li Zefan581bb052011-04-20 10:06:11 +08006597 err = btrfs_find_free_ino(root, &objectid);
6598 if (err)
6599 goto out_unlock;
6600
Josef Bacikaec74772008-07-24 12:12:38 -04006601 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
David Sterbaf85b7372017-01-20 14:54:07 +01006602 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6603 mode, &index);
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04006604 if (IS_ERR(inode)) {
6605 err = PTR_ERR(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04006606 goto out_unlock;
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04006607 }
Filipe Brandenburger43baa572012-11-30 03:40:09 +00006608 drop_inode_on_err = 1;
Casey Schauflerad19db72011-12-15 10:09:07 -05006609 /*
6610 * If the active LSM wants to access the inode during
6611 * d_instantiate it needs these. Smack checks to see
6612 * if the filesystem supports xattrs by looking at the
6613 * ops vector.
6614 */
6615 inode->i_fop = &btrfs_file_operations;
6616 inode->i_op = &btrfs_file_inode_operations;
Chris Masonb0d5d102014-09-08 13:08:51 -07006617 inode->i_mapping->a_ops = &btrfs_aops;
Chris Masonb0d5d102014-09-08 13:08:51 -07006618
6619 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6620 if (err)
6621 goto out_unlock_inode;
6622
6623 err = btrfs_update_inode(trans, root, inode);
6624 if (err)
6625 goto out_unlock_inode;
Casey Schauflerad19db72011-12-15 10:09:07 -05006626
Nikolay Borisovcef415a2017-02-20 13:51:09 +02006627 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6628 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04006629 if (err)
Chris Masonb0d5d102014-09-08 13:08:51 -07006630 goto out_unlock_inode;
Filipe Brandenburger43baa572012-11-30 03:40:09 +00006631
Filipe Brandenburger43baa572012-11-30 03:40:09 +00006632 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
Chris Masonb0d5d102014-09-08 13:08:51 -07006633 unlock_new_inode(inode);
Filipe Brandenburger43baa572012-11-30 03:40:09 +00006634 d_instantiate(dentry, inode);
6635
Chris Mason39279cc2007-06-12 06:35:45 -04006636out_unlock:
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04006637 btrfs_end_transaction(trans);
Filipe Brandenburger43baa572012-11-30 03:40:09 +00006638 if (err && drop_inode_on_err) {
Chris Mason39279cc2007-06-12 06:35:45 -04006639 inode_dec_link_count(inode);
6640 iput(inode);
6641 }
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04006642 btrfs_balance_delayed_items(fs_info);
6643 btrfs_btree_balance_dirty(fs_info);
Chris Mason39279cc2007-06-12 06:35:45 -04006644 return err;
Chris Masonb0d5d102014-09-08 13:08:51 -07006645
6646out_unlock_inode:
6647 unlock_new_inode(inode);
6648 goto out_unlock;
6649
Chris Mason39279cc2007-06-12 06:35:45 -04006650}
6651
6652static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6653 struct dentry *dentry)
6654{
Filipe Manana271dba42016-01-05 16:24:05 +00006655 struct btrfs_trans_handle *trans = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04006656 struct btrfs_root *root = BTRFS_I(dir)->root;
David Howells2b0143b2015-03-17 22:25:59 +00006657 struct inode *inode = d_inode(old_dentry);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04006658 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Chris Mason00e4e6b2008-08-05 11:18:09 -04006659 u64 index;
Chris Mason39279cc2007-06-12 06:35:45 -04006660 int err;
6661 int drop_inode = 0;
6662
TARUISI Hiroaki4a8be422009-11-12 07:14:26 +00006663 /* do not allow sys_link's with other subvols of the same device */
6664 if (root->objectid != BTRFS_I(inode)->root->objectid)
Mark Fasheh3ab35642011-03-22 17:20:26 +00006665 return -EXDEV;
TARUISI Hiroaki4a8be422009-11-12 07:14:26 +00006666
Mark Fashehf1863732012-08-08 11:32:27 -07006667 if (inode->i_nlink >= BTRFS_LINK_MAX)
Al Viroc055e992011-03-04 17:15:18 +00006668 return -EMLINK;
Josef Bacik9ed74f22009-09-11 16:12:44 -04006669
Nikolay Borisov877574e2017-02-20 13:50:33 +02006670 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
Josef Bacikaec74772008-07-24 12:12:38 -04006671 if (err)
6672 goto fail;
6673
Yan, Zhenga22285a2010-05-16 10:48:46 -04006674 /*
Miao Xie7e6b6462011-02-18 09:21:17 +00006675 * 2 items for inode and inode ref
Yan, Zhenga22285a2010-05-16 10:48:46 -04006676 * 2 items for dir items
Miao Xie7e6b6462011-02-18 09:21:17 +00006677 * 1 item for parent inode
Yan, Zhenga22285a2010-05-16 10:48:46 -04006678 */
Miao Xie7e6b6462011-02-18 09:21:17 +00006679 trans = btrfs_start_transaction(root, 5);
Yan, Zhenga22285a2010-05-16 10:48:46 -04006680 if (IS_ERR(trans)) {
6681 err = PTR_ERR(trans);
Filipe Manana271dba42016-01-05 16:24:05 +00006682 trans = NULL;
Yan, Zhenga22285a2010-05-16 10:48:46 -04006683 goto fail;
6684 }
Chris Mason5f39d392007-10-15 16:14:19 -04006685
Miao Xie67de1172013-12-26 13:07:06 +08006686 /* There are several dir indexes for this inode, clear the cache. */
6687 BTRFS_I(inode)->dir_index = 0ULL;
Zach Brown8b558c52013-10-16 12:10:34 -07006688 inc_nlink(inode);
Josef Bacik0c4d2d92012-04-05 15:03:02 -04006689 inode_inc_iversion(inode);
Deepa Dinamanic2050a42016-09-14 07:48:06 -07006690 inode->i_ctime = current_time(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04006691 ihold(inode);
Josef Bacike9976152012-10-11 15:53:56 -04006692 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
Josef Bacikaec74772008-07-24 12:12:38 -04006693
Nikolay Borisovcef415a2017-02-20 13:51:09 +02006694 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6695 1, index);
Chris Mason5f39d392007-10-15 16:14:19 -04006696
Yan, Zhenga5719522009-09-24 09:17:31 -04006697 if (err) {
Chris Mason39279cc2007-06-12 06:35:45 -04006698 drop_inode = 1;
Yan, Zhenga5719522009-09-24 09:17:31 -04006699 } else {
Al Viro10d9f302011-07-16 23:09:10 -04006700 struct dentry *parent = dentry->d_parent;
Yan, Zhenga5719522009-09-24 09:17:31 -04006701 err = btrfs_update_inode(trans, root, inode);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006702 if (err)
6703 goto fail;
Filipe Mananaef3b9af2014-04-27 20:40:45 +01006704 if (inode->i_nlink == 1) {
6705 /*
6706 * If new hard link count is 1, it's a file created
6707 * with open(2) O_TMPFILE flag.
6708 */
Nikolay Borisov3d6ae7b2017-02-20 13:50:58 +02006709 err = btrfs_orphan_del(trans, BTRFS_I(inode));
Filipe Mananaef3b9af2014-04-27 20:40:45 +01006710 if (err)
6711 goto fail;
6712 }
Al Viro08c422c2011-12-23 07:58:13 -05006713 d_instantiate(dentry, inode);
Nikolay Borisov9ca5fbfb2017-01-18 00:31:31 +02006714 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
Yan, Zhenga5719522009-09-24 09:17:31 -04006715 }
Chris Mason39279cc2007-06-12 06:35:45 -04006716
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04006717 btrfs_balance_delayed_items(fs_info);
Chris Mason1832a6d2007-12-21 16:27:21 -05006718fail:
Filipe Manana271dba42016-01-05 16:24:05 +00006719 if (trans)
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04006720 btrfs_end_transaction(trans);
Chris Mason39279cc2007-06-12 06:35:45 -04006721 if (drop_inode) {
6722 inode_dec_link_count(inode);
6723 iput(inode);
6724 }
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04006725 btrfs_btree_balance_dirty(fs_info);
Chris Mason39279cc2007-06-12 06:35:45 -04006726 return err;
6727}
6728
Al Viro18bb1db2011-07-26 01:41:39 -04006729static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Chris Mason39279cc2007-06-12 06:35:45 -04006730{
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04006731 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
Chris Masonb9d86662008-05-02 16:13:49 -04006732 struct inode *inode = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04006733 struct btrfs_trans_handle *trans;
6734 struct btrfs_root *root = BTRFS_I(dir)->root;
6735 int err = 0;
6736 int drop_on_err = 0;
Chris Masonb9d86662008-05-02 16:13:49 -04006737 u64 objectid = 0;
Chris Mason00e4e6b2008-08-05 11:18:09 -04006738 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04006739
Josef Bacik9ed74f22009-09-11 16:12:44 -04006740 /*
6741 * 2 items for inode and ref
6742 * 2 items for dir items
6743 * 1 for xattr if selinux is on
6744 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04006745 trans = btrfs_start_transaction(root, 5);
6746 if (IS_ERR(trans))
6747 return PTR_ERR(trans);
Chris Mason39279cc2007-06-12 06:35:45 -04006748
Li Zefan581bb052011-04-20 10:06:11 +08006749 err = btrfs_find_free_ino(root, &objectid);
6750 if (err)
6751 goto out_fail;
6752
Josef Bacikaec74772008-07-24 12:12:38 -04006753 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
David Sterbaf85b7372017-01-20 14:54:07 +01006754 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6755 S_IFDIR | mode, &index);
Chris Mason39279cc2007-06-12 06:35:45 -04006756 if (IS_ERR(inode)) {
6757 err = PTR_ERR(inode);
6758 goto out_fail;
6759 }
Chris Mason5f39d392007-10-15 16:14:19 -04006760
Chris Mason39279cc2007-06-12 06:35:45 -04006761 drop_on_err = 1;
Chris Masonb0d5d102014-09-08 13:08:51 -07006762 /* these must be set before we unlock the inode */
6763 inode->i_op = &btrfs_dir_inode_operations;
6764 inode->i_fop = &btrfs_dir_file_operations;
Josef Bacik33268ea2008-07-24 12:16:36 -04006765
Eric Paris2a7dba32011-02-01 11:05:39 -05006766 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
Josef Bacik33268ea2008-07-24 12:16:36 -04006767 if (err)
Chris Masonb0d5d102014-09-08 13:08:51 -07006768 goto out_fail_inode;
Chris Mason39279cc2007-06-12 06:35:45 -04006769
Nikolay Borisov6ef06d22017-02-20 13:50:34 +02006770 btrfs_i_size_write(BTRFS_I(inode), 0);
Chris Mason39279cc2007-06-12 06:35:45 -04006771 err = btrfs_update_inode(trans, root, inode);
6772 if (err)
Chris Masonb0d5d102014-09-08 13:08:51 -07006773 goto out_fail_inode;
Chris Mason5f39d392007-10-15 16:14:19 -04006774
Nikolay Borisovdb0a6692017-02-20 13:51:08 +02006775 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6776 dentry->d_name.name,
6777 dentry->d_name.len, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04006778 if (err)
Chris Masonb0d5d102014-09-08 13:08:51 -07006779 goto out_fail_inode;
Chris Mason5f39d392007-10-15 16:14:19 -04006780
Chris Mason39279cc2007-06-12 06:35:45 -04006781 d_instantiate(dentry, inode);
Chris Masonb0d5d102014-09-08 13:08:51 -07006782 /*
6783 * mkdir is special. We're unlocking after we call d_instantiate
6784 * to avoid a race with nfsd calling d_instantiate.
6785 */
6786 unlock_new_inode(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04006787 drop_on_err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04006788
6789out_fail:
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04006790 btrfs_end_transaction(trans);
Wang Shilongc7cfb8a2014-12-24 14:45:30 +08006791 if (drop_on_err) {
6792 inode_dec_link_count(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04006793 iput(inode);
Wang Shilongc7cfb8a2014-12-24 14:45:30 +08006794 }
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04006795 btrfs_balance_delayed_items(fs_info);
6796 btrfs_btree_balance_dirty(fs_info);
Chris Mason39279cc2007-06-12 06:35:45 -04006797 return err;
Chris Masonb0d5d102014-09-08 13:08:51 -07006798
6799out_fail_inode:
6800 unlock_new_inode(inode);
6801 goto out_fail;
Chris Mason39279cc2007-06-12 06:35:45 -04006802}
6803
Qu Wenruoe6c4efd2014-09-17 11:53:35 +08006804/* Find next extent map of a given extent map, caller needs to ensure locks */
6805static struct extent_map *next_extent_map(struct extent_map *em)
6806{
6807 struct rb_node *next;
6808
6809 next = rb_next(&em->rb_node);
6810 if (!next)
6811 return NULL;
6812 return container_of(next, struct extent_map, rb_node);
6813}
6814
6815static struct extent_map *prev_extent_map(struct extent_map *em)
6816{
6817 struct rb_node *prev;
6818
6819 prev = rb_prev(&em->rb_node);
6820 if (!prev)
6821 return NULL;
6822 return container_of(prev, struct extent_map, rb_node);
6823}
6824
Chris Masond352ac62008-09-29 15:18:18 -04006825/* helper for btfs_get_extent. Given an existing extent in the tree,
Qu Wenruoe6c4efd2014-09-17 11:53:35 +08006826 * the existing extent is the nearest extent to map_start,
Chris Masond352ac62008-09-29 15:18:18 -04006827 * and an extent that you want to insert, deal with overlap and insert
Qu Wenruoe6c4efd2014-09-17 11:53:35 +08006828 * the best fitted new extent into the tree.
Chris Masond352ac62008-09-29 15:18:18 -04006829 */
Chris Mason3b951512008-04-17 11:29:12 -04006830static int merge_extent_mapping(struct extent_map_tree *em_tree,
6831 struct extent_map *existing,
Chris Masone6dcd2d2008-07-17 12:53:50 -04006832 struct extent_map *em,
Qu Wenruo51f395a2014-08-08 13:06:20 +08006833 u64 map_start)
Chris Mason3b951512008-04-17 11:29:12 -04006834{
Qu Wenruoe6c4efd2014-09-17 11:53:35 +08006835 struct extent_map *prev;
6836 struct extent_map *next;
6837 u64 start;
6838 u64 end;
Chris Mason3b951512008-04-17 11:29:12 -04006839 u64 start_diff;
Chris Mason3b951512008-04-17 11:29:12 -04006840
Chris Masone6dcd2d2008-07-17 12:53:50 -04006841 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
Qu Wenruoe6c4efd2014-09-17 11:53:35 +08006842
6843 if (existing->start > map_start) {
6844 next = existing;
6845 prev = prev_extent_map(next);
6846 } else {
6847 prev = existing;
6848 next = next_extent_map(prev);
6849 }
6850
6851 start = prev ? extent_map_end(prev) : em->start;
6852 start = max_t(u64, start, em->start);
6853 end = next ? next->start : extent_map_end(em);
6854 end = min_t(u64, end, extent_map_end(em));
6855 start_diff = start - em->start;
6856 em->start = start;
6857 em->len = end - start;
Chris Masonc8b97812008-10-29 14:49:59 -04006858 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6859 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
Chris Masone6dcd2d2008-07-17 12:53:50 -04006860 em->block_start += start_diff;
Chris Masonc8b97812008-10-29 14:49:59 -04006861 em->block_len -= start_diff;
6862 }
Josef Bacik09a2a8f92013-04-05 16:51:15 -04006863 return add_extent_mapping(em_tree, em, 0);
Chris Mason3b951512008-04-17 11:29:12 -04006864}
6865
Chris Masonc8b97812008-10-29 14:49:59 -04006866static noinline int uncompress_inline(struct btrfs_path *path,
Byongho Leee40da0e2015-05-19 23:46:45 +09006867 struct page *page,
Chris Masonc8b97812008-10-29 14:49:59 -04006868 size_t pg_offset, u64 extent_offset,
6869 struct btrfs_file_extent_item *item)
6870{
6871 int ret;
6872 struct extent_buffer *leaf = path->nodes[0];
6873 char *tmp;
6874 size_t max_size;
6875 unsigned long inline_size;
6876 unsigned long ptr;
Li Zefan261507a02010-12-17 14:21:50 +08006877 int compress_type;
Chris Masonc8b97812008-10-29 14:49:59 -04006878
6879 WARN_ON(pg_offset != 0);
Li Zefan261507a02010-12-17 14:21:50 +08006880 compress_type = btrfs_file_extent_compression(leaf, item);
Chris Masonc8b97812008-10-29 14:49:59 -04006881 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6882 inline_size = btrfs_file_extent_inline_item_len(leaf,
Ross Kirkdd3cc162013-09-16 15:58:09 +01006883 btrfs_item_nr(path->slots[0]));
Chris Masonc8b97812008-10-29 14:49:59 -04006884 tmp = kmalloc(inline_size, GFP_NOFS);
Tsutomu Itoh8d413712011-04-25 19:43:52 -04006885 if (!tmp)
6886 return -ENOMEM;
Chris Masonc8b97812008-10-29 14:49:59 -04006887 ptr = btrfs_file_extent_inline_start(item);
6888
6889 read_extent_buffer(leaf, tmp, ptr, inline_size);
6890
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03006891 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
Li Zefan261507a02010-12-17 14:21:50 +08006892 ret = btrfs_decompress(compress_type, tmp, page,
6893 extent_offset, inline_size, max_size);
Zygo Blaxelle1699d22017-03-10 16:45:44 -05006894
6895 /*
6896 * decompression code contains a memset to fill in any space between the end
6897 * of the uncompressed data and the end of max_size in case the decompressed
6898 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6899 * the end of an inline extent and the beginning of the next block, so we
6900 * cover that region here.
6901 */
6902
6903 if (max_size + pg_offset < PAGE_SIZE) {
6904 char *map = kmap(page);
6905 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6906 kunmap(page);
6907 }
Chris Masonc8b97812008-10-29 14:49:59 -04006908 kfree(tmp);
Zach Brown166ae5a2014-05-09 17:15:10 -04006909 return ret;
Chris Masonc8b97812008-10-29 14:49:59 -04006910}
6911
Chris Masond352ac62008-09-29 15:18:18 -04006912/*
6913 * a bit scary, this does extent mapping from logical file offset to the disk.
Chris Masond3977122009-01-05 21:25:51 -05006914 * the ugly parts come from merging extents from the disk with the in-ram
6915 * representation. This gets more complex because of the data=ordered code,
Chris Masond352ac62008-09-29 15:18:18 -04006916 * where the in-ram extents might be locked pending data=ordered completion.
6917 *
6918 * This also copies inline extents directly into the page.
6919 */
Nikolay Borisovfc4f21b2017-02-20 13:51:06 +02006920struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6921 struct page *page,
6922 size_t pg_offset, u64 start, u64 len,
6923 int create)
Chris Masona52d9a82007-08-27 16:49:44 -04006924{
Nikolay Borisovfc4f21b2017-02-20 13:51:06 +02006925 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
Chris Masona52d9a82007-08-27 16:49:44 -04006926 int ret;
6927 int err = 0;
Chris Masona52d9a82007-08-27 16:49:44 -04006928 u64 extent_start = 0;
6929 u64 extent_end = 0;
Nikolay Borisovfc4f21b2017-02-20 13:51:06 +02006930 u64 objectid = btrfs_ino(inode);
Chris Masona52d9a82007-08-27 16:49:44 -04006931 u32 found_type;
Chris Masonf4219502008-07-22 11:18:09 -04006932 struct btrfs_path *path = NULL;
Nikolay Borisovfc4f21b2017-02-20 13:51:06 +02006933 struct btrfs_root *root = inode->root;
Chris Masona52d9a82007-08-27 16:49:44 -04006934 struct btrfs_file_extent_item *item;
Chris Mason5f39d392007-10-15 16:14:19 -04006935 struct extent_buffer *leaf;
6936 struct btrfs_key found_key;
Chris Masona52d9a82007-08-27 16:49:44 -04006937 struct extent_map *em = NULL;
Nikolay Borisovfc4f21b2017-02-20 13:51:06 +02006938 struct extent_map_tree *em_tree = &inode->extent_tree;
6939 struct extent_io_tree *io_tree = &inode->io_tree;
Chris Masona52d9a82007-08-27 16:49:44 -04006940 struct btrfs_trans_handle *trans = NULL;
Filipe Manana7ffbb592014-06-09 03:48:05 +01006941 const bool new_inline = !page || create;
Chris Masona52d9a82007-08-27 16:49:44 -04006942
Chris Masona52d9a82007-08-27 16:49:44 -04006943again:
Chris Mason890871b2009-09-02 16:24:52 -04006944 read_lock(&em_tree->lock);
Chris Masond1310b22008-01-24 16:13:08 -05006945 em = lookup_extent_mapping(em_tree, start, len);
Chris Masona061fc82008-05-07 11:43:44 -04006946 if (em)
Jeff Mahoney0b246af2016-06-22 18:54:23 -04006947 em->bdev = fs_info->fs_devices->latest_bdev;
Chris Mason890871b2009-09-02 16:24:52 -04006948 read_unlock(&em_tree->lock);
Chris Masond1310b22008-01-24 16:13:08 -05006949
Chris Masona52d9a82007-08-27 16:49:44 -04006950 if (em) {
Chris Masone1c4b742008-04-22 13:26:46 -04006951 if (em->start > start || em->start + em->len <= start)
6952 free_extent_map(em);
6953 else if (em->block_start == EXTENT_MAP_INLINE && page)
Chris Mason70dec802008-01-29 09:59:12 -05006954 free_extent_map(em);
6955 else
6956 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04006957 }
David Sterba172ddd62011-04-21 00:48:27 +02006958 em = alloc_extent_map();
Chris Masona52d9a82007-08-27 16:49:44 -04006959 if (!em) {
Chris Masond1310b22008-01-24 16:13:08 -05006960 err = -ENOMEM;
6961 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04006962 }
Jeff Mahoney0b246af2016-06-22 18:54:23 -04006963 em->bdev = fs_info->fs_devices->latest_bdev;
Chris Masond1310b22008-01-24 16:13:08 -05006964 em->start = EXTENT_MAP_HOLE;
Chris Mason445a6942008-11-10 11:53:33 -05006965 em->orig_start = EXTENT_MAP_HOLE;
Chris Masond1310b22008-01-24 16:13:08 -05006966 em->len = (u64)-1;
Chris Masonc8b97812008-10-29 14:49:59 -04006967 em->block_len = (u64)-1;
Chris Masonf4219502008-07-22 11:18:09 -04006968
6969 if (!path) {
6970 path = btrfs_alloc_path();
Josef Bacik026fd312011-05-13 10:32:11 -04006971 if (!path) {
6972 err = -ENOMEM;
6973 goto out;
6974 }
6975 /*
6976 * Chances are we'll be called again, so go ahead and do
6977 * readahead
6978 */
David Sterbae4058b52015-11-27 16:31:35 +01006979 path->reada = READA_FORWARD;
Chris Masonf4219502008-07-22 11:18:09 -04006980 }
6981
Chris Mason179e29e2007-11-01 11:28:41 -04006982 ret = btrfs_lookup_file_extent(trans, root, path,
6983 objectid, start, trans != NULL);
Chris Masona52d9a82007-08-27 16:49:44 -04006984 if (ret < 0) {
6985 err = ret;
6986 goto out;
6987 }
6988
6989 if (ret != 0) {
6990 if (path->slots[0] == 0)
6991 goto not_found;
6992 path->slots[0]--;
6993 }
6994
Chris Mason5f39d392007-10-15 16:14:19 -04006995 leaf = path->nodes[0];
6996 item = btrfs_item_ptr(leaf, path->slots[0],
Chris Masona52d9a82007-08-27 16:49:44 -04006997 struct btrfs_file_extent_item);
Chris Masona52d9a82007-08-27 16:49:44 -04006998 /* are we inside the extent that was found? */
Chris Mason5f39d392007-10-15 16:14:19 -04006999 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
David Sterba962a2982014-06-04 18:41:45 +02007000 found_type = found_key.type;
Chris Mason5f39d392007-10-15 16:14:19 -04007001 if (found_key.objectid != objectid ||
Chris Masona52d9a82007-08-27 16:49:44 -04007002 found_type != BTRFS_EXTENT_DATA_KEY) {
Josef Bacik25a50342013-10-14 12:08:38 -04007003 /*
7004 * If we backup past the first extent we want to move forward
7005 * and see if there is an extent in front of us, otherwise we'll
7006 * say there is a hole for our whole search range which can
7007 * cause problems.
7008 */
7009 extent_end = start;
7010 goto next;
Chris Masona52d9a82007-08-27 16:49:44 -04007011 }
7012
Chris Mason5f39d392007-10-15 16:14:19 -04007013 found_type = btrfs_file_extent_type(leaf, item);
7014 extent_start = found_key.offset;
Yan Zhengd899e052008-10-30 14:25:28 -04007015 if (found_type == BTRFS_FILE_EXTENT_REG ||
7016 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Chris Masona52d9a82007-08-27 16:49:44 -04007017 extent_end = extent_start +
Chris Masondb945352007-10-15 16:15:53 -04007018 btrfs_file_extent_num_bytes(leaf, item);
Liu Bo09ed2f12017-03-10 11:09:48 -08007019
7020 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
7021 extent_start);
Yan Zheng9036c102008-10-30 14:19:41 -04007022 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
7023 size_t size;
Chris Mason514ac8a2014-01-03 21:07:00 -08007024 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
Jeff Mahoneyda170662016-06-15 09:22:56 -04007025 extent_end = ALIGN(extent_start + size,
Jeff Mahoney0b246af2016-06-22 18:54:23 -04007026 fs_info->sectorsize);
Liu Bo09ed2f12017-03-10 11:09:48 -08007027
7028 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
7029 path->slots[0],
7030 extent_start);
Yan Zheng9036c102008-10-30 14:19:41 -04007031 }
Josef Bacik25a50342013-10-14 12:08:38 -04007032next:
Yan Zheng9036c102008-10-30 14:19:41 -04007033 if (start >= extent_end) {
7034 path->slots[0]++;
7035 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
7036 ret = btrfs_next_leaf(root, path);
7037 if (ret < 0) {
7038 err = ret;
7039 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04007040 }
Yan Zheng9036c102008-10-30 14:19:41 -04007041 if (ret > 0)
7042 goto not_found;
7043 leaf = path->nodes[0];
Chris Masona52d9a82007-08-27 16:49:44 -04007044 }
Yan Zheng9036c102008-10-30 14:19:41 -04007045 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7046 if (found_key.objectid != objectid ||
7047 found_key.type != BTRFS_EXTENT_DATA_KEY)
7048 goto not_found;
7049 if (start + len <= found_key.offset)
7050 goto not_found;
Wang Shilonge2eca692014-07-17 11:44:14 +08007051 if (start > found_key.offset)
7052 goto next;
Yan Zheng9036c102008-10-30 14:19:41 -04007053 em->start = start;
Josef Bacik70c8a912012-10-11 16:54:30 -04007054 em->orig_start = start;
Yan Zheng9036c102008-10-30 14:19:41 -04007055 em->len = found_key.offset - start;
7056 goto not_found_em;
7057 }
7058
Nikolay Borisovfc4f21b2017-02-20 13:51:06 +02007059 btrfs_extent_item_to_extent_map(inode, path, item,
Nikolay Borisov9cdc5122017-02-20 13:51:02 +02007060 new_inline, em);
Filipe Manana7ffbb592014-06-09 03:48:05 +01007061
Yan Zhengd899e052008-10-30 14:25:28 -04007062 if (found_type == BTRFS_FILE_EXTENT_REG ||
7063 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Chris Masona52d9a82007-08-27 16:49:44 -04007064 goto insert;
7065 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason5f39d392007-10-15 16:14:19 -04007066 unsigned long ptr;
Chris Masona52d9a82007-08-27 16:49:44 -04007067 char *map;
Chris Mason3326d1b2007-10-15 16:18:25 -04007068 size_t size;
7069 size_t extent_offset;
7070 size_t copy_size;
Chris Masona52d9a82007-08-27 16:49:44 -04007071
Filipe Manana7ffbb592014-06-09 03:48:05 +01007072 if (new_inline)
Yan689f9342007-10-29 11:41:07 -04007073 goto out;
Yan689f9342007-10-29 11:41:07 -04007074
Chris Mason514ac8a2014-01-03 21:07:00 -08007075 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
Yan Zheng9036c102008-10-30 14:19:41 -04007076 extent_offset = page_offset(page) + pg_offset - extent_start;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03007077 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
7078 size - extent_offset);
Chris Mason3326d1b2007-10-15 16:18:25 -04007079 em->start = extent_start + extent_offset;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04007080 em->len = ALIGN(copy_size, fs_info->sectorsize);
Josef Bacikb4939682012-12-03 10:31:19 -05007081 em->orig_block_len = em->len;
Josef Bacik70c8a912012-10-11 16:54:30 -04007082 em->orig_start = em->start;
Yan689f9342007-10-29 11:41:07 -04007083 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
Chris Mason179e29e2007-11-01 11:28:41 -04007084 if (create == 0 && !PageUptodate(page)) {
Li Zefan261507a02010-12-17 14:21:50 +08007085 if (btrfs_file_extent_compression(leaf, item) !=
7086 BTRFS_COMPRESS_NONE) {
Byongho Leee40da0e2015-05-19 23:46:45 +09007087 ret = uncompress_inline(path, page, pg_offset,
Chris Masonc8b97812008-10-29 14:49:59 -04007088 extent_offset, item);
Zach Brown166ae5a2014-05-09 17:15:10 -04007089 if (ret) {
7090 err = ret;
7091 goto out;
7092 }
Chris Masonc8b97812008-10-29 14:49:59 -04007093 } else {
7094 map = kmap(page);
7095 read_extent_buffer(leaf, map + pg_offset, ptr,
7096 copy_size);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03007097 if (pg_offset + copy_size < PAGE_SIZE) {
Chris Mason93c82d52009-09-11 12:36:29 -04007098 memset(map + pg_offset + copy_size, 0,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03007099 PAGE_SIZE - pg_offset -
Chris Mason93c82d52009-09-11 12:36:29 -04007100 copy_size);
7101 }
Chris Masonc8b97812008-10-29 14:49:59 -04007102 kunmap(page);
7103 }
Chris Mason179e29e2007-11-01 11:28:41 -04007104 flush_dcache_page(page);
7105 } else if (create && PageUptodate(page)) {
Jan Schmidt6bf7e082011-12-01 14:35:19 +01007106 BUG();
Chris Mason179e29e2007-11-01 11:28:41 -04007107 if (!trans) {
7108 kunmap(page);
7109 free_extent_map(em);
7110 em = NULL;
Chris Masonff5714c2011-05-28 07:00:39 -04007111
David Sterbab3b4aa72011-04-21 01:20:15 +02007112 btrfs_release_path(path);
Josef Bacik7a7eaa42011-04-13 12:54:33 -04007113 trans = btrfs_join_transaction(root);
Chris Masonff5714c2011-05-28 07:00:39 -04007114
Tsutomu Itoh3612b492011-01-25 02:51:38 +00007115 if (IS_ERR(trans))
7116 return ERR_CAST(trans);
Chris Mason179e29e2007-11-01 11:28:41 -04007117 goto again;
7118 }
Chris Masonc8b97812008-10-29 14:49:59 -04007119 map = kmap(page);
Chris Mason70dec802008-01-29 09:59:12 -05007120 write_extent_buffer(leaf, map + pg_offset, ptr,
Chris Mason179e29e2007-11-01 11:28:41 -04007121 copy_size);
Chris Masonc8b97812008-10-29 14:49:59 -04007122 kunmap(page);
Chris Mason179e29e2007-11-01 11:28:41 -04007123 btrfs_mark_buffer_dirty(leaf);
Chris Masona52d9a82007-08-27 16:49:44 -04007124 }
Chris Masond1310b22008-01-24 16:13:08 -05007125 set_extent_uptodate(io_tree, em->start,
Arne Jansen507903b2011-04-06 10:02:20 +00007126 extent_map_end(em) - 1, NULL, GFP_NOFS);
Chris Masona52d9a82007-08-27 16:49:44 -04007127 goto insert;
Chris Masona52d9a82007-08-27 16:49:44 -04007128 }
7129not_found:
7130 em->start = start;
Josef Bacik70c8a912012-10-11 16:54:30 -04007131 em->orig_start = start;
Chris Masond1310b22008-01-24 16:13:08 -05007132 em->len = len;
Chris Masona52d9a82007-08-27 16:49:44 -04007133not_found_em:
Chris Mason5f39d392007-10-15 16:14:19 -04007134 em->block_start = EXTENT_MAP_HOLE;
Yan Zheng9036c102008-10-30 14:19:41 -04007135 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
Chris Masona52d9a82007-08-27 16:49:44 -04007136insert:
David Sterbab3b4aa72011-04-21 01:20:15 +02007137 btrfs_release_path(path);
Chris Masond1310b22008-01-24 16:13:08 -05007138 if (em->start > start || extent_map_end(em) <= start) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04007139 btrfs_err(fs_info,
Jeff Mahoney5d163e02016-09-20 10:05:00 -04007140 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7141 em->start, em->len, start, len);
Chris Masona52d9a82007-08-27 16:49:44 -04007142 err = -EIO;
7143 goto out;
7144 }
Chris Masond1310b22008-01-24 16:13:08 -05007145
7146 err = 0;
Chris Mason890871b2009-09-02 16:24:52 -04007147 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -04007148 ret = add_extent_mapping(em_tree, em, 0);
Chris Mason3b951512008-04-17 11:29:12 -04007149 /* it is possible that someone inserted the extent into the tree
7150 * while we had the lock dropped. It is also possible that
7151 * an overlapping map exists in the tree
7152 */
Chris Masona52d9a82007-08-27 16:49:44 -04007153 if (ret == -EEXIST) {
Chris Mason3b951512008-04-17 11:29:12 -04007154 struct extent_map *existing;
Chris Masone6dcd2d2008-07-17 12:53:50 -04007155
7156 ret = 0;
7157
Qu Wenruoe6c4efd2014-09-17 11:53:35 +08007158 existing = search_extent_mapping(em_tree, start, len);
7159 /*
7160 * existing will always be non-NULL, since there must be
7161 * extent causing the -EEXIST.
7162 */
Chris Mason8dff9c82015-09-19 11:28:25 -07007163 if (existing->start == em->start &&
Omar Sandoval8e2bd3b2016-11-09 15:26:50 -08007164 extent_map_end(existing) >= extent_map_end(em) &&
Chris Mason8dff9c82015-09-19 11:28:25 -07007165 em->block_start == existing->block_start) {
7166 /*
Omar Sandoval8e2bd3b2016-11-09 15:26:50 -08007167 * The existing extent map already encompasses the
7168 * entire extent map we tried to add.
Chris Mason8dff9c82015-09-19 11:28:25 -07007169 */
7170 free_extent_map(em);
7171 em = existing;
7172 err = 0;
7173
7174 } else if (start >= extent_map_end(existing) ||
Qu Wenruo32be3a12014-09-22 09:13:03 +08007175 start <= existing->start) {
Qu Wenruoe6c4efd2014-09-17 11:53:35 +08007176 /*
7177 * The existing extent map is the one nearest to
7178 * the [start, start + len) range which overlaps
7179 */
7180 err = merge_extent_mapping(em_tree, existing,
7181 em, start);
Chris Masone1c4b742008-04-22 13:26:46 -04007182 free_extent_map(existing);
Qu Wenruoe6c4efd2014-09-17 11:53:35 +08007183 if (err) {
Chris Mason3b951512008-04-17 11:29:12 -04007184 free_extent_map(em);
7185 em = NULL;
7186 }
7187 } else {
7188 free_extent_map(em);
7189 em = existing;
Chris Masone6dcd2d2008-07-17 12:53:50 -04007190 err = 0;
Chris Masona52d9a82007-08-27 16:49:44 -04007191 }
Chris Masona52d9a82007-08-27 16:49:44 -04007192 }
Chris Mason890871b2009-09-02 16:24:52 -04007193 write_unlock(&em_tree->lock);
Chris Masona52d9a82007-08-27 16:49:44 -04007194out:
liubo1abe9b82011-03-24 11:18:59 +00007195
Nikolay Borisovfc4f21b2017-02-20 13:51:06 +02007196 trace_btrfs_get_extent(root, inode, em);
liubo1abe9b82011-03-24 11:18:59 +00007197
Tsutomu Itoh527afb42015-08-19 14:55:00 +09007198 btrfs_free_path(path);
Chris Masona52d9a82007-08-27 16:49:44 -04007199 if (trans) {
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04007200 ret = btrfs_end_transaction(trans);
Chris Masond3977122009-01-05 21:25:51 -05007201 if (!err)
Chris Masona52d9a82007-08-27 16:49:44 -04007202 err = ret;
7203 }
Chris Masona52d9a82007-08-27 16:49:44 -04007204 if (err) {
7205 free_extent_map(em);
Chris Masona52d9a82007-08-27 16:49:44 -04007206 return ERR_PTR(err);
7207 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007208 BUG_ON(!em); /* Error is always set */
Chris Masona52d9a82007-08-27 16:49:44 -04007209 return em;
7210}
7211
Nikolay Borisovfc4f21b2017-02-20 13:51:06 +02007212struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7213 struct page *page,
7214 size_t pg_offset, u64 start, u64 len,
7215 int create)
Chris Masonec29ed52011-02-23 16:23:20 -05007216{
7217 struct extent_map *em;
7218 struct extent_map *hole_em = NULL;
7219 u64 range_start = start;
7220 u64 end;
7221 u64 found;
7222 u64 found_end;
7223 int err = 0;
7224
7225 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7226 if (IS_ERR(em))
7227 return em;
Dan Carpenter99862772017-04-11 11:57:15 +03007228 /*
7229 * If our em maps to:
7230 * - a hole or
7231 * - a pre-alloc extent,
7232 * there might actually be delalloc bytes behind it.
7233 */
7234 if (em->block_start != EXTENT_MAP_HOLE &&
7235 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7236 return em;
7237 else
7238 hole_em = em;
Chris Masonec29ed52011-02-23 16:23:20 -05007239
7240 /* check to see if we've wrapped (len == -1 or similar) */
7241 end = start + len;
7242 if (end < start)
7243 end = (u64)-1;
7244 else
7245 end -= 1;
7246
7247 em = NULL;
7248
7249 /* ok, we didn't find anything, lets look for delalloc */
Nikolay Borisovfc4f21b2017-02-20 13:51:06 +02007250 found = count_range_bits(&inode->io_tree, &range_start,
Chris Masonec29ed52011-02-23 16:23:20 -05007251 end, len, EXTENT_DELALLOC, 1);
7252 found_end = range_start + found;
7253 if (found_end < range_start)
7254 found_end = (u64)-1;
7255
7256 /*
7257 * we didn't find anything useful, return
7258 * the original results from get_extent()
7259 */
7260 if (range_start > end || found_end <= start) {
7261 em = hole_em;
7262 hole_em = NULL;
7263 goto out;
7264 }
7265
7266 /* adjust the range_start to make sure it doesn't
7267 * go backwards from the start they passed in
7268 */
Dulshani Gunawardhana67871252013-10-31 10:33:04 +05307269 range_start = max(start, range_start);
Chris Masonec29ed52011-02-23 16:23:20 -05007270 found = found_end - range_start;
7271
7272 if (found > 0) {
7273 u64 hole_start = start;
7274 u64 hole_len = len;
7275
David Sterba172ddd62011-04-21 00:48:27 +02007276 em = alloc_extent_map();
Chris Masonec29ed52011-02-23 16:23:20 -05007277 if (!em) {
7278 err = -ENOMEM;
7279 goto out;
7280 }
7281 /*
7282 * when btrfs_get_extent can't find anything it
7283 * returns one huge hole
7284 *
7285 * make sure what it found really fits our range, and
7286 * adjust to make sure it is based on the start from
7287 * the caller
7288 */
7289 if (hole_em) {
7290 u64 calc_end = extent_map_end(hole_em);
7291
7292 if (calc_end <= start || (hole_em->start > end)) {
7293 free_extent_map(hole_em);
7294 hole_em = NULL;
7295 } else {
7296 hole_start = max(hole_em->start, start);
7297 hole_len = calc_end - hole_start;
7298 }
7299 }
7300 em->bdev = NULL;
7301 if (hole_em && range_start > hole_start) {
7302 /* our hole starts before our delalloc, so we
7303 * have to return just the parts of the hole
7304 * that go until the delalloc starts
7305 */
7306 em->len = min(hole_len,
7307 range_start - hole_start);
7308 em->start = hole_start;
7309 em->orig_start = hole_start;
7310 /*
7311 * don't adjust block start at all,
7312 * it is fixed at EXTENT_MAP_HOLE
7313 */
7314 em->block_start = hole_em->block_start;
7315 em->block_len = hole_len;
Liu Bof9e4fb52013-01-07 10:10:12 +00007316 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7317 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
Chris Masonec29ed52011-02-23 16:23:20 -05007318 } else {
7319 em->start = range_start;
7320 em->len = found;
7321 em->orig_start = range_start;
7322 em->block_start = EXTENT_MAP_DELALLOC;
7323 em->block_len = found;
7324 }
7325 } else if (hole_em) {
7326 return hole_em;
7327 }
7328out:
7329
7330 free_extent_map(hole_em);
7331 if (err) {
7332 free_extent_map(em);
7333 return ERR_PTR(err);
7334 }
7335 return em;
7336}
7337
Filipe Manana5f9a8a52016-05-12 13:53:36 +01007338static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7339 const u64 start,
7340 const u64 len,
7341 const u64 orig_start,
7342 const u64 block_start,
7343 const u64 block_len,
7344 const u64 orig_block_len,
7345 const u64 ram_bytes,
7346 const int type)
7347{
7348 struct extent_map *em = NULL;
7349 int ret;
7350
Filipe Manana5f9a8a52016-05-12 13:53:36 +01007351 if (type != BTRFS_ORDERED_NOCOW) {
Liu Bo6f9994d2017-01-31 07:50:22 -08007352 em = create_io_em(inode, start, len, orig_start,
7353 block_start, block_len, orig_block_len,
7354 ram_bytes,
7355 BTRFS_COMPRESS_NONE, /* compress_type */
7356 type);
Filipe Manana5f9a8a52016-05-12 13:53:36 +01007357 if (IS_ERR(em))
7358 goto out;
7359 }
7360 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7361 len, block_len, type);
7362 if (ret) {
7363 if (em) {
7364 free_extent_map(em);
Nikolay Borisovdcdbc052017-02-20 13:50:45 +02007365 btrfs_drop_extent_cache(BTRFS_I(inode), start,
Filipe Manana5f9a8a52016-05-12 13:53:36 +01007366 start + len - 1, 0);
7367 }
7368 em = ERR_PTR(ret);
7369 }
7370 out:
Filipe Manana5f9a8a52016-05-12 13:53:36 +01007371
7372 return em;
7373}
7374
Josef Bacik4b46fce2010-05-23 11:00:55 -04007375static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7376 u64 start, u64 len)
7377{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04007378 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Josef Bacik4b46fce2010-05-23 11:00:55 -04007379 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik70c8a912012-10-11 16:54:30 -04007380 struct extent_map *em;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007381 struct btrfs_key ins;
7382 u64 alloc_hint;
7383 int ret;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007384
Josef Bacik4b46fce2010-05-23 11:00:55 -04007385 alloc_hint = get_extent_allocation_hint(inode, start, len);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04007386 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
Jeff Mahoneyda170662016-06-15 09:22:56 -04007387 0, alloc_hint, &ins, 1, 1);
Josef Bacik00361582013-08-14 14:02:47 -04007388 if (ret)
7389 return ERR_PTR(ret);
Josef Bacik4b46fce2010-05-23 11:00:55 -04007390
Filipe Manana5f9a8a52016-05-12 13:53:36 +01007391 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7392 ins.objectid, ins.offset, ins.offset,
Liu Bo6288d6e2017-02-21 12:12:58 -08007393 ins.offset, BTRFS_ORDERED_REGULAR);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04007394 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
Filipe Manana5f9a8a52016-05-12 13:53:36 +01007395 if (IS_ERR(em))
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04007396 btrfs_free_reserved_extent(fs_info, ins.objectid,
7397 ins.offset, 1);
Josef Bacik00361582013-08-14 14:02:47 -04007398
Josef Bacik4b46fce2010-05-23 11:00:55 -04007399 return em;
7400}
7401
Chris Mason46bfbb52010-05-26 11:04:10 -04007402/*
7403 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7404 * block must be cow'd
7405 */
Josef Bacik00361582013-08-14 14:02:47 -04007406noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
Josef Bacik7ee9e442013-06-21 16:37:03 -04007407 u64 *orig_start, u64 *orig_block_len,
7408 u64 *ram_bytes)
Chris Mason46bfbb52010-05-26 11:04:10 -04007409{
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04007410 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Chris Mason46bfbb52010-05-26 11:04:10 -04007411 struct btrfs_path *path;
7412 int ret;
7413 struct extent_buffer *leaf;
7414 struct btrfs_root *root = BTRFS_I(inode)->root;
Miao Xie7b2b7082014-02-27 13:58:05 +08007415 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Chris Mason46bfbb52010-05-26 11:04:10 -04007416 struct btrfs_file_extent_item *fi;
7417 struct btrfs_key key;
7418 u64 disk_bytenr;
7419 u64 backref_offset;
7420 u64 extent_end;
7421 u64 num_bytes;
7422 int slot;
7423 int found_type;
Josef Bacik7ee9e442013-06-21 16:37:03 -04007424 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
Miao Xiee77751a2013-12-27 21:11:50 +08007425
Chris Mason46bfbb52010-05-26 11:04:10 -04007426 path = btrfs_alloc_path();
7427 if (!path)
7428 return -ENOMEM;
7429
David Sterbaf85b7372017-01-20 14:54:07 +01007430 ret = btrfs_lookup_file_extent(NULL, root, path,
7431 btrfs_ino(BTRFS_I(inode)), offset, 0);
Chris Mason46bfbb52010-05-26 11:04:10 -04007432 if (ret < 0)
7433 goto out;
7434
7435 slot = path->slots[0];
7436 if (ret == 1) {
7437 if (slot == 0) {
7438 /* can't find the item, must cow */
7439 ret = 0;
7440 goto out;
7441 }
7442 slot--;
7443 }
7444 ret = 0;
7445 leaf = path->nodes[0];
7446 btrfs_item_key_to_cpu(leaf, &key, slot);
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02007447 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
Chris Mason46bfbb52010-05-26 11:04:10 -04007448 key.type != BTRFS_EXTENT_DATA_KEY) {
7449 /* not our file or wrong item type, must cow */
7450 goto out;
7451 }
7452
7453 if (key.offset > offset) {
7454 /* Wrong offset, must cow */
7455 goto out;
7456 }
7457
7458 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7459 found_type = btrfs_file_extent_type(leaf, fi);
7460 if (found_type != BTRFS_FILE_EXTENT_REG &&
7461 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7462 /* not a regular extent, must cow */
7463 goto out;
7464 }
Josef Bacik7ee9e442013-06-21 16:37:03 -04007465
7466 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7467 goto out;
7468
Miao Xiee77751a2013-12-27 21:11:50 +08007469 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7470 if (extent_end <= offset)
7471 goto out;
7472
Chris Mason46bfbb52010-05-26 11:04:10 -04007473 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
Josef Bacik7ee9e442013-06-21 16:37:03 -04007474 if (disk_bytenr == 0)
7475 goto out;
7476
7477 if (btrfs_file_extent_compression(leaf, fi) ||
7478 btrfs_file_extent_encryption(leaf, fi) ||
7479 btrfs_file_extent_other_encoding(leaf, fi))
7480 goto out;
7481
Chris Mason46bfbb52010-05-26 11:04:10 -04007482 backref_offset = btrfs_file_extent_offset(leaf, fi);
7483
Josef Bacik7ee9e442013-06-21 16:37:03 -04007484 if (orig_start) {
7485 *orig_start = key.offset - backref_offset;
7486 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7487 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7488 }
Josef Bacikeb384b52013-04-24 16:32:55 -04007489
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04007490 if (btrfs_extent_readonly(fs_info, disk_bytenr))
Chris Mason46bfbb52010-05-26 11:04:10 -04007491 goto out;
Miao Xie7b2b7082014-02-27 13:58:05 +08007492
7493 num_bytes = min(offset + *len, extent_end) - offset;
7494 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7495 u64 range_end;
7496
Jeff Mahoneyda170662016-06-15 09:22:56 -04007497 range_end = round_up(offset + num_bytes,
7498 root->fs_info->sectorsize) - 1;
Miao Xie7b2b7082014-02-27 13:58:05 +08007499 ret = test_range_bit(io_tree, offset, range_end,
7500 EXTENT_DELALLOC, 0, NULL);
7501 if (ret) {
7502 ret = -EAGAIN;
7503 goto out;
7504 }
7505 }
7506
Josef Bacik1bda19e2013-10-18 12:10:36 -04007507 btrfs_release_path(path);
Chris Mason46bfbb52010-05-26 11:04:10 -04007508
7509 /*
7510 * look for other files referencing this extent, if we
7511 * find any we must cow
7512 */
Josef Bacik00361582013-08-14 14:02:47 -04007513
Liu Boe4c3b2d2017-01-30 12:25:28 -08007514 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
Josef Bacik00361582013-08-14 14:02:47 -04007515 key.offset - backref_offset, disk_bytenr);
Josef Bacik00361582013-08-14 14:02:47 -04007516 if (ret) {
7517 ret = 0;
7518 goto out;
7519 }
Chris Mason46bfbb52010-05-26 11:04:10 -04007520
7521 /*
7522 * adjust disk_bytenr and num_bytes to cover just the bytes
7523 * in this extent we are about to write. If there
7524 * are any csums in that range we have to cow in order
7525 * to keep the csums correct
7526 */
7527 disk_bytenr += backref_offset;
7528 disk_bytenr += offset - key.offset;
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04007529 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7530 goto out;
Chris Mason46bfbb52010-05-26 11:04:10 -04007531 /*
7532 * all of the above have passed, it is safe to overwrite this extent
7533 * without cow
7534 */
Josef Bacikeb384b52013-04-24 16:32:55 -04007535 *len = num_bytes;
Chris Mason46bfbb52010-05-26 11:04:10 -04007536 ret = 1;
7537out:
7538 btrfs_free_path(path);
7539 return ret;
7540}
7541
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07007542bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7543{
7544 struct radix_tree_root *root = &inode->i_mapping->page_tree;
David Sterbae03733d2017-05-12 01:02:22 +02007545 bool found = false;
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07007546 void **pagep = NULL;
7547 struct page *page = NULL;
David Sterbacc2b7022017-05-12 01:03:52 +02007548 unsigned long start_idx;
7549 unsigned long end_idx;
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07007550
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03007551 start_idx = start >> PAGE_SHIFT;
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07007552
7553 /*
7554 * end is the last byte in the last page. end == start is legal
7555 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03007556 end_idx = end >> PAGE_SHIFT;
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07007557
7558 rcu_read_lock();
7559
7560 /* Most of the code in this while loop is lifted from
7561 * find_get_page. It's been modified to begin searching from a
7562 * page and return just the first page found in that range. If the
7563 * found idx is less than or equal to the end idx then we know that
7564 * a page exists. If no pages are found or if those pages are
7565 * outside of the range then we're fine (yay!) */
7566 while (page == NULL &&
7567 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7568 page = radix_tree_deref_slot(pagep);
7569 if (unlikely(!page))
7570 break;
7571
7572 if (radix_tree_exception(page)) {
Filipe Manana809f9012014-06-05 13:22:25 +01007573 if (radix_tree_deref_retry(page)) {
7574 page = NULL;
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07007575 continue;
Filipe Manana809f9012014-06-05 13:22:25 +01007576 }
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07007577 /*
7578 * Otherwise, shmem/tmpfs must be storing a swap entry
7579 * here as an exceptional entry: so return it without
7580 * attempting to raise page count.
7581 */
Filipe Manana6fdef6d2014-06-05 13:22:26 +01007582 page = NULL;
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07007583 break; /* TODO: Is this relevant for this use case? */
7584 }
7585
Filipe Manana91405152014-06-05 13:22:24 +01007586 if (!page_cache_get_speculative(page)) {
7587 page = NULL;
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07007588 continue;
Filipe Manana91405152014-06-05 13:22:24 +01007589 }
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07007590
7591 /*
7592 * Has the page moved?
7593 * This is part of the lockless pagecache protocol. See
7594 * include/linux/pagemap.h for details.
7595 */
7596 if (unlikely(page != *pagep)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03007597 put_page(page);
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07007598 page = NULL;
7599 }
7600 }
7601
7602 if (page) {
7603 if (page->index <= end_idx)
7604 found = true;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03007605 put_page(page);
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07007606 }
7607
7608 rcu_read_unlock();
7609 return found;
7610}
7611
Josef Bacikeb838e72012-07-31 16:28:48 -04007612static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7613 struct extent_state **cached_state, int writing)
7614{
7615 struct btrfs_ordered_extent *ordered;
7616 int ret = 0;
7617
7618 while (1) {
7619 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
David Sterbaff13db42015-12-03 14:30:40 +01007620 cached_state);
Josef Bacikeb838e72012-07-31 16:28:48 -04007621 /*
7622 * We're concerned with the entire range that we're going to be
Nicholas D Steeves01327612016-05-19 21:18:45 -04007623 * doing DIO to, so we need to make sure there's no ordered
Josef Bacikeb838e72012-07-31 16:28:48 -04007624 * extents in this range.
7625 */
Nikolay Borisova776c6f2017-02-20 13:50:49 +02007626 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
Josef Bacikeb838e72012-07-31 16:28:48 -04007627 lockend - lockstart + 1);
7628
7629 /*
7630 * We need to make sure there are no buffered pages in this
7631 * range either, we could have raced between the invalidate in
7632 * generic_file_direct_write and locking the extent. The
7633 * invalidate needs to happen so that reads after a write do not
7634 * get stale data.
7635 */
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07007636 if (!ordered &&
7637 (!writing ||
7638 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
Josef Bacikeb838e72012-07-31 16:28:48 -04007639 break;
7640
7641 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7642 cached_state, GFP_NOFS);
7643
7644 if (ordered) {
Filipe Mananaade77022016-02-18 14:28:55 +00007645 /*
7646 * If we are doing a DIO read and the ordered extent we
7647 * found is for a buffered write, we can not wait for it
7648 * to complete and retry, because if we do so we can
7649 * deadlock with concurrent buffered writes on page
7650 * locks. This happens only if our DIO read covers more
7651 * than one extent map, if at this point has already
7652 * created an ordered extent for a previous extent map
7653 * and locked its range in the inode's io tree, and a
7654 * concurrent write against that previous extent map's
7655 * range and this range started (we unlock the ranges
7656 * in the io tree only when the bios complete and
7657 * buffered writes always lock pages before attempting
7658 * to lock range in the io tree).
7659 */
7660 if (writing ||
7661 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7662 btrfs_start_ordered_extent(inode, ordered, 1);
7663 else
7664 ret = -ENOTBLK;
Josef Bacikeb838e72012-07-31 16:28:48 -04007665 btrfs_put_ordered_extent(ordered);
7666 } else {
Josef Bacikeb838e72012-07-31 16:28:48 -04007667 /*
Filipe Mananab850ae12015-12-08 16:23:16 +00007668 * We could trigger writeback for this range (and wait
7669 * for it to complete) and then invalidate the pages for
7670 * this range (through invalidate_inode_pages2_range()),
7671 * but that can lead us to a deadlock with a concurrent
7672 * call to readpages() (a buffered read or a defrag call
7673 * triggered a readahead) on a page lock due to an
7674 * ordered dio extent we created before but did not have
7675 * yet a corresponding bio submitted (whence it can not
7676 * complete), which makes readpages() wait for that
7677 * ordered extent to complete while holding a lock on
7678 * that page.
Josef Bacikeb838e72012-07-31 16:28:48 -04007679 */
Filipe Mananab850ae12015-12-08 16:23:16 +00007680 ret = -ENOTBLK;
Josef Bacikeb838e72012-07-31 16:28:48 -04007681 }
7682
Filipe Mananaade77022016-02-18 14:28:55 +00007683 if (ret)
7684 break;
7685
Josef Bacikeb838e72012-07-31 16:28:48 -04007686 cond_resched();
7687 }
7688
7689 return ret;
7690}
7691
Liu Bo6f9994d2017-01-31 07:50:22 -08007692/* The callers of this must take lock_extent() */
7693static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7694 u64 orig_start, u64 block_start,
7695 u64 block_len, u64 orig_block_len,
7696 u64 ram_bytes, int compress_type,
7697 int type)
Josef Bacik69ffb542012-09-11 15:40:07 -04007698{
7699 struct extent_map_tree *em_tree;
7700 struct extent_map *em;
7701 struct btrfs_root *root = BTRFS_I(inode)->root;
7702 int ret;
7703
Liu Bo6f9994d2017-01-31 07:50:22 -08007704 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7705 type == BTRFS_ORDERED_COMPRESSED ||
7706 type == BTRFS_ORDERED_NOCOW ||
Liu Bo1af4a0a2017-02-13 15:35:09 -08007707 type == BTRFS_ORDERED_REGULAR);
Liu Bo6f9994d2017-01-31 07:50:22 -08007708
Josef Bacik69ffb542012-09-11 15:40:07 -04007709 em_tree = &BTRFS_I(inode)->extent_tree;
7710 em = alloc_extent_map();
7711 if (!em)
7712 return ERR_PTR(-ENOMEM);
7713
7714 em->start = start;
7715 em->orig_start = orig_start;
7716 em->len = len;
7717 em->block_len = block_len;
7718 em->block_start = block_start;
7719 em->bdev = root->fs_info->fs_devices->latest_bdev;
Josef Bacikb4939682012-12-03 10:31:19 -05007720 em->orig_block_len = orig_block_len;
Josef Bacikcc95bef2013-04-04 14:31:27 -04007721 em->ram_bytes = ram_bytes;
Josef Bacik70c8a912012-10-11 16:54:30 -04007722 em->generation = -1;
Josef Bacik69ffb542012-09-11 15:40:07 -04007723 set_bit(EXTENT_FLAG_PINNED, &em->flags);
Liu Bo1af4a0a2017-02-13 15:35:09 -08007724 if (type == BTRFS_ORDERED_PREALLOC) {
Josef Bacikb11e2342012-12-03 10:58:15 -05007725 set_bit(EXTENT_FLAG_FILLING, &em->flags);
Liu Bo1af4a0a2017-02-13 15:35:09 -08007726 } else if (type == BTRFS_ORDERED_COMPRESSED) {
Liu Bo6f9994d2017-01-31 07:50:22 -08007727 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7728 em->compress_type = compress_type;
7729 }
Josef Bacik69ffb542012-09-11 15:40:07 -04007730
7731 do {
Nikolay Borisovdcdbc052017-02-20 13:50:45 +02007732 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
Josef Bacik69ffb542012-09-11 15:40:07 -04007733 em->start + em->len - 1, 0);
7734 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -04007735 ret = add_extent_mapping(em_tree, em, 1);
Josef Bacik69ffb542012-09-11 15:40:07 -04007736 write_unlock(&em_tree->lock);
Liu Bo6f9994d2017-01-31 07:50:22 -08007737 /*
7738 * The caller has taken lock_extent(), who could race with us
7739 * to add em?
7740 */
Josef Bacik69ffb542012-09-11 15:40:07 -04007741 } while (ret == -EEXIST);
7742
7743 if (ret) {
7744 free_extent_map(em);
7745 return ERR_PTR(ret);
7746 }
7747
Liu Bo6f9994d2017-01-31 07:50:22 -08007748 /* em got 2 refs now, callers needs to do free_extent_map once. */
Josef Bacik69ffb542012-09-11 15:40:07 -04007749 return em;
7750}
7751
Josef Bacik4b46fce2010-05-23 11:00:55 -04007752static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7753 struct buffer_head *bh_result, int create)
7754{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04007755 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Josef Bacik4b46fce2010-05-23 11:00:55 -04007756 struct extent_map *em;
Josef Bacikeb838e72012-07-31 16:28:48 -04007757 struct extent_state *cached_state = NULL;
chandan50745b02015-08-28 21:10:13 +05307758 struct btrfs_dio_data *dio_data = NULL;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007759 u64 start = iblock << inode->i_blkbits;
Josef Bacikeb838e72012-07-31 16:28:48 -04007760 u64 lockstart, lockend;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007761 u64 len = bh_result->b_size;
Josef Bacikeb838e72012-07-31 16:28:48 -04007762 int unlock_bits = EXTENT_LOCKED;
Miao Xie09348562013-02-07 10:12:07 +00007763 int ret = 0;
Josef Bacikeb838e72012-07-31 16:28:48 -04007764
Miao Xie172a5042013-02-21 02:48:22 -07007765 if (create)
Josef Bacik32667892015-02-11 15:08:58 -05007766 unlock_bits |= EXTENT_DIRTY;
Miao Xie172a5042013-02-21 02:48:22 -07007767 else
Jeff Mahoney0b246af2016-06-22 18:54:23 -04007768 len = min_t(u64, len, fs_info->sectorsize);
Josef Bacikeb838e72012-07-31 16:28:48 -04007769
Josef Bacikc3298612012-08-03 16:49:19 -04007770 lockstart = start;
7771 lockend = start + len - 1;
7772
Josef Bacike1cbbfa2015-03-17 10:52:28 -04007773 if (current->journal_info) {
7774 /*
7775 * Need to pull our outstanding extents and set journal_info to NULL so
Nicholas D Steeves01327612016-05-19 21:18:45 -04007776 * that anything that needs to check if there's a transaction doesn't get
Josef Bacike1cbbfa2015-03-17 10:52:28 -04007777 * confused.
7778 */
chandan50745b02015-08-28 21:10:13 +05307779 dio_data = current->journal_info;
Josef Bacike1cbbfa2015-03-17 10:52:28 -04007780 current->journal_info = NULL;
7781 }
7782
Josef Bacikeb838e72012-07-31 16:28:48 -04007783 /*
7784 * If this errors out it's because we couldn't invalidate pagecache for
7785 * this range and we need to fallback to buffered.
7786 */
Filipe Manana9c9464c2015-11-04 09:52:04 +00007787 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7788 create)) {
7789 ret = -ENOTBLK;
7790 goto err;
7791 }
Josef Bacikeb838e72012-07-31 16:28:48 -04007792
Nikolay Borisovfc4f21b2017-02-20 13:51:06 +02007793 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
Josef Bacikeb838e72012-07-31 16:28:48 -04007794 if (IS_ERR(em)) {
7795 ret = PTR_ERR(em);
7796 goto unlock_err;
7797 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04007798
7799 /*
7800 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7801 * io. INLINE is special, and we could probably kludge it in here, but
7802 * it's still buffered so for safety lets just fall back to the generic
7803 * buffered path.
7804 *
7805 * For COMPRESSED we _have_ to read the entire extent in so we can
7806 * decompress it, so there will be buffering required no matter what we
7807 * do, so go ahead and fallback to buffered.
7808 *
Nicholas D Steeves01327612016-05-19 21:18:45 -04007809 * We return -ENOTBLK because that's what makes DIO go ahead and go back
Josef Bacik4b46fce2010-05-23 11:00:55 -04007810 * to buffered IO. Don't blame me, this is the price we pay for using
7811 * the generic code.
7812 */
7813 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7814 em->block_start == EXTENT_MAP_INLINE) {
7815 free_extent_map(em);
Josef Bacikeb838e72012-07-31 16:28:48 -04007816 ret = -ENOTBLK;
7817 goto unlock_err;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007818 }
7819
7820 /* Just a good old fashioned hole, return */
7821 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7822 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7823 free_extent_map(em);
Josef Bacikeb838e72012-07-31 16:28:48 -04007824 goto unlock_err;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007825 }
7826
7827 /*
7828 * We don't allocate a new extent in the following cases
7829 *
7830 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7831 * existing extent.
7832 * 2) The extent is marked as PREALLOC. We're good to go here and can
7833 * just use the extent.
7834 *
7835 */
Chris Mason46bfbb52010-05-26 11:04:10 -04007836 if (!create) {
Josef Bacikeb838e72012-07-31 16:28:48 -04007837 len = min(len, em->len - (start - em->start));
7838 lockstart = start + len;
7839 goto unlock;
Chris Mason46bfbb52010-05-26 11:04:10 -04007840 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04007841
7842 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7843 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7844 em->block_start != EXTENT_MAP_HOLE)) {
Josef Bacik4b46fce2010-05-23 11:00:55 -04007845 int type;
Josef Bacikeb384b52013-04-24 16:32:55 -04007846 u64 block_start, orig_start, orig_block_len, ram_bytes;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007847
7848 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7849 type = BTRFS_ORDERED_PREALLOC;
7850 else
7851 type = BTRFS_ORDERED_NOCOW;
Chris Mason46bfbb52010-05-26 11:04:10 -04007852 len = min(len, em->len - (start - em->start));
Josef Bacik4b46fce2010-05-23 11:00:55 -04007853 block_start = em->block_start + (start - em->start);
Chris Mason46bfbb52010-05-26 11:04:10 -04007854
Josef Bacik00361582013-08-14 14:02:47 -04007855 if (can_nocow_extent(inode, start, &len, &orig_start,
Filipe Mananaf78c4362016-05-09 13:15:41 +01007856 &orig_block_len, &ram_bytes) == 1 &&
Jeff Mahoney0b246af2016-06-22 18:54:23 -04007857 btrfs_inc_nocow_writers(fs_info, block_start)) {
Filipe Manana5f9a8a52016-05-12 13:53:36 +01007858 struct extent_map *em2;
Filipe Manana0b901912016-05-09 13:15:27 +01007859
Filipe Manana5f9a8a52016-05-12 13:53:36 +01007860 em2 = btrfs_create_dio_extent(inode, start, len,
7861 orig_start, block_start,
7862 len, orig_block_len,
7863 ram_bytes, type);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04007864 btrfs_dec_nocow_writers(fs_info, block_start);
Josef Bacik69ffb542012-09-11 15:40:07 -04007865 if (type == BTRFS_ORDERED_PREALLOC) {
7866 free_extent_map(em);
Filipe Manana5f9a8a52016-05-12 13:53:36 +01007867 em = em2;
Josef Bacik69ffb542012-09-11 15:40:07 -04007868 }
Filipe Manana5f9a8a52016-05-12 13:53:36 +01007869 if (em2 && IS_ERR(em2)) {
7870 ret = PTR_ERR(em2);
Josef Bacikeb838e72012-07-31 16:28:48 -04007871 goto unlock_err;
Chris Mason46bfbb52010-05-26 11:04:10 -04007872 }
Wang Xiaoguang18513092016-07-25 15:51:40 +08007873 /*
7874 * For inode marked NODATACOW or extent marked PREALLOC,
7875 * use the existing or preallocated extent, so does not
7876 * need to adjust btrfs_space_info's bytes_may_use.
7877 */
7878 btrfs_free_reserved_data_space_noquota(inode,
7879 start, len);
Chris Mason46bfbb52010-05-26 11:04:10 -04007880 goto unlock;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007881 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04007882 }
Josef Bacik00361582013-08-14 14:02:47 -04007883
Chris Mason46bfbb52010-05-26 11:04:10 -04007884 /*
7885 * this will cow the extent, reset the len in case we changed
7886 * it above
7887 */
7888 len = bh_result->b_size;
Josef Bacik70c8a912012-10-11 16:54:30 -04007889 free_extent_map(em);
7890 em = btrfs_new_extent_direct(inode, start, len);
Josef Bacikeb838e72012-07-31 16:28:48 -04007891 if (IS_ERR(em)) {
7892 ret = PTR_ERR(em);
7893 goto unlock_err;
7894 }
Chris Mason46bfbb52010-05-26 11:04:10 -04007895 len = min(len, em->len - (start - em->start));
7896unlock:
Josef Bacik4b46fce2010-05-23 11:00:55 -04007897 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7898 inode->i_blkbits;
Chris Mason46bfbb52010-05-26 11:04:10 -04007899 bh_result->b_size = len;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007900 bh_result->b_bdev = em->bdev;
7901 set_buffer_mapped(bh_result);
Josef Bacikc3473e82012-06-19 10:59:00 -04007902 if (create) {
7903 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7904 set_buffer_new(bh_result);
7905
7906 /*
7907 * Need to update the i_size under the extent lock so buffered
7908 * readers will get the updated i_size when we unlock.
7909 */
Liu Bo4aaedfb2016-12-14 22:36:05 -08007910 if (!dio_data->overwrite && start + len > i_size_read(inode))
Josef Bacikc3473e82012-06-19 10:59:00 -04007911 i_size_write(inode, start + len);
Miao Xie09348562013-02-07 10:12:07 +00007912
chandan50745b02015-08-28 21:10:13 +05307913 WARN_ON(dio_data->reserve < len);
7914 dio_data->reserve -= len;
Filipe Mananaf28a4922015-12-08 19:23:20 +00007915 dio_data->unsubmitted_oe_range_end = start + len;
chandan50745b02015-08-28 21:10:13 +05307916 current->journal_info = dio_data;
Josef Bacikc3473e82012-06-19 10:59:00 -04007917 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04007918
Josef Bacikeb838e72012-07-31 16:28:48 -04007919 /*
7920 * In the case of write we need to clear and unlock the entire range,
7921 * in the case of read we need to unlock only the end area that we
7922 * aren't using if there is any left over space.
7923 */
Liu Bo24c03fa2012-08-22 20:10:38 -06007924 if (lockstart < lockend) {
Miao Xie09348562013-02-07 10:12:07 +00007925 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7926 lockend, unlock_bits, 1, 0,
7927 &cached_state, GFP_NOFS);
Liu Bo24c03fa2012-08-22 20:10:38 -06007928 } else {
Josef Bacikeb838e72012-07-31 16:28:48 -04007929 free_extent_state(cached_state);
Liu Bo24c03fa2012-08-22 20:10:38 -06007930 }
Josef Bacikeb838e72012-07-31 16:28:48 -04007931
Josef Bacik4b46fce2010-05-23 11:00:55 -04007932 free_extent_map(em);
7933
7934 return 0;
Josef Bacikeb838e72012-07-31 16:28:48 -04007935
7936unlock_err:
Josef Bacikeb838e72012-07-31 16:28:48 -04007937 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7938 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
Filipe Manana9c9464c2015-11-04 09:52:04 +00007939err:
chandan50745b02015-08-28 21:10:13 +05307940 if (dio_data)
7941 current->journal_info = dio_data;
Josef Bacikeb838e72012-07-31 16:28:48 -04007942 return ret;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007943}
7944
Omar Sandoval58efbc92017-08-22 23:45:59 -07007945static inline blk_status_t submit_dio_repair_bio(struct inode *inode,
7946 struct bio *bio,
7947 int mirror_num)
Miao Xie8b110e32014-09-12 18:44:03 +08007948{
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04007949 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Omar Sandoval58efbc92017-08-22 23:45:59 -07007950 blk_status_t ret;
Miao Xie8b110e32014-09-12 18:44:03 +08007951
Mike Christie37226b22016-06-05 14:31:52 -05007952 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
Miao Xie8b110e32014-09-12 18:44:03 +08007953
7954 bio_get(bio);
7955
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04007956 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
Miao Xie8b110e32014-09-12 18:44:03 +08007957 if (ret)
7958 goto err;
7959
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04007960 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
Miao Xie8b110e32014-09-12 18:44:03 +08007961err:
7962 bio_put(bio);
7963 return ret;
7964}
7965
7966static int btrfs_check_dio_repairable(struct inode *inode,
7967 struct bio *failed_bio,
7968 struct io_failure_record *failrec,
7969 int failed_mirror)
7970{
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04007971 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Miao Xie8b110e32014-09-12 18:44:03 +08007972 int num_copies;
7973
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04007974 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
Miao Xie8b110e32014-09-12 18:44:03 +08007975 if (num_copies == 1) {
7976 /*
7977 * we only have a single copy of the data, so don't bother with
7978 * all the retry and error correction code that follows. no
7979 * matter what the error is, it is very likely to persist.
7980 */
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04007981 btrfs_debug(fs_info,
7982 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7983 num_copies, failrec->this_mirror, failed_mirror);
Miao Xie8b110e32014-09-12 18:44:03 +08007984 return 0;
7985 }
7986
7987 failrec->failed_mirror = failed_mirror;
7988 failrec->this_mirror++;
7989 if (failrec->this_mirror == failed_mirror)
7990 failrec->this_mirror++;
7991
7992 if (failrec->this_mirror > num_copies) {
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04007993 btrfs_debug(fs_info,
7994 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7995 num_copies, failrec->this_mirror, failed_mirror);
Miao Xie8b110e32014-09-12 18:44:03 +08007996 return 0;
7997 }
7998
7999 return 1;
8000}
8001
Omar Sandoval58efbc92017-08-22 23:45:59 -07008002static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio,
8003 struct page *page, unsigned int pgoff,
8004 u64 start, u64 end, int failed_mirror,
8005 bio_end_io_t *repair_endio, void *repair_arg)
Miao Xie8b110e32014-09-12 18:44:03 +08008006{
8007 struct io_failure_record *failrec;
Josef Bacik7870d082017-05-05 11:57:15 -04008008 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8009 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
Miao Xie8b110e32014-09-12 18:44:03 +08008010 struct bio *bio;
8011 int isector;
David Sterbaf1c77c52017-06-06 19:03:49 +02008012 unsigned int read_mode = 0;
Liu Bo17347ce2017-05-15 15:33:27 -07008013 int segs;
Miao Xie8b110e32014-09-12 18:44:03 +08008014 int ret;
Omar Sandoval58efbc92017-08-22 23:45:59 -07008015 blk_status_t status;
Miao Xie8b110e32014-09-12 18:44:03 +08008016
Mike Christie37226b22016-06-05 14:31:52 -05008017 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
Miao Xie8b110e32014-09-12 18:44:03 +08008018
8019 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
8020 if (ret)
Omar Sandoval58efbc92017-08-22 23:45:59 -07008021 return errno_to_blk_status(ret);
Miao Xie8b110e32014-09-12 18:44:03 +08008022
8023 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
8024 failed_mirror);
8025 if (!ret) {
Josef Bacik7870d082017-05-05 11:57:15 -04008026 free_io_failure(failure_tree, io_tree, failrec);
Omar Sandoval58efbc92017-08-22 23:45:59 -07008027 return BLK_STS_IOERR;
Miao Xie8b110e32014-09-12 18:44:03 +08008028 }
8029
Liu Bo17347ce2017-05-15 15:33:27 -07008030 segs = bio_segments(failed_bio);
8031 if (segs > 1 ||
8032 (failed_bio->bi_io_vec->bv_len > btrfs_inode_sectorsize(inode)))
Christoph Hellwig70fd7612016-11-01 07:40:10 -06008033 read_mode |= REQ_FAILFAST_DEV;
Miao Xie8b110e32014-09-12 18:44:03 +08008034
8035 isector = start - btrfs_io_bio(failed_bio)->logical;
8036 isector >>= inode->i_sb->s_blocksize_bits;
8037 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308038 pgoff, isector, repair_endio, repair_arg);
Mike Christie37226b22016-06-05 14:31:52 -05008039 bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
Miao Xie8b110e32014-09-12 18:44:03 +08008040
8041 btrfs_debug(BTRFS_I(inode)->root->fs_info,
David Sterba913e1532017-07-13 15:32:18 +02008042 "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
Miao Xie8b110e32014-09-12 18:44:03 +08008043 read_mode, failrec->this_mirror, failrec->in_validation);
8044
Omar Sandoval58efbc92017-08-22 23:45:59 -07008045 status = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
8046 if (status) {
Josef Bacik7870d082017-05-05 11:57:15 -04008047 free_io_failure(failure_tree, io_tree, failrec);
Miao Xie8b110e32014-09-12 18:44:03 +08008048 bio_put(bio);
8049 }
8050
Omar Sandoval58efbc92017-08-22 23:45:59 -07008051 return status;
Miao Xie8b110e32014-09-12 18:44:03 +08008052}
8053
8054struct btrfs_retry_complete {
8055 struct completion done;
8056 struct inode *inode;
8057 u64 start;
8058 int uptodate;
8059};
8060
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02008061static void btrfs_retry_endio_nocsum(struct bio *bio)
Miao Xie8b110e32014-09-12 18:44:03 +08008062{
8063 struct btrfs_retry_complete *done = bio->bi_private;
Josef Bacik7870d082017-05-05 11:57:15 -04008064 struct inode *inode = done->inode;
Miao Xie8b110e32014-09-12 18:44:03 +08008065 struct bio_vec *bvec;
Josef Bacik7870d082017-05-05 11:57:15 -04008066 struct extent_io_tree *io_tree, *failure_tree;
Miao Xie8b110e32014-09-12 18:44:03 +08008067 int i;
8068
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02008069 if (bio->bi_status)
Miao Xie8b110e32014-09-12 18:44:03 +08008070 goto end;
8071
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308072 ASSERT(bio->bi_vcnt == 1);
Josef Bacik7870d082017-05-05 11:57:15 -04008073 io_tree = &BTRFS_I(inode)->io_tree;
8074 failure_tree = &BTRFS_I(inode)->io_failure_tree;
8075 ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(inode));
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308076
Miao Xie8b110e32014-09-12 18:44:03 +08008077 done->uptodate = 1;
David Sterbac09abff2017-07-13 18:10:07 +02008078 ASSERT(!bio_flagged(bio, BIO_CLONED));
Miao Xie8b110e32014-09-12 18:44:03 +08008079 bio_for_each_segment_all(bvec, bio, i)
Josef Bacik7870d082017-05-05 11:57:15 -04008080 clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
8081 io_tree, done->start, bvec->bv_page,
8082 btrfs_ino(BTRFS_I(inode)), 0);
Miao Xie8b110e32014-09-12 18:44:03 +08008083end:
8084 complete(&done->done);
8085 bio_put(bio);
8086}
8087
Omar Sandoval58efbc92017-08-22 23:45:59 -07008088static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode,
8089 struct btrfs_io_bio *io_bio)
Josef Bacik4b46fce2010-05-23 11:00:55 -04008090{
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308091 struct btrfs_fs_info *fs_info;
Liu Bo17347ce2017-05-15 15:33:27 -07008092 struct bio_vec bvec;
8093 struct bvec_iter iter;
Miao Xie8b110e32014-09-12 18:44:03 +08008094 struct btrfs_retry_complete done;
Josef Bacik4b46fce2010-05-23 11:00:55 -04008095 u64 start;
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308096 unsigned int pgoff;
8097 u32 sectorsize;
8098 int nr_sectors;
Omar Sandoval58efbc92017-08-22 23:45:59 -07008099 blk_status_t ret;
8100 blk_status_t err = BLK_STS_OK;
Miao Xiedc380ae2014-09-12 18:43:55 +08008101
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308102 fs_info = BTRFS_I(inode)->root->fs_info;
Jeff Mahoneyda170662016-06-15 09:22:56 -04008103 sectorsize = fs_info->sectorsize;
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308104
Miao Xiec1dc0892014-09-12 18:43:56 +08008105 start = io_bio->logical;
Miao Xie8b110e32014-09-12 18:44:03 +08008106 done.inode = inode;
Liu Bo17347ce2017-05-15 15:33:27 -07008107 io_bio->bio.bi_iter = io_bio->iter;
Miao Xie8b110e32014-09-12 18:44:03 +08008108
Liu Bo17347ce2017-05-15 15:33:27 -07008109 bio_for_each_segment(bvec, &io_bio->bio, iter) {
8110 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8111 pgoff = bvec.bv_offset;
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308112
8113next_block_or_try_again:
Miao Xie8b110e32014-09-12 18:44:03 +08008114 done.uptodate = 0;
8115 done.start = start;
8116 init_completion(&done.done);
8117
Liu Bo17347ce2017-05-15 15:33:27 -07008118 ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308119 pgoff, start, start + sectorsize - 1,
8120 io_bio->mirror_num,
8121 btrfs_retry_endio_nocsum, &done);
Liu Bo629ebf42017-05-15 17:20:07 -07008122 if (ret) {
8123 err = ret;
8124 goto next;
8125 }
Miao Xie8b110e32014-09-12 18:44:03 +08008126
David Sterba9c17f6c2017-07-19 19:26:45 +02008127 wait_for_completion_io(&done.done);
Miao Xie8b110e32014-09-12 18:44:03 +08008128
8129 if (!done.uptodate) {
8130 /* We might have another mirror, so try again */
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308131 goto next_block_or_try_again;
Miao Xie8b110e32014-09-12 18:44:03 +08008132 }
8133
Liu Bo629ebf42017-05-15 17:20:07 -07008134next:
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308135 start += sectorsize;
8136
Liu Bo97bf5a52017-04-07 13:11:10 -07008137 nr_sectors--;
8138 if (nr_sectors) {
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308139 pgoff += sectorsize;
Liu Bo97bf5a52017-04-07 13:11:10 -07008140 ASSERT(pgoff < PAGE_SIZE);
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308141 goto next_block_or_try_again;
8142 }
Miao Xie8b110e32014-09-12 18:44:03 +08008143 }
8144
Liu Bo629ebf42017-05-15 17:20:07 -07008145 return err;
Miao Xie8b110e32014-09-12 18:44:03 +08008146}
8147
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02008148static void btrfs_retry_endio(struct bio *bio)
Miao Xie8b110e32014-09-12 18:44:03 +08008149{
8150 struct btrfs_retry_complete *done = bio->bi_private;
8151 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
Josef Bacik7870d082017-05-05 11:57:15 -04008152 struct extent_io_tree *io_tree, *failure_tree;
8153 struct inode *inode = done->inode;
Miao Xie8b110e32014-09-12 18:44:03 +08008154 struct bio_vec *bvec;
8155 int uptodate;
8156 int ret;
8157 int i;
8158
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02008159 if (bio->bi_status)
Miao Xie8b110e32014-09-12 18:44:03 +08008160 goto end;
8161
8162 uptodate = 1;
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308163
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308164 ASSERT(bio->bi_vcnt == 1);
Liu Bo2e949b02017-04-05 14:04:19 -07008165 ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(done->inode));
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308166
Josef Bacik7870d082017-05-05 11:57:15 -04008167 io_tree = &BTRFS_I(inode)->io_tree;
8168 failure_tree = &BTRFS_I(inode)->io_failure_tree;
8169
David Sterbac09abff2017-07-13 18:10:07 +02008170 ASSERT(!bio_flagged(bio, BIO_CLONED));
Miao Xie8b110e32014-09-12 18:44:03 +08008171 bio_for_each_segment_all(bvec, bio, i) {
Josef Bacik7870d082017-05-05 11:57:15 -04008172 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
8173 bvec->bv_offset, done->start,
8174 bvec->bv_len);
Miao Xie8b110e32014-09-12 18:44:03 +08008175 if (!ret)
Josef Bacik7870d082017-05-05 11:57:15 -04008176 clean_io_failure(BTRFS_I(inode)->root->fs_info,
8177 failure_tree, io_tree, done->start,
8178 bvec->bv_page,
8179 btrfs_ino(BTRFS_I(inode)),
8180 bvec->bv_offset);
Miao Xie8b110e32014-09-12 18:44:03 +08008181 else
8182 uptodate = 0;
8183 }
8184
8185 done->uptodate = uptodate;
8186end:
8187 complete(&done->done);
8188 bio_put(bio);
8189}
8190
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02008191static blk_status_t __btrfs_subio_endio_read(struct inode *inode,
8192 struct btrfs_io_bio *io_bio, blk_status_t err)
Miao Xie8b110e32014-09-12 18:44:03 +08008193{
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308194 struct btrfs_fs_info *fs_info;
Liu Bo17347ce2017-05-15 15:33:27 -07008195 struct bio_vec bvec;
8196 struct bvec_iter iter;
Miao Xie8b110e32014-09-12 18:44:03 +08008197 struct btrfs_retry_complete done;
8198 u64 start;
8199 u64 offset = 0;
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308200 u32 sectorsize;
8201 int nr_sectors;
8202 unsigned int pgoff;
8203 int csum_pos;
Liu Boef7cdac12017-04-13 18:11:48 -07008204 bool uptodate = (err == 0);
Miao Xie8b110e32014-09-12 18:44:03 +08008205 int ret;
Omar Sandoval58efbc92017-08-22 23:45:59 -07008206 blk_status_t status;
Miao Xie8b110e32014-09-12 18:44:03 +08008207
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308208 fs_info = BTRFS_I(inode)->root->fs_info;
Jeff Mahoneyda170662016-06-15 09:22:56 -04008209 sectorsize = fs_info->sectorsize;
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308210
Omar Sandoval58efbc92017-08-22 23:45:59 -07008211 err = BLK_STS_OK;
Miao Xie8b110e32014-09-12 18:44:03 +08008212 start = io_bio->logical;
8213 done.inode = inode;
Liu Bo17347ce2017-05-15 15:33:27 -07008214 io_bio->bio.bi_iter = io_bio->iter;
Miao Xie8b110e32014-09-12 18:44:03 +08008215
Liu Bo17347ce2017-05-15 15:33:27 -07008216 bio_for_each_segment(bvec, &io_bio->bio, iter) {
8217 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308218
Liu Bo17347ce2017-05-15 15:33:27 -07008219 pgoff = bvec.bv_offset;
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308220next_block:
Liu Boef7cdac12017-04-13 18:11:48 -07008221 if (uptodate) {
8222 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8223 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8224 bvec.bv_page, pgoff, start, sectorsize);
8225 if (likely(!ret))
8226 goto next;
8227 }
Miao Xie8b110e32014-09-12 18:44:03 +08008228try_again:
8229 done.uptodate = 0;
8230 done.start = start;
8231 init_completion(&done.done);
8232
Omar Sandoval58efbc92017-08-22 23:45:59 -07008233 status = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8234 pgoff, start, start + sectorsize - 1,
8235 io_bio->mirror_num, btrfs_retry_endio,
8236 &done);
8237 if (status) {
8238 err = status;
Miao Xie8b110e32014-09-12 18:44:03 +08008239 goto next;
8240 }
8241
David Sterba9c17f6c2017-07-19 19:26:45 +02008242 wait_for_completion_io(&done.done);
Miao Xie8b110e32014-09-12 18:44:03 +08008243
8244 if (!done.uptodate) {
8245 /* We might have another mirror, so try again */
8246 goto try_again;
8247 }
8248next:
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308249 offset += sectorsize;
8250 start += sectorsize;
8251
8252 ASSERT(nr_sectors);
8253
Liu Bo97bf5a52017-04-07 13:11:10 -07008254 nr_sectors--;
8255 if (nr_sectors) {
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308256 pgoff += sectorsize;
Liu Bo97bf5a52017-04-07 13:11:10 -07008257 ASSERT(pgoff < PAGE_SIZE);
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308258 goto next_block;
8259 }
Kent Overstreet2c30c712013-11-07 12:20:26 -08008260 }
Miao Xiec1dc0892014-09-12 18:43:56 +08008261
8262 return err;
8263}
8264
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02008265static blk_status_t btrfs_subio_endio_read(struct inode *inode,
8266 struct btrfs_io_bio *io_bio, blk_status_t err)
Miao Xie8b110e32014-09-12 18:44:03 +08008267{
8268 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8269
8270 if (skip_csum) {
8271 if (unlikely(err))
8272 return __btrfs_correct_data_nocsum(inode, io_bio);
8273 else
Omar Sandoval58efbc92017-08-22 23:45:59 -07008274 return BLK_STS_OK;
Miao Xie8b110e32014-09-12 18:44:03 +08008275 } else {
8276 return __btrfs_subio_endio_read(inode, io_bio, err);
8277 }
8278}
8279
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02008280static void btrfs_endio_direct_read(struct bio *bio)
Miao Xiec1dc0892014-09-12 18:43:56 +08008281{
8282 struct btrfs_dio_private *dip = bio->bi_private;
8283 struct inode *inode = dip->inode;
8284 struct bio *dio_bio;
8285 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02008286 blk_status_t err = bio->bi_status;
Miao Xiec1dc0892014-09-12 18:43:56 +08008287
Liu Bo99c4e3b2017-09-15 15:06:51 -06008288 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
Miao Xie8b110e32014-09-12 18:44:03 +08008289 err = btrfs_subio_endio_read(inode, io_bio, err);
Miao Xiec1dc0892014-09-12 18:43:56 +08008290
Josef Bacik4b46fce2010-05-23 11:00:55 -04008291 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01008292 dip->logical_offset + dip->bytes - 1);
Chris Mason9be33952013-05-17 18:30:14 -04008293 dio_bio = dip->dio_bio;
Josef Bacik4b46fce2010-05-23 11:00:55 -04008294
Josef Bacik4b46fce2010-05-23 11:00:55 -04008295 kfree(dip);
Josef Bacikc0da7aa2011-03-22 11:05:07 -04008296
Liu Bo99c4e3b2017-09-15 15:06:51 -06008297 dio_bio->bi_status = err;
Christoph Hellwig40553512017-06-03 09:37:58 +02008298 dio_end_io(dio_bio);
Miao Xie23ea8e52014-09-12 18:43:54 +08008299
8300 if (io_bio->end_io)
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02008301 io_bio->end_io(io_bio, blk_status_to_errno(err));
Chris Mason9be33952013-05-17 18:30:14 -04008302 bio_put(bio);
Josef Bacik4b46fce2010-05-23 11:00:55 -04008303}
8304
Qu Wenruo524272602017-03-08 10:25:52 +08008305static void __endio_write_update_ordered(struct inode *inode,
8306 const u64 offset, const u64 bytes,
8307 const bool uptodate)
Josef Bacik4b46fce2010-05-23 11:00:55 -04008308{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04008309 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Josef Bacik4b46fce2010-05-23 11:00:55 -04008310 struct btrfs_ordered_extent *ordered = NULL;
Qu Wenruo524272602017-03-08 10:25:52 +08008311 struct btrfs_workqueue *wq;
8312 btrfs_work_func_t func;
Filipe Manana14543772015-11-24 16:23:54 +00008313 u64 ordered_offset = offset;
8314 u64 ordered_bytes = bytes;
Naohiro Aota67c003f2017-09-01 17:59:07 +09008315 u64 last_offset;
Josef Bacik4b46fce2010-05-23 11:00:55 -04008316 int ret;
8317
Qu Wenruo524272602017-03-08 10:25:52 +08008318 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
8319 wq = fs_info->endio_freespace_worker;
8320 func = btrfs_freespace_write_helper;
8321 } else {
8322 wq = fs_info->endio_write_workers;
8323 func = btrfs_endio_write_helper;
8324 }
8325
Chris Mason163cf092010-11-28 19:56:33 -05008326again:
Naohiro Aota67c003f2017-09-01 17:59:07 +09008327 last_offset = ordered_offset;
Chris Mason163cf092010-11-28 19:56:33 -05008328 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8329 &ordered_offset,
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02008330 ordered_bytes,
Filipe Manana14543772015-11-24 16:23:54 +00008331 uptodate);
Josef Bacik4b46fce2010-05-23 11:00:55 -04008332 if (!ret)
Chris Mason163cf092010-11-28 19:56:33 -05008333 goto out_test;
Josef Bacik4b46fce2010-05-23 11:00:55 -04008334
Qu Wenruo524272602017-03-08 10:25:52 +08008335 btrfs_init_work(&ordered->work, func, finish_ordered_fn, NULL, NULL);
8336 btrfs_queue_work(wq, &ordered->work);
Chris Mason163cf092010-11-28 19:56:33 -05008337out_test:
8338 /*
Naohiro Aota67c003f2017-09-01 17:59:07 +09008339 * If btrfs_dec_test_ordered_pending does not find any ordered extent
8340 * in the range, we can exit.
8341 */
8342 if (ordered_offset == last_offset)
8343 return;
8344 /*
Chris Mason163cf092010-11-28 19:56:33 -05008345 * our bio might span multiple ordered extents. If we haven't
8346 * completed the accounting for the whole dio, go back and try again
8347 */
Filipe Manana14543772015-11-24 16:23:54 +00008348 if (ordered_offset < offset + bytes) {
8349 ordered_bytes = offset + bytes - ordered_offset;
Josef Bacik5fd02042012-05-02 14:00:54 -04008350 ordered = NULL;
Chris Mason163cf092010-11-28 19:56:33 -05008351 goto again;
8352 }
Filipe Manana14543772015-11-24 16:23:54 +00008353}
8354
8355static void btrfs_endio_direct_write(struct bio *bio)
8356{
8357 struct btrfs_dio_private *dip = bio->bi_private;
8358 struct bio *dio_bio = dip->dio_bio;
8359
Qu Wenruo524272602017-03-08 10:25:52 +08008360 __endio_write_update_ordered(dip->inode, dip->logical_offset,
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02008361 dip->bytes, !bio->bi_status);
Josef Bacik4b46fce2010-05-23 11:00:55 -04008362
Josef Bacik4b46fce2010-05-23 11:00:55 -04008363 kfree(dip);
Josef Bacikc0da7aa2011-03-22 11:05:07 -04008364
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02008365 dio_bio->bi_status = bio->bi_status;
Christoph Hellwig40553512017-06-03 09:37:58 +02008366 dio_end_io(dio_bio);
Chris Mason9be33952013-05-17 18:30:14 -04008367 bio_put(bio);
Josef Bacik4b46fce2010-05-23 11:00:55 -04008368}
8369
Linus Torvalds8c27cb32017-07-05 16:41:23 -07008370static blk_status_t __btrfs_submit_bio_start_direct_io(void *private_data,
Chris Masoneaf25d92010-05-25 09:48:28 -04008371 struct bio *bio, int mirror_num,
8372 unsigned long bio_flags, u64 offset)
8373{
Josef Bacikc6100a42017-05-05 11:57:13 -04008374 struct inode *inode = private_data;
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02008375 blk_status_t ret;
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04008376 ret = btrfs_csum_one_bio(inode, bio, offset, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01008377 BUG_ON(ret); /* -ENOMEM */
Chris Masoneaf25d92010-05-25 09:48:28 -04008378 return 0;
8379}
8380
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02008381static void btrfs_end_dio_bio(struct bio *bio)
Miao Xiee65e1532010-11-22 03:04:43 +00008382{
8383 struct btrfs_dio_private *dip = bio->bi_private;
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02008384 blk_status_t err = bio->bi_status;
Miao Xiee65e1532010-11-22 03:04:43 +00008385
Miao Xie8b110e32014-09-12 18:44:03 +08008386 if (err)
8387 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
Mike Christie6296b962016-06-05 14:32:21 -05008388 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
David Sterbaf85b7372017-01-20 14:54:07 +01008389 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8390 bio->bi_opf,
Miao Xie8b110e32014-09-12 18:44:03 +08008391 (unsigned long long)bio->bi_iter.bi_sector,
8392 bio->bi_iter.bi_size, err);
8393
8394 if (dip->subio_endio)
8395 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
Miao Xiec1dc0892014-09-12 18:43:56 +08008396
8397 if (err) {
Miao Xiee65e1532010-11-22 03:04:43 +00008398 dip->errors = 1;
8399
8400 /*
8401 * before atomic variable goto zero, we must make sure
8402 * dip->errors is perceived to be set.
8403 */
Peter Zijlstra4e857c52014-03-17 18:06:10 +01008404 smp_mb__before_atomic();
Miao Xiee65e1532010-11-22 03:04:43 +00008405 }
8406
8407 /* if there are more bios still pending for this dio, just exit */
8408 if (!atomic_dec_and_test(&dip->pending_bios))
8409 goto out;
8410
Chris Mason9be33952013-05-17 18:30:14 -04008411 if (dip->errors) {
Miao Xiee65e1532010-11-22 03:04:43 +00008412 bio_io_error(dip->orig_bio);
Chris Mason9be33952013-05-17 18:30:14 -04008413 } else {
Anand Jain2dbe0c72017-10-14 08:35:56 +08008414 dip->dio_bio->bi_status = BLK_STS_OK;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02008415 bio_endio(dip->orig_bio);
Miao Xiee65e1532010-11-22 03:04:43 +00008416 }
8417out:
8418 bio_put(bio);
8419}
8420
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02008421static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode,
Miao Xiec1dc0892014-09-12 18:43:56 +08008422 struct btrfs_dio_private *dip,
8423 struct bio *bio,
8424 u64 file_offset)
8425{
8426 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8427 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02008428 blk_status_t ret;
Miao Xiec1dc0892014-09-12 18:43:56 +08008429
8430 /*
8431 * We load all the csum data we need when we submit
8432 * the first bio to reduce the csum tree search and
8433 * contention.
8434 */
8435 if (dip->logical_offset == file_offset) {
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04008436 ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
Miao Xiec1dc0892014-09-12 18:43:56 +08008437 file_offset);
8438 if (ret)
8439 return ret;
8440 }
8441
8442 if (bio == dip->orig_bio)
8443 return 0;
8444
8445 file_offset -= dip->logical_offset;
8446 file_offset >>= inode->i_sb->s_blocksize_bits;
8447 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8448
8449 return 0;
8450}
8451
Omar Sandoval58efbc92017-08-22 23:45:59 -07008452static inline blk_status_t
8453__btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, u64 file_offset,
Linus Torvalds66ba7722017-09-09 13:27:51 -07008454 int async_submit)
Miao Xiee65e1532010-11-22 03:04:43 +00008455{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04008456 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Miao Xiefacc8a222013-07-25 19:22:34 +08008457 struct btrfs_dio_private *dip = bio->bi_private;
Mike Christie37226b22016-06-05 14:31:52 -05008458 bool write = bio_op(bio) == REQ_OP_WRITE;
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02008459 blk_status_t ret;
Miao Xiee65e1532010-11-22 03:04:43 +00008460
Josef Bacikb812ce22012-11-16 13:56:32 -05008461 if (async_submit)
8462 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8463
Miao Xiee65e1532010-11-22 03:04:43 +00008464 bio_get(bio);
Josef Bacik5fd02042012-05-02 14:00:54 -04008465
8466 if (!write) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04008467 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
Josef Bacik5fd02042012-05-02 14:00:54 -04008468 if (ret)
8469 goto err;
8470 }
Miao Xiee65e1532010-11-22 03:04:43 +00008471
Nikolay Borisove6961ca2017-08-03 15:44:58 +03008472 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
Josef Bacik1ae39932011-04-06 14:41:34 -04008473 goto map;
8474
8475 if (write && async_submit) {
Josef Bacikc6100a42017-05-05 11:57:13 -04008476 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
8477 file_offset, inode,
Jeff Mahoney0b246af2016-06-22 18:54:23 -04008478 __btrfs_submit_bio_start_direct_io,
8479 __btrfs_submit_bio_done);
Miao Xiee65e1532010-11-22 03:04:43 +00008480 goto err;
Josef Bacik1ae39932011-04-06 14:41:34 -04008481 } else if (write) {
8482 /*
8483 * If we aren't doing async submit, calculate the csum of the
8484 * bio now.
8485 */
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04008486 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
Josef Bacik1ae39932011-04-06 14:41:34 -04008487 if (ret)
8488 goto err;
Miao Xie23ea8e52014-09-12 18:43:54 +08008489 } else {
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04008490 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
Miao Xiec1dc0892014-09-12 18:43:56 +08008491 file_offset);
Tsutomu Itohc2db1072011-03-01 06:48:31 +00008492 if (ret)
8493 goto err;
8494 }
Josef Bacik1ae39932011-04-06 14:41:34 -04008495map:
Liu Bo9b4a9b22017-08-18 11:54:02 -06008496 ret = btrfs_map_bio(fs_info, bio, 0, 0);
Miao Xiee65e1532010-11-22 03:04:43 +00008497err:
8498 bio_put(bio);
8499 return ret;
8500}
8501
Nikolay Borisove6961ca2017-08-03 15:44:58 +03008502static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip)
Miao Xiee65e1532010-11-22 03:04:43 +00008503{
8504 struct inode *inode = dip->inode;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04008505 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Miao Xiee65e1532010-11-22 03:04:43 +00008506 struct bio *bio;
8507 struct bio *orig_bio = dip->orig_bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07008508 u64 start_sector = orig_bio->bi_iter.bi_sector;
Miao Xiee65e1532010-11-22 03:04:43 +00008509 u64 file_offset = dip->logical_offset;
Miao Xiee65e1532010-11-22 03:04:43 +00008510 u64 map_length;
Josef Bacik1ae39932011-04-06 14:41:34 -04008511 int async_submit = 0;
Liu Bo725130b2017-05-16 09:51:39 -07008512 u64 submit_len;
8513 int clone_offset = 0;
8514 int clone_len;
Chandan Rajendra5f4dc8f2016-01-21 15:56:00 +05308515 int ret;
Omar Sandoval58efbc92017-08-22 23:45:59 -07008516 blk_status_t status;
Miao Xiee65e1532010-11-22 03:04:43 +00008517
Kent Overstreet4f024f32013-10-11 15:44:27 -07008518 map_length = orig_bio->bi_iter.bi_size;
Liu Bo725130b2017-05-16 09:51:39 -07008519 submit_len = map_length;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04008520 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8521 &map_length, NULL, 0);
Miao Xie7a5c3c92014-06-17 18:58:59 +08008522 if (ret)
Miao Xiee65e1532010-11-22 03:04:43 +00008523 return -EIO;
Miao Xiefacc8a222013-07-25 19:22:34 +08008524
Liu Bo725130b2017-05-16 09:51:39 -07008525 if (map_length >= submit_len) {
Josef Bacik02f57c72011-04-06 14:25:44 -04008526 bio = orig_bio;
Miao Xiec1dc0892014-09-12 18:43:56 +08008527 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
Josef Bacik02f57c72011-04-06 14:25:44 -04008528 goto submit;
8529 }
8530
David Woodhouse53b381b2013-01-29 18:40:14 -05008531 /* async crcs make it difficult to collect full stripe writes. */
Jeff Mahoney1b868262017-05-17 11:38:35 -04008532 if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
David Woodhouse53b381b2013-01-29 18:40:14 -05008533 async_submit = 0;
8534 else
8535 async_submit = 1;
8536
Liu Bo725130b2017-05-16 09:51:39 -07008537 /* bio split */
8538 ASSERT(map_length <= INT_MAX);
Josef Bacik02f57c72011-04-06 14:25:44 -04008539 atomic_inc(&dip->pending_bios);
Arnd Bergmann3c91ee62017-05-18 15:33:29 +02008540 do {
Liu Bo725130b2017-05-16 09:51:39 -07008541 clone_len = min_t(int, submit_len, map_length);
Josef Bacik02f57c72011-04-06 14:25:44 -04008542
Liu Bo725130b2017-05-16 09:51:39 -07008543 /*
8544 * This will never fail as it's passing GPF_NOFS and
8545 * the allocation is backed by btrfs_bioset.
8546 */
Liu Boe4770942017-05-16 10:57:14 -07008547 bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
Liu Bo725130b2017-05-16 09:51:39 -07008548 clone_len);
8549 bio->bi_private = dip;
8550 bio->bi_end_io = btrfs_end_dio_bio;
8551 btrfs_io_bio(bio)->logical = file_offset;
Miao Xiee65e1532010-11-22 03:04:43 +00008552
Liu Bo725130b2017-05-16 09:51:39 -07008553 ASSERT(submit_len >= clone_len);
8554 submit_len -= clone_len;
8555 if (submit_len == 0)
8556 break;
Miao Xiee65e1532010-11-22 03:04:43 +00008557
Liu Bo725130b2017-05-16 09:51:39 -07008558 /*
8559 * Increase the count before we submit the bio so we know
8560 * the end IO handler won't happen before we increase the
8561 * count. Otherwise, the dip might get freed before we're
8562 * done setting it up.
8563 */
8564 atomic_inc(&dip->pending_bios);
Miao Xiee65e1532010-11-22 03:04:43 +00008565
Linus Torvalds66ba7722017-09-09 13:27:51 -07008566 status = __btrfs_submit_dio_bio(bio, inode, file_offset,
Omar Sandoval58efbc92017-08-22 23:45:59 -07008567 async_submit);
8568 if (status) {
Liu Bo725130b2017-05-16 09:51:39 -07008569 bio_put(bio);
8570 atomic_dec(&dip->pending_bios);
8571 goto out_err;
Miao Xiee65e1532010-11-22 03:04:43 +00008572 }
Liu Bo725130b2017-05-16 09:51:39 -07008573
8574 clone_offset += clone_len;
8575 start_sector += clone_len >> 9;
8576 file_offset += clone_len;
8577
8578 map_length = submit_len;
8579 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8580 start_sector << 9, &map_length, NULL, 0);
8581 if (ret)
8582 goto out_err;
Arnd Bergmann3c91ee62017-05-18 15:33:29 +02008583 } while (submit_len > 0);
Miao Xiee65e1532010-11-22 03:04:43 +00008584
Josef Bacik02f57c72011-04-06 14:25:44 -04008585submit:
Linus Torvalds66ba7722017-09-09 13:27:51 -07008586 status = __btrfs_submit_dio_bio(bio, inode, file_offset, async_submit);
Omar Sandoval58efbc92017-08-22 23:45:59 -07008587 if (!status)
Miao Xiee65e1532010-11-22 03:04:43 +00008588 return 0;
8589
8590 bio_put(bio);
8591out_err:
8592 dip->errors = 1;
8593 /*
8594 * before atomic variable goto zero, we must
8595 * make sure dip->errors is perceived to be set.
8596 */
Peter Zijlstra4e857c52014-03-17 18:06:10 +01008597 smp_mb__before_atomic();
Miao Xiee65e1532010-11-22 03:04:43 +00008598 if (atomic_dec_and_test(&dip->pending_bios))
8599 bio_io_error(dip->orig_bio);
8600
8601 /* bio_end_io() will handle error, so we needn't return it */
8602 return 0;
8603}
8604
Mike Christie8a4c1e42016-06-05 14:31:50 -05008605static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8606 loff_t file_offset)
Josef Bacik4b46fce2010-05-23 11:00:55 -04008607{
Filipe Manana61de7182015-07-01 12:13:10 +01008608 struct btrfs_dio_private *dip = NULL;
Liu Bo3892ac92017-04-17 15:00:28 -07008609 struct bio *bio = NULL;
8610 struct btrfs_io_bio *io_bio;
Mike Christie8a4c1e42016-06-05 14:31:50 -05008611 bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
Josef Bacik4b46fce2010-05-23 11:00:55 -04008612 int ret = 0;
8613
David Sterba8b6c1d52017-06-02 17:48:13 +02008614 bio = btrfs_bio_clone(dio_bio);
Josef Bacik4b46fce2010-05-23 11:00:55 -04008615
Miao Xiec1dc0892014-09-12 18:43:56 +08008616 dip = kzalloc(sizeof(*dip), GFP_NOFS);
Chris Mason9be33952013-05-17 18:30:14 -04008617 if (!dip) {
8618 ret = -ENOMEM;
Filipe Manana61de7182015-07-01 12:13:10 +01008619 goto free_ordered;
Chris Mason9be33952013-05-17 18:30:14 -04008620 }
8621
8622 dip->private = dio_bio->bi_private;
Josef Bacik4b46fce2010-05-23 11:00:55 -04008623 dip->inode = inode;
8624 dip->logical_offset = file_offset;
Kent Overstreet4f024f32013-10-11 15:44:27 -07008625 dip->bytes = dio_bio->bi_iter.bi_size;
8626 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
Liu Bo3892ac92017-04-17 15:00:28 -07008627 bio->bi_private = dip;
8628 dip->orig_bio = bio;
Chris Mason9be33952013-05-17 18:30:14 -04008629 dip->dio_bio = dio_bio;
Miao Xiee65e1532010-11-22 03:04:43 +00008630 atomic_set(&dip->pending_bios, 0);
Liu Bo3892ac92017-04-17 15:00:28 -07008631 io_bio = btrfs_io_bio(bio);
8632 io_bio->logical = file_offset;
Josef Bacik4b46fce2010-05-23 11:00:55 -04008633
Miao Xiec1dc0892014-09-12 18:43:56 +08008634 if (write) {
Liu Bo3892ac92017-04-17 15:00:28 -07008635 bio->bi_end_io = btrfs_endio_direct_write;
Miao Xiec1dc0892014-09-12 18:43:56 +08008636 } else {
Liu Bo3892ac92017-04-17 15:00:28 -07008637 bio->bi_end_io = btrfs_endio_direct_read;
Miao Xiec1dc0892014-09-12 18:43:56 +08008638 dip->subio_endio = btrfs_subio_endio_read;
8639 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04008640
Filipe Mananaf28a4922015-12-08 19:23:20 +00008641 /*
8642 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8643 * even if we fail to submit a bio, because in such case we do the
8644 * corresponding error handling below and it must not be done a second
8645 * time by btrfs_direct_IO().
8646 */
8647 if (write) {
8648 struct btrfs_dio_data *dio_data = current->journal_info;
8649
8650 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8651 dip->bytes;
8652 dio_data->unsubmitted_oe_range_start =
8653 dio_data->unsubmitted_oe_range_end;
8654 }
8655
Nikolay Borisove6961ca2017-08-03 15:44:58 +03008656 ret = btrfs_submit_direct_hook(dip);
Miao Xiee65e1532010-11-22 03:04:43 +00008657 if (!ret)
Chris Masoneaf25d92010-05-25 09:48:28 -04008658 return;
Chris Mason9be33952013-05-17 18:30:14 -04008659
Liu Bo3892ac92017-04-17 15:00:28 -07008660 if (io_bio->end_io)
8661 io_bio->end_io(io_bio, ret);
Chris Mason9be33952013-05-17 18:30:14 -04008662
Josef Bacik4b46fce2010-05-23 11:00:55 -04008663free_ordered:
8664 /*
Filipe Manana61de7182015-07-01 12:13:10 +01008665 * If we arrived here it means either we failed to submit the dip
8666 * or we either failed to clone the dio_bio or failed to allocate the
8667 * dip. If we cloned the dio_bio and allocated the dip, we can just
8668 * call bio_endio against our io_bio so that we get proper resource
8669 * cleanup if we fail to submit the dip, otherwise, we must do the
8670 * same as btrfs_endio_direct_[write|read] because we can't call these
8671 * callbacks - they require an allocated dip and a clone of dio_bio.
Josef Bacik4b46fce2010-05-23 11:00:55 -04008672 */
Liu Bo3892ac92017-04-17 15:00:28 -07008673 if (bio && dip) {
Guoqing Jiang054ec2f2017-06-02 16:08:50 +08008674 bio_io_error(bio);
Filipe Manana61de7182015-07-01 12:13:10 +01008675 /*
Liu Bo3892ac92017-04-17 15:00:28 -07008676 * The end io callbacks free our dip, do the final put on bio
Filipe Manana61de7182015-07-01 12:13:10 +01008677 * and all the cleanup and final put for dio_bio (through
8678 * dio_end_io()).
8679 */
8680 dip = NULL;
Liu Bo3892ac92017-04-17 15:00:28 -07008681 bio = NULL;
Filipe Manana61de7182015-07-01 12:13:10 +01008682 } else {
Filipe Manana14543772015-11-24 16:23:54 +00008683 if (write)
Qu Wenruo524272602017-03-08 10:25:52 +08008684 __endio_write_update_ordered(inode,
Filipe Manana14543772015-11-24 16:23:54 +00008685 file_offset,
8686 dio_bio->bi_iter.bi_size,
Qu Wenruo524272602017-03-08 10:25:52 +08008687 false);
Filipe Manana14543772015-11-24 16:23:54 +00008688 else
Filipe Manana61de7182015-07-01 12:13:10 +01008689 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8690 file_offset + dio_bio->bi_iter.bi_size - 1);
Filipe Manana14543772015-11-24 16:23:54 +00008691
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02008692 dio_bio->bi_status = BLK_STS_IOERR;
Filipe Manana61de7182015-07-01 12:13:10 +01008693 /*
8694 * Releases and cleans up our dio_bio, no need to bio_put()
8695 * nor bio_endio()/bio_io_error() against dio_bio.
8696 */
Christoph Hellwig40553512017-06-03 09:37:58 +02008697 dio_end_io(dio_bio);
Josef Bacik4b46fce2010-05-23 11:00:55 -04008698 }
Liu Bo3892ac92017-04-17 15:00:28 -07008699 if (bio)
8700 bio_put(bio);
Filipe Manana61de7182015-07-01 12:13:10 +01008701 kfree(dip);
Josef Bacik4b46fce2010-05-23 11:00:55 -04008702}
8703
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04008704static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04008705 const struct iov_iter *iter, loff_t offset)
Chris Mason5a5f79b2010-05-26 21:33:37 -04008706{
8707 int seg;
Josef Bacika1b75f72011-04-08 15:51:18 +00008708 int i;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04008709 unsigned int blocksize_mask = fs_info->sectorsize - 1;
Chris Mason5a5f79b2010-05-26 21:33:37 -04008710 ssize_t retval = -EINVAL;
Chris Mason5a5f79b2010-05-26 21:33:37 -04008711
8712 if (offset & blocksize_mask)
8713 goto out;
8714
Al Viro28060d52014-03-22 05:15:17 -04008715 if (iov_iter_alignment(iter) & blocksize_mask)
8716 goto out;
Josef Bacika1b75f72011-04-08 15:51:18 +00008717
Al Viro28060d52014-03-22 05:15:17 -04008718 /* If this is a write we don't need to check anymore */
Al Virocd27e452016-10-10 13:39:05 -04008719 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
Al Viro28060d52014-03-22 05:15:17 -04008720 return 0;
8721 /*
8722 * Check to make sure we don't have duplicate iov_base's in this
8723 * iovec, if so return EINVAL, otherwise we'll get csum errors
8724 * when reading back.
8725 */
8726 for (seg = 0; seg < iter->nr_segs; seg++) {
8727 for (i = seg + 1; i < iter->nr_segs; i++) {
8728 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
Josef Bacika1b75f72011-04-08 15:51:18 +00008729 goto out;
8730 }
Chris Mason5a5f79b2010-05-26 21:33:37 -04008731 }
8732 retval = 0;
8733out:
8734 return retval;
8735}
Josef Bacikeb838e72012-07-31 16:28:48 -04008736
Christoph Hellwigc8b8e322016-04-07 08:51:58 -07008737static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
Chris Mason16432982008-04-10 10:23:21 -04008738{
Josef Bacik4b46fce2010-05-23 11:00:55 -04008739 struct file *file = iocb->ki_filp;
8740 struct inode *inode = file->f_mapping->host;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04008741 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
chandan50745b02015-08-28 21:10:13 +05308742 struct btrfs_dio_data dio_data = { 0 };
Qu Wenruo364ecf32017-02-27 15:10:38 +08008743 struct extent_changeset *data_reserved = NULL;
Christoph Hellwigc8b8e322016-04-07 08:51:58 -07008744 loff_t offset = iocb->ki_pos;
Miao Xie09348562013-02-07 10:12:07 +00008745 size_t count = 0;
Miao Xie2e60a512013-02-08 07:01:08 +00008746 int flags = 0;
Miao Xie38851cc2013-02-08 07:04:11 +00008747 bool wakeup = true;
8748 bool relock = false;
Miao Xie09348562013-02-07 10:12:07 +00008749 ssize_t ret;
Josef Bacik4b46fce2010-05-23 11:00:55 -04008750
Nikolay Borisov8c70c9f2017-08-21 12:43:46 +03008751 if (check_direct_IO(fs_info, iter, offset))
Chris Mason5a5f79b2010-05-26 21:33:37 -04008752 return 0;
Chris Mason5a5f79b2010-05-26 21:33:37 -04008753
Jens Axboefe0f07d2015-04-15 17:05:48 -06008754 inode_dio_begin(inode);
Miao Xie38851cc2013-02-08 07:04:11 +00008755
Josef Bacik0e267c42013-07-02 10:38:02 -04008756 /*
Miao Xie41bd9ca2014-03-06 13:54:57 +08008757 * The generic stuff only does filemap_write_and_wait_range, which
8758 * isn't enough if we've written compressed pages to this area, so
8759 * we need to flush the dirty pages again to make absolutely sure
8760 * that any outstanding dirty pages are on disk.
Josef Bacik0e267c42013-07-02 10:38:02 -04008761 */
Al Viroa6cbcd42014-03-04 22:38:00 -05008762 count = iov_iter_count(iter);
Miao Xie41bd9ca2014-03-06 13:54:57 +08008763 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8764 &BTRFS_I(inode)->runtime_flags))
Wang Shilong9a025a02014-07-17 11:44:13 +08008765 filemap_fdatawrite_range(inode->i_mapping, offset,
8766 offset + count - 1);
Josef Bacik0e267c42013-07-02 10:38:02 -04008767
Omar Sandoval6f673762015-03-16 04:33:52 -07008768 if (iov_iter_rw(iter) == WRITE) {
Miao Xie38851cc2013-02-08 07:04:11 +00008769 /*
8770 * If the write DIO is beyond the EOF, we need update
8771 * the isize, but it is protected by i_mutex. So we can
8772 * not unlock the i_mutex at this case.
8773 */
8774 if (offset + count <= inode->i_size) {
Liu Bo4aaedfb2016-12-14 22:36:05 -08008775 dio_data.overwrite = 1;
Al Viro59551022016-01-22 15:40:57 -05008776 inode_unlock(inode);
Miao Xie38851cc2013-02-08 07:04:11 +00008777 relock = true;
Goldwyn Rodriguesedf064e2017-06-20 07:05:49 -05008778 } else if (iocb->ki_flags & IOCB_NOWAIT) {
8779 ret = -EAGAIN;
8780 goto out;
Miao Xie38851cc2013-02-08 07:04:11 +00008781 }
Qu Wenruo364ecf32017-02-27 15:10:38 +08008782 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
8783 offset, count);
Miao Xie09348562013-02-07 10:12:07 +00008784 if (ret)
Miao Xie38851cc2013-02-08 07:04:11 +00008785 goto out;
Josef Bacike1cbbfa2015-03-17 10:52:28 -04008786
8787 /*
8788 * We need to know how many extents we reserved so that we can
8789 * do the accounting properly if we go over the number we
8790 * originally calculated. Abuse current->journal_info for this.
8791 */
Jeff Mahoneyda170662016-06-15 09:22:56 -04008792 dio_data.reserve = round_up(count,
Jeff Mahoney0b246af2016-06-22 18:54:23 -04008793 fs_info->sectorsize);
Filipe Mananaf28a4922015-12-08 19:23:20 +00008794 dio_data.unsubmitted_oe_range_start = (u64)offset;
8795 dio_data.unsubmitted_oe_range_end = (u64)offset;
chandan50745b02015-08-28 21:10:13 +05308796 current->journal_info = &dio_data;
Chandan Rajendra97dcdea2016-12-23 15:00:18 +05308797 down_read(&BTRFS_I(inode)->dio_sem);
David Sterbaee39b432014-09-30 01:33:33 +02008798 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8799 &BTRFS_I(inode)->runtime_flags)) {
Jens Axboefe0f07d2015-04-15 17:05:48 -06008800 inode_dio_end(inode);
Miao Xie38851cc2013-02-08 07:04:11 +00008801 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8802 wakeup = false;
Miao Xie09348562013-02-07 10:12:07 +00008803 }
8804
Omar Sandoval17f8c842015-03-16 04:33:50 -07008805 ret = __blockdev_direct_IO(iocb, inode,
Jeff Mahoney0b246af2016-06-22 18:54:23 -04008806 fs_info->fs_devices->latest_bdev,
Christoph Hellwigc8b8e322016-04-07 08:51:58 -07008807 iter, btrfs_get_blocks_direct, NULL,
Omar Sandoval17f8c842015-03-16 04:33:50 -07008808 btrfs_submit_direct, flags);
Omar Sandoval6f673762015-03-16 04:33:52 -07008809 if (iov_iter_rw(iter) == WRITE) {
Chandan Rajendra97dcdea2016-12-23 15:00:18 +05308810 up_read(&BTRFS_I(inode)->dio_sem);
Josef Bacike1cbbfa2015-03-17 10:52:28 -04008811 current->journal_info = NULL;
Liu Boddba1bf2015-06-17 16:59:58 +08008812 if (ret < 0 && ret != -EIOCBQUEUED) {
chandan50745b02015-08-28 21:10:13 +05308813 if (dio_data.reserve)
Qu Wenruobc42bda2017-02-27 15:10:39 +08008814 btrfs_delalloc_release_space(inode, data_reserved,
8815 offset, dio_data.reserve);
Filipe Mananaf28a4922015-12-08 19:23:20 +00008816 /*
8817 * On error we might have left some ordered extents
8818 * without submitting corresponding bios for them, so
8819 * cleanup them up to avoid other tasks getting them
8820 * and waiting for them to complete forever.
8821 */
8822 if (dio_data.unsubmitted_oe_range_start <
8823 dio_data.unsubmitted_oe_range_end)
Qu Wenruo524272602017-03-08 10:25:52 +08008824 __endio_write_update_ordered(inode,
Filipe Mananaf28a4922015-12-08 19:23:20 +00008825 dio_data.unsubmitted_oe_range_start,
8826 dio_data.unsubmitted_oe_range_end -
8827 dio_data.unsubmitted_oe_range_start,
Qu Wenruo524272602017-03-08 10:25:52 +08008828 false);
Liu Boddba1bf2015-06-17 16:59:58 +08008829 } else if (ret >= 0 && (size_t)ret < count)
Qu Wenruobc42bda2017-02-27 15:10:39 +08008830 btrfs_delalloc_release_space(inode, data_reserved,
8831 offset, count - (size_t)ret);
Josef Bacik69fe2d72017-10-19 14:15:57 -04008832 btrfs_delalloc_release_extents(BTRFS_I(inode), count);
Miao Xie09348562013-02-07 10:12:07 +00008833 }
Miao Xie38851cc2013-02-08 07:04:11 +00008834out:
Miao Xie2e60a512013-02-08 07:01:08 +00008835 if (wakeup)
Jens Axboefe0f07d2015-04-15 17:05:48 -06008836 inode_dio_end(inode);
Miao Xie38851cc2013-02-08 07:04:11 +00008837 if (relock)
Al Viro59551022016-01-22 15:40:57 -05008838 inode_lock(inode);
Miao Xie09348562013-02-07 10:12:07 +00008839
Qu Wenruo364ecf32017-02-27 15:10:38 +08008840 extent_changeset_free(data_reserved);
Miao Xie09348562013-02-07 10:12:07 +00008841 return ret;
Chris Mason16432982008-04-10 10:23:21 -04008842}
8843
Tsutomu Itoh05dadc02012-11-29 05:08:26 +00008844#define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8845
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -05008846static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8847 __u64 start, __u64 len)
8848{
Tsutomu Itoh05dadc02012-11-29 05:08:26 +00008849 int ret;
8850
8851 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8852 if (ret)
8853 return ret;
8854
Chris Masonec29ed52011-02-23 16:23:20 -05008855 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -05008856}
8857
Chris Mason9ebefb182007-06-15 13:50:00 -04008858int btrfs_readpage(struct file *file, struct page *page)
8859{
Chris Masond1310b22008-01-24 16:13:08 -05008860 struct extent_io_tree *tree;
8861 tree = &BTRFS_I(page->mapping->host)->io_tree;
Jan Schmidt8ddc7d92011-06-13 20:02:58 +02008862 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04008863}
Chris Mason1832a6d2007-12-21 16:27:21 -05008864
Chris Mason39279cc2007-06-12 06:35:45 -04008865static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8866{
Chris Masond1310b22008-01-24 16:13:08 -05008867 struct extent_io_tree *tree;
Josef Bacikbe7bd732015-10-22 15:05:09 -04008868 struct inode *inode = page->mapping->host;
8869 int ret;
Chris Masonb888db22007-08-27 16:49:44 -04008870
8871 if (current->flags & PF_MEMALLOC) {
8872 redirty_page_for_writepage(wbc, page);
8873 unlock_page(page);
8874 return 0;
8875 }
Josef Bacikbe7bd732015-10-22 15:05:09 -04008876
8877 /*
8878 * If we are under memory pressure we will call this directly from the
8879 * VM, we need to make sure we have the inode referenced for the ordered
8880 * extent. If not just return like we didn't do anything.
8881 */
8882 if (!igrab(inode)) {
8883 redirty_page_for_writepage(wbc, page);
8884 return AOP_WRITEPAGE_ACTIVATE;
8885 }
Chris Masond1310b22008-01-24 16:13:08 -05008886 tree = &BTRFS_I(page->mapping->host)->io_tree;
Josef Bacikbe7bd732015-10-22 15:05:09 -04008887 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8888 btrfs_add_delayed_iput(inode);
8889 return ret;
Chris Masona52d9a82007-08-27 16:49:44 -04008890}
Chris Mason39279cc2007-06-12 06:35:45 -04008891
Eric Sandeen48a3b632013-04-25 20:41:01 +00008892static int btrfs_writepages(struct address_space *mapping,
8893 struct writeback_control *wbc)
Chris Masonb293f022007-11-01 19:45:34 -04008894{
Chris Masond1310b22008-01-24 16:13:08 -05008895 struct extent_io_tree *tree;
Chris Mason771ed682008-11-06 22:02:51 -05008896
Chris Masond1310b22008-01-24 16:13:08 -05008897 tree = &BTRFS_I(mapping->host)->io_tree;
Chris Masonb293f022007-11-01 19:45:34 -04008898 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8899}
8900
Chris Mason3ab2fb52007-11-08 10:59:22 -05008901static int
8902btrfs_readpages(struct file *file, struct address_space *mapping,
8903 struct list_head *pages, unsigned nr_pages)
8904{
Chris Masond1310b22008-01-24 16:13:08 -05008905 struct extent_io_tree *tree;
8906 tree = &BTRFS_I(mapping->host)->io_tree;
Chris Mason3ab2fb52007-11-08 10:59:22 -05008907 return extent_readpages(tree, mapping, pages, nr_pages,
8908 btrfs_get_extent);
8909}
Chris Masone6dcd2d2008-07-17 12:53:50 -04008910static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
Chris Masona52d9a82007-08-27 16:49:44 -04008911{
Chris Masond1310b22008-01-24 16:13:08 -05008912 struct extent_io_tree *tree;
8913 struct extent_map_tree *map;
Chris Masona52d9a82007-08-27 16:49:44 -04008914 int ret;
Chris Mason39279cc2007-06-12 06:35:45 -04008915
Chris Masond1310b22008-01-24 16:13:08 -05008916 tree = &BTRFS_I(page->mapping->host)->io_tree;
8917 map = &BTRFS_I(page->mapping->host)->extent_tree;
Chris Mason70dec802008-01-29 09:59:12 -05008918 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
Chris Masona52d9a82007-08-27 16:49:44 -04008919 if (ret == 1) {
8920 ClearPagePrivate(page);
8921 set_page_private(page, 0);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03008922 put_page(page);
Chris Masona52d9a82007-08-27 16:49:44 -04008923 }
8924 return ret;
8925}
Chris Mason39279cc2007-06-12 06:35:45 -04008926
Chris Masone6dcd2d2008-07-17 12:53:50 -04008927static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8928{
Chris Mason98509cf2008-09-11 15:51:43 -04008929 if (PageWriteback(page) || PageDirty(page))
8930 return 0;
Michal Hocko3ba7ab22017-01-09 15:39:02 +01008931 return __btrfs_releasepage(page, gfp_flags);
Chris Masone6dcd2d2008-07-17 12:53:50 -04008932}
8933
Lukas Czernerd47992f2013-05-21 23:17:23 -04008934static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8935 unsigned int length)
Chris Masona52d9a82007-08-27 16:49:44 -04008936{
Josef Bacik5fd02042012-05-02 14:00:54 -04008937 struct inode *inode = page->mapping->host;
Chris Masond1310b22008-01-24 16:13:08 -05008938 struct extent_io_tree *tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -04008939 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00008940 struct extent_state *cached_state = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04008941 u64 page_start = page_offset(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03008942 u64 page_end = page_start + PAGE_SIZE - 1;
Chandan Rajendradbfdb6d2016-01-21 15:55:58 +05308943 u64 start;
8944 u64 end;
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00008945 int inode_evicting = inode->i_state & I_FREEING;
Chris Masona52d9a82007-08-27 16:49:44 -04008946
Chris Mason8b62b722009-09-02 16:53:46 -04008947 /*
8948 * we have the page locked, so new writeback can't start,
8949 * and the dirty bit won't be cleared while we are here.
8950 *
8951 * Wait for IO on this page so that we can safely clear
8952 * the PagePrivate2 bit and do ordered accounting
8953 */
Chris Masone6dcd2d2008-07-17 12:53:50 -04008954 wait_on_page_writeback(page);
Chris Mason8b62b722009-09-02 16:53:46 -04008955
Josef Bacik5fd02042012-05-02 14:00:54 -04008956 tree = &BTRFS_I(inode)->io_tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -04008957 if (offset) {
8958 btrfs_releasepage(page, GFP_NOFS);
8959 return;
8960 }
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00008961
8962 if (!inode_evicting)
David Sterbaff13db42015-12-03 14:30:40 +01008963 lock_extent_bits(tree, page_start, page_end, &cached_state);
Chandan Rajendradbfdb6d2016-01-21 15:55:58 +05308964again:
8965 start = page_start;
Nikolay Borisova776c6f2017-02-20 13:50:49 +02008966 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
Chandan Rajendradbfdb6d2016-01-21 15:55:58 +05308967 page_end - start + 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04008968 if (ordered) {
Chandan Rajendradbfdb6d2016-01-21 15:55:58 +05308969 end = min(page_end, ordered->file_offset + ordered->len - 1);
Chris Masoneb84ae02008-07-17 13:53:27 -04008970 /*
8971 * IO on this page will never be started, so we need
8972 * to account for any ordered extents now
8973 */
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00008974 if (!inode_evicting)
Chandan Rajendradbfdb6d2016-01-21 15:55:58 +05308975 clear_extent_bit(tree, start, end,
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00008976 EXTENT_DIRTY | EXTENT_DELALLOC |
Filipe Mananaa7e3b972017-04-03 10:45:46 +01008977 EXTENT_DELALLOC_NEW |
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00008978 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8979 EXTENT_DEFRAG, 1, 0, &cached_state,
8980 GFP_NOFS);
Chris Mason8b62b722009-09-02 16:53:46 -04008981 /*
8982 * whoever cleared the private bit is responsible
8983 * for the finish_ordered_io
8984 */
Josef Bacik77cef2e2013-08-29 13:57:21 -04008985 if (TestClearPagePrivate2(page)) {
8986 struct btrfs_ordered_inode_tree *tree;
8987 u64 new_len;
8988
8989 tree = &BTRFS_I(inode)->ordered_tree;
8990
8991 spin_lock_irq(&tree->lock);
8992 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
Chandan Rajendradbfdb6d2016-01-21 15:55:58 +05308993 new_len = start - ordered->file_offset;
Josef Bacik77cef2e2013-08-29 13:57:21 -04008994 if (new_len < ordered->truncated_len)
8995 ordered->truncated_len = new_len;
8996 spin_unlock_irq(&tree->lock);
8997
8998 if (btrfs_dec_test_ordered_pending(inode, &ordered,
Chandan Rajendradbfdb6d2016-01-21 15:55:58 +05308999 start,
9000 end - start + 1, 1))
Josef Bacik77cef2e2013-08-29 13:57:21 -04009001 btrfs_finish_ordered_io(ordered);
Chris Mason8b62b722009-09-02 16:53:46 -04009002 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04009003 btrfs_put_ordered_extent(ordered);
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00009004 if (!inode_evicting) {
9005 cached_state = NULL;
Chandan Rajendradbfdb6d2016-01-21 15:55:58 +05309006 lock_extent_bits(tree, start, end,
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00009007 &cached_state);
9008 }
Chandan Rajendradbfdb6d2016-01-21 15:55:58 +05309009
9010 start = end + 1;
9011 if (start < page_end)
9012 goto again;
Chris Masone6dcd2d2008-07-17 12:53:50 -04009013 }
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00009014
Qu Wenruob9d0b382015-09-29 10:35:16 +08009015 /*
9016 * Qgroup reserved space handler
9017 * Page here will be either
9018 * 1) Already written to disk
9019 * In this case, its reserved space is released from data rsv map
9020 * and will be freed by delayed_ref handler finally.
9021 * So even we call qgroup_free_data(), it won't decrease reserved
9022 * space.
9023 * 2) Not written to disk
Goldwyn Rodrigues0b34c262016-09-30 10:40:52 -05009024 * This means the reserved space should be freed here. However,
9025 * if a truncate invalidates the page (by clearing PageDirty)
9026 * and the page is accounted for while allocating extent
9027 * in btrfs_check_data_free_space() we let delayed_ref to
9028 * free the entire extent.
Qu Wenruob9d0b382015-09-29 10:35:16 +08009029 */
Goldwyn Rodrigues0b34c262016-09-30 10:40:52 -05009030 if (PageDirty(page))
Qu Wenruobc42bda2017-02-27 15:10:39 +08009031 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00009032 if (!inode_evicting) {
9033 clear_extent_bit(tree, page_start, page_end,
9034 EXTENT_LOCKED | EXTENT_DIRTY |
Filipe Mananaa7e3b972017-04-03 10:45:46 +01009035 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
9036 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00009037 &cached_state, GFP_NOFS);
9038
9039 __btrfs_releasepage(page, GFP_NOFS);
9040 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04009041
Chris Mason4a096752008-07-21 10:29:44 -04009042 ClearPageChecked(page);
Chris Mason9ad6b7b2008-04-18 16:11:30 -04009043 if (PagePrivate(page)) {
Chris Mason9ad6b7b2008-04-18 16:11:30 -04009044 ClearPagePrivate(page);
9045 set_page_private(page, 0);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03009046 put_page(page);
Chris Mason9ad6b7b2008-04-18 16:11:30 -04009047 }
Chris Mason39279cc2007-06-12 06:35:45 -04009048}
9049
Chris Mason9ebefb182007-06-15 13:50:00 -04009050/*
9051 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
9052 * called from a page fault handler when a page is first dirtied. Hence we must
9053 * be careful to check for EOF conditions here. We set the page up correctly
9054 * for a written page which means we get ENOSPC checking when writing into
9055 * holes and correct delalloc and unwritten extent mapping on filesystems that
9056 * support these features.
9057 *
9058 * We are not allowed to take the i_mutex here so we have to play games to
9059 * protect against truncate races as the page could now be beyond EOF. Because
9060 * vmtruncate() writes the inode size before removing pages, once we have the
9061 * page lock we can determine safely if the page is beyond EOF. If it is not
9062 * beyond EOF, then the page is guaranteed safe against truncation until we
9063 * unlock the page.
9064 */
Dave Jiang11bac802017-02-24 14:56:41 -08009065int btrfs_page_mkwrite(struct vm_fault *vmf)
Chris Mason9ebefb182007-06-15 13:50:00 -04009066{
Nick Pigginc2ec1752009-03-31 15:23:21 -07009067 struct page *page = vmf->page;
Dave Jiang11bac802017-02-24 14:56:41 -08009068 struct inode *inode = file_inode(vmf->vma->vm_file);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04009069 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Chris Masone6dcd2d2008-07-17 12:53:50 -04009070 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
9071 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00009072 struct extent_state *cached_state = NULL;
Qu Wenruo364ecf32017-02-27 15:10:38 +08009073 struct extent_changeset *data_reserved = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04009074 char *kaddr;
9075 unsigned long zero_start;
Chris Mason9ebefb182007-06-15 13:50:00 -04009076 loff_t size;
Chris Mason1832a6d2007-12-21 16:27:21 -05009077 int ret;
Chris Mason9998eb72012-01-25 13:47:40 -05009078 int reserved = 0;
Chandan Rajendrad0b7da882016-01-21 15:55:57 +05309079 u64 reserved_space;
Chris Masona52d9a82007-08-27 16:49:44 -04009080 u64 page_start;
Chris Masone6dcd2d2008-07-17 12:53:50 -04009081 u64 page_end;
Chandan Rajendrad0b7da882016-01-21 15:55:57 +05309082 u64 end;
9083
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03009084 reserved_space = PAGE_SIZE;
Chris Mason9ebefb182007-06-15 13:50:00 -04009085
Jan Karab2b5ef52012-06-12 16:20:45 +02009086 sb_start_pagefault(inode->i_sb);
Qu Wenruodf480632015-09-08 17:25:54 +08009087 page_start = page_offset(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03009088 page_end = page_start + PAGE_SIZE - 1;
Chandan Rajendrad0b7da882016-01-21 15:55:57 +05309089 end = page_end;
Qu Wenruodf480632015-09-08 17:25:54 +08009090
Chandan Rajendrad0b7da882016-01-21 15:55:57 +05309091 /*
9092 * Reserving delalloc space after obtaining the page lock can lead to
9093 * deadlock. For example, if a dirty page is locked by this function
9094 * and the call to btrfs_delalloc_reserve_space() ends up triggering
9095 * dirty page write out, then the btrfs_writepage() function could
9096 * end up waiting indefinitely to get a lock on the page currently
9097 * being processed by btrfs_page_mkwrite() function.
9098 */
Qu Wenruo364ecf32017-02-27 15:10:38 +08009099 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
Chandan Rajendrad0b7da882016-01-21 15:55:57 +05309100 reserved_space);
Chris Mason9998eb72012-01-25 13:47:40 -05009101 if (!ret) {
Dave Jiang11bac802017-02-24 14:56:41 -08009102 ret = file_update_time(vmf->vma->vm_file);
Chris Mason9998eb72012-01-25 13:47:40 -05009103 reserved = 1;
9104 }
Nick Piggin56a76f82009-03-31 15:23:23 -07009105 if (ret) {
9106 if (ret == -ENOMEM)
9107 ret = VM_FAULT_OOM;
9108 else /* -ENOSPC, -EIO, etc */
9109 ret = VM_FAULT_SIGBUS;
Chris Mason9998eb72012-01-25 13:47:40 -05009110 if (reserved)
9111 goto out;
9112 goto out_noreserve;
Nick Piggin56a76f82009-03-31 15:23:23 -07009113 }
Chris Mason1832a6d2007-12-21 16:27:21 -05009114
Nick Piggin56a76f82009-03-31 15:23:23 -07009115 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
Chris Masone6dcd2d2008-07-17 12:53:50 -04009116again:
Chris Mason9ebefb182007-06-15 13:50:00 -04009117 lock_page(page);
Chris Mason9ebefb182007-06-15 13:50:00 -04009118 size = i_size_read(inode);
Chris Masona52d9a82007-08-27 16:49:44 -04009119
Chris Mason9ebefb182007-06-15 13:50:00 -04009120 if ((page->mapping != inode->i_mapping) ||
Chris Masone6dcd2d2008-07-17 12:53:50 -04009121 (page_start >= size)) {
Chris Mason9ebefb182007-06-15 13:50:00 -04009122 /* page got truncated out from underneath us */
9123 goto out_unlock;
9124 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04009125 wait_on_page_writeback(page);
9126
David Sterbaff13db42015-12-03 14:30:40 +01009127 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
Chris Masone6dcd2d2008-07-17 12:53:50 -04009128 set_page_extent_mapped(page);
9129
Chris Masoneb84ae02008-07-17 13:53:27 -04009130 /*
9131 * we can't set the delalloc bits if there are pending ordered
9132 * extents. Drop our locks and wait for them to finish
9133 */
Nikolay Borisova776c6f2017-02-20 13:50:49 +02009134 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
9135 PAGE_SIZE);
Chris Masone6dcd2d2008-07-17 12:53:50 -04009136 if (ordered) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00009137 unlock_extent_cached(io_tree, page_start, page_end,
9138 &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04009139 unlock_page(page);
Chris Masoneb84ae02008-07-17 13:53:27 -04009140 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04009141 btrfs_put_ordered_extent(ordered);
9142 goto again;
9143 }
9144
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03009145 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
Jeff Mahoneyda170662016-06-15 09:22:56 -04009146 reserved_space = round_up(size - page_start,
Jeff Mahoney0b246af2016-06-22 18:54:23 -04009147 fs_info->sectorsize);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03009148 if (reserved_space < PAGE_SIZE) {
Chandan Rajendrad0b7da882016-01-21 15:55:57 +05309149 end = page_start + reserved_space - 1;
Qu Wenruobc42bda2017-02-27 15:10:39 +08009150 btrfs_delalloc_release_space(inode, data_reserved,
9151 page_start, PAGE_SIZE - reserved_space);
Chandan Rajendrad0b7da882016-01-21 15:55:57 +05309152 }
9153 }
9154
Josef Bacikfbf19082009-10-01 17:10:23 -04009155 /*
Liu Bo54160342016-12-13 12:15:19 -08009156 * page_mkwrite gets called when the page is firstly dirtied after it's
9157 * faulted in, but write(2) could also dirty a page and set delalloc
9158 * bits, thus in this case for space account reason, we still need to
9159 * clear any delalloc bits within this page range since we have to
9160 * reserve data&meta space before lock_page() (see above comments).
Josef Bacikfbf19082009-10-01 17:10:23 -04009161 */
Chandan Rajendrad0b7da882016-01-21 15:55:57 +05309162 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
Liu Bo9e8a4a82012-09-05 19:10:51 -06009163 EXTENT_DIRTY | EXTENT_DELALLOC |
9164 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
Josef Bacik2ac55d42010-02-03 19:33:23 +00009165 0, 0, &cached_state, GFP_NOFS);
Josef Bacikfbf19082009-10-01 17:10:23 -04009166
Filipe Mananae3b8a482017-11-04 00:16:59 +00009167 ret = btrfs_set_extent_delalloc(inode, page_start, end, 0,
Qu Wenruoba8b04c2016-07-19 16:50:36 +08009168 &cached_state, 0);
Josef Bacik9ed74f22009-09-11 16:12:44 -04009169 if (ret) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00009170 unlock_extent_cached(io_tree, page_start, page_end,
9171 &cached_state, GFP_NOFS);
Josef Bacik9ed74f22009-09-11 16:12:44 -04009172 ret = VM_FAULT_SIGBUS;
9173 goto out_unlock;
9174 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04009175 ret = 0;
Chris Mason9ebefb182007-06-15 13:50:00 -04009176
9177 /* page is wholly or partially inside EOF */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03009178 if (page_start + PAGE_SIZE > size)
9179 zero_start = size & ~PAGE_MASK;
Chris Mason9ebefb182007-06-15 13:50:00 -04009180 else
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03009181 zero_start = PAGE_SIZE;
Chris Mason9ebefb182007-06-15 13:50:00 -04009182
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03009183 if (zero_start != PAGE_SIZE) {
Chris Masone6dcd2d2008-07-17 12:53:50 -04009184 kaddr = kmap(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03009185 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
Chris Masone6dcd2d2008-07-17 12:53:50 -04009186 flush_dcache_page(page);
9187 kunmap(page);
9188 }
Chris Mason247e7432008-07-17 12:53:51 -04009189 ClearPageChecked(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04009190 set_page_dirty(page);
Chris Mason50a9b212009-09-11 12:33:12 -04009191 SetPageUptodate(page);
Chris Mason5a3f23d2009-03-31 13:27:11 -04009192
Jeff Mahoney0b246af2016-06-22 18:54:23 -04009193 BTRFS_I(inode)->last_trans = fs_info->generation;
Chris Mason257c62e2009-10-13 13:21:08 -04009194 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
Liu Bo46d8bc32012-08-29 01:07:55 -06009195 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
Chris Mason257c62e2009-10-13 13:21:08 -04009196
Josef Bacik2ac55d42010-02-03 19:33:23 +00009197 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
Chris Mason9ebefb182007-06-15 13:50:00 -04009198
9199out_unlock:
Jan Karab2b5ef52012-06-12 16:20:45 +02009200 if (!ret) {
Josef Bacik8b62f872017-10-19 14:15:55 -04009201 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
Jan Karab2b5ef52012-06-12 16:20:45 +02009202 sb_end_pagefault(inode->i_sb);
Qu Wenruo364ecf32017-02-27 15:10:38 +08009203 extent_changeset_free(data_reserved);
Chris Mason50a9b212009-09-11 12:33:12 -04009204 return VM_FAULT_LOCKED;
Jan Karab2b5ef52012-06-12 16:20:45 +02009205 }
Chris Mason9ebefb182007-06-15 13:50:00 -04009206 unlock_page(page);
Chris Mason1832a6d2007-12-21 16:27:21 -05009207out:
Josef Bacik8b62f872017-10-19 14:15:55 -04009208 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
Qu Wenruobc42bda2017-02-27 15:10:39 +08009209 btrfs_delalloc_release_space(inode, data_reserved, page_start,
9210 reserved_space);
Chris Mason9998eb72012-01-25 13:47:40 -05009211out_noreserve:
Jan Karab2b5ef52012-06-12 16:20:45 +02009212 sb_end_pagefault(inode->i_sb);
Qu Wenruo364ecf32017-02-27 15:10:38 +08009213 extent_changeset_free(data_reserved);
Chris Mason9ebefb182007-06-15 13:50:00 -04009214 return ret;
9215}
9216
Josef Bacika41ad392011-01-31 15:30:16 -05009217static int btrfs_truncate(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04009218{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04009219 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Chris Mason39279cc2007-06-12 06:35:45 -04009220 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacikfcb80c22011-05-03 10:40:22 -04009221 struct btrfs_block_rsv *rsv;
Josef Bacika71754f2013-06-17 17:14:39 -04009222 int ret = 0;
Josef Bacik3893e332011-01-31 16:03:11 -05009223 int err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04009224 struct btrfs_trans_handle *trans;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04009225 u64 mask = fs_info->sectorsize - 1;
9226 u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
Chris Mason39279cc2007-06-12 06:35:45 -04009227
Josef Bacik0ef8b722013-10-25 16:13:35 -04009228 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9229 (u64)-1);
9230 if (ret)
9231 return ret;
Chris Mason39279cc2007-06-12 06:35:45 -04009232
Josef Bacikfcb80c22011-05-03 10:40:22 -04009233 /*
Nicholas D Steeves01327612016-05-19 21:18:45 -04009234 * Yes ladies and gentlemen, this is indeed ugly. The fact is we have
Josef Bacikfcb80c22011-05-03 10:40:22 -04009235 * 3 things going on here
9236 *
9237 * 1) We need to reserve space for our orphan item and the space to
9238 * delete our orphan item. Lord knows we don't want to have a dangling
9239 * orphan item because we didn't reserve space to remove it.
9240 *
9241 * 2) We need to reserve space to update our inode.
9242 *
9243 * 3) We need to have something to cache all the space that is going to
9244 * be free'd up by the truncate operation, but also have some slack
9245 * space reserved in case it uses space during the truncate (thank you
9246 * very much snapshotting).
9247 *
Nicholas D Steeves01327612016-05-19 21:18:45 -04009248 * And we need these to all be separate. The fact is we can use a lot of
Josef Bacikfcb80c22011-05-03 10:40:22 -04009249 * space doing the truncate, and we have no earthly idea how much space
Nicholas D Steeves01327612016-05-19 21:18:45 -04009250 * we will use, so we need the truncate reservation to be separate so it
Josef Bacikfcb80c22011-05-03 10:40:22 -04009251 * doesn't end up using space reserved for updating the inode or
9252 * removing the orphan item. We also need to be able to stop the
9253 * transaction and start a new one, which means we need to be able to
9254 * update the inode several times, and we have no idea of knowing how
9255 * many times that will be, so we can't just reserve 1 item for the
Nicholas D Steeves01327612016-05-19 21:18:45 -04009256 * entirety of the operation, so that has to be done separately as well.
Josef Bacikfcb80c22011-05-03 10:40:22 -04009257 * Then there is the orphan item, which does indeed need to be held on
9258 * to for the whole operation, and we need nobody to touch this reserved
9259 * space except the orphan code.
9260 *
9261 * So that leaves us with
9262 *
9263 * 1) root->orphan_block_rsv - for the orphan deletion.
9264 * 2) rsv - for the truncate reservation, which we will steal from the
9265 * transaction reservation.
9266 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9267 * updating the inode.
9268 */
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04009269 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
Josef Bacikfcb80c22011-05-03 10:40:22 -04009270 if (!rsv)
9271 return -ENOMEM;
Josef Bacik4a338542011-08-29 11:01:31 -04009272 rsv->size = min_size;
Josef Bacikca7e70f2012-08-27 17:48:15 -04009273 rsv->failfast = 1;
Josef Bacikf0cd8462011-03-04 14:37:08 -05009274
Josef Bacik907cbce2011-08-08 13:46:15 -04009275 /*
Josef Bacik07127182011-08-19 10:29:59 -04009276 * 1 for the truncate slack space
Josef Bacik907cbce2011-08-08 13:46:15 -04009277 * 1 for updating the inode.
9278 */
Josef Bacikf3fe8202013-01-07 17:03:21 -05009279 trans = btrfs_start_transaction(root, 2);
Josef Bacikfcb80c22011-05-03 10:40:22 -04009280 if (IS_ERR(trans)) {
9281 err = PTR_ERR(trans);
9282 goto out;
9283 }
Josef Bacikf0cd8462011-03-04 14:37:08 -05009284
Josef Bacik907cbce2011-08-08 13:46:15 -04009285 /* Migrate the slack space for the truncate to our reserve */
Jeff Mahoney0b246af2016-06-22 18:54:23 -04009286 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
Josef Bacik25d609f2016-03-25 13:25:48 -04009287 min_size, 0);
Josef Bacikfcb80c22011-05-03 10:40:22 -04009288 BUG_ON(ret);
Josef Bacikf0cd8462011-03-04 14:37:08 -05009289
Chris Mason5a3f23d2009-03-31 13:27:11 -04009290 /*
Josef Bacik5dc562c2012-08-17 13:14:17 -04009291 * So if we truncate and then write and fsync we normally would just
9292 * write the extents that changed, which is a problem if we need to
9293 * first truncate that entire inode. So set this flag so we write out
9294 * all of the extents in the inode to the sync log so we're completely
9295 * safe.
9296 */
9297 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
Josef Bacikca7e70f2012-08-27 17:48:15 -04009298 trans->block_rsv = rsv;
Josef Bacik5dc562c2012-08-17 13:14:17 -04009299
Yan, Zheng80825102009-11-12 09:35:36 +00009300 while (1) {
9301 ret = btrfs_truncate_inode_items(trans, root, inode,
9302 inode->i_size,
9303 BTRFS_EXTENT_DATA_KEY);
Josef Bacikddfae632017-10-19 14:16:02 -04009304 trans->block_rsv = &fs_info->trans_block_rsv;
Chris Mason28ed1342014-12-17 09:41:04 -08009305 if (ret != -ENOSPC && ret != -EAGAIN) {
Josef Bacik3893e332011-01-31 16:03:11 -05009306 err = ret;
Yan, Zheng80825102009-11-12 09:35:36 +00009307 break;
Josef Bacik3893e332011-01-31 16:03:11 -05009308 }
Chris Mason39279cc2007-06-12 06:35:45 -04009309
Yan, Zheng80825102009-11-12 09:35:36 +00009310 ret = btrfs_update_inode(trans, root, inode);
Josef Bacik3893e332011-01-31 16:03:11 -05009311 if (ret) {
9312 err = ret;
9313 break;
9314 }
Josef Bacikca7e70f2012-08-27 17:48:15 -04009315
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04009316 btrfs_end_transaction(trans);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04009317 btrfs_btree_balance_dirty(fs_info);
Josef Bacikca7e70f2012-08-27 17:48:15 -04009318
9319 trans = btrfs_start_transaction(root, 2);
9320 if (IS_ERR(trans)) {
9321 ret = err = PTR_ERR(trans);
9322 trans = NULL;
9323 break;
9324 }
9325
Wang Xiaoguang47b5d642016-09-07 20:17:38 +08009326 btrfs_block_rsv_release(fs_info, rsv, -1);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04009327 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
Josef Bacik25d609f2016-03-25 13:25:48 -04009328 rsv, min_size, 0);
Josef Bacikca7e70f2012-08-27 17:48:15 -04009329 BUG_ON(ret); /* shouldn't happen */
9330 trans->block_rsv = rsv;
Yan, Zheng80825102009-11-12 09:35:36 +00009331 }
9332
Josef Bacikddfae632017-10-19 14:16:02 -04009333 /*
9334 * We can't call btrfs_truncate_block inside a trans handle as we could
9335 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
9336 * we've truncated everything except the last little bit, and can do
9337 * btrfs_truncate_block and then update the disk_i_size.
9338 */
9339 if (ret == NEED_TRUNCATE_BLOCK) {
9340 btrfs_end_transaction(trans);
9341 btrfs_btree_balance_dirty(fs_info);
9342
9343 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
9344 if (ret)
9345 goto out;
9346 trans = btrfs_start_transaction(root, 1);
9347 if (IS_ERR(trans)) {
9348 ret = PTR_ERR(trans);
9349 goto out;
9350 }
9351 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
9352 }
9353
Yan, Zheng80825102009-11-12 09:35:36 +00009354 if (ret == 0 && inode->i_nlink > 0) {
Josef Bacikfcb80c22011-05-03 10:40:22 -04009355 trans->block_rsv = root->orphan_block_rsv;
Nikolay Borisov3d6ae7b2017-02-20 13:50:58 +02009356 ret = btrfs_orphan_del(trans, BTRFS_I(inode));
Josef Bacik3893e332011-01-31 16:03:11 -05009357 if (ret)
9358 err = ret;
Yan, Zheng80825102009-11-12 09:35:36 +00009359 }
9360
Chris Mason917c16b2011-11-08 14:49:59 -05009361 if (trans) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04009362 trans->block_rsv = &fs_info->trans_block_rsv;
Chris Mason917c16b2011-11-08 14:49:59 -05009363 ret = btrfs_update_inode(trans, root, inode);
9364 if (ret && !err)
9365 err = ret;
Josef Bacik7b128762008-07-24 12:17:14 -04009366
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04009367 ret = btrfs_end_transaction(trans);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04009368 btrfs_btree_balance_dirty(fs_info);
Chris Mason917c16b2011-11-08 14:49:59 -05009369 }
Josef Bacikfcb80c22011-05-03 10:40:22 -04009370out:
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04009371 btrfs_free_block_rsv(fs_info, rsv);
Josef Bacikfcb80c22011-05-03 10:40:22 -04009372
Josef Bacik3893e332011-01-31 16:03:11 -05009373 if (ret && !err)
9374 err = ret;
Josef Bacika41ad392011-01-31 15:30:16 -05009375
Josef Bacik3893e332011-01-31 16:03:11 -05009376 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04009377}
9378
Sven Wegener3b963622008-06-09 21:57:42 -04009379/*
Chris Masond352ac62008-09-29 15:18:18 -04009380 * create a new subvolume directory/inode (helper for the ioctl).
9381 */
Yan Zhengd2fb3432008-12-11 16:30:39 -05009382int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
Filipe David Borba Manana63541922014-01-07 11:47:46 +00009383 struct btrfs_root *new_root,
9384 struct btrfs_root *parent_root,
9385 u64 new_dirid)
Chris Mason39279cc2007-06-12 06:35:45 -04009386{
Chris Mason39279cc2007-06-12 06:35:45 -04009387 struct inode *inode;
Yan, Zheng76dda932009-09-21 16:00:26 -04009388 int err;
Chris Mason00e4e6b2008-08-05 11:18:09 -04009389 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04009390
Florian Albrechtskirchinger12fc9d02012-02-10 22:15:54 +01009391 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9392 new_dirid, new_dirid,
9393 S_IFDIR | (~current_umask() & S_IRWXUGO),
9394 &index);
Chris Mason54aa1f42007-06-22 14:16:25 -04009395 if (IS_ERR(inode))
Christoph Hellwigf46b5a62008-06-11 21:53:53 -04009396 return PTR_ERR(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04009397 inode->i_op = &btrfs_dir_inode_operations;
9398 inode->i_fop = &btrfs_dir_file_operations;
9399
Miklos Szeredibfe86842011-10-28 14:13:29 +02009400 set_nlink(inode, 1);
Nikolay Borisov6ef06d22017-02-20 13:50:34 +02009401 btrfs_i_size_write(BTRFS_I(inode), 0);
Chris Masonb0d5d102014-09-08 13:08:51 -07009402 unlock_new_inode(inode);
Sven Wegener3b963622008-06-09 21:57:42 -04009403
Filipe David Borba Manana63541922014-01-07 11:47:46 +00009404 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9405 if (err)
9406 btrfs_err(new_root->fs_info,
David Sterba351fd352014-05-15 16:48:20 +02009407 "error inheriting subvolume %llu properties: %d",
Filipe David Borba Manana63541922014-01-07 11:47:46 +00009408 new_root->root_key.objectid, err);
9409
Yan, Zheng76dda932009-09-21 16:00:26 -04009410 err = btrfs_update_inode(trans, new_root, inode);
Christoph Hellwigcb8e7092008-10-09 13:39:39 -04009411
Yan, Zheng76dda932009-09-21 16:00:26 -04009412 iput(inode);
Mark Fashehce598972011-07-26 11:32:23 -07009413 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04009414}
9415
Chris Mason39279cc2007-06-12 06:35:45 -04009416struct inode *btrfs_alloc_inode(struct super_block *sb)
9417{
Josef Bacik69fe2d72017-10-19 14:15:57 -04009418 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
Chris Mason39279cc2007-06-12 06:35:45 -04009419 struct btrfs_inode *ei;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009420 struct inode *inode;
Chris Mason39279cc2007-06-12 06:35:45 -04009421
9422 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9423 if (!ei)
9424 return NULL;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009425
9426 ei->root = NULL;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009427 ei->generation = 0;
Josef Bacik15ee9bc2007-08-10 16:22:09 -04009428 ei->last_trans = 0;
Chris Mason257c62e2009-10-13 13:21:08 -04009429 ei->last_sub_trans = 0;
Chris Masone02119d2008-09-05 16:13:11 -04009430 ei->logged_trans = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009431 ei->delalloc_bytes = 0;
Filipe Mananaa7e3b972017-04-03 10:45:46 +01009432 ei->new_delalloc_bytes = 0;
Wang Shilong47059d92014-07-03 18:22:07 +08009433 ei->defrag_bytes = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009434 ei->disk_i_size = 0;
9435 ei->flags = 0;
Josef Bacik7709cde2011-08-04 10:25:02 -04009436 ei->csum_bytes = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009437 ei->index_cnt = (u64)-1;
Miao Xie67de1172013-12-26 13:07:06 +08009438 ei->dir_index = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009439 ei->last_unlink_trans = 0;
Liu Bo46d8bc32012-08-29 01:07:55 -06009440 ei->last_log_commit = 0;
David Sterba8089fe62015-11-19 14:15:51 +01009441 ei->delayed_iput_count = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009442
Josef Bacik9e0baf62011-07-15 15:16:44 +00009443 spin_lock_init(&ei->lock);
9444 ei->outstanding_extents = 0;
Josef Bacik69fe2d72017-10-19 14:15:57 -04009445 if (sb->s_magic != BTRFS_TEST_MAGIC)
9446 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
9447 BTRFS_BLOCK_RSV_DELALLOC);
Josef Bacik72ac3c02012-05-23 14:13:11 -04009448 ei->runtime_flags = 0;
David Sterbab52aa8c2017-07-17 19:17:20 +02009449 ei->prop_compress = BTRFS_COMPRESS_NONE;
David Sterbaeec63c62017-07-17 19:41:31 +02009450 ei->defrag_compress = BTRFS_COMPRESS_NONE;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009451
Miao Xie16cdcec2011-04-22 18:12:22 +08009452 ei->delayed_node = NULL;
9453
chandan r9cc97d62012-07-04 12:48:07 +05309454 ei->i_otime.tv_sec = 0;
9455 ei->i_otime.tv_nsec = 0;
9456
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009457 inode = &ei->vfs_inode;
David Sterbaa8067e02011-04-21 00:34:43 +02009458 extent_map_tree_init(&ei->extent_tree);
Josef Bacikc6100a42017-05-05 11:57:13 -04009459 extent_io_tree_init(&ei->io_tree, inode);
9460 extent_io_tree_init(&ei->io_failure_tree, inode);
Josef Bacik0b32f4b2012-03-13 09:38:00 -04009461 ei->io_tree.track_uptodate = 1;
9462 ei->io_failure_tree.track_uptodate = 1;
Josef Bacikb812ce22012-11-16 13:56:32 -05009463 atomic_set(&ei->sync_writers, 0);
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009464 mutex_init(&ei->log_mutex);
Josef Bacikf2486792012-01-13 12:09:22 -05009465 mutex_init(&ei->delalloc_mutex);
Chris Masone6dcd2d2008-07-17 12:53:50 -04009466 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009467 INIT_LIST_HEAD(&ei->delalloc_inodes);
David Sterba8089fe62015-11-19 14:15:51 +01009468 INIT_LIST_HEAD(&ei->delayed_iput);
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009469 RB_CLEAR_NODE(&ei->rb_node);
Filipe Manana5f9a8a52016-05-12 13:53:36 +01009470 init_rwsem(&ei->dio_sem);
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009471
9472 return inode;
Chris Mason39279cc2007-06-12 06:35:45 -04009473}
9474
Josef Bacikaaedb552013-10-11 14:44:09 -04009475#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9476void btrfs_test_destroy_inode(struct inode *inode)
9477{
Nikolay Borisovdcdbc052017-02-20 13:50:45 +02009478 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
Josef Bacikaaedb552013-10-11 14:44:09 -04009479 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9480}
9481#endif
9482
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11009483static void btrfs_i_callback(struct rcu_head *head)
9484{
9485 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11009486 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9487}
9488
Chris Mason39279cc2007-06-12 06:35:45 -04009489void btrfs_destroy_inode(struct inode *inode)
9490{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04009491 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Chris Masone6dcd2d2008-07-17 12:53:50 -04009492 struct btrfs_ordered_extent *ordered;
Chris Mason5a3f23d2009-03-31 13:27:11 -04009493 struct btrfs_root *root = BTRFS_I(inode)->root;
9494
Al Virob3d9b7a2012-06-09 13:51:19 -04009495 WARN_ON(!hlist_empty(&inode->i_dentry));
Chris Mason39279cc2007-06-12 06:35:45 -04009496 WARN_ON(inode->i_data.nrpages);
Josef Bacik69fe2d72017-10-19 14:15:57 -04009497 WARN_ON(BTRFS_I(inode)->block_rsv.reserved);
9498 WARN_ON(BTRFS_I(inode)->block_rsv.size);
Josef Bacik9e0baf62011-07-15 15:16:44 +00009499 WARN_ON(BTRFS_I(inode)->outstanding_extents);
Josef Bacik7709cde2011-08-04 10:25:02 -04009500 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
Filipe Mananaa7e3b972017-04-03 10:45:46 +01009501 WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
Josef Bacik7709cde2011-08-04 10:25:02 -04009502 WARN_ON(BTRFS_I(inode)->csum_bytes);
Wang Shilong47059d92014-07-03 18:22:07 +08009503 WARN_ON(BTRFS_I(inode)->defrag_bytes);
Chris Mason39279cc2007-06-12 06:35:45 -04009504
Chris Mason5a3f23d2009-03-31 13:27:11 -04009505 /*
Josef Bacika6dbd422009-11-11 15:53:34 -05009506 * This can happen where we create an inode, but somebody else also
9507 * created the same inode and we need to destroy the one we already
9508 * created.
9509 */
9510 if (!root)
9511 goto free;
9512
Josef Bacik8a35d952012-05-23 14:26:42 -04009513 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9514 &BTRFS_I(inode)->runtime_flags)) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04009515 btrfs_info(fs_info, "inode %llu still on the orphan list",
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02009516 btrfs_ino(BTRFS_I(inode)));
Josef Bacik8a35d952012-05-23 14:26:42 -04009517 atomic_dec(&root->orphan_inodes);
Josef Bacik7b128762008-07-24 12:17:14 -04009518 }
Josef Bacik7b128762008-07-24 12:17:14 -04009519
Chris Masond3977122009-01-05 21:25:51 -05009520 while (1) {
Chris Masone6dcd2d2008-07-17 12:53:50 -04009521 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9522 if (!ordered)
9523 break;
9524 else {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04009525 btrfs_err(fs_info,
Jeff Mahoney5d163e02016-09-20 10:05:00 -04009526 "found ordered extent %llu %llu on inode cleanup",
9527 ordered->file_offset, ordered->len);
Chris Masone6dcd2d2008-07-17 12:53:50 -04009528 btrfs_remove_ordered_extent(inode, ordered);
9529 btrfs_put_ordered_extent(ordered);
9530 btrfs_put_ordered_extent(ordered);
9531 }
9532 }
Qu Wenruo56fa9d02015-10-13 09:53:10 +08009533 btrfs_qgroup_check_reserved_leak(inode);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04009534 inode_tree_del(inode);
Nikolay Borisovdcdbc052017-02-20 13:50:45 +02009535 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
Josef Bacika6dbd422009-11-11 15:53:34 -05009536free:
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11009537 call_rcu(&inode->i_rcu, btrfs_i_callback);
Chris Mason39279cc2007-06-12 06:35:45 -04009538}
9539
Al Viro45321ac2010-06-07 13:43:19 -04009540int btrfs_drop_inode(struct inode *inode)
Yan, Zheng76dda932009-09-21 16:00:26 -04009541{
9542 struct btrfs_root *root = BTRFS_I(inode)->root;
Al Viro45321ac2010-06-07 13:43:19 -04009543
Naohiro Aota6379ef92013-06-06 09:56:34 +00009544 if (root == NULL)
9545 return 1;
9546
Liu Bofa6ac872013-02-20 14:10:23 +00009547 /* the snap/subvol tree is on deleting */
Stefan Behrens69e9c6c2013-09-05 16:58:43 +02009548 if (btrfs_root_refs(&root->root_item) == 0)
Al Viro45321ac2010-06-07 13:43:19 -04009549 return 1;
Yan, Zheng76dda932009-09-21 16:00:26 -04009550 else
Al Viro45321ac2010-06-07 13:43:19 -04009551 return generic_drop_inode(inode);
Yan, Zheng76dda932009-09-21 16:00:26 -04009552}
9553
Sven Wegener0ee0fda2008-07-30 16:54:26 -04009554static void init_once(void *foo)
Chris Mason39279cc2007-06-12 06:35:45 -04009555{
9556 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9557
9558 inode_init_once(&ei->vfs_inode);
9559}
9560
9561void btrfs_destroy_cachep(void)
9562{
Kirill A. Shutemov8c0a8532012-09-26 11:33:07 +10009563 /*
9564 * Make sure all delayed rcu free inodes are flushed before we
9565 * destroy cache.
9566 */
9567 rcu_barrier();
Kinglong Mee5598e902016-01-29 21:36:35 +08009568 kmem_cache_destroy(btrfs_inode_cachep);
9569 kmem_cache_destroy(btrfs_trans_handle_cachep);
Kinglong Mee5598e902016-01-29 21:36:35 +08009570 kmem_cache_destroy(btrfs_path_cachep);
9571 kmem_cache_destroy(btrfs_free_space_cachep);
Chris Mason39279cc2007-06-12 06:35:45 -04009572}
9573
9574int btrfs_init_cachep(void)
9575{
David Sterba837e1972012-09-07 03:00:48 -06009576 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02009577 sizeof(struct btrfs_inode), 0,
Vladimir Davydov5d097052016-01-14 15:18:21 -08009578 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9579 init_once);
Chris Mason39279cc2007-06-12 06:35:45 -04009580 if (!btrfs_inode_cachep)
9581 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02009582
David Sterba837e1972012-09-07 03:00:48 -06009583 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02009584 sizeof(struct btrfs_trans_handle), 0,
Nikolay Borisovfba4b692016-06-23 21:17:08 +03009585 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04009586 if (!btrfs_trans_handle_cachep)
9587 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02009588
David Sterba837e1972012-09-07 03:00:48 -06009589 btrfs_path_cachep = kmem_cache_create("btrfs_path",
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02009590 sizeof(struct btrfs_path), 0,
Nikolay Borisovfba4b692016-06-23 21:17:08 +03009591 SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04009592 if (!btrfs_path_cachep)
9593 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02009594
David Sterba837e1972012-09-07 03:00:48 -06009595 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
Josef Bacikdc89e982011-01-28 17:05:48 -05009596 sizeof(struct btrfs_free_space), 0,
Nikolay Borisovfba4b692016-06-23 21:17:08 +03009597 SLAB_MEM_SPREAD, NULL);
Josef Bacikdc89e982011-01-28 17:05:48 -05009598 if (!btrfs_free_space_cachep)
9599 goto fail;
9600
Chris Mason39279cc2007-06-12 06:35:45 -04009601 return 0;
9602fail:
9603 btrfs_destroy_cachep();
9604 return -ENOMEM;
9605}
9606
David Howellsa528d352017-01-31 16:46:22 +00009607static int btrfs_getattr(const struct path *path, struct kstat *stat,
9608 u32 request_mask, unsigned int flags)
Chris Mason39279cc2007-06-12 06:35:45 -04009609{
Miao Xiedf0af1a2013-01-29 10:11:59 +00009610 u64 delalloc_bytes;
David Howellsa528d352017-01-31 16:46:22 +00009611 struct inode *inode = d_inode(path->dentry);
David Sterbafadc0d82011-11-20 07:33:38 -05009612 u32 blocksize = inode->i_sb->s_blocksize;
Yonghong Song04a87e32017-05-12 15:07:43 -07009613 u32 bi_flags = BTRFS_I(inode)->flags;
9614
9615 stat->result_mask |= STATX_BTIME;
9616 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9617 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9618 if (bi_flags & BTRFS_INODE_APPEND)
9619 stat->attributes |= STATX_ATTR_APPEND;
9620 if (bi_flags & BTRFS_INODE_COMPRESS)
9621 stat->attributes |= STATX_ATTR_COMPRESSED;
9622 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9623 stat->attributes |= STATX_ATTR_IMMUTABLE;
9624 if (bi_flags & BTRFS_INODE_NODUMP)
9625 stat->attributes |= STATX_ATTR_NODUMP;
9626
9627 stat->attributes_mask |= (STATX_ATTR_APPEND |
9628 STATX_ATTR_COMPRESSED |
9629 STATX_ATTR_IMMUTABLE |
9630 STATX_ATTR_NODUMP);
David Sterbafadc0d82011-11-20 07:33:38 -05009631
Chris Mason39279cc2007-06-12 06:35:45 -04009632 generic_fillattr(inode, stat);
Al Viro0ee5dc62011-07-07 15:44:25 -04009633 stat->dev = BTRFS_I(inode)->root->anon_dev;
Miao Xiedf0af1a2013-01-29 10:11:59 +00009634
9635 spin_lock(&BTRFS_I(inode)->lock);
Filipe Mananaa7e3b972017-04-03 10:45:46 +01009636 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
Miao Xiedf0af1a2013-01-29 10:11:59 +00009637 spin_unlock(&BTRFS_I(inode)->lock);
David Sterbafadc0d82011-11-20 07:33:38 -05009638 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
Miao Xiedf0af1a2013-01-29 10:11:59 +00009639 ALIGN(delalloc_bytes, blocksize)) >> 9;
Chris Mason39279cc2007-06-12 06:35:45 -04009640 return 0;
9641}
9642
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009643static int btrfs_rename_exchange(struct inode *old_dir,
9644 struct dentry *old_dentry,
9645 struct inode *new_dir,
9646 struct dentry *new_dentry)
Chris Mason39279cc2007-06-12 06:35:45 -04009647{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04009648 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
Chris Mason39279cc2007-06-12 06:35:45 -04009649 struct btrfs_trans_handle *trans;
9650 struct btrfs_root *root = BTRFS_I(old_dir)->root;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04009651 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009652 struct inode *new_inode = new_dentry->d_inode;
9653 struct inode *old_inode = old_dentry->d_inode;
Deepa Dinamanic2050a42016-09-14 07:48:06 -07009654 struct timespec ctime = current_time(old_inode);
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009655 struct dentry *parent;
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02009656 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9657 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009658 u64 old_idx = 0;
9659 u64 new_idx = 0;
9660 u64 root_objectid;
9661 int ret;
Filipe Manana86e8aa02016-05-05 02:02:27 +01009662 bool root_log_pinned = false;
9663 bool dest_log_pinned = false;
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009664
9665 /* we only allow rename subvolume link between subvolumes */
9666 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9667 return -EXDEV;
9668
9669 /* close the race window with snapshot create/destroy ioctl */
9670 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
Jeff Mahoney0b246af2016-06-22 18:54:23 -04009671 down_read(&fs_info->subvol_sem);
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009672 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
Jeff Mahoney0b246af2016-06-22 18:54:23 -04009673 down_read(&fs_info->subvol_sem);
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009674
9675 /*
9676 * We want to reserve the absolute worst case amount of items. So if
9677 * both inodes are subvols and we need to unlink them then that would
9678 * require 4 item modifications, but if they are both normal inodes it
9679 * would require 5 item modifications, so we'll assume their normal
9680 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9681 * should cover the worst case number of items we'll modify.
9682 */
9683 trans = btrfs_start_transaction(root, 12);
9684 if (IS_ERR(trans)) {
9685 ret = PTR_ERR(trans);
9686 goto out_notrans;
9687 }
9688
9689 /*
9690 * We need to find a free sequence number both in the source and
9691 * in the destination directory for the exchange.
9692 */
Nikolay Borisov877574e2017-02-20 13:50:33 +02009693 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009694 if (ret)
9695 goto out_fail;
Nikolay Borisov877574e2017-02-20 13:50:33 +02009696 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009697 if (ret)
9698 goto out_fail;
9699
9700 BTRFS_I(old_inode)->dir_index = 0ULL;
9701 BTRFS_I(new_inode)->dir_index = 0ULL;
9702
9703 /* Reference for the source. */
9704 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9705 /* force full log commit if subvolume involved. */
Jeff Mahoney0b246af2016-06-22 18:54:23 -04009706 btrfs_set_log_full_commit(fs_info, trans);
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009707 } else {
Filipe Manana376e5a52016-05-05 02:08:56 +01009708 btrfs_pin_log_trans(root);
9709 root_log_pinned = true;
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009710 ret = btrfs_insert_inode_ref(trans, dest,
9711 new_dentry->d_name.name,
9712 new_dentry->d_name.len,
9713 old_ino,
David Sterbaf85b7372017-01-20 14:54:07 +01009714 btrfs_ino(BTRFS_I(new_dir)),
9715 old_idx);
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009716 if (ret)
9717 goto out_fail;
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009718 }
9719
9720 /* And now for the dest. */
9721 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9722 /* force full log commit if subvolume involved. */
Jeff Mahoney0b246af2016-06-22 18:54:23 -04009723 btrfs_set_log_full_commit(fs_info, trans);
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009724 } else {
Filipe Manana376e5a52016-05-05 02:08:56 +01009725 btrfs_pin_log_trans(dest);
9726 dest_log_pinned = true;
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009727 ret = btrfs_insert_inode_ref(trans, root,
9728 old_dentry->d_name.name,
9729 old_dentry->d_name.len,
9730 new_ino,
David Sterbaf85b7372017-01-20 14:54:07 +01009731 btrfs_ino(BTRFS_I(old_dir)),
9732 new_idx);
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009733 if (ret)
9734 goto out_fail;
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009735 }
9736
9737 /* Update inode version and ctime/mtime. */
9738 inode_inc_iversion(old_dir);
9739 inode_inc_iversion(new_dir);
9740 inode_inc_iversion(old_inode);
9741 inode_inc_iversion(new_inode);
9742 old_dir->i_ctime = old_dir->i_mtime = ctime;
9743 new_dir->i_ctime = new_dir->i_mtime = ctime;
9744 old_inode->i_ctime = ctime;
9745 new_inode->i_ctime = ctime;
9746
9747 if (old_dentry->d_parent != new_dentry->d_parent) {
David Sterbaf85b7372017-01-20 14:54:07 +01009748 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9749 BTRFS_I(old_inode), 1);
9750 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9751 BTRFS_I(new_inode), 1);
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009752 }
9753
9754 /* src is a subvolume */
9755 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9756 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9757 ret = btrfs_unlink_subvol(trans, root, old_dir,
9758 root_objectid,
9759 old_dentry->d_name.name,
9760 old_dentry->d_name.len);
9761 } else { /* src is an inode */
Nikolay Borisov4ec59342017-01-18 00:31:44 +02009762 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9763 BTRFS_I(old_dentry->d_inode),
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009764 old_dentry->d_name.name,
9765 old_dentry->d_name.len);
9766 if (!ret)
9767 ret = btrfs_update_inode(trans, root, old_inode);
9768 }
9769 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04009770 btrfs_abort_transaction(trans, ret);
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009771 goto out_fail;
9772 }
9773
9774 /* dest is a subvolume */
9775 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9776 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9777 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9778 root_objectid,
9779 new_dentry->d_name.name,
9780 new_dentry->d_name.len);
9781 } else { /* dest is an inode */
Nikolay Borisov4ec59342017-01-18 00:31:44 +02009782 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9783 BTRFS_I(new_dentry->d_inode),
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009784 new_dentry->d_name.name,
9785 new_dentry->d_name.len);
9786 if (!ret)
9787 ret = btrfs_update_inode(trans, dest, new_inode);
9788 }
9789 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04009790 btrfs_abort_transaction(trans, ret);
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009791 goto out_fail;
9792 }
9793
Nikolay Borisovdb0a6692017-02-20 13:51:08 +02009794 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009795 new_dentry->d_name.name,
9796 new_dentry->d_name.len, 0, old_idx);
9797 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04009798 btrfs_abort_transaction(trans, ret);
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009799 goto out_fail;
9800 }
9801
Nikolay Borisovdb0a6692017-02-20 13:51:08 +02009802 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009803 old_dentry->d_name.name,
9804 old_dentry->d_name.len, 0, new_idx);
9805 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04009806 btrfs_abort_transaction(trans, ret);
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009807 goto out_fail;
9808 }
9809
9810 if (old_inode->i_nlink == 1)
9811 BTRFS_I(old_inode)->dir_index = old_idx;
9812 if (new_inode->i_nlink == 1)
9813 BTRFS_I(new_inode)->dir_index = new_idx;
9814
Filipe Manana86e8aa02016-05-05 02:02:27 +01009815 if (root_log_pinned) {
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009816 parent = new_dentry->d_parent;
David Sterbaf85b7372017-01-20 14:54:07 +01009817 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9818 parent);
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009819 btrfs_end_log_trans(root);
Filipe Manana86e8aa02016-05-05 02:02:27 +01009820 root_log_pinned = false;
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009821 }
Filipe Manana86e8aa02016-05-05 02:02:27 +01009822 if (dest_log_pinned) {
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009823 parent = old_dentry->d_parent;
David Sterbaf85b7372017-01-20 14:54:07 +01009824 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9825 parent);
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009826 btrfs_end_log_trans(dest);
Filipe Manana86e8aa02016-05-05 02:02:27 +01009827 dest_log_pinned = false;
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009828 }
9829out_fail:
Filipe Manana86e8aa02016-05-05 02:02:27 +01009830 /*
9831 * If we have pinned a log and an error happened, we unpin tasks
9832 * trying to sync the log and force them to fallback to a transaction
9833 * commit if the log currently contains any of the inodes involved in
9834 * this rename operation (to ensure we do not persist a log with an
9835 * inconsistent state for any of these inodes or leading to any
9836 * inconsistencies when replayed). If the transaction was aborted, the
9837 * abortion reason is propagated to userspace when attempting to commit
9838 * the transaction. If the log does not contain any of these inodes, we
9839 * allow the tasks to sync it.
9840 */
9841 if (ret && (root_log_pinned || dest_log_pinned)) {
Nikolay Borisov0f8939b2017-01-18 00:31:30 +02009842 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9843 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9844 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
Filipe Manana86e8aa02016-05-05 02:02:27 +01009845 (new_inode &&
Nikolay Borisov0f8939b2017-01-18 00:31:30 +02009846 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
Jeff Mahoney0b246af2016-06-22 18:54:23 -04009847 btrfs_set_log_full_commit(fs_info, trans);
Filipe Manana86e8aa02016-05-05 02:02:27 +01009848
9849 if (root_log_pinned) {
9850 btrfs_end_log_trans(root);
9851 root_log_pinned = false;
9852 }
9853 if (dest_log_pinned) {
9854 btrfs_end_log_trans(dest);
9855 dest_log_pinned = false;
9856 }
9857 }
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04009858 ret = btrfs_end_transaction(trans);
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009859out_notrans:
9860 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
Jeff Mahoney0b246af2016-06-22 18:54:23 -04009861 up_read(&fs_info->subvol_sem);
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009862 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
Jeff Mahoney0b246af2016-06-22 18:54:23 -04009863 up_read(&fs_info->subvol_sem);
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009864
9865 return ret;
9866}
9867
9868static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9869 struct btrfs_root *root,
9870 struct inode *dir,
9871 struct dentry *dentry)
9872{
9873 int ret;
9874 struct inode *inode;
9875 u64 objectid;
9876 u64 index;
9877
9878 ret = btrfs_find_free_ino(root, &objectid);
9879 if (ret)
9880 return ret;
9881
9882 inode = btrfs_new_inode(trans, root, dir,
9883 dentry->d_name.name,
9884 dentry->d_name.len,
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02009885 btrfs_ino(BTRFS_I(dir)),
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009886 objectid,
9887 S_IFCHR | WHITEOUT_MODE,
9888 &index);
9889
9890 if (IS_ERR(inode)) {
9891 ret = PTR_ERR(inode);
9892 return ret;
9893 }
9894
9895 inode->i_op = &btrfs_special_inode_operations;
9896 init_special_inode(inode, inode->i_mode,
9897 WHITEOUT_DEV);
9898
9899 ret = btrfs_init_inode_security(trans, inode, dir,
9900 &dentry->d_name);
9901 if (ret)
Filipe Mananac9901612016-05-05 01:41:57 +01009902 goto out;
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009903
Nikolay Borisovcef415a2017-02-20 13:51:09 +02009904 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9905 BTRFS_I(inode), 0, index);
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009906 if (ret)
Filipe Mananac9901612016-05-05 01:41:57 +01009907 goto out;
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009908
9909 ret = btrfs_update_inode(trans, root, inode);
Filipe Mananac9901612016-05-05 01:41:57 +01009910out:
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009911 unlock_new_inode(inode);
Filipe Mananac9901612016-05-05 01:41:57 +01009912 if (ret)
9913 inode_dec_link_count(inode);
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009914 iput(inode);
9915
Filipe Mananac9901612016-05-05 01:41:57 +01009916 return ret;
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009917}
9918
Chris Mason39279cc2007-06-12 06:35:45 -04009919static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009920 struct inode *new_dir, struct dentry *new_dentry,
9921 unsigned int flags)
Chris Mason39279cc2007-06-12 06:35:45 -04009922{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04009923 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
Chris Mason39279cc2007-06-12 06:35:45 -04009924 struct btrfs_trans_handle *trans;
Filipe Manana5062af32016-05-05 10:26:26 +01009925 unsigned int trans_num_items;
Chris Mason39279cc2007-06-12 06:35:45 -04009926 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9927 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
David Howells2b0143b2015-03-17 22:25:59 +00009928 struct inode *new_inode = d_inode(new_dentry);
9929 struct inode *old_inode = d_inode(old_dentry);
Chris Mason00e4e6b2008-08-05 11:18:09 -04009930 u64 index = 0;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04009931 u64 root_objectid;
Chris Mason39279cc2007-06-12 06:35:45 -04009932 int ret;
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02009933 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
Filipe Manana3dc9e8f2016-04-29 11:34:22 +01009934 bool log_pinned = false;
Chris Mason39279cc2007-06-12 06:35:45 -04009935
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02009936 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
Yan, Zhengf679a842009-09-24 09:17:31 -04009937 return -EPERM;
9938
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04009939 /* we only allow rename subvolume link between subvolumes */
Li Zefan33345d012011-04-20 10:31:50 +08009940 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
Chris Mason3394e162008-11-17 20:42:26 -05009941 return -EXDEV;
9942
Li Zefan33345d012011-04-20 10:31:50 +08009943 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +02009944 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04009945 return -ENOTEMPTY;
9946
Chris Mason39279cc2007-06-12 06:35:45 -04009947 if (S_ISDIR(old_inode->i_mode) && new_inode &&
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04009948 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
Chris Mason39279cc2007-06-12 06:35:45 -04009949 return -ENOTEMPTY;
Chris Mason9c520572012-12-17 14:26:57 -05009950
9951
9952 /* check for collisions, even if the name isn't there */
Josef Bacik4871c152013-10-09 12:24:04 -04009953 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
Chris Mason9c520572012-12-17 14:26:57 -05009954 new_dentry->d_name.name,
9955 new_dentry->d_name.len);
9956
9957 if (ret) {
9958 if (ret == -EEXIST) {
9959 /* we shouldn't get
9960 * eexist without a new_inode */
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05309961 if (WARN_ON(!new_inode)) {
Chris Mason9c520572012-12-17 14:26:57 -05009962 return ret;
9963 }
9964 } else {
9965 /* maybe -EOVERFLOW */
9966 return ret;
9967 }
9968 }
9969 ret = 0;
9970
Chris Mason5a3f23d2009-03-31 13:27:11 -04009971 /*
Chris Mason8d875f92014-08-12 10:47:42 -07009972 * we're using rename to replace one file with another. Start IO on it
9973 * now so we don't add too much work to the end of the transaction
Chris Mason5a3f23d2009-03-31 13:27:11 -04009974 */
Chris Mason8d875f92014-08-12 10:47:42 -07009975 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
Chris Mason5a3f23d2009-03-31 13:27:11 -04009976 filemap_flush(old_inode->i_mapping);
9977
Yan, Zheng76dda932009-09-21 16:00:26 -04009978 /* close the racy window with snapshot create/destroy ioctl */
Li Zefan33345d012011-04-20 10:31:50 +08009979 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
Jeff Mahoney0b246af2016-06-22 18:54:23 -04009980 down_read(&fs_info->subvol_sem);
Yan, Zhenga22285a2010-05-16 10:48:46 -04009981 /*
9982 * We want to reserve the absolute worst case amount of items. So if
9983 * both inodes are subvols and we need to unlink them then that would
9984 * require 4 item modifications, but if they are both normal inodes it
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009985 * would require 5 item modifications, so we'll assume they are normal
Yan, Zhenga22285a2010-05-16 10:48:46 -04009986 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9987 * should cover the worst case number of items we'll modify.
Filipe Manana5062af32016-05-05 10:26:26 +01009988 * If our rename has the whiteout flag, we need more 5 units for the
9989 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9990 * when selinux is enabled).
Yan, Zhenga22285a2010-05-16 10:48:46 -04009991 */
Filipe Manana5062af32016-05-05 10:26:26 +01009992 trans_num_items = 11;
9993 if (flags & RENAME_WHITEOUT)
9994 trans_num_items += 5;
9995 trans = btrfs_start_transaction(root, trans_num_items);
Johann Lombardib44c59a2011-03-31 13:23:47 +00009996 if (IS_ERR(trans)) {
Dan Fuhrycdd1fed2016-03-17 15:23:38 +01009997 ret = PTR_ERR(trans);
9998 goto out_notrans;
9999 }
Chris Mason5f39d392007-10-15 16:14:19 -040010000
Yan, Zheng4df27c4d2009-09-21 15:56:00 -040010001 if (dest != root)
10002 btrfs_record_root_in_trans(trans, dest);
10003
Nikolay Borisov877574e2017-02-20 13:50:33 +020010004 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
Yan, Zhenga5719522009-09-24 09:17:31 -040010005 if (ret)
10006 goto out_fail;
Chris Mason5a3f23d2009-03-31 13:27:11 -040010007
Miao Xie67de1172013-12-26 13:07:06 +080010008 BTRFS_I(old_inode)->dir_index = 0ULL;
Li Zefan33345d012011-04-20 10:31:50 +080010009 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -040010010 /* force full log commit if subvolume involved. */
Jeff Mahoney0b246af2016-06-22 18:54:23 -040010011 btrfs_set_log_full_commit(fs_info, trans);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -040010012 } else {
Filipe Mananac4aba952016-04-29 13:14:42 +010010013 btrfs_pin_log_trans(root);
10014 log_pinned = true;
Yan, Zhenga5719522009-09-24 09:17:31 -040010015 ret = btrfs_insert_inode_ref(trans, dest,
10016 new_dentry->d_name.name,
10017 new_dentry->d_name.len,
Li Zefan33345d012011-04-20 10:31:50 +080010018 old_ino,
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +020010019 btrfs_ino(BTRFS_I(new_dir)), index);
Yan, Zhenga5719522009-09-24 09:17:31 -040010020 if (ret)
10021 goto out_fail;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -040010022 }
Chris Mason5a3f23d2009-03-31 13:27:11 -040010023
Josef Bacik0c4d2d92012-04-05 15:03:02 -040010024 inode_inc_iversion(old_dir);
10025 inode_inc_iversion(new_dir);
10026 inode_inc_iversion(old_inode);
Deepa Dinamani04b285f2016-02-06 23:57:21 -080010027 old_dir->i_ctime = old_dir->i_mtime =
10028 new_dir->i_ctime = new_dir->i_mtime =
Deepa Dinamanic2050a42016-09-14 07:48:06 -070010029 old_inode->i_ctime = current_time(old_dir);
Chris Mason5f39d392007-10-15 16:14:19 -040010030
Chris Mason12fcfd22009-03-24 10:24:20 -040010031 if (old_dentry->d_parent != new_dentry->d_parent)
David Sterbaf85b7372017-01-20 14:54:07 +010010032 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
10033 BTRFS_I(old_inode), 1);
Chris Mason12fcfd22009-03-24 10:24:20 -040010034
Li Zefan33345d012011-04-20 10:31:50 +080010035 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -040010036 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
10037 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
10038 old_dentry->d_name.name,
10039 old_dentry->d_name.len);
10040 } else {
Nikolay Borisov4ec59342017-01-18 00:31:44 +020010041 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
10042 BTRFS_I(d_inode(old_dentry)),
Al Viro92986792011-03-04 17:14:37 +000010043 old_dentry->d_name.name,
10044 old_dentry->d_name.len);
10045 if (!ret)
10046 ret = btrfs_update_inode(trans, root, old_inode);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -040010047 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +010010048 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -040010049 btrfs_abort_transaction(trans, ret);
Jeff Mahoney79787ea2012-03-12 16:03:00 +010010050 goto out_fail;
10051 }
Chris Mason39279cc2007-06-12 06:35:45 -040010052
10053 if (new_inode) {
Josef Bacik0c4d2d92012-04-05 15:03:02 -040010054 inode_inc_iversion(new_inode);
Deepa Dinamanic2050a42016-09-14 07:48:06 -070010055 new_inode->i_ctime = current_time(new_inode);
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +020010056 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
Yan, Zheng4df27c4d2009-09-21 15:56:00 -040010057 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
10058 root_objectid = BTRFS_I(new_inode)->location.objectid;
10059 ret = btrfs_unlink_subvol(trans, dest, new_dir,
10060 root_objectid,
10061 new_dentry->d_name.name,
10062 new_dentry->d_name.len);
10063 BUG_ON(new_inode->i_nlink == 0);
10064 } else {
Nikolay Borisov4ec59342017-01-18 00:31:44 +020010065 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
10066 BTRFS_I(d_inode(new_dentry)),
Yan, Zheng4df27c4d2009-09-21 15:56:00 -040010067 new_dentry->d_name.name,
10068 new_dentry->d_name.len);
10069 }
Josef Bacik4ef31a42013-08-13 14:10:08 -040010070 if (!ret && new_inode->i_nlink == 0)
Nikolay Borisov73f2e542017-02-20 13:50:59 +020010071 ret = btrfs_orphan_add(trans,
10072 BTRFS_I(d_inode(new_dentry)));
Jeff Mahoney79787ea2012-03-12 16:03:00 +010010073 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -040010074 btrfs_abort_transaction(trans, ret);
Jeff Mahoney79787ea2012-03-12 16:03:00 +010010075 goto out_fail;
10076 }
Chris Mason39279cc2007-06-12 06:35:45 -040010077 }
Josef Bacikaec74772008-07-24 12:12:38 -040010078
Nikolay Borisovdb0a6692017-02-20 13:51:08 +020010079 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
Yan, Zheng4df27c4d2009-09-21 15:56:00 -040010080 new_dentry->d_name.name,
Yan, Zhenga5719522009-09-24 09:17:31 -040010081 new_dentry->d_name.len, 0, index);
Jeff Mahoney79787ea2012-03-12 16:03:00 +010010082 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -040010083 btrfs_abort_transaction(trans, ret);
Jeff Mahoney79787ea2012-03-12 16:03:00 +010010084 goto out_fail;
10085 }
Chris Mason39279cc2007-06-12 06:35:45 -040010086
Miao Xie67de1172013-12-26 13:07:06 +080010087 if (old_inode->i_nlink == 1)
10088 BTRFS_I(old_inode)->dir_index = index;
10089
Filipe Manana3dc9e8f2016-04-29 11:34:22 +010010090 if (log_pinned) {
Al Viro10d9f302011-07-16 23:09:10 -040010091 struct dentry *parent = new_dentry->d_parent;
Filipe Manana3dc9e8f2016-04-29 11:34:22 +010010092
David Sterbaf85b7372017-01-20 14:54:07 +010010093 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
10094 parent);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -040010095 btrfs_end_log_trans(root);
Filipe Manana3dc9e8f2016-04-29 11:34:22 +010010096 log_pinned = false;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -040010097 }
Dan Fuhrycdd1fed2016-03-17 15:23:38 +010010098
10099 if (flags & RENAME_WHITEOUT) {
10100 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
10101 old_dentry);
10102
10103 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -040010104 btrfs_abort_transaction(trans, ret);
Dan Fuhrycdd1fed2016-03-17 15:23:38 +010010105 goto out_fail;
10106 }
Chris Mason12fcfd22009-03-24 10:24:20 -040010107 }
Chris Mason39279cc2007-06-12 06:35:45 -040010108out_fail:
Filipe Manana3dc9e8f2016-04-29 11:34:22 +010010109 /*
10110 * If we have pinned the log and an error happened, we unpin tasks
10111 * trying to sync the log and force them to fallback to a transaction
10112 * commit if the log currently contains any of the inodes involved in
10113 * this rename operation (to ensure we do not persist a log with an
10114 * inconsistent state for any of these inodes or leading to any
10115 * inconsistencies when replayed). If the transaction was aborted, the
10116 * abortion reason is propagated to userspace when attempting to commit
10117 * the transaction. If the log does not contain any of these inodes, we
10118 * allow the tasks to sync it.
10119 */
10120 if (ret && log_pinned) {
Nikolay Borisov0f8939b2017-01-18 00:31:30 +020010121 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
10122 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
10123 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
Filipe Manana3dc9e8f2016-04-29 11:34:22 +010010124 (new_inode &&
Nikolay Borisov0f8939b2017-01-18 00:31:30 +020010125 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
Jeff Mahoney0b246af2016-06-22 18:54:23 -040010126 btrfs_set_log_full_commit(fs_info, trans);
Filipe Manana3dc9e8f2016-04-29 11:34:22 +010010127
10128 btrfs_end_log_trans(root);
10129 log_pinned = false;
10130 }
Jeff Mahoney3a45bb22016-09-09 21:39:03 -040010131 btrfs_end_transaction(trans);
Johann Lombardib44c59a2011-03-31 13:23:47 +000010132out_notrans:
Li Zefan33345d012011-04-20 10:31:50 +080010133 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
Jeff Mahoney0b246af2016-06-22 18:54:23 -040010134 up_read(&fs_info->subvol_sem);
Josef Bacik9ed74f22009-09-11 16:12:44 -040010135
Chris Mason39279cc2007-06-12 06:35:45 -040010136 return ret;
10137}
10138
Miklos Szeredi80ace852014-07-23 15:15:32 +020010139static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
10140 struct inode *new_dir, struct dentry *new_dentry,
10141 unsigned int flags)
10142{
Dan Fuhrycdd1fed2016-03-17 15:23:38 +010010143 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredi80ace852014-07-23 15:15:32 +020010144 return -EINVAL;
10145
Dan Fuhrycdd1fed2016-03-17 15:23:38 +010010146 if (flags & RENAME_EXCHANGE)
10147 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
10148 new_dentry);
10149
10150 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
Miklos Szeredi80ace852014-07-23 15:15:32 +020010151}
10152
Miao Xie8ccf6f192012-10-25 09:28:04 +000010153static void btrfs_run_delalloc_work(struct btrfs_work *work)
10154{
10155 struct btrfs_delalloc_work *delalloc_work;
Josef Bacik9f23e282013-10-28 15:03:41 -040010156 struct inode *inode;
Miao Xie8ccf6f192012-10-25 09:28:04 +000010157
10158 delalloc_work = container_of(work, struct btrfs_delalloc_work,
10159 work);
Josef Bacik9f23e282013-10-28 15:03:41 -040010160 inode = delalloc_work->inode;
David Sterba30424602015-11-27 19:27:11 +010010161 filemap_flush(inode->i_mapping);
10162 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
10163 &BTRFS_I(inode)->runtime_flags))
Josef Bacik9f23e282013-10-28 15:03:41 -040010164 filemap_flush(inode->i_mapping);
Miao Xie8ccf6f192012-10-25 09:28:04 +000010165
10166 if (delalloc_work->delay_iput)
Josef Bacik9f23e282013-10-28 15:03:41 -040010167 btrfs_add_delayed_iput(inode);
Miao Xie8ccf6f192012-10-25 09:28:04 +000010168 else
Josef Bacik9f23e282013-10-28 15:03:41 -040010169 iput(inode);
Miao Xie8ccf6f192012-10-25 09:28:04 +000010170 complete(&delalloc_work->completion);
10171}
10172
10173struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
David Sterba651d4942015-11-27 19:24:16 +010010174 int delay_iput)
Miao Xie8ccf6f192012-10-25 09:28:04 +000010175{
10176 struct btrfs_delalloc_work *work;
10177
David Sterba100d5702015-12-08 14:39:32 +010010178 work = kmalloc(sizeof(*work), GFP_NOFS);
Miao Xie8ccf6f192012-10-25 09:28:04 +000010179 if (!work)
10180 return NULL;
10181
10182 init_completion(&work->completion);
10183 INIT_LIST_HEAD(&work->list);
10184 work->inode = inode;
Miao Xie8ccf6f192012-10-25 09:28:04 +000010185 work->delay_iput = delay_iput;
Liu Bo9e0af232014-08-15 23:36:53 +080010186 WARN_ON_ONCE(!inode);
10187 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
10188 btrfs_run_delalloc_work, NULL, NULL);
Miao Xie8ccf6f192012-10-25 09:28:04 +000010189
10190 return work;
10191}
10192
10193void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
10194{
10195 wait_for_completion(&work->completion);
David Sterba100d5702015-12-08 14:39:32 +010010196 kfree(work);
Miao Xie8ccf6f192012-10-25 09:28:04 +000010197}
10198
Chris Masond352ac62008-09-29 15:18:18 -040010199/*
10200 * some fairly slow code that needs optimization. This walks the list
10201 * of all the inodes with pending delalloc and forces them to disk.
10202 */
Miao Xie6c255e62014-03-06 13:55:01 +080010203static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
10204 int nr)
Chris Masonea8c2812008-08-04 23:17:27 -040010205{
Chris Masonea8c2812008-08-04 23:17:27 -040010206 struct btrfs_inode *binode;
Zheng Yan5b21f2e2008-09-26 10:05:38 -040010207 struct inode *inode;
Miao Xie8ccf6f192012-10-25 09:28:04 +000010208 struct btrfs_delalloc_work *work, *next;
10209 struct list_head works;
Miao Xie1eafa6c2013-01-22 10:49:00 +000010210 struct list_head splice;
Miao Xie8ccf6f192012-10-25 09:28:04 +000010211 int ret = 0;
Chris Masonea8c2812008-08-04 23:17:27 -040010212
Miao Xie8ccf6f192012-10-25 09:28:04 +000010213 INIT_LIST_HEAD(&works);
Miao Xie1eafa6c2013-01-22 10:49:00 +000010214 INIT_LIST_HEAD(&splice);
Miao Xie63607cc2013-01-22 10:50:35 +000010215
Miao Xie573bfb72014-03-06 13:55:03 +080010216 mutex_lock(&root->delalloc_mutex);
Miao Xieeb73c1b2013-05-15 07:48:22 +000010217 spin_lock(&root->delalloc_lock);
10218 list_splice_init(&root->delalloc_inodes, &splice);
Miao Xie1eafa6c2013-01-22 10:49:00 +000010219 while (!list_empty(&splice)) {
10220 binode = list_entry(splice.next, struct btrfs_inode,
Chris Masonea8c2812008-08-04 23:17:27 -040010221 delalloc_inodes);
Miao Xie1eafa6c2013-01-22 10:49:00 +000010222
Miao Xieeb73c1b2013-05-15 07:48:22 +000010223 list_move_tail(&binode->delalloc_inodes,
10224 &root->delalloc_inodes);
Zheng Yan5b21f2e2008-09-26 10:05:38 -040010225 inode = igrab(&binode->vfs_inode);
Miao Xiedf0af1a2013-01-29 10:11:59 +000010226 if (!inode) {
Miao Xieeb73c1b2013-05-15 07:48:22 +000010227 cond_resched_lock(&root->delalloc_lock);
Miao Xie1eafa6c2013-01-22 10:49:00 +000010228 continue;
Miao Xiedf0af1a2013-01-29 10:11:59 +000010229 }
Miao Xieeb73c1b2013-05-15 07:48:22 +000010230 spin_unlock(&root->delalloc_lock);
Miao Xie1eafa6c2013-01-22 10:49:00 +000010231
David Sterba651d4942015-11-27 19:24:16 +010010232 work = btrfs_alloc_delalloc_work(inode, delay_iput);
David Sterba5d99a9982014-09-29 19:20:37 +020010233 if (!work) {
Josef Bacikf4ab9ea2013-09-17 11:25:44 -040010234 if (delay_iput)
10235 btrfs_add_delayed_iput(inode);
10236 else
10237 iput(inode);
Miao Xie1eafa6c2013-01-22 10:49:00 +000010238 ret = -ENOMEM;
Wang Shilonga1ecaab2014-04-02 19:53:32 +080010239 goto out;
Zheng Yan5b21f2e2008-09-26 10:05:38 -040010240 }
Miao Xie1eafa6c2013-01-22 10:49:00 +000010241 list_add_tail(&work->list, &works);
Qu Wenruoa44903a2014-02-28 10:46:09 +080010242 btrfs_queue_work(root->fs_info->flush_workers,
10243 &work->work);
Miao Xie6c255e62014-03-06 13:55:01 +080010244 ret++;
10245 if (nr != -1 && ret >= nr)
Wang Shilonga1ecaab2014-04-02 19:53:32 +080010246 goto out;
Zheng Yan5b21f2e2008-09-26 10:05:38 -040010247 cond_resched();
Miao Xieeb73c1b2013-05-15 07:48:22 +000010248 spin_lock(&root->delalloc_lock);
Chris Masonea8c2812008-08-04 23:17:27 -040010249 }
Miao Xieeb73c1b2013-05-15 07:48:22 +000010250 spin_unlock(&root->delalloc_lock);
Chris Mason8c8bee12008-09-29 11:19:10 -040010251
Wang Shilonga1ecaab2014-04-02 19:53:32 +080010252out:
Miao Xie1eafa6c2013-01-22 10:49:00 +000010253 list_for_each_entry_safe(work, next, &works, list) {
10254 list_del_init(&work->list);
10255 btrfs_wait_and_free_delalloc_work(work);
10256 }
10257
Miao Xieeb73c1b2013-05-15 07:48:22 +000010258 if (!list_empty_careful(&splice)) {
10259 spin_lock(&root->delalloc_lock);
10260 list_splice_tail(&splice, &root->delalloc_inodes);
10261 spin_unlock(&root->delalloc_lock);
10262 }
Miao Xie573bfb72014-03-06 13:55:03 +080010263 mutex_unlock(&root->delalloc_mutex);
Miao Xieeb73c1b2013-05-15 07:48:22 +000010264 return ret;
10265}
10266
10267int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
10268{
Jeff Mahoney0b246af2016-06-22 18:54:23 -040010269 struct btrfs_fs_info *fs_info = root->fs_info;
Miao Xieeb73c1b2013-05-15 07:48:22 +000010270 int ret;
10271
Jeff Mahoney0b246af2016-06-22 18:54:23 -040010272 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
Miao Xieeb73c1b2013-05-15 07:48:22 +000010273 return -EROFS;
10274
Miao Xie6c255e62014-03-06 13:55:01 +080010275 ret = __start_delalloc_inodes(root, delay_iput, -1);
10276 if (ret > 0)
10277 ret = 0;
Miao Xieeb73c1b2013-05-15 07:48:22 +000010278 return ret;
10279}
10280
Miao Xie6c255e62014-03-06 13:55:01 +080010281int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
10282 int nr)
Miao Xieeb73c1b2013-05-15 07:48:22 +000010283{
10284 struct btrfs_root *root;
10285 struct list_head splice;
10286 int ret;
10287
Wang Shilong2c21b4d2014-01-14 19:42:20 +080010288 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
Miao Xieeb73c1b2013-05-15 07:48:22 +000010289 return -EROFS;
10290
10291 INIT_LIST_HEAD(&splice);
10292
Miao Xie573bfb72014-03-06 13:55:03 +080010293 mutex_lock(&fs_info->delalloc_root_mutex);
Miao Xieeb73c1b2013-05-15 07:48:22 +000010294 spin_lock(&fs_info->delalloc_root_lock);
10295 list_splice_init(&fs_info->delalloc_roots, &splice);
Miao Xie6c255e62014-03-06 13:55:01 +080010296 while (!list_empty(&splice) && nr) {
Miao Xieeb73c1b2013-05-15 07:48:22 +000010297 root = list_first_entry(&splice, struct btrfs_root,
10298 delalloc_root);
10299 root = btrfs_grab_fs_root(root);
10300 BUG_ON(!root);
10301 list_move_tail(&root->delalloc_root,
10302 &fs_info->delalloc_roots);
10303 spin_unlock(&fs_info->delalloc_root_lock);
10304
Miao Xie6c255e62014-03-06 13:55:01 +080010305 ret = __start_delalloc_inodes(root, delay_iput, nr);
Miao Xieeb73c1b2013-05-15 07:48:22 +000010306 btrfs_put_fs_root(root);
Miao Xie6c255e62014-03-06 13:55:01 +080010307 if (ret < 0)
Miao Xieeb73c1b2013-05-15 07:48:22 +000010308 goto out;
10309
Miao Xie6c255e62014-03-06 13:55:01 +080010310 if (nr != -1) {
10311 nr -= ret;
10312 WARN_ON(nr < 0);
10313 }
Miao Xieeb73c1b2013-05-15 07:48:22 +000010314 spin_lock(&fs_info->delalloc_root_lock);
10315 }
10316 spin_unlock(&fs_info->delalloc_root_lock);
10317
Miao Xie6c255e62014-03-06 13:55:01 +080010318 ret = 0;
Miao Xie8ccf6f192012-10-25 09:28:04 +000010319out:
Miao Xie1eafa6c2013-01-22 10:49:00 +000010320 if (!list_empty_careful(&splice)) {
Miao Xieeb73c1b2013-05-15 07:48:22 +000010321 spin_lock(&fs_info->delalloc_root_lock);
10322 list_splice_tail(&splice, &fs_info->delalloc_roots);
10323 spin_unlock(&fs_info->delalloc_root_lock);
Miao Xie1eafa6c2013-01-22 10:49:00 +000010324 }
Miao Xie573bfb72014-03-06 13:55:03 +080010325 mutex_unlock(&fs_info->delalloc_root_mutex);
Miao Xie8ccf6f192012-10-25 09:28:04 +000010326 return ret;
Chris Masonea8c2812008-08-04 23:17:27 -040010327}
10328
Chris Mason39279cc2007-06-12 06:35:45 -040010329static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10330 const char *symname)
10331{
Jeff Mahoney0b246af2016-06-22 18:54:23 -040010332 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
Chris Mason39279cc2007-06-12 06:35:45 -040010333 struct btrfs_trans_handle *trans;
10334 struct btrfs_root *root = BTRFS_I(dir)->root;
10335 struct btrfs_path *path;
10336 struct btrfs_key key;
Chris Mason1832a6d2007-12-21 16:27:21 -050010337 struct inode *inode = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -040010338 int err;
10339 int drop_inode = 0;
10340 u64 objectid;
Dulshani Gunawardhana67871252013-10-31 10:33:04 +053010341 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -040010342 int name_len;
10343 int datasize;
Chris Mason5f39d392007-10-15 16:14:19 -040010344 unsigned long ptr;
Chris Mason39279cc2007-06-12 06:35:45 -040010345 struct btrfs_file_extent_item *ei;
Chris Mason5f39d392007-10-15 16:14:19 -040010346 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -040010347
Filipe David Borba Mananaf06becc2013-09-16 09:53:28 +010010348 name_len = strlen(symname);
Jeff Mahoney0b246af2016-06-22 18:54:23 -040010349 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
Chris Mason39279cc2007-06-12 06:35:45 -040010350 return -ENAMETOOLONG;
Chris Mason1832a6d2007-12-21 16:27:21 -050010351
Josef Bacik9ed74f22009-09-11 16:12:44 -040010352 /*
10353 * 2 items for inode item and ref
10354 * 2 items for dir items
Filipe Manana9269d122015-12-31 18:16:29 +000010355 * 1 item for updating parent inode item
10356 * 1 item for the inline extent item
Josef Bacik9ed74f22009-09-11 16:12:44 -040010357 * 1 item for xattr if selinux is on
10358 */
Filipe Manana9269d122015-12-31 18:16:29 +000010359 trans = btrfs_start_transaction(root, 7);
Yan, Zhenga22285a2010-05-16 10:48:46 -040010360 if (IS_ERR(trans))
10361 return PTR_ERR(trans);
Chris Mason1832a6d2007-12-21 16:27:21 -050010362
Li Zefan581bb052011-04-20 10:06:11 +080010363 err = btrfs_find_free_ino(root, &objectid);
10364 if (err)
10365 goto out_unlock;
10366
Josef Bacikaec74772008-07-24 12:12:38 -040010367 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
David Sterbaf85b7372017-01-20 14:54:07 +010010368 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10369 objectid, S_IFLNK|S_IRWXUGO, &index);
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -040010370 if (IS_ERR(inode)) {
10371 err = PTR_ERR(inode);
Chris Mason39279cc2007-06-12 06:35:45 -040010372 goto out_unlock;
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -040010373 }
Chris Mason39279cc2007-06-12 06:35:45 -040010374
Casey Schauflerad19db72011-12-15 10:09:07 -050010375 /*
10376 * If the active LSM wants to access the inode during
10377 * d_instantiate it needs these. Smack checks to see
10378 * if the filesystem supports xattrs by looking at the
10379 * ops vector.
10380 */
10381 inode->i_fop = &btrfs_file_operations;
10382 inode->i_op = &btrfs_file_inode_operations;
Chris Masonb0d5d102014-09-08 13:08:51 -070010383 inode->i_mapping->a_ops = &btrfs_aops;
Chris Masonb0d5d102014-09-08 13:08:51 -070010384 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10385
10386 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10387 if (err)
10388 goto out_unlock_inode;
Casey Schauflerad19db72011-12-15 10:09:07 -050010389
Chris Mason39279cc2007-06-12 06:35:45 -040010390 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -070010391 if (!path) {
10392 err = -ENOMEM;
Chris Masonb0d5d102014-09-08 13:08:51 -070010393 goto out_unlock_inode;
Mark Fashehd8926bb2011-07-13 10:38:47 -070010394 }
Nikolay Borisov4a0cc7c2017-01-10 20:35:31 +020010395 key.objectid = btrfs_ino(BTRFS_I(inode));
Chris Mason39279cc2007-06-12 06:35:45 -040010396 key.offset = 0;
David Sterba962a2982014-06-04 18:41:45 +020010397 key.type = BTRFS_EXTENT_DATA_KEY;
Chris Mason39279cc2007-06-12 06:35:45 -040010398 datasize = btrfs_file_extent_calc_inline_size(name_len);
10399 err = btrfs_insert_empty_item(trans, root, path, &key,
10400 datasize);
Chris Mason54aa1f42007-06-22 14:16:25 -040010401 if (err) {
Julia Lawallb0839162011-05-14 07:10:51 +000010402 btrfs_free_path(path);
Chris Masonb0d5d102014-09-08 13:08:51 -070010403 goto out_unlock_inode;
Chris Mason54aa1f42007-06-22 14:16:25 -040010404 }
Chris Mason5f39d392007-10-15 16:14:19 -040010405 leaf = path->nodes[0];
10406 ei = btrfs_item_ptr(leaf, path->slots[0],
10407 struct btrfs_file_extent_item);
10408 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10409 btrfs_set_file_extent_type(leaf, ei,
Chris Mason39279cc2007-06-12 06:35:45 -040010410 BTRFS_FILE_EXTENT_INLINE);
Chris Masonc8b97812008-10-29 14:49:59 -040010411 btrfs_set_file_extent_encryption(leaf, ei, 0);
10412 btrfs_set_file_extent_compression(leaf, ei, 0);
10413 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10414 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10415
Chris Mason39279cc2007-06-12 06:35:45 -040010416 ptr = btrfs_file_extent_inline_start(ei);
Chris Mason5f39d392007-10-15 16:14:19 -040010417 write_extent_buffer(leaf, symname, ptr, name_len);
10418 btrfs_mark_buffer_dirty(leaf);
Chris Mason39279cc2007-06-12 06:35:45 -040010419 btrfs_free_path(path);
Chris Mason5f39d392007-10-15 16:14:19 -040010420
Chris Mason39279cc2007-06-12 06:35:45 -040010421 inode->i_op = &btrfs_symlink_inode_operations;
Al Viro21fc61c2015-11-17 01:07:57 -050010422 inode_nohighmem(inode);
Chris Mason39279cc2007-06-12 06:35:45 -040010423 inode->i_mapping->a_ops = &btrfs_symlink_aops;
Yan Zhengd899e052008-10-30 14:25:28 -040010424 inode_set_bytes(inode, name_len);
Nikolay Borisov6ef06d22017-02-20 13:50:34 +020010425 btrfs_i_size_write(BTRFS_I(inode), name_len);
Chris Mason54aa1f42007-06-22 14:16:25 -040010426 err = btrfs_update_inode(trans, root, inode);
Filipe Mananad50866d2015-12-31 18:08:24 +000010427 /*
10428 * Last step, add directory indexes for our symlink inode. This is the
10429 * last step to avoid extra cleanup of these indexes if an error happens
10430 * elsewhere above.
10431 */
10432 if (!err)
Nikolay Borisovcef415a2017-02-20 13:51:09 +020010433 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10434 BTRFS_I(inode), 0, index);
Chris Masonb0d5d102014-09-08 13:08:51 -070010435 if (err) {
Chris Mason54aa1f42007-06-22 14:16:25 -040010436 drop_inode = 1;
Chris Masonb0d5d102014-09-08 13:08:51 -070010437 goto out_unlock_inode;
10438 }
10439
10440 unlock_new_inode(inode);
10441 d_instantiate(dentry, inode);
Chris Mason39279cc2007-06-12 06:35:45 -040010442
10443out_unlock:
Jeff Mahoney3a45bb22016-09-09 21:39:03 -040010444 btrfs_end_transaction(trans);
Chris Mason39279cc2007-06-12 06:35:45 -040010445 if (drop_inode) {
10446 inode_dec_link_count(inode);
10447 iput(inode);
10448 }
Jeff Mahoney2ff7e612016-06-22 18:54:24 -040010449 btrfs_btree_balance_dirty(fs_info);
Chris Mason39279cc2007-06-12 06:35:45 -040010450 return err;
Chris Masonb0d5d102014-09-08 13:08:51 -070010451
10452out_unlock_inode:
10453 drop_inode = 1;
10454 unlock_new_inode(inode);
10455 goto out_unlock;
Chris Mason39279cc2007-06-12 06:35:45 -040010456}
Chris Mason16432982008-04-10 10:23:21 -040010457
Josef Bacik0af3d002010-06-21 14:48:16 -040010458static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10459 u64 start, u64 num_bytes, u64 min_size,
10460 loff_t actual_len, u64 *alloc_hint,
10461 struct btrfs_trans_handle *trans)
Yan Zhengd899e052008-10-30 14:25:28 -040010462{
Jeff Mahoney0b246af2016-06-22 18:54:23 -040010463 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Josef Bacik5dc562c2012-08-17 13:14:17 -040010464 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10465 struct extent_map *em;
Yan Zhengd899e052008-10-30 14:25:28 -040010466 struct btrfs_root *root = BTRFS_I(inode)->root;
10467 struct btrfs_key ins;
Yan Zhengd899e052008-10-30 14:25:28 -040010468 u64 cur_offset = start;
Josef Bacik55a61d12010-11-22 18:50:32 +000010469 u64 i_size;
Chris Mason154ea282013-03-05 11:11:26 -050010470 u64 cur_bytes;
Josef Bacik0b670dc2015-09-23 17:11:16 -040010471 u64 last_alloc = (u64)-1;
Yan Zhengd899e052008-10-30 14:25:28 -040010472 int ret = 0;
Josef Bacik0af3d002010-06-21 14:48:16 -040010473 bool own_trans = true;
Wang Xiaoguang18513092016-07-25 15:51:40 +080010474 u64 end = start + num_bytes - 1;
Yan Zhengd899e052008-10-30 14:25:28 -040010475
Josef Bacik0af3d002010-06-21 14:48:16 -040010476 if (trans)
10477 own_trans = false;
Yan Zhengd899e052008-10-30 14:25:28 -040010478 while (num_bytes > 0) {
Josef Bacik0af3d002010-06-21 14:48:16 -040010479 if (own_trans) {
10480 trans = btrfs_start_transaction(root, 3);
10481 if (IS_ERR(trans)) {
10482 ret = PTR_ERR(trans);
10483 break;
10484 }
Yan Zhengd899e052008-10-30 14:25:28 -040010485 }
Yan, Zheng5a303d52009-11-12 09:34:52 +000010486
Byongho Leeee221842015-12-15 01:42:10 +090010487 cur_bytes = min_t(u64, num_bytes, SZ_256M);
Chris Mason154ea282013-03-05 11:11:26 -050010488 cur_bytes = max(cur_bytes, min_size);
Josef Bacik0b670dc2015-09-23 17:11:16 -040010489 /*
10490 * If we are severely fragmented we could end up with really
10491 * small allocations, so if the allocator is returning small
10492 * chunks lets make its job easier by only searching for those
10493 * sized chunks.
10494 */
10495 cur_bytes = min(cur_bytes, last_alloc);
Wang Xiaoguang18513092016-07-25 15:51:40 +080010496 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10497 min_size, 0, *alloc_hint, &ins, 1, 0);
Yan, Zheng5a303d52009-11-12 09:34:52 +000010498 if (ret) {
Josef Bacik0af3d002010-06-21 14:48:16 -040010499 if (own_trans)
Jeff Mahoney3a45bb22016-09-09 21:39:03 -040010500 btrfs_end_transaction(trans);
Yan, Zhenga22285a2010-05-16 10:48:46 -040010501 break;
Yan, Zheng5a303d52009-11-12 09:34:52 +000010502 }
Jeff Mahoney0b246af2016-06-22 18:54:23 -040010503 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
Yan, Zheng5a303d52009-11-12 09:34:52 +000010504
Josef Bacik0b670dc2015-09-23 17:11:16 -040010505 last_alloc = ins.offset;
Yan Zhengd899e052008-10-30 14:25:28 -040010506 ret = insert_reserved_file_extent(trans, inode,
10507 cur_offset, ins.objectid,
10508 ins.offset, ins.offset,
Yan, Zheng920bbbf2009-11-12 09:34:08 +000010509 ins.offset, 0, 0, 0,
Yan Zhengd899e052008-10-30 14:25:28 -040010510 BTRFS_FILE_EXTENT_PREALLOC);
Jeff Mahoney79787ea2012-03-12 16:03:00 +010010511 if (ret) {
Jeff Mahoney2ff7e612016-06-22 18:54:24 -040010512 btrfs_free_reserved_extent(fs_info, ins.objectid,
Miao Xiee570fd22014-06-19 10:42:50 +080010513 ins.offset, 0);
Jeff Mahoney66642832016-06-10 18:19:25 -040010514 btrfs_abort_transaction(trans, ret);
Jeff Mahoney79787ea2012-03-12 16:03:00 +010010515 if (own_trans)
Jeff Mahoney3a45bb22016-09-09 21:39:03 -040010516 btrfs_end_transaction(trans);
Jeff Mahoney79787ea2012-03-12 16:03:00 +010010517 break;
10518 }
Dongsheng Yang31193212014-12-12 16:44:35 +080010519
Nikolay Borisovdcdbc052017-02-20 13:50:45 +020010520 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
Chris Masona1ed8352009-09-11 12:27:37 -040010521 cur_offset + ins.offset -1, 0);
Yan, Zheng5a303d52009-11-12 09:34:52 +000010522
Josef Bacik5dc562c2012-08-17 13:14:17 -040010523 em = alloc_extent_map();
10524 if (!em) {
10525 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10526 &BTRFS_I(inode)->runtime_flags);
10527 goto next;
10528 }
10529
10530 em->start = cur_offset;
10531 em->orig_start = cur_offset;
10532 em->len = ins.offset;
10533 em->block_start = ins.objectid;
10534 em->block_len = ins.offset;
Josef Bacikb4939682012-12-03 10:31:19 -050010535 em->orig_block_len = ins.offset;
Josef Bacikcc95bef2013-04-04 14:31:27 -040010536 em->ram_bytes = ins.offset;
Jeff Mahoney0b246af2016-06-22 18:54:23 -040010537 em->bdev = fs_info->fs_devices->latest_bdev;
Josef Bacik5dc562c2012-08-17 13:14:17 -040010538 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10539 em->generation = trans->transid;
10540
10541 while (1) {
10542 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -040010543 ret = add_extent_mapping(em_tree, em, 1);
Josef Bacik5dc562c2012-08-17 13:14:17 -040010544 write_unlock(&em_tree->lock);
10545 if (ret != -EEXIST)
10546 break;
Nikolay Borisovdcdbc052017-02-20 13:50:45 +020010547 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
Josef Bacik5dc562c2012-08-17 13:14:17 -040010548 cur_offset + ins.offset - 1,
10549 0);
10550 }
10551 free_extent_map(em);
10552next:
Yan Zhengd899e052008-10-30 14:25:28 -040010553 num_bytes -= ins.offset;
10554 cur_offset += ins.offset;
Yan, Zhengefa56462010-05-16 10:49:59 -040010555 *alloc_hint = ins.objectid + ins.offset;
Yan, Zheng5a303d52009-11-12 09:34:52 +000010556
Josef Bacik0c4d2d92012-04-05 15:03:02 -040010557 inode_inc_iversion(inode);
Deepa Dinamanic2050a42016-09-14 07:48:06 -070010558 inode->i_ctime = current_time(inode);
Christoph Hellwig6cbff002009-04-17 10:37:41 +020010559 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
Yan Zhengd899e052008-10-30 14:25:28 -040010560 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
Yan, Zhengefa56462010-05-16 10:49:59 -040010561 (actual_len > inode->i_size) &&
10562 (cur_offset > inode->i_size)) {
Aneesh Kumar K.Vd1ea6a62010-01-20 07:28:54 +000010563 if (cur_offset > actual_len)
Josef Bacik55a61d12010-11-22 18:50:32 +000010564 i_size = actual_len;
Aneesh Kumar K.Vd1ea6a62010-01-20 07:28:54 +000010565 else
Josef Bacik55a61d12010-11-22 18:50:32 +000010566 i_size = cur_offset;
10567 i_size_write(inode, i_size);
10568 btrfs_ordered_update_i_size(inode, i_size, NULL);
Yan, Zheng5a303d52009-11-12 09:34:52 +000010569 }
10570
Yan Zhengd899e052008-10-30 14:25:28 -040010571 ret = btrfs_update_inode(trans, root, inode);
Jeff Mahoney79787ea2012-03-12 16:03:00 +010010572
10573 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -040010574 btrfs_abort_transaction(trans, ret);
Jeff Mahoney79787ea2012-03-12 16:03:00 +010010575 if (own_trans)
Jeff Mahoney3a45bb22016-09-09 21:39:03 -040010576 btrfs_end_transaction(trans);
Jeff Mahoney79787ea2012-03-12 16:03:00 +010010577 break;
10578 }
Yan Zhengd899e052008-10-30 14:25:28 -040010579
Josef Bacik0af3d002010-06-21 14:48:16 -040010580 if (own_trans)
Jeff Mahoney3a45bb22016-09-09 21:39:03 -040010581 btrfs_end_transaction(trans);
Yan, Zheng5a303d52009-11-12 09:34:52 +000010582 }
Wang Xiaoguang18513092016-07-25 15:51:40 +080010583 if (cur_offset < end)
Qu Wenruobc42bda2017-02-27 15:10:39 +080010584 btrfs_free_reserved_data_space(inode, NULL, cur_offset,
Wang Xiaoguang18513092016-07-25 15:51:40 +080010585 end - cur_offset + 1);
Yan Zhengd899e052008-10-30 14:25:28 -040010586 return ret;
10587}
10588
Josef Bacik0af3d002010-06-21 14:48:16 -040010589int btrfs_prealloc_file_range(struct inode *inode, int mode,
10590 u64 start, u64 num_bytes, u64 min_size,
10591 loff_t actual_len, u64 *alloc_hint)
10592{
10593 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10594 min_size, actual_len, alloc_hint,
10595 NULL);
10596}
10597
10598int btrfs_prealloc_file_range_trans(struct inode *inode,
10599 struct btrfs_trans_handle *trans, int mode,
10600 u64 start, u64 num_bytes, u64 min_size,
10601 loff_t actual_len, u64 *alloc_hint)
10602{
10603 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10604 min_size, actual_len, alloc_hint, trans);
10605}
10606
Chris Masone6dcd2d2008-07-17 12:53:50 -040010607static int btrfs_set_page_dirty(struct page *page)
10608{
Chris Masone6dcd2d2008-07-17 12:53:50 -040010609 return __set_page_dirty_nobuffers(page);
10610}
10611
Al Viro10556cb2011-06-20 19:28:19 -040010612static int btrfs_permission(struct inode *inode, int mask)
Yanfdebe2b2008-01-14 13:26:08 -050010613{
Li Zefanb83cc962010-12-20 16:04:08 +080010614 struct btrfs_root *root = BTRFS_I(inode)->root;
Jeff Mahoneycb6db4e2011-08-15 17:27:21 +000010615 umode_t mode = inode->i_mode;
Li Zefanb83cc962010-12-20 16:04:08 +080010616
Jeff Mahoneycb6db4e2011-08-15 17:27:21 +000010617 if (mask & MAY_WRITE &&
10618 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10619 if (btrfs_root_readonly(root))
10620 return -EROFS;
10621 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10622 return -EACCES;
10623 }
Al Viro2830ba72011-06-20 19:16:29 -040010624 return generic_permission(inode, mask);
Yanfdebe2b2008-01-14 13:26:08 -050010625}
Chris Mason39279cc2007-06-12 06:35:45 -040010626
Filipe Mananaef3b9af2014-04-27 20:40:45 +010010627static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10628{
Jeff Mahoney2ff7e612016-06-22 18:54:24 -040010629 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
Filipe Mananaef3b9af2014-04-27 20:40:45 +010010630 struct btrfs_trans_handle *trans;
10631 struct btrfs_root *root = BTRFS_I(dir)->root;
10632 struct inode *inode = NULL;
10633 u64 objectid;
10634 u64 index;
10635 int ret = 0;
10636
10637 /*
10638 * 5 units required for adding orphan entry
10639 */
10640 trans = btrfs_start_transaction(root, 5);
10641 if (IS_ERR(trans))
10642 return PTR_ERR(trans);
10643
10644 ret = btrfs_find_free_ino(root, &objectid);
10645 if (ret)
10646 goto out;
10647
10648 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
David Sterbaf85b7372017-01-20 14:54:07 +010010649 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
Filipe Mananaef3b9af2014-04-27 20:40:45 +010010650 if (IS_ERR(inode)) {
10651 ret = PTR_ERR(inode);
10652 inode = NULL;
10653 goto out;
10654 }
10655
Filipe Mananaef3b9af2014-04-27 20:40:45 +010010656 inode->i_fop = &btrfs_file_operations;
10657 inode->i_op = &btrfs_file_inode_operations;
10658
10659 inode->i_mapping->a_ops = &btrfs_aops;
Filipe Mananaef3b9af2014-04-27 20:40:45 +010010660 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10661
Chris Masonb0d5d102014-09-08 13:08:51 -070010662 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10663 if (ret)
10664 goto out_inode;
10665
10666 ret = btrfs_update_inode(trans, root, inode);
10667 if (ret)
10668 goto out_inode;
Nikolay Borisov73f2e542017-02-20 13:50:59 +020010669 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
Filipe Mananaef3b9af2014-04-27 20:40:45 +010010670 if (ret)
Chris Masonb0d5d102014-09-08 13:08:51 -070010671 goto out_inode;
Filipe Mananaef3b9af2014-04-27 20:40:45 +010010672
Filipe Manana5762b5c2014-08-01 00:10:32 +010010673 /*
10674 * We set number of links to 0 in btrfs_new_inode(), and here we set
10675 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10676 * through:
10677 *
10678 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10679 */
10680 set_nlink(inode, 1);
Chris Masonb0d5d102014-09-08 13:08:51 -070010681 unlock_new_inode(inode);
Filipe Mananaef3b9af2014-04-27 20:40:45 +010010682 d_tmpfile(dentry, inode);
10683 mark_inode_dirty(inode);
10684
10685out:
Jeff Mahoney3a45bb22016-09-09 21:39:03 -040010686 btrfs_end_transaction(trans);
Filipe Mananaef3b9af2014-04-27 20:40:45 +010010687 if (ret)
10688 iput(inode);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -040010689 btrfs_balance_delayed_items(fs_info);
10690 btrfs_btree_balance_dirty(fs_info);
Filipe Mananaef3b9af2014-04-27 20:40:45 +010010691 return ret;
Chris Masonb0d5d102014-09-08 13:08:51 -070010692
10693out_inode:
10694 unlock_new_inode(inode);
10695 goto out;
10696
Filipe Mananaef3b9af2014-04-27 20:40:45 +010010697}
10698
David Sterba20a7db82017-02-17 16:24:29 +010010699__attribute__((const))
Liu Bo9d0d1c82017-03-24 15:04:50 -070010700static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror)
David Sterba20a7db82017-02-17 16:24:29 +010010701{
Liu Bo9d0d1c82017-03-24 15:04:50 -070010702 return -EAGAIN;
David Sterba20a7db82017-02-17 16:24:29 +010010703}
10704
Josef Bacikc6100a42017-05-05 11:57:13 -040010705static struct btrfs_fs_info *iotree_fs_info(void *private_data)
10706{
10707 struct inode *inode = private_data;
10708 return btrfs_sb(inode->i_sb);
10709}
10710
10711static void btrfs_check_extent_io_range(void *private_data, const char *caller,
10712 u64 start, u64 end)
10713{
10714 struct inode *inode = private_data;
10715 u64 isize;
10716
10717 isize = i_size_read(inode);
10718 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
10719 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
10720 "%s: ino %llu isize %llu odd range [%llu,%llu]",
10721 caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
10722 }
10723}
10724
10725void btrfs_set_range_writeback(void *private_data, u64 start, u64 end)
10726{
10727 struct inode *inode = private_data;
10728 unsigned long index = start >> PAGE_SHIFT;
10729 unsigned long end_index = end >> PAGE_SHIFT;
10730 struct page *page;
10731
10732 while (index <= end_index) {
10733 page = find_get_page(inode->i_mapping, index);
10734 ASSERT(page); /* Pages should be in the extent_io_tree */
10735 set_page_writeback(page);
10736 put_page(page);
10737 index++;
10738 }
10739}
10740
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -070010741static const struct inode_operations btrfs_dir_inode_operations = {
Chris Mason3394e162008-11-17 20:42:26 -050010742 .getattr = btrfs_getattr,
Chris Mason39279cc2007-06-12 06:35:45 -040010743 .lookup = btrfs_lookup,
10744 .create = btrfs_create,
10745 .unlink = btrfs_unlink,
10746 .link = btrfs_link,
10747 .mkdir = btrfs_mkdir,
10748 .rmdir = btrfs_rmdir,
Miklos Szeredi2773bf02016-09-27 11:03:58 +020010749 .rename = btrfs_rename2,
Chris Mason39279cc2007-06-12 06:35:45 -040010750 .symlink = btrfs_symlink,
10751 .setattr = btrfs_setattr,
Josef Bacik618e21d2007-07-11 10:18:17 -040010752 .mknod = btrfs_mknod,
Josef Bacik5103e942007-11-16 11:45:54 -050010753 .listxattr = btrfs_listxattr,
Yanfdebe2b2008-01-14 13:26:08 -050010754 .permission = btrfs_permission,
Christoph Hellwig4e34e712011-07-23 17:37:31 +020010755 .get_acl = btrfs_get_acl,
Christoph Hellwig996a7102013-12-20 05:16:43 -080010756 .set_acl = btrfs_set_acl,
Guangyu Sun93fd63c2013-09-16 10:42:03 -070010757 .update_time = btrfs_update_time,
Filipe Mananaef3b9af2014-04-27 20:40:45 +010010758 .tmpfile = btrfs_tmpfile,
Chris Mason39279cc2007-06-12 06:35:45 -040010759};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -070010760static const struct inode_operations btrfs_dir_ro_inode_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -040010761 .lookup = btrfs_lookup,
Yanfdebe2b2008-01-14 13:26:08 -050010762 .permission = btrfs_permission,
Guangyu Sun93fd63c2013-09-16 10:42:03 -070010763 .update_time = btrfs_update_time,
Chris Mason39279cc2007-06-12 06:35:45 -040010764};
Yan, Zheng76dda932009-09-21 16:00:26 -040010765
Alexey Dobriyan828c0952009-10-01 15:43:56 -070010766static const struct file_operations btrfs_dir_file_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -040010767 .llseek = generic_file_llseek,
10768 .read = generic_read_dir,
Omar Sandoval02dbfc92016-05-20 13:50:33 -070010769 .iterate_shared = btrfs_real_readdir,
Josef Bacik23b5ec72017-07-24 15:14:25 -040010770 .open = btrfs_opendir,
Christoph Hellwig34287aa2007-09-14 10:22:47 -040010771 .unlocked_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -040010772#ifdef CONFIG_COMPAT
Luke Dashjr4c63c242015-10-29 08:22:21 +000010773 .compat_ioctl = btrfs_compat_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -040010774#endif
Sage Weil6bf13c02008-06-10 10:07:39 -040010775 .release = btrfs_release_file,
Chris Masone02119d2008-09-05 16:13:11 -040010776 .fsync = btrfs_sync_file,
Chris Mason39279cc2007-06-12 06:35:45 -040010777};
10778
David Sterba20e55062015-11-19 11:42:28 +010010779static const struct extent_io_ops btrfs_extent_io_ops = {
David Sterba4d53ddd2017-02-17 15:27:44 +010010780 /* mandatory callbacks */
Chris Mason065631f2008-02-20 12:07:25 -050010781 .submit_bio_hook = btrfs_submit_bio_hook,
Chris Mason07157aa2007-08-30 08:50:51 -040010782 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
David Sterba4d53ddd2017-02-17 15:27:44 +010010783 .merge_bio_hook = btrfs_merge_bio_hook,
Liu Bo9d0d1c82017-03-24 15:04:50 -070010784 .readpage_io_failed_hook = btrfs_readpage_io_failed_hook,
Josef Bacikc6100a42017-05-05 11:57:13 -040010785 .tree_fs_info = iotree_fs_info,
10786 .set_range_writeback = btrfs_set_range_writeback,
David Sterba4d53ddd2017-02-17 15:27:44 +010010787
10788 /* optional callbacks */
10789 .fill_delalloc = run_delalloc_range,
Chris Masone6dcd2d2008-07-17 12:53:50 -040010790 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
Chris Mason247e7432008-07-17 12:53:51 -040010791 .writepage_start_hook = btrfs_writepage_start_hook,
Chris Masonb0c68f82008-01-31 11:05:37 -050010792 .set_bit_hook = btrfs_set_bit_hook,
10793 .clear_bit_hook = btrfs_clear_bit_hook,
Josef Bacik9ed74f22009-09-11 16:12:44 -040010794 .merge_extent_hook = btrfs_merge_extent_hook,
10795 .split_extent_hook = btrfs_split_extent_hook,
Josef Bacikc6100a42017-05-05 11:57:13 -040010796 .check_extent_io_range = btrfs_check_extent_io_range,
Chris Mason07157aa2007-08-30 08:50:51 -040010797};
10798
Chris Mason35054392009-01-21 13:11:13 -050010799/*
10800 * btrfs doesn't support the bmap operation because swapfiles
10801 * use bmap to make a mapping of extents in the file. They assume
10802 * these extents won't change over the life of the file and they
10803 * use the bmap result to do IO directly to the drive.
10804 *
10805 * the btrfs bmap call would return logical addresses that aren't
10806 * suitable for IO and they also will change frequently as COW
10807 * operations happen. So, swapfile + btrfs == corruption.
10808 *
10809 * For now we're avoiding this by dropping bmap.
10810 */
Alexey Dobriyan7f094102009-09-21 17:01:10 -070010811static const struct address_space_operations btrfs_aops = {
Chris Mason39279cc2007-06-12 06:35:45 -040010812 .readpage = btrfs_readpage,
10813 .writepage = btrfs_writepage,
Chris Masonb293f022007-11-01 19:45:34 -040010814 .writepages = btrfs_writepages,
Chris Mason3ab2fb52007-11-08 10:59:22 -050010815 .readpages = btrfs_readpages,
Chris Mason16432982008-04-10 10:23:21 -040010816 .direct_IO = btrfs_direct_IO,
Chris Masona52d9a82007-08-27 16:49:44 -040010817 .invalidatepage = btrfs_invalidatepage,
10818 .releasepage = btrfs_releasepage,
Chris Masone6dcd2d2008-07-17 12:53:50 -040010819 .set_page_dirty = btrfs_set_page_dirty,
Andi Kleen465fdd92009-09-16 11:50:18 +020010820 .error_remove_page = generic_error_remove_page,
Chris Mason39279cc2007-06-12 06:35:45 -040010821};
10822
Alexey Dobriyan7f094102009-09-21 17:01:10 -070010823static const struct address_space_operations btrfs_symlink_aops = {
Chris Mason39279cc2007-06-12 06:35:45 -040010824 .readpage = btrfs_readpage,
10825 .writepage = btrfs_writepage,
Chris Mason2bf5a722007-08-30 11:54:02 -040010826 .invalidatepage = btrfs_invalidatepage,
10827 .releasepage = btrfs_releasepage,
Chris Mason39279cc2007-06-12 06:35:45 -040010828};
10829
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -070010830static const struct inode_operations btrfs_file_inode_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -040010831 .getattr = btrfs_getattr,
10832 .setattr = btrfs_setattr,
Josef Bacik5103e942007-11-16 11:45:54 -050010833 .listxattr = btrfs_listxattr,
Yanfdebe2b2008-01-14 13:26:08 -050010834 .permission = btrfs_permission,
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -050010835 .fiemap = btrfs_fiemap,
Christoph Hellwig4e34e712011-07-23 17:37:31 +020010836 .get_acl = btrfs_get_acl,
Christoph Hellwig996a7102013-12-20 05:16:43 -080010837 .set_acl = btrfs_set_acl,
Josef Bacike41f9412012-03-26 09:46:47 -040010838 .update_time = btrfs_update_time,
Chris Mason39279cc2007-06-12 06:35:45 -040010839};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -070010840static const struct inode_operations btrfs_special_inode_operations = {
Josef Bacik618e21d2007-07-11 10:18:17 -040010841 .getattr = btrfs_getattr,
10842 .setattr = btrfs_setattr,
Yanfdebe2b2008-01-14 13:26:08 -050010843 .permission = btrfs_permission,
Josef Bacik33268ea2008-07-24 12:16:36 -040010844 .listxattr = btrfs_listxattr,
Christoph Hellwig4e34e712011-07-23 17:37:31 +020010845 .get_acl = btrfs_get_acl,
Christoph Hellwig996a7102013-12-20 05:16:43 -080010846 .set_acl = btrfs_set_acl,
Josef Bacike41f9412012-03-26 09:46:47 -040010847 .update_time = btrfs_update_time,
Josef Bacik618e21d2007-07-11 10:18:17 -040010848};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -070010849static const struct inode_operations btrfs_symlink_inode_operations = {
Al Viro6b255392015-11-17 10:20:54 -050010850 .get_link = page_get_link,
Li Zefanf2095612010-11-19 02:05:24 +000010851 .getattr = btrfs_getattr,
Josef Bacik22c44fe2011-11-30 10:45:38 -050010852 .setattr = btrfs_setattr,
Yanfdebe2b2008-01-14 13:26:08 -050010853 .permission = btrfs_permission,
Jim Owens0279b4c2009-02-04 09:29:13 -050010854 .listxattr = btrfs_listxattr,
Josef Bacike41f9412012-03-26 09:46:47 -040010855 .update_time = btrfs_update_time,
Chris Mason39279cc2007-06-12 06:35:45 -040010856};
Yan, Zheng76dda932009-09-21 16:00:26 -040010857
Alexey Dobriyan82d339d2009-10-09 09:54:36 -040010858const struct dentry_operations btrfs_dentry_operations = {
Yan, Zheng76dda932009-09-21 16:00:26 -040010859 .d_delete = btrfs_dentry_delete,
Josef Bacikb4aff1f2011-06-28 16:18:59 -040010860 .d_release = btrfs_dentry_release,
Yan, Zheng76dda932009-09-21 16:00:26 -040010861};