blob: 2aaba58b4856be3fc6b179ad3e2d0e94e559dc04 [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>
33#include <linux/statfs.h>
34#include <linux/compat.h>
Chris Mason9ebefb182007-06-15 13:50:00 -040035#include <linux/bit_spinlock.h>
Josef Bacik5103e942007-11-16 11:45:54 -050036#include <linux/xattr.h>
Josef Bacik33268ea2008-07-24 12:16:36 -040037#include <linux/posix_acl.h>
Yan Zhengd899e052008-10-30 14:25:28 -040038#include <linux/falloc.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090039#include <linux/slab.h>
David Sterba7a36dde2011-05-06 15:33:15 +020040#include <linux/ratelimit.h>
Josef Bacik22c44fe2011-11-30 10:45:38 -050041#include <linux/mount.h>
Filipe Brandenburger55e301f2013-01-29 06:04:50 +000042#include <linux/btrfs.h>
David Woodhouse53b381b2013-01-29 18:40:14 -050043#include <linux/blkdev.h>
Josef Bacikf23b5a52013-06-19 10:16:26 -040044#include <linux/posix_acl_xattr.h>
Christoph Hellwige2e40f22015-02-22 08:58:50 -080045#include <linux/uio.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"
Chris Mason39279cc2007-06-12 06:35:45 -040063
64struct btrfs_iget_args {
Chris Mason90d3e592014-01-09 17:28:00 -080065 struct btrfs_key *location;
Chris Mason39279cc2007-06-12 06:35:45 -040066 struct btrfs_root *root;
67};
68
Filipe Mananaf28a4922015-12-08 19:23:20 +000069struct btrfs_dio_data {
70 u64 outstanding_extents;
71 u64 reserve;
72 u64 unsubmitted_oe_range_start;
73 u64 unsubmitted_oe_range_end;
74};
75
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -070076static const struct inode_operations btrfs_dir_inode_operations;
77static const struct inode_operations btrfs_symlink_inode_operations;
78static const struct inode_operations btrfs_dir_ro_inode_operations;
79static const struct inode_operations btrfs_special_inode_operations;
80static const struct inode_operations btrfs_file_inode_operations;
Alexey Dobriyan7f094102009-09-21 17:01:10 -070081static const struct address_space_operations btrfs_aops;
82static const struct address_space_operations btrfs_symlink_aops;
Alexey Dobriyan828c0952009-10-01 15:43:56 -070083static const struct file_operations btrfs_dir_file_operations;
David Sterba20e55062015-11-19 11:42:28 +010084static const struct extent_io_ops btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -040085
86static struct kmem_cache *btrfs_inode_cachep;
87struct kmem_cache *btrfs_trans_handle_cachep;
88struct kmem_cache *btrfs_transaction_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,
108 u64 start, u64 end, int *page_started,
109 unsigned long *nr_written, int unlock);
Josef Bacik70c8a912012-10-11 16:54:30 -0400110static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
111 u64 len, u64 orig_start,
112 u64 block_start, u64 block_len,
Josef Bacikcc95bef2013-04-04 14:31:27 -0400113 u64 orig_block_len, u64 ram_bytes,
114 int type);
Josef Bacik7b128762008-07-24 12:17:14 -0400115
Eric Sandeen48a3b632013-04-25 20:41:01 +0000116static int btrfs_dirty_inode(struct inode *inode);
Chris Mason39279cc2007-06-12 06:35:45 -0400117
Josef Bacik6a3891c2015-03-16 17:38:52 -0400118#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
119void btrfs_test_inode_set_ops(struct inode *inode)
120{
121 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
122}
123#endif
124
Yan, Zhengf34f57a2009-11-12 09:35:27 +0000125static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
Eric Paris2a7dba32011-02-01 11:05:39 -0500126 struct inode *inode, struct inode *dir,
127 const struct qstr *qstr)
Jim Owens0279b4c2009-02-04 09:29:13 -0500128{
129 int err;
130
Yan, Zhengf34f57a2009-11-12 09:35:27 +0000131 err = btrfs_init_acl(trans, inode, dir);
Jim Owens0279b4c2009-02-04 09:29:13 -0500132 if (!err)
Eric Paris2a7dba32011-02-01 11:05:39 -0500133 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
Jim Owens0279b4c2009-02-04 09:29:13 -0500134 return err;
135}
136
Chris Masond352ac62008-09-29 15:18:18 -0400137/*
Chris Masonc8b97812008-10-29 14:49:59 -0400138 * this does all the hard work for inserting an inline extent into
139 * the btree. The caller should have done a btrfs_drop_extents so that
140 * no overlapping inline items exist in the btree
141 */
Chris Mason40f76582014-05-21 13:35:51 -0700142static int insert_inline_extent(struct btrfs_trans_handle *trans,
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000143 struct btrfs_path *path, int extent_inserted,
Chris Masonc8b97812008-10-29 14:49:59 -0400144 struct btrfs_root *root, struct inode *inode,
145 u64 start, size_t size, size_t compressed_size,
Li Zefanfe3f5662011-03-28 08:30:38 +0000146 int compress_type,
Chris Masonc8b97812008-10-29 14:49:59 -0400147 struct page **compressed_pages)
148{
Chris Masonc8b97812008-10-29 14:49:59 -0400149 struct extent_buffer *leaf;
150 struct page *page = NULL;
151 char *kaddr;
152 unsigned long ptr;
153 struct btrfs_file_extent_item *ei;
154 int err = 0;
155 int ret;
156 size_t cur_size = size;
Chris Masonc8b97812008-10-29 14:49:59 -0400157 unsigned long offset;
Chris Masonc8b97812008-10-29 14:49:59 -0400158
Li Zefanfe3f5662011-03-28 08:30:38 +0000159 if (compressed_size && compressed_pages)
Chris Masonc8b97812008-10-29 14:49:59 -0400160 cur_size = compressed_size;
Chris Masonc8b97812008-10-29 14:49:59 -0400161
Chris Masonc8b97812008-10-29 14:49:59 -0400162 inode_add_bytes(inode, size);
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000163
164 if (!extent_inserted) {
165 struct btrfs_key key;
166 size_t datasize;
167
168 key.objectid = btrfs_ino(inode);
169 key.offset = start;
David Sterba962a2982014-06-04 18:41:45 +0200170 key.type = BTRFS_EXTENT_DATA_KEY;
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000171
172 datasize = btrfs_file_extent_calc_inline_size(cur_size);
173 path->leave_spinning = 1;
174 ret = btrfs_insert_empty_item(trans, root, path, &key,
175 datasize);
176 if (ret) {
177 err = ret;
178 goto fail;
179 }
Chris Masonc8b97812008-10-29 14:49:59 -0400180 }
181 leaf = path->nodes[0];
182 ei = btrfs_item_ptr(leaf, path->slots[0],
183 struct btrfs_file_extent_item);
184 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
185 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
186 btrfs_set_file_extent_encryption(leaf, ei, 0);
187 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
188 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
189 ptr = btrfs_file_extent_inline_start(ei);
190
Li Zefan261507a02010-12-17 14:21:50 +0800191 if (compress_type != BTRFS_COMPRESS_NONE) {
Chris Masonc8b97812008-10-29 14:49:59 -0400192 struct page *cpage;
193 int i = 0;
Chris Masond3977122009-01-05 21:25:51 -0500194 while (compressed_size > 0) {
Chris Masonc8b97812008-10-29 14:49:59 -0400195 cpage = compressed_pages[i];
Chris Mason5b050f02008-11-11 09:34:41 -0500196 cur_size = min_t(unsigned long, compressed_size,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300197 PAGE_SIZE);
Chris Masonc8b97812008-10-29 14:49:59 -0400198
Cong Wang7ac687d2011-11-25 23:14:28 +0800199 kaddr = kmap_atomic(cpage);
Chris Masonc8b97812008-10-29 14:49:59 -0400200 write_extent_buffer(leaf, kaddr, ptr, cur_size);
Cong Wang7ac687d2011-11-25 23:14:28 +0800201 kunmap_atomic(kaddr);
Chris Masonc8b97812008-10-29 14:49:59 -0400202
203 i++;
204 ptr += cur_size;
205 compressed_size -= cur_size;
206 }
207 btrfs_set_file_extent_compression(leaf, ei,
Li Zefan261507a02010-12-17 14:21:50 +0800208 compress_type);
Chris Masonc8b97812008-10-29 14:49:59 -0400209 } else {
210 page = find_get_page(inode->i_mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300211 start >> PAGE_SHIFT);
Chris Masonc8b97812008-10-29 14:49:59 -0400212 btrfs_set_file_extent_compression(leaf, ei, 0);
Cong Wang7ac687d2011-11-25 23:14:28 +0800213 kaddr = kmap_atomic(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300214 offset = start & (PAGE_SIZE - 1);
Chris Masonc8b97812008-10-29 14:49:59 -0400215 write_extent_buffer(leaf, kaddr + offset, ptr, size);
Cong Wang7ac687d2011-11-25 23:14:28 +0800216 kunmap_atomic(kaddr);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300217 put_page(page);
Chris Masonc8b97812008-10-29 14:49:59 -0400218 }
219 btrfs_mark_buffer_dirty(leaf);
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000220 btrfs_release_path(path);
Chris Masonc8b97812008-10-29 14:49:59 -0400221
Yan, Zhengc2167752009-11-12 09:34:21 +0000222 /*
223 * we're an inline extent, so nobody can
224 * extend the file past i_size without locking
225 * a page we already have locked.
226 *
227 * We must do any isize and inode updates
228 * before we unlock the pages. Otherwise we
229 * could end up racing with unlink.
230 */
Chris Masonc8b97812008-10-29 14:49:59 -0400231 BTRFS_I(inode)->disk_i_size = inode->i_size;
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100232 ret = btrfs_update_inode(trans, root, inode);
Yan, Zhengc2167752009-11-12 09:34:21 +0000233
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100234 return ret;
Chris Masonc8b97812008-10-29 14:49:59 -0400235fail:
Chris Masonc8b97812008-10-29 14:49:59 -0400236 return err;
237}
238
239
240/*
241 * conditionally insert an inline extent into the file. This
242 * does the checks required to make sure the data is small enough
243 * to fit as an inline extent.
244 */
Josef Bacik00361582013-08-14 14:02:47 -0400245static noinline int cow_file_range_inline(struct btrfs_root *root,
246 struct inode *inode, u64 start,
247 u64 end, size_t compressed_size,
248 int compress_type,
249 struct page **compressed_pages)
Chris Masonc8b97812008-10-29 14:49:59 -0400250{
Josef Bacik00361582013-08-14 14:02:47 -0400251 struct btrfs_trans_handle *trans;
Chris Masonc8b97812008-10-29 14:49:59 -0400252 u64 isize = i_size_read(inode);
253 u64 actual_end = min(end + 1, isize);
254 u64 inline_len = actual_end - start;
Qu Wenruofda28322013-02-26 08:10:22 +0000255 u64 aligned_end = ALIGN(end, root->sectorsize);
Chris Masonc8b97812008-10-29 14:49:59 -0400256 u64 data_len = inline_len;
257 int ret;
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000258 struct btrfs_path *path;
259 int extent_inserted = 0;
260 u32 extent_item_size;
Chris Masonc8b97812008-10-29 14:49:59 -0400261
262 if (compressed_size)
263 data_len = compressed_size;
264
265 if (start > 0 ||
Chandan Rajendra0c29ba92016-01-21 15:56:01 +0530266 actual_end > root->sectorsize ||
Wang Shilong354877b2014-07-17 11:44:11 +0800267 data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
Chris Masonc8b97812008-10-29 14:49:59 -0400268 (!compressed_size &&
269 (actual_end & (root->sectorsize - 1)) == 0) ||
270 end + 1 < isize ||
271 data_len > root->fs_info->max_inline) {
272 return 1;
273 }
274
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000275 path = btrfs_alloc_path();
276 if (!path)
277 return -ENOMEM;
278
Josef Bacik00361582013-08-14 14:02:47 -0400279 trans = btrfs_join_transaction(root);
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000280 if (IS_ERR(trans)) {
281 btrfs_free_path(path);
Josef Bacik00361582013-08-14 14:02:47 -0400282 return PTR_ERR(trans);
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000283 }
Josef Bacik00361582013-08-14 14:02:47 -0400284 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
285
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000286 if (compressed_size && compressed_pages)
287 extent_item_size = btrfs_file_extent_calc_inline_size(
288 compressed_size);
289 else
290 extent_item_size = btrfs_file_extent_calc_inline_size(
291 inline_len);
292
293 ret = __btrfs_drop_extents(trans, root, inode, path,
294 start, aligned_end, NULL,
295 1, 1, extent_item_size, &extent_inserted);
Josef Bacik00361582013-08-14 14:02:47 -0400296 if (ret) {
297 btrfs_abort_transaction(trans, root, ret);
298 goto out;
299 }
Chris Masonc8b97812008-10-29 14:49:59 -0400300
301 if (isize > actual_end)
302 inline_len = min_t(u64, isize, actual_end);
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000303 ret = insert_inline_extent(trans, path, extent_inserted,
304 root, inode, start,
Chris Masonc8b97812008-10-29 14:49:59 -0400305 inline_len, compressed_size,
Li Zefanfe3f5662011-03-28 08:30:38 +0000306 compress_type, compressed_pages);
Josef Bacik2adcac12012-05-23 16:10:14 -0400307 if (ret && ret != -ENOSPC) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100308 btrfs_abort_transaction(trans, root, ret);
Josef Bacik00361582013-08-14 14:02:47 -0400309 goto out;
Josef Bacik2adcac12012-05-23 16:10:14 -0400310 } else if (ret == -ENOSPC) {
Josef Bacik00361582013-08-14 14:02:47 -0400311 ret = 1;
312 goto out;
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100313 }
Josef Bacik2adcac12012-05-23 16:10:14 -0400314
Josef Bacikbdc20e62013-02-28 13:23:38 -0500315 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -0400316 btrfs_delalloc_release_metadata(inode, end + 1 - start);
Chris Masona1ed8352009-09-11 12:27:37 -0400317 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
Josef Bacik00361582013-08-14 14:02:47 -0400318out:
Qu Wenruo94ed9382015-09-08 17:25:56 +0800319 /*
320 * Don't forget to free the reserved space, as for inlined extent
321 * it won't count as data extent, free them directly here.
322 * And at reserve time, it's always aligned to page size, so
323 * just free one page here.
324 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300325 btrfs_qgroup_free_data(inode, 0, PAGE_SIZE);
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000326 btrfs_free_path(path);
Josef Bacik00361582013-08-14 14:02:47 -0400327 btrfs_end_transaction(trans, root);
328 return ret;
Chris Masonc8b97812008-10-29 14:49:59 -0400329}
330
Chris Mason771ed682008-11-06 22:02:51 -0500331struct async_extent {
332 u64 start;
333 u64 ram_size;
334 u64 compressed_size;
335 struct page **pages;
336 unsigned long nr_pages;
Li Zefan261507a02010-12-17 14:21:50 +0800337 int compress_type;
Chris Mason771ed682008-11-06 22:02:51 -0500338 struct list_head list;
339};
340
341struct async_cow {
342 struct inode *inode;
343 struct btrfs_root *root;
344 struct page *locked_page;
345 u64 start;
346 u64 end;
347 struct list_head extents;
348 struct btrfs_work work;
349};
350
351static noinline int add_async_extent(struct async_cow *cow,
352 u64 start, u64 ram_size,
353 u64 compressed_size,
354 struct page **pages,
Li Zefan261507a02010-12-17 14:21:50 +0800355 unsigned long nr_pages,
356 int compress_type)
Chris Mason771ed682008-11-06 22:02:51 -0500357{
358 struct async_extent *async_extent;
359
360 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100361 BUG_ON(!async_extent); /* -ENOMEM */
Chris Mason771ed682008-11-06 22:02:51 -0500362 async_extent->start = start;
363 async_extent->ram_size = ram_size;
364 async_extent->compressed_size = compressed_size;
365 async_extent->pages = pages;
366 async_extent->nr_pages = nr_pages;
Li Zefan261507a02010-12-17 14:21:50 +0800367 async_extent->compress_type = compress_type;
Chris Mason771ed682008-11-06 22:02:51 -0500368 list_add_tail(&async_extent->list, &cow->extents);
369 return 0;
370}
371
Wang Shilongf79707b2014-07-17 11:44:09 +0800372static inline int inode_need_compress(struct inode *inode)
373{
374 struct btrfs_root *root = BTRFS_I(inode)->root;
375
376 /* force compress */
377 if (btrfs_test_opt(root, FORCE_COMPRESS))
378 return 1;
379 /* bad compression ratios */
380 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
381 return 0;
382 if (btrfs_test_opt(root, COMPRESS) ||
383 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
384 BTRFS_I(inode)->force_compress)
385 return 1;
386 return 0;
387}
388
Chris Masonc8b97812008-10-29 14:49:59 -0400389/*
Chris Mason771ed682008-11-06 22:02:51 -0500390 * we create compressed extents in two phases. The first
391 * phase compresses a range of pages that have already been
392 * locked (both pages and state bits are locked).
Chris Masonc8b97812008-10-29 14:49:59 -0400393 *
Chris Mason771ed682008-11-06 22:02:51 -0500394 * This is done inside an ordered work queue, and the compression
395 * is spread across many cpus. The actual IO submission is step
396 * two, and the ordered work queue takes care of making sure that
397 * happens in the same order things were put onto the queue by
398 * writepages and friends.
Chris Masonc8b97812008-10-29 14:49:59 -0400399 *
Chris Mason771ed682008-11-06 22:02:51 -0500400 * If this code finds it can't get good compression, it puts an
401 * entry onto the work queue to write the uncompressed bytes. This
402 * makes sure that both compressed inodes and uncompressed inodes
Artem Bityutskiyb2570312012-07-25 18:12:06 +0300403 * are written in the same order that the flusher thread sent them
404 * down.
Chris Masond352ac62008-09-29 15:18:18 -0400405 */
Filipe Mananac44f6492014-10-09 21:15:44 +0100406static noinline void compress_file_range(struct inode *inode,
Chris Mason771ed682008-11-06 22:02:51 -0500407 struct page *locked_page,
408 u64 start, u64 end,
409 struct async_cow *async_cow,
410 int *num_added)
Chris Masonb888db22007-08-27 16:49:44 -0400411{
412 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masondb945352007-10-15 16:15:53 -0400413 u64 num_bytes;
Chris Masondb945352007-10-15 16:15:53 -0400414 u64 blocksize = root->sectorsize;
Chris Masonc8b97812008-10-29 14:49:59 -0400415 u64 actual_end;
Chris Mason42dc7ba2008-12-15 11:44:56 -0500416 u64 isize = i_size_read(inode);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400417 int ret = 0;
Chris Masonc8b97812008-10-29 14:49:59 -0400418 struct page **pages = NULL;
419 unsigned long nr_pages;
420 unsigned long nr_pages_ret = 0;
421 unsigned long total_compressed = 0;
422 unsigned long total_in = 0;
Byongho Leeee221842015-12-15 01:42:10 +0900423 unsigned long max_compressed = SZ_128K;
424 unsigned long max_uncompressed = SZ_128K;
Chris Masonc8b97812008-10-29 14:49:59 -0400425 int i;
426 int will_compress;
Li Zefan261507a02010-12-17 14:21:50 +0800427 int compress_type = root->fs_info->compress_type;
Chris Mason4adaa612013-03-26 13:07:00 -0400428 int redirty = 0;
Chris Masonb888db22007-08-27 16:49:44 -0400429
Liu Bo4cb13e52012-03-29 09:57:45 -0400430 /* if this is a small write inside eof, kick off a defrag */
Byongho Leeee221842015-12-15 01:42:10 +0900431 if ((end - start + 1) < SZ_16K &&
Liu Bo4cb13e52012-03-29 09:57:45 -0400432 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
Chris Mason4cb53002011-05-24 15:35:30 -0400433 btrfs_add_inode_defrag(NULL, inode);
434
Chris Mason42dc7ba2008-12-15 11:44:56 -0500435 actual_end = min_t(u64, isize, end + 1);
Chris Masonc8b97812008-10-29 14:49:59 -0400436again:
437 will_compress = 0;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300438 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
439 nr_pages = min_t(unsigned long, nr_pages, SZ_128K / PAGE_SIZE);
Chris Masonc8b97812008-10-29 14:49:59 -0400440
Chris Masonf03d9302009-02-04 09:31:06 -0500441 /*
442 * we don't want to send crud past the end of i_size through
443 * compression, that's just a waste of CPU time. So, if the
444 * end of the file is before the start of our current
445 * requested range of bytes, we bail out to the uncompressed
446 * cleanup code that can deal with all of this.
447 *
448 * It isn't really the fastest way to fix things, but this is a
449 * very uncommon corner.
450 */
451 if (actual_end <= start)
452 goto cleanup_and_bail_uncompressed;
453
Chris Masonc8b97812008-10-29 14:49:59 -0400454 total_compressed = actual_end - start;
455
Shilong Wang4bcbb332014-10-07 18:44:35 -0400456 /*
457 * skip compression for a small file range(<=blocksize) that
458 * isn't an inline extent, since it dosen't save disk space at all.
459 */
460 if (total_compressed <= blocksize &&
461 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
462 goto cleanup_and_bail_uncompressed;
463
Chris Masonc8b97812008-10-29 14:49:59 -0400464 /* we want to make sure that amount of ram required to uncompress
465 * an extent is reasonable, so we limit the total size in ram
Chris Mason771ed682008-11-06 22:02:51 -0500466 * of a compressed extent to 128k. This is a crucial number
467 * because it also controls how easily we can spread reads across
468 * cpus for decompression.
469 *
470 * We also want to make sure the amount of IO required to do
471 * a random read is reasonably small, so we limit the size of
472 * a compressed extent to 128k.
Chris Masonc8b97812008-10-29 14:49:59 -0400473 */
474 total_compressed = min(total_compressed, max_uncompressed);
Qu Wenruofda28322013-02-26 08:10:22 +0000475 num_bytes = ALIGN(end - start + 1, blocksize);
Chris Masonbe20aa92007-12-17 20:14:01 -0500476 num_bytes = max(blocksize, num_bytes);
Chris Masonc8b97812008-10-29 14:49:59 -0400477 total_in = 0;
478 ret = 0;
Chris Masondb945352007-10-15 16:15:53 -0400479
Chris Mason771ed682008-11-06 22:02:51 -0500480 /*
481 * we do compression for mount -o compress and when the
482 * inode has not been flagged as nocompress. This flag can
483 * change at any time if we discover bad compression ratios.
Chris Masonc8b97812008-10-29 14:49:59 -0400484 */
Wang Shilongf79707b2014-07-17 11:44:09 +0800485 if (inode_need_compress(inode)) {
Chris Masonc8b97812008-10-29 14:49:59 -0400486 WARN_ON(pages);
David Sterba31e818f2015-02-20 18:00:26 +0100487 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
Li Zefan560f7d72011-09-08 10:22:01 +0800488 if (!pages) {
489 /* just bail out to the uncompressed code */
490 goto cont;
491 }
Chris Mason179e29e2007-11-01 11:28:41 -0400492
Li Zefan261507a02010-12-17 14:21:50 +0800493 if (BTRFS_I(inode)->force_compress)
494 compress_type = BTRFS_I(inode)->force_compress;
495
Chris Mason4adaa612013-03-26 13:07:00 -0400496 /*
497 * we need to call clear_page_dirty_for_io on each
498 * page in the range. Otherwise applications with the file
499 * mmap'd can wander in and change the page contents while
500 * we are compressing them.
501 *
502 * If the compression fails for any reason, we set the pages
503 * dirty again later on.
504 */
505 extent_range_clear_dirty_for_io(inode, start, end);
506 redirty = 1;
Li Zefan261507a02010-12-17 14:21:50 +0800507 ret = btrfs_compress_pages(compress_type,
508 inode->i_mapping, start,
509 total_compressed, pages,
510 nr_pages, &nr_pages_ret,
511 &total_in,
512 &total_compressed,
513 max_compressed);
Chris Masonc8b97812008-10-29 14:49:59 -0400514
515 if (!ret) {
516 unsigned long offset = total_compressed &
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300517 (PAGE_SIZE - 1);
Chris Masonc8b97812008-10-29 14:49:59 -0400518 struct page *page = pages[nr_pages_ret - 1];
519 char *kaddr;
520
521 /* zero the tail end of the last page, we might be
522 * sending it down to disk
523 */
524 if (offset) {
Cong Wang7ac687d2011-11-25 23:14:28 +0800525 kaddr = kmap_atomic(page);
Chris Masonc8b97812008-10-29 14:49:59 -0400526 memset(kaddr + offset, 0,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300527 PAGE_SIZE - offset);
Cong Wang7ac687d2011-11-25 23:14:28 +0800528 kunmap_atomic(kaddr);
Chris Masonc8b97812008-10-29 14:49:59 -0400529 }
530 will_compress = 1;
531 }
532 }
Li Zefan560f7d72011-09-08 10:22:01 +0800533cont:
Chris Masonc8b97812008-10-29 14:49:59 -0400534 if (start == 0) {
535 /* lets try to make an inline extent */
Chris Mason771ed682008-11-06 22:02:51 -0500536 if (ret || total_in < (actual_end - start)) {
Chris Masonc8b97812008-10-29 14:49:59 -0400537 /* we didn't compress the entire range, try
Chris Mason771ed682008-11-06 22:02:51 -0500538 * to make an uncompressed inline extent.
Chris Masonc8b97812008-10-29 14:49:59 -0400539 */
Josef Bacik00361582013-08-14 14:02:47 -0400540 ret = cow_file_range_inline(root, inode, start, end,
541 0, 0, NULL);
Chris Masonc8b97812008-10-29 14:49:59 -0400542 } else {
Chris Mason771ed682008-11-06 22:02:51 -0500543 /* try making a compressed inline extent */
Josef Bacik00361582013-08-14 14:02:47 -0400544 ret = cow_file_range_inline(root, inode, start, end,
Li Zefanfe3f5662011-03-28 08:30:38 +0000545 total_compressed,
546 compress_type, pages);
Chris Masonc8b97812008-10-29 14:49:59 -0400547 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100548 if (ret <= 0) {
Josef Bacik151a41b2013-07-29 13:22:24 -0400549 unsigned long clear_flags = EXTENT_DELALLOC |
550 EXTENT_DEFRAG;
Filipe Mananae6eb4312014-10-10 10:45:12 +0100551 unsigned long page_error_op;
552
Josef Bacik151a41b2013-07-29 13:22:24 -0400553 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
Filipe Mananae6eb4312014-10-10 10:45:12 +0100554 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
Josef Bacik151a41b2013-07-29 13:22:24 -0400555
Chris Mason771ed682008-11-06 22:02:51 -0500556 /*
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100557 * inline extent creation worked or returned error,
558 * we don't need to create any more async work items.
559 * Unlock and free up our temp pages.
Chris Mason771ed682008-11-06 22:02:51 -0500560 */
Josef Bacikc2790a22013-07-29 11:20:47 -0400561 extent_clear_unlock_delalloc(inode, start, end, NULL,
Josef Bacik151a41b2013-07-29 13:22:24 -0400562 clear_flags, PAGE_UNLOCK |
Josef Bacikc2790a22013-07-29 11:20:47 -0400563 PAGE_CLEAR_DIRTY |
564 PAGE_SET_WRITEBACK |
Filipe Mananae6eb4312014-10-10 10:45:12 +0100565 page_error_op |
Josef Bacikc2790a22013-07-29 11:20:47 -0400566 PAGE_END_WRITEBACK);
Chris Masonc8b97812008-10-29 14:49:59 -0400567 goto free_pages_out;
568 }
569 }
570
571 if (will_compress) {
572 /*
573 * we aren't doing an inline extent round the compressed size
574 * up to a block size boundary so the allocator does sane
575 * things
576 */
Qu Wenruofda28322013-02-26 08:10:22 +0000577 total_compressed = ALIGN(total_compressed, blocksize);
Chris Masonc8b97812008-10-29 14:49:59 -0400578
579 /*
580 * one last check to make sure the compression is really a
581 * win, compare the page count read with the blocks on disk
582 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300583 total_in = ALIGN(total_in, PAGE_SIZE);
Chris Masonc8b97812008-10-29 14:49:59 -0400584 if (total_compressed >= total_in) {
585 will_compress = 0;
586 } else {
Chris Masonc8b97812008-10-29 14:49:59 -0400587 num_bytes = total_in;
588 }
589 }
590 if (!will_compress && pages) {
591 /*
592 * the compression code ran but failed to make things smaller,
593 * free any pages it allocated and our page pointer array
594 */
595 for (i = 0; i < nr_pages_ret; i++) {
Chris Mason70b99e62008-10-31 12:46:39 -0400596 WARN_ON(pages[i]->mapping);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300597 put_page(pages[i]);
Chris Masonc8b97812008-10-29 14:49:59 -0400598 }
599 kfree(pages);
600 pages = NULL;
601 total_compressed = 0;
602 nr_pages_ret = 0;
603
604 /* flag the file so we don't compress in the future */
Chris Mason1e701a32010-03-11 09:42:04 -0500605 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
606 !(BTRFS_I(inode)->force_compress)) {
Chris Masona555f812010-01-28 16:18:15 -0500607 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
Chris Mason1e701a32010-03-11 09:42:04 -0500608 }
Chris Masonc8b97812008-10-29 14:49:59 -0400609 }
Chris Mason771ed682008-11-06 22:02:51 -0500610 if (will_compress) {
611 *num_added += 1;
612
613 /* the async work queues will take care of doing actual
614 * allocation on disk for these compressed pages,
615 * and will submit them to the elevator.
616 */
617 add_async_extent(async_cow, start, num_bytes,
Li Zefan261507a02010-12-17 14:21:50 +0800618 total_compressed, pages, nr_pages_ret,
619 compress_type);
Chris Mason771ed682008-11-06 22:02:51 -0500620
Yan, Zheng24ae6362010-12-06 07:02:36 +0000621 if (start + num_bytes < end) {
Chris Mason771ed682008-11-06 22:02:51 -0500622 start += num_bytes;
623 pages = NULL;
624 cond_resched();
625 goto again;
626 }
627 } else {
Chris Masonf03d9302009-02-04 09:31:06 -0500628cleanup_and_bail_uncompressed:
Chris Mason771ed682008-11-06 22:02:51 -0500629 /*
630 * No compression, but we still need to write the pages in
631 * the file we've been given so far. redirty the locked
632 * page if it corresponds to our extent and set things up
633 * for the async work queue to run cow_file_range to do
634 * the normal delalloc dance
635 */
636 if (page_offset(locked_page) >= start &&
637 page_offset(locked_page) <= end) {
638 __set_page_dirty_nobuffers(locked_page);
639 /* unlocked later on in the async handlers */
640 }
Chris Mason4adaa612013-03-26 13:07:00 -0400641 if (redirty)
642 extent_range_redirty_for_io(inode, start, end);
Li Zefan261507a02010-12-17 14:21:50 +0800643 add_async_extent(async_cow, start, end - start + 1,
644 0, NULL, 0, BTRFS_COMPRESS_NONE);
Chris Mason771ed682008-11-06 22:02:51 -0500645 *num_added += 1;
646 }
647
Filipe Mananac44f6492014-10-09 21:15:44 +0100648 return;
Chris Mason771ed682008-11-06 22:02:51 -0500649
650free_pages_out:
651 for (i = 0; i < nr_pages_ret; i++) {
652 WARN_ON(pages[i]->mapping);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300653 put_page(pages[i]);
Chris Mason771ed682008-11-06 22:02:51 -0500654 }
Chris Masond3977122009-01-05 21:25:51 -0500655 kfree(pages);
Chris Mason771ed682008-11-06 22:02:51 -0500656}
Chris Mason771ed682008-11-06 22:02:51 -0500657
Filipe Manana40ae8372014-10-06 22:14:24 +0100658static void free_async_extent_pages(struct async_extent *async_extent)
659{
660 int i;
661
662 if (!async_extent->pages)
663 return;
664
665 for (i = 0; i < async_extent->nr_pages; i++) {
666 WARN_ON(async_extent->pages[i]->mapping);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300667 put_page(async_extent->pages[i]);
Filipe Manana40ae8372014-10-06 22:14:24 +0100668 }
669 kfree(async_extent->pages);
670 async_extent->nr_pages = 0;
671 async_extent->pages = NULL;
Chris Mason771ed682008-11-06 22:02:51 -0500672}
673
674/*
675 * phase two of compressed writeback. This is the ordered portion
676 * of the code, which only gets called in the order the work was
677 * queued. We walk all the async extents created by compress_file_range
678 * and send them down to the disk.
679 */
Filipe Mananadec8f172014-10-06 22:14:26 +0100680static noinline void submit_compressed_extents(struct inode *inode,
Chris Mason771ed682008-11-06 22:02:51 -0500681 struct async_cow *async_cow)
682{
683 struct async_extent *async_extent;
684 u64 alloc_hint = 0;
Chris Mason771ed682008-11-06 22:02:51 -0500685 struct btrfs_key ins;
686 struct extent_map *em;
687 struct btrfs_root *root = BTRFS_I(inode)->root;
688 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
689 struct extent_io_tree *io_tree;
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500690 int ret = 0;
Chris Mason771ed682008-11-06 22:02:51 -0500691
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500692again:
Chris Masond3977122009-01-05 21:25:51 -0500693 while (!list_empty(&async_cow->extents)) {
Chris Mason771ed682008-11-06 22:02:51 -0500694 async_extent = list_entry(async_cow->extents.next,
695 struct async_extent, list);
696 list_del(&async_extent->list);
697
698 io_tree = &BTRFS_I(inode)->io_tree;
699
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500700retry:
Chris Mason771ed682008-11-06 22:02:51 -0500701 /* did the compression code fall back to uncompressed IO? */
702 if (!async_extent->pages) {
703 int page_started = 0;
704 unsigned long nr_written = 0;
705
706 lock_extent(io_tree, async_extent->start,
Josef Bacik2ac55d42010-02-03 19:33:23 +0000707 async_extent->start +
Jeff Mahoneyd0082372012-03-01 14:57:19 +0100708 async_extent->ram_size - 1);
Chris Mason771ed682008-11-06 22:02:51 -0500709
710 /* allocate blocks */
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500711 ret = cow_file_range(inode, async_cow->locked_page,
712 async_extent->start,
713 async_extent->start +
714 async_extent->ram_size - 1,
715 &page_started, &nr_written, 0);
Chris Mason771ed682008-11-06 22:02:51 -0500716
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100717 /* JDM XXX */
718
Chris Mason771ed682008-11-06 22:02:51 -0500719 /*
720 * if page_started, cow_file_range inserted an
721 * inline extent and took care of all the unlocking
722 * and IO for us. Otherwise, we need to submit
723 * all those pages down to the drive.
724 */
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500725 if (!page_started && !ret)
Chris Mason771ed682008-11-06 22:02:51 -0500726 extent_write_locked_range(io_tree,
727 inode, async_extent->start,
Chris Masond3977122009-01-05 21:25:51 -0500728 async_extent->start +
Chris Mason771ed682008-11-06 22:02:51 -0500729 async_extent->ram_size - 1,
730 btrfs_get_extent,
731 WB_SYNC_ALL);
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500732 else if (ret)
733 unlock_page(async_cow->locked_page);
Chris Mason771ed682008-11-06 22:02:51 -0500734 kfree(async_extent);
735 cond_resched();
736 continue;
737 }
738
739 lock_extent(io_tree, async_extent->start,
Jeff Mahoneyd0082372012-03-01 14:57:19 +0100740 async_extent->start + async_extent->ram_size - 1);
Chris Mason771ed682008-11-06 22:02:51 -0500741
Josef Bacik00361582013-08-14 14:02:47 -0400742 ret = btrfs_reserve_extent(root,
Chris Mason771ed682008-11-06 22:02:51 -0500743 async_extent->compressed_size,
744 async_extent->compressed_size,
Miao Xiee570fd22014-06-19 10:42:50 +0800745 0, alloc_hint, &ins, 1, 1);
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500746 if (ret) {
Filipe Manana40ae8372014-10-06 22:14:24 +0100747 free_async_extent_pages(async_extent);
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500748
Josef Bacikfdf8e2e2013-06-14 16:58:23 -0400749 if (ret == -ENOSPC) {
750 unlock_extent(io_tree, async_extent->start,
751 async_extent->start +
752 async_extent->ram_size - 1);
Liu Boce620032014-07-24 22:48:05 +0800753
754 /*
755 * we need to redirty the pages if we decide to
756 * fallback to uncompressed IO, otherwise we
757 * will not submit these pages down to lower
758 * layers.
759 */
760 extent_range_redirty_for_io(inode,
761 async_extent->start,
762 async_extent->start +
763 async_extent->ram_size - 1);
764
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100765 goto retry;
Josef Bacikfdf8e2e2013-06-14 16:58:23 -0400766 }
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500767 goto out_free;
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500768 }
Yan, Zhengc2167752009-11-12 09:34:21 +0000769 /*
770 * here we're doing allocation and writeback of the
771 * compressed pages
772 */
773 btrfs_drop_extent_cache(inode, async_extent->start,
774 async_extent->start +
775 async_extent->ram_size - 1, 0);
776
David Sterba172ddd62011-04-21 00:48:27 +0200777 em = alloc_extent_map();
Liu Bob9aa55b2013-05-14 02:12:15 +0000778 if (!em) {
779 ret = -ENOMEM;
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500780 goto out_free_reserve;
Liu Bob9aa55b2013-05-14 02:12:15 +0000781 }
Chris Mason771ed682008-11-06 22:02:51 -0500782 em->start = async_extent->start;
783 em->len = async_extent->ram_size;
Chris Mason445a6942008-11-10 11:53:33 -0500784 em->orig_start = em->start;
Josef Bacik2ab28f32012-10-12 15:27:49 -0400785 em->mod_start = em->start;
786 em->mod_len = em->len;
Chris Mason771ed682008-11-06 22:02:51 -0500787
788 em->block_start = ins.objectid;
789 em->block_len = ins.offset;
Josef Bacikb4939682012-12-03 10:31:19 -0500790 em->orig_block_len = ins.offset;
Josef Bacikcc95bef2013-04-04 14:31:27 -0400791 em->ram_bytes = async_extent->ram_size;
Chris Mason771ed682008-11-06 22:02:51 -0500792 em->bdev = root->fs_info->fs_devices->latest_bdev;
Li Zefan261507a02010-12-17 14:21:50 +0800793 em->compress_type = async_extent->compress_type;
Chris Mason771ed682008-11-06 22:02:51 -0500794 set_bit(EXTENT_FLAG_PINNED, &em->flags);
795 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
Josef Bacik70c8a912012-10-11 16:54:30 -0400796 em->generation = -1;
Chris Mason771ed682008-11-06 22:02:51 -0500797
Chris Masond3977122009-01-05 21:25:51 -0500798 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -0400799 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -0400800 ret = add_extent_mapping(em_tree, em, 1);
Chris Mason890871b2009-09-02 16:24:52 -0400801 write_unlock(&em_tree->lock);
Chris Mason771ed682008-11-06 22:02:51 -0500802 if (ret != -EEXIST) {
803 free_extent_map(em);
804 break;
805 }
806 btrfs_drop_extent_cache(inode, async_extent->start,
807 async_extent->start +
808 async_extent->ram_size - 1, 0);
809 }
810
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500811 if (ret)
812 goto out_free_reserve;
813
Li Zefan261507a02010-12-17 14:21:50 +0800814 ret = btrfs_add_ordered_extent_compress(inode,
815 async_extent->start,
816 ins.objectid,
817 async_extent->ram_size,
818 ins.offset,
819 BTRFS_ORDERED_COMPRESSED,
820 async_extent->compress_type);
Filipe Mananad9f85962014-08-25 10:43:00 +0100821 if (ret) {
822 btrfs_drop_extent_cache(inode, async_extent->start,
823 async_extent->start +
824 async_extent->ram_size - 1, 0);
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500825 goto out_free_reserve;
Filipe Mananad9f85962014-08-25 10:43:00 +0100826 }
Chris Mason771ed682008-11-06 22:02:51 -0500827
Chris Mason771ed682008-11-06 22:02:51 -0500828 /*
829 * clear dirty, set writeback and unlock the pages.
830 */
Josef Bacikc2790a22013-07-29 11:20:47 -0400831 extent_clear_unlock_delalloc(inode, async_extent->start,
Chris Masona791e352009-10-08 11:27:10 -0400832 async_extent->start +
833 async_extent->ram_size - 1,
Josef Bacik151a41b2013-07-29 13:22:24 -0400834 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
835 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
Josef Bacikc2790a22013-07-29 11:20:47 -0400836 PAGE_SET_WRITEBACK);
Chris Mason771ed682008-11-06 22:02:51 -0500837 ret = btrfs_submit_compressed_write(inode,
Chris Masond3977122009-01-05 21:25:51 -0500838 async_extent->start,
839 async_extent->ram_size,
840 ins.objectid,
841 ins.offset, async_extent->pages,
842 async_extent->nr_pages);
Filipe Mananafce2a4e2014-10-06 22:14:23 +0100843 if (ret) {
844 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
845 struct page *p = async_extent->pages[0];
846 const u64 start = async_extent->start;
847 const u64 end = start + async_extent->ram_size - 1;
848
849 p->mapping = inode->i_mapping;
850 tree->ops->writepage_end_io_hook(p, start, end,
851 NULL, 0);
852 p->mapping = NULL;
853 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
854 PAGE_END_WRITEBACK |
855 PAGE_SET_ERROR);
Filipe Manana40ae8372014-10-06 22:14:24 +0100856 free_async_extent_pages(async_extent);
Filipe Mananafce2a4e2014-10-06 22:14:23 +0100857 }
Chris Mason771ed682008-11-06 22:02:51 -0500858 alloc_hint = ins.objectid + ins.offset;
859 kfree(async_extent);
860 cond_resched();
861 }
Filipe Mananadec8f172014-10-06 22:14:26 +0100862 return;
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500863out_free_reserve:
Miao Xiee570fd22014-06-19 10:42:50 +0800864 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100865out_free:
Josef Bacikc2790a22013-07-29 11:20:47 -0400866 extent_clear_unlock_delalloc(inode, async_extent->start,
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500867 async_extent->start +
868 async_extent->ram_size - 1,
Josef Bacikc2790a22013-07-29 11:20:47 -0400869 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
Josef Bacik151a41b2013-07-29 13:22:24 -0400870 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
871 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
Filipe Manana704de492014-10-06 22:14:22 +0100872 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
873 PAGE_SET_ERROR);
Filipe Manana40ae8372014-10-06 22:14:24 +0100874 free_async_extent_pages(async_extent);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100875 kfree(async_extent);
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500876 goto again;
Chris Mason771ed682008-11-06 22:02:51 -0500877}
878
Josef Bacik4b46fce2010-05-23 11:00:55 -0400879static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
880 u64 num_bytes)
881{
882 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
883 struct extent_map *em;
884 u64 alloc_hint = 0;
885
886 read_lock(&em_tree->lock);
887 em = search_extent_mapping(em_tree, start, num_bytes);
888 if (em) {
889 /*
890 * if block start isn't an actual block number then find the
891 * first block in this inode and use that as a hint. If that
892 * block is also bogus then just don't worry about it.
893 */
894 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
895 free_extent_map(em);
896 em = search_extent_mapping(em_tree, 0, 0);
897 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
898 alloc_hint = em->block_start;
899 if (em)
900 free_extent_map(em);
901 } else {
902 alloc_hint = em->block_start;
903 free_extent_map(em);
904 }
905 }
906 read_unlock(&em_tree->lock);
907
908 return alloc_hint;
909}
910
Chris Mason771ed682008-11-06 22:02:51 -0500911/*
912 * when extent_io.c finds a delayed allocation range in the file,
913 * the call backs end up in this code. The basic idea is to
914 * allocate extents on disk for the range, and create ordered data structs
915 * in ram to track those extents.
916 *
917 * locked_page is the page that writepage had locked already. We use
918 * it to make sure we don't do extra locks or unlocks.
919 *
920 * *page_started is set to one if we unlock locked_page and do everything
921 * required to start IO on it. It may be clean and already done with
922 * IO when we return.
923 */
Josef Bacik00361582013-08-14 14:02:47 -0400924static noinline int cow_file_range(struct inode *inode,
925 struct page *locked_page,
926 u64 start, u64 end, int *page_started,
927 unsigned long *nr_written,
928 int unlock)
Chris Mason771ed682008-11-06 22:02:51 -0500929{
Josef Bacik00361582013-08-14 14:02:47 -0400930 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason771ed682008-11-06 22:02:51 -0500931 u64 alloc_hint = 0;
932 u64 num_bytes;
933 unsigned long ram_size;
934 u64 disk_num_bytes;
935 u64 cur_alloc_size;
936 u64 blocksize = root->sectorsize;
Chris Mason771ed682008-11-06 22:02:51 -0500937 struct btrfs_key ins;
938 struct extent_map *em;
939 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
940 int ret = 0;
941
Josef Bacik02ecd2c2013-10-25 16:19:08 -0400942 if (btrfs_is_free_space_inode(inode)) {
943 WARN_ON_ONCE(1);
Josef Bacik29bce2f2014-02-07 12:21:23 -0500944 ret = -EINVAL;
945 goto out_unlock;
Josef Bacik02ecd2c2013-10-25 16:19:08 -0400946 }
Chris Mason771ed682008-11-06 22:02:51 -0500947
Qu Wenruofda28322013-02-26 08:10:22 +0000948 num_bytes = ALIGN(end - start + 1, blocksize);
Chris Mason771ed682008-11-06 22:02:51 -0500949 num_bytes = max(blocksize, num_bytes);
950 disk_num_bytes = num_bytes;
Chris Mason771ed682008-11-06 22:02:51 -0500951
Chris Mason4cb53002011-05-24 15:35:30 -0400952 /* if this is a small write inside eof, kick off defrag */
Byongho Leeee221842015-12-15 01:42:10 +0900953 if (num_bytes < SZ_64K &&
Liu Bo4cb13e52012-03-29 09:57:45 -0400954 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
Josef Bacik00361582013-08-14 14:02:47 -0400955 btrfs_add_inode_defrag(NULL, inode);
Chris Mason4cb53002011-05-24 15:35:30 -0400956
Chris Mason771ed682008-11-06 22:02:51 -0500957 if (start == 0) {
958 /* lets try to make an inline extent */
Josef Bacik00361582013-08-14 14:02:47 -0400959 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
960 NULL);
Chris Mason771ed682008-11-06 22:02:51 -0500961 if (ret == 0) {
Josef Bacikc2790a22013-07-29 11:20:47 -0400962 extent_clear_unlock_delalloc(inode, start, end, NULL,
963 EXTENT_LOCKED | EXTENT_DELALLOC |
Josef Bacik151a41b2013-07-29 13:22:24 -0400964 EXTENT_DEFRAG, PAGE_UNLOCK |
Josef Bacikc2790a22013-07-29 11:20:47 -0400965 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
966 PAGE_END_WRITEBACK);
Yan, Zhengc2167752009-11-12 09:34:21 +0000967
Chris Mason771ed682008-11-06 22:02:51 -0500968 *nr_written = *nr_written +
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300969 (end - start + PAGE_SIZE) / PAGE_SIZE;
Chris Mason771ed682008-11-06 22:02:51 -0500970 *page_started = 1;
Chris Mason771ed682008-11-06 22:02:51 -0500971 goto out;
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100972 } else if (ret < 0) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100973 goto out_unlock;
Chris Mason771ed682008-11-06 22:02:51 -0500974 }
975 }
Chris Masonc8b97812008-10-29 14:49:59 -0400976
977 BUG_ON(disk_num_bytes >
David Sterba6c417612011-04-13 15:41:04 +0200978 btrfs_super_total_bytes(root->fs_info->super_copy));
Chris Masonc8b97812008-10-29 14:49:59 -0400979
Josef Bacik4b46fce2010-05-23 11:00:55 -0400980 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400981 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
Chris Mason3b951512008-04-17 11:29:12 -0400982
Chris Masond3977122009-01-05 21:25:51 -0500983 while (disk_num_bytes > 0) {
Chris Masona791e352009-10-08 11:27:10 -0400984 unsigned long op;
985
Josef Bacik287a0ab2010-03-19 18:07:23 +0000986 cur_alloc_size = disk_num_bytes;
Josef Bacik00361582013-08-14 14:02:47 -0400987 ret = btrfs_reserve_extent(root, cur_alloc_size,
Chris Mason771ed682008-11-06 22:02:51 -0500988 root->sectorsize, 0, alloc_hint,
Miao Xiee570fd22014-06-19 10:42:50 +0800989 &ins, 1, 1);
Josef Bacik00361582013-08-14 14:02:47 -0400990 if (ret < 0)
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100991 goto out_unlock;
Chris Masond3977122009-01-05 21:25:51 -0500992
David Sterba172ddd62011-04-21 00:48:27 +0200993 em = alloc_extent_map();
Liu Bob9aa55b2013-05-14 02:12:15 +0000994 if (!em) {
995 ret = -ENOMEM;
Liu Boace68ba2013-04-22 10:53:47 +0000996 goto out_reserve;
Liu Bob9aa55b2013-05-14 02:12:15 +0000997 }
Chris Masone6dcd2d2008-07-17 12:53:50 -0400998 em->start = start;
Chris Mason445a6942008-11-10 11:53:33 -0500999 em->orig_start = em->start;
Chris Mason771ed682008-11-06 22:02:51 -05001000 ram_size = ins.offset;
1001 em->len = ins.offset;
Josef Bacik2ab28f32012-10-12 15:27:49 -04001002 em->mod_start = em->start;
1003 em->mod_len = em->len;
Chris Masonc8b97812008-10-29 14:49:59 -04001004
Chris Masone6dcd2d2008-07-17 12:53:50 -04001005 em->block_start = ins.objectid;
Chris Masonc8b97812008-10-29 14:49:59 -04001006 em->block_len = ins.offset;
Josef Bacikb4939682012-12-03 10:31:19 -05001007 em->orig_block_len = ins.offset;
Josef Bacikcc95bef2013-04-04 14:31:27 -04001008 em->ram_bytes = ram_size;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001009 em->bdev = root->fs_info->fs_devices->latest_bdev;
Chris Mason7f3c74f2008-07-18 12:01:11 -04001010 set_bit(EXTENT_FLAG_PINNED, &em->flags);
Josef Bacik70c8a912012-10-11 16:54:30 -04001011 em->generation = -1;
Chris Masonc8b97812008-10-29 14:49:59 -04001012
Chris Masond3977122009-01-05 21:25:51 -05001013 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -04001014 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -04001015 ret = add_extent_mapping(em_tree, em, 1);
Chris Mason890871b2009-09-02 16:24:52 -04001016 write_unlock(&em_tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001017 if (ret != -EEXIST) {
1018 free_extent_map(em);
1019 break;
1020 }
1021 btrfs_drop_extent_cache(inode, start,
Chris Masonc8b97812008-10-29 14:49:59 -04001022 start + ram_size - 1, 0);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001023 }
Liu Boace68ba2013-04-22 10:53:47 +00001024 if (ret)
1025 goto out_reserve;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001026
Chris Mason98d20f62008-04-14 09:46:10 -04001027 cur_alloc_size = ins.offset;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001028 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
Chris Mason771ed682008-11-06 22:02:51 -05001029 ram_size, cur_alloc_size, 0);
Liu Boace68ba2013-04-22 10:53:47 +00001030 if (ret)
Filipe Mananad9f85962014-08-25 10:43:00 +01001031 goto out_drop_extent_cache;
Chris Masonc8b97812008-10-29 14:49:59 -04001032
Yan Zheng17d217f2008-12-12 10:03:38 -05001033 if (root->root_key.objectid ==
1034 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1035 ret = btrfs_reloc_clone_csums(inode, start,
1036 cur_alloc_size);
Josef Bacik00361582013-08-14 14:02:47 -04001037 if (ret)
Filipe Mananad9f85962014-08-25 10:43:00 +01001038 goto out_drop_extent_cache;
Yan Zheng17d217f2008-12-12 10:03:38 -05001039 }
1040
Chris Masond3977122009-01-05 21:25:51 -05001041 if (disk_num_bytes < cur_alloc_size)
Chris Mason3b951512008-04-17 11:29:12 -04001042 break;
Chris Masond3977122009-01-05 21:25:51 -05001043
Chris Masonc8b97812008-10-29 14:49:59 -04001044 /* we're not doing compressed IO, don't unlock the first
1045 * page (which the caller expects to stay locked), don't
1046 * clear any dirty bits and don't set any writeback bits
Chris Mason8b62b722009-09-02 16:53:46 -04001047 *
1048 * Do set the Private2 bit so we know this page was properly
1049 * setup for writepage
Chris Masonc8b97812008-10-29 14:49:59 -04001050 */
Josef Bacikc2790a22013-07-29 11:20:47 -04001051 op = unlock ? PAGE_UNLOCK : 0;
1052 op |= PAGE_SET_PRIVATE2;
Chris Masona791e352009-10-08 11:27:10 -04001053
Josef Bacikc2790a22013-07-29 11:20:47 -04001054 extent_clear_unlock_delalloc(inode, start,
1055 start + ram_size - 1, locked_page,
1056 EXTENT_LOCKED | EXTENT_DELALLOC,
1057 op);
Chris Masonc8b97812008-10-29 14:49:59 -04001058 disk_num_bytes -= cur_alloc_size;
Chris Masonc59f8952007-12-17 20:14:04 -05001059 num_bytes -= cur_alloc_size;
1060 alloc_hint = ins.objectid + ins.offset;
1061 start += cur_alloc_size;
Chris Masonb888db22007-08-27 16:49:44 -04001062 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001063out:
Chris Masonbe20aa92007-12-17 20:14:01 -05001064 return ret;
Miao Xieb7d5b0a2012-11-01 07:32:18 +00001065
Filipe Mananad9f85962014-08-25 10:43:00 +01001066out_drop_extent_cache:
1067 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
Liu Boace68ba2013-04-22 10:53:47 +00001068out_reserve:
Miao Xiee570fd22014-06-19 10:42:50 +08001069 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001070out_unlock:
Josef Bacikc2790a22013-07-29 11:20:47 -04001071 extent_clear_unlock_delalloc(inode, start, end, locked_page,
Josef Bacik151a41b2013-07-29 13:22:24 -04001072 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1073 EXTENT_DELALLOC | EXTENT_DEFRAG,
1074 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1075 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001076 goto out;
Chris Mason771ed682008-11-06 22:02:51 -05001077}
Chris Masonc8b97812008-10-29 14:49:59 -04001078
Chris Mason771ed682008-11-06 22:02:51 -05001079/*
1080 * work queue call back to started compression on a file and pages
1081 */
1082static noinline void async_cow_start(struct btrfs_work *work)
1083{
1084 struct async_cow *async_cow;
1085 int num_added = 0;
1086 async_cow = container_of(work, struct async_cow, work);
1087
1088 compress_file_range(async_cow->inode, async_cow->locked_page,
1089 async_cow->start, async_cow->end, async_cow,
1090 &num_added);
Josef Bacik8180ef82012-06-08 15:16:12 -04001091 if (num_added == 0) {
Josef Bacikcb77fcd2012-06-15 12:19:48 -06001092 btrfs_add_delayed_iput(async_cow->inode);
Chris Mason771ed682008-11-06 22:02:51 -05001093 async_cow->inode = NULL;
Josef Bacik8180ef82012-06-08 15:16:12 -04001094 }
Chris Mason771ed682008-11-06 22:02:51 -05001095}
1096
1097/*
1098 * work queue call back to submit previously compressed pages
1099 */
1100static noinline void async_cow_submit(struct btrfs_work *work)
1101{
1102 struct async_cow *async_cow;
1103 struct btrfs_root *root;
1104 unsigned long nr_pages;
1105
1106 async_cow = container_of(work, struct async_cow, work);
1107
1108 root = async_cow->root;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001109 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1110 PAGE_SHIFT;
Chris Mason771ed682008-11-06 22:02:51 -05001111
David Sterbaee863952015-02-16 19:41:40 +01001112 /*
1113 * atomic_sub_return implies a barrier for waitqueue_active
1114 */
Josef Bacik66657b32012-08-01 15:36:24 -04001115 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
Byongho Leeee221842015-12-15 01:42:10 +09001116 5 * SZ_1M &&
Chris Mason771ed682008-11-06 22:02:51 -05001117 waitqueue_active(&root->fs_info->async_submit_wait))
1118 wake_up(&root->fs_info->async_submit_wait);
1119
Chris Masond3977122009-01-05 21:25:51 -05001120 if (async_cow->inode)
Chris Mason771ed682008-11-06 22:02:51 -05001121 submit_compressed_extents(async_cow->inode, async_cow);
Chris Mason771ed682008-11-06 22:02:51 -05001122}
Chris Masonc8b97812008-10-29 14:49:59 -04001123
Chris Mason771ed682008-11-06 22:02:51 -05001124static noinline void async_cow_free(struct btrfs_work *work)
1125{
1126 struct async_cow *async_cow;
1127 async_cow = container_of(work, struct async_cow, work);
Josef Bacik8180ef82012-06-08 15:16:12 -04001128 if (async_cow->inode)
Josef Bacikcb77fcd2012-06-15 12:19:48 -06001129 btrfs_add_delayed_iput(async_cow->inode);
Chris Mason771ed682008-11-06 22:02:51 -05001130 kfree(async_cow);
1131}
1132
1133static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1134 u64 start, u64 end, int *page_started,
1135 unsigned long *nr_written)
1136{
1137 struct async_cow *async_cow;
1138 struct btrfs_root *root = BTRFS_I(inode)->root;
1139 unsigned long nr_pages;
1140 u64 cur_end;
Byongho Leeee221842015-12-15 01:42:10 +09001141 int limit = 10 * SZ_1M;
Chris Mason771ed682008-11-06 22:02:51 -05001142
Chris Masona3429ab2009-10-08 12:30:20 -04001143 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1144 1, 0, NULL, GFP_NOFS);
Chris Masond3977122009-01-05 21:25:51 -05001145 while (start < end) {
Chris Mason771ed682008-11-06 22:02:51 -05001146 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001147 BUG_ON(!async_cow); /* -ENOMEM */
Josef Bacik8180ef82012-06-08 15:16:12 -04001148 async_cow->inode = igrab(inode);
Chris Mason771ed682008-11-06 22:02:51 -05001149 async_cow->root = root;
1150 async_cow->locked_page = locked_page;
1151 async_cow->start = start;
1152
Wang Shilongf79707b2014-07-17 11:44:09 +08001153 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1154 !btrfs_test_opt(root, FORCE_COMPRESS))
Chris Mason771ed682008-11-06 22:02:51 -05001155 cur_end = end;
1156 else
Byongho Leeee221842015-12-15 01:42:10 +09001157 cur_end = min(end, start + SZ_512K - 1);
Chris Mason771ed682008-11-06 22:02:51 -05001158
1159 async_cow->end = cur_end;
1160 INIT_LIST_HEAD(&async_cow->extents);
1161
Liu Bo9e0af232014-08-15 23:36:53 +08001162 btrfs_init_work(&async_cow->work,
1163 btrfs_delalloc_helper,
1164 async_cow_start, async_cow_submit,
1165 async_cow_free);
Chris Mason771ed682008-11-06 22:02:51 -05001166
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001167 nr_pages = (cur_end - start + PAGE_SIZE) >>
1168 PAGE_SHIFT;
Chris Mason771ed682008-11-06 22:02:51 -05001169 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1170
Qu Wenruoafe3d242014-02-28 10:46:07 +08001171 btrfs_queue_work(root->fs_info->delalloc_workers,
1172 &async_cow->work);
Chris Mason771ed682008-11-06 22:02:51 -05001173
1174 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1175 wait_event(root->fs_info->async_submit_wait,
1176 (atomic_read(&root->fs_info->async_delalloc_pages) <
1177 limit));
1178 }
1179
Chris Masond3977122009-01-05 21:25:51 -05001180 while (atomic_read(&root->fs_info->async_submit_draining) &&
Chris Mason771ed682008-11-06 22:02:51 -05001181 atomic_read(&root->fs_info->async_delalloc_pages)) {
1182 wait_event(root->fs_info->async_submit_wait,
1183 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1184 0));
1185 }
1186
1187 *nr_written += nr_pages;
1188 start = cur_end + 1;
1189 }
1190 *page_started = 1;
1191 return 0;
Chris Masonbe20aa92007-12-17 20:14:01 -05001192}
1193
Chris Masond3977122009-01-05 21:25:51 -05001194static noinline int csum_exist_in_range(struct btrfs_root *root,
Yan Zheng17d217f2008-12-12 10:03:38 -05001195 u64 bytenr, u64 num_bytes)
1196{
1197 int ret;
1198 struct btrfs_ordered_sum *sums;
1199 LIST_HEAD(list);
1200
Yan Zheng07d400a2009-01-06 11:42:00 -05001201 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
Arne Jansena2de7332011-03-08 14:14:00 +01001202 bytenr + num_bytes - 1, &list, 0);
Yan Zheng17d217f2008-12-12 10:03:38 -05001203 if (ret == 0 && list_empty(&list))
1204 return 0;
1205
1206 while (!list_empty(&list)) {
1207 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1208 list_del(&sums->list);
1209 kfree(sums);
1210 }
1211 return 1;
1212}
1213
Chris Masond352ac62008-09-29 15:18:18 -04001214/*
1215 * when nowcow writeback call back. This checks for snapshots or COW copies
1216 * of the extents that exist in the file, and COWs the file as required.
1217 *
1218 * If no cow copies or snapshots exist, we write directly to the existing
1219 * blocks on disk
1220 */
Chris Mason7f366cf2009-03-12 20:12:45 -04001221static noinline int run_delalloc_nocow(struct inode *inode,
1222 struct page *locked_page,
Chris Mason771ed682008-11-06 22:02:51 -05001223 u64 start, u64 end, int *page_started, int force,
1224 unsigned long *nr_written)
Chris Masonbe20aa92007-12-17 20:14:01 -05001225{
Chris Masonbe20aa92007-12-17 20:14:01 -05001226 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001227 struct btrfs_trans_handle *trans;
Chris Masonbe20aa92007-12-17 20:14:01 -05001228 struct extent_buffer *leaf;
Chris Masonbe20aa92007-12-17 20:14:01 -05001229 struct btrfs_path *path;
Yan Zheng80ff3852008-10-30 14:20:02 -04001230 struct btrfs_file_extent_item *fi;
Chris Masonbe20aa92007-12-17 20:14:01 -05001231 struct btrfs_key found_key;
Yan Zheng80ff3852008-10-30 14:20:02 -04001232 u64 cow_start;
1233 u64 cur_offset;
1234 u64 extent_end;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001235 u64 extent_offset;
Yan Zheng80ff3852008-10-30 14:20:02 -04001236 u64 disk_bytenr;
1237 u64 num_bytes;
Josef Bacikb4939682012-12-03 10:31:19 -05001238 u64 disk_num_bytes;
Josef Bacikcc95bef2013-04-04 14:31:27 -04001239 u64 ram_bytes;
Yan Zheng80ff3852008-10-30 14:20:02 -04001240 int extent_type;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001241 int ret, err;
Yan Zhengd899e052008-10-30 14:25:28 -04001242 int type;
Yan Zheng80ff3852008-10-30 14:20:02 -04001243 int nocow;
1244 int check_prev = 1;
Li Zefan82d59022011-04-20 10:33:24 +08001245 bool nolock;
Li Zefan33345d012011-04-20 10:31:50 +08001246 u64 ino = btrfs_ino(inode);
Chris Masonbe20aa92007-12-17 20:14:01 -05001247
1248 path = btrfs_alloc_path();
Josef Bacik17ca04a2012-05-31 15:58:55 -04001249 if (!path) {
Josef Bacikc2790a22013-07-29 11:20:47 -04001250 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1251 EXTENT_LOCKED | EXTENT_DELALLOC |
Josef Bacik151a41b2013-07-29 13:22:24 -04001252 EXTENT_DO_ACCOUNTING |
1253 EXTENT_DEFRAG, PAGE_UNLOCK |
Josef Bacikc2790a22013-07-29 11:20:47 -04001254 PAGE_CLEAR_DIRTY |
1255 PAGE_SET_WRITEBACK |
1256 PAGE_END_WRITEBACK);
Mark Fashehd8926bb2011-07-13 10:38:47 -07001257 return -ENOMEM;
Josef Bacik17ca04a2012-05-31 15:58:55 -04001258 }
Li Zefan82d59022011-04-20 10:33:24 +08001259
Liu Bo83eea1f2012-07-10 05:28:39 -06001260 nolock = btrfs_is_free_space_inode(inode);
Li Zefan82d59022011-04-20 10:33:24 +08001261
1262 if (nolock)
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001263 trans = btrfs_join_transaction_nolock(root);
Li Zefan82d59022011-04-20 10:33:24 +08001264 else
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001265 trans = btrfs_join_transaction(root);
Chris Masonff5714c2011-05-28 07:00:39 -04001266
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001267 if (IS_ERR(trans)) {
Josef Bacikc2790a22013-07-29 11:20:47 -04001268 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1269 EXTENT_LOCKED | EXTENT_DELALLOC |
Josef Bacik151a41b2013-07-29 13:22:24 -04001270 EXTENT_DO_ACCOUNTING |
1271 EXTENT_DEFRAG, PAGE_UNLOCK |
Josef Bacikc2790a22013-07-29 11:20:47 -04001272 PAGE_CLEAR_DIRTY |
1273 PAGE_SET_WRITEBACK |
1274 PAGE_END_WRITEBACK);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001275 btrfs_free_path(path);
1276 return PTR_ERR(trans);
1277 }
1278
Josef Bacik74b21072011-04-13 12:02:53 -04001279 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
Chris Masonbe20aa92007-12-17 20:14:01 -05001280
Yan Zheng80ff3852008-10-30 14:20:02 -04001281 cow_start = (u64)-1;
1282 cur_offset = start;
1283 while (1) {
Li Zefan33345d012011-04-20 10:31:50 +08001284 ret = btrfs_lookup_file_extent(trans, root, path, ino,
Yan Zheng80ff3852008-10-30 14:20:02 -04001285 cur_offset, 0);
Josef Bacikd788a342013-10-25 16:55:08 -04001286 if (ret < 0)
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001287 goto error;
Yan Zheng80ff3852008-10-30 14:20:02 -04001288 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1289 leaf = path->nodes[0];
1290 btrfs_item_key_to_cpu(leaf, &found_key,
1291 path->slots[0] - 1);
Li Zefan33345d012011-04-20 10:31:50 +08001292 if (found_key.objectid == ino &&
Yan Zheng80ff3852008-10-30 14:20:02 -04001293 found_key.type == BTRFS_EXTENT_DATA_KEY)
1294 path->slots[0]--;
1295 }
1296 check_prev = 0;
1297next_slot:
1298 leaf = path->nodes[0];
1299 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1300 ret = btrfs_next_leaf(root, path);
Josef Bacikd788a342013-10-25 16:55:08 -04001301 if (ret < 0)
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001302 goto error;
Yan Zheng80ff3852008-10-30 14:20:02 -04001303 if (ret > 0)
1304 break;
1305 leaf = path->nodes[0];
1306 }
Chris Masonbe20aa92007-12-17 20:14:01 -05001307
Yan Zheng80ff3852008-10-30 14:20:02 -04001308 nocow = 0;
1309 disk_bytenr = 0;
Yan Zheng17d217f2008-12-12 10:03:38 -05001310 num_bytes = 0;
Yan Zheng80ff3852008-10-30 14:20:02 -04001311 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
Chris Masonbe20aa92007-12-17 20:14:01 -05001312
Filipe Manana1d512cb2015-11-09 00:33:58 +00001313 if (found_key.objectid > ino)
1314 break;
1315 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1316 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1317 path->slots[0]++;
1318 goto next_slot;
1319 }
1320 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
Yan Zheng80ff3852008-10-30 14:20:02 -04001321 found_key.offset > end)
1322 break;
Chris Masonbe20aa92007-12-17 20:14:01 -05001323
Yan Zheng80ff3852008-10-30 14:20:02 -04001324 if (found_key.offset > cur_offset) {
1325 extent_end = found_key.offset;
Chris Masone9061e22009-10-09 09:57:45 -04001326 extent_type = 0;
Yan Zheng80ff3852008-10-30 14:20:02 -04001327 goto out_check;
1328 }
Chris Masonc31f8832008-01-08 15:46:31 -05001329
Yan Zheng80ff3852008-10-30 14:20:02 -04001330 fi = btrfs_item_ptr(leaf, path->slots[0],
1331 struct btrfs_file_extent_item);
1332 extent_type = btrfs_file_extent_type(leaf, fi);
Chris Masonbe20aa92007-12-17 20:14:01 -05001333
Josef Bacikcc95bef2013-04-04 14:31:27 -04001334 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
Yan Zhengd899e052008-10-30 14:25:28 -04001335 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1336 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng80ff3852008-10-30 14:20:02 -04001337 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001338 extent_offset = btrfs_file_extent_offset(leaf, fi);
Yan Zheng80ff3852008-10-30 14:20:02 -04001339 extent_end = found_key.offset +
1340 btrfs_file_extent_num_bytes(leaf, fi);
Josef Bacikb4939682012-12-03 10:31:19 -05001341 disk_num_bytes =
1342 btrfs_file_extent_disk_num_bytes(leaf, fi);
Yan Zheng80ff3852008-10-30 14:20:02 -04001343 if (extent_end <= start) {
1344 path->slots[0]++;
1345 goto next_slot;
1346 }
Yan Zheng17d217f2008-12-12 10:03:38 -05001347 if (disk_bytenr == 0)
1348 goto out_check;
Yan Zheng80ff3852008-10-30 14:20:02 -04001349 if (btrfs_file_extent_compression(leaf, fi) ||
1350 btrfs_file_extent_encryption(leaf, fi) ||
1351 btrfs_file_extent_other_encoding(leaf, fi))
1352 goto out_check;
Yan Zhengd899e052008-10-30 14:25:28 -04001353 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1354 goto out_check;
Yan Zhengd2fb3432008-12-11 16:30:39 -05001355 if (btrfs_extent_readonly(root, disk_bytenr))
Yan Zheng80ff3852008-10-30 14:20:02 -04001356 goto out_check;
Li Zefan33345d012011-04-20 10:31:50 +08001357 if (btrfs_cross_ref_exist(trans, root, ino,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001358 found_key.offset -
1359 extent_offset, disk_bytenr))
Yan Zheng17d217f2008-12-12 10:03:38 -05001360 goto out_check;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001361 disk_bytenr += extent_offset;
Yan Zheng17d217f2008-12-12 10:03:38 -05001362 disk_bytenr += cur_offset - found_key.offset;
1363 num_bytes = min(end + 1, extent_end) - cur_offset;
1364 /*
Wang Shilonge9894fd2014-03-27 11:12:25 +08001365 * if there are pending snapshots for this root,
1366 * we fall into common COW way.
1367 */
1368 if (!nolock) {
Filipe Manana9ea24bb2014-10-29 11:57:59 +00001369 err = btrfs_start_write_no_snapshoting(root);
Wang Shilonge9894fd2014-03-27 11:12:25 +08001370 if (!err)
1371 goto out_check;
1372 }
1373 /*
Yan Zheng17d217f2008-12-12 10:03:38 -05001374 * force cow if csum exists in the range.
1375 * this ensure that csum for a given extent are
1376 * either valid or do not exist.
1377 */
1378 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1379 goto out_check;
Yan Zheng80ff3852008-10-30 14:20:02 -04001380 nocow = 1;
1381 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1382 extent_end = found_key.offset +
Chris Mason514ac8a2014-01-03 21:07:00 -08001383 btrfs_file_extent_inline_len(leaf,
1384 path->slots[0], fi);
Yan Zheng80ff3852008-10-30 14:20:02 -04001385 extent_end = ALIGN(extent_end, root->sectorsize);
1386 } else {
1387 BUG_ON(1);
1388 }
1389out_check:
1390 if (extent_end <= start) {
1391 path->slots[0]++;
Wang Shilonge9894fd2014-03-27 11:12:25 +08001392 if (!nolock && nocow)
Filipe Manana9ea24bb2014-10-29 11:57:59 +00001393 btrfs_end_write_no_snapshoting(root);
Yan Zheng80ff3852008-10-30 14:20:02 -04001394 goto next_slot;
1395 }
1396 if (!nocow) {
1397 if (cow_start == (u64)-1)
1398 cow_start = cur_offset;
1399 cur_offset = extent_end;
1400 if (cur_offset > end)
1401 break;
1402 path->slots[0]++;
1403 goto next_slot;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001404 }
Chris Masonbe20aa92007-12-17 20:14:01 -05001405
David Sterbab3b4aa72011-04-21 01:20:15 +02001406 btrfs_release_path(path);
Yan Zheng80ff3852008-10-30 14:20:02 -04001407 if (cow_start != (u64)-1) {
Josef Bacik00361582013-08-14 14:02:47 -04001408 ret = cow_file_range(inode, locked_page,
1409 cow_start, found_key.offset - 1,
1410 page_started, nr_written, 1);
Wang Shilonge9894fd2014-03-27 11:12:25 +08001411 if (ret) {
1412 if (!nolock && nocow)
Filipe Manana9ea24bb2014-10-29 11:57:59 +00001413 btrfs_end_write_no_snapshoting(root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001414 goto error;
Wang Shilonge9894fd2014-03-27 11:12:25 +08001415 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001416 cow_start = (u64)-1;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001417 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001418
Yan Zhengd899e052008-10-30 14:25:28 -04001419 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1420 struct extent_map *em;
1421 struct extent_map_tree *em_tree;
1422 em_tree = &BTRFS_I(inode)->extent_tree;
David Sterba172ddd62011-04-21 00:48:27 +02001423 em = alloc_extent_map();
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001424 BUG_ON(!em); /* -ENOMEM */
Yan Zhengd899e052008-10-30 14:25:28 -04001425 em->start = cur_offset;
Josef Bacik70c8a912012-10-11 16:54:30 -04001426 em->orig_start = found_key.offset - extent_offset;
Yan Zhengd899e052008-10-30 14:25:28 -04001427 em->len = num_bytes;
1428 em->block_len = num_bytes;
1429 em->block_start = disk_bytenr;
Josef Bacikb4939682012-12-03 10:31:19 -05001430 em->orig_block_len = disk_num_bytes;
Josef Bacikcc95bef2013-04-04 14:31:27 -04001431 em->ram_bytes = ram_bytes;
Yan Zhengd899e052008-10-30 14:25:28 -04001432 em->bdev = root->fs_info->fs_devices->latest_bdev;
Josef Bacik2ab28f32012-10-12 15:27:49 -04001433 em->mod_start = em->start;
1434 em->mod_len = em->len;
Yan Zhengd899e052008-10-30 14:25:28 -04001435 set_bit(EXTENT_FLAG_PINNED, &em->flags);
Josef Bacikb11e2342012-12-03 10:58:15 -05001436 set_bit(EXTENT_FLAG_FILLING, &em->flags);
Josef Bacik70c8a912012-10-11 16:54:30 -04001437 em->generation = -1;
Yan Zhengd899e052008-10-30 14:25:28 -04001438 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -04001439 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -04001440 ret = add_extent_mapping(em_tree, em, 1);
Chris Mason890871b2009-09-02 16:24:52 -04001441 write_unlock(&em_tree->lock);
Yan Zhengd899e052008-10-30 14:25:28 -04001442 if (ret != -EEXIST) {
1443 free_extent_map(em);
1444 break;
1445 }
1446 btrfs_drop_extent_cache(inode, em->start,
1447 em->start + em->len - 1, 0);
1448 }
1449 type = BTRFS_ORDERED_PREALLOC;
1450 } else {
1451 type = BTRFS_ORDERED_NOCOW;
1452 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001453
1454 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
Yan Zhengd899e052008-10-30 14:25:28 -04001455 num_bytes, num_bytes, type);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001456 BUG_ON(ret); /* -ENOMEM */
Chris Mason771ed682008-11-06 22:02:51 -05001457
Yan, Zhengefa56462010-05-16 10:49:59 -04001458 if (root->root_key.objectid ==
1459 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1460 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1461 num_bytes);
Wang Shilonge9894fd2014-03-27 11:12:25 +08001462 if (ret) {
1463 if (!nolock && nocow)
Filipe Manana9ea24bb2014-10-29 11:57:59 +00001464 btrfs_end_write_no_snapshoting(root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001465 goto error;
Wang Shilonge9894fd2014-03-27 11:12:25 +08001466 }
Yan, Zhengefa56462010-05-16 10:49:59 -04001467 }
1468
Josef Bacikc2790a22013-07-29 11:20:47 -04001469 extent_clear_unlock_delalloc(inode, cur_offset,
1470 cur_offset + num_bytes - 1,
1471 locked_page, EXTENT_LOCKED |
1472 EXTENT_DELALLOC, PAGE_UNLOCK |
1473 PAGE_SET_PRIVATE2);
Wang Shilonge9894fd2014-03-27 11:12:25 +08001474 if (!nolock && nocow)
Filipe Manana9ea24bb2014-10-29 11:57:59 +00001475 btrfs_end_write_no_snapshoting(root);
Yan Zheng80ff3852008-10-30 14:20:02 -04001476 cur_offset = extent_end;
1477 if (cur_offset > end)
1478 break;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001479 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001480 btrfs_release_path(path);
Yan Zheng80ff3852008-10-30 14:20:02 -04001481
Josef Bacik17ca04a2012-05-31 15:58:55 -04001482 if (cur_offset <= end && cow_start == (u64)-1) {
Yan Zheng80ff3852008-10-30 14:20:02 -04001483 cow_start = cur_offset;
Josef Bacik17ca04a2012-05-31 15:58:55 -04001484 cur_offset = end;
1485 }
1486
Yan Zheng80ff3852008-10-30 14:20:02 -04001487 if (cow_start != (u64)-1) {
Josef Bacik00361582013-08-14 14:02:47 -04001488 ret = cow_file_range(inode, locked_page, cow_start, end,
1489 page_started, nr_written, 1);
Josef Bacikd788a342013-10-25 16:55:08 -04001490 if (ret)
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001491 goto error;
Yan Zheng80ff3852008-10-30 14:20:02 -04001492 }
1493
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001494error:
Miao Xiea698d0752012-09-20 01:51:59 -06001495 err = btrfs_end_transaction(trans, root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001496 if (!ret)
1497 ret = err;
1498
Josef Bacik17ca04a2012-05-31 15:58:55 -04001499 if (ret && cur_offset < end)
Josef Bacikc2790a22013-07-29 11:20:47 -04001500 extent_clear_unlock_delalloc(inode, cur_offset, end,
1501 locked_page, EXTENT_LOCKED |
Josef Bacik151a41b2013-07-29 13:22:24 -04001502 EXTENT_DELALLOC | EXTENT_DEFRAG |
1503 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1504 PAGE_CLEAR_DIRTY |
Josef Bacikc2790a22013-07-29 11:20:47 -04001505 PAGE_SET_WRITEBACK |
1506 PAGE_END_WRITEBACK);
Yan Zheng7ea394f2008-08-05 13:05:02 -04001507 btrfs_free_path(path);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001508 return ret;
Chris Masonbe20aa92007-12-17 20:14:01 -05001509}
1510
Wang Shilong47059d92014-07-03 18:22:07 +08001511static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1512{
1513
1514 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1515 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1516 return 0;
1517
1518 /*
1519 * @defrag_bytes is a hint value, no spinlock held here,
1520 * if is not zero, it means the file is defragging.
1521 * Force cow if given extent needs to be defragged.
1522 */
1523 if (BTRFS_I(inode)->defrag_bytes &&
1524 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1525 EXTENT_DEFRAG, 0, NULL))
1526 return 1;
1527
1528 return 0;
1529}
1530
Chris Masond352ac62008-09-29 15:18:18 -04001531/*
1532 * extent_io.c call back to do delayed allocation processing
1533 */
Chris Masonc8b97812008-10-29 14:49:59 -04001534static int run_delalloc_range(struct inode *inode, struct page *locked_page,
Chris Mason771ed682008-11-06 22:02:51 -05001535 u64 start, u64 end, int *page_started,
1536 unsigned long *nr_written)
Chris Masonbe20aa92007-12-17 20:14:01 -05001537{
Chris Masonbe20aa92007-12-17 20:14:01 -05001538 int ret;
Wang Shilong47059d92014-07-03 18:22:07 +08001539 int force_cow = need_force_cow(inode, start, end);
Chris Masona2135012008-06-25 16:01:30 -04001540
Wang Shilong47059d92014-07-03 18:22:07 +08001541 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
Chris Masonc8b97812008-10-29 14:49:59 -04001542 ret = run_delalloc_nocow(inode, locked_page, start, end,
Chris Masond3977122009-01-05 21:25:51 -05001543 page_started, 1, nr_written);
Wang Shilong47059d92014-07-03 18:22:07 +08001544 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
Yan Zhengd899e052008-10-30 14:25:28 -04001545 ret = run_delalloc_nocow(inode, locked_page, start, end,
Chris Masond3977122009-01-05 21:25:51 -05001546 page_started, 0, nr_written);
Wang Shilong78160302014-07-17 11:44:10 +08001547 } else if (!inode_need_compress(inode)) {
Chris Mason7f366cf2009-03-12 20:12:45 -04001548 ret = cow_file_range(inode, locked_page, start, end,
1549 page_started, nr_written, 1);
Josef Bacik7ddf5a42012-06-08 15:26:47 -04001550 } else {
1551 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1552 &BTRFS_I(inode)->runtime_flags);
Chris Mason771ed682008-11-06 22:02:51 -05001553 ret = cow_file_range_async(inode, locked_page, start, end,
Chris Masond3977122009-01-05 21:25:51 -05001554 page_started, nr_written);
Josef Bacik7ddf5a42012-06-08 15:26:47 -04001555 }
Chris Masonb888db22007-08-27 16:49:44 -04001556 return ret;
1557}
1558
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001559static void btrfs_split_extent_hook(struct inode *inode,
1560 struct extent_state *orig, u64 split)
Josef Bacik9ed74f22009-09-11 16:12:44 -04001561{
Josef Bacikdcab6a32015-02-11 15:08:59 -05001562 u64 size;
1563
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001564 /* not delalloc, ignore it */
Josef Bacik9ed74f22009-09-11 16:12:44 -04001565 if (!(orig->state & EXTENT_DELALLOC))
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001566 return;
Josef Bacik9ed74f22009-09-11 16:12:44 -04001567
Josef Bacikdcab6a32015-02-11 15:08:59 -05001568 size = orig->end - orig->start + 1;
1569 if (size > BTRFS_MAX_EXTENT_SIZE) {
1570 u64 num_extents;
1571 u64 new_size;
1572
1573 /*
Josef Bacikba117212015-03-13 15:01:24 -04001574 * See the explanation in btrfs_merge_extent_hook, the same
1575 * applies here, just in reverse.
Josef Bacikdcab6a32015-02-11 15:08:59 -05001576 */
1577 new_size = orig->end - split + 1;
Josef Bacikba117212015-03-13 15:01:24 -04001578 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
Josef Bacikdcab6a32015-02-11 15:08:59 -05001579 BTRFS_MAX_EXTENT_SIZE);
Josef Bacikba117212015-03-13 15:01:24 -04001580 new_size = split - orig->start;
1581 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1582 BTRFS_MAX_EXTENT_SIZE);
1583 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1584 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
Josef Bacikdcab6a32015-02-11 15:08:59 -05001585 return;
1586 }
1587
Josef Bacik9e0baf62011-07-15 15:16:44 +00001588 spin_lock(&BTRFS_I(inode)->lock);
1589 BTRFS_I(inode)->outstanding_extents++;
1590 spin_unlock(&BTRFS_I(inode)->lock);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001591}
1592
1593/*
1594 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1595 * extents so we can keep track of new extents that are just merged onto old
1596 * extents, such as when we are doing sequential writes, so we can properly
1597 * account for the metadata space we'll need.
1598 */
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001599static void btrfs_merge_extent_hook(struct inode *inode,
1600 struct extent_state *new,
1601 struct extent_state *other)
Josef Bacik9ed74f22009-09-11 16:12:44 -04001602{
Josef Bacikdcab6a32015-02-11 15:08:59 -05001603 u64 new_size, old_size;
1604 u64 num_extents;
1605
Josef Bacik9ed74f22009-09-11 16:12:44 -04001606 /* not delalloc, ignore it */
1607 if (!(other->state & EXTENT_DELALLOC))
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001608 return;
Josef Bacik9ed74f22009-09-11 16:12:44 -04001609
Josef Bacik8461a3d2015-03-13 15:12:08 -04001610 if (new->start > other->start)
1611 new_size = new->end - other->start + 1;
1612 else
1613 new_size = other->end - new->start + 1;
Josef Bacikdcab6a32015-02-11 15:08:59 -05001614
1615 /* we're not bigger than the max, unreserve the space and go */
1616 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1617 spin_lock(&BTRFS_I(inode)->lock);
1618 BTRFS_I(inode)->outstanding_extents--;
1619 spin_unlock(&BTRFS_I(inode)->lock);
1620 return;
1621 }
1622
1623 /*
Josef Bacikba117212015-03-13 15:01:24 -04001624 * We have to add up either side to figure out how many extents were
1625 * accounted for before we merged into one big extent. If the number of
1626 * extents we accounted for is <= the amount we need for the new range
1627 * then we can return, otherwise drop. Think of it like this
1628 *
1629 * [ 4k][MAX_SIZE]
1630 *
1631 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1632 * need 2 outstanding extents, on one side we have 1 and the other side
1633 * we have 1 so they are == and we can return. But in this case
1634 *
1635 * [MAX_SIZE+4k][MAX_SIZE+4k]
1636 *
1637 * Each range on their own accounts for 2 extents, but merged together
1638 * they are only 3 extents worth of accounting, so we need to drop in
1639 * this case.
Josef Bacikdcab6a32015-02-11 15:08:59 -05001640 */
Josef Bacikba117212015-03-13 15:01:24 -04001641 old_size = other->end - other->start + 1;
Josef Bacikdcab6a32015-02-11 15:08:59 -05001642 num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1643 BTRFS_MAX_EXTENT_SIZE);
Josef Bacikba117212015-03-13 15:01:24 -04001644 old_size = new->end - new->start + 1;
1645 num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1646 BTRFS_MAX_EXTENT_SIZE);
1647
Josef Bacikdcab6a32015-02-11 15:08:59 -05001648 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
Josef Bacikba117212015-03-13 15:01:24 -04001649 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
Josef Bacikdcab6a32015-02-11 15:08:59 -05001650 return;
1651
Josef Bacik9e0baf62011-07-15 15:16:44 +00001652 spin_lock(&BTRFS_I(inode)->lock);
1653 BTRFS_I(inode)->outstanding_extents--;
1654 spin_unlock(&BTRFS_I(inode)->lock);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001655}
1656
Miao Xieeb73c1b2013-05-15 07:48:22 +00001657static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1658 struct inode *inode)
1659{
1660 spin_lock(&root->delalloc_lock);
1661 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1662 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1663 &root->delalloc_inodes);
1664 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1665 &BTRFS_I(inode)->runtime_flags);
1666 root->nr_delalloc_inodes++;
1667 if (root->nr_delalloc_inodes == 1) {
1668 spin_lock(&root->fs_info->delalloc_root_lock);
1669 BUG_ON(!list_empty(&root->delalloc_root));
1670 list_add_tail(&root->delalloc_root,
1671 &root->fs_info->delalloc_roots);
1672 spin_unlock(&root->fs_info->delalloc_root_lock);
1673 }
1674 }
1675 spin_unlock(&root->delalloc_lock);
1676}
1677
1678static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1679 struct inode *inode)
1680{
1681 spin_lock(&root->delalloc_lock);
1682 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1683 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1684 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1685 &BTRFS_I(inode)->runtime_flags);
1686 root->nr_delalloc_inodes--;
1687 if (!root->nr_delalloc_inodes) {
1688 spin_lock(&root->fs_info->delalloc_root_lock);
1689 BUG_ON(list_empty(&root->delalloc_root));
1690 list_del_init(&root->delalloc_root);
1691 spin_unlock(&root->fs_info->delalloc_root_lock);
1692 }
1693 }
1694 spin_unlock(&root->delalloc_lock);
1695}
1696
Chris Masond352ac62008-09-29 15:18:18 -04001697/*
1698 * extent_io.c set_bit_hook, used to track delayed allocation
1699 * bytes in this file, and to maintain the list of inodes that
1700 * have pending delalloc work to be done.
1701 */
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001702static void btrfs_set_bit_hook(struct inode *inode,
David Sterba9ee49a042015-01-14 19:52:13 +01001703 struct extent_state *state, unsigned *bits)
Chris Mason291d6732008-01-29 15:55:23 -05001704{
Josef Bacik9ed74f22009-09-11 16:12:44 -04001705
Wang Shilong47059d92014-07-03 18:22:07 +08001706 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1707 WARN_ON(1);
Chris Mason75eff682008-12-15 15:54:40 -05001708 /*
1709 * set_bit and clear bit hooks normally require _irqsave/restore
Sergei Trofimovich27160b62011-05-20 20:20:32 +00001710 * but in this case, we are only testing for the DELALLOC
Chris Mason75eff682008-12-15 15:54:40 -05001711 * bit, which is only set or cleared with irqs on
1712 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001713 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
Chris Mason291d6732008-01-29 15:55:23 -05001714 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001715 u64 len = state->end + 1 - state->start;
Liu Bo83eea1f2012-07-10 05:28:39 -06001716 bool do_list = !btrfs_is_free_space_inode(inode);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001717
Josef Bacik9e0baf62011-07-15 15:16:44 +00001718 if (*bits & EXTENT_FIRST_DELALLOC) {
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001719 *bits &= ~EXTENT_FIRST_DELALLOC;
Josef Bacik9e0baf62011-07-15 15:16:44 +00001720 } else {
1721 spin_lock(&BTRFS_I(inode)->lock);
1722 BTRFS_I(inode)->outstanding_extents++;
1723 spin_unlock(&BTRFS_I(inode)->lock);
1724 }
Josef Bacik287a0ab2010-03-19 18:07:23 +00001725
Josef Bacik6a3891c2015-03-16 17:38:52 -04001726 /* For sanity tests */
1727 if (btrfs_test_is_dummy_root(root))
1728 return;
1729
Miao Xie963d6782013-01-29 10:10:51 +00001730 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1731 root->fs_info->delalloc_batch);
Miao Xiedf0af1a2013-01-29 10:11:59 +00001732 spin_lock(&BTRFS_I(inode)->lock);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001733 BTRFS_I(inode)->delalloc_bytes += len;
Wang Shilong47059d92014-07-03 18:22:07 +08001734 if (*bits & EXTENT_DEFRAG)
1735 BTRFS_I(inode)->defrag_bytes += len;
Miao Xiedf0af1a2013-01-29 10:11:59 +00001736 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
Miao Xieeb73c1b2013-05-15 07:48:22 +00001737 &BTRFS_I(inode)->runtime_flags))
1738 btrfs_add_delalloc_inodes(root, inode);
Miao Xiedf0af1a2013-01-29 10:11:59 +00001739 spin_unlock(&BTRFS_I(inode)->lock);
Chris Mason291d6732008-01-29 15:55:23 -05001740 }
Chris Mason291d6732008-01-29 15:55:23 -05001741}
1742
Chris Masond352ac62008-09-29 15:18:18 -04001743/*
1744 * extent_io.c clear_bit_hook, see set_bit_hook for why
1745 */
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001746static void btrfs_clear_bit_hook(struct inode *inode,
David Sterba41074882013-04-29 13:38:46 +00001747 struct extent_state *state,
David Sterba9ee49a042015-01-14 19:52:13 +01001748 unsigned *bits)
Chris Mason291d6732008-01-29 15:55:23 -05001749{
Wang Shilong47059d92014-07-03 18:22:07 +08001750 u64 len = state->end + 1 - state->start;
Josef Bacikdcab6a32015-02-11 15:08:59 -05001751 u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1752 BTRFS_MAX_EXTENT_SIZE);
Wang Shilong47059d92014-07-03 18:22:07 +08001753
1754 spin_lock(&BTRFS_I(inode)->lock);
1755 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1756 BTRFS_I(inode)->defrag_bytes -= len;
1757 spin_unlock(&BTRFS_I(inode)->lock);
1758
Chris Mason75eff682008-12-15 15:54:40 -05001759 /*
1760 * set_bit and clear bit hooks normally require _irqsave/restore
Sergei Trofimovich27160b62011-05-20 20:20:32 +00001761 * but in this case, we are only testing for the DELALLOC
Chris Mason75eff682008-12-15 15:54:40 -05001762 * bit, which is only set or cleared with irqs on
1763 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001764 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
Chris Mason291d6732008-01-29 15:55:23 -05001765 struct btrfs_root *root = BTRFS_I(inode)->root;
Liu Bo83eea1f2012-07-10 05:28:39 -06001766 bool do_list = !btrfs_is_free_space_inode(inode);
Chris Masonbcbfce82008-04-22 13:26:47 -04001767
Josef Bacik9e0baf62011-07-15 15:16:44 +00001768 if (*bits & EXTENT_FIRST_DELALLOC) {
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001769 *bits &= ~EXTENT_FIRST_DELALLOC;
Josef Bacik9e0baf62011-07-15 15:16:44 +00001770 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1771 spin_lock(&BTRFS_I(inode)->lock);
Josef Bacikdcab6a32015-02-11 15:08:59 -05001772 BTRFS_I(inode)->outstanding_extents -= num_extents;
Josef Bacik9e0baf62011-07-15 15:16:44 +00001773 spin_unlock(&BTRFS_I(inode)->lock);
1774 }
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001775
Josef Bacikb6d08f02013-09-27 14:57:43 -04001776 /*
1777 * We don't reserve metadata space for space cache inodes so we
1778 * don't need to call dellalloc_release_metadata if there is an
1779 * error.
1780 */
1781 if (*bits & EXTENT_DO_ACCOUNTING &&
1782 root != root->fs_info->tree_root)
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001783 btrfs_delalloc_release_metadata(inode, len);
1784
Josef Bacik6a3891c2015-03-16 17:38:52 -04001785 /* For sanity tests. */
1786 if (btrfs_test_is_dummy_root(root))
1787 return;
1788
Josef Bacik0cb59c92010-07-02 12:14:14 -04001789 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
Josef Bacik7ee9e442013-06-21 16:37:03 -04001790 && do_list && !(state->state & EXTENT_NORESERVE))
Qu Wenruo51773be2015-10-08 18:19:37 +08001791 btrfs_free_reserved_data_space_noquota(inode,
1792 state->start, len);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001793
Miao Xie963d6782013-01-29 10:10:51 +00001794 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1795 root->fs_info->delalloc_batch);
Miao Xiedf0af1a2013-01-29 10:11:59 +00001796 spin_lock(&BTRFS_I(inode)->lock);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001797 BTRFS_I(inode)->delalloc_bytes -= len;
Josef Bacik0cb59c92010-07-02 12:14:14 -04001798 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
Miao Xiedf0af1a2013-01-29 10:11:59 +00001799 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
Miao Xieeb73c1b2013-05-15 07:48:22 +00001800 &BTRFS_I(inode)->runtime_flags))
1801 btrfs_del_delalloc_inode(root, inode);
Miao Xiedf0af1a2013-01-29 10:11:59 +00001802 spin_unlock(&BTRFS_I(inode)->lock);
Chris Mason291d6732008-01-29 15:55:23 -05001803 }
Chris Mason291d6732008-01-29 15:55:23 -05001804}
1805
Chris Masond352ac62008-09-29 15:18:18 -04001806/*
1807 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1808 * we don't create bios that span stripes or chunks
1809 */
David Woodhouse64a16702009-07-15 23:29:37 +01001810int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
Chris Masonc8b97812008-10-29 14:49:59 -04001811 size_t size, struct bio *bio,
1812 unsigned long bio_flags)
Chris Mason239b14b2008-03-24 15:02:07 -04001813{
1814 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001815 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
Chris Mason239b14b2008-03-24 15:02:07 -04001816 u64 length = 0;
1817 u64 map_length;
Chris Mason239b14b2008-03-24 15:02:07 -04001818 int ret;
1819
Chris Mason771ed682008-11-06 22:02:51 -05001820 if (bio_flags & EXTENT_BIO_COMPRESSED)
1821 return 0;
1822
Kent Overstreet4f024f32013-10-11 15:44:27 -07001823 length = bio->bi_iter.bi_size;
Chris Mason239b14b2008-03-24 15:02:07 -04001824 map_length = length;
David Woodhouse64a16702009-07-15 23:29:37 +01001825 ret = btrfs_map_block(root->fs_info, rw, logical,
Chris Masonf1885912008-04-09 16:28:12 -04001826 &map_length, NULL, 0);
Stefan Behrens3ec706c2012-11-05 15:46:42 +01001827 /* Will always return 0 with map_multi == NULL */
Jeff Mahoney3444a972011-10-03 23:23:13 -04001828 BUG_ON(ret < 0);
Chris Masond3977122009-01-05 21:25:51 -05001829 if (map_length < length + size)
Chris Mason239b14b2008-03-24 15:02:07 -04001830 return 1;
Jeff Mahoney3444a972011-10-03 23:23:13 -04001831 return 0;
Chris Mason239b14b2008-03-24 15:02:07 -04001832}
1833
Chris Masond352ac62008-09-29 15:18:18 -04001834/*
1835 * in order to insert checksums into the metadata in large chunks,
1836 * we wait until bio submission time. All the pages in the bio are
1837 * checksummed and sums are attached onto the ordered extent record.
1838 *
1839 * At IO completion time the cums attached on the ordered extent record
1840 * are inserted into the btree
1841 */
Chris Masond3977122009-01-05 21:25:51 -05001842static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1843 struct bio *bio, int mirror_num,
Chris Masoneaf25d92010-05-25 09:48:28 -04001844 unsigned long bio_flags,
1845 u64 bio_offset)
Chris Mason065631f2008-02-20 12:07:25 -05001846{
Chris Mason065631f2008-02-20 12:07:25 -05001847 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason065631f2008-02-20 12:07:25 -05001848 int ret = 0;
Chris Masone0156402008-04-16 11:15:20 -04001849
Chris Masond20f7042008-12-08 16:58:54 -05001850 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001851 BUG_ON(ret); /* -ENOMEM */
Chris Mason4a69a412008-11-06 22:03:00 -05001852 return 0;
1853}
Chris Masone0156402008-04-16 11:15:20 -04001854
Chris Mason4a69a412008-11-06 22:03:00 -05001855/*
1856 * in order to insert checksums into the metadata in large chunks,
1857 * we wait until bio submission time. All the pages in the bio are
1858 * checksummed and sums are attached onto the ordered extent record.
1859 *
1860 * At IO completion time the cums attached on the ordered extent record
1861 * are inserted into the btree
1862 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001863static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
Chris Masoneaf25d92010-05-25 09:48:28 -04001864 int mirror_num, unsigned long bio_flags,
1865 u64 bio_offset)
Chris Mason4a69a412008-11-06 22:03:00 -05001866{
1867 struct btrfs_root *root = BTRFS_I(inode)->root;
Stefan Behrens61891922012-11-05 18:51:52 +01001868 int ret;
1869
1870 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001871 if (ret) {
1872 bio->bi_error = ret;
1873 bio_endio(bio);
1874 }
Stefan Behrens61891922012-11-05 18:51:52 +01001875 return ret;
Chris Mason44b8bd72008-04-16 11:14:51 -04001876}
1877
Chris Masond352ac62008-09-29 15:18:18 -04001878/*
Chris Masoncad321a2008-12-17 14:51:42 -05001879 * extent_io.c submission hook. This does the right thing for csum calculation
1880 * on write, or reading the csums from the tree before a read
Chris Masond352ac62008-09-29 15:18:18 -04001881 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001882static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
Chris Masoneaf25d92010-05-25 09:48:28 -04001883 int mirror_num, unsigned long bio_flags,
1884 u64 bio_offset)
Chris Mason44b8bd72008-04-16 11:14:51 -04001885{
1886 struct btrfs_root *root = BTRFS_I(inode)->root;
Chandan Rajendra0d51e282015-07-27 15:26:43 +05301887 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
Chris Mason44b8bd72008-04-16 11:14:51 -04001888 int ret = 0;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001889 int skip_sum;
Josef Bacikb812ce22012-11-16 13:56:32 -05001890 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
Chris Mason44b8bd72008-04-16 11:14:51 -04001891
Christoph Hellwig6cbff002009-04-17 10:37:41 +02001892 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
Chris Masoncad321a2008-12-17 14:51:42 -05001893
Liu Bo83eea1f2012-07-10 05:28:39 -06001894 if (btrfs_is_free_space_inode(inode))
Chandan Rajendra0d51e282015-07-27 15:26:43 +05301895 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
Jeff Mahoney04173412011-10-03 23:23:12 -04001896
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001897 if (!(rw & REQ_WRITE)) {
Josef Bacik5fd02042012-05-02 14:00:54 -04001898 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1899 if (ret)
Stefan Behrens61891922012-11-05 18:51:52 +01001900 goto out;
Josef Bacik5fd02042012-05-02 14:00:54 -04001901
Chris Masond20f7042008-12-08 16:58:54 -05001902 if (bio_flags & EXTENT_BIO_COMPRESSED) {
Stefan Behrens61891922012-11-05 18:51:52 +01001903 ret = btrfs_submit_compressed_read(inode, bio,
1904 mirror_num,
1905 bio_flags);
1906 goto out;
Tsutomu Itohc2db1072011-03-01 06:48:31 +00001907 } else if (!skip_sum) {
1908 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1909 if (ret)
Stefan Behrens61891922012-11-05 18:51:52 +01001910 goto out;
Tsutomu Itohc2db1072011-03-01 06:48:31 +00001911 }
Chris Mason4d1b5fb2008-08-20 09:44:52 -04001912 goto mapit;
Josef Bacikb812ce22012-11-16 13:56:32 -05001913 } else if (async && !skip_sum) {
Yan Zheng17d217f2008-12-12 10:03:38 -05001914 /* csum items have already been cloned */
1915 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1916 goto mapit;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001917 /* we're doing a write, do the async checksumming */
Stefan Behrens61891922012-11-05 18:51:52 +01001918 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
Chris Mason44b8bd72008-04-16 11:14:51 -04001919 inode, rw, bio, mirror_num,
Chris Masoneaf25d92010-05-25 09:48:28 -04001920 bio_flags, bio_offset,
1921 __btrfs_submit_bio_start,
Chris Mason4a69a412008-11-06 22:03:00 -05001922 __btrfs_submit_bio_done);
Stefan Behrens61891922012-11-05 18:51:52 +01001923 goto out;
Josef Bacikb812ce22012-11-16 13:56:32 -05001924 } else if (!skip_sum) {
1925 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1926 if (ret)
1927 goto out;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001928 }
1929
Chris Mason0b86a832008-03-24 15:01:56 -04001930mapit:
Stefan Behrens61891922012-11-05 18:51:52 +01001931 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1932
1933out:
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001934 if (ret < 0) {
1935 bio->bi_error = ret;
1936 bio_endio(bio);
1937 }
Stefan Behrens61891922012-11-05 18:51:52 +01001938 return ret;
Chris Mason065631f2008-02-20 12:07:25 -05001939}
Chris Mason6885f302008-02-20 16:11:05 -05001940
Chris Masond352ac62008-09-29 15:18:18 -04001941/*
1942 * given a list of ordered sums record them in the inode. This happens
1943 * at IO completion time based on sums calculated at bio submission time.
1944 */
Chris Masonba1da2f2008-07-17 12:54:15 -04001945static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
Chris Masone6dcd2d2008-07-17 12:53:50 -04001946 struct inode *inode, u64 file_offset,
1947 struct list_head *list)
1948{
Chris Masone6dcd2d2008-07-17 12:53:50 -04001949 struct btrfs_ordered_sum *sum;
1950
Qinghuang Fengc6e30872009-01-21 10:59:08 -05001951 list_for_each_entry(sum, list, list) {
Miao Xie39847c42013-03-28 08:08:20 +00001952 trans->adding_csums = 1;
Chris Masond20f7042008-12-08 16:58:54 -05001953 btrfs_csum_file_blocks(trans,
1954 BTRFS_I(inode)->root->fs_info->csum_root, sum);
Miao Xie39847c42013-03-28 08:08:20 +00001955 trans->adding_csums = 0;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001956 }
1957 return 0;
1958}
1959
Josef Bacik2ac55d42010-02-03 19:33:23 +00001960int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1961 struct extent_state **cached_state)
Chris Masonea8c2812008-08-04 23:17:27 -04001962{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001963 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
Chris Masonea8c2812008-08-04 23:17:27 -04001964 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
Josef Bacik2ac55d42010-02-03 19:33:23 +00001965 cached_state, GFP_NOFS);
Chris Masonea8c2812008-08-04 23:17:27 -04001966}
1967
Chris Masond352ac62008-09-29 15:18:18 -04001968/* see btrfs_writepage_start_hook for details on why this is required */
Chris Mason247e7432008-07-17 12:53:51 -04001969struct btrfs_writepage_fixup {
1970 struct page *page;
1971 struct btrfs_work work;
1972};
1973
Christoph Hellwigb2950862008-12-02 09:54:17 -05001974static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
Chris Mason247e7432008-07-17 12:53:51 -04001975{
1976 struct btrfs_writepage_fixup *fixup;
1977 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00001978 struct extent_state *cached_state = NULL;
Chris Mason247e7432008-07-17 12:53:51 -04001979 struct page *page;
1980 struct inode *inode;
1981 u64 page_start;
1982 u64 page_end;
Jeff Mahoney87826df2012-02-15 16:23:57 +01001983 int ret;
Chris Mason247e7432008-07-17 12:53:51 -04001984
1985 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1986 page = fixup->page;
Chris Mason4a096752008-07-21 10:29:44 -04001987again:
Chris Mason247e7432008-07-17 12:53:51 -04001988 lock_page(page);
1989 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1990 ClearPageChecked(page);
1991 goto out_page;
1992 }
1993
1994 inode = page->mapping->host;
1995 page_start = page_offset(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001996 page_end = page_offset(page) + PAGE_SIZE - 1;
Chris Mason247e7432008-07-17 12:53:51 -04001997
David Sterbaff13db42015-12-03 14:30:40 +01001998 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01001999 &cached_state);
Chris Mason4a096752008-07-21 10:29:44 -04002000
2001 /* already ordered? We're done */
Chris Mason8b62b722009-09-02 16:53:46 -04002002 if (PagePrivate2(page))
Chris Mason247e7432008-07-17 12:53:51 -04002003 goto out;
Chris Mason4a096752008-07-21 10:29:44 -04002004
Chandan Rajendradbfdb6d2016-01-21 15:55:58 +05302005 ordered = btrfs_lookup_ordered_range(inode, page_start,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002006 PAGE_SIZE);
Chris Mason4a096752008-07-21 10:29:44 -04002007 if (ordered) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00002008 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2009 page_end, &cached_state, GFP_NOFS);
Chris Mason4a096752008-07-21 10:29:44 -04002010 unlock_page(page);
2011 btrfs_start_ordered_extent(inode, ordered, 1);
Jeff Mahoney87826df2012-02-15 16:23:57 +01002012 btrfs_put_ordered_extent(ordered);
Chris Mason4a096752008-07-21 10:29:44 -04002013 goto again;
2014 }
Chris Mason247e7432008-07-17 12:53:51 -04002015
Qu Wenruo7cf5b972015-09-08 17:25:55 +08002016 ret = btrfs_delalloc_reserve_space(inode, page_start,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002017 PAGE_SIZE);
Jeff Mahoney87826df2012-02-15 16:23:57 +01002018 if (ret) {
2019 mapping_set_error(page->mapping, ret);
2020 end_extent_writepage(page, ret, page_start, page_end);
2021 ClearPageChecked(page);
2022 goto out;
2023 }
2024
Josef Bacik2ac55d42010-02-03 19:33:23 +00002025 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
Chris Mason247e7432008-07-17 12:53:51 -04002026 ClearPageChecked(page);
Jeff Mahoney87826df2012-02-15 16:23:57 +01002027 set_page_dirty(page);
Chris Mason247e7432008-07-17 12:53:51 -04002028out:
Josef Bacik2ac55d42010-02-03 19:33:23 +00002029 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2030 &cached_state, GFP_NOFS);
Chris Mason247e7432008-07-17 12:53:51 -04002031out_page:
2032 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002033 put_page(page);
Miao Xieb897abe2011-01-26 16:19:22 +08002034 kfree(fixup);
Chris Mason247e7432008-07-17 12:53:51 -04002035}
2036
2037/*
2038 * There are a few paths in the higher layers of the kernel that directly
2039 * set the page dirty bit without asking the filesystem if it is a
2040 * good idea. This causes problems because we want to make sure COW
2041 * properly happens and the data=ordered rules are followed.
2042 *
Chris Masonc8b97812008-10-29 14:49:59 -04002043 * In our case any range that doesn't have the ORDERED bit set
Chris Mason247e7432008-07-17 12:53:51 -04002044 * hasn't been properly setup for IO. We kick off an async process
2045 * to fix it up. The async helper will wait for ordered extents, set
2046 * the delalloc bit and make it safe to write the page.
2047 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05002048static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
Chris Mason247e7432008-07-17 12:53:51 -04002049{
2050 struct inode *inode = page->mapping->host;
2051 struct btrfs_writepage_fixup *fixup;
2052 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason247e7432008-07-17 12:53:51 -04002053
Chris Mason8b62b722009-09-02 16:53:46 -04002054 /* this page is properly in the ordered list */
2055 if (TestClearPagePrivate2(page))
Chris Mason247e7432008-07-17 12:53:51 -04002056 return 0;
2057
2058 if (PageChecked(page))
2059 return -EAGAIN;
2060
2061 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2062 if (!fixup)
2063 return -EAGAIN;
Chris Masonf4219502008-07-22 11:18:09 -04002064
Chris Mason247e7432008-07-17 12:53:51 -04002065 SetPageChecked(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002066 get_page(page);
Liu Bo9e0af232014-08-15 23:36:53 +08002067 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2068 btrfs_writepage_fixup_worker, NULL, NULL);
Chris Mason247e7432008-07-17 12:53:51 -04002069 fixup->page = page;
Qu Wenruodc6e3202014-02-28 10:46:14 +08002070 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
Jeff Mahoney87826df2012-02-15 16:23:57 +01002071 return -EBUSY;
Chris Mason247e7432008-07-17 12:53:51 -04002072}
2073
Yan Zhengd899e052008-10-30 14:25:28 -04002074static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2075 struct inode *inode, u64 file_pos,
2076 u64 disk_bytenr, u64 disk_num_bytes,
2077 u64 num_bytes, u64 ram_bytes,
2078 u8 compression, u8 encryption,
2079 u16 other_encoding, int extent_type)
2080{
2081 struct btrfs_root *root = BTRFS_I(inode)->root;
2082 struct btrfs_file_extent_item *fi;
2083 struct btrfs_path *path;
2084 struct extent_buffer *leaf;
2085 struct btrfs_key ins;
Filipe David Borba Manana1acae572014-01-07 11:42:27 +00002086 int extent_inserted = 0;
Yan Zhengd899e052008-10-30 14:25:28 -04002087 int ret;
2088
2089 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07002090 if (!path)
2091 return -ENOMEM;
Yan Zhengd899e052008-10-30 14:25:28 -04002092
Chris Masona1ed8352009-09-11 12:27:37 -04002093 /*
2094 * we may be replacing one extent in the tree with another.
2095 * The new extent is pinned in the extent map, and we don't want
2096 * to drop it from the cache until it is completely in the btree.
2097 *
2098 * So, tell btrfs_drop_extents to leave this extent in the cache.
2099 * the caller is expected to unpin it and allow it to be merged
2100 * with the others.
2101 */
Filipe David Borba Manana1acae572014-01-07 11:42:27 +00002102 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2103 file_pos + num_bytes, NULL, 0,
2104 1, sizeof(*fi), &extent_inserted);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002105 if (ret)
2106 goto out;
Yan Zhengd899e052008-10-30 14:25:28 -04002107
Filipe David Borba Manana1acae572014-01-07 11:42:27 +00002108 if (!extent_inserted) {
2109 ins.objectid = btrfs_ino(inode);
2110 ins.offset = file_pos;
2111 ins.type = BTRFS_EXTENT_DATA_KEY;
2112
2113 path->leave_spinning = 1;
2114 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2115 sizeof(*fi));
2116 if (ret)
2117 goto out;
2118 }
Yan Zhengd899e052008-10-30 14:25:28 -04002119 leaf = path->nodes[0];
2120 fi = btrfs_item_ptr(leaf, path->slots[0],
2121 struct btrfs_file_extent_item);
2122 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2123 btrfs_set_file_extent_type(leaf, fi, extent_type);
2124 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2125 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2126 btrfs_set_file_extent_offset(leaf, fi, 0);
2127 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2128 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2129 btrfs_set_file_extent_compression(leaf, fi, compression);
2130 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2131 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
Chris Masonb9473432009-03-13 11:00:37 -04002132
Yan Zhengd899e052008-10-30 14:25:28 -04002133 btrfs_mark_buffer_dirty(leaf);
Josef Bacikce195332012-09-25 15:26:16 -04002134 btrfs_release_path(path);
Yan Zhengd899e052008-10-30 14:25:28 -04002135
2136 inode_add_bytes(inode, num_bytes);
Yan Zhengd899e052008-10-30 14:25:28 -04002137
2138 ins.objectid = disk_bytenr;
2139 ins.offset = disk_num_bytes;
2140 ins.type = BTRFS_EXTENT_ITEM_KEY;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002141 ret = btrfs_alloc_reserved_file_extent(trans, root,
2142 root->root_key.objectid,
Qu Wenruo5846a3c2015-10-26 14:11:18 +08002143 btrfs_ino(inode), file_pos,
2144 ram_bytes, &ins);
Qu Wenruo297d7502015-09-08 17:08:37 +08002145 /*
Qu Wenruo5846a3c2015-10-26 14:11:18 +08002146 * Release the reserved range from inode dirty range map, as it is
2147 * already moved into delayed_ref_head
Qu Wenruo297d7502015-09-08 17:08:37 +08002148 */
2149 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002150out:
Yan Zhengd899e052008-10-30 14:25:28 -04002151 btrfs_free_path(path);
Chris Masonb9473432009-03-13 11:00:37 -04002152
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002153 return ret;
Yan Zhengd899e052008-10-30 14:25:28 -04002154}
2155
Liu Bo38c227d2013-01-29 03:18:40 +00002156/* snapshot-aware defrag */
2157struct sa_defrag_extent_backref {
2158 struct rb_node node;
2159 struct old_sa_defrag_extent *old;
2160 u64 root_id;
2161 u64 inum;
2162 u64 file_pos;
2163 u64 extent_offset;
2164 u64 num_bytes;
2165 u64 generation;
2166};
2167
2168struct old_sa_defrag_extent {
2169 struct list_head list;
2170 struct new_sa_defrag_extent *new;
2171
2172 u64 extent_offset;
2173 u64 bytenr;
2174 u64 offset;
2175 u64 len;
2176 int count;
2177};
2178
2179struct new_sa_defrag_extent {
2180 struct rb_root root;
2181 struct list_head head;
2182 struct btrfs_path *path;
2183 struct inode *inode;
2184 u64 file_pos;
2185 u64 len;
2186 u64 bytenr;
2187 u64 disk_len;
2188 u8 compress_type;
2189};
2190
2191static int backref_comp(struct sa_defrag_extent_backref *b1,
2192 struct sa_defrag_extent_backref *b2)
2193{
2194 if (b1->root_id < b2->root_id)
2195 return -1;
2196 else if (b1->root_id > b2->root_id)
2197 return 1;
2198
2199 if (b1->inum < b2->inum)
2200 return -1;
2201 else if (b1->inum > b2->inum)
2202 return 1;
2203
2204 if (b1->file_pos < b2->file_pos)
2205 return -1;
2206 else if (b1->file_pos > b2->file_pos)
2207 return 1;
2208
2209 /*
2210 * [------------------------------] ===> (a range of space)
2211 * |<--->| |<---->| =============> (fs/file tree A)
2212 * |<---------------------------->| ===> (fs/file tree B)
2213 *
2214 * A range of space can refer to two file extents in one tree while
2215 * refer to only one file extent in another tree.
2216 *
2217 * So we may process a disk offset more than one time(two extents in A)
2218 * and locate at the same extent(one extent in B), then insert two same
2219 * backrefs(both refer to the extent in B).
2220 */
2221 return 0;
2222}
2223
2224static void backref_insert(struct rb_root *root,
2225 struct sa_defrag_extent_backref *backref)
2226{
2227 struct rb_node **p = &root->rb_node;
2228 struct rb_node *parent = NULL;
2229 struct sa_defrag_extent_backref *entry;
2230 int ret;
2231
2232 while (*p) {
2233 parent = *p;
2234 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2235
2236 ret = backref_comp(backref, entry);
2237 if (ret < 0)
2238 p = &(*p)->rb_left;
2239 else
2240 p = &(*p)->rb_right;
2241 }
2242
2243 rb_link_node(&backref->node, parent, p);
2244 rb_insert_color(&backref->node, root);
2245}
2246
2247/*
2248 * Note the backref might has changed, and in this case we just return 0.
2249 */
2250static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2251 void *ctx)
2252{
2253 struct btrfs_file_extent_item *extent;
2254 struct btrfs_fs_info *fs_info;
2255 struct old_sa_defrag_extent *old = ctx;
2256 struct new_sa_defrag_extent *new = old->new;
2257 struct btrfs_path *path = new->path;
2258 struct btrfs_key key;
2259 struct btrfs_root *root;
2260 struct sa_defrag_extent_backref *backref;
2261 struct extent_buffer *leaf;
2262 struct inode *inode = new->inode;
2263 int slot;
2264 int ret;
2265 u64 extent_offset;
2266 u64 num_bytes;
2267
2268 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2269 inum == btrfs_ino(inode))
2270 return 0;
2271
2272 key.objectid = root_id;
2273 key.type = BTRFS_ROOT_ITEM_KEY;
2274 key.offset = (u64)-1;
2275
2276 fs_info = BTRFS_I(inode)->root->fs_info;
2277 root = btrfs_read_fs_root_no_name(fs_info, &key);
2278 if (IS_ERR(root)) {
2279 if (PTR_ERR(root) == -ENOENT)
2280 return 0;
2281 WARN_ON(1);
2282 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2283 inum, offset, root_id);
2284 return PTR_ERR(root);
2285 }
2286
2287 key.objectid = inum;
2288 key.type = BTRFS_EXTENT_DATA_KEY;
2289 if (offset > (u64)-1 << 32)
2290 key.offset = 0;
2291 else
2292 key.offset = offset;
2293
2294 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05302295 if (WARN_ON(ret < 0))
Liu Bo38c227d2013-01-29 03:18:40 +00002296 return ret;
Josef Bacik50f13192013-07-22 12:50:37 -04002297 ret = 0;
Liu Bo38c227d2013-01-29 03:18:40 +00002298
2299 while (1) {
2300 cond_resched();
2301
2302 leaf = path->nodes[0];
2303 slot = path->slots[0];
2304
2305 if (slot >= btrfs_header_nritems(leaf)) {
2306 ret = btrfs_next_leaf(root, path);
2307 if (ret < 0) {
2308 goto out;
2309 } else if (ret > 0) {
2310 ret = 0;
2311 goto out;
2312 }
2313 continue;
2314 }
2315
2316 path->slots[0]++;
2317
2318 btrfs_item_key_to_cpu(leaf, &key, slot);
2319
2320 if (key.objectid > inum)
2321 goto out;
2322
2323 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2324 continue;
2325
2326 extent = btrfs_item_ptr(leaf, slot,
2327 struct btrfs_file_extent_item);
2328
2329 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2330 continue;
2331
Liu Boe68afa42013-07-01 22:13:26 +08002332 /*
2333 * 'offset' refers to the exact key.offset,
2334 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2335 * (key.offset - extent_offset).
2336 */
2337 if (key.offset != offset)
Liu Bo38c227d2013-01-29 03:18:40 +00002338 continue;
2339
Liu Boe68afa42013-07-01 22:13:26 +08002340 extent_offset = btrfs_file_extent_offset(leaf, extent);
Liu Bo38c227d2013-01-29 03:18:40 +00002341 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
Liu Boe68afa42013-07-01 22:13:26 +08002342
Liu Bo38c227d2013-01-29 03:18:40 +00002343 if (extent_offset >= old->extent_offset + old->offset +
2344 old->len || extent_offset + num_bytes <=
2345 old->extent_offset + old->offset)
2346 continue;
Liu Bo38c227d2013-01-29 03:18:40 +00002347 break;
2348 }
2349
2350 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2351 if (!backref) {
2352 ret = -ENOENT;
2353 goto out;
2354 }
2355
2356 backref->root_id = root_id;
2357 backref->inum = inum;
Liu Boe68afa42013-07-01 22:13:26 +08002358 backref->file_pos = offset;
Liu Bo38c227d2013-01-29 03:18:40 +00002359 backref->num_bytes = num_bytes;
2360 backref->extent_offset = extent_offset;
2361 backref->generation = btrfs_file_extent_generation(leaf, extent);
2362 backref->old = old;
2363 backref_insert(&new->root, backref);
2364 old->count++;
2365out:
2366 btrfs_release_path(path);
2367 WARN_ON(ret);
2368 return ret;
2369}
2370
2371static noinline bool record_extent_backrefs(struct btrfs_path *path,
2372 struct new_sa_defrag_extent *new)
2373{
2374 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2375 struct old_sa_defrag_extent *old, *tmp;
2376 int ret;
2377
2378 new->path = path;
2379
2380 list_for_each_entry_safe(old, tmp, &new->head, list) {
Liu Boe68afa42013-07-01 22:13:26 +08002381 ret = iterate_inodes_from_logical(old->bytenr +
2382 old->extent_offset, fs_info,
Liu Bo38c227d2013-01-29 03:18:40 +00002383 path, record_one_backref,
2384 old);
Josef Bacik4724b102013-11-05 11:11:40 -05002385 if (ret < 0 && ret != -ENOENT)
2386 return false;
Liu Bo38c227d2013-01-29 03:18:40 +00002387
2388 /* no backref to be processed for this extent */
2389 if (!old->count) {
2390 list_del(&old->list);
2391 kfree(old);
2392 }
2393 }
2394
2395 if (list_empty(&new->head))
2396 return false;
2397
2398 return true;
2399}
2400
2401static int relink_is_mergable(struct extent_buffer *leaf,
2402 struct btrfs_file_extent_item *fi,
Liu Bo116e0022013-08-02 16:30:40 +08002403 struct new_sa_defrag_extent *new)
Liu Bo38c227d2013-01-29 03:18:40 +00002404{
Liu Bo116e0022013-08-02 16:30:40 +08002405 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
Liu Bo38c227d2013-01-29 03:18:40 +00002406 return 0;
2407
2408 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2409 return 0;
2410
Liu Bo116e0022013-08-02 16:30:40 +08002411 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2412 return 0;
2413
2414 if (btrfs_file_extent_encryption(leaf, fi) ||
Liu Bo38c227d2013-01-29 03:18:40 +00002415 btrfs_file_extent_other_encoding(leaf, fi))
2416 return 0;
2417
2418 return 1;
2419}
2420
2421/*
2422 * Note the backref might has changed, and in this case we just return 0.
2423 */
2424static noinline int relink_extent_backref(struct btrfs_path *path,
2425 struct sa_defrag_extent_backref *prev,
2426 struct sa_defrag_extent_backref *backref)
2427{
2428 struct btrfs_file_extent_item *extent;
2429 struct btrfs_file_extent_item *item;
2430 struct btrfs_ordered_extent *ordered;
2431 struct btrfs_trans_handle *trans;
2432 struct btrfs_fs_info *fs_info;
2433 struct btrfs_root *root;
2434 struct btrfs_key key;
2435 struct extent_buffer *leaf;
2436 struct old_sa_defrag_extent *old = backref->old;
2437 struct new_sa_defrag_extent *new = old->new;
2438 struct inode *src_inode = new->inode;
2439 struct inode *inode;
2440 struct extent_state *cached = NULL;
2441 int ret = 0;
2442 u64 start;
2443 u64 len;
2444 u64 lock_start;
2445 u64 lock_end;
2446 bool merge = false;
2447 int index;
2448
2449 if (prev && prev->root_id == backref->root_id &&
2450 prev->inum == backref->inum &&
2451 prev->file_pos + prev->num_bytes == backref->file_pos)
2452 merge = true;
2453
2454 /* step 1: get root */
2455 key.objectid = backref->root_id;
2456 key.type = BTRFS_ROOT_ITEM_KEY;
2457 key.offset = (u64)-1;
2458
2459 fs_info = BTRFS_I(src_inode)->root->fs_info;
2460 index = srcu_read_lock(&fs_info->subvol_srcu);
2461
2462 root = btrfs_read_fs_root_no_name(fs_info, &key);
2463 if (IS_ERR(root)) {
2464 srcu_read_unlock(&fs_info->subvol_srcu, index);
2465 if (PTR_ERR(root) == -ENOENT)
2466 return 0;
2467 return PTR_ERR(root);
2468 }
Liu Bo38c227d2013-01-29 03:18:40 +00002469
Wang Shilongbcbba5e2014-02-08 23:46:35 +08002470 if (btrfs_root_readonly(root)) {
2471 srcu_read_unlock(&fs_info->subvol_srcu, index);
2472 return 0;
2473 }
2474
Liu Bo38c227d2013-01-29 03:18:40 +00002475 /* step 2: get inode */
2476 key.objectid = backref->inum;
2477 key.type = BTRFS_INODE_ITEM_KEY;
2478 key.offset = 0;
2479
2480 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2481 if (IS_ERR(inode)) {
2482 srcu_read_unlock(&fs_info->subvol_srcu, index);
2483 return 0;
2484 }
2485
2486 srcu_read_unlock(&fs_info->subvol_srcu, index);
2487
2488 /* step 3: relink backref */
2489 lock_start = backref->file_pos;
2490 lock_end = backref->file_pos + backref->num_bytes - 1;
2491 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
David Sterbaff13db42015-12-03 14:30:40 +01002492 &cached);
Liu Bo38c227d2013-01-29 03:18:40 +00002493
2494 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2495 if (ordered) {
2496 btrfs_put_ordered_extent(ordered);
2497 goto out_unlock;
2498 }
2499
2500 trans = btrfs_join_transaction(root);
2501 if (IS_ERR(trans)) {
2502 ret = PTR_ERR(trans);
2503 goto out_unlock;
2504 }
2505
2506 key.objectid = backref->inum;
2507 key.type = BTRFS_EXTENT_DATA_KEY;
2508 key.offset = backref->file_pos;
2509
2510 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2511 if (ret < 0) {
2512 goto out_free_path;
2513 } else if (ret > 0) {
2514 ret = 0;
2515 goto out_free_path;
2516 }
2517
2518 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2519 struct btrfs_file_extent_item);
2520
2521 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2522 backref->generation)
2523 goto out_free_path;
2524
2525 btrfs_release_path(path);
2526
2527 start = backref->file_pos;
2528 if (backref->extent_offset < old->extent_offset + old->offset)
2529 start += old->extent_offset + old->offset -
2530 backref->extent_offset;
2531
2532 len = min(backref->extent_offset + backref->num_bytes,
2533 old->extent_offset + old->offset + old->len);
2534 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2535
2536 ret = btrfs_drop_extents(trans, root, inode, start,
2537 start + len, 1);
2538 if (ret)
2539 goto out_free_path;
2540again:
2541 key.objectid = btrfs_ino(inode);
2542 key.type = BTRFS_EXTENT_DATA_KEY;
2543 key.offset = start;
2544
Liu Boa09a0a72013-03-11 09:20:58 +00002545 path->leave_spinning = 1;
Liu Bo38c227d2013-01-29 03:18:40 +00002546 if (merge) {
2547 struct btrfs_file_extent_item *fi;
2548 u64 extent_len;
2549 struct btrfs_key found_key;
2550
Gui Hecheng3c9665d2014-01-23 13:41:09 +08002551 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
Liu Bo38c227d2013-01-29 03:18:40 +00002552 if (ret < 0)
2553 goto out_free_path;
2554
2555 path->slots[0]--;
2556 leaf = path->nodes[0];
2557 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2558
2559 fi = btrfs_item_ptr(leaf, path->slots[0],
2560 struct btrfs_file_extent_item);
2561 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2562
Liu Bo116e0022013-08-02 16:30:40 +08002563 if (extent_len + found_key.offset == start &&
2564 relink_is_mergable(leaf, fi, new)) {
Liu Bo38c227d2013-01-29 03:18:40 +00002565 btrfs_set_file_extent_num_bytes(leaf, fi,
2566 extent_len + len);
2567 btrfs_mark_buffer_dirty(leaf);
2568 inode_add_bytes(inode, len);
2569
2570 ret = 1;
2571 goto out_free_path;
2572 } else {
2573 merge = false;
2574 btrfs_release_path(path);
2575 goto again;
2576 }
2577 }
2578
2579 ret = btrfs_insert_empty_item(trans, root, path, &key,
2580 sizeof(*extent));
2581 if (ret) {
2582 btrfs_abort_transaction(trans, root, ret);
2583 goto out_free_path;
2584 }
2585
2586 leaf = path->nodes[0];
2587 item = btrfs_item_ptr(leaf, path->slots[0],
2588 struct btrfs_file_extent_item);
2589 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2590 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2591 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2592 btrfs_set_file_extent_num_bytes(leaf, item, len);
2593 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2594 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2595 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2596 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2597 btrfs_set_file_extent_encryption(leaf, item, 0);
2598 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2599
2600 btrfs_mark_buffer_dirty(leaf);
2601 inode_add_bytes(inode, len);
Liu Boa09a0a72013-03-11 09:20:58 +00002602 btrfs_release_path(path);
Liu Bo38c227d2013-01-29 03:18:40 +00002603
2604 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2605 new->disk_len, 0,
2606 backref->root_id, backref->inum,
Filipe Mananab06c4bf2015-10-23 07:52:54 +01002607 new->file_pos); /* start - extent_offset */
Liu Bo38c227d2013-01-29 03:18:40 +00002608 if (ret) {
2609 btrfs_abort_transaction(trans, root, ret);
2610 goto out_free_path;
2611 }
2612
2613 ret = 1;
2614out_free_path:
2615 btrfs_release_path(path);
Liu Boa09a0a72013-03-11 09:20:58 +00002616 path->leave_spinning = 0;
Liu Bo38c227d2013-01-29 03:18:40 +00002617 btrfs_end_transaction(trans, root);
2618out_unlock:
2619 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2620 &cached, GFP_NOFS);
2621 iput(inode);
2622 return ret;
2623}
2624
Liu Bo6f519562013-10-29 10:45:05 +08002625static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2626{
2627 struct old_sa_defrag_extent *old, *tmp;
2628
2629 if (!new)
2630 return;
2631
2632 list_for_each_entry_safe(old, tmp, &new->head, list) {
Liu Bo6f519562013-10-29 10:45:05 +08002633 kfree(old);
2634 }
2635 kfree(new);
2636}
2637
Liu Bo38c227d2013-01-29 03:18:40 +00002638static void relink_file_extents(struct new_sa_defrag_extent *new)
2639{
2640 struct btrfs_path *path;
Liu Bo38c227d2013-01-29 03:18:40 +00002641 struct sa_defrag_extent_backref *backref;
2642 struct sa_defrag_extent_backref *prev = NULL;
2643 struct inode *inode;
2644 struct btrfs_root *root;
2645 struct rb_node *node;
2646 int ret;
2647
2648 inode = new->inode;
2649 root = BTRFS_I(inode)->root;
2650
2651 path = btrfs_alloc_path();
2652 if (!path)
2653 return;
2654
2655 if (!record_extent_backrefs(path, new)) {
2656 btrfs_free_path(path);
2657 goto out;
2658 }
2659 btrfs_release_path(path);
2660
2661 while (1) {
2662 node = rb_first(&new->root);
2663 if (!node)
2664 break;
2665 rb_erase(node, &new->root);
2666
2667 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2668
2669 ret = relink_extent_backref(path, prev, backref);
2670 WARN_ON(ret < 0);
2671
2672 kfree(prev);
2673
2674 if (ret == 1)
2675 prev = backref;
2676 else
2677 prev = NULL;
2678 cond_resched();
2679 }
2680 kfree(prev);
2681
2682 btrfs_free_path(path);
Liu Bo38c227d2013-01-29 03:18:40 +00002683out:
Liu Bo6f519562013-10-29 10:45:05 +08002684 free_sa_defrag_extent(new);
2685
Liu Bo38c227d2013-01-29 03:18:40 +00002686 atomic_dec(&root->fs_info->defrag_running);
2687 wake_up(&root->fs_info->transaction_wait);
Liu Bo38c227d2013-01-29 03:18:40 +00002688}
2689
2690static struct new_sa_defrag_extent *
2691record_old_file_extents(struct inode *inode,
2692 struct btrfs_ordered_extent *ordered)
2693{
2694 struct btrfs_root *root = BTRFS_I(inode)->root;
2695 struct btrfs_path *path;
2696 struct btrfs_key key;
Liu Bo6f519562013-10-29 10:45:05 +08002697 struct old_sa_defrag_extent *old;
Liu Bo38c227d2013-01-29 03:18:40 +00002698 struct new_sa_defrag_extent *new;
2699 int ret;
2700
2701 new = kmalloc(sizeof(*new), GFP_NOFS);
2702 if (!new)
2703 return NULL;
2704
2705 new->inode = inode;
2706 new->file_pos = ordered->file_offset;
2707 new->len = ordered->len;
2708 new->bytenr = ordered->start;
2709 new->disk_len = ordered->disk_len;
2710 new->compress_type = ordered->compress_type;
2711 new->root = RB_ROOT;
2712 INIT_LIST_HEAD(&new->head);
2713
2714 path = btrfs_alloc_path();
2715 if (!path)
2716 goto out_kfree;
2717
2718 key.objectid = btrfs_ino(inode);
2719 key.type = BTRFS_EXTENT_DATA_KEY;
2720 key.offset = new->file_pos;
2721
2722 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2723 if (ret < 0)
2724 goto out_free_path;
2725 if (ret > 0 && path->slots[0] > 0)
2726 path->slots[0]--;
2727
2728 /* find out all the old extents for the file range */
2729 while (1) {
2730 struct btrfs_file_extent_item *extent;
2731 struct extent_buffer *l;
2732 int slot;
2733 u64 num_bytes;
2734 u64 offset;
2735 u64 end;
2736 u64 disk_bytenr;
2737 u64 extent_offset;
2738
2739 l = path->nodes[0];
2740 slot = path->slots[0];
2741
2742 if (slot >= btrfs_header_nritems(l)) {
2743 ret = btrfs_next_leaf(root, path);
2744 if (ret < 0)
Liu Bo6f519562013-10-29 10:45:05 +08002745 goto out_free_path;
Liu Bo38c227d2013-01-29 03:18:40 +00002746 else if (ret > 0)
2747 break;
2748 continue;
2749 }
2750
2751 btrfs_item_key_to_cpu(l, &key, slot);
2752
2753 if (key.objectid != btrfs_ino(inode))
2754 break;
2755 if (key.type != BTRFS_EXTENT_DATA_KEY)
2756 break;
2757 if (key.offset >= new->file_pos + new->len)
2758 break;
2759
2760 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2761
2762 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2763 if (key.offset + num_bytes < new->file_pos)
2764 goto next;
2765
2766 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2767 if (!disk_bytenr)
2768 goto next;
2769
2770 extent_offset = btrfs_file_extent_offset(l, extent);
2771
2772 old = kmalloc(sizeof(*old), GFP_NOFS);
2773 if (!old)
Liu Bo6f519562013-10-29 10:45:05 +08002774 goto out_free_path;
Liu Bo38c227d2013-01-29 03:18:40 +00002775
2776 offset = max(new->file_pos, key.offset);
2777 end = min(new->file_pos + new->len, key.offset + num_bytes);
2778
2779 old->bytenr = disk_bytenr;
2780 old->extent_offset = extent_offset;
2781 old->offset = offset - key.offset;
2782 old->len = end - offset;
2783 old->new = new;
2784 old->count = 0;
2785 list_add_tail(&old->list, &new->head);
2786next:
2787 path->slots[0]++;
2788 cond_resched();
2789 }
2790
2791 btrfs_free_path(path);
2792 atomic_inc(&root->fs_info->defrag_running);
2793
2794 return new;
2795
Liu Bo38c227d2013-01-29 03:18:40 +00002796out_free_path:
2797 btrfs_free_path(path);
2798out_kfree:
Liu Bo6f519562013-10-29 10:45:05 +08002799 free_sa_defrag_extent(new);
Liu Bo38c227d2013-01-29 03:18:40 +00002800 return NULL;
2801}
2802
Miao Xiee570fd22014-06-19 10:42:50 +08002803static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2804 u64 start, u64 len)
2805{
2806 struct btrfs_block_group_cache *cache;
2807
2808 cache = btrfs_lookup_block_group(root->fs_info, start);
2809 ASSERT(cache);
2810
2811 spin_lock(&cache->lock);
2812 cache->delalloc_bytes -= len;
2813 spin_unlock(&cache->lock);
2814
2815 btrfs_put_block_group(cache);
2816}
2817
Chris Masond352ac62008-09-29 15:18:18 -04002818/* as ordered data IO finishes, this gets called so we can finish
2819 * an ordered extent if the range of bytes in the file it covers are
2820 * fully written.
2821 */
Josef Bacik5fd02042012-05-02 14:00:54 -04002822static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
Chris Masone6dcd2d2008-07-17 12:53:50 -04002823{
Josef Bacik5fd02042012-05-02 14:00:54 -04002824 struct inode *inode = ordered_extent->inode;
Chris Masone6dcd2d2008-07-17 12:53:50 -04002825 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04002826 struct btrfs_trans_handle *trans = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04002827 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Josef Bacik2ac55d42010-02-03 19:33:23 +00002828 struct extent_state *cached_state = NULL;
Liu Bo38c227d2013-01-29 03:18:40 +00002829 struct new_sa_defrag_extent *new = NULL;
Li Zefan261507a02010-12-17 14:21:50 +08002830 int compress_type = 0;
Josef Bacik77cef2e2013-08-29 13:57:21 -04002831 int ret = 0;
2832 u64 logical_len = ordered_extent->len;
Li Zefan82d59022011-04-20 10:33:24 +08002833 bool nolock;
Josef Bacik77cef2e2013-08-29 13:57:21 -04002834 bool truncated = false;
Chris Masone6dcd2d2008-07-17 12:53:50 -04002835
Liu Bo83eea1f2012-07-10 05:28:39 -06002836 nolock = btrfs_is_free_space_inode(inode);
Josef Bacik0cb59c92010-07-02 12:14:14 -04002837
Josef Bacik5fd02042012-05-02 14:00:54 -04002838 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2839 ret = -EIO;
2840 goto out;
2841 }
2842
Miao Xief6124962014-09-12 18:44:04 +08002843 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2844 ordered_extent->file_offset +
2845 ordered_extent->len - 1);
2846
Josef Bacik77cef2e2013-08-29 13:57:21 -04002847 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2848 truncated = true;
2849 logical_len = ordered_extent->truncated_len;
2850 /* Truncated the entire extent, don't bother adding */
2851 if (!logical_len)
2852 goto out;
2853 }
2854
Yan, Zhengc2167752009-11-12 09:34:21 +00002855 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002856 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
Qu Wenruo94ed9382015-09-08 17:25:56 +08002857
2858 /*
2859 * For mwrite(mmap + memset to write) case, we still reserve
2860 * space for NOCOW range.
2861 * As NOCOW won't cause a new delayed ref, just free the space
2862 */
2863 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2864 ordered_extent->len);
Josef Bacik6c760c02012-11-09 10:53:21 -05002865 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2866 if (nolock)
2867 trans = btrfs_join_transaction_nolock(root);
2868 else
2869 trans = btrfs_join_transaction(root);
2870 if (IS_ERR(trans)) {
2871 ret = PTR_ERR(trans);
2872 trans = NULL;
2873 goto out;
Yan, Zhengc2167752009-11-12 09:34:21 +00002874 }
Josef Bacik6c760c02012-11-09 10:53:21 -05002875 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2876 ret = btrfs_update_inode_fallback(trans, root, inode);
2877 if (ret) /* -ENOMEM or corruption */
2878 btrfs_abort_transaction(trans, root, ret);
Yan, Zhengc2167752009-11-12 09:34:21 +00002879 goto out;
2880 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04002881
Josef Bacik2ac55d42010-02-03 19:33:23 +00002882 lock_extent_bits(io_tree, ordered_extent->file_offset,
2883 ordered_extent->file_offset + ordered_extent->len - 1,
David Sterbaff13db42015-12-03 14:30:40 +01002884 &cached_state);
Chris Masone6dcd2d2008-07-17 12:53:50 -04002885
Liu Bo38c227d2013-01-29 03:18:40 +00002886 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2887 ordered_extent->file_offset + ordered_extent->len - 1,
2888 EXTENT_DEFRAG, 1, cached_state);
2889 if (ret) {
2890 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
Josef Bacik8101c8d2014-01-29 16:05:30 -05002891 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
Liu Bo38c227d2013-01-29 03:18:40 +00002892 /* the inode is shared */
2893 new = record_old_file_extents(inode, ordered_extent);
2894
2895 clear_extent_bit(io_tree, ordered_extent->file_offset,
2896 ordered_extent->file_offset + ordered_extent->len - 1,
2897 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2898 }
2899
Josef Bacik0cb59c92010-07-02 12:14:14 -04002900 if (nolock)
Josef Bacik7a7eaa42011-04-13 12:54:33 -04002901 trans = btrfs_join_transaction_nolock(root);
Josef Bacik0cb59c92010-07-02 12:14:14 -04002902 else
Josef Bacik7a7eaa42011-04-13 12:54:33 -04002903 trans = btrfs_join_transaction(root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002904 if (IS_ERR(trans)) {
2905 ret = PTR_ERR(trans);
2906 trans = NULL;
2907 goto out_unlock;
2908 }
Chris Masona79b7d42014-05-22 16:18:52 -07002909
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04002910 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
Yan, Zhengc2167752009-11-12 09:34:21 +00002911
Chris Masonc8b97812008-10-29 14:49:59 -04002912 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
Li Zefan261507a02010-12-17 14:21:50 +08002913 compress_type = ordered_extent->compress_type;
Yan Zhengd899e052008-10-30 14:25:28 -04002914 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
Li Zefan261507a02010-12-17 14:21:50 +08002915 BUG_ON(compress_type);
Yan, Zheng920bbbf2009-11-12 09:34:08 +00002916 ret = btrfs_mark_extent_written(trans, inode,
Yan Zhengd899e052008-10-30 14:25:28 -04002917 ordered_extent->file_offset,
2918 ordered_extent->file_offset +
Josef Bacik77cef2e2013-08-29 13:57:21 -04002919 logical_len);
Yan Zhengd899e052008-10-30 14:25:28 -04002920 } else {
Josef Bacik0af3d002010-06-21 14:48:16 -04002921 BUG_ON(root == root->fs_info->tree_root);
Yan Zhengd899e052008-10-30 14:25:28 -04002922 ret = insert_reserved_file_extent(trans, inode,
2923 ordered_extent->file_offset,
2924 ordered_extent->start,
2925 ordered_extent->disk_len,
Josef Bacik77cef2e2013-08-29 13:57:21 -04002926 logical_len, logical_len,
Li Zefan261507a02010-12-17 14:21:50 +08002927 compress_type, 0, 0,
Yan Zhengd899e052008-10-30 14:25:28 -04002928 BTRFS_FILE_EXTENT_REG);
Miao Xiee570fd22014-06-19 10:42:50 +08002929 if (!ret)
2930 btrfs_release_delalloc_bytes(root,
2931 ordered_extent->start,
2932 ordered_extent->disk_len);
Yan Zhengd899e052008-10-30 14:25:28 -04002933 }
Josef Bacik5dc562c2012-08-17 13:14:17 -04002934 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2935 ordered_extent->file_offset, ordered_extent->len,
2936 trans->transid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002937 if (ret < 0) {
2938 btrfs_abort_transaction(trans, root, ret);
Josef Bacik5fd02042012-05-02 14:00:54 -04002939 goto out_unlock;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002940 }
Josef Bacik2ac55d42010-02-03 19:33:23 +00002941
Chris Masone6dcd2d2008-07-17 12:53:50 -04002942 add_pending_csums(trans, inode, ordered_extent->file_offset,
2943 &ordered_extent->list);
2944
Josef Bacik6c760c02012-11-09 10:53:21 -05002945 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2946 ret = btrfs_update_inode_fallback(trans, root, inode);
2947 if (ret) { /* -ENOMEM or corruption */
2948 btrfs_abort_transaction(trans, root, ret);
2949 goto out_unlock;
Josef Bacik1ef30be2011-04-05 19:25:36 -04002950 }
2951 ret = 0;
Josef Bacik5fd02042012-05-02 14:00:54 -04002952out_unlock:
2953 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2954 ordered_extent->file_offset +
2955 ordered_extent->len - 1, &cached_state, GFP_NOFS);
Yan, Zhengc2167752009-11-12 09:34:21 +00002956out:
Josef Bacik5b0e95b2011-10-06 08:58:24 -04002957 if (root != root->fs_info->tree_root)
Josef Bacik0cb59c92010-07-02 12:14:14 -04002958 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
Miao Xiea698d0752012-09-20 01:51:59 -06002959 if (trans)
2960 btrfs_end_transaction(trans, root);
Josef Bacik0cb59c92010-07-02 12:14:14 -04002961
Josef Bacik77cef2e2013-08-29 13:57:21 -04002962 if (ret || truncated) {
2963 u64 start, end;
2964
2965 if (truncated)
2966 start = ordered_extent->file_offset + logical_len;
2967 else
2968 start = ordered_extent->file_offset;
2969 end = ordered_extent->file_offset + ordered_extent->len - 1;
2970 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2971
2972 /* Drop the cache for the part of the extent we didn't write. */
2973 btrfs_drop_extent_cache(inode, start, end, 0);
Josef Bacik5fd02042012-05-02 14:00:54 -04002974
Josef Bacik0bec9ef2013-01-31 14:58:00 -05002975 /*
2976 * If the ordered extent had an IOERR or something else went
2977 * wrong we need to return the space for this ordered extent
Josef Bacik77cef2e2013-08-29 13:57:21 -04002978 * back to the allocator. We only free the extent in the
2979 * truncated case if we didn't write out the extent at all.
Josef Bacik0bec9ef2013-01-31 14:58:00 -05002980 */
Josef Bacik77cef2e2013-08-29 13:57:21 -04002981 if ((ret || !logical_len) &&
2982 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
Josef Bacik0bec9ef2013-01-31 14:58:00 -05002983 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2984 btrfs_free_reserved_extent(root, ordered_extent->start,
Miao Xiee570fd22014-06-19 10:42:50 +08002985 ordered_extent->disk_len, 1);
Josef Bacik0bec9ef2013-01-31 14:58:00 -05002986 }
2987
2988
Josef Bacik5fd02042012-05-02 14:00:54 -04002989 /*
Liu Bo8bad3c02012-06-18 12:14:23 +08002990 * This needs to be done to make sure anybody waiting knows we are done
2991 * updating everything for this ordered extent.
Josef Bacik5fd02042012-05-02 14:00:54 -04002992 */
2993 btrfs_remove_ordered_extent(inode, ordered_extent);
2994
Liu Bo38c227d2013-01-29 03:18:40 +00002995 /* for snapshot-aware defrag */
Liu Bo6f519562013-10-29 10:45:05 +08002996 if (new) {
2997 if (ret) {
2998 free_sa_defrag_extent(new);
2999 atomic_dec(&root->fs_info->defrag_running);
3000 } else {
3001 relink_file_extents(new);
3002 }
3003 }
Liu Bo38c227d2013-01-29 03:18:40 +00003004
Chris Masone6dcd2d2008-07-17 12:53:50 -04003005 /* once for us */
3006 btrfs_put_ordered_extent(ordered_extent);
3007 /* once for the tree */
3008 btrfs_put_ordered_extent(ordered_extent);
3009
Josef Bacik5fd02042012-05-02 14:00:54 -04003010 return ret;
3011}
3012
3013static void finish_ordered_fn(struct btrfs_work *work)
3014{
3015 struct btrfs_ordered_extent *ordered_extent;
3016 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3017 btrfs_finish_ordered_io(ordered_extent);
Chris Masone6dcd2d2008-07-17 12:53:50 -04003018}
3019
Christoph Hellwigb2950862008-12-02 09:54:17 -05003020static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
Chris Mason211f90e2008-07-18 11:56:15 -04003021 struct extent_state *state, int uptodate)
3022{
Josef Bacik5fd02042012-05-02 14:00:54 -04003023 struct inode *inode = page->mapping->host;
3024 struct btrfs_root *root = BTRFS_I(inode)->root;
3025 struct btrfs_ordered_extent *ordered_extent = NULL;
Liu Bo9e0af232014-08-15 23:36:53 +08003026 struct btrfs_workqueue *wq;
3027 btrfs_work_func_t func;
Josef Bacik5fd02042012-05-02 14:00:54 -04003028
liubo1abe9b82011-03-24 11:18:59 +00003029 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3030
Chris Mason8b62b722009-09-02 16:53:46 -04003031 ClearPagePrivate2(page);
Josef Bacik5fd02042012-05-02 14:00:54 -04003032 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3033 end - start + 1, uptodate))
3034 return 0;
3035
Liu Bo9e0af232014-08-15 23:36:53 +08003036 if (btrfs_is_free_space_inode(inode)) {
3037 wq = root->fs_info->endio_freespace_worker;
3038 func = btrfs_freespace_write_helper;
3039 } else {
3040 wq = root->fs_info->endio_write_workers;
3041 func = btrfs_endio_write_helper;
3042 }
Josef Bacik5fd02042012-05-02 14:00:54 -04003043
Liu Bo9e0af232014-08-15 23:36:53 +08003044 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3045 NULL);
3046 btrfs_queue_work(wq, &ordered_extent->work);
Josef Bacik5fd02042012-05-02 14:00:54 -04003047
3048 return 0;
Chris Mason211f90e2008-07-18 11:56:15 -04003049}
3050
Miao Xiedc380ae2014-09-12 18:43:55 +08003051static int __readpage_endio_check(struct inode *inode,
3052 struct btrfs_io_bio *io_bio,
3053 int icsum, struct page *page,
3054 int pgoff, u64 start, size_t len)
3055{
3056 char *kaddr;
3057 u32 csum_expected;
3058 u32 csum = ~(u32)0;
Miao Xiedc380ae2014-09-12 18:43:55 +08003059
3060 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3061
3062 kaddr = kmap_atomic(page);
3063 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3064 btrfs_csum_final(csum, (char *)&csum);
3065 if (csum != csum_expected)
3066 goto zeroit;
3067
3068 kunmap_atomic(kaddr);
3069 return 0;
3070zeroit:
David Sterba94647322015-10-08 11:01:36 +02003071 btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3072 "csum failed ino %llu off %llu csum %u expected csum %u",
Miao Xiedc380ae2014-09-12 18:43:55 +08003073 btrfs_ino(inode), start, csum, csum_expected);
3074 memset(kaddr + pgoff, 1, len);
3075 flush_dcache_page(page);
3076 kunmap_atomic(kaddr);
3077 if (csum_expected == 0)
3078 return 0;
3079 return -EIO;
3080}
3081
Chris Masond352ac62008-09-29 15:18:18 -04003082/*
Chris Masond352ac62008-09-29 15:18:18 -04003083 * when reads are done, we need to check csums to verify the data is correct
Jan Schmidt4a54c8c2011-07-22 15:41:52 +02003084 * if there's a match, we allow the bio to finish. If not, the code in
3085 * extent_io.c will try to find good copies for us.
Chris Masond352ac62008-09-29 15:18:18 -04003086 */
Miao Xiefacc8a222013-07-25 19:22:34 +08003087static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3088 u64 phy_offset, struct page *page,
3089 u64 start, u64 end, int mirror)
Chris Mason07157aa2007-08-30 08:50:51 -04003090{
Miao Xie4eee4fa2012-12-21 09:17:45 +00003091 size_t offset = start - page_offset(page);
Chris Mason07157aa2007-08-30 08:50:51 -04003092 struct inode *inode = page->mapping->host;
Chris Masond1310b22008-01-24 16:13:08 -05003093 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Chris Masonff79f812007-10-15 16:22:25 -04003094 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masond1310b22008-01-24 16:13:08 -05003095
Chris Masond20f7042008-12-08 16:58:54 -05003096 if (PageChecked(page)) {
3097 ClearPageChecked(page);
Miao Xiedc380ae2014-09-12 18:43:55 +08003098 return 0;
Chris Masond20f7042008-12-08 16:58:54 -05003099 }
Christoph Hellwig6cbff002009-04-17 10:37:41 +02003100
3101 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
Miao Xiedc380ae2014-09-12 18:43:55 +08003102 return 0;
Chris Masond20f7042008-12-08 16:58:54 -05003103
Yan Zheng17d217f2008-12-12 10:03:38 -05003104 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
Chris Mason9655d292009-09-02 15:22:30 -04003105 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
Yan Zheng17d217f2008-12-12 10:03:38 -05003106 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
3107 GFP_NOFS);
3108 return 0;
3109 }
3110
Miao Xiefacc8a222013-07-25 19:22:34 +08003111 phy_offset >>= inode->i_sb->s_blocksize_bits;
Miao Xiedc380ae2014-09-12 18:43:55 +08003112 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3113 start, (size_t)(end - start + 1));
Chris Mason07157aa2007-08-30 08:50:51 -04003114}
Chris Masonb888db22007-08-27 16:49:44 -04003115
Yan, Zheng24bbcf02009-11-12 09:36:34 +00003116void btrfs_add_delayed_iput(struct inode *inode)
3117{
3118 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
David Sterba8089fe62015-11-19 14:15:51 +01003119 struct btrfs_inode *binode = BTRFS_I(inode);
Yan, Zheng24bbcf02009-11-12 09:36:34 +00003120
3121 if (atomic_add_unless(&inode->i_count, -1, 1))
3122 return;
3123
Yan, Zheng24bbcf02009-11-12 09:36:34 +00003124 spin_lock(&fs_info->delayed_iput_lock);
David Sterba8089fe62015-11-19 14:15:51 +01003125 if (binode->delayed_iput_count == 0) {
3126 ASSERT(list_empty(&binode->delayed_iput));
3127 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3128 } else {
3129 binode->delayed_iput_count++;
3130 }
Yan, Zheng24bbcf02009-11-12 09:36:34 +00003131 spin_unlock(&fs_info->delayed_iput_lock);
3132}
3133
3134void btrfs_run_delayed_iputs(struct btrfs_root *root)
3135{
Yan, Zheng24bbcf02009-11-12 09:36:34 +00003136 struct btrfs_fs_info *fs_info = root->fs_info;
Yan, Zheng24bbcf02009-11-12 09:36:34 +00003137
Yan, Zheng24bbcf02009-11-12 09:36:34 +00003138 spin_lock(&fs_info->delayed_iput_lock);
David Sterba8089fe62015-11-19 14:15:51 +01003139 while (!list_empty(&fs_info->delayed_iputs)) {
3140 struct btrfs_inode *inode;
Yan, Zheng24bbcf02009-11-12 09:36:34 +00003141
David Sterba8089fe62015-11-19 14:15:51 +01003142 inode = list_first_entry(&fs_info->delayed_iputs,
3143 struct btrfs_inode, delayed_iput);
3144 if (inode->delayed_iput_count) {
3145 inode->delayed_iput_count--;
3146 list_move_tail(&inode->delayed_iput,
3147 &fs_info->delayed_iputs);
3148 } else {
3149 list_del_init(&inode->delayed_iput);
3150 }
3151 spin_unlock(&fs_info->delayed_iput_lock);
3152 iput(&inode->vfs_inode);
3153 spin_lock(&fs_info->delayed_iput_lock);
Yan, Zheng24bbcf02009-11-12 09:36:34 +00003154 }
David Sterba8089fe62015-11-19 14:15:51 +01003155 spin_unlock(&fs_info->delayed_iput_lock);
Yan, Zheng24bbcf02009-11-12 09:36:34 +00003156}
3157
Yan, Zhengd68fc572010-05-16 10:49:58 -04003158/*
Justin P. Mattock42b2aa82011-11-28 20:31:00 -08003159 * This is called in transaction commit time. If there are no orphan
Yan, Zhengd68fc572010-05-16 10:49:58 -04003160 * files in the subvolume, it removes orphan item and frees block_rsv
3161 * structure.
3162 */
3163void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3164 struct btrfs_root *root)
3165{
Josef Bacik90290e12011-12-02 15:44:12 -05003166 struct btrfs_block_rsv *block_rsv;
Yan, Zhengd68fc572010-05-16 10:49:58 -04003167 int ret;
3168
Josef Bacik8a35d952012-05-23 14:26:42 -04003169 if (atomic_read(&root->orphan_inodes) ||
Yan, Zhengd68fc572010-05-16 10:49:58 -04003170 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3171 return;
3172
Josef Bacik90290e12011-12-02 15:44:12 -05003173 spin_lock(&root->orphan_lock);
Josef Bacik8a35d952012-05-23 14:26:42 -04003174 if (atomic_read(&root->orphan_inodes)) {
Josef Bacik90290e12011-12-02 15:44:12 -05003175 spin_unlock(&root->orphan_lock);
3176 return;
3177 }
3178
3179 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3180 spin_unlock(&root->orphan_lock);
3181 return;
3182 }
3183
3184 block_rsv = root->orphan_block_rsv;
3185 root->orphan_block_rsv = NULL;
3186 spin_unlock(&root->orphan_lock);
3187
Miao Xie27cdeb72014-04-02 19:51:05 +08003188 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
Yan, Zhengd68fc572010-05-16 10:49:58 -04003189 btrfs_root_refs(&root->root_item) > 0) {
3190 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3191 root->root_key.objectid);
Josef Bacik4ef31a42013-08-13 14:10:08 -04003192 if (ret)
3193 btrfs_abort_transaction(trans, root, ret);
3194 else
Miao Xie27cdeb72014-04-02 19:51:05 +08003195 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3196 &root->state);
Yan, Zhengd68fc572010-05-16 10:49:58 -04003197 }
3198
Josef Bacik90290e12011-12-02 15:44:12 -05003199 if (block_rsv) {
3200 WARN_ON(block_rsv->size > 0);
3201 btrfs_free_block_rsv(root, block_rsv);
Yan, Zhengd68fc572010-05-16 10:49:58 -04003202 }
3203}
3204
3205/*
Josef Bacik7b128762008-07-24 12:17:14 -04003206 * This creates an orphan entry for the given inode in case something goes
3207 * wrong in the middle of an unlink/truncate.
Yan, Zhengd68fc572010-05-16 10:49:58 -04003208 *
3209 * NOTE: caller of this function should reserve 5 units of metadata for
3210 * this function.
Josef Bacik7b128762008-07-24 12:17:14 -04003211 */
3212int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3213{
3214 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zhengd68fc572010-05-16 10:49:58 -04003215 struct btrfs_block_rsv *block_rsv = NULL;
3216 int reserve = 0;
3217 int insert = 0;
3218 int ret;
Josef Bacik7b128762008-07-24 12:17:14 -04003219
Yan, Zhengd68fc572010-05-16 10:49:58 -04003220 if (!root->orphan_block_rsv) {
Miao Xie66d8f3d2012-09-06 04:02:28 -06003221 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
Tsutomu Itohb5324022011-07-19 07:27:20 +00003222 if (!block_rsv)
3223 return -ENOMEM;
Josef Bacik7b128762008-07-24 12:17:14 -04003224 }
3225
Yan, Zhengd68fc572010-05-16 10:49:58 -04003226 spin_lock(&root->orphan_lock);
3227 if (!root->orphan_block_rsv) {
3228 root->orphan_block_rsv = block_rsv;
3229 } else if (block_rsv) {
3230 btrfs_free_block_rsv(root, block_rsv);
3231 block_rsv = NULL;
3232 }
Josef Bacik7b128762008-07-24 12:17:14 -04003233
Josef Bacik8a35d952012-05-23 14:26:42 -04003234 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3235 &BTRFS_I(inode)->runtime_flags)) {
Yan, Zhengd68fc572010-05-16 10:49:58 -04003236#if 0
3237 /*
3238 * For proper ENOSPC handling, we should do orphan
3239 * cleanup when mounting. But this introduces backward
3240 * compatibility issue.
3241 */
3242 if (!xchg(&root->orphan_item_inserted, 1))
3243 insert = 2;
3244 else
3245 insert = 1;
3246#endif
3247 insert = 1;
Miao Xie321f0e72012-08-28 22:13:02 -06003248 atomic_inc(&root->orphan_inodes);
Yan, Zhengd68fc572010-05-16 10:49:58 -04003249 }
Josef Bacik7b128762008-07-24 12:17:14 -04003250
Josef Bacik72ac3c02012-05-23 14:13:11 -04003251 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3252 &BTRFS_I(inode)->runtime_flags))
Yan, Zhengd68fc572010-05-16 10:49:58 -04003253 reserve = 1;
Yan, Zhengd68fc572010-05-16 10:49:58 -04003254 spin_unlock(&root->orphan_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04003255
Yan, Zhengd68fc572010-05-16 10:49:58 -04003256 /* grab metadata reservation from transaction handle */
3257 if (reserve) {
3258 ret = btrfs_orphan_reserve_metadata(trans, inode);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003259 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
Yan, Zhengd68fc572010-05-16 10:49:58 -04003260 }
3261
3262 /* insert an orphan item to track this unlinked/truncated file */
3263 if (insert >= 1) {
Li Zefan33345d012011-04-20 10:31:50 +08003264 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
Josef Bacik4ef31a42013-08-13 14:10:08 -04003265 if (ret) {
Filipe David Borba Manana703c88e2013-09-22 21:54:55 +01003266 atomic_dec(&root->orphan_inodes);
Josef Bacik4ef31a42013-08-13 14:10:08 -04003267 if (reserve) {
3268 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3269 &BTRFS_I(inode)->runtime_flags);
3270 btrfs_orphan_release_metadata(inode);
3271 }
3272 if (ret != -EEXIST) {
Josef Bacike8e7cff2013-08-21 15:54:00 -04003273 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3274 &BTRFS_I(inode)->runtime_flags);
Josef Bacik4ef31a42013-08-13 14:10:08 -04003275 btrfs_abort_transaction(trans, root, ret);
3276 return ret;
3277 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003278 }
3279 ret = 0;
Yan, Zhengd68fc572010-05-16 10:49:58 -04003280 }
3281
3282 /* insert an orphan item to track subvolume contains orphan files */
3283 if (insert >= 2) {
3284 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3285 root->root_key.objectid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003286 if (ret && ret != -EEXIST) {
3287 btrfs_abort_transaction(trans, root, ret);
3288 return ret;
3289 }
Yan, Zhengd68fc572010-05-16 10:49:58 -04003290 }
3291 return 0;
Josef Bacik7b128762008-07-24 12:17:14 -04003292}
3293
3294/*
3295 * We have done the truncate/delete so we can go ahead and remove the orphan
3296 * item for this particular inode.
3297 */
Eric Sandeen48a3b632013-04-25 20:41:01 +00003298static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3299 struct inode *inode)
Josef Bacik7b128762008-07-24 12:17:14 -04003300{
3301 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zhengd68fc572010-05-16 10:49:58 -04003302 int delete_item = 0;
3303 int release_rsv = 0;
Josef Bacik7b128762008-07-24 12:17:14 -04003304 int ret = 0;
3305
Yan, Zhengd68fc572010-05-16 10:49:58 -04003306 spin_lock(&root->orphan_lock);
Josef Bacik8a35d952012-05-23 14:26:42 -04003307 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3308 &BTRFS_I(inode)->runtime_flags))
Yan, Zhengd68fc572010-05-16 10:49:58 -04003309 delete_item = 1;
Josef Bacik7b128762008-07-24 12:17:14 -04003310
Josef Bacik72ac3c02012-05-23 14:13:11 -04003311 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3312 &BTRFS_I(inode)->runtime_flags))
Yan, Zhengd68fc572010-05-16 10:49:58 -04003313 release_rsv = 1;
Yan, Zhengd68fc572010-05-16 10:49:58 -04003314 spin_unlock(&root->orphan_lock);
3315
Filipe David Borba Manana703c88e2013-09-22 21:54:55 +01003316 if (delete_item) {
Josef Bacik8a35d952012-05-23 14:26:42 -04003317 atomic_dec(&root->orphan_inodes);
Filipe David Borba Manana703c88e2013-09-22 21:54:55 +01003318 if (trans)
3319 ret = btrfs_del_orphan_item(trans, root,
3320 btrfs_ino(inode));
Josef Bacik8a35d952012-05-23 14:26:42 -04003321 }
Josef Bacik7b128762008-07-24 12:17:14 -04003322
Filipe David Borba Manana703c88e2013-09-22 21:54:55 +01003323 if (release_rsv)
3324 btrfs_orphan_release_metadata(inode);
3325
Josef Bacik4ef31a42013-08-13 14:10:08 -04003326 return ret;
Josef Bacik7b128762008-07-24 12:17:14 -04003327}
3328
3329/*
3330 * this cleans up any orphans that may be left on the list from the last use
3331 * of this root.
3332 */
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003333int btrfs_orphan_cleanup(struct btrfs_root *root)
Josef Bacik7b128762008-07-24 12:17:14 -04003334{
3335 struct btrfs_path *path;
3336 struct extent_buffer *leaf;
Josef Bacik7b128762008-07-24 12:17:14 -04003337 struct btrfs_key key, found_key;
3338 struct btrfs_trans_handle *trans;
3339 struct inode *inode;
Josef Bacik8f6d7f42011-09-26 15:55:20 -04003340 u64 last_objectid = 0;
Josef Bacik7b128762008-07-24 12:17:14 -04003341 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3342
Yan, Zhengd68fc572010-05-16 10:49:58 -04003343 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003344 return 0;
Yan, Zhengc71bf092009-11-12 09:34:40 +00003345
3346 path = btrfs_alloc_path();
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003347 if (!path) {
3348 ret = -ENOMEM;
3349 goto out;
3350 }
David Sterbae4058b52015-11-27 16:31:35 +01003351 path->reada = READA_BACK;
Josef Bacik7b128762008-07-24 12:17:14 -04003352
3353 key.objectid = BTRFS_ORPHAN_OBJECTID;
David Sterba962a2982014-06-04 18:41:45 +02003354 key.type = BTRFS_ORPHAN_ITEM_KEY;
Josef Bacik7b128762008-07-24 12:17:14 -04003355 key.offset = (u64)-1;
3356
Josef Bacik7b128762008-07-24 12:17:14 -04003357 while (1) {
3358 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003359 if (ret < 0)
3360 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04003361
3362 /*
3363 * if ret == 0 means we found what we were searching for, which
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003364 * is weird, but possible, so only screw with path if we didn't
Josef Bacik7b128762008-07-24 12:17:14 -04003365 * find the key and see if we have stuff that matches
3366 */
3367 if (ret > 0) {
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003368 ret = 0;
Josef Bacik7b128762008-07-24 12:17:14 -04003369 if (path->slots[0] == 0)
3370 break;
3371 path->slots[0]--;
3372 }
3373
3374 /* pull out the item */
3375 leaf = path->nodes[0];
Josef Bacik7b128762008-07-24 12:17:14 -04003376 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3377
3378 /* make sure the item matches what we want */
3379 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3380 break;
David Sterba962a2982014-06-04 18:41:45 +02003381 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
Josef Bacik7b128762008-07-24 12:17:14 -04003382 break;
3383
3384 /* release the path since we're done with it */
David Sterbab3b4aa72011-04-21 01:20:15 +02003385 btrfs_release_path(path);
Josef Bacik7b128762008-07-24 12:17:14 -04003386
3387 /*
3388 * this is where we are basically btrfs_lookup, without the
3389 * crossing root thing. we store the inode number in the
3390 * offset of the orphan item.
3391 */
Josef Bacik8f6d7f42011-09-26 15:55:20 -04003392
3393 if (found_key.offset == last_objectid) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00003394 btrfs_err(root->fs_info,
3395 "Error removing orphan entry, stopping orphan cleanup");
Josef Bacik8f6d7f42011-09-26 15:55:20 -04003396 ret = -EINVAL;
3397 goto out;
3398 }
3399
3400 last_objectid = found_key.offset;
3401
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003402 found_key.objectid = found_key.offset;
3403 found_key.type = BTRFS_INODE_ITEM_KEY;
3404 found_key.offset = 0;
Josef Bacik73f73412009-12-04 17:38:27 +00003405 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
Rusty Russell8c6ffba2013-07-15 11:20:32 +09303406 ret = PTR_ERR_OR_ZERO(inode);
Josef Bacika8c9e572011-09-21 16:55:59 -04003407 if (ret && ret != -ESTALE)
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003408 goto out;
Josef Bacika8c9e572011-09-21 16:55:59 -04003409
Arne Jansenf8e9e0b2011-12-14 20:12:02 -05003410 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3411 struct btrfs_root *dead_root;
3412 struct btrfs_fs_info *fs_info = root->fs_info;
3413 int is_dead_root = 0;
3414
3415 /*
3416 * this is an orphan in the tree root. Currently these
3417 * could come from 2 sources:
3418 * a) a snapshot deletion in progress
3419 * b) a free space cache inode
3420 * We need to distinguish those two, as the snapshot
3421 * orphan must not get deleted.
3422 * find_dead_roots already ran before us, so if this
3423 * is a snapshot deletion, we should find the root
3424 * in the dead_roots list
3425 */
3426 spin_lock(&fs_info->trans_lock);
3427 list_for_each_entry(dead_root, &fs_info->dead_roots,
3428 root_list) {
3429 if (dead_root->root_key.objectid ==
3430 found_key.objectid) {
3431 is_dead_root = 1;
3432 break;
3433 }
3434 }
3435 spin_unlock(&fs_info->trans_lock);
3436 if (is_dead_root) {
3437 /* prevent this orphan from being found again */
3438 key.offset = found_key.objectid - 1;
3439 continue;
3440 }
3441 }
Josef Bacika8c9e572011-09-21 16:55:59 -04003442 /*
3443 * Inode is already gone but the orphan item is still there,
3444 * kill the orphan item.
3445 */
3446 if (ret == -ESTALE) {
3447 trans = btrfs_start_transaction(root, 1);
3448 if (IS_ERR(trans)) {
3449 ret = PTR_ERR(trans);
3450 goto out;
3451 }
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00003452 btrfs_debug(root->fs_info, "auto deleting %Lu",
3453 found_key.objectid);
Josef Bacika8c9e572011-09-21 16:55:59 -04003454 ret = btrfs_del_orphan_item(trans, root,
3455 found_key.objectid);
Josef Bacika8c9e572011-09-21 16:55:59 -04003456 btrfs_end_transaction(trans, root);
Josef Bacik4ef31a42013-08-13 14:10:08 -04003457 if (ret)
3458 goto out;
Josef Bacika8c9e572011-09-21 16:55:59 -04003459 continue;
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003460 }
Josef Bacik7b128762008-07-24 12:17:14 -04003461
Josef Bacik7b128762008-07-24 12:17:14 -04003462 /*
3463 * add this inode to the orphan list so btrfs_orphan_del does
3464 * the proper thing when we hit it
3465 */
Josef Bacik8a35d952012-05-23 14:26:42 -04003466 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3467 &BTRFS_I(inode)->runtime_flags);
Josef Bacik925396e2013-02-01 15:57:47 -05003468 atomic_inc(&root->orphan_inodes);
Josef Bacik7b128762008-07-24 12:17:14 -04003469
Josef Bacik7b128762008-07-24 12:17:14 -04003470 /* if we have links, this was a truncate, lets do that */
3471 if (inode->i_nlink) {
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05303472 if (WARN_ON(!S_ISREG(inode->i_mode))) {
Josef Bacika41ad392011-01-31 15:30:16 -05003473 iput(inode);
3474 continue;
3475 }
Josef Bacik7b128762008-07-24 12:17:14 -04003476 nr_truncate++;
Josef Bacikf3fe8202013-01-07 17:03:21 -05003477
3478 /* 1 for the orphan item deletion. */
3479 trans = btrfs_start_transaction(root, 1);
3480 if (IS_ERR(trans)) {
Josef Bacikc69b26b2013-06-03 16:51:23 -04003481 iput(inode);
Josef Bacikf3fe8202013-01-07 17:03:21 -05003482 ret = PTR_ERR(trans);
3483 goto out;
3484 }
3485 ret = btrfs_orphan_add(trans, inode);
3486 btrfs_end_transaction(trans, root);
Josef Bacikc69b26b2013-06-03 16:51:23 -04003487 if (ret) {
3488 iput(inode);
Josef Bacikf3fe8202013-01-07 17:03:21 -05003489 goto out;
Josef Bacikc69b26b2013-06-03 16:51:23 -04003490 }
Josef Bacikf3fe8202013-01-07 17:03:21 -05003491
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003492 ret = btrfs_truncate(inode);
Josef Bacik4a7d0f62013-02-07 16:27:28 -05003493 if (ret)
3494 btrfs_orphan_del(NULL, inode);
Josef Bacik7b128762008-07-24 12:17:14 -04003495 } else {
3496 nr_unlink++;
3497 }
3498
3499 /* this will do delete_inode and everything for us */
3500 iput(inode);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003501 if (ret)
3502 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04003503 }
Miao Xie3254c872011-11-10 20:45:05 -05003504 /* release the path since we're done with it */
3505 btrfs_release_path(path);
3506
Yan, Zhengd68fc572010-05-16 10:49:58 -04003507 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3508
3509 if (root->orphan_block_rsv)
3510 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3511 (u64)-1);
3512
Miao Xie27cdeb72014-04-02 19:51:05 +08003513 if (root->orphan_block_rsv ||
3514 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
Josef Bacik7a7eaa42011-04-13 12:54:33 -04003515 trans = btrfs_join_transaction(root);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003516 if (!IS_ERR(trans))
3517 btrfs_end_transaction(trans, root);
Yan, Zhengd68fc572010-05-16 10:49:58 -04003518 }
Josef Bacik7b128762008-07-24 12:17:14 -04003519
3520 if (nr_unlink)
David Sterba4884b472013-03-20 13:31:27 +00003521 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
Josef Bacik7b128762008-07-24 12:17:14 -04003522 if (nr_truncate)
David Sterba4884b472013-03-20 13:31:27 +00003523 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003524
3525out:
3526 if (ret)
David Sterba68b663d2014-12-19 18:38:37 +01003527 btrfs_err(root->fs_info,
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00003528 "could not do orphan cleanup %d", ret);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003529 btrfs_free_path(path);
3530 return ret;
Josef Bacik7b128762008-07-24 12:17:14 -04003531}
3532
Chris Masond352ac62008-09-29 15:18:18 -04003533/*
Chris Mason46a53cc2009-04-27 11:47:50 -04003534 * very simple check to peek ahead in the leaf looking for xattrs. If we
3535 * don't find any xattrs, we know there can't be any acls.
3536 *
3537 * slot is the slot the inode is in, objectid is the objectid of the inode
3538 */
3539static noinline int acls_after_inode_item(struct extent_buffer *leaf,
Filipe David Borba Manana63541922014-01-07 11:47:46 +00003540 int slot, u64 objectid,
3541 int *first_xattr_slot)
Chris Mason46a53cc2009-04-27 11:47:50 -04003542{
3543 u32 nritems = btrfs_header_nritems(leaf);
3544 struct btrfs_key found_key;
Josef Bacikf23b5a52013-06-19 10:16:26 -04003545 static u64 xattr_access = 0;
3546 static u64 xattr_default = 0;
Chris Mason46a53cc2009-04-27 11:47:50 -04003547 int scanned = 0;
3548
Josef Bacikf23b5a52013-06-19 10:16:26 -04003549 if (!xattr_access) {
Andreas Gruenbacher97d79292015-12-02 14:44:35 +01003550 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3551 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3552 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3553 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
Josef Bacikf23b5a52013-06-19 10:16:26 -04003554 }
3555
Chris Mason46a53cc2009-04-27 11:47:50 -04003556 slot++;
Filipe David Borba Manana63541922014-01-07 11:47:46 +00003557 *first_xattr_slot = -1;
Chris Mason46a53cc2009-04-27 11:47:50 -04003558 while (slot < nritems) {
3559 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3560
3561 /* we found a different objectid, there must not be acls */
3562 if (found_key.objectid != objectid)
3563 return 0;
3564
3565 /* we found an xattr, assume we've got an acl */
Josef Bacikf23b5a52013-06-19 10:16:26 -04003566 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
Filipe David Borba Manana63541922014-01-07 11:47:46 +00003567 if (*first_xattr_slot == -1)
3568 *first_xattr_slot = slot;
Josef Bacikf23b5a52013-06-19 10:16:26 -04003569 if (found_key.offset == xattr_access ||
3570 found_key.offset == xattr_default)
3571 return 1;
3572 }
Chris Mason46a53cc2009-04-27 11:47:50 -04003573
3574 /*
3575 * we found a key greater than an xattr key, there can't
3576 * be any acls later on
3577 */
3578 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3579 return 0;
3580
3581 slot++;
3582 scanned++;
3583
3584 /*
3585 * it goes inode, inode backrefs, xattrs, extents,
3586 * so if there are a ton of hard links to an inode there can
3587 * be a lot of backrefs. Don't waste time searching too hard,
3588 * this is just an optimization
3589 */
3590 if (scanned >= 8)
3591 break;
3592 }
3593 /* we hit the end of the leaf before we found an xattr or
3594 * something larger than an xattr. We have to assume the inode
3595 * has acls
3596 */
Filipe David Borba Manana63541922014-01-07 11:47:46 +00003597 if (*first_xattr_slot == -1)
3598 *first_xattr_slot = slot;
Chris Mason46a53cc2009-04-27 11:47:50 -04003599 return 1;
3600}
3601
3602/*
Chris Masond352ac62008-09-29 15:18:18 -04003603 * read an inode from the btree into the in-memory inode
3604 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003605static void btrfs_read_locked_inode(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04003606{
3607 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04003608 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04003609 struct btrfs_inode_item *inode_item;
3610 struct btrfs_root *root = BTRFS_I(inode)->root;
3611 struct btrfs_key location;
Miao Xie67de1172013-12-26 13:07:06 +08003612 unsigned long ptr;
Chris Mason46a53cc2009-04-27 11:47:50 -04003613 int maybe_acls;
Josef Bacik618e21d2007-07-11 10:18:17 -04003614 u32 rdev;
Chris Mason39279cc2007-06-12 06:35:45 -04003615 int ret;
Miao Xie2f7e33d2011-06-23 07:27:13 +00003616 bool filled = false;
Filipe David Borba Manana63541922014-01-07 11:47:46 +00003617 int first_xattr_slot;
Miao Xie2f7e33d2011-06-23 07:27:13 +00003618
3619 ret = btrfs_fill_inode(inode, &rdev);
3620 if (!ret)
3621 filled = true;
Chris Mason39279cc2007-06-12 06:35:45 -04003622
3623 path = btrfs_alloc_path();
Mark Fasheh1748f842011-07-12 11:25:31 -07003624 if (!path)
3625 goto make_bad;
3626
Chris Mason39279cc2007-06-12 06:35:45 -04003627 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
Chris Masondc17ff82008-01-08 15:46:30 -05003628
Chris Mason39279cc2007-06-12 06:35:45 -04003629 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
Chris Mason5f39d392007-10-15 16:14:19 -04003630 if (ret)
Chris Mason39279cc2007-06-12 06:35:45 -04003631 goto make_bad;
Chris Mason39279cc2007-06-12 06:35:45 -04003632
Chris Mason5f39d392007-10-15 16:14:19 -04003633 leaf = path->nodes[0];
Miao Xie2f7e33d2011-06-23 07:27:13 +00003634
3635 if (filled)
Miao Xie67de1172013-12-26 13:07:06 +08003636 goto cache_index;
Miao Xie2f7e33d2011-06-23 07:27:13 +00003637
Chris Mason5f39d392007-10-15 16:14:19 -04003638 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3639 struct btrfs_inode_item);
Chris Mason5f39d392007-10-15 16:14:19 -04003640 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
Miklos Szeredibfe86842011-10-28 14:13:29 +02003641 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
Eric W. Biederman2f2f43d2012-02-10 11:05:07 -08003642 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3643 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
Chris Masondbe674a2008-07-17 12:54:05 -04003644 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
Chris Mason5f39d392007-10-15 16:14:19 -04003645
David Sterbaa937b972014-12-12 17:39:12 +01003646 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3647 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
Chris Mason5f39d392007-10-15 16:14:19 -04003648
David Sterbaa937b972014-12-12 17:39:12 +01003649 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3650 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
Chris Mason5f39d392007-10-15 16:14:19 -04003651
David Sterbaa937b972014-12-12 17:39:12 +01003652 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3653 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
Chris Mason5f39d392007-10-15 16:14:19 -04003654
chandan r9cc97d62012-07-04 12:48:07 +05303655 BTRFS_I(inode)->i_otime.tv_sec =
3656 btrfs_timespec_sec(leaf, &inode_item->otime);
3657 BTRFS_I(inode)->i_otime.tv_nsec =
3658 btrfs_timespec_nsec(leaf, &inode_item->otime);
Chris Mason5f39d392007-10-15 16:14:19 -04003659
Yan Zhenga76a3cd2008-10-09 11:46:29 -04003660 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
Chris Masone02119d2008-09-05 16:13:11 -04003661 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
Josef Bacik5dc562c2012-08-17 13:14:17 -04003662 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3663
Josef Bacik0c4d2d92012-04-05 15:03:02 -04003664 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
Chris Masone02119d2008-09-05 16:13:11 -04003665 inode->i_generation = BTRFS_I(inode)->generation;
Josef Bacik618e21d2007-07-11 10:18:17 -04003666 inode->i_rdev = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04003667 rdev = btrfs_inode_rdev(leaf, inode_item);
3668
Josef Bacikaec74772008-07-24 12:12:38 -04003669 BTRFS_I(inode)->index_cnt = (u64)-1;
Yan Zhengd2fb3432008-12-11 16:30:39 -05003670 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
Miao Xie67de1172013-12-26 13:07:06 +08003671
3672cache_index:
Yang Dongsheng6e17d302015-04-09 12:08:43 +08003673 /*
3674 * If we were modified in the current generation and evicted from memory
3675 * and then re-read we need to do a full sync since we don't have any
3676 * idea about which extents were modified before we were evicted from
3677 * cache.
3678 *
3679 * This is required for both inode re-read from disk and delayed inode
3680 * in delayed_nodes_tree.
3681 */
3682 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3683 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3684 &BTRFS_I(inode)->runtime_flags);
3685
Filipe Mananabde6c242015-07-24 00:00:19 +01003686 /*
3687 * We don't persist the id of the transaction where an unlink operation
3688 * against the inode was last made. So here we assume the inode might
3689 * have been evicted, and therefore the exact value of last_unlink_trans
3690 * lost, and set it to last_trans to avoid metadata inconsistencies
3691 * between the inode and its parent if the inode is fsync'ed and the log
3692 * replayed. For example, in the scenario:
3693 *
3694 * touch mydir/foo
3695 * ln mydir/foo mydir/bar
3696 * sync
3697 * unlink mydir/bar
3698 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3699 * xfs_io -c fsync mydir/foo
3700 * <power failure>
3701 * mount fs, triggers fsync log replay
3702 *
3703 * We must make sure that when we fsync our inode foo we also log its
3704 * parent inode, otherwise after log replay the parent still has the
3705 * dentry with the "bar" name but our inode foo has a link count of 1
3706 * and doesn't have an inode ref with the name "bar" anymore.
3707 *
3708 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3709 * but it guarantees correctness at the expense of ocassional full
3710 * transaction commits on fsync if our inode is a directory, or if our
3711 * inode is not a directory, logging its parent unnecessarily.
3712 */
3713 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3714
Miao Xie67de1172013-12-26 13:07:06 +08003715 path->slots[0]++;
3716 if (inode->i_nlink != 1 ||
3717 path->slots[0] >= btrfs_header_nritems(leaf))
3718 goto cache_acl;
3719
3720 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3721 if (location.objectid != btrfs_ino(inode))
3722 goto cache_acl;
3723
3724 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3725 if (location.type == BTRFS_INODE_REF_KEY) {
3726 struct btrfs_inode_ref *ref;
3727
3728 ref = (struct btrfs_inode_ref *)ptr;
3729 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3730 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3731 struct btrfs_inode_extref *extref;
3732
3733 extref = (struct btrfs_inode_extref *)ptr;
3734 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3735 extref);
3736 }
Miao Xie2f7e33d2011-06-23 07:27:13 +00003737cache_acl:
Chris Mason46a53cc2009-04-27 11:47:50 -04003738 /*
3739 * try to precache a NULL acl entry for files that don't have
3740 * any xattrs or acls
3741 */
Li Zefan33345d012011-04-20 10:31:50 +08003742 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
Filipe David Borba Manana63541922014-01-07 11:47:46 +00003743 btrfs_ino(inode), &first_xattr_slot);
3744 if (first_xattr_slot != -1) {
3745 path->slots[0] = first_xattr_slot;
3746 ret = btrfs_load_inode_props(inode, path);
3747 if (ret)
3748 btrfs_err(root->fs_info,
David Sterba351fd352014-05-15 16:48:20 +02003749 "error loading props for ino %llu (root %llu): %d",
Filipe David Borba Manana63541922014-01-07 11:47:46 +00003750 btrfs_ino(inode),
3751 root->root_key.objectid, ret);
3752 }
3753 btrfs_free_path(path);
3754
Al Viro72c04902009-06-24 16:58:48 -04003755 if (!maybe_acls)
3756 cache_no_acl(inode);
Chris Mason46a53cc2009-04-27 11:47:50 -04003757
Chris Mason39279cc2007-06-12 06:35:45 -04003758 switch (inode->i_mode & S_IFMT) {
Chris Mason39279cc2007-06-12 06:35:45 -04003759 case S_IFREG:
3760 inode->i_mapping->a_ops = &btrfs_aops;
Chris Masond1310b22008-01-24 16:13:08 -05003761 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -04003762 inode->i_fop = &btrfs_file_operations;
3763 inode->i_op = &btrfs_file_inode_operations;
3764 break;
3765 case S_IFDIR:
3766 inode->i_fop = &btrfs_dir_file_operations;
3767 if (root == root->fs_info->tree_root)
3768 inode->i_op = &btrfs_dir_ro_inode_operations;
3769 else
3770 inode->i_op = &btrfs_dir_inode_operations;
3771 break;
3772 case S_IFLNK:
3773 inode->i_op = &btrfs_symlink_inode_operations;
Al Viro21fc61c2015-11-17 01:07:57 -05003774 inode_nohighmem(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04003775 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3776 break;
Josef Bacik618e21d2007-07-11 10:18:17 -04003777 default:
Jim Owens0279b4c2009-02-04 09:29:13 -05003778 inode->i_op = &btrfs_special_inode_operations;
Josef Bacik618e21d2007-07-11 10:18:17 -04003779 init_special_inode(inode, inode->i_mode, rdev);
3780 break;
Chris Mason39279cc2007-06-12 06:35:45 -04003781 }
Christoph Hellwig6cbff002009-04-17 10:37:41 +02003782
3783 btrfs_update_iflags(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04003784 return;
3785
3786make_bad:
Chris Mason39279cc2007-06-12 06:35:45 -04003787 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04003788 make_bad_inode(inode);
3789}
3790
Chris Masond352ac62008-09-29 15:18:18 -04003791/*
3792 * given a leaf and an inode, copy the inode fields into the leaf
3793 */
Chris Masone02119d2008-09-05 16:13:11 -04003794static void fill_inode_item(struct btrfs_trans_handle *trans,
3795 struct extent_buffer *leaf,
Chris Mason5f39d392007-10-15 16:14:19 -04003796 struct btrfs_inode_item *item,
Chris Mason39279cc2007-06-12 06:35:45 -04003797 struct inode *inode)
3798{
Liu Bo51fab692012-12-27 09:01:21 +00003799 struct btrfs_map_token token;
Chris Mason5f39d392007-10-15 16:14:19 -04003800
Liu Bo51fab692012-12-27 09:01:21 +00003801 btrfs_init_map_token(&token);
Chris Mason5f39d392007-10-15 16:14:19 -04003802
Liu Bo51fab692012-12-27 09:01:21 +00003803 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3804 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3805 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3806 &token);
3807 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3808 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
Chris Mason5f39d392007-10-15 16:14:19 -04003809
David Sterbaa937b972014-12-12 17:39:12 +01003810 btrfs_set_token_timespec_sec(leaf, &item->atime,
Liu Bo51fab692012-12-27 09:01:21 +00003811 inode->i_atime.tv_sec, &token);
David Sterbaa937b972014-12-12 17:39:12 +01003812 btrfs_set_token_timespec_nsec(leaf, &item->atime,
Liu Bo51fab692012-12-27 09:01:21 +00003813 inode->i_atime.tv_nsec, &token);
Chris Mason5f39d392007-10-15 16:14:19 -04003814
David Sterbaa937b972014-12-12 17:39:12 +01003815 btrfs_set_token_timespec_sec(leaf, &item->mtime,
Liu Bo51fab692012-12-27 09:01:21 +00003816 inode->i_mtime.tv_sec, &token);
David Sterbaa937b972014-12-12 17:39:12 +01003817 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
Liu Bo51fab692012-12-27 09:01:21 +00003818 inode->i_mtime.tv_nsec, &token);
3819
David Sterbaa937b972014-12-12 17:39:12 +01003820 btrfs_set_token_timespec_sec(leaf, &item->ctime,
Liu Bo51fab692012-12-27 09:01:21 +00003821 inode->i_ctime.tv_sec, &token);
David Sterbaa937b972014-12-12 17:39:12 +01003822 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
Liu Bo51fab692012-12-27 09:01:21 +00003823 inode->i_ctime.tv_nsec, &token);
3824
chandan r9cc97d62012-07-04 12:48:07 +05303825 btrfs_set_token_timespec_sec(leaf, &item->otime,
3826 BTRFS_I(inode)->i_otime.tv_sec, &token);
3827 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3828 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3829
Liu Bo51fab692012-12-27 09:01:21 +00003830 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3831 &token);
3832 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3833 &token);
3834 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3835 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3836 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3837 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3838 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
Chris Mason39279cc2007-06-12 06:35:45 -04003839}
3840
Chris Masond352ac62008-09-29 15:18:18 -04003841/*
3842 * copy everything in the in-memory inode into the btree.
3843 */
Chris Mason21151332011-11-10 20:39:08 -05003844static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
Chris Masond3977122009-01-05 21:25:51 -05003845 struct btrfs_root *root, struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04003846{
3847 struct btrfs_inode_item *inode_item;
3848 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04003849 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04003850 int ret;
3851
3852 path = btrfs_alloc_path();
Miao Xie16cdcec2011-04-22 18:12:22 +08003853 if (!path)
3854 return -ENOMEM;
3855
Chris Masonb9473432009-03-13 11:00:37 -04003856 path->leave_spinning = 1;
Miao Xie16cdcec2011-04-22 18:12:22 +08003857 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3858 1);
Chris Mason39279cc2007-06-12 06:35:45 -04003859 if (ret) {
3860 if (ret > 0)
3861 ret = -ENOENT;
3862 goto failed;
3863 }
3864
Chris Mason5f39d392007-10-15 16:14:19 -04003865 leaf = path->nodes[0];
3866 inode_item = btrfs_item_ptr(leaf, path->slots[0],
Miao Xie16cdcec2011-04-22 18:12:22 +08003867 struct btrfs_inode_item);
Chris Mason39279cc2007-06-12 06:35:45 -04003868
Chris Masone02119d2008-09-05 16:13:11 -04003869 fill_inode_item(trans, leaf, inode_item, inode);
Chris Mason5f39d392007-10-15 16:14:19 -04003870 btrfs_mark_buffer_dirty(leaf);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04003871 btrfs_set_inode_last_trans(trans, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04003872 ret = 0;
3873failed:
Chris Mason39279cc2007-06-12 06:35:45 -04003874 btrfs_free_path(path);
3875 return ret;
3876}
3877
Chris Masond352ac62008-09-29 15:18:18 -04003878/*
Chris Mason21151332011-11-10 20:39:08 -05003879 * copy everything in the in-memory inode into the btree.
3880 */
3881noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3882 struct btrfs_root *root, struct inode *inode)
3883{
3884 int ret;
3885
3886 /*
3887 * If the inode is a free space inode, we can deadlock during commit
3888 * if we put it into the delayed code.
3889 *
3890 * The data relocation inode should also be directly updated
3891 * without delay
3892 */
Liu Bo83eea1f2012-07-10 05:28:39 -06003893 if (!btrfs_is_free_space_inode(inode)
Josef Bacik1d52c782014-09-18 11:30:44 -04003894 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3895 && !root->fs_info->log_root_recovering) {
Alexander Block8ea05e32012-07-25 17:35:53 +02003896 btrfs_update_root_times(trans, root);
3897
Chris Mason21151332011-11-10 20:39:08 -05003898 ret = btrfs_delayed_update_inode(trans, root, inode);
3899 if (!ret)
3900 btrfs_set_inode_last_trans(trans, inode);
3901 return ret;
3902 }
3903
3904 return btrfs_update_inode_item(trans, root, inode);
3905}
3906
Josef Bacikbe6aef62012-10-22 15:43:12 -04003907noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3908 struct btrfs_root *root,
3909 struct inode *inode)
Chris Mason21151332011-11-10 20:39:08 -05003910{
3911 int ret;
3912
3913 ret = btrfs_update_inode(trans, root, inode);
3914 if (ret == -ENOSPC)
3915 return btrfs_update_inode_item(trans, root, inode);
3916 return ret;
3917}
3918
3919/*
Chris Masond352ac62008-09-29 15:18:18 -04003920 * unlink helper that gets used here in inode.c and in the tree logging
3921 * recovery code. It remove a link in a directory with a given name, and
3922 * also drops the back refs in the inode to the directory
3923 */
Al Viro92986792011-03-04 17:14:37 +00003924static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3925 struct btrfs_root *root,
3926 struct inode *dir, struct inode *inode,
3927 const char *name, int name_len)
Chris Mason39279cc2007-06-12 06:35:45 -04003928{
3929 struct btrfs_path *path;
Chris Mason39279cc2007-06-12 06:35:45 -04003930 int ret = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04003931 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04003932 struct btrfs_dir_item *di;
Chris Mason5f39d392007-10-15 16:14:19 -04003933 struct btrfs_key key;
Josef Bacikaec74772008-07-24 12:12:38 -04003934 u64 index;
Li Zefan33345d012011-04-20 10:31:50 +08003935 u64 ino = btrfs_ino(inode);
3936 u64 dir_ino = btrfs_ino(dir);
Chris Mason39279cc2007-06-12 06:35:45 -04003937
3938 path = btrfs_alloc_path();
Chris Mason54aa1f42007-06-22 14:16:25 -04003939 if (!path) {
3940 ret = -ENOMEM;
Tsutomu Itoh554233a2011-02-03 03:16:25 +00003941 goto out;
Chris Mason54aa1f42007-06-22 14:16:25 -04003942 }
3943
Chris Masonb9473432009-03-13 11:00:37 -04003944 path->leave_spinning = 1;
Li Zefan33345d012011-04-20 10:31:50 +08003945 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
Chris Mason39279cc2007-06-12 06:35:45 -04003946 name, name_len, -1);
3947 if (IS_ERR(di)) {
3948 ret = PTR_ERR(di);
3949 goto err;
3950 }
3951 if (!di) {
3952 ret = -ENOENT;
3953 goto err;
3954 }
Chris Mason5f39d392007-10-15 16:14:19 -04003955 leaf = path->nodes[0];
3956 btrfs_dir_item_key_to_cpu(leaf, di, &key);
Chris Mason39279cc2007-06-12 06:35:45 -04003957 ret = btrfs_delete_one_dir_name(trans, root, path, di);
Chris Mason54aa1f42007-06-22 14:16:25 -04003958 if (ret)
3959 goto err;
David Sterbab3b4aa72011-04-21 01:20:15 +02003960 btrfs_release_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04003961
Miao Xie67de1172013-12-26 13:07:06 +08003962 /*
3963 * If we don't have dir index, we have to get it by looking up
3964 * the inode ref, since we get the inode ref, remove it directly,
3965 * it is unnecessary to do delayed deletion.
3966 *
3967 * But if we have dir index, needn't search inode ref to get it.
3968 * Since the inode ref is close to the inode item, it is better
3969 * that we delay to delete it, and just do this deletion when
3970 * we update the inode item.
3971 */
3972 if (BTRFS_I(inode)->dir_index) {
3973 ret = btrfs_delayed_delete_inode_ref(inode);
3974 if (!ret) {
3975 index = BTRFS_I(inode)->dir_index;
3976 goto skip_backref;
3977 }
3978 }
3979
Li Zefan33345d012011-04-20 10:31:50 +08003980 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3981 dir_ino, &index);
Josef Bacikaec74772008-07-24 12:12:38 -04003982 if (ret) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00003983 btrfs_info(root->fs_info,
3984 "failed to delete reference to %.*s, inode %llu parent %llu",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02003985 name_len, name, ino, dir_ino);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003986 btrfs_abort_transaction(trans, root, ret);
Josef Bacikaec74772008-07-24 12:12:38 -04003987 goto err;
3988 }
Miao Xie67de1172013-12-26 13:07:06 +08003989skip_backref:
Miao Xie16cdcec2011-04-22 18:12:22 +08003990 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003991 if (ret) {
3992 btrfs_abort_transaction(trans, root, ret);
Chris Mason39279cc2007-06-12 06:35:45 -04003993 goto err;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003994 }
Chris Mason39279cc2007-06-12 06:35:45 -04003995
Chris Masone02119d2008-09-05 16:13:11 -04003996 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
Li Zefan33345d012011-04-20 10:31:50 +08003997 inode, dir_ino);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003998 if (ret != 0 && ret != -ENOENT) {
3999 btrfs_abort_transaction(trans, root, ret);
4000 goto err;
4001 }
Chris Masone02119d2008-09-05 16:13:11 -04004002
4003 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4004 dir, index);
Chris Mason6418c962010-10-30 07:34:24 -04004005 if (ret == -ENOENT)
4006 ret = 0;
Zach Brownd4e39912013-04-02 21:02:16 +00004007 else if (ret)
4008 btrfs_abort_transaction(trans, root, ret);
Chris Mason39279cc2007-06-12 06:35:45 -04004009err:
4010 btrfs_free_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04004011 if (ret)
4012 goto out;
4013
4014 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
Josef Bacik0c4d2d92012-04-05 15:03:02 -04004015 inode_inc_iversion(inode);
4016 inode_inc_iversion(dir);
Deepa Dinamani04b285f2016-02-06 23:57:21 -08004017 inode->i_ctime = dir->i_mtime =
4018 dir->i_ctime = current_fs_time(inode->i_sb);
Tsutomu Itohb9959292012-06-25 21:25:22 -06004019 ret = btrfs_update_inode(trans, root, dir);
Chris Masone02119d2008-09-05 16:13:11 -04004020out:
Chris Mason39279cc2007-06-12 06:35:45 -04004021 return ret;
4022}
4023
Al Viro92986792011-03-04 17:14:37 +00004024int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4025 struct btrfs_root *root,
4026 struct inode *dir, struct inode *inode,
4027 const char *name, int name_len)
4028{
4029 int ret;
4030 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4031 if (!ret) {
Zach Brown8b558c52013-10-16 12:10:34 -07004032 drop_nlink(inode);
Al Viro92986792011-03-04 17:14:37 +00004033 ret = btrfs_update_inode(trans, root, inode);
4034 }
4035 return ret;
4036}
Yan, Zhenga22285a2010-05-16 10:48:46 -04004037
4038/*
4039 * helper to start transaction for unlink and rmdir.
4040 *
Josef Bacikd52be812013-05-29 14:54:47 -04004041 * unlink and rmdir are special in btrfs, they do not always free space, so
4042 * if we cannot make our reservations the normal way try and see if there is
4043 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4044 * allow the unlink to occur.
Yan, Zhenga22285a2010-05-16 10:48:46 -04004045 */
Josef Bacikd52be812013-05-29 14:54:47 -04004046static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
Yan, Zhenga22285a2010-05-16 10:48:46 -04004047{
Yan, Zhenga22285a2010-05-16 10:48:46 -04004048 struct btrfs_root *root = BTRFS_I(dir)->root;
Yan, Zhenga22285a2010-05-16 10:48:46 -04004049
Josef Bacike70bea52011-10-11 14:18:24 -04004050 /*
4051 * 1 for the possible orphan item
4052 * 1 for the dir item
4053 * 1 for the dir index
4054 * 1 for the inode ref
Josef Bacike70bea52011-10-11 14:18:24 -04004055 * 1 for the inode
4056 */
Filipe Manana8eab77f2015-11-13 23:57:16 +00004057 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
Yan, Zhenga22285a2010-05-16 10:48:46 -04004058}
4059
Chris Mason39279cc2007-06-12 06:35:45 -04004060static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4061{
Yan, Zhenga22285a2010-05-16 10:48:46 -04004062 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04004063 struct btrfs_trans_handle *trans;
David Howells2b0143b2015-03-17 22:25:59 +00004064 struct inode *inode = d_inode(dentry);
Chris Mason39279cc2007-06-12 06:35:45 -04004065 int ret;
4066
Josef Bacikd52be812013-05-29 14:54:47 -04004067 trans = __unlink_start_trans(dir);
Yan, Zhenga22285a2010-05-16 10:48:46 -04004068 if (IS_ERR(trans))
Josef Bacik5df6a9f2009-11-10 21:23:48 -05004069 return PTR_ERR(trans);
Chris Mason5f39d392007-10-15 16:14:19 -04004070
David Howells2b0143b2015-03-17 22:25:59 +00004071 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
Chris Mason12fcfd22009-03-24 10:24:20 -04004072
David Howells2b0143b2015-03-17 22:25:59 +00004073 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
Chris Masone02119d2008-09-05 16:13:11 -04004074 dentry->d_name.name, dentry->d_name.len);
Tsutomu Itohb5324022011-07-19 07:27:20 +00004075 if (ret)
4076 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04004077
Yan, Zhenga22285a2010-05-16 10:48:46 -04004078 if (inode->i_nlink == 0) {
Josef Bacik7b128762008-07-24 12:17:14 -04004079 ret = btrfs_orphan_add(trans, inode);
Tsutomu Itohb5324022011-07-19 07:27:20 +00004080 if (ret)
4081 goto out;
Yan, Zhenga22285a2010-05-16 10:48:46 -04004082 }
Josef Bacik7b128762008-07-24 12:17:14 -04004083
Tsutomu Itohb5324022011-07-19 07:27:20 +00004084out:
Josef Bacikd52be812013-05-29 14:54:47 -04004085 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00004086 btrfs_btree_balance_dirty(root);
Chris Mason39279cc2007-06-12 06:35:45 -04004087 return ret;
4088}
4089
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004090int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4091 struct btrfs_root *root,
4092 struct inode *dir, u64 objectid,
4093 const char *name, int name_len)
4094{
4095 struct btrfs_path *path;
4096 struct extent_buffer *leaf;
4097 struct btrfs_dir_item *di;
4098 struct btrfs_key key;
4099 u64 index;
4100 int ret;
Li Zefan33345d012011-04-20 10:31:50 +08004101 u64 dir_ino = btrfs_ino(dir);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004102
4103 path = btrfs_alloc_path();
4104 if (!path)
4105 return -ENOMEM;
4106
Li Zefan33345d012011-04-20 10:31:50 +08004107 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004108 name, name_len, -1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004109 if (IS_ERR_OR_NULL(di)) {
4110 if (!di)
4111 ret = -ENOENT;
4112 else
4113 ret = PTR_ERR(di);
4114 goto out;
4115 }
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004116
4117 leaf = path->nodes[0];
4118 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4119 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4120 ret = btrfs_delete_one_dir_name(trans, root, path, di);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004121 if (ret) {
4122 btrfs_abort_transaction(trans, root, ret);
4123 goto out;
4124 }
David Sterbab3b4aa72011-04-21 01:20:15 +02004125 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004126
4127 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4128 objectid, root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +08004129 dir_ino, &index, name, name_len);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004130 if (ret < 0) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004131 if (ret != -ENOENT) {
4132 btrfs_abort_transaction(trans, root, ret);
4133 goto out;
4134 }
Li Zefan33345d012011-04-20 10:31:50 +08004135 di = btrfs_search_dir_index_item(root, path, dir_ino,
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004136 name, name_len);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004137 if (IS_ERR_OR_NULL(di)) {
4138 if (!di)
4139 ret = -ENOENT;
4140 else
4141 ret = PTR_ERR(di);
4142 btrfs_abort_transaction(trans, root, ret);
4143 goto out;
4144 }
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004145
4146 leaf = path->nodes[0];
4147 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02004148 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004149 index = key.offset;
4150 }
David Sterbab3b4aa72011-04-21 01:20:15 +02004151 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004152
Miao Xie16cdcec2011-04-22 18:12:22 +08004153 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004154 if (ret) {
4155 btrfs_abort_transaction(trans, root, ret);
4156 goto out;
4157 }
Miao Xie16cdcec2011-04-22 18:12:22 +08004158
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004159 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
Josef Bacik0c4d2d92012-04-05 15:03:02 -04004160 inode_inc_iversion(dir);
Deepa Dinamani04b285f2016-02-06 23:57:21 -08004161 dir->i_mtime = dir->i_ctime = current_fs_time(dir->i_sb);
Josef Bacik5a24e842012-08-08 10:12:59 -06004162 ret = btrfs_update_inode_fallback(trans, root, dir);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004163 if (ret)
4164 btrfs_abort_transaction(trans, root, ret);
4165out:
Josef Bacik71d7aed2011-06-14 14:24:32 -04004166 btrfs_free_path(path);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004167 return ret;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004168}
4169
Chris Mason39279cc2007-06-12 06:35:45 -04004170static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4171{
David Howells2b0143b2015-03-17 22:25:59 +00004172 struct inode *inode = d_inode(dentry);
Chris Mason1832a6d2007-12-21 16:27:21 -05004173 int err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004174 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04004175 struct btrfs_trans_handle *trans;
Chris Mason39279cc2007-06-12 06:35:45 -04004176
David Sterbab3ae2442012-09-13 16:04:34 -06004177 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
Yan134d4512007-10-25 15:49:25 -04004178 return -ENOTEMPTY;
David Sterbab3ae2442012-09-13 16:04:34 -06004179 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4180 return -EPERM;
Yan134d4512007-10-25 15:49:25 -04004181
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);
Josef Bacik5df6a9f2009-11-10 21:23:48 -05004185
Li Zefan33345d012011-04-20 10:31:50 +08004186 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004187 err = btrfs_unlink_subvol(trans, root, dir,
4188 BTRFS_I(inode)->location.objectid,
4189 dentry->d_name.name,
4190 dentry->d_name.len);
4191 goto out;
4192 }
4193
Josef Bacik7b128762008-07-24 12:17:14 -04004194 err = btrfs_orphan_add(trans, inode);
4195 if (err)
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004196 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04004197
Chris Mason39279cc2007-06-12 06:35:45 -04004198 /* now the directory is empty */
David Howells2b0143b2015-03-17 22:25:59 +00004199 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
Chris Masone02119d2008-09-05 16:13:11 -04004200 dentry->d_name.name, dentry->d_name.len);
Chris Masond3977122009-01-05 21:25:51 -05004201 if (!err)
Chris Masondbe674a2008-07-17 12:54:05 -04004202 btrfs_i_size_write(inode, 0);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004203out:
Josef Bacikd52be812013-05-29 14:54:47 -04004204 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00004205 btrfs_btree_balance_dirty(root);
Chris Mason39544012007-12-12 14:38:19 -05004206
Chris Mason39279cc2007-06-12 06:35:45 -04004207 return err;
4208}
4209
Chris Mason28f75a02015-02-04 06:59:29 -08004210static int truncate_space_check(struct btrfs_trans_handle *trans,
4211 struct btrfs_root *root,
4212 u64 bytes_deleted)
4213{
4214 int ret;
4215
Josef Bacikdc95f7b2016-01-13 11:48:06 -05004216 /*
4217 * This is only used to apply pressure to the enospc system, we don't
4218 * intend to use this reservation at all.
4219 */
Chris Mason28f75a02015-02-04 06:59:29 -08004220 bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
Josef Bacikdc95f7b2016-01-13 11:48:06 -05004221 bytes_deleted *= root->nodesize;
Chris Mason28f75a02015-02-04 06:59:29 -08004222 ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4223 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
Josef Bacikdc95f7b2016-01-13 11:48:06 -05004224 if (!ret) {
4225 trace_btrfs_space_reservation(root->fs_info, "transaction",
4226 trans->transid,
4227 bytes_deleted, 1);
Chris Mason28f75a02015-02-04 06:59:29 -08004228 trans->bytes_reserved += bytes_deleted;
Josef Bacikdc95f7b2016-01-13 11:48:06 -05004229 }
Chris Mason28f75a02015-02-04 06:59:29 -08004230 return ret;
4231
4232}
4233
Filipe Manana0305cd52015-10-16 12:34:25 +01004234static int truncate_inline_extent(struct inode *inode,
4235 struct btrfs_path *path,
4236 struct btrfs_key *found_key,
4237 const u64 item_end,
4238 const u64 new_size)
4239{
4240 struct extent_buffer *leaf = path->nodes[0];
4241 int slot = path->slots[0];
4242 struct btrfs_file_extent_item *fi;
4243 u32 size = (u32)(new_size - found_key->offset);
4244 struct btrfs_root *root = BTRFS_I(inode)->root;
4245
4246 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4247
4248 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4249 loff_t offset = new_size;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004250 loff_t page_end = ALIGN(offset, PAGE_SIZE);
Filipe Manana0305cd52015-10-16 12:34:25 +01004251
4252 /*
4253 * Zero out the remaining of the last page of our inline extent,
4254 * instead of directly truncating our inline extent here - that
4255 * would be much more complex (decompressing all the data, then
4256 * compressing the truncated data, which might be bigger than
4257 * the size of the inline extent, resize the extent, etc).
4258 * We release the path because to get the page we might need to
4259 * read the extent item from disk (data not in the page cache).
4260 */
4261 btrfs_release_path(path);
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304262 return btrfs_truncate_block(inode, offset, page_end - offset,
4263 0);
Filipe Manana0305cd52015-10-16 12:34:25 +01004264 }
4265
4266 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4267 size = btrfs_file_extent_calc_inline_size(size);
4268 btrfs_truncate_item(root, path, size, 1);
4269
4270 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4271 inode_sub_bytes(inode, item_end + 1 - new_size);
4272
4273 return 0;
4274}
4275
Chris Mason323ac952008-10-01 19:05:46 -04004276/*
Chris Mason39279cc2007-06-12 06:35:45 -04004277 * this can truncate away extent items, csum items and directory items.
4278 * It starts at a high offset and removes keys until it can't find
Chris Masond352ac62008-09-29 15:18:18 -04004279 * any higher than new_size
Chris Mason39279cc2007-06-12 06:35:45 -04004280 *
4281 * csum items that cross the new i_size are truncated to the new size
4282 * as well.
Josef Bacik7b128762008-07-24 12:17:14 -04004283 *
4284 * min_type is the minimum key type to truncate down to. If set to 0, this
4285 * will kill all the items on this inode, including the INODE_ITEM_KEY.
Chris Mason39279cc2007-06-12 06:35:45 -04004286 */
Yan, Zheng80825102009-11-12 09:35:36 +00004287int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4288 struct btrfs_root *root,
4289 struct inode *inode,
4290 u64 new_size, u32 min_type)
Chris Mason39279cc2007-06-12 06:35:45 -04004291{
Chris Mason39279cc2007-06-12 06:35:45 -04004292 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04004293 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04004294 struct btrfs_file_extent_item *fi;
Yan, Zheng80825102009-11-12 09:35:36 +00004295 struct btrfs_key key;
4296 struct btrfs_key found_key;
Chris Mason39279cc2007-06-12 06:35:45 -04004297 u64 extent_start = 0;
Chris Masondb945352007-10-15 16:15:53 -04004298 u64 extent_num_bytes = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004299 u64 extent_offset = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004300 u64 item_end = 0;
Filipe Mananac1aa4572015-06-20 18:20:09 +01004301 u64 last_size = new_size;
Yan, Zheng80825102009-11-12 09:35:36 +00004302 u32 found_type = (u8)-1;
Chris Mason39279cc2007-06-12 06:35:45 -04004303 int found_extent;
4304 int del_item;
Chris Mason85e21ba2008-01-29 15:11:36 -05004305 int pending_del_nr = 0;
4306 int pending_del_slot = 0;
Chris Mason179e29e2007-11-01 11:28:41 -04004307 int extent_type = -1;
Yan, Zheng80825102009-11-12 09:35:36 +00004308 int ret;
4309 int err = 0;
Li Zefan33345d012011-04-20 10:31:50 +08004310 u64 ino = btrfs_ino(inode);
Chris Mason28ed1342014-12-17 09:41:04 -08004311 u64 bytes_deleted = 0;
Josef Bacik12621332015-02-03 07:50:16 -08004312 bool be_nice = 0;
4313 bool should_throttle = 0;
Chris Mason28f75a02015-02-04 06:59:29 -08004314 bool should_end = 0;
Yan, Zheng80825102009-11-12 09:35:36 +00004315
4316 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
Chris Mason39279cc2007-06-12 06:35:45 -04004317
Chris Mason28ed1342014-12-17 09:41:04 -08004318 /*
4319 * for non-free space inodes and ref cows, we want to back off from
4320 * time to time
4321 */
4322 if (!btrfs_is_free_space_inode(inode) &&
4323 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4324 be_nice = 1;
4325
Mark Fasheh0eb0e192011-07-12 16:44:10 -07004326 path = btrfs_alloc_path();
4327 if (!path)
4328 return -ENOMEM;
David Sterbae4058b52015-11-27 16:31:35 +01004329 path->reada = READA_BACK;
Mark Fasheh0eb0e192011-07-12 16:44:10 -07004330
Josef Bacik5dc562c2012-08-17 13:14:17 -04004331 /*
4332 * We want to drop from the next block forward in case this new size is
4333 * not block aligned since we will be keeping the last block of the
4334 * extent just the way it is.
4335 */
Miao Xie27cdeb72014-04-02 19:51:05 +08004336 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4337 root == root->fs_info->tree_root)
Qu Wenruofda28322013-02-26 08:10:22 +00004338 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4339 root->sectorsize), (u64)-1, 0);
Yan, Zheng80825102009-11-12 09:35:36 +00004340
Miao Xie16cdcec2011-04-22 18:12:22 +08004341 /*
4342 * This function is also used to drop the items in the log tree before
4343 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4344 * it is used to drop the loged items. So we shouldn't kill the delayed
4345 * items.
4346 */
4347 if (min_type == 0 && root == BTRFS_I(inode)->root)
4348 btrfs_kill_delayed_inode_items(inode);
4349
Li Zefan33345d012011-04-20 10:31:50 +08004350 key.objectid = ino;
Chris Mason39279cc2007-06-12 06:35:45 -04004351 key.offset = (u64)-1;
Chris Mason5f39d392007-10-15 16:14:19 -04004352 key.type = (u8)-1;
4353
Chris Mason85e21ba2008-01-29 15:11:36 -05004354search_again:
Chris Mason28ed1342014-12-17 09:41:04 -08004355 /*
4356 * with a 16K leaf size and 128MB extents, you can actually queue
4357 * up a huge file in a single leaf. Most of the time that
4358 * bytes_deleted is > 0, it will be huge by the time we get here
4359 */
Byongho Leeee221842015-12-15 01:42:10 +09004360 if (be_nice && bytes_deleted > SZ_32M) {
Chris Mason28ed1342014-12-17 09:41:04 -08004361 if (btrfs_should_end_transaction(trans, root)) {
4362 err = -EAGAIN;
4363 goto error;
4364 }
4365 }
4366
4367
Chris Masonb9473432009-03-13 11:00:37 -04004368 path->leave_spinning = 1;
Chris Mason85e21ba2008-01-29 15:11:36 -05004369 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
Yan, Zheng80825102009-11-12 09:35:36 +00004370 if (ret < 0) {
4371 err = ret;
4372 goto out;
4373 }
Chris Masond3977122009-01-05 21:25:51 -05004374
Chris Mason85e21ba2008-01-29 15:11:36 -05004375 if (ret > 0) {
Chris Masone02119d2008-09-05 16:13:11 -04004376 /* there are no items in the tree for us to truncate, we're
4377 * done
4378 */
Yan, Zheng80825102009-11-12 09:35:36 +00004379 if (path->slots[0] == 0)
4380 goto out;
Chris Mason85e21ba2008-01-29 15:11:36 -05004381 path->slots[0]--;
4382 }
4383
Chris Masond3977122009-01-05 21:25:51 -05004384 while (1) {
Chris Mason39279cc2007-06-12 06:35:45 -04004385 fi = NULL;
Chris Mason5f39d392007-10-15 16:14:19 -04004386 leaf = path->nodes[0];
4387 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
David Sterba962a2982014-06-04 18:41:45 +02004388 found_type = found_key.type;
Chris Mason39279cc2007-06-12 06:35:45 -04004389
Li Zefan33345d012011-04-20 10:31:50 +08004390 if (found_key.objectid != ino)
Chris Mason39279cc2007-06-12 06:35:45 -04004391 break;
Chris Mason5f39d392007-10-15 16:14:19 -04004392
Chris Mason85e21ba2008-01-29 15:11:36 -05004393 if (found_type < min_type)
Chris Mason39279cc2007-06-12 06:35:45 -04004394 break;
4395
Chris Mason5f39d392007-10-15 16:14:19 -04004396 item_end = found_key.offset;
Chris Mason39279cc2007-06-12 06:35:45 -04004397 if (found_type == BTRFS_EXTENT_DATA_KEY) {
Chris Mason5f39d392007-10-15 16:14:19 -04004398 fi = btrfs_item_ptr(leaf, path->slots[0],
Chris Mason39279cc2007-06-12 06:35:45 -04004399 struct btrfs_file_extent_item);
Chris Mason179e29e2007-11-01 11:28:41 -04004400 extent_type = btrfs_file_extent_type(leaf, fi);
4401 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
Chris Mason5f39d392007-10-15 16:14:19 -04004402 item_end +=
Chris Masondb945352007-10-15 16:15:53 -04004403 btrfs_file_extent_num_bytes(leaf, fi);
Chris Mason179e29e2007-11-01 11:28:41 -04004404 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason179e29e2007-11-01 11:28:41 -04004405 item_end += btrfs_file_extent_inline_len(leaf,
Chris Mason514ac8a2014-01-03 21:07:00 -08004406 path->slots[0], fi);
Chris Mason39279cc2007-06-12 06:35:45 -04004407 }
Yan008630c2007-11-07 13:31:09 -05004408 item_end--;
Chris Mason39279cc2007-06-12 06:35:45 -04004409 }
Yan, Zheng80825102009-11-12 09:35:36 +00004410 if (found_type > min_type) {
Chris Mason39279cc2007-06-12 06:35:45 -04004411 del_item = 1;
Yan, Zheng80825102009-11-12 09:35:36 +00004412 } else {
4413 if (item_end < new_size)
4414 break;
4415 if (found_key.offset >= new_size)
4416 del_item = 1;
4417 else
4418 del_item = 0;
4419 }
Chris Mason39279cc2007-06-12 06:35:45 -04004420 found_extent = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004421 /* FIXME, shrink the extent if the ref count is only 1 */
Chris Mason179e29e2007-11-01 11:28:41 -04004422 if (found_type != BTRFS_EXTENT_DATA_KEY)
4423 goto delete;
4424
Josef Bacik7f4f6e02013-08-29 16:43:28 -04004425 if (del_item)
4426 last_size = found_key.offset;
4427 else
4428 last_size = new_size;
4429
Chris Mason179e29e2007-11-01 11:28:41 -04004430 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
Chris Mason39279cc2007-06-12 06:35:45 -04004431 u64 num_dec;
Chris Masondb945352007-10-15 16:15:53 -04004432 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
Miao Xief70a9a6b2012-01-12 19:10:12 -05004433 if (!del_item) {
Chris Masondb945352007-10-15 16:15:53 -04004434 u64 orig_num_bytes =
4435 btrfs_file_extent_num_bytes(leaf, fi);
Qu Wenruofda28322013-02-26 08:10:22 +00004436 extent_num_bytes = ALIGN(new_size -
4437 found_key.offset,
4438 root->sectorsize);
Chris Masondb945352007-10-15 16:15:53 -04004439 btrfs_set_file_extent_num_bytes(leaf, fi,
4440 extent_num_bytes);
4441 num_dec = (orig_num_bytes -
Chris Mason90692182008-02-08 13:49:28 -05004442 extent_num_bytes);
Miao Xie27cdeb72014-04-02 19:51:05 +08004443 if (test_bit(BTRFS_ROOT_REF_COWS,
4444 &root->state) &&
4445 extent_start != 0)
Yan Zhenga76a3cd2008-10-09 11:46:29 -04004446 inode_sub_bytes(inode, num_dec);
Chris Mason5f39d392007-10-15 16:14:19 -04004447 btrfs_mark_buffer_dirty(leaf);
Chris Mason39279cc2007-06-12 06:35:45 -04004448 } else {
Chris Masondb945352007-10-15 16:15:53 -04004449 extent_num_bytes =
4450 btrfs_file_extent_disk_num_bytes(leaf,
4451 fi);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004452 extent_offset = found_key.offset -
4453 btrfs_file_extent_offset(leaf, fi);
4454
Chris Mason39279cc2007-06-12 06:35:45 -04004455 /* FIXME blocksize != 4096 */
Chris Mason90692182008-02-08 13:49:28 -05004456 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
Chris Mason39279cc2007-06-12 06:35:45 -04004457 if (extent_start != 0) {
4458 found_extent = 1;
Miao Xie27cdeb72014-04-02 19:51:05 +08004459 if (test_bit(BTRFS_ROOT_REF_COWS,
4460 &root->state))
Yan Zhenga76a3cd2008-10-09 11:46:29 -04004461 inode_sub_bytes(inode, num_dec);
Chris Mason39279cc2007-06-12 06:35:45 -04004462 }
4463 }
Chris Mason90692182008-02-08 13:49:28 -05004464 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Masonc8b97812008-10-29 14:49:59 -04004465 /*
4466 * we can't truncate inline items that have had
4467 * special encodings
4468 */
4469 if (!del_item &&
Chris Masonc8b97812008-10-29 14:49:59 -04004470 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4471 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
Chris Mason514ac8a2014-01-03 21:07:00 -08004472
4473 /*
Filipe Manana0305cd52015-10-16 12:34:25 +01004474 * Need to release path in order to truncate a
4475 * compressed extent. So delete any accumulated
4476 * extent items so far.
Chris Mason514ac8a2014-01-03 21:07:00 -08004477 */
Filipe Manana0305cd52015-10-16 12:34:25 +01004478 if (btrfs_file_extent_compression(leaf, fi) !=
4479 BTRFS_COMPRESS_NONE && pending_del_nr) {
4480 err = btrfs_del_items(trans, root, path,
4481 pending_del_slot,
4482 pending_del_nr);
4483 if (err) {
4484 btrfs_abort_transaction(trans,
4485 root,
4486 err);
4487 goto error;
4488 }
4489 pending_del_nr = 0;
4490 }
4491
4492 err = truncate_inline_extent(inode, path,
4493 &found_key,
4494 item_end,
4495 new_size);
4496 if (err) {
4497 btrfs_abort_transaction(trans,
4498 root, err);
4499 goto error;
4500 }
Miao Xie27cdeb72014-04-02 19:51:05 +08004501 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4502 &root->state)) {
Filipe Manana0305cd52015-10-16 12:34:25 +01004503 inode_sub_bytes(inode, item_end + 1 - new_size);
Chris Mason90692182008-02-08 13:49:28 -05004504 }
Chris Mason39279cc2007-06-12 06:35:45 -04004505 }
Chris Mason179e29e2007-11-01 11:28:41 -04004506delete:
Chris Mason39279cc2007-06-12 06:35:45 -04004507 if (del_item) {
Chris Mason85e21ba2008-01-29 15:11:36 -05004508 if (!pending_del_nr) {
4509 /* no pending yet, add ourselves */
4510 pending_del_slot = path->slots[0];
4511 pending_del_nr = 1;
4512 } else if (pending_del_nr &&
4513 path->slots[0] + 1 == pending_del_slot) {
4514 /* hop on the pending chunk */
4515 pending_del_nr++;
4516 pending_del_slot = path->slots[0];
4517 } else {
Chris Masond3977122009-01-05 21:25:51 -05004518 BUG();
Chris Mason85e21ba2008-01-29 15:11:36 -05004519 }
Chris Mason39279cc2007-06-12 06:35:45 -04004520 } else {
4521 break;
4522 }
Chris Mason28f75a02015-02-04 06:59:29 -08004523 should_throttle = 0;
4524
Miao Xie27cdeb72014-04-02 19:51:05 +08004525 if (found_extent &&
4526 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4527 root == root->fs_info->tree_root)) {
Chris Masonb9473432009-03-13 11:00:37 -04004528 btrfs_set_path_blocking(path);
Chris Mason28ed1342014-12-17 09:41:04 -08004529 bytes_deleted += extent_num_bytes;
Chris Mason39279cc2007-06-12 06:35:45 -04004530 ret = btrfs_free_extent(trans, root, extent_start,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004531 extent_num_bytes, 0,
4532 btrfs_header_owner(leaf),
Filipe Mananab06c4bf2015-10-23 07:52:54 +01004533 ino, extent_offset);
Chris Mason39279cc2007-06-12 06:35:45 -04004534 BUG_ON(ret);
Josef Bacik12621332015-02-03 07:50:16 -08004535 if (btrfs_should_throttle_delayed_refs(trans, root))
Chris Mason28ed1342014-12-17 09:41:04 -08004536 btrfs_async_run_delayed_refs(root,
4537 trans->delayed_ref_updates * 2, 0);
Chris Mason28f75a02015-02-04 06:59:29 -08004538 if (be_nice) {
4539 if (truncate_space_check(trans, root,
4540 extent_num_bytes)) {
4541 should_end = 1;
4542 }
4543 if (btrfs_should_throttle_delayed_refs(trans,
4544 root)) {
4545 should_throttle = 1;
4546 }
4547 }
Chris Mason39279cc2007-06-12 06:35:45 -04004548 }
Chris Mason85e21ba2008-01-29 15:11:36 -05004549
Yan, Zheng80825102009-11-12 09:35:36 +00004550 if (found_type == BTRFS_INODE_ITEM_KEY)
4551 break;
4552
4553 if (path->slots[0] == 0 ||
Josef Bacik12621332015-02-03 07:50:16 -08004554 path->slots[0] != pending_del_slot ||
Chris Mason28f75a02015-02-04 06:59:29 -08004555 should_throttle || should_end) {
Yan, Zheng80825102009-11-12 09:35:36 +00004556 if (pending_del_nr) {
4557 ret = btrfs_del_items(trans, root, path,
4558 pending_del_slot,
4559 pending_del_nr);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004560 if (ret) {
4561 btrfs_abort_transaction(trans,
4562 root, ret);
4563 goto error;
4564 }
Yan, Zheng80825102009-11-12 09:35:36 +00004565 pending_del_nr = 0;
4566 }
David Sterbab3b4aa72011-04-21 01:20:15 +02004567 btrfs_release_path(path);
Chris Mason28f75a02015-02-04 06:59:29 -08004568 if (should_throttle) {
Josef Bacik12621332015-02-03 07:50:16 -08004569 unsigned long updates = trans->delayed_ref_updates;
4570 if (updates) {
4571 trans->delayed_ref_updates = 0;
4572 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4573 if (ret && !err)
4574 err = ret;
4575 }
4576 }
Chris Mason28f75a02015-02-04 06:59:29 -08004577 /*
4578 * if we failed to refill our space rsv, bail out
4579 * and let the transaction restart
4580 */
4581 if (should_end) {
4582 err = -EAGAIN;
4583 goto error;
4584 }
Chris Mason85e21ba2008-01-29 15:11:36 -05004585 goto search_again;
Yan, Zheng80825102009-11-12 09:35:36 +00004586 } else {
4587 path->slots[0]--;
Chris Mason85e21ba2008-01-29 15:11:36 -05004588 }
Chris Mason39279cc2007-06-12 06:35:45 -04004589 }
Yan, Zheng80825102009-11-12 09:35:36 +00004590out:
Chris Mason85e21ba2008-01-29 15:11:36 -05004591 if (pending_del_nr) {
4592 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4593 pending_del_nr);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004594 if (ret)
4595 btrfs_abort_transaction(trans, root, ret);
Chris Mason85e21ba2008-01-29 15:11:36 -05004596 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004597error:
Filipe Mananac1aa4572015-06-20 18:20:09 +01004598 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
Josef Bacik7f4f6e02013-08-29 16:43:28 -04004599 btrfs_ordered_update_i_size(inode, last_size, NULL);
Chris Mason28ed1342014-12-17 09:41:04 -08004600
Chris Mason39279cc2007-06-12 06:35:45 -04004601 btrfs_free_path(path);
Chris Mason28ed1342014-12-17 09:41:04 -08004602
Byongho Leeee221842015-12-15 01:42:10 +09004603 if (be_nice && bytes_deleted > SZ_32M) {
Chris Mason28ed1342014-12-17 09:41:04 -08004604 unsigned long updates = trans->delayed_ref_updates;
4605 if (updates) {
4606 trans->delayed_ref_updates = 0;
4607 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4608 if (ret && !err)
4609 err = ret;
4610 }
4611 }
Yan, Zheng80825102009-11-12 09:35:36 +00004612 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04004613}
4614
Chris Masona52d9a82007-08-27 16:49:44 -04004615/*
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304616 * btrfs_truncate_block - read, zero a chunk and write a block
Josef Bacik2aaa6652012-08-29 14:27:18 -04004617 * @inode - inode that we're zeroing
4618 * @from - the offset to start zeroing
4619 * @len - the length to zero, 0 to zero the entire range respective to the
4620 * offset
4621 * @front - zero up to the offset instead of from the offset on
4622 *
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304623 * This will find the block for the "from" offset and cow the block and zero the
Josef Bacik2aaa6652012-08-29 14:27:18 -04004624 * part we want to zero. This is used with truncate and hole punching.
Chris Masona52d9a82007-08-27 16:49:44 -04004625 */
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304626int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
Josef Bacik2aaa6652012-08-29 14:27:18 -04004627 int front)
Chris Masona52d9a82007-08-27 16:49:44 -04004628{
Josef Bacik2aaa6652012-08-29 14:27:18 -04004629 struct address_space *mapping = inode->i_mapping;
Chris Masondb945352007-10-15 16:15:53 -04004630 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004631 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4632 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00004633 struct extent_state *cached_state = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004634 char *kaddr;
Chris Masondb945352007-10-15 16:15:53 -04004635 u32 blocksize = root->sectorsize;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004636 pgoff_t index = from >> PAGE_SHIFT;
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304637 unsigned offset = from & (blocksize - 1);
Chris Masona52d9a82007-08-27 16:49:44 -04004638 struct page *page;
Josef Bacik3b16a4e2011-09-21 15:05:58 -04004639 gfp_t mask = btrfs_alloc_write_mask(mapping);
Chris Masona52d9a82007-08-27 16:49:44 -04004640 int ret = 0;
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304641 u64 block_start;
4642 u64 block_end;
Chris Masona52d9a82007-08-27 16:49:44 -04004643
Josef Bacik2aaa6652012-08-29 14:27:18 -04004644 if ((offset & (blocksize - 1)) == 0 &&
4645 (!len || ((len & (blocksize - 1)) == 0)))
Chris Masona52d9a82007-08-27 16:49:44 -04004646 goto out;
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304647
Qu Wenruo7cf5b972015-09-08 17:25:55 +08004648 ret = btrfs_delalloc_reserve_space(inode,
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304649 round_down(from, blocksize), blocksize);
Josef Bacik5d5e1032009-10-13 16:46:49 -04004650 if (ret)
4651 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04004652
Chris Mason211c17f2008-05-15 09:13:45 -04004653again:
Josef Bacik3b16a4e2011-09-21 15:05:58 -04004654 page = find_or_create_page(mapping, index, mask);
Josef Bacik5d5e1032009-10-13 16:46:49 -04004655 if (!page) {
Qu Wenruo7cf5b972015-09-08 17:25:55 +08004656 btrfs_delalloc_release_space(inode,
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304657 round_down(from, blocksize),
4658 blocksize);
Miao Xieac6a2b32012-12-05 10:56:13 +00004659 ret = -ENOMEM;
Chris Masona52d9a82007-08-27 16:49:44 -04004660 goto out;
Josef Bacik5d5e1032009-10-13 16:46:49 -04004661 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04004662
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304663 block_start = round_down(from, blocksize);
4664 block_end = block_start + blocksize - 1;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004665
Chris Masona52d9a82007-08-27 16:49:44 -04004666 if (!PageUptodate(page)) {
4667 ret = btrfs_readpage(NULL, page);
4668 lock_page(page);
Chris Mason211c17f2008-05-15 09:13:45 -04004669 if (page->mapping != mapping) {
4670 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004671 put_page(page);
Chris Mason211c17f2008-05-15 09:13:45 -04004672 goto again;
4673 }
Chris Masona52d9a82007-08-27 16:49:44 -04004674 if (!PageUptodate(page)) {
4675 ret = -EIO;
Chris Mason89642222008-07-24 09:41:53 -04004676 goto out_unlock;
Chris Masona52d9a82007-08-27 16:49:44 -04004677 }
4678 }
Chris Mason211c17f2008-05-15 09:13:45 -04004679 wait_on_page_writeback(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004680
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304681 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004682 set_page_extent_mapped(page);
4683
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304684 ordered = btrfs_lookup_ordered_extent(inode, block_start);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004685 if (ordered) {
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304686 unlock_extent_cached(io_tree, block_start, block_end,
Josef Bacik2ac55d42010-02-03 19:33:23 +00004687 &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004688 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004689 put_page(page);
Chris Masoneb84ae02008-07-17 13:53:27 -04004690 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004691 btrfs_put_ordered_extent(ordered);
4692 goto again;
4693 }
4694
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304695 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
Liu Bo9e8a4a82012-09-05 19:10:51 -06004696 EXTENT_DIRTY | EXTENT_DELALLOC |
4697 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
Josef Bacik2ac55d42010-02-03 19:33:23 +00004698 0, 0, &cached_state, GFP_NOFS);
Josef Bacik5d5e1032009-10-13 16:46:49 -04004699
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304700 ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
Josef Bacik2ac55d42010-02-03 19:33:23 +00004701 &cached_state);
Josef Bacik9ed74f22009-09-11 16:12:44 -04004702 if (ret) {
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304703 unlock_extent_cached(io_tree, block_start, block_end,
Josef Bacik2ac55d42010-02-03 19:33:23 +00004704 &cached_state, GFP_NOFS);
Josef Bacik9ed74f22009-09-11 16:12:44 -04004705 goto out_unlock;
4706 }
4707
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304708 if (offset != blocksize) {
Josef Bacik2aaa6652012-08-29 14:27:18 -04004709 if (!len)
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304710 len = blocksize - offset;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004711 kaddr = kmap(page);
Josef Bacik2aaa6652012-08-29 14:27:18 -04004712 if (front)
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304713 memset(kaddr + (block_start - page_offset(page)),
4714 0, offset);
Josef Bacik2aaa6652012-08-29 14:27:18 -04004715 else
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304716 memset(kaddr + (block_start - page_offset(page)) + offset,
4717 0, len);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004718 flush_dcache_page(page);
4719 kunmap(page);
4720 }
Chris Mason247e7432008-07-17 12:53:51 -04004721 ClearPageChecked(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004722 set_page_dirty(page);
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304723 unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
Josef Bacik2ac55d42010-02-03 19:33:23 +00004724 GFP_NOFS);
Chris Mason39279cc2007-06-12 06:35:45 -04004725
Chris Mason89642222008-07-24 09:41:53 -04004726out_unlock:
Josef Bacik5d5e1032009-10-13 16:46:49 -04004727 if (ret)
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304728 btrfs_delalloc_release_space(inode, block_start,
4729 blocksize);
Chris Mason39279cc2007-06-12 06:35:45 -04004730 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004731 put_page(page);
Chris Mason39279cc2007-06-12 06:35:45 -04004732out:
4733 return ret;
4734}
4735
Josef Bacik16e75492013-10-22 12:18:51 -04004736static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4737 u64 offset, u64 len)
4738{
4739 struct btrfs_trans_handle *trans;
4740 int ret;
4741
4742 /*
4743 * Still need to make sure the inode looks like it's been updated so
4744 * that any holes get logged if we fsync.
4745 */
4746 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4747 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4748 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4749 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4750 return 0;
4751 }
4752
4753 /*
4754 * 1 - for the one we're dropping
4755 * 1 - for the one we're adding
4756 * 1 - for updating the inode.
4757 */
4758 trans = btrfs_start_transaction(root, 3);
4759 if (IS_ERR(trans))
4760 return PTR_ERR(trans);
4761
4762 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4763 if (ret) {
4764 btrfs_abort_transaction(trans, root, ret);
4765 btrfs_end_transaction(trans, root);
4766 return ret;
4767 }
4768
4769 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4770 0, 0, len, 0, len, 0, 0, 0);
4771 if (ret)
4772 btrfs_abort_transaction(trans, root, ret);
4773 else
4774 btrfs_update_inode(trans, root, inode);
4775 btrfs_end_transaction(trans, root);
4776 return ret;
4777}
4778
Josef Bacik695a0d02011-03-04 15:46:53 -05004779/*
4780 * This function puts in dummy file extents for the area we're creating a hole
4781 * for. So if we are truncating this file to a larger size we need to insert
4782 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4783 * the range between oldsize and size
4784 */
Josef Bacika41ad392011-01-31 15:30:16 -05004785int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
Yan Zheng9036c102008-10-30 14:19:41 -04004786{
Yan Zheng9036c102008-10-30 14:19:41 -04004787 struct btrfs_root *root = BTRFS_I(inode)->root;
4788 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Yan, Zhenga22285a2010-05-16 10:48:46 -04004789 struct extent_map *em = NULL;
Josef Bacik2ac55d42010-02-03 19:33:23 +00004790 struct extent_state *cached_state = NULL;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004791 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Qu Wenruofda28322013-02-26 08:10:22 +00004792 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4793 u64 block_end = ALIGN(size, root->sectorsize);
Yan Zheng9036c102008-10-30 14:19:41 -04004794 u64 last_byte;
4795 u64 cur_offset;
4796 u64 hole_size;
Josef Bacik9ed74f22009-09-11 16:12:44 -04004797 int err = 0;
Yan Zheng9036c102008-10-30 14:19:41 -04004798
Josef Bacika71754f2013-06-17 17:14:39 -04004799 /*
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304800 * If our size started in the middle of a block we need to zero out the
4801 * rest of the block before we expand the i_size, otherwise we could
Josef Bacika71754f2013-06-17 17:14:39 -04004802 * expose stale data.
4803 */
Chandan Rajendra9703fef2016-01-21 15:55:56 +05304804 err = btrfs_truncate_block(inode, oldsize, 0, 0);
Josef Bacika71754f2013-06-17 17:14:39 -04004805 if (err)
4806 return err;
4807
Yan Zheng9036c102008-10-30 14:19:41 -04004808 if (size <= hole_start)
4809 return 0;
4810
Yan Zheng9036c102008-10-30 14:19:41 -04004811 while (1) {
4812 struct btrfs_ordered_extent *ordered;
Miao Xiefa7c1492013-09-26 13:15:27 +08004813
David Sterbaff13db42015-12-03 14:30:40 +01004814 lock_extent_bits(io_tree, hole_start, block_end - 1,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01004815 &cached_state);
Miao Xiefa7c1492013-09-26 13:15:27 +08004816 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4817 block_end - hole_start);
Yan Zheng9036c102008-10-30 14:19:41 -04004818 if (!ordered)
4819 break;
Josef Bacik2ac55d42010-02-03 19:33:23 +00004820 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4821 &cached_state, GFP_NOFS);
Miao Xiefa7c1492013-09-26 13:15:27 +08004822 btrfs_start_ordered_extent(inode, ordered, 1);
Yan Zheng9036c102008-10-30 14:19:41 -04004823 btrfs_put_ordered_extent(ordered);
4824 }
4825
Yan Zheng9036c102008-10-30 14:19:41 -04004826 cur_offset = hole_start;
4827 while (1) {
4828 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4829 block_end - cur_offset, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004830 if (IS_ERR(em)) {
4831 err = PTR_ERR(em);
Zach Brownf2767952013-01-08 19:37:58 +00004832 em = NULL;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004833 break;
4834 }
Yan Zheng9036c102008-10-30 14:19:41 -04004835 last_byte = min(extent_map_end(em), block_end);
Qu Wenruofda28322013-02-26 08:10:22 +00004836 last_byte = ALIGN(last_byte , root->sectorsize);
Yan, Zheng80825102009-11-12 09:35:36 +00004837 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
Josef Bacik5dc562c2012-08-17 13:14:17 -04004838 struct extent_map *hole_em;
Yan Zheng9036c102008-10-30 14:19:41 -04004839 hole_size = last_byte - cur_offset;
Yan, Zheng80825102009-11-12 09:35:36 +00004840
Josef Bacik16e75492013-10-22 12:18:51 -04004841 err = maybe_insert_hole(root, inode, cur_offset,
4842 hole_size);
4843 if (err)
Chris Mason771ed682008-11-06 22:02:51 -05004844 break;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004845 btrfs_drop_extent_cache(inode, cur_offset,
4846 cur_offset + hole_size - 1, 0);
4847 hole_em = alloc_extent_map();
4848 if (!hole_em) {
4849 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4850 &BTRFS_I(inode)->runtime_flags);
4851 goto next;
4852 }
4853 hole_em->start = cur_offset;
4854 hole_em->len = hole_size;
4855 hole_em->orig_start = cur_offset;
Yan, Zheng80825102009-11-12 09:35:36 +00004856
Josef Bacik5dc562c2012-08-17 13:14:17 -04004857 hole_em->block_start = EXTENT_MAP_HOLE;
4858 hole_em->block_len = 0;
Josef Bacikb4939682012-12-03 10:31:19 -05004859 hole_em->orig_block_len = 0;
Josef Bacikcc95bef2013-04-04 14:31:27 -04004860 hole_em->ram_bytes = hole_size;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004861 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4862 hole_em->compress_type = BTRFS_COMPRESS_NONE;
Josef Bacik16e75492013-10-22 12:18:51 -04004863 hole_em->generation = root->fs_info->generation;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004864
4865 while (1) {
4866 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -04004867 err = add_extent_mapping(em_tree, hole_em, 1);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004868 write_unlock(&em_tree->lock);
4869 if (err != -EEXIST)
4870 break;
4871 btrfs_drop_extent_cache(inode, cur_offset,
4872 cur_offset +
4873 hole_size - 1, 0);
4874 }
4875 free_extent_map(hole_em);
Yan Zheng9036c102008-10-30 14:19:41 -04004876 }
Josef Bacik16e75492013-10-22 12:18:51 -04004877next:
Yan Zheng9036c102008-10-30 14:19:41 -04004878 free_extent_map(em);
Yan, Zhenga22285a2010-05-16 10:48:46 -04004879 em = NULL;
Yan Zheng9036c102008-10-30 14:19:41 -04004880 cur_offset = last_byte;
Yan, Zheng80825102009-11-12 09:35:36 +00004881 if (cur_offset >= block_end)
Yan Zheng9036c102008-10-30 14:19:41 -04004882 break;
4883 }
Yan, Zhenga22285a2010-05-16 10:48:46 -04004884 free_extent_map(em);
Josef Bacik2ac55d42010-02-03 19:33:23 +00004885 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4886 GFP_NOFS);
Yan Zheng9036c102008-10-30 14:19:41 -04004887 return err;
4888}
4889
Eric Sandeen3972f262013-01-12 02:57:22 +00004890static int btrfs_setsize(struct inode *inode, struct iattr *attr)
Yan, Zheng80825102009-11-12 09:35:36 +00004891{
Miao Xief4a2f4c2011-12-14 20:12:01 -05004892 struct btrfs_root *root = BTRFS_I(inode)->root;
4893 struct btrfs_trans_handle *trans;
Josef Bacika41ad392011-01-31 15:30:16 -05004894 loff_t oldsize = i_size_read(inode);
Eric Sandeen3972f262013-01-12 02:57:22 +00004895 loff_t newsize = attr->ia_size;
4896 int mask = attr->ia_valid;
Yan, Zheng80825102009-11-12 09:35:36 +00004897 int ret;
4898
Eric Sandeen3972f262013-01-12 02:57:22 +00004899 /*
4900 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4901 * special case where we need to update the times despite not having
4902 * these flags set. For all other operations the VFS set these flags
4903 * explicitly if it wants a timestamp update.
4904 */
Christoph Hellwigdff6efc2013-11-19 07:17:07 -08004905 if (newsize != oldsize) {
4906 inode_inc_iversion(inode);
4907 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4908 inode->i_ctime = inode->i_mtime =
4909 current_fs_time(inode->i_sb);
4910 }
Eric Sandeen3972f262013-01-12 02:57:22 +00004911
Josef Bacika41ad392011-01-31 15:30:16 -05004912 if (newsize > oldsize) {
Filipe Manana9ea24bb2014-10-29 11:57:59 +00004913 /*
4914 * Don't do an expanding truncate while snapshoting is ongoing.
4915 * This is to ensure the snapshot captures a fully consistent
4916 * state of this file - if the snapshot captures this expanding
4917 * truncation, it must capture all writes that happened before
4918 * this truncation.
4919 */
Zhao Lei0bc19f902016-01-06 18:56:36 +08004920 btrfs_wait_for_snapshot_creation(root);
Josef Bacika41ad392011-01-31 15:30:16 -05004921 ret = btrfs_cont_expand(inode, oldsize, newsize);
Filipe Manana9ea24bb2014-10-29 11:57:59 +00004922 if (ret) {
4923 btrfs_end_write_no_snapshoting(root);
Yan, Zheng80825102009-11-12 09:35:36 +00004924 return ret;
Filipe Manana9ea24bb2014-10-29 11:57:59 +00004925 }
Yan, Zheng80825102009-11-12 09:35:36 +00004926
Miao Xief4a2f4c2011-12-14 20:12:01 -05004927 trans = btrfs_start_transaction(root, 1);
Filipe Manana9ea24bb2014-10-29 11:57:59 +00004928 if (IS_ERR(trans)) {
4929 btrfs_end_write_no_snapshoting(root);
Miao Xief4a2f4c2011-12-14 20:12:01 -05004930 return PTR_ERR(trans);
Filipe Manana9ea24bb2014-10-29 11:57:59 +00004931 }
Miao Xief4a2f4c2011-12-14 20:12:01 -05004932
4933 i_size_write(inode, newsize);
4934 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
Chandan Rajendra27772b62016-01-21 15:56:03 +05304935 pagecache_isize_extended(inode, oldsize, newsize);
Miao Xief4a2f4c2011-12-14 20:12:01 -05004936 ret = btrfs_update_inode(trans, root, inode);
Filipe Manana9ea24bb2014-10-29 11:57:59 +00004937 btrfs_end_write_no_snapshoting(root);
Josef Bacik7ad85bb2012-01-12 19:10:12 -05004938 btrfs_end_transaction(trans, root);
Josef Bacika41ad392011-01-31 15:30:16 -05004939 } else {
Yan, Zheng80825102009-11-12 09:35:36 +00004940
Josef Bacika41ad392011-01-31 15:30:16 -05004941 /*
4942 * We're truncating a file that used to have good data down to
4943 * zero. Make sure it gets into the ordered flush list so that
4944 * any new writes get down to disk quickly.
4945 */
4946 if (newsize == 0)
Josef Bacik72ac3c02012-05-23 14:13:11 -04004947 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4948 &BTRFS_I(inode)->runtime_flags);
Yan, Zheng80825102009-11-12 09:35:36 +00004949
Josef Bacikf3fe8202013-01-07 17:03:21 -05004950 /*
4951 * 1 for the orphan item we're going to add
4952 * 1 for the orphan item deletion.
4953 */
4954 trans = btrfs_start_transaction(root, 2);
4955 if (IS_ERR(trans))
4956 return PTR_ERR(trans);
4957
4958 /*
4959 * We need to do this in case we fail at _any_ point during the
4960 * actual truncate. Once we do the truncate_setsize we could
4961 * invalidate pages which forces any outstanding ordered io to
4962 * be instantly completed which will give us extents that need
4963 * to be truncated. If we fail to get an orphan inode down we
4964 * could have left over extents that were never meant to live,
4965 * so we need to garuntee from this point on that everything
4966 * will be consistent.
4967 */
4968 ret = btrfs_orphan_add(trans, inode);
4969 btrfs_end_transaction(trans, root);
4970 if (ret)
4971 return ret;
4972
Josef Bacika41ad392011-01-31 15:30:16 -05004973 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4974 truncate_setsize(inode, newsize);
Miao Xie2e60a512013-02-08 07:01:08 +00004975
4976 /* Disable nonlocked read DIO to avoid the end less truncate */
4977 btrfs_inode_block_unlocked_dio(inode);
4978 inode_dio_wait(inode);
4979 btrfs_inode_resume_unlocked_dio(inode);
4980
Josef Bacika41ad392011-01-31 15:30:16 -05004981 ret = btrfs_truncate(inode);
Josef Bacik7f4f6e02013-08-29 16:43:28 -04004982 if (ret && inode->i_nlink) {
4983 int err;
4984
4985 /*
4986 * failed to truncate, disk_i_size is only adjusted down
4987 * as we remove extents, so it should represent the true
4988 * size of the inode, so reset the in memory size and
4989 * delete our orphan entry.
4990 */
4991 trans = btrfs_join_transaction(root);
4992 if (IS_ERR(trans)) {
4993 btrfs_orphan_del(NULL, inode);
4994 return ret;
4995 }
4996 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4997 err = btrfs_orphan_del(trans, inode);
4998 if (err)
4999 btrfs_abort_transaction(trans, root, err);
5000 btrfs_end_transaction(trans, root);
5001 }
Yan, Zheng80825102009-11-12 09:35:36 +00005002 }
5003
Josef Bacika41ad392011-01-31 15:30:16 -05005004 return ret;
Yan, Zheng80825102009-11-12 09:35:36 +00005005}
5006
Chris Mason39279cc2007-06-12 06:35:45 -04005007static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5008{
David Howells2b0143b2015-03-17 22:25:59 +00005009 struct inode *inode = d_inode(dentry);
Li Zefanb83cc962010-12-20 16:04:08 +08005010 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04005011 int err;
5012
Li Zefanb83cc962010-12-20 16:04:08 +08005013 if (btrfs_root_readonly(root))
5014 return -EROFS;
5015
Chris Mason39279cc2007-06-12 06:35:45 -04005016 err = inode_change_ok(inode, attr);
5017 if (err)
5018 return err;
5019
Chris Mason5a3f23d2009-03-31 13:27:11 -04005020 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
Eric Sandeen3972f262013-01-12 02:57:22 +00005021 err = btrfs_setsize(inode, attr);
Yan, Zheng80825102009-11-12 09:35:36 +00005022 if (err)
5023 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04005024 }
Yan Zheng9036c102008-10-30 14:19:41 -04005025
Christoph Hellwig10257742010-06-04 11:30:02 +02005026 if (attr->ia_valid) {
5027 setattr_copy(inode, attr);
Josef Bacik0c4d2d92012-04-05 15:03:02 -04005028 inode_inc_iversion(inode);
Josef Bacik22c44fe2011-11-30 10:45:38 -05005029 err = btrfs_dirty_inode(inode);
Josef Bacik33268ea2008-07-24 12:16:36 -04005030
Josef Bacik22c44fe2011-11-30 10:45:38 -05005031 if (!err && attr->ia_valid & ATTR_MODE)
Christoph Hellwig996a7102013-12-20 05:16:43 -08005032 err = posix_acl_chmod(inode, inode->i_mode);
Christoph Hellwig10257742010-06-04 11:30:02 +02005033 }
5034
Chris Mason39279cc2007-06-12 06:35:45 -04005035 return err;
5036}
Chris Mason61295eb2008-01-14 16:24:38 -05005037
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00005038/*
5039 * While truncating the inode pages during eviction, we get the VFS calling
5040 * btrfs_invalidatepage() against each page of the inode. This is slow because
5041 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5042 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5043 * extent_state structures over and over, wasting lots of time.
5044 *
5045 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5046 * those expensive operations on a per page basis and do only the ordered io
5047 * finishing, while we release here the extent_map and extent_state structures,
5048 * without the excessive merging and splitting.
5049 */
5050static void evict_inode_truncate_pages(struct inode *inode)
5051{
5052 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5053 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5054 struct rb_node *node;
5055
5056 ASSERT(inode->i_state & I_FREEING);
Johannes Weiner91b0abe2014-04-03 14:47:49 -07005057 truncate_inode_pages_final(&inode->i_data);
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00005058
5059 write_lock(&map_tree->lock);
5060 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5061 struct extent_map *em;
5062
5063 node = rb_first(&map_tree->map);
5064 em = rb_entry(node, struct extent_map, rb_node);
Wang Shilong180589e2013-12-14 15:27:31 +08005065 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5066 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00005067 remove_extent_mapping(map_tree, em);
5068 free_extent_map(em);
Filipe Manana7064dd52014-08-08 02:47:05 +01005069 if (need_resched()) {
5070 write_unlock(&map_tree->lock);
5071 cond_resched();
5072 write_lock(&map_tree->lock);
5073 }
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00005074 }
5075 write_unlock(&map_tree->lock);
5076
Filipe Manana6ca07092015-05-26 00:55:42 +01005077 /*
5078 * Keep looping until we have no more ranges in the io tree.
5079 * We can have ongoing bios started by readpages (called from readahead)
Filipe Manana9c6429d2015-06-10 12:55:41 +01005080 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5081 * still in progress (unlocked the pages in the bio but did not yet
5082 * unlocked the ranges in the io tree). Therefore this means some
Filipe Manana6ca07092015-05-26 00:55:42 +01005083 * ranges can still be locked and eviction started because before
5084 * submitting those bios, which are executed by a separate task (work
5085 * queue kthread), inode references (inode->i_count) were not taken
5086 * (which would be dropped in the end io callback of each bio).
5087 * Therefore here we effectively end up waiting for those bios and
5088 * anyone else holding locked ranges without having bumped the inode's
5089 * reference count - if we don't do it, when they access the inode's
5090 * io_tree to unlock a range it may be too late, leading to an
5091 * use-after-free issue.
5092 */
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00005093 spin_lock(&io_tree->lock);
5094 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5095 struct extent_state *state;
5096 struct extent_state *cached_state = NULL;
Filipe Manana6ca07092015-05-26 00:55:42 +01005097 u64 start;
5098 u64 end;
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00005099
5100 node = rb_first(&io_tree->state);
5101 state = rb_entry(node, struct extent_state, rb_node);
Filipe Manana6ca07092015-05-26 00:55:42 +01005102 start = state->start;
5103 end = state->end;
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00005104 spin_unlock(&io_tree->lock);
5105
David Sterbaff13db42015-12-03 14:30:40 +01005106 lock_extent_bits(io_tree, start, end, &cached_state);
Qu Wenruob9d0b382015-09-29 10:35:16 +08005107
5108 /*
5109 * If still has DELALLOC flag, the extent didn't reach disk,
5110 * and its reserved space won't be freed by delayed_ref.
5111 * So we need to free its reserved space here.
5112 * (Refer to comment in btrfs_invalidatepage, case 2)
5113 *
5114 * Note, end is the bytenr of last byte, so we need + 1 here.
5115 */
5116 if (state->state & EXTENT_DELALLOC)
5117 btrfs_qgroup_free_data(inode, start, end - start + 1);
5118
Filipe Manana6ca07092015-05-26 00:55:42 +01005119 clear_extent_bit(io_tree, start, end,
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00005120 EXTENT_LOCKED | EXTENT_DIRTY |
5121 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5122 EXTENT_DEFRAG, 1, 1,
5123 &cached_state, GFP_NOFS);
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00005124
Filipe Manana7064dd52014-08-08 02:47:05 +01005125 cond_resched();
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00005126 spin_lock(&io_tree->lock);
5127 }
5128 spin_unlock(&io_tree->lock);
5129}
5130
Al Virobd555972010-06-07 11:35:40 -04005131void btrfs_evict_inode(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04005132{
5133 struct btrfs_trans_handle *trans;
5134 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik726c35f2011-09-26 15:46:06 -04005135 struct btrfs_block_rsv *rsv, *global_rsv;
Josef Bacik3bce8762015-02-24 12:35:51 -08005136 int steal_from_global = 0;
Josef Bacik07127182011-08-19 10:29:59 -04005137 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
Chris Mason39279cc2007-06-12 06:35:45 -04005138 int ret;
5139
liubo1abe9b82011-03-24 11:18:59 +00005140 trace_btrfs_inode_evict(inode);
5141
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00005142 evict_inode_truncate_pages(inode);
5143
Stefan Behrens69e9c6c2013-09-05 16:58:43 +02005144 if (inode->i_nlink &&
5145 ((btrfs_root_refs(&root->root_item) != 0 &&
5146 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5147 btrfs_is_free_space_inode(inode)))
Al Virobd555972010-06-07 11:35:40 -04005148 goto no_delete;
5149
Chris Mason39279cc2007-06-12 06:35:45 -04005150 if (is_bad_inode(inode)) {
Josef Bacik7b128762008-07-24 12:17:14 -04005151 btrfs_orphan_del(NULL, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04005152 goto no_delete;
5153 }
Al Virobd555972010-06-07 11:35:40 -04005154 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
Jeff Mahoneya30e5772015-09-11 21:44:17 -04005155 if (!special_file(inode->i_mode))
5156 btrfs_wait_ordered_range(inode, 0, (u64)-1);
Chris Mason5f39d392007-10-15 16:14:19 -04005157
Miao Xief6124962014-09-12 18:44:04 +08005158 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5159
Yan, Zhengc71bf092009-11-12 09:34:40 +00005160 if (root->fs_info->log_root_recovering) {
Liu Bo6bf02312012-06-25 21:59:09 -06005161 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
Josef Bacik8a35d952012-05-23 14:26:42 -04005162 &BTRFS_I(inode)->runtime_flags));
Yan, Zhengc71bf092009-11-12 09:34:40 +00005163 goto no_delete;
5164 }
5165
Yan, Zheng76dda932009-09-21 16:00:26 -04005166 if (inode->i_nlink > 0) {
Stefan Behrens69e9c6c2013-09-05 16:58:43 +02005167 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5168 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
Yan, Zheng76dda932009-09-21 16:00:26 -04005169 goto no_delete;
5170 }
5171
Miao Xie0e8c36a2012-12-19 06:59:51 +00005172 ret = btrfs_commit_inode_delayed_inode(inode);
5173 if (ret) {
5174 btrfs_orphan_del(NULL, inode);
5175 goto no_delete;
5176 }
5177
Miao Xie66d8f3d2012-09-06 04:02:28 -06005178 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
Josef Bacik4289a662011-08-05 13:22:24 -04005179 if (!rsv) {
5180 btrfs_orphan_del(NULL, inode);
5181 goto no_delete;
5182 }
Josef Bacik4a338542011-08-29 11:01:31 -04005183 rsv->size = min_size;
Josef Bacikca7e70f2012-08-27 17:48:15 -04005184 rsv->failfast = 1;
Josef Bacik726c35f2011-09-26 15:46:06 -04005185 global_rsv = &root->fs_info->global_block_rsv;
Josef Bacik4289a662011-08-05 13:22:24 -04005186
Chris Masondbe674a2008-07-17 12:54:05 -04005187 btrfs_i_size_write(inode, 0);
Chris Mason5f39d392007-10-15 16:14:19 -04005188
Josef Bacik4289a662011-08-05 13:22:24 -04005189 /*
Miao Xie8407aa42012-09-07 01:43:32 -06005190 * This is a bit simpler than btrfs_truncate since we've already
5191 * reserved our space for our orphan item in the unlink, so we just
5192 * need to reserve some slack space in case we add bytes and update
5193 * inode item when doing the truncate.
Josef Bacik4289a662011-08-05 13:22:24 -04005194 */
Yan, Zheng80825102009-11-12 09:35:36 +00005195 while (1) {
Miao Xie08e007d2012-10-16 11:33:38 +00005196 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5197 BTRFS_RESERVE_FLUSH_LIMIT);
Yan, Zheng80825102009-11-12 09:35:36 +00005198
Josef Bacik726c35f2011-09-26 15:46:06 -04005199 /*
5200 * Try and steal from the global reserve since we will
5201 * likely not use this space anyway, we want to try as
5202 * hard as possible to get this to work.
5203 */
5204 if (ret)
Josef Bacik3bce8762015-02-24 12:35:51 -08005205 steal_from_global++;
5206 else
5207 steal_from_global = 0;
5208 ret = 0;
Josef Bacik726c35f2011-09-26 15:46:06 -04005209
Josef Bacik3bce8762015-02-24 12:35:51 -08005210 /*
5211 * steal_from_global == 0: we reserved stuff, hooray!
5212 * steal_from_global == 1: we didn't reserve stuff, boo!
5213 * steal_from_global == 2: we've committed, still not a lot of
5214 * room but maybe we'll have room in the global reserve this
5215 * time.
5216 * steal_from_global == 3: abandon all hope!
5217 */
5218 if (steal_from_global > 2) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00005219 btrfs_warn(root->fs_info,
5220 "Could not get space for a delete, will truncate on mount %d",
5221 ret);
Josef Bacik4289a662011-08-05 13:22:24 -04005222 btrfs_orphan_del(NULL, inode);
5223 btrfs_free_block_rsv(root, rsv);
5224 goto no_delete;
Yan, Zhengd68fc572010-05-16 10:49:58 -04005225 }
5226
Miao Xie0e8c36a2012-12-19 06:59:51 +00005227 trans = btrfs_join_transaction(root);
Josef Bacik4289a662011-08-05 13:22:24 -04005228 if (IS_ERR(trans)) {
5229 btrfs_orphan_del(NULL, inode);
5230 btrfs_free_block_rsv(root, rsv);
5231 goto no_delete;
5232 }
5233
Josef Bacik3bce8762015-02-24 12:35:51 -08005234 /*
5235 * We can't just steal from the global reserve, we need tomake
5236 * sure there is room to do it, if not we need to commit and try
5237 * again.
5238 */
5239 if (steal_from_global) {
5240 if (!btrfs_check_space_for_delayed_refs(trans, root))
5241 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5242 min_size);
5243 else
5244 ret = -ENOSPC;
5245 }
5246
5247 /*
5248 * Couldn't steal from the global reserve, we have too much
5249 * pending stuff built up, commit the transaction and try it
5250 * again.
5251 */
5252 if (ret) {
5253 ret = btrfs_commit_transaction(trans, root);
5254 if (ret) {
5255 btrfs_orphan_del(NULL, inode);
5256 btrfs_free_block_rsv(root, rsv);
5257 goto no_delete;
5258 }
5259 continue;
5260 } else {
5261 steal_from_global = 0;
5262 }
5263
Josef Bacik4289a662011-08-05 13:22:24 -04005264 trans->block_rsv = rsv;
5265
Yan, Zhengd68fc572010-05-16 10:49:58 -04005266 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
Chris Mason28ed1342014-12-17 09:41:04 -08005267 if (ret != -ENOSPC && ret != -EAGAIN)
Yan, Zheng80825102009-11-12 09:35:36 +00005268 break;
5269
Miao Xie8407aa42012-09-07 01:43:32 -06005270 trans->block_rsv = &root->fs_info->trans_block_rsv;
Yan, Zheng80825102009-11-12 09:35:36 +00005271 btrfs_end_transaction(trans, root);
5272 trans = NULL;
Liu Bob53d3f52012-11-14 14:34:34 +00005273 btrfs_btree_balance_dirty(root);
Josef Bacik7b128762008-07-24 12:17:14 -04005274 }
5275
Josef Bacik4289a662011-08-05 13:22:24 -04005276 btrfs_free_block_rsv(root, rsv);
5277
Josef Bacik4ef31a42013-08-13 14:10:08 -04005278 /*
5279 * Errors here aren't a big deal, it just means we leave orphan items
5280 * in the tree. They will be cleaned up on the next mount.
5281 */
Yan, Zheng80825102009-11-12 09:35:36 +00005282 if (ret == 0) {
Josef Bacik4289a662011-08-05 13:22:24 -04005283 trans->block_rsv = root->orphan_block_rsv;
Josef Bacik4ef31a42013-08-13 14:10:08 -04005284 btrfs_orphan_del(trans, inode);
5285 } else {
5286 btrfs_orphan_del(NULL, inode);
Yan, Zheng80825102009-11-12 09:35:36 +00005287 }
Chris Mason85e21ba2008-01-29 15:11:36 -05005288
Josef Bacik4289a662011-08-05 13:22:24 -04005289 trans->block_rsv = &root->fs_info->trans_block_rsv;
Li Zefan581bb052011-04-20 10:06:11 +08005290 if (!(root == root->fs_info->tree_root ||
5291 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
Li Zefan33345d012011-04-20 10:31:50 +08005292 btrfs_return_ino(root, btrfs_ino(inode));
Li Zefan581bb052011-04-20 10:06:11 +08005293
Chris Mason54aa1f42007-06-22 14:16:25 -04005294 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00005295 btrfs_btree_balance_dirty(root);
Chris Mason39279cc2007-06-12 06:35:45 -04005296no_delete:
Miao Xie89042e52013-05-15 07:48:15 +00005297 btrfs_remove_delayed_node(inode);
Jan Karadbd57682012-05-03 14:48:02 +02005298 clear_inode(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04005299}
5300
5301/*
5302 * this returns the key found in the dir entry in the location pointer.
5303 * If no dir entries were found, location->objectid is 0.
5304 */
5305static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5306 struct btrfs_key *location)
5307{
5308 const char *name = dentry->d_name.name;
5309 int namelen = dentry->d_name.len;
5310 struct btrfs_dir_item *di;
5311 struct btrfs_path *path;
5312 struct btrfs_root *root = BTRFS_I(dir)->root;
Yan0d9f7f32007-10-25 15:48:28 -04005313 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04005314
5315 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07005316 if (!path)
5317 return -ENOMEM;
Chris Mason39544012007-12-12 14:38:19 -05005318
Li Zefan33345d012011-04-20 10:31:50 +08005319 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
Chris Mason39279cc2007-06-12 06:35:45 -04005320 namelen, 0);
Yan0d9f7f32007-10-25 15:48:28 -04005321 if (IS_ERR(di))
5322 ret = PTR_ERR(di);
Chris Masond3977122009-01-05 21:25:51 -05005323
David Sterbac7040052011-04-19 18:00:01 +02005324 if (IS_ERR_OR_NULL(di))
Chris Mason39544012007-12-12 14:38:19 -05005325 goto out_err;
Chris Masond3977122009-01-05 21:25:51 -05005326
Chris Mason5f39d392007-10-15 16:14:19 -04005327 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
Chris Mason39279cc2007-06-12 06:35:45 -04005328out:
Chris Mason39279cc2007-06-12 06:35:45 -04005329 btrfs_free_path(path);
5330 return ret;
Chris Mason39544012007-12-12 14:38:19 -05005331out_err:
5332 location->objectid = 0;
5333 goto out;
Chris Mason39279cc2007-06-12 06:35:45 -04005334}
5335
5336/*
5337 * when we hit a tree root in a directory, the btrfs part of the inode
5338 * needs to be changed to reflect the root directory of the tree root. This
5339 * is kind of like crossing a mount point.
5340 */
5341static int fixup_tree_root_location(struct btrfs_root *root,
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005342 struct inode *dir,
5343 struct dentry *dentry,
5344 struct btrfs_key *location,
5345 struct btrfs_root **sub_root)
Chris Mason39279cc2007-06-12 06:35:45 -04005346{
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005347 struct btrfs_path *path;
5348 struct btrfs_root *new_root;
5349 struct btrfs_root_ref *ref;
5350 struct extent_buffer *leaf;
David Sterba1d4c08e2015-01-02 19:36:14 +01005351 struct btrfs_key key;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005352 int ret;
5353 int err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04005354
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005355 path = btrfs_alloc_path();
5356 if (!path) {
5357 err = -ENOMEM;
5358 goto out;
5359 }
Chris Mason39279cc2007-06-12 06:35:45 -04005360
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005361 err = -ENOENT;
David Sterba1d4c08e2015-01-02 19:36:14 +01005362 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5363 key.type = BTRFS_ROOT_REF_KEY;
5364 key.offset = location->objectid;
5365
5366 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5367 0, 0);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005368 if (ret) {
5369 if (ret < 0)
5370 err = ret;
5371 goto out;
5372 }
Chris Mason39279cc2007-06-12 06:35:45 -04005373
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005374 leaf = path->nodes[0];
5375 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
Li Zefan33345d012011-04-20 10:31:50 +08005376 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005377 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5378 goto out;
5379
5380 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5381 (unsigned long)(ref + 1),
5382 dentry->d_name.len);
5383 if (ret)
5384 goto out;
5385
David Sterbab3b4aa72011-04-21 01:20:15 +02005386 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005387
5388 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5389 if (IS_ERR(new_root)) {
5390 err = PTR_ERR(new_root);
5391 goto out;
5392 }
5393
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005394 *sub_root = new_root;
5395 location->objectid = btrfs_root_dirid(&new_root->root_item);
5396 location->type = BTRFS_INODE_ITEM_KEY;
Chris Mason39279cc2007-06-12 06:35:45 -04005397 location->offset = 0;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005398 err = 0;
5399out:
5400 btrfs_free_path(path);
5401 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04005402}
5403
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005404static void inode_tree_add(struct inode *inode)
5405{
5406 struct btrfs_root *root = BTRFS_I(inode)->root;
5407 struct btrfs_inode *entry;
From: Nick Piggin03e860b2009-08-21 10:09:44 +02005408 struct rb_node **p;
5409 struct rb_node *parent;
Filipe David Borba Mananacef21932013-09-02 12:19:13 +01005410 struct rb_node *new = &BTRFS_I(inode)->rb_node;
Li Zefan33345d012011-04-20 10:31:50 +08005411 u64 ino = btrfs_ino(inode);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005412
Al Viro1d3382cb2010-10-23 15:19:20 -04005413 if (inode_unhashed(inode))
Yan, Zheng76dda932009-09-21 16:00:26 -04005414 return;
Miao Xiee1409ce2013-05-15 07:48:16 +00005415 parent = NULL;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005416 spin_lock(&root->inode_lock);
Miao Xiee1409ce2013-05-15 07:48:16 +00005417 p = &root->inode_tree.rb_node;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005418 while (*p) {
5419 parent = *p;
5420 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5421
Li Zefan33345d012011-04-20 10:31:50 +08005422 if (ino < btrfs_ino(&entry->vfs_inode))
From: Nick Piggin03e860b2009-08-21 10:09:44 +02005423 p = &parent->rb_left;
Li Zefan33345d012011-04-20 10:31:50 +08005424 else if (ino > btrfs_ino(&entry->vfs_inode))
From: Nick Piggin03e860b2009-08-21 10:09:44 +02005425 p = &parent->rb_right;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005426 else {
5427 WARN_ON(!(entry->vfs_inode.i_state &
Al Viroa4ffdde2010-06-02 17:38:30 -04005428 (I_WILL_FREE | I_FREEING)));
Filipe David Borba Mananacef21932013-09-02 12:19:13 +01005429 rb_replace_node(parent, new, &root->inode_tree);
From: Nick Piggin03e860b2009-08-21 10:09:44 +02005430 RB_CLEAR_NODE(parent);
5431 spin_unlock(&root->inode_lock);
Filipe David Borba Mananacef21932013-09-02 12:19:13 +01005432 return;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005433 }
5434 }
Filipe David Borba Mananacef21932013-09-02 12:19:13 +01005435 rb_link_node(new, parent, p);
5436 rb_insert_color(new, &root->inode_tree);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005437 spin_unlock(&root->inode_lock);
5438}
5439
5440static void inode_tree_del(struct inode *inode)
5441{
5442 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng76dda932009-09-21 16:00:26 -04005443 int empty = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005444
From: Nick Piggin03e860b2009-08-21 10:09:44 +02005445 spin_lock(&root->inode_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005446 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005447 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005448 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
Yan, Zheng76dda932009-09-21 16:00:26 -04005449 empty = RB_EMPTY_ROOT(&root->inode_tree);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005450 }
From: Nick Piggin03e860b2009-08-21 10:09:44 +02005451 spin_unlock(&root->inode_lock);
Yan, Zheng76dda932009-09-21 16:00:26 -04005452
Stefan Behrens69e9c6c2013-09-05 16:58:43 +02005453 if (empty && btrfs_root_refs(&root->root_item) == 0) {
Yan, Zheng76dda932009-09-21 16:00:26 -04005454 synchronize_srcu(&root->fs_info->subvol_srcu);
5455 spin_lock(&root->inode_lock);
5456 empty = RB_EMPTY_ROOT(&root->inode_tree);
5457 spin_unlock(&root->inode_lock);
5458 if (empty)
5459 btrfs_add_dead_root(root);
5460 }
5461}
5462
Jeff Mahoney143bede2012-03-01 14:56:26 +01005463void btrfs_invalidate_inodes(struct btrfs_root *root)
Yan, Zheng76dda932009-09-21 16:00:26 -04005464{
5465 struct rb_node *node;
5466 struct rb_node *prev;
5467 struct btrfs_inode *entry;
5468 struct inode *inode;
5469 u64 objectid = 0;
5470
Liu Bo7813b3d2014-02-10 17:37:25 +08005471 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5472 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
Yan, Zheng76dda932009-09-21 16:00:26 -04005473
5474 spin_lock(&root->inode_lock);
5475again:
5476 node = root->inode_tree.rb_node;
5477 prev = NULL;
5478 while (node) {
5479 prev = node;
5480 entry = rb_entry(node, struct btrfs_inode, rb_node);
5481
Li Zefan33345d012011-04-20 10:31:50 +08005482 if (objectid < btrfs_ino(&entry->vfs_inode))
Yan, Zheng76dda932009-09-21 16:00:26 -04005483 node = node->rb_left;
Li Zefan33345d012011-04-20 10:31:50 +08005484 else if (objectid > btrfs_ino(&entry->vfs_inode))
Yan, Zheng76dda932009-09-21 16:00:26 -04005485 node = node->rb_right;
5486 else
5487 break;
5488 }
5489 if (!node) {
5490 while (prev) {
5491 entry = rb_entry(prev, struct btrfs_inode, rb_node);
Li Zefan33345d012011-04-20 10:31:50 +08005492 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
Yan, Zheng76dda932009-09-21 16:00:26 -04005493 node = prev;
5494 break;
5495 }
5496 prev = rb_next(prev);
5497 }
5498 }
5499 while (node) {
5500 entry = rb_entry(node, struct btrfs_inode, rb_node);
Li Zefan33345d012011-04-20 10:31:50 +08005501 objectid = btrfs_ino(&entry->vfs_inode) + 1;
Yan, Zheng76dda932009-09-21 16:00:26 -04005502 inode = igrab(&entry->vfs_inode);
5503 if (inode) {
5504 spin_unlock(&root->inode_lock);
5505 if (atomic_read(&inode->i_count) > 1)
5506 d_prune_aliases(inode);
5507 /*
Al Viro45321ac2010-06-07 13:43:19 -04005508 * btrfs_drop_inode will have it removed from
Yan, Zheng76dda932009-09-21 16:00:26 -04005509 * the inode cache when its usage count
5510 * hits zero.
5511 */
5512 iput(inode);
5513 cond_resched();
5514 spin_lock(&root->inode_lock);
5515 goto again;
5516 }
5517
5518 if (cond_resched_lock(&root->inode_lock))
5519 goto again;
5520
5521 node = rb_next(node);
5522 }
5523 spin_unlock(&root->inode_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005524}
5525
Chris Masone02119d2008-09-05 16:13:11 -04005526static int btrfs_init_locked_inode(struct inode *inode, void *p)
5527{
5528 struct btrfs_iget_args *args = p;
Chris Mason90d3e592014-01-09 17:28:00 -08005529 inode->i_ino = args->location->objectid;
5530 memcpy(&BTRFS_I(inode)->location, args->location,
5531 sizeof(*args->location));
Chris Masone02119d2008-09-05 16:13:11 -04005532 BTRFS_I(inode)->root = args->root;
Chris Mason39279cc2007-06-12 06:35:45 -04005533 return 0;
5534}
5535
5536static int btrfs_find_actor(struct inode *inode, void *opaque)
5537{
5538 struct btrfs_iget_args *args = opaque;
Chris Mason90d3e592014-01-09 17:28:00 -08005539 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
Chris Masond3977122009-01-05 21:25:51 -05005540 args->root == BTRFS_I(inode)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04005541}
5542
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005543static struct inode *btrfs_iget_locked(struct super_block *s,
Chris Mason90d3e592014-01-09 17:28:00 -08005544 struct btrfs_key *location,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005545 struct btrfs_root *root)
Chris Mason39279cc2007-06-12 06:35:45 -04005546{
5547 struct inode *inode;
5548 struct btrfs_iget_args args;
Chris Mason90d3e592014-01-09 17:28:00 -08005549 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
Filipe David Borba Manana778ba822013-10-06 22:22:33 +01005550
Chris Mason90d3e592014-01-09 17:28:00 -08005551 args.location = location;
Chris Mason39279cc2007-06-12 06:35:45 -04005552 args.root = root;
5553
Filipe David Borba Manana778ba822013-10-06 22:22:33 +01005554 inode = iget5_locked(s, hashval, btrfs_find_actor,
Chris Mason39279cc2007-06-12 06:35:45 -04005555 btrfs_init_locked_inode,
5556 (void *)&args);
5557 return inode;
5558}
5559
Balaji Rao1a54ef82008-07-21 02:01:04 +05305560/* Get an inode object given its location and corresponding root.
5561 * Returns in *is_new if the inode was read from disk
5562 */
5563struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
Josef Bacik73f73412009-12-04 17:38:27 +00005564 struct btrfs_root *root, int *new)
Balaji Rao1a54ef82008-07-21 02:01:04 +05305565{
5566 struct inode *inode;
5567
Chris Mason90d3e592014-01-09 17:28:00 -08005568 inode = btrfs_iget_locked(s, location, root);
Balaji Rao1a54ef82008-07-21 02:01:04 +05305569 if (!inode)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005570 return ERR_PTR(-ENOMEM);
Balaji Rao1a54ef82008-07-21 02:01:04 +05305571
5572 if (inode->i_state & I_NEW) {
Balaji Rao1a54ef82008-07-21 02:01:04 +05305573 btrfs_read_locked_inode(inode);
Mark Fasheh1748f842011-07-12 11:25:31 -07005574 if (!is_bad_inode(inode)) {
5575 inode_tree_add(inode);
5576 unlock_new_inode(inode);
5577 if (new)
5578 *new = 1;
5579 } else {
Sergei Trofimoviche0b6d652011-09-11 10:52:24 -04005580 unlock_new_inode(inode);
5581 iput(inode);
5582 inode = ERR_PTR(-ESTALE);
Mark Fasheh1748f842011-07-12 11:25:31 -07005583 }
5584 }
5585
Balaji Rao1a54ef82008-07-21 02:01:04 +05305586 return inode;
5587}
5588
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005589static struct inode *new_simple_dir(struct super_block *s,
5590 struct btrfs_key *key,
5591 struct btrfs_root *root)
5592{
5593 struct inode *inode = new_inode(s);
5594
5595 if (!inode)
5596 return ERR_PTR(-ENOMEM);
5597
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005598 BTRFS_I(inode)->root = root;
5599 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
Josef Bacik72ac3c02012-05-23 14:13:11 -04005600 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005601
5602 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
Li Zefan848cce02012-02-21 17:04:28 +08005603 inode->i_op = &btrfs_dir_ro_inode_operations;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005604 inode->i_fop = &simple_dir_operations;
5605 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
Deepa Dinamani04b285f2016-02-06 23:57:21 -08005606 inode->i_mtime = current_fs_time(inode->i_sb);
chandan r9cc97d62012-07-04 12:48:07 +05305607 inode->i_atime = inode->i_mtime;
5608 inode->i_ctime = inode->i_mtime;
5609 BTRFS_I(inode)->i_otime = inode->i_mtime;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005610
5611 return inode;
5612}
5613
Chris Mason3de45862008-11-17 21:02:50 -05005614struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
Chris Mason39279cc2007-06-12 06:35:45 -04005615{
Chris Masond3977122009-01-05 21:25:51 -05005616 struct inode *inode;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005617 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04005618 struct btrfs_root *sub_root = root;
5619 struct btrfs_key location;
Yan, Zheng76dda932009-09-21 16:00:26 -04005620 int index;
Josef Bacikb4aff1f2011-06-28 16:18:59 -04005621 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04005622
5623 if (dentry->d_name.len > BTRFS_NAME_LEN)
5624 return ERR_PTR(-ENAMETOOLONG);
Chris Mason5f39d392007-10-15 16:14:19 -04005625
Jeff Layton39e3c952012-11-28 11:30:53 -05005626 ret = btrfs_inode_by_name(dir, dentry, &location);
Chris Mason39279cc2007-06-12 06:35:45 -04005627 if (ret < 0)
5628 return ERR_PTR(ret);
Chris Mason5f39d392007-10-15 16:14:19 -04005629
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005630 if (location.objectid == 0)
Tsutomu Itoh5662344b32013-12-13 09:51:42 +09005631 return ERR_PTR(-ENOENT);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005632
5633 if (location.type == BTRFS_INODE_ITEM_KEY) {
Josef Bacik73f73412009-12-04 17:38:27 +00005634 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005635 return inode;
Chris Mason39279cc2007-06-12 06:35:45 -04005636 }
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005637
5638 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5639
Yan, Zheng76dda932009-09-21 16:00:26 -04005640 index = srcu_read_lock(&root->fs_info->subvol_srcu);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005641 ret = fixup_tree_root_location(root, dir, dentry,
5642 &location, &sub_root);
5643 if (ret < 0) {
5644 if (ret != -ENOENT)
5645 inode = ERR_PTR(ret);
5646 else
5647 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5648 } else {
Josef Bacik73f73412009-12-04 17:38:27 +00005649 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04005650 }
Yan, Zheng76dda932009-09-21 16:00:26 -04005651 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5652
Julia Lawall34d19ba2011-01-24 19:55:19 +00005653 if (!IS_ERR(inode) && root != sub_root) {
Yan, Zhengc71bf092009-11-12 09:34:40 +00005654 down_read(&root->fs_info->cleanup_work_sem);
5655 if (!(inode->i_sb->s_flags & MS_RDONLY))
Josef Bacik66b4ffd2011-01-31 16:22:42 -05005656 ret = btrfs_orphan_cleanup(sub_root);
Yan, Zhengc71bf092009-11-12 09:34:40 +00005657 up_read(&root->fs_info->cleanup_work_sem);
Josef Bacik01cd3362013-06-03 21:39:49 -04005658 if (ret) {
5659 iput(inode);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05005660 inode = ERR_PTR(ret);
Josef Bacik01cd3362013-06-03 21:39:49 -04005661 }
Yan, Zhengc71bf092009-11-12 09:34:40 +00005662 }
5663
Chris Mason3de45862008-11-17 21:02:50 -05005664 return inode;
5665}
5666
Nick Pigginfe15ce42011-01-07 17:49:23 +11005667static int btrfs_dentry_delete(const struct dentry *dentry)
Yan, Zheng76dda932009-09-21 16:00:26 -04005668{
5669 struct btrfs_root *root;
David Howells2b0143b2015-03-17 22:25:59 +00005670 struct inode *inode = d_inode(dentry);
Yan, Zheng76dda932009-09-21 16:00:26 -04005671
Li Zefan848cce02012-02-21 17:04:28 +08005672 if (!inode && !IS_ROOT(dentry))
David Howells2b0143b2015-03-17 22:25:59 +00005673 inode = d_inode(dentry->d_parent);
Yan, Zheng76dda932009-09-21 16:00:26 -04005674
Li Zefan848cce02012-02-21 17:04:28 +08005675 if (inode) {
5676 root = BTRFS_I(inode)->root;
Yan, Zhengefefb142009-10-09 09:25:16 -04005677 if (btrfs_root_refs(&root->root_item) == 0)
5678 return 1;
Li Zefan848cce02012-02-21 17:04:28 +08005679
5680 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5681 return 1;
Yan, Zhengefefb142009-10-09 09:25:16 -04005682 }
Yan, Zheng76dda932009-09-21 16:00:26 -04005683 return 0;
5684}
5685
Josef Bacikb4aff1f2011-06-28 16:18:59 -04005686static void btrfs_dentry_release(struct dentry *dentry)
5687{
Daeseok Youn944a4512014-04-14 15:37:02 +09005688 kfree(dentry->d_fsdata);
Josef Bacikb4aff1f2011-06-28 16:18:59 -04005689}
5690
Chris Mason3de45862008-11-17 21:02:50 -05005691static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
Al Viro00cd8dd2012-06-10 17:13:09 -04005692 unsigned int flags)
Chris Mason3de45862008-11-17 21:02:50 -05005693{
Tsutomu Itoh5662344b32013-12-13 09:51:42 +09005694 struct inode *inode;
Josef Bacika66e7cc2011-09-18 10:34:03 -04005695
Tsutomu Itoh5662344b32013-12-13 09:51:42 +09005696 inode = btrfs_lookup_dentry(dir, dentry);
5697 if (IS_ERR(inode)) {
5698 if (PTR_ERR(inode) == -ENOENT)
5699 inode = NULL;
5700 else
5701 return ERR_CAST(inode);
5702 }
5703
Al Viro41d28bc2014-10-12 22:24:21 -04005704 return d_splice_alias(inode, dentry);
Chris Mason39279cc2007-06-12 06:35:45 -04005705}
5706
Miao Xie16cdcec2011-04-22 18:12:22 +08005707unsigned char btrfs_filetype_table[] = {
Chris Mason39279cc2007-06-12 06:35:45 -04005708 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5709};
5710
Al Viro9cdda8d2013-05-22 16:48:09 -04005711static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
Chris Mason39279cc2007-06-12 06:35:45 -04005712{
Al Viro9cdda8d2013-05-22 16:48:09 -04005713 struct inode *inode = file_inode(file);
Chris Mason39279cc2007-06-12 06:35:45 -04005714 struct btrfs_root *root = BTRFS_I(inode)->root;
5715 struct btrfs_item *item;
5716 struct btrfs_dir_item *di;
5717 struct btrfs_key key;
Chris Mason5f39d392007-10-15 16:14:19 -04005718 struct btrfs_key found_key;
Chris Mason39279cc2007-06-12 06:35:45 -04005719 struct btrfs_path *path;
Miao Xie16cdcec2011-04-22 18:12:22 +08005720 struct list_head ins_list;
5721 struct list_head del_list;
Chris Mason39279cc2007-06-12 06:35:45 -04005722 int ret;
Chris Mason5f39d392007-10-15 16:14:19 -04005723 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04005724 int slot;
Chris Mason39279cc2007-06-12 06:35:45 -04005725 unsigned char d_type;
5726 int over = 0;
5727 u32 di_cur;
5728 u32 di_total;
5729 u32 di_len;
5730 int key_type = BTRFS_DIR_INDEX_KEY;
Chris Mason5f39d392007-10-15 16:14:19 -04005731 char tmp_name[32];
5732 char *name_ptr;
5733 int name_len;
Al Viro9cdda8d2013-05-22 16:48:09 -04005734 int is_curr = 0; /* ctx->pos points to the current index? */
David Sterbabc4ef752015-11-13 13:44:28 +01005735 bool emitted;
Chris Mason39279cc2007-06-12 06:35:45 -04005736
5737 /* FIXME, use a real flag for deciding about the key type */
5738 if (root->fs_info->tree_root == root)
5739 key_type = BTRFS_DIR_ITEM_KEY;
Chris Mason5f39d392007-10-15 16:14:19 -04005740
Al Viro9cdda8d2013-05-22 16:48:09 -04005741 if (!dir_emit_dots(file, ctx))
5742 return 0;
5743
David Woodhouse49593bf2008-08-17 17:08:36 +01005744 path = btrfs_alloc_path();
Miao Xie16cdcec2011-04-22 18:12:22 +08005745 if (!path)
5746 return -ENOMEM;
Chris Masonff5714c2011-05-28 07:00:39 -04005747
David Sterbae4058b52015-11-27 16:31:35 +01005748 path->reada = READA_FORWARD;
David Woodhouse49593bf2008-08-17 17:08:36 +01005749
Miao Xie16cdcec2011-04-22 18:12:22 +08005750 if (key_type == BTRFS_DIR_INDEX_KEY) {
5751 INIT_LIST_HEAD(&ins_list);
5752 INIT_LIST_HEAD(&del_list);
5753 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5754 }
5755
David Sterba962a2982014-06-04 18:41:45 +02005756 key.type = key_type;
Al Viro9cdda8d2013-05-22 16:48:09 -04005757 key.offset = ctx->pos;
Li Zefan33345d012011-04-20 10:31:50 +08005758 key.objectid = btrfs_ino(inode);
Chris Mason5f39d392007-10-15 16:14:19 -04005759
Chris Mason39279cc2007-06-12 06:35:45 -04005760 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5761 if (ret < 0)
5762 goto err;
David Woodhouse49593bf2008-08-17 17:08:36 +01005763
David Sterbabc4ef752015-11-13 13:44:28 +01005764 emitted = false;
David Woodhouse49593bf2008-08-17 17:08:36 +01005765 while (1) {
Chris Mason5f39d392007-10-15 16:14:19 -04005766 leaf = path->nodes[0];
Chris Mason39279cc2007-06-12 06:35:45 -04005767 slot = path->slots[0];
Li Zefanb9e03af2011-03-23 10:43:58 +08005768 if (slot >= btrfs_header_nritems(leaf)) {
5769 ret = btrfs_next_leaf(root, path);
5770 if (ret < 0)
5771 goto err;
5772 else if (ret > 0)
5773 break;
5774 continue;
Chris Mason39279cc2007-06-12 06:35:45 -04005775 }
Chris Mason3de45862008-11-17 21:02:50 -05005776
Ross Kirkdd3cc162013-09-16 15:58:09 +01005777 item = btrfs_item_nr(slot);
Chris Mason5f39d392007-10-15 16:14:19 -04005778 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5779
5780 if (found_key.objectid != key.objectid)
Chris Mason39279cc2007-06-12 06:35:45 -04005781 break;
David Sterba962a2982014-06-04 18:41:45 +02005782 if (found_key.type != key_type)
Chris Mason39279cc2007-06-12 06:35:45 -04005783 break;
Al Viro9cdda8d2013-05-22 16:48:09 -04005784 if (found_key.offset < ctx->pos)
Li Zefanb9e03af2011-03-23 10:43:58 +08005785 goto next;
Miao Xie16cdcec2011-04-22 18:12:22 +08005786 if (key_type == BTRFS_DIR_INDEX_KEY &&
5787 btrfs_should_delete_dir_index(&del_list,
5788 found_key.offset))
5789 goto next;
Chris Mason5f39d392007-10-15 16:14:19 -04005790
Al Viro9cdda8d2013-05-22 16:48:09 -04005791 ctx->pos = found_key.offset;
Miao Xie16cdcec2011-04-22 18:12:22 +08005792 is_curr = 1;
David Woodhouse49593bf2008-08-17 17:08:36 +01005793
Chris Mason39279cc2007-06-12 06:35:45 -04005794 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5795 di_cur = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04005796 di_total = btrfs_item_size(leaf, item);
David Woodhouse49593bf2008-08-17 17:08:36 +01005797
5798 while (di_cur < di_total) {
Chris Mason5f39d392007-10-15 16:14:19 -04005799 struct btrfs_key location;
5800
Josef Bacik22a94d42011-03-16 16:47:17 -04005801 if (verify_dir_item(root, leaf, di))
5802 break;
5803
Chris Mason5f39d392007-10-15 16:14:19 -04005804 name_len = btrfs_dir_name_len(leaf, di);
David Woodhouse49593bf2008-08-17 17:08:36 +01005805 if (name_len <= sizeof(tmp_name)) {
Chris Mason5f39d392007-10-15 16:14:19 -04005806 name_ptr = tmp_name;
5807 } else {
David Sterba49e350a2016-02-11 14:25:38 +01005808 name_ptr = kmalloc(name_len, GFP_KERNEL);
David Woodhouse49593bf2008-08-17 17:08:36 +01005809 if (!name_ptr) {
5810 ret = -ENOMEM;
5811 goto err;
5812 }
Chris Mason5f39d392007-10-15 16:14:19 -04005813 }
5814 read_extent_buffer(leaf, name_ptr,
5815 (unsigned long)(di + 1), name_len);
5816
5817 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5818 btrfs_dir_item_key_to_cpu(leaf, di, &location);
Chris Mason3de45862008-11-17 21:02:50 -05005819
Josef Bacikb4aff1f2011-06-28 16:18:59 -04005820
Chris Mason3de45862008-11-17 21:02:50 -05005821 /* is this a reference to our own snapshot? If so
Arne Jansen8c9c2bf2012-02-25 09:09:30 +01005822 * skip it.
5823 *
5824 * In contrast to old kernels, we insert the snapshot's
5825 * dir item and dir index after it has been created, so
5826 * we won't find a reference to our own snapshot. We
5827 * still keep the following code for backward
5828 * compatibility.
Chris Mason3de45862008-11-17 21:02:50 -05005829 */
5830 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5831 location.objectid == root->root_key.objectid) {
5832 over = 0;
5833 goto skip;
5834 }
Al Viro9cdda8d2013-05-22 16:48:09 -04005835 over = !dir_emit(ctx, name_ptr, name_len,
5836 location.objectid, d_type);
Chris Mason5f39d392007-10-15 16:14:19 -04005837
Chris Mason3de45862008-11-17 21:02:50 -05005838skip:
Chris Mason5f39d392007-10-15 16:14:19 -04005839 if (name_ptr != tmp_name)
5840 kfree(name_ptr);
5841
Chris Mason39279cc2007-06-12 06:35:45 -04005842 if (over)
5843 goto nopos;
David Sterbabc4ef752015-11-13 13:44:28 +01005844 emitted = true;
Josef Bacik5103e942007-11-16 11:45:54 -05005845 di_len = btrfs_dir_name_len(leaf, di) +
David Woodhouse49593bf2008-08-17 17:08:36 +01005846 btrfs_dir_data_len(leaf, di) + sizeof(*di);
Chris Mason39279cc2007-06-12 06:35:45 -04005847 di_cur += di_len;
5848 di = (struct btrfs_dir_item *)((char *)di + di_len);
5849 }
Li Zefanb9e03af2011-03-23 10:43:58 +08005850next:
5851 path->slots[0]++;
Chris Mason39279cc2007-06-12 06:35:45 -04005852 }
David Woodhouse49593bf2008-08-17 17:08:36 +01005853
Miao Xie16cdcec2011-04-22 18:12:22 +08005854 if (key_type == BTRFS_DIR_INDEX_KEY) {
5855 if (is_curr)
Al Viro9cdda8d2013-05-22 16:48:09 -04005856 ctx->pos++;
David Sterbabc4ef752015-11-13 13:44:28 +01005857 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list, &emitted);
Miao Xie16cdcec2011-04-22 18:12:22 +08005858 if (ret)
5859 goto nopos;
5860 }
5861
David Sterbabc4ef752015-11-13 13:44:28 +01005862 /*
5863 * If we haven't emitted any dir entry, we must not touch ctx->pos as
5864 * it was was set to the termination value in previous call. We assume
5865 * that "." and ".." were emitted if we reach this point and set the
5866 * termination value as well for an empty directory.
5867 */
5868 if (ctx->pos > 2 && !emitted)
5869 goto nopos;
5870
David Woodhouse49593bf2008-08-17 17:08:36 +01005871 /* Reached end of directory/root. Bump pos past the last item. */
Zach Browndb62efb2013-07-11 16:19:42 -07005872 ctx->pos++;
5873
5874 /*
5875 * Stop new entries from being returned after we return the last
5876 * entry.
5877 *
5878 * New directory entries are assigned a strictly increasing
5879 * offset. This means that new entries created during readdir
5880 * are *guaranteed* to be seen in the future by that readdir.
5881 * This has broken buggy programs which operate on names as
5882 * they're returned by readdir. Until we re-use freed offsets
5883 * we have this hack to stop new entries from being returned
5884 * under the assumption that they'll never reach this huge
5885 * offset.
5886 *
5887 * This is being careful not to overflow 32bit loff_t unless the
5888 * last entry requires it because doing so has broken 32bit apps
5889 * in the past.
5890 */
5891 if (key_type == BTRFS_DIR_INDEX_KEY) {
5892 if (ctx->pos >= INT_MAX)
5893 ctx->pos = LLONG_MAX;
5894 else
5895 ctx->pos = INT_MAX;
5896 }
Chris Mason39279cc2007-06-12 06:35:45 -04005897nopos:
5898 ret = 0;
5899err:
Miao Xie16cdcec2011-04-22 18:12:22 +08005900 if (key_type == BTRFS_DIR_INDEX_KEY)
5901 btrfs_put_delayed_items(&ins_list, &del_list);
Chris Mason39279cc2007-06-12 06:35:45 -04005902 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04005903 return ret;
5904}
5905
Christoph Hellwiga9185b42010-03-05 09:21:37 +01005906int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
Chris Mason39279cc2007-06-12 06:35:45 -04005907{
5908 struct btrfs_root *root = BTRFS_I(inode)->root;
5909 struct btrfs_trans_handle *trans;
5910 int ret = 0;
Josef Bacik0af3d002010-06-21 14:48:16 -04005911 bool nolock = false;
Chris Mason39279cc2007-06-12 06:35:45 -04005912
Josef Bacik72ac3c02012-05-23 14:13:11 -04005913 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
Chris Mason4ca8b412008-08-05 13:30:48 -04005914 return 0;
5915
Liu Bo83eea1f2012-07-10 05:28:39 -06005916 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
Li Zefan82d59022011-04-20 10:33:24 +08005917 nolock = true;
Josef Bacik0af3d002010-06-21 14:48:16 -04005918
Christoph Hellwiga9185b42010-03-05 09:21:37 +01005919 if (wbc->sync_mode == WB_SYNC_ALL) {
Josef Bacik0af3d002010-06-21 14:48:16 -04005920 if (nolock)
Josef Bacik7a7eaa42011-04-13 12:54:33 -04005921 trans = btrfs_join_transaction_nolock(root);
Josef Bacik0af3d002010-06-21 14:48:16 -04005922 else
Josef Bacik7a7eaa42011-04-13 12:54:33 -04005923 trans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00005924 if (IS_ERR(trans))
5925 return PTR_ERR(trans);
Miao Xiea698d0752012-09-20 01:51:59 -06005926 ret = btrfs_commit_transaction(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04005927 }
5928 return ret;
5929}
5930
5931/*
Chris Mason54aa1f42007-06-22 14:16:25 -04005932 * This is somewhat expensive, updating the tree every time the
Chris Mason39279cc2007-06-12 06:35:45 -04005933 * inode changes. But, it is most likely to find the inode in cache.
5934 * FIXME, needs more benchmarking...there are no reasons other than performance
5935 * to keep or drop this code.
5936 */
Eric Sandeen48a3b632013-04-25 20:41:01 +00005937static int btrfs_dirty_inode(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04005938{
5939 struct btrfs_root *root = BTRFS_I(inode)->root;
5940 struct btrfs_trans_handle *trans;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04005941 int ret;
5942
Josef Bacik72ac3c02012-05-23 14:13:11 -04005943 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
Josef Bacik22c44fe2011-11-30 10:45:38 -05005944 return 0;
Chris Mason39279cc2007-06-12 06:35:45 -04005945
Josef Bacik7a7eaa42011-04-13 12:54:33 -04005946 trans = btrfs_join_transaction(root);
Josef Bacik22c44fe2011-11-30 10:45:38 -05005947 if (IS_ERR(trans))
5948 return PTR_ERR(trans);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04005949
5950 ret = btrfs_update_inode(trans, root, inode);
Chris Mason94b60442010-05-26 11:02:00 -04005951 if (ret && ret == -ENOSPC) {
5952 /* whoops, lets try again with the full transaction */
5953 btrfs_end_transaction(trans, root);
5954 trans = btrfs_start_transaction(root, 1);
Josef Bacik22c44fe2011-11-30 10:45:38 -05005955 if (IS_ERR(trans))
5956 return PTR_ERR(trans);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04005957
Chris Mason94b60442010-05-26 11:02:00 -04005958 ret = btrfs_update_inode(trans, root, inode);
Chris Mason94b60442010-05-26 11:02:00 -04005959 }
Chris Mason39279cc2007-06-12 06:35:45 -04005960 btrfs_end_transaction(trans, root);
Miao Xie16cdcec2011-04-22 18:12:22 +08005961 if (BTRFS_I(inode)->delayed_node)
5962 btrfs_balance_delayed_items(root);
Josef Bacik22c44fe2011-11-30 10:45:38 -05005963
5964 return ret;
5965}
5966
5967/*
5968 * This is a copy of file_update_time. We need this so we can return error on
5969 * ENOSPC for updating the inode in the case of file write and mmap writes.
5970 */
Josef Bacike41f9412012-03-26 09:46:47 -04005971static int btrfs_update_time(struct inode *inode, struct timespec *now,
5972 int flags)
Josef Bacik22c44fe2011-11-30 10:45:38 -05005973{
Alexander Block2bc5565282012-06-15 09:49:33 +02005974 struct btrfs_root *root = BTRFS_I(inode)->root;
5975
5976 if (btrfs_root_readonly(root))
5977 return -EROFS;
5978
Josef Bacike41f9412012-03-26 09:46:47 -04005979 if (flags & S_VERSION)
Josef Bacik22c44fe2011-11-30 10:45:38 -05005980 inode_inc_iversion(inode);
Josef Bacike41f9412012-03-26 09:46:47 -04005981 if (flags & S_CTIME)
5982 inode->i_ctime = *now;
5983 if (flags & S_MTIME)
5984 inode->i_mtime = *now;
5985 if (flags & S_ATIME)
5986 inode->i_atime = *now;
5987 return btrfs_dirty_inode(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04005988}
5989
Chris Masond352ac62008-09-29 15:18:18 -04005990/*
5991 * find the highest existing sequence number in a directory
5992 * and then set the in-memory index_cnt variable to reflect
5993 * free sequence numbers
5994 */
Josef Bacikaec74772008-07-24 12:12:38 -04005995static int btrfs_set_inode_index_count(struct inode *inode)
5996{
5997 struct btrfs_root *root = BTRFS_I(inode)->root;
5998 struct btrfs_key key, found_key;
5999 struct btrfs_path *path;
6000 struct extent_buffer *leaf;
6001 int ret;
6002
Li Zefan33345d012011-04-20 10:31:50 +08006003 key.objectid = btrfs_ino(inode);
David Sterba962a2982014-06-04 18:41:45 +02006004 key.type = BTRFS_DIR_INDEX_KEY;
Josef Bacikaec74772008-07-24 12:12:38 -04006005 key.offset = (u64)-1;
6006
6007 path = btrfs_alloc_path();
6008 if (!path)
6009 return -ENOMEM;
6010
6011 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6012 if (ret < 0)
6013 goto out;
6014 /* FIXME: we should be able to handle this */
6015 if (ret == 0)
6016 goto out;
6017 ret = 0;
6018
6019 /*
6020 * MAGIC NUMBER EXPLANATION:
6021 * since we search a directory based on f_pos we have to start at 2
6022 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6023 * else has to start at 2
6024 */
6025 if (path->slots[0] == 0) {
6026 BTRFS_I(inode)->index_cnt = 2;
6027 goto out;
6028 }
6029
6030 path->slots[0]--;
6031
6032 leaf = path->nodes[0];
6033 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6034
Li Zefan33345d012011-04-20 10:31:50 +08006035 if (found_key.objectid != btrfs_ino(inode) ||
David Sterba962a2982014-06-04 18:41:45 +02006036 found_key.type != BTRFS_DIR_INDEX_KEY) {
Josef Bacikaec74772008-07-24 12:12:38 -04006037 BTRFS_I(inode)->index_cnt = 2;
6038 goto out;
6039 }
6040
6041 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
6042out:
6043 btrfs_free_path(path);
6044 return ret;
6045}
6046
Chris Masond352ac62008-09-29 15:18:18 -04006047/*
6048 * helper to find a free sequence number in a given directory. This current
6049 * code is very simple, later versions will do smarter things in the btree
6050 */
Chris Mason3de45862008-11-17 21:02:50 -05006051int btrfs_set_inode_index(struct inode *dir, u64 *index)
Josef Bacikaec74772008-07-24 12:12:38 -04006052{
6053 int ret = 0;
6054
6055 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
Miao Xie16cdcec2011-04-22 18:12:22 +08006056 ret = btrfs_inode_delayed_dir_index_count(dir);
6057 if (ret) {
6058 ret = btrfs_set_inode_index_count(dir);
6059 if (ret)
6060 return ret;
6061 }
Josef Bacikaec74772008-07-24 12:12:38 -04006062 }
6063
Chris Mason00e4e6b2008-08-05 11:18:09 -04006064 *index = BTRFS_I(dir)->index_cnt;
Josef Bacikaec74772008-07-24 12:12:38 -04006065 BTRFS_I(dir)->index_cnt++;
6066
6067 return ret;
6068}
6069
Chris Masonb0d5d102014-09-08 13:08:51 -07006070static int btrfs_insert_inode_locked(struct inode *inode)
6071{
6072 struct btrfs_iget_args args;
6073 args.location = &BTRFS_I(inode)->location;
6074 args.root = BTRFS_I(inode)->root;
6075
6076 return insert_inode_locked4(inode,
6077 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6078 btrfs_find_actor, &args);
6079}
6080
Chris Mason39279cc2007-06-12 06:35:45 -04006081static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6082 struct btrfs_root *root,
Josef Bacikaec74772008-07-24 12:12:38 -04006083 struct inode *dir,
Chris Mason9c583092008-01-29 15:15:18 -05006084 const char *name, int name_len,
Al Viro175a4eb2011-07-26 03:30:54 -04006085 u64 ref_objectid, u64 objectid,
6086 umode_t mode, u64 *index)
Chris Mason39279cc2007-06-12 06:35:45 -04006087{
6088 struct inode *inode;
Chris Mason5f39d392007-10-15 16:14:19 -04006089 struct btrfs_inode_item *inode_item;
Chris Mason39279cc2007-06-12 06:35:45 -04006090 struct btrfs_key *location;
Chris Mason5f39d392007-10-15 16:14:19 -04006091 struct btrfs_path *path;
Chris Mason9c583092008-01-29 15:15:18 -05006092 struct btrfs_inode_ref *ref;
6093 struct btrfs_key key[2];
6094 u32 sizes[2];
Filipe Mananaef3b9af2014-04-27 20:40:45 +01006095 int nitems = name ? 2 : 1;
Chris Mason9c583092008-01-29 15:15:18 -05006096 unsigned long ptr;
Chris Mason39279cc2007-06-12 06:35:45 -04006097 int ret;
Chris Mason39279cc2007-06-12 06:35:45 -04006098
Chris Mason5f39d392007-10-15 16:14:19 -04006099 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07006100 if (!path)
6101 return ERR_PTR(-ENOMEM);
Chris Mason5f39d392007-10-15 16:14:19 -04006102
Chris Mason39279cc2007-06-12 06:35:45 -04006103 inode = new_inode(root->fs_info->sb);
Yoshinori Sano8fb27642011-04-09 02:30:07 +00006104 if (!inode) {
6105 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04006106 return ERR_PTR(-ENOMEM);
Yoshinori Sano8fb27642011-04-09 02:30:07 +00006107 }
Chris Mason39279cc2007-06-12 06:35:45 -04006108
Li Zefan581bb052011-04-20 10:06:11 +08006109 /*
Filipe Manana5762b5c2014-08-01 00:10:32 +01006110 * O_TMPFILE, set link count to 0, so that after this point,
6111 * we fill in an inode item with the correct link count.
6112 */
6113 if (!name)
6114 set_nlink(inode, 0);
6115
6116 /*
Li Zefan581bb052011-04-20 10:06:11 +08006117 * we have to initialize this early, so we can reclaim the inode
6118 * number if we fail afterwards in this function.
6119 */
6120 inode->i_ino = objectid;
6121
Filipe Mananaef3b9af2014-04-27 20:40:45 +01006122 if (dir && name) {
liubo1abe9b82011-03-24 11:18:59 +00006123 trace_btrfs_inode_request(dir);
6124
Chris Mason3de45862008-11-17 21:02:50 -05006125 ret = btrfs_set_inode_index(dir, index);
Shen Feng09771432009-04-02 16:46:06 -04006126 if (ret) {
Yoshinori Sano8fb27642011-04-09 02:30:07 +00006127 btrfs_free_path(path);
Shen Feng09771432009-04-02 16:46:06 -04006128 iput(inode);
Josef Bacikaec74772008-07-24 12:12:38 -04006129 return ERR_PTR(ret);
Shen Feng09771432009-04-02 16:46:06 -04006130 }
Filipe Mananaef3b9af2014-04-27 20:40:45 +01006131 } else if (dir) {
6132 *index = 0;
Josef Bacikaec74772008-07-24 12:12:38 -04006133 }
6134 /*
6135 * index_cnt is ignored for everything but a dir,
6136 * btrfs_get_inode_index_count has an explanation for the magic
6137 * number
6138 */
6139 BTRFS_I(inode)->index_cnt = 2;
Miao Xie67de1172013-12-26 13:07:06 +08006140 BTRFS_I(inode)->dir_index = *index;
Chris Mason39279cc2007-06-12 06:35:45 -04006141 BTRFS_I(inode)->root = root;
Chris Masone02119d2008-09-05 16:13:11 -04006142 BTRFS_I(inode)->generation = trans->transid;
Josef Bacik76195852010-11-19 02:18:02 +00006143 inode->i_generation = BTRFS_I(inode)->generation;
Chris Masonb888db22007-08-27 16:49:44 -04006144
Josef Bacik5dc562c2012-08-17 13:14:17 -04006145 /*
6146 * We could have gotten an inode number from somebody who was fsynced
6147 * and then removed in this same transaction, so let's just set full
6148 * sync since it will be a full sync anyway and this will blow away the
6149 * old info in the log.
6150 */
6151 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6152
Chris Mason9c583092008-01-29 15:15:18 -05006153 key[0].objectid = objectid;
David Sterba962a2982014-06-04 18:41:45 +02006154 key[0].type = BTRFS_INODE_ITEM_KEY;
Chris Mason9c583092008-01-29 15:15:18 -05006155 key[0].offset = 0;
6156
Chris Mason9c583092008-01-29 15:15:18 -05006157 sizes[0] = sizeof(struct btrfs_inode_item);
Filipe Mananaef3b9af2014-04-27 20:40:45 +01006158
6159 if (name) {
6160 /*
6161 * Start new inodes with an inode_ref. This is slightly more
6162 * efficient for small numbers of hard links since they will
6163 * be packed into one item. Extended refs will kick in if we
6164 * add more hard links than can fit in the ref item.
6165 */
6166 key[1].objectid = objectid;
David Sterba962a2982014-06-04 18:41:45 +02006167 key[1].type = BTRFS_INODE_REF_KEY;
Filipe Mananaef3b9af2014-04-27 20:40:45 +01006168 key[1].offset = ref_objectid;
6169
6170 sizes[1] = name_len + sizeof(*ref);
6171 }
Chris Mason9c583092008-01-29 15:15:18 -05006172
Chris Masonb0d5d102014-09-08 13:08:51 -07006173 location = &BTRFS_I(inode)->location;
6174 location->objectid = objectid;
6175 location->offset = 0;
David Sterba962a2982014-06-04 18:41:45 +02006176 location->type = BTRFS_INODE_ITEM_KEY;
Chris Masonb0d5d102014-09-08 13:08:51 -07006177
6178 ret = btrfs_insert_inode_locked(inode);
6179 if (ret < 0)
6180 goto fail;
6181
Chris Masonb9473432009-03-13 11:00:37 -04006182 path->leave_spinning = 1;
Filipe Mananaef3b9af2014-04-27 20:40:45 +01006183 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
Chris Mason9c583092008-01-29 15:15:18 -05006184 if (ret != 0)
Chris Masonb0d5d102014-09-08 13:08:51 -07006185 goto fail_unlock;
Chris Mason5f39d392007-10-15 16:14:19 -04006186
Dmitry Monakhovecc11fab2010-03-04 17:31:47 +03006187 inode_init_owner(inode, dir, mode);
Yan Zhenga76a3cd2008-10-09 11:46:29 -04006188 inode_set_bytes(inode, 0);
chandan r9cc97d62012-07-04 12:48:07 +05306189
Deepa Dinamani04b285f2016-02-06 23:57:21 -08006190 inode->i_mtime = current_fs_time(inode->i_sb);
chandan r9cc97d62012-07-04 12:48:07 +05306191 inode->i_atime = inode->i_mtime;
6192 inode->i_ctime = inode->i_mtime;
6193 BTRFS_I(inode)->i_otime = inode->i_mtime;
6194
Chris Mason5f39d392007-10-15 16:14:19 -04006195 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6196 struct btrfs_inode_item);
Li Zefan293f7e02012-07-10 00:58:58 -06006197 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6198 sizeof(*inode_item));
Chris Masone02119d2008-09-05 16:13:11 -04006199 fill_inode_item(trans, path->nodes[0], inode_item, inode);
Chris Mason9c583092008-01-29 15:15:18 -05006200
Filipe Mananaef3b9af2014-04-27 20:40:45 +01006201 if (name) {
6202 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6203 struct btrfs_inode_ref);
6204 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6205 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6206 ptr = (unsigned long)(ref + 1);
6207 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6208 }
Chris Mason9c583092008-01-29 15:15:18 -05006209
Chris Mason5f39d392007-10-15 16:14:19 -04006210 btrfs_mark_buffer_dirty(path->nodes[0]);
6211 btrfs_free_path(path);
6212
Christoph Hellwig6cbff002009-04-17 10:37:41 +02006213 btrfs_inherit_iflags(inode, dir);
6214
Al Viro569254b2011-07-24 17:08:40 -04006215 if (S_ISREG(mode)) {
Chris Mason94272162009-07-02 12:26:06 -04006216 if (btrfs_test_opt(root, NODATASUM))
6217 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
Liu Bo213490b2012-09-11 08:33:50 -06006218 if (btrfs_test_opt(root, NODATACOW))
Josef Bacikf2bdf9a2013-02-21 15:28:28 -05006219 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6220 BTRFS_INODE_NODATASUM;
Chris Mason94272162009-07-02 12:26:06 -04006221 }
6222
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006223 inode_tree_add(inode);
liubo1abe9b82011-03-24 11:18:59 +00006224
6225 trace_btrfs_inode_new(inode);
Chris Mason1973f0f2011-06-24 13:13:29 -04006226 btrfs_set_inode_last_trans(trans, inode);
liubo1abe9b82011-03-24 11:18:59 +00006227
Alexander Block8ea05e32012-07-25 17:35:53 +02006228 btrfs_update_root_times(trans, root);
6229
Filipe David Borba Manana63541922014-01-07 11:47:46 +00006230 ret = btrfs_inode_inherit_props(trans, inode, dir);
6231 if (ret)
6232 btrfs_err(root->fs_info,
6233 "error inheriting props for ino %llu (root %llu): %d",
6234 btrfs_ino(inode), root->root_key.objectid, ret);
6235
Chris Mason39279cc2007-06-12 06:35:45 -04006236 return inode;
Chris Masonb0d5d102014-09-08 13:08:51 -07006237
6238fail_unlock:
6239 unlock_new_inode(inode);
Chris Mason5f39d392007-10-15 16:14:19 -04006240fail:
Filipe Mananaef3b9af2014-04-27 20:40:45 +01006241 if (dir && name)
Josef Bacikaec74772008-07-24 12:12:38 -04006242 BTRFS_I(dir)->index_cnt--;
Chris Mason5f39d392007-10-15 16:14:19 -04006243 btrfs_free_path(path);
Shen Feng09771432009-04-02 16:46:06 -04006244 iput(inode);
Chris Mason5f39d392007-10-15 16:14:19 -04006245 return ERR_PTR(ret);
Chris Mason39279cc2007-06-12 06:35:45 -04006246}
6247
6248static inline u8 btrfs_inode_type(struct inode *inode)
6249{
6250 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6251}
6252
Chris Masond352ac62008-09-29 15:18:18 -04006253/*
6254 * utility function to add 'inode' into 'parent_inode' with
6255 * a give name and a given sequence number.
6256 * if 'add_backref' is true, also insert a backref from the
6257 * inode to the parent directory.
6258 */
Chris Masone02119d2008-09-05 16:13:11 -04006259int btrfs_add_link(struct btrfs_trans_handle *trans,
6260 struct inode *parent_inode, struct inode *inode,
6261 const char *name, int name_len, int add_backref, u64 index)
Chris Mason39279cc2007-06-12 06:35:45 -04006262{
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006263 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04006264 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -04006265 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
Li Zefan33345d012011-04-20 10:31:50 +08006266 u64 ino = btrfs_ino(inode);
6267 u64 parent_ino = btrfs_ino(parent_inode);
Chris Mason5f39d392007-10-15 16:14:19 -04006268
Li Zefan33345d012011-04-20 10:31:50 +08006269 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006270 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6271 } else {
Li Zefan33345d012011-04-20 10:31:50 +08006272 key.objectid = ino;
David Sterba962a2982014-06-04 18:41:45 +02006273 key.type = BTRFS_INODE_ITEM_KEY;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006274 key.offset = 0;
6275 }
Chris Mason39279cc2007-06-12 06:35:45 -04006276
Li Zefan33345d012011-04-20 10:31:50 +08006277 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006278 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6279 key.objectid, root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +08006280 parent_ino, index, name, name_len);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006281 } else if (add_backref) {
Li Zefan33345d012011-04-20 10:31:50 +08006282 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6283 parent_ino, index);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006284 }
6285
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006286 /* Nothing to clean up yet */
6287 if (ret)
6288 return ret;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04006289
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006290 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6291 parent_inode, &key,
6292 btrfs_inode_type(inode), index);
Chris Mason9c520572012-12-17 14:26:57 -05006293 if (ret == -EEXIST || ret == -EOVERFLOW)
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006294 goto fail_dir_item;
6295 else if (ret) {
6296 btrfs_abort_transaction(trans, root, ret);
6297 return ret;
Chris Mason39279cc2007-06-12 06:35:45 -04006298 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006299
6300 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6301 name_len * 2);
Josef Bacik0c4d2d92012-04-05 15:03:02 -04006302 inode_inc_iversion(parent_inode);
Deepa Dinamani04b285f2016-02-06 23:57:21 -08006303 parent_inode->i_mtime = parent_inode->i_ctime =
6304 current_fs_time(parent_inode->i_sb);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006305 ret = btrfs_update_inode(trans, root, parent_inode);
6306 if (ret)
6307 btrfs_abort_transaction(trans, root, ret);
Chris Mason39279cc2007-06-12 06:35:45 -04006308 return ret;
Chris Masonfe66a052012-02-20 08:40:56 -05006309
6310fail_dir_item:
6311 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6312 u64 local_index;
6313 int err;
6314 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6315 key.objectid, root->root_key.objectid,
6316 parent_ino, &local_index, name, name_len);
6317
6318 } else if (add_backref) {
6319 u64 local_index;
6320 int err;
6321
6322 err = btrfs_del_inode_ref(trans, root, name, name_len,
6323 ino, parent_ino, &local_index);
6324 }
6325 return ret;
Chris Mason39279cc2007-06-12 06:35:45 -04006326}
6327
6328static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
Josef Bacika1b075d2010-11-19 20:36:11 +00006329 struct inode *dir, struct dentry *dentry,
6330 struct inode *inode, int backref, u64 index)
Chris Mason39279cc2007-06-12 06:35:45 -04006331{
Josef Bacika1b075d2010-11-19 20:36:11 +00006332 int err = btrfs_add_link(trans, dir, inode,
6333 dentry->d_name.name, dentry->d_name.len,
6334 backref, index);
Chris Mason39279cc2007-06-12 06:35:45 -04006335 if (err > 0)
6336 err = -EEXIST;
6337 return err;
6338}
6339
Josef Bacik618e21d2007-07-11 10:18:17 -04006340static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
Al Viro1a67aaf2011-07-26 01:52:52 -04006341 umode_t mode, dev_t rdev)
Josef Bacik618e21d2007-07-11 10:18:17 -04006342{
6343 struct btrfs_trans_handle *trans;
6344 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason1832a6d2007-12-21 16:27:21 -05006345 struct inode *inode = NULL;
Josef Bacik618e21d2007-07-11 10:18:17 -04006346 int err;
6347 int drop_inode = 0;
6348 u64 objectid;
Chris Mason00e4e6b2008-08-05 11:18:09 -04006349 u64 index = 0;
Josef Bacik618e21d2007-07-11 10:18:17 -04006350
Josef Bacik9ed74f22009-09-11 16:12:44 -04006351 /*
6352 * 2 for inode item and ref
6353 * 2 for dir items
6354 * 1 for xattr if selinux is on
6355 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04006356 trans = btrfs_start_transaction(root, 5);
6357 if (IS_ERR(trans))
6358 return PTR_ERR(trans);
Chris Mason1832a6d2007-12-21 16:27:21 -05006359
Li Zefan581bb052011-04-20 10:06:11 +08006360 err = btrfs_find_free_ino(root, &objectid);
6361 if (err)
6362 goto out_unlock;
6363
Josef Bacikaec74772008-07-24 12:12:38 -04006364 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Li Zefan33345d012011-04-20 10:31:50 +08006365 dentry->d_name.len, btrfs_ino(dir), objectid,
Josef Bacikd82a6f12011-05-11 15:26:06 -04006366 mode, &index);
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04006367 if (IS_ERR(inode)) {
6368 err = PTR_ERR(inode);
Josef Bacik618e21d2007-07-11 10:18:17 -04006369 goto out_unlock;
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04006370 }
Josef Bacik618e21d2007-07-11 10:18:17 -04006371
Casey Schauflerad19db72011-12-15 10:09:07 -05006372 /*
6373 * If the active LSM wants to access the inode during
6374 * d_instantiate it needs these. Smack checks to see
6375 * if the filesystem supports xattrs by looking at the
6376 * ops vector.
6377 */
Casey Schauflerad19db72011-12-15 10:09:07 -05006378 inode->i_op = &btrfs_special_inode_operations;
Chris Masonb0d5d102014-09-08 13:08:51 -07006379 init_special_inode(inode, inode->i_mode, rdev);
6380
6381 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
Josef Bacik618e21d2007-07-11 10:18:17 -04006382 if (err)
Chris Masonb0d5d102014-09-08 13:08:51 -07006383 goto out_unlock_inode;
6384
6385 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6386 if (err) {
6387 goto out_unlock_inode;
6388 } else {
Yan1b4ab1b2007-08-29 09:11:44 -04006389 btrfs_update_inode(trans, root, inode);
Chris Masonb0d5d102014-09-08 13:08:51 -07006390 unlock_new_inode(inode);
Al Viro08c422c2011-12-23 07:58:13 -05006391 d_instantiate(dentry, inode);
Josef Bacik618e21d2007-07-11 10:18:17 -04006392 }
Chris Masonb0d5d102014-09-08 13:08:51 -07006393
Josef Bacik618e21d2007-07-11 10:18:17 -04006394out_unlock:
Josef Bacik7ad85bb2012-01-12 19:10:12 -05006395 btrfs_end_transaction(trans, root);
Josef Bacikc581afc2014-02-06 16:06:06 -05006396 btrfs_balance_delayed_items(root);
Liu Bob53d3f52012-11-14 14:34:34 +00006397 btrfs_btree_balance_dirty(root);
Josef Bacik618e21d2007-07-11 10:18:17 -04006398 if (drop_inode) {
6399 inode_dec_link_count(inode);
6400 iput(inode);
6401 }
Josef Bacik618e21d2007-07-11 10:18:17 -04006402 return err;
Chris Masonb0d5d102014-09-08 13:08:51 -07006403
6404out_unlock_inode:
6405 drop_inode = 1;
6406 unlock_new_inode(inode);
6407 goto out_unlock;
6408
Josef Bacik618e21d2007-07-11 10:18:17 -04006409}
6410
Chris Mason39279cc2007-06-12 06:35:45 -04006411static int btrfs_create(struct inode *dir, struct dentry *dentry,
Al Viroebfc3b42012-06-10 18:05:36 -04006412 umode_t mode, bool excl)
Chris Mason39279cc2007-06-12 06:35:45 -04006413{
6414 struct btrfs_trans_handle *trans;
6415 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason1832a6d2007-12-21 16:27:21 -05006416 struct inode *inode = NULL;
Filipe Brandenburger43baa572012-11-30 03:40:09 +00006417 int drop_inode_on_err = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -04006418 int err;
Chris Mason39279cc2007-06-12 06:35:45 -04006419 u64 objectid;
Chris Mason00e4e6b2008-08-05 11:18:09 -04006420 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04006421
Josef Bacik9ed74f22009-09-11 16:12:44 -04006422 /*
6423 * 2 for inode item and ref
6424 * 2 for dir items
6425 * 1 for xattr if selinux is on
6426 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04006427 trans = btrfs_start_transaction(root, 5);
6428 if (IS_ERR(trans))
6429 return PTR_ERR(trans);
Josef Bacik9ed74f22009-09-11 16:12:44 -04006430
Li Zefan581bb052011-04-20 10:06:11 +08006431 err = btrfs_find_free_ino(root, &objectid);
6432 if (err)
6433 goto out_unlock;
6434
Josef Bacikaec74772008-07-24 12:12:38 -04006435 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Li Zefan33345d012011-04-20 10:31:50 +08006436 dentry->d_name.len, btrfs_ino(dir), objectid,
Josef Bacikd82a6f12011-05-11 15:26:06 -04006437 mode, &index);
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04006438 if (IS_ERR(inode)) {
6439 err = PTR_ERR(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04006440 goto out_unlock;
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04006441 }
Filipe Brandenburger43baa572012-11-30 03:40:09 +00006442 drop_inode_on_err = 1;
Casey Schauflerad19db72011-12-15 10:09:07 -05006443 /*
6444 * If the active LSM wants to access the inode during
6445 * d_instantiate it needs these. Smack checks to see
6446 * if the filesystem supports xattrs by looking at the
6447 * ops vector.
6448 */
6449 inode->i_fop = &btrfs_file_operations;
6450 inode->i_op = &btrfs_file_inode_operations;
Chris Masonb0d5d102014-09-08 13:08:51 -07006451 inode->i_mapping->a_ops = &btrfs_aops;
Chris Masonb0d5d102014-09-08 13:08:51 -07006452
6453 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6454 if (err)
6455 goto out_unlock_inode;
6456
6457 err = btrfs_update_inode(trans, root, inode);
6458 if (err)
6459 goto out_unlock_inode;
Casey Schauflerad19db72011-12-15 10:09:07 -05006460
Josef Bacika1b075d2010-11-19 20:36:11 +00006461 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04006462 if (err)
Chris Masonb0d5d102014-09-08 13:08:51 -07006463 goto out_unlock_inode;
Filipe Brandenburger43baa572012-11-30 03:40:09 +00006464
Filipe Brandenburger43baa572012-11-30 03:40:09 +00006465 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
Chris Masonb0d5d102014-09-08 13:08:51 -07006466 unlock_new_inode(inode);
Filipe Brandenburger43baa572012-11-30 03:40:09 +00006467 d_instantiate(dentry, inode);
6468
Chris Mason39279cc2007-06-12 06:35:45 -04006469out_unlock:
Josef Bacik7ad85bb2012-01-12 19:10:12 -05006470 btrfs_end_transaction(trans, root);
Filipe Brandenburger43baa572012-11-30 03:40:09 +00006471 if (err && drop_inode_on_err) {
Chris Mason39279cc2007-06-12 06:35:45 -04006472 inode_dec_link_count(inode);
6473 iput(inode);
6474 }
Josef Bacikc581afc2014-02-06 16:06:06 -05006475 btrfs_balance_delayed_items(root);
Liu Bob53d3f52012-11-14 14:34:34 +00006476 btrfs_btree_balance_dirty(root);
Chris Mason39279cc2007-06-12 06:35:45 -04006477 return err;
Chris Masonb0d5d102014-09-08 13:08:51 -07006478
6479out_unlock_inode:
6480 unlock_new_inode(inode);
6481 goto out_unlock;
6482
Chris Mason39279cc2007-06-12 06:35:45 -04006483}
6484
6485static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6486 struct dentry *dentry)
6487{
Filipe Manana271dba42016-01-05 16:24:05 +00006488 struct btrfs_trans_handle *trans = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04006489 struct btrfs_root *root = BTRFS_I(dir)->root;
David Howells2b0143b2015-03-17 22:25:59 +00006490 struct inode *inode = d_inode(old_dentry);
Chris Mason00e4e6b2008-08-05 11:18:09 -04006491 u64 index;
Chris Mason39279cc2007-06-12 06:35:45 -04006492 int err;
6493 int drop_inode = 0;
6494
TARUISI Hiroaki4a8be422009-11-12 07:14:26 +00006495 /* do not allow sys_link's with other subvols of the same device */
6496 if (root->objectid != BTRFS_I(inode)->root->objectid)
Mark Fasheh3ab35642011-03-22 17:20:26 +00006497 return -EXDEV;
TARUISI Hiroaki4a8be422009-11-12 07:14:26 +00006498
Mark Fashehf1863732012-08-08 11:32:27 -07006499 if (inode->i_nlink >= BTRFS_LINK_MAX)
Al Viroc055e992011-03-04 17:15:18 +00006500 return -EMLINK;
Josef Bacik9ed74f22009-09-11 16:12:44 -04006501
Chris Mason3de45862008-11-17 21:02:50 -05006502 err = btrfs_set_inode_index(dir, &index);
Josef Bacikaec74772008-07-24 12:12:38 -04006503 if (err)
6504 goto fail;
6505
Yan, Zhenga22285a2010-05-16 10:48:46 -04006506 /*
Miao Xie7e6b6462011-02-18 09:21:17 +00006507 * 2 items for inode and inode ref
Yan, Zhenga22285a2010-05-16 10:48:46 -04006508 * 2 items for dir items
Miao Xie7e6b6462011-02-18 09:21:17 +00006509 * 1 item for parent inode
Yan, Zhenga22285a2010-05-16 10:48:46 -04006510 */
Miao Xie7e6b6462011-02-18 09:21:17 +00006511 trans = btrfs_start_transaction(root, 5);
Yan, Zhenga22285a2010-05-16 10:48:46 -04006512 if (IS_ERR(trans)) {
6513 err = PTR_ERR(trans);
Filipe Manana271dba42016-01-05 16:24:05 +00006514 trans = NULL;
Yan, Zhenga22285a2010-05-16 10:48:46 -04006515 goto fail;
6516 }
Chris Mason5f39d392007-10-15 16:14:19 -04006517
Miao Xie67de1172013-12-26 13:07:06 +08006518 /* There are several dir indexes for this inode, clear the cache. */
6519 BTRFS_I(inode)->dir_index = 0ULL;
Zach Brown8b558c52013-10-16 12:10:34 -07006520 inc_nlink(inode);
Josef Bacik0c4d2d92012-04-05 15:03:02 -04006521 inode_inc_iversion(inode);
Deepa Dinamani04b285f2016-02-06 23:57:21 -08006522 inode->i_ctime = current_fs_time(inode->i_sb);
Al Viro7de9c6ee2010-10-23 11:11:40 -04006523 ihold(inode);
Josef Bacike9976152012-10-11 15:53:56 -04006524 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
Josef Bacikaec74772008-07-24 12:12:38 -04006525
Josef Bacika1b075d2010-11-19 20:36:11 +00006526 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
Chris Mason5f39d392007-10-15 16:14:19 -04006527
Yan, Zhenga5719522009-09-24 09:17:31 -04006528 if (err) {
Chris Mason39279cc2007-06-12 06:35:45 -04006529 drop_inode = 1;
Yan, Zhenga5719522009-09-24 09:17:31 -04006530 } else {
Al Viro10d9f302011-07-16 23:09:10 -04006531 struct dentry *parent = dentry->d_parent;
Yan, Zhenga5719522009-09-24 09:17:31 -04006532 err = btrfs_update_inode(trans, root, inode);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006533 if (err)
6534 goto fail;
Filipe Mananaef3b9af2014-04-27 20:40:45 +01006535 if (inode->i_nlink == 1) {
6536 /*
6537 * If new hard link count is 1, it's a file created
6538 * with open(2) O_TMPFILE flag.
6539 */
6540 err = btrfs_orphan_del(trans, inode);
6541 if (err)
6542 goto fail;
6543 }
Al Viro08c422c2011-12-23 07:58:13 -05006544 d_instantiate(dentry, inode);
Josef Bacik6a912212010-11-20 09:48:00 +00006545 btrfs_log_new_name(trans, inode, NULL, parent);
Yan, Zhenga5719522009-09-24 09:17:31 -04006546 }
Chris Mason39279cc2007-06-12 06:35:45 -04006547
Josef Bacikc581afc2014-02-06 16:06:06 -05006548 btrfs_balance_delayed_items(root);
Chris Mason1832a6d2007-12-21 16:27:21 -05006549fail:
Filipe Manana271dba42016-01-05 16:24:05 +00006550 if (trans)
6551 btrfs_end_transaction(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04006552 if (drop_inode) {
6553 inode_dec_link_count(inode);
6554 iput(inode);
6555 }
Liu Bob53d3f52012-11-14 14:34:34 +00006556 btrfs_btree_balance_dirty(root);
Chris Mason39279cc2007-06-12 06:35:45 -04006557 return err;
6558}
6559
Al Viro18bb1db2011-07-26 01:41:39 -04006560static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Chris Mason39279cc2007-06-12 06:35:45 -04006561{
Chris Masonb9d86662008-05-02 16:13:49 -04006562 struct inode *inode = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04006563 struct btrfs_trans_handle *trans;
6564 struct btrfs_root *root = BTRFS_I(dir)->root;
6565 int err = 0;
6566 int drop_on_err = 0;
Chris Masonb9d86662008-05-02 16:13:49 -04006567 u64 objectid = 0;
Chris Mason00e4e6b2008-08-05 11:18:09 -04006568 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04006569
Josef Bacik9ed74f22009-09-11 16:12:44 -04006570 /*
6571 * 2 items for inode and ref
6572 * 2 items for dir items
6573 * 1 for xattr if selinux is on
6574 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04006575 trans = btrfs_start_transaction(root, 5);
6576 if (IS_ERR(trans))
6577 return PTR_ERR(trans);
Chris Mason39279cc2007-06-12 06:35:45 -04006578
Li Zefan581bb052011-04-20 10:06:11 +08006579 err = btrfs_find_free_ino(root, &objectid);
6580 if (err)
6581 goto out_fail;
6582
Josef Bacikaec74772008-07-24 12:12:38 -04006583 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Li Zefan33345d012011-04-20 10:31:50 +08006584 dentry->d_name.len, btrfs_ino(dir), objectid,
Josef Bacikd82a6f12011-05-11 15:26:06 -04006585 S_IFDIR | mode, &index);
Chris Mason39279cc2007-06-12 06:35:45 -04006586 if (IS_ERR(inode)) {
6587 err = PTR_ERR(inode);
6588 goto out_fail;
6589 }
Chris Mason5f39d392007-10-15 16:14:19 -04006590
Chris Mason39279cc2007-06-12 06:35:45 -04006591 drop_on_err = 1;
Chris Masonb0d5d102014-09-08 13:08:51 -07006592 /* these must be set before we unlock the inode */
6593 inode->i_op = &btrfs_dir_inode_operations;
6594 inode->i_fop = &btrfs_dir_file_operations;
Josef Bacik33268ea2008-07-24 12:16:36 -04006595
Eric Paris2a7dba32011-02-01 11:05:39 -05006596 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
Josef Bacik33268ea2008-07-24 12:16:36 -04006597 if (err)
Chris Masonb0d5d102014-09-08 13:08:51 -07006598 goto out_fail_inode;
Chris Mason39279cc2007-06-12 06:35:45 -04006599
Chris Masondbe674a2008-07-17 12:54:05 -04006600 btrfs_i_size_write(inode, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04006601 err = btrfs_update_inode(trans, root, inode);
6602 if (err)
Chris Masonb0d5d102014-09-08 13:08:51 -07006603 goto out_fail_inode;
Chris Mason5f39d392007-10-15 16:14:19 -04006604
Josef Bacika1b075d2010-11-19 20:36:11 +00006605 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6606 dentry->d_name.len, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04006607 if (err)
Chris Masonb0d5d102014-09-08 13:08:51 -07006608 goto out_fail_inode;
Chris Mason5f39d392007-10-15 16:14:19 -04006609
Chris Mason39279cc2007-06-12 06:35:45 -04006610 d_instantiate(dentry, inode);
Chris Masonb0d5d102014-09-08 13:08:51 -07006611 /*
6612 * mkdir is special. We're unlocking after we call d_instantiate
6613 * to avoid a race with nfsd calling d_instantiate.
6614 */
6615 unlock_new_inode(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04006616 drop_on_err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04006617
6618out_fail:
Josef Bacik7ad85bb2012-01-12 19:10:12 -05006619 btrfs_end_transaction(trans, root);
Wang Shilongc7cfb8a2014-12-24 14:45:30 +08006620 if (drop_on_err) {
6621 inode_dec_link_count(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04006622 iput(inode);
Wang Shilongc7cfb8a2014-12-24 14:45:30 +08006623 }
Josef Bacikc581afc2014-02-06 16:06:06 -05006624 btrfs_balance_delayed_items(root);
Liu Bob53d3f52012-11-14 14:34:34 +00006625 btrfs_btree_balance_dirty(root);
Chris Mason39279cc2007-06-12 06:35:45 -04006626 return err;
Chris Masonb0d5d102014-09-08 13:08:51 -07006627
6628out_fail_inode:
6629 unlock_new_inode(inode);
6630 goto out_fail;
Chris Mason39279cc2007-06-12 06:35:45 -04006631}
6632
Qu Wenruoe6c4efd2014-09-17 11:53:35 +08006633/* Find next extent map of a given extent map, caller needs to ensure locks */
6634static struct extent_map *next_extent_map(struct extent_map *em)
6635{
6636 struct rb_node *next;
6637
6638 next = rb_next(&em->rb_node);
6639 if (!next)
6640 return NULL;
6641 return container_of(next, struct extent_map, rb_node);
6642}
6643
6644static struct extent_map *prev_extent_map(struct extent_map *em)
6645{
6646 struct rb_node *prev;
6647
6648 prev = rb_prev(&em->rb_node);
6649 if (!prev)
6650 return NULL;
6651 return container_of(prev, struct extent_map, rb_node);
6652}
6653
Chris Masond352ac62008-09-29 15:18:18 -04006654/* helper for btfs_get_extent. Given an existing extent in the tree,
Qu Wenruoe6c4efd2014-09-17 11:53:35 +08006655 * the existing extent is the nearest extent to map_start,
Chris Masond352ac62008-09-29 15:18:18 -04006656 * and an extent that you want to insert, deal with overlap and insert
Qu Wenruoe6c4efd2014-09-17 11:53:35 +08006657 * the best fitted new extent into the tree.
Chris Masond352ac62008-09-29 15:18:18 -04006658 */
Chris Mason3b951512008-04-17 11:29:12 -04006659static int merge_extent_mapping(struct extent_map_tree *em_tree,
6660 struct extent_map *existing,
Chris Masone6dcd2d2008-07-17 12:53:50 -04006661 struct extent_map *em,
Qu Wenruo51f395a2014-08-08 13:06:20 +08006662 u64 map_start)
Chris Mason3b951512008-04-17 11:29:12 -04006663{
Qu Wenruoe6c4efd2014-09-17 11:53:35 +08006664 struct extent_map *prev;
6665 struct extent_map *next;
6666 u64 start;
6667 u64 end;
Chris Mason3b951512008-04-17 11:29:12 -04006668 u64 start_diff;
Chris Mason3b951512008-04-17 11:29:12 -04006669
Chris Masone6dcd2d2008-07-17 12:53:50 -04006670 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
Qu Wenruoe6c4efd2014-09-17 11:53:35 +08006671
6672 if (existing->start > map_start) {
6673 next = existing;
6674 prev = prev_extent_map(next);
6675 } else {
6676 prev = existing;
6677 next = next_extent_map(prev);
6678 }
6679
6680 start = prev ? extent_map_end(prev) : em->start;
6681 start = max_t(u64, start, em->start);
6682 end = next ? next->start : extent_map_end(em);
6683 end = min_t(u64, end, extent_map_end(em));
6684 start_diff = start - em->start;
6685 em->start = start;
6686 em->len = end - start;
Chris Masonc8b97812008-10-29 14:49:59 -04006687 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6688 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
Chris Masone6dcd2d2008-07-17 12:53:50 -04006689 em->block_start += start_diff;
Chris Masonc8b97812008-10-29 14:49:59 -04006690 em->block_len -= start_diff;
6691 }
Josef Bacik09a2a8f92013-04-05 16:51:15 -04006692 return add_extent_mapping(em_tree, em, 0);
Chris Mason3b951512008-04-17 11:29:12 -04006693}
6694
Chris Masonc8b97812008-10-29 14:49:59 -04006695static noinline int uncompress_inline(struct btrfs_path *path,
Byongho Leee40da0e2015-05-19 23:46:45 +09006696 struct page *page,
Chris Masonc8b97812008-10-29 14:49:59 -04006697 size_t pg_offset, u64 extent_offset,
6698 struct btrfs_file_extent_item *item)
6699{
6700 int ret;
6701 struct extent_buffer *leaf = path->nodes[0];
6702 char *tmp;
6703 size_t max_size;
6704 unsigned long inline_size;
6705 unsigned long ptr;
Li Zefan261507a02010-12-17 14:21:50 +08006706 int compress_type;
Chris Masonc8b97812008-10-29 14:49:59 -04006707
6708 WARN_ON(pg_offset != 0);
Li Zefan261507a02010-12-17 14:21:50 +08006709 compress_type = btrfs_file_extent_compression(leaf, item);
Chris Masonc8b97812008-10-29 14:49:59 -04006710 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6711 inline_size = btrfs_file_extent_inline_item_len(leaf,
Ross Kirkdd3cc162013-09-16 15:58:09 +01006712 btrfs_item_nr(path->slots[0]));
Chris Masonc8b97812008-10-29 14:49:59 -04006713 tmp = kmalloc(inline_size, GFP_NOFS);
Tsutomu Itoh8d413712011-04-25 19:43:52 -04006714 if (!tmp)
6715 return -ENOMEM;
Chris Masonc8b97812008-10-29 14:49:59 -04006716 ptr = btrfs_file_extent_inline_start(item);
6717
6718 read_extent_buffer(leaf, tmp, ptr, inline_size);
6719
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03006720 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
Li Zefan261507a02010-12-17 14:21:50 +08006721 ret = btrfs_decompress(compress_type, tmp, page,
6722 extent_offset, inline_size, max_size);
Chris Masonc8b97812008-10-29 14:49:59 -04006723 kfree(tmp);
Zach Brown166ae5a2014-05-09 17:15:10 -04006724 return ret;
Chris Masonc8b97812008-10-29 14:49:59 -04006725}
6726
Chris Masond352ac62008-09-29 15:18:18 -04006727/*
6728 * a bit scary, this does extent mapping from logical file offset to the disk.
Chris Masond3977122009-01-05 21:25:51 -05006729 * the ugly parts come from merging extents from the disk with the in-ram
6730 * representation. This gets more complex because of the data=ordered code,
Chris Masond352ac62008-09-29 15:18:18 -04006731 * where the in-ram extents might be locked pending data=ordered completion.
6732 *
6733 * This also copies inline extents directly into the page.
6734 */
Chris Masond3977122009-01-05 21:25:51 -05006735
Chris Masona52d9a82007-08-27 16:49:44 -04006736struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
Chris Mason70dec802008-01-29 09:59:12 -05006737 size_t pg_offset, u64 start, u64 len,
Chris Masona52d9a82007-08-27 16:49:44 -04006738 int create)
6739{
6740 int ret;
6741 int err = 0;
Chris Masona52d9a82007-08-27 16:49:44 -04006742 u64 extent_start = 0;
6743 u64 extent_end = 0;
Li Zefan33345d012011-04-20 10:31:50 +08006744 u64 objectid = btrfs_ino(inode);
Chris Masona52d9a82007-08-27 16:49:44 -04006745 u32 found_type;
Chris Masonf4219502008-07-22 11:18:09 -04006746 struct btrfs_path *path = NULL;
Chris Masona52d9a82007-08-27 16:49:44 -04006747 struct btrfs_root *root = BTRFS_I(inode)->root;
6748 struct btrfs_file_extent_item *item;
Chris Mason5f39d392007-10-15 16:14:19 -04006749 struct extent_buffer *leaf;
6750 struct btrfs_key found_key;
Chris Masona52d9a82007-08-27 16:49:44 -04006751 struct extent_map *em = NULL;
6752 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Chris Masond1310b22008-01-24 16:13:08 -05006753 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Chris Masona52d9a82007-08-27 16:49:44 -04006754 struct btrfs_trans_handle *trans = NULL;
Filipe Manana7ffbb592014-06-09 03:48:05 +01006755 const bool new_inline = !page || create;
Chris Masona52d9a82007-08-27 16:49:44 -04006756
Chris Masona52d9a82007-08-27 16:49:44 -04006757again:
Chris Mason890871b2009-09-02 16:24:52 -04006758 read_lock(&em_tree->lock);
Chris Masond1310b22008-01-24 16:13:08 -05006759 em = lookup_extent_mapping(em_tree, start, len);
Chris Masona061fc82008-05-07 11:43:44 -04006760 if (em)
6761 em->bdev = root->fs_info->fs_devices->latest_bdev;
Chris Mason890871b2009-09-02 16:24:52 -04006762 read_unlock(&em_tree->lock);
Chris Masond1310b22008-01-24 16:13:08 -05006763
Chris Masona52d9a82007-08-27 16:49:44 -04006764 if (em) {
Chris Masone1c4b742008-04-22 13:26:46 -04006765 if (em->start > start || em->start + em->len <= start)
6766 free_extent_map(em);
6767 else if (em->block_start == EXTENT_MAP_INLINE && page)
Chris Mason70dec802008-01-29 09:59:12 -05006768 free_extent_map(em);
6769 else
6770 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04006771 }
David Sterba172ddd62011-04-21 00:48:27 +02006772 em = alloc_extent_map();
Chris Masona52d9a82007-08-27 16:49:44 -04006773 if (!em) {
Chris Masond1310b22008-01-24 16:13:08 -05006774 err = -ENOMEM;
6775 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04006776 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04006777 em->bdev = root->fs_info->fs_devices->latest_bdev;
Chris Masond1310b22008-01-24 16:13:08 -05006778 em->start = EXTENT_MAP_HOLE;
Chris Mason445a6942008-11-10 11:53:33 -05006779 em->orig_start = EXTENT_MAP_HOLE;
Chris Masond1310b22008-01-24 16:13:08 -05006780 em->len = (u64)-1;
Chris Masonc8b97812008-10-29 14:49:59 -04006781 em->block_len = (u64)-1;
Chris Masonf4219502008-07-22 11:18:09 -04006782
6783 if (!path) {
6784 path = btrfs_alloc_path();
Josef Bacik026fd312011-05-13 10:32:11 -04006785 if (!path) {
6786 err = -ENOMEM;
6787 goto out;
6788 }
6789 /*
6790 * Chances are we'll be called again, so go ahead and do
6791 * readahead
6792 */
David Sterbae4058b52015-11-27 16:31:35 +01006793 path->reada = READA_FORWARD;
Chris Masonf4219502008-07-22 11:18:09 -04006794 }
6795
Chris Mason179e29e2007-11-01 11:28:41 -04006796 ret = btrfs_lookup_file_extent(trans, root, path,
6797 objectid, start, trans != NULL);
Chris Masona52d9a82007-08-27 16:49:44 -04006798 if (ret < 0) {
6799 err = ret;
6800 goto out;
6801 }
6802
6803 if (ret != 0) {
6804 if (path->slots[0] == 0)
6805 goto not_found;
6806 path->slots[0]--;
6807 }
6808
Chris Mason5f39d392007-10-15 16:14:19 -04006809 leaf = path->nodes[0];
6810 item = btrfs_item_ptr(leaf, path->slots[0],
Chris Masona52d9a82007-08-27 16:49:44 -04006811 struct btrfs_file_extent_item);
Chris Masona52d9a82007-08-27 16:49:44 -04006812 /* are we inside the extent that was found? */
Chris Mason5f39d392007-10-15 16:14:19 -04006813 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
David Sterba962a2982014-06-04 18:41:45 +02006814 found_type = found_key.type;
Chris Mason5f39d392007-10-15 16:14:19 -04006815 if (found_key.objectid != objectid ||
Chris Masona52d9a82007-08-27 16:49:44 -04006816 found_type != BTRFS_EXTENT_DATA_KEY) {
Josef Bacik25a50342013-10-14 12:08:38 -04006817 /*
6818 * If we backup past the first extent we want to move forward
6819 * and see if there is an extent in front of us, otherwise we'll
6820 * say there is a hole for our whole search range which can
6821 * cause problems.
6822 */
6823 extent_end = start;
6824 goto next;
Chris Masona52d9a82007-08-27 16:49:44 -04006825 }
6826
Chris Mason5f39d392007-10-15 16:14:19 -04006827 found_type = btrfs_file_extent_type(leaf, item);
6828 extent_start = found_key.offset;
Yan Zhengd899e052008-10-30 14:25:28 -04006829 if (found_type == BTRFS_FILE_EXTENT_REG ||
6830 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Chris Masona52d9a82007-08-27 16:49:44 -04006831 extent_end = extent_start +
Chris Masondb945352007-10-15 16:15:53 -04006832 btrfs_file_extent_num_bytes(leaf, item);
Yan Zheng9036c102008-10-30 14:19:41 -04006833 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6834 size_t size;
Chris Mason514ac8a2014-01-03 21:07:00 -08006835 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
Qu Wenruofda28322013-02-26 08:10:22 +00006836 extent_end = ALIGN(extent_start + size, root->sectorsize);
Yan Zheng9036c102008-10-30 14:19:41 -04006837 }
Josef Bacik25a50342013-10-14 12:08:38 -04006838next:
Yan Zheng9036c102008-10-30 14:19:41 -04006839 if (start >= extent_end) {
6840 path->slots[0]++;
6841 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6842 ret = btrfs_next_leaf(root, path);
6843 if (ret < 0) {
6844 err = ret;
6845 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04006846 }
Yan Zheng9036c102008-10-30 14:19:41 -04006847 if (ret > 0)
6848 goto not_found;
6849 leaf = path->nodes[0];
Chris Masona52d9a82007-08-27 16:49:44 -04006850 }
Yan Zheng9036c102008-10-30 14:19:41 -04006851 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6852 if (found_key.objectid != objectid ||
6853 found_key.type != BTRFS_EXTENT_DATA_KEY)
6854 goto not_found;
6855 if (start + len <= found_key.offset)
6856 goto not_found;
Wang Shilonge2eca692014-07-17 11:44:14 +08006857 if (start > found_key.offset)
6858 goto next;
Yan Zheng9036c102008-10-30 14:19:41 -04006859 em->start = start;
Josef Bacik70c8a912012-10-11 16:54:30 -04006860 em->orig_start = start;
Yan Zheng9036c102008-10-30 14:19:41 -04006861 em->len = found_key.offset - start;
6862 goto not_found_em;
6863 }
6864
Filipe Manana7ffbb592014-06-09 03:48:05 +01006865 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6866
Yan Zhengd899e052008-10-30 14:25:28 -04006867 if (found_type == BTRFS_FILE_EXTENT_REG ||
6868 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Chris Masona52d9a82007-08-27 16:49:44 -04006869 goto insert;
6870 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason5f39d392007-10-15 16:14:19 -04006871 unsigned long ptr;
Chris Masona52d9a82007-08-27 16:49:44 -04006872 char *map;
Chris Mason3326d1b2007-10-15 16:18:25 -04006873 size_t size;
6874 size_t extent_offset;
6875 size_t copy_size;
Chris Masona52d9a82007-08-27 16:49:44 -04006876
Filipe Manana7ffbb592014-06-09 03:48:05 +01006877 if (new_inline)
Yan689f9342007-10-29 11:41:07 -04006878 goto out;
Yan689f9342007-10-29 11:41:07 -04006879
Chris Mason514ac8a2014-01-03 21:07:00 -08006880 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
Yan Zheng9036c102008-10-30 14:19:41 -04006881 extent_offset = page_offset(page) + pg_offset - extent_start;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03006882 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6883 size - extent_offset);
Chris Mason3326d1b2007-10-15 16:18:25 -04006884 em->start = extent_start + extent_offset;
Qu Wenruofda28322013-02-26 08:10:22 +00006885 em->len = ALIGN(copy_size, root->sectorsize);
Josef Bacikb4939682012-12-03 10:31:19 -05006886 em->orig_block_len = em->len;
Josef Bacik70c8a912012-10-11 16:54:30 -04006887 em->orig_start = em->start;
Yan689f9342007-10-29 11:41:07 -04006888 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
Chris Mason179e29e2007-11-01 11:28:41 -04006889 if (create == 0 && !PageUptodate(page)) {
Li Zefan261507a02010-12-17 14:21:50 +08006890 if (btrfs_file_extent_compression(leaf, item) !=
6891 BTRFS_COMPRESS_NONE) {
Byongho Leee40da0e2015-05-19 23:46:45 +09006892 ret = uncompress_inline(path, page, pg_offset,
Chris Masonc8b97812008-10-29 14:49:59 -04006893 extent_offset, item);
Zach Brown166ae5a2014-05-09 17:15:10 -04006894 if (ret) {
6895 err = ret;
6896 goto out;
6897 }
Chris Masonc8b97812008-10-29 14:49:59 -04006898 } else {
6899 map = kmap(page);
6900 read_extent_buffer(leaf, map + pg_offset, ptr,
6901 copy_size);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03006902 if (pg_offset + copy_size < PAGE_SIZE) {
Chris Mason93c82d52009-09-11 12:36:29 -04006903 memset(map + pg_offset + copy_size, 0,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03006904 PAGE_SIZE - pg_offset -
Chris Mason93c82d52009-09-11 12:36:29 -04006905 copy_size);
6906 }
Chris Masonc8b97812008-10-29 14:49:59 -04006907 kunmap(page);
6908 }
Chris Mason179e29e2007-11-01 11:28:41 -04006909 flush_dcache_page(page);
6910 } else if (create && PageUptodate(page)) {
Jan Schmidt6bf7e082011-12-01 14:35:19 +01006911 BUG();
Chris Mason179e29e2007-11-01 11:28:41 -04006912 if (!trans) {
6913 kunmap(page);
6914 free_extent_map(em);
6915 em = NULL;
Chris Masonff5714c2011-05-28 07:00:39 -04006916
David Sterbab3b4aa72011-04-21 01:20:15 +02006917 btrfs_release_path(path);
Josef Bacik7a7eaa42011-04-13 12:54:33 -04006918 trans = btrfs_join_transaction(root);
Chris Masonff5714c2011-05-28 07:00:39 -04006919
Tsutomu Itoh3612b492011-01-25 02:51:38 +00006920 if (IS_ERR(trans))
6921 return ERR_CAST(trans);
Chris Mason179e29e2007-11-01 11:28:41 -04006922 goto again;
6923 }
Chris Masonc8b97812008-10-29 14:49:59 -04006924 map = kmap(page);
Chris Mason70dec802008-01-29 09:59:12 -05006925 write_extent_buffer(leaf, map + pg_offset, ptr,
Chris Mason179e29e2007-11-01 11:28:41 -04006926 copy_size);
Chris Masonc8b97812008-10-29 14:49:59 -04006927 kunmap(page);
Chris Mason179e29e2007-11-01 11:28:41 -04006928 btrfs_mark_buffer_dirty(leaf);
Chris Masona52d9a82007-08-27 16:49:44 -04006929 }
Chris Masond1310b22008-01-24 16:13:08 -05006930 set_extent_uptodate(io_tree, em->start,
Arne Jansen507903b2011-04-06 10:02:20 +00006931 extent_map_end(em) - 1, NULL, GFP_NOFS);
Chris Masona52d9a82007-08-27 16:49:44 -04006932 goto insert;
Chris Masona52d9a82007-08-27 16:49:44 -04006933 }
6934not_found:
6935 em->start = start;
Josef Bacik70c8a912012-10-11 16:54:30 -04006936 em->orig_start = start;
Chris Masond1310b22008-01-24 16:13:08 -05006937 em->len = len;
Chris Masona52d9a82007-08-27 16:49:44 -04006938not_found_em:
Chris Mason5f39d392007-10-15 16:14:19 -04006939 em->block_start = EXTENT_MAP_HOLE;
Yan Zheng9036c102008-10-30 14:19:41 -04006940 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
Chris Masona52d9a82007-08-27 16:49:44 -04006941insert:
David Sterbab3b4aa72011-04-21 01:20:15 +02006942 btrfs_release_path(path);
Chris Masond1310b22008-01-24 16:13:08 -05006943 if (em->start > start || extent_map_end(em) <= start) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00006944 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02006945 em->start, em->len, start, len);
Chris Masona52d9a82007-08-27 16:49:44 -04006946 err = -EIO;
6947 goto out;
6948 }
Chris Masond1310b22008-01-24 16:13:08 -05006949
6950 err = 0;
Chris Mason890871b2009-09-02 16:24:52 -04006951 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -04006952 ret = add_extent_mapping(em_tree, em, 0);
Chris Mason3b951512008-04-17 11:29:12 -04006953 /* it is possible that someone inserted the extent into the tree
6954 * while we had the lock dropped. It is also possible that
6955 * an overlapping map exists in the tree
6956 */
Chris Masona52d9a82007-08-27 16:49:44 -04006957 if (ret == -EEXIST) {
Chris Mason3b951512008-04-17 11:29:12 -04006958 struct extent_map *existing;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006959
6960 ret = 0;
6961
Qu Wenruoe6c4efd2014-09-17 11:53:35 +08006962 existing = search_extent_mapping(em_tree, start, len);
6963 /*
6964 * existing will always be non-NULL, since there must be
6965 * extent causing the -EEXIST.
6966 */
6967 if (start >= extent_map_end(existing) ||
Qu Wenruo32be3a12014-09-22 09:13:03 +08006968 start <= existing->start) {
Qu Wenruoe6c4efd2014-09-17 11:53:35 +08006969 /*
6970 * The existing extent map is the one nearest to
6971 * the [start, start + len) range which overlaps
6972 */
6973 err = merge_extent_mapping(em_tree, existing,
6974 em, start);
Chris Masone1c4b742008-04-22 13:26:46 -04006975 free_extent_map(existing);
Qu Wenruoe6c4efd2014-09-17 11:53:35 +08006976 if (err) {
Chris Mason3b951512008-04-17 11:29:12 -04006977 free_extent_map(em);
6978 em = NULL;
6979 }
6980 } else {
6981 free_extent_map(em);
6982 em = existing;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006983 err = 0;
Chris Masona52d9a82007-08-27 16:49:44 -04006984 }
Chris Masona52d9a82007-08-27 16:49:44 -04006985 }
Chris Mason890871b2009-09-02 16:24:52 -04006986 write_unlock(&em_tree->lock);
Chris Masona52d9a82007-08-27 16:49:44 -04006987out:
liubo1abe9b82011-03-24 11:18:59 +00006988
Steven Rostedt4cd8587c2013-11-14 22:57:29 -05006989 trace_btrfs_get_extent(root, em);
liubo1abe9b82011-03-24 11:18:59 +00006990
Tsutomu Itoh527afb42015-08-19 14:55:00 +09006991 btrfs_free_path(path);
Chris Masona52d9a82007-08-27 16:49:44 -04006992 if (trans) {
6993 ret = btrfs_end_transaction(trans, root);
Chris Masond3977122009-01-05 21:25:51 -05006994 if (!err)
Chris Masona52d9a82007-08-27 16:49:44 -04006995 err = ret;
6996 }
Chris Masona52d9a82007-08-27 16:49:44 -04006997 if (err) {
6998 free_extent_map(em);
Chris Masona52d9a82007-08-27 16:49:44 -04006999 return ERR_PTR(err);
7000 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007001 BUG_ON(!em); /* Error is always set */
Chris Masona52d9a82007-08-27 16:49:44 -04007002 return em;
7003}
7004
Chris Masonec29ed52011-02-23 16:23:20 -05007005struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
7006 size_t pg_offset, u64 start, u64 len,
7007 int create)
7008{
7009 struct extent_map *em;
7010 struct extent_map *hole_em = NULL;
7011 u64 range_start = start;
7012 u64 end;
7013 u64 found;
7014 u64 found_end;
7015 int err = 0;
7016
7017 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7018 if (IS_ERR(em))
7019 return em;
7020 if (em) {
7021 /*
Liu Bof9e4fb52013-01-07 10:10:12 +00007022 * if our em maps to
7023 * - a hole or
7024 * - a pre-alloc extent,
7025 * there might actually be delalloc bytes behind it.
Chris Masonec29ed52011-02-23 16:23:20 -05007026 */
Liu Bof9e4fb52013-01-07 10:10:12 +00007027 if (em->block_start != EXTENT_MAP_HOLE &&
7028 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
Chris Masonec29ed52011-02-23 16:23:20 -05007029 return em;
7030 else
7031 hole_em = em;
7032 }
7033
7034 /* check to see if we've wrapped (len == -1 or similar) */
7035 end = start + len;
7036 if (end < start)
7037 end = (u64)-1;
7038 else
7039 end -= 1;
7040
7041 em = NULL;
7042
7043 /* ok, we didn't find anything, lets look for delalloc */
7044 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
7045 end, len, EXTENT_DELALLOC, 1);
7046 found_end = range_start + found;
7047 if (found_end < range_start)
7048 found_end = (u64)-1;
7049
7050 /*
7051 * we didn't find anything useful, return
7052 * the original results from get_extent()
7053 */
7054 if (range_start > end || found_end <= start) {
7055 em = hole_em;
7056 hole_em = NULL;
7057 goto out;
7058 }
7059
7060 /* adjust the range_start to make sure it doesn't
7061 * go backwards from the start they passed in
7062 */
Dulshani Gunawardhana67871252013-10-31 10:33:04 +05307063 range_start = max(start, range_start);
Chris Masonec29ed52011-02-23 16:23:20 -05007064 found = found_end - range_start;
7065
7066 if (found > 0) {
7067 u64 hole_start = start;
7068 u64 hole_len = len;
7069
David Sterba172ddd62011-04-21 00:48:27 +02007070 em = alloc_extent_map();
Chris Masonec29ed52011-02-23 16:23:20 -05007071 if (!em) {
7072 err = -ENOMEM;
7073 goto out;
7074 }
7075 /*
7076 * when btrfs_get_extent can't find anything it
7077 * returns one huge hole
7078 *
7079 * make sure what it found really fits our range, and
7080 * adjust to make sure it is based on the start from
7081 * the caller
7082 */
7083 if (hole_em) {
7084 u64 calc_end = extent_map_end(hole_em);
7085
7086 if (calc_end <= start || (hole_em->start > end)) {
7087 free_extent_map(hole_em);
7088 hole_em = NULL;
7089 } else {
7090 hole_start = max(hole_em->start, start);
7091 hole_len = calc_end - hole_start;
7092 }
7093 }
7094 em->bdev = NULL;
7095 if (hole_em && range_start > hole_start) {
7096 /* our hole starts before our delalloc, so we
7097 * have to return just the parts of the hole
7098 * that go until the delalloc starts
7099 */
7100 em->len = min(hole_len,
7101 range_start - hole_start);
7102 em->start = hole_start;
7103 em->orig_start = hole_start;
7104 /*
7105 * don't adjust block start at all,
7106 * it is fixed at EXTENT_MAP_HOLE
7107 */
7108 em->block_start = hole_em->block_start;
7109 em->block_len = hole_len;
Liu Bof9e4fb52013-01-07 10:10:12 +00007110 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7111 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
Chris Masonec29ed52011-02-23 16:23:20 -05007112 } else {
7113 em->start = range_start;
7114 em->len = found;
7115 em->orig_start = range_start;
7116 em->block_start = EXTENT_MAP_DELALLOC;
7117 em->block_len = found;
7118 }
7119 } else if (hole_em) {
7120 return hole_em;
7121 }
7122out:
7123
7124 free_extent_map(hole_em);
7125 if (err) {
7126 free_extent_map(em);
7127 return ERR_PTR(err);
7128 }
7129 return em;
7130}
7131
Josef Bacik4b46fce2010-05-23 11:00:55 -04007132static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7133 u64 start, u64 len)
7134{
7135 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik70c8a912012-10-11 16:54:30 -04007136 struct extent_map *em;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007137 struct btrfs_key ins;
7138 u64 alloc_hint;
7139 int ret;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007140
Josef Bacik4b46fce2010-05-23 11:00:55 -04007141 alloc_hint = get_extent_allocation_hint(inode, start, len);
Josef Bacik00361582013-08-14 14:02:47 -04007142 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
Miao Xiee570fd22014-06-19 10:42:50 +08007143 alloc_hint, &ins, 1, 1);
Josef Bacik00361582013-08-14 14:02:47 -04007144 if (ret)
7145 return ERR_PTR(ret);
Josef Bacik4b46fce2010-05-23 11:00:55 -04007146
Filipe Mananade0ee0e2016-01-21 10:17:54 +00007147 /*
7148 * Create the ordered extent before the extent map. This is to avoid
7149 * races with the fast fsync path that would lead to it logging file
7150 * extent items that point to disk extents that were not yet written to.
7151 * The fast fsync path collects ordered extents into a local list and
7152 * then collects all the new extent maps, so we must create the ordered
7153 * extent first and make sure the fast fsync path collects any new
7154 * ordered extents after collecting new extent maps as well.
7155 * The fsync path simply can not rely on inode_dio_wait() because it
7156 * causes deadlock with AIO.
7157 */
Josef Bacik4b46fce2010-05-23 11:00:55 -04007158 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
7159 ins.offset, ins.offset, 0);
7160 if (ret) {
Miao Xiee570fd22014-06-19 10:42:50 +08007161 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
Josef Bacik00361582013-08-14 14:02:47 -04007162 return ERR_PTR(ret);
Josef Bacik4b46fce2010-05-23 11:00:55 -04007163 }
Josef Bacik00361582013-08-14 14:02:47 -04007164
Filipe Mananade0ee0e2016-01-21 10:17:54 +00007165 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
7166 ins.offset, ins.offset, ins.offset, 0);
7167 if (IS_ERR(em)) {
7168 struct btrfs_ordered_extent *oe;
7169
7170 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7171 oe = btrfs_lookup_ordered_extent(inode, start);
7172 ASSERT(oe);
7173 if (WARN_ON(!oe))
7174 return em;
7175 set_bit(BTRFS_ORDERED_IOERR, &oe->flags);
7176 set_bit(BTRFS_ORDERED_IO_DONE, &oe->flags);
7177 btrfs_remove_ordered_extent(inode, oe);
7178 /* Once for our lookup and once for the ordered extents tree. */
7179 btrfs_put_ordered_extent(oe);
7180 btrfs_put_ordered_extent(oe);
7181 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04007182 return em;
7183}
7184
Chris Mason46bfbb52010-05-26 11:04:10 -04007185/*
7186 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7187 * block must be cow'd
7188 */
Josef Bacik00361582013-08-14 14:02:47 -04007189noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
Josef Bacik7ee9e442013-06-21 16:37:03 -04007190 u64 *orig_start, u64 *orig_block_len,
7191 u64 *ram_bytes)
Chris Mason46bfbb52010-05-26 11:04:10 -04007192{
Josef Bacik00361582013-08-14 14:02:47 -04007193 struct btrfs_trans_handle *trans;
Chris Mason46bfbb52010-05-26 11:04:10 -04007194 struct btrfs_path *path;
7195 int ret;
7196 struct extent_buffer *leaf;
7197 struct btrfs_root *root = BTRFS_I(inode)->root;
Miao Xie7b2b7082014-02-27 13:58:05 +08007198 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Chris Mason46bfbb52010-05-26 11:04:10 -04007199 struct btrfs_file_extent_item *fi;
7200 struct btrfs_key key;
7201 u64 disk_bytenr;
7202 u64 backref_offset;
7203 u64 extent_end;
7204 u64 num_bytes;
7205 int slot;
7206 int found_type;
Josef Bacik7ee9e442013-06-21 16:37:03 -04007207 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
Miao Xiee77751a2013-12-27 21:11:50 +08007208
Chris Mason46bfbb52010-05-26 11:04:10 -04007209 path = btrfs_alloc_path();
7210 if (!path)
7211 return -ENOMEM;
7212
Josef Bacik00361582013-08-14 14:02:47 -04007213 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
Chris Mason46bfbb52010-05-26 11:04:10 -04007214 offset, 0);
7215 if (ret < 0)
7216 goto out;
7217
7218 slot = path->slots[0];
7219 if (ret == 1) {
7220 if (slot == 0) {
7221 /* can't find the item, must cow */
7222 ret = 0;
7223 goto out;
7224 }
7225 slot--;
7226 }
7227 ret = 0;
7228 leaf = path->nodes[0];
7229 btrfs_item_key_to_cpu(leaf, &key, slot);
Li Zefan33345d012011-04-20 10:31:50 +08007230 if (key.objectid != btrfs_ino(inode) ||
Chris Mason46bfbb52010-05-26 11:04:10 -04007231 key.type != BTRFS_EXTENT_DATA_KEY) {
7232 /* not our file or wrong item type, must cow */
7233 goto out;
7234 }
7235
7236 if (key.offset > offset) {
7237 /* Wrong offset, must cow */
7238 goto out;
7239 }
7240
7241 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7242 found_type = btrfs_file_extent_type(leaf, fi);
7243 if (found_type != BTRFS_FILE_EXTENT_REG &&
7244 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7245 /* not a regular extent, must cow */
7246 goto out;
7247 }
Josef Bacik7ee9e442013-06-21 16:37:03 -04007248
7249 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7250 goto out;
7251
Miao Xiee77751a2013-12-27 21:11:50 +08007252 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7253 if (extent_end <= offset)
7254 goto out;
7255
Chris Mason46bfbb52010-05-26 11:04:10 -04007256 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
Josef Bacik7ee9e442013-06-21 16:37:03 -04007257 if (disk_bytenr == 0)
7258 goto out;
7259
7260 if (btrfs_file_extent_compression(leaf, fi) ||
7261 btrfs_file_extent_encryption(leaf, fi) ||
7262 btrfs_file_extent_other_encoding(leaf, fi))
7263 goto out;
7264
Chris Mason46bfbb52010-05-26 11:04:10 -04007265 backref_offset = btrfs_file_extent_offset(leaf, fi);
7266
Josef Bacik7ee9e442013-06-21 16:37:03 -04007267 if (orig_start) {
7268 *orig_start = key.offset - backref_offset;
7269 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7270 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7271 }
Josef Bacikeb384b52013-04-24 16:32:55 -04007272
Chris Mason46bfbb52010-05-26 11:04:10 -04007273 if (btrfs_extent_readonly(root, disk_bytenr))
7274 goto out;
Miao Xie7b2b7082014-02-27 13:58:05 +08007275
7276 num_bytes = min(offset + *len, extent_end) - offset;
7277 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7278 u64 range_end;
7279
7280 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7281 ret = test_range_bit(io_tree, offset, range_end,
7282 EXTENT_DELALLOC, 0, NULL);
7283 if (ret) {
7284 ret = -EAGAIN;
7285 goto out;
7286 }
7287 }
7288
Josef Bacik1bda19e2013-10-18 12:10:36 -04007289 btrfs_release_path(path);
Chris Mason46bfbb52010-05-26 11:04:10 -04007290
7291 /*
7292 * look for other files referencing this extent, if we
7293 * find any we must cow
7294 */
Josef Bacik00361582013-08-14 14:02:47 -04007295 trans = btrfs_join_transaction(root);
7296 if (IS_ERR(trans)) {
7297 ret = 0;
Chris Mason46bfbb52010-05-26 11:04:10 -04007298 goto out;
Josef Bacik00361582013-08-14 14:02:47 -04007299 }
7300
7301 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7302 key.offset - backref_offset, disk_bytenr);
7303 btrfs_end_transaction(trans, root);
7304 if (ret) {
7305 ret = 0;
7306 goto out;
7307 }
Chris Mason46bfbb52010-05-26 11:04:10 -04007308
7309 /*
7310 * adjust disk_bytenr and num_bytes to cover just the bytes
7311 * in this extent we are about to write. If there
7312 * are any csums in that range we have to cow in order
7313 * to keep the csums correct
7314 */
7315 disk_bytenr += backref_offset;
7316 disk_bytenr += offset - key.offset;
Chris Mason46bfbb52010-05-26 11:04:10 -04007317 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7318 goto out;
7319 /*
7320 * all of the above have passed, it is safe to overwrite this extent
7321 * without cow
7322 */
Josef Bacikeb384b52013-04-24 16:32:55 -04007323 *len = num_bytes;
Chris Mason46bfbb52010-05-26 11:04:10 -04007324 ret = 1;
7325out:
7326 btrfs_free_path(path);
7327 return ret;
7328}
7329
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07007330bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7331{
7332 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7333 int found = false;
7334 void **pagep = NULL;
7335 struct page *page = NULL;
7336 int start_idx;
7337 int end_idx;
7338
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03007339 start_idx = start >> PAGE_SHIFT;
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07007340
7341 /*
7342 * end is the last byte in the last page. end == start is legal
7343 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03007344 end_idx = end >> PAGE_SHIFT;
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07007345
7346 rcu_read_lock();
7347
7348 /* Most of the code in this while loop is lifted from
7349 * find_get_page. It's been modified to begin searching from a
7350 * page and return just the first page found in that range. If the
7351 * found idx is less than or equal to the end idx then we know that
7352 * a page exists. If no pages are found or if those pages are
7353 * outside of the range then we're fine (yay!) */
7354 while (page == NULL &&
7355 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7356 page = radix_tree_deref_slot(pagep);
7357 if (unlikely(!page))
7358 break;
7359
7360 if (radix_tree_exception(page)) {
Filipe Manana809f9012014-06-05 13:22:25 +01007361 if (radix_tree_deref_retry(page)) {
7362 page = NULL;
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07007363 continue;
Filipe Manana809f9012014-06-05 13:22:25 +01007364 }
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07007365 /*
7366 * Otherwise, shmem/tmpfs must be storing a swap entry
7367 * here as an exceptional entry: so return it without
7368 * attempting to raise page count.
7369 */
Filipe Manana6fdef6d2014-06-05 13:22:26 +01007370 page = NULL;
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07007371 break; /* TODO: Is this relevant for this use case? */
7372 }
7373
Filipe Manana91405152014-06-05 13:22:24 +01007374 if (!page_cache_get_speculative(page)) {
7375 page = NULL;
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07007376 continue;
Filipe Manana91405152014-06-05 13:22:24 +01007377 }
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07007378
7379 /*
7380 * Has the page moved?
7381 * This is part of the lockless pagecache protocol. See
7382 * include/linux/pagemap.h for details.
7383 */
7384 if (unlikely(page != *pagep)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03007385 put_page(page);
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07007386 page = NULL;
7387 }
7388 }
7389
7390 if (page) {
7391 if (page->index <= end_idx)
7392 found = true;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03007393 put_page(page);
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07007394 }
7395
7396 rcu_read_unlock();
7397 return found;
7398}
7399
Josef Bacikeb838e72012-07-31 16:28:48 -04007400static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7401 struct extent_state **cached_state, int writing)
7402{
7403 struct btrfs_ordered_extent *ordered;
7404 int ret = 0;
7405
7406 while (1) {
7407 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
David Sterbaff13db42015-12-03 14:30:40 +01007408 cached_state);
Josef Bacikeb838e72012-07-31 16:28:48 -04007409 /*
7410 * We're concerned with the entire range that we're going to be
7411 * doing DIO to, so we need to make sure theres no ordered
7412 * extents in this range.
7413 */
7414 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7415 lockend - lockstart + 1);
7416
7417 /*
7418 * We need to make sure there are no buffered pages in this
7419 * range either, we could have raced between the invalidate in
7420 * generic_file_direct_write and locking the extent. The
7421 * invalidate needs to happen so that reads after a write do not
7422 * get stale data.
7423 */
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07007424 if (!ordered &&
7425 (!writing ||
7426 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
Josef Bacikeb838e72012-07-31 16:28:48 -04007427 break;
7428
7429 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7430 cached_state, GFP_NOFS);
7431
7432 if (ordered) {
Filipe Mananaade77022016-02-18 14:28:55 +00007433 /*
7434 * If we are doing a DIO read and the ordered extent we
7435 * found is for a buffered write, we can not wait for it
7436 * to complete and retry, because if we do so we can
7437 * deadlock with concurrent buffered writes on page
7438 * locks. This happens only if our DIO read covers more
7439 * than one extent map, if at this point has already
7440 * created an ordered extent for a previous extent map
7441 * and locked its range in the inode's io tree, and a
7442 * concurrent write against that previous extent map's
7443 * range and this range started (we unlock the ranges
7444 * in the io tree only when the bios complete and
7445 * buffered writes always lock pages before attempting
7446 * to lock range in the io tree).
7447 */
7448 if (writing ||
7449 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7450 btrfs_start_ordered_extent(inode, ordered, 1);
7451 else
7452 ret = -ENOTBLK;
Josef Bacikeb838e72012-07-31 16:28:48 -04007453 btrfs_put_ordered_extent(ordered);
7454 } else {
Josef Bacikeb838e72012-07-31 16:28:48 -04007455 /*
Filipe Mananab850ae12015-12-08 16:23:16 +00007456 * We could trigger writeback for this range (and wait
7457 * for it to complete) and then invalidate the pages for
7458 * this range (through invalidate_inode_pages2_range()),
7459 * but that can lead us to a deadlock with a concurrent
7460 * call to readpages() (a buffered read or a defrag call
7461 * triggered a readahead) on a page lock due to an
7462 * ordered dio extent we created before but did not have
7463 * yet a corresponding bio submitted (whence it can not
7464 * complete), which makes readpages() wait for that
7465 * ordered extent to complete while holding a lock on
7466 * that page.
Josef Bacikeb838e72012-07-31 16:28:48 -04007467 */
Filipe Mananab850ae12015-12-08 16:23:16 +00007468 ret = -ENOTBLK;
Josef Bacikeb838e72012-07-31 16:28:48 -04007469 }
7470
Filipe Mananaade77022016-02-18 14:28:55 +00007471 if (ret)
7472 break;
7473
Josef Bacikeb838e72012-07-31 16:28:48 -04007474 cond_resched();
7475 }
7476
7477 return ret;
7478}
7479
Josef Bacik69ffb542012-09-11 15:40:07 -04007480static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7481 u64 len, u64 orig_start,
7482 u64 block_start, u64 block_len,
Josef Bacikcc95bef2013-04-04 14:31:27 -04007483 u64 orig_block_len, u64 ram_bytes,
7484 int type)
Josef Bacik69ffb542012-09-11 15:40:07 -04007485{
7486 struct extent_map_tree *em_tree;
7487 struct extent_map *em;
7488 struct btrfs_root *root = BTRFS_I(inode)->root;
7489 int ret;
7490
7491 em_tree = &BTRFS_I(inode)->extent_tree;
7492 em = alloc_extent_map();
7493 if (!em)
7494 return ERR_PTR(-ENOMEM);
7495
7496 em->start = start;
7497 em->orig_start = orig_start;
Josef Bacik2ab28f32012-10-12 15:27:49 -04007498 em->mod_start = start;
7499 em->mod_len = len;
Josef Bacik69ffb542012-09-11 15:40:07 -04007500 em->len = len;
7501 em->block_len = block_len;
7502 em->block_start = block_start;
7503 em->bdev = root->fs_info->fs_devices->latest_bdev;
Josef Bacikb4939682012-12-03 10:31:19 -05007504 em->orig_block_len = orig_block_len;
Josef Bacikcc95bef2013-04-04 14:31:27 -04007505 em->ram_bytes = ram_bytes;
Josef Bacik70c8a912012-10-11 16:54:30 -04007506 em->generation = -1;
Josef Bacik69ffb542012-09-11 15:40:07 -04007507 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7508 if (type == BTRFS_ORDERED_PREALLOC)
Josef Bacikb11e2342012-12-03 10:58:15 -05007509 set_bit(EXTENT_FLAG_FILLING, &em->flags);
Josef Bacik69ffb542012-09-11 15:40:07 -04007510
7511 do {
7512 btrfs_drop_extent_cache(inode, em->start,
7513 em->start + em->len - 1, 0);
7514 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -04007515 ret = add_extent_mapping(em_tree, em, 1);
Josef Bacik69ffb542012-09-11 15:40:07 -04007516 write_unlock(&em_tree->lock);
7517 } while (ret == -EEXIST);
7518
7519 if (ret) {
7520 free_extent_map(em);
7521 return ERR_PTR(ret);
7522 }
7523
7524 return em;
7525}
7526
Filipe Manana9c9464c2015-11-04 09:52:04 +00007527static void adjust_dio_outstanding_extents(struct inode *inode,
7528 struct btrfs_dio_data *dio_data,
7529 const u64 len)
7530{
7531 unsigned num_extents;
7532
7533 num_extents = (unsigned) div64_u64(len + BTRFS_MAX_EXTENT_SIZE - 1,
7534 BTRFS_MAX_EXTENT_SIZE);
7535 /*
7536 * If we have an outstanding_extents count still set then we're
7537 * within our reservation, otherwise we need to adjust our inode
7538 * counter appropriately.
7539 */
7540 if (dio_data->outstanding_extents) {
7541 dio_data->outstanding_extents -= num_extents;
7542 } else {
7543 spin_lock(&BTRFS_I(inode)->lock);
7544 BTRFS_I(inode)->outstanding_extents += num_extents;
7545 spin_unlock(&BTRFS_I(inode)->lock);
7546 }
7547}
7548
Josef Bacik4b46fce2010-05-23 11:00:55 -04007549static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7550 struct buffer_head *bh_result, int create)
7551{
7552 struct extent_map *em;
7553 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacikeb838e72012-07-31 16:28:48 -04007554 struct extent_state *cached_state = NULL;
chandan50745b02015-08-28 21:10:13 +05307555 struct btrfs_dio_data *dio_data = NULL;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007556 u64 start = iblock << inode->i_blkbits;
Josef Bacikeb838e72012-07-31 16:28:48 -04007557 u64 lockstart, lockend;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007558 u64 len = bh_result->b_size;
Josef Bacikeb838e72012-07-31 16:28:48 -04007559 int unlock_bits = EXTENT_LOCKED;
Miao Xie09348562013-02-07 10:12:07 +00007560 int ret = 0;
Josef Bacikeb838e72012-07-31 16:28:48 -04007561
Miao Xie172a5042013-02-21 02:48:22 -07007562 if (create)
Josef Bacik32667892015-02-11 15:08:58 -05007563 unlock_bits |= EXTENT_DIRTY;
Miao Xie172a5042013-02-21 02:48:22 -07007564 else
Josef Bacikc3298612012-08-03 16:49:19 -04007565 len = min_t(u64, len, root->sectorsize);
Josef Bacikeb838e72012-07-31 16:28:48 -04007566
Josef Bacikc3298612012-08-03 16:49:19 -04007567 lockstart = start;
7568 lockend = start + len - 1;
7569
Josef Bacike1cbbfa2015-03-17 10:52:28 -04007570 if (current->journal_info) {
7571 /*
7572 * Need to pull our outstanding extents and set journal_info to NULL so
7573 * that anything that needs to check if there's a transction doesn't get
7574 * confused.
7575 */
chandan50745b02015-08-28 21:10:13 +05307576 dio_data = current->journal_info;
Josef Bacike1cbbfa2015-03-17 10:52:28 -04007577 current->journal_info = NULL;
7578 }
7579
Josef Bacikeb838e72012-07-31 16:28:48 -04007580 /*
7581 * If this errors out it's because we couldn't invalidate pagecache for
7582 * this range and we need to fallback to buffered.
7583 */
Filipe Manana9c9464c2015-11-04 09:52:04 +00007584 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7585 create)) {
7586 ret = -ENOTBLK;
7587 goto err;
7588 }
Josef Bacikeb838e72012-07-31 16:28:48 -04007589
Josef Bacik4b46fce2010-05-23 11:00:55 -04007590 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
Josef Bacikeb838e72012-07-31 16:28:48 -04007591 if (IS_ERR(em)) {
7592 ret = PTR_ERR(em);
7593 goto unlock_err;
7594 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04007595
7596 /*
7597 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7598 * io. INLINE is special, and we could probably kludge it in here, but
7599 * it's still buffered so for safety lets just fall back to the generic
7600 * buffered path.
7601 *
7602 * For COMPRESSED we _have_ to read the entire extent in so we can
7603 * decompress it, so there will be buffering required no matter what we
7604 * do, so go ahead and fallback to buffered.
7605 *
7606 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7607 * to buffered IO. Don't blame me, this is the price we pay for using
7608 * the generic code.
7609 */
7610 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7611 em->block_start == EXTENT_MAP_INLINE) {
7612 free_extent_map(em);
Josef Bacikeb838e72012-07-31 16:28:48 -04007613 ret = -ENOTBLK;
7614 goto unlock_err;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007615 }
7616
7617 /* Just a good old fashioned hole, return */
7618 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7619 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7620 free_extent_map(em);
Josef Bacikeb838e72012-07-31 16:28:48 -04007621 goto unlock_err;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007622 }
7623
7624 /*
7625 * We don't allocate a new extent in the following cases
7626 *
7627 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7628 * existing extent.
7629 * 2) The extent is marked as PREALLOC. We're good to go here and can
7630 * just use the extent.
7631 *
7632 */
Chris Mason46bfbb52010-05-26 11:04:10 -04007633 if (!create) {
Josef Bacikeb838e72012-07-31 16:28:48 -04007634 len = min(len, em->len - (start - em->start));
7635 lockstart = start + len;
7636 goto unlock;
Chris Mason46bfbb52010-05-26 11:04:10 -04007637 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04007638
7639 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7640 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7641 em->block_start != EXTENT_MAP_HOLE)) {
Josef Bacik4b46fce2010-05-23 11:00:55 -04007642 int type;
Josef Bacikeb384b52013-04-24 16:32:55 -04007643 u64 block_start, orig_start, orig_block_len, ram_bytes;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007644
7645 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7646 type = BTRFS_ORDERED_PREALLOC;
7647 else
7648 type = BTRFS_ORDERED_NOCOW;
Chris Mason46bfbb52010-05-26 11:04:10 -04007649 len = min(len, em->len - (start - em->start));
Josef Bacik4b46fce2010-05-23 11:00:55 -04007650 block_start = em->block_start + (start - em->start);
Chris Mason46bfbb52010-05-26 11:04:10 -04007651
Josef Bacik00361582013-08-14 14:02:47 -04007652 if (can_nocow_extent(inode, start, &len, &orig_start,
Josef Bacik7ee9e442013-06-21 16:37:03 -04007653 &orig_block_len, &ram_bytes) == 1) {
Josef Bacik69ffb542012-09-11 15:40:07 -04007654 if (type == BTRFS_ORDERED_PREALLOC) {
7655 free_extent_map(em);
7656 em = create_pinned_em(inode, start, len,
7657 orig_start,
Josef Bacikb4939682012-12-03 10:31:19 -05007658 block_start, len,
Josef Bacikcc95bef2013-04-04 14:31:27 -04007659 orig_block_len,
7660 ram_bytes, type);
Filipe Manana555e1282014-07-07 12:35:21 +01007661 if (IS_ERR(em)) {
7662 ret = PTR_ERR(em);
Josef Bacik69ffb542012-09-11 15:40:07 -04007663 goto unlock_err;
Filipe Manana555e1282014-07-07 12:35:21 +01007664 }
Josef Bacik69ffb542012-09-11 15:40:07 -04007665 }
7666
Chris Mason46bfbb52010-05-26 11:04:10 -04007667 ret = btrfs_add_ordered_extent_dio(inode, start,
7668 block_start, len, len, type);
Chris Mason46bfbb52010-05-26 11:04:10 -04007669 if (ret) {
7670 free_extent_map(em);
Josef Bacikeb838e72012-07-31 16:28:48 -04007671 goto unlock_err;
Chris Mason46bfbb52010-05-26 11:04:10 -04007672 }
7673 goto unlock;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007674 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04007675 }
Josef Bacik00361582013-08-14 14:02:47 -04007676
Chris Mason46bfbb52010-05-26 11:04:10 -04007677 /*
7678 * this will cow the extent, reset the len in case we changed
7679 * it above
7680 */
7681 len = bh_result->b_size;
Josef Bacik70c8a912012-10-11 16:54:30 -04007682 free_extent_map(em);
7683 em = btrfs_new_extent_direct(inode, start, len);
Josef Bacikeb838e72012-07-31 16:28:48 -04007684 if (IS_ERR(em)) {
7685 ret = PTR_ERR(em);
7686 goto unlock_err;
7687 }
Chris Mason46bfbb52010-05-26 11:04:10 -04007688 len = min(len, em->len - (start - em->start));
7689unlock:
Josef Bacik4b46fce2010-05-23 11:00:55 -04007690 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7691 inode->i_blkbits;
Chris Mason46bfbb52010-05-26 11:04:10 -04007692 bh_result->b_size = len;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007693 bh_result->b_bdev = em->bdev;
7694 set_buffer_mapped(bh_result);
Josef Bacikc3473e82012-06-19 10:59:00 -04007695 if (create) {
7696 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7697 set_buffer_new(bh_result);
7698
7699 /*
7700 * Need to update the i_size under the extent lock so buffered
7701 * readers will get the updated i_size when we unlock.
7702 */
7703 if (start + len > i_size_read(inode))
7704 i_size_write(inode, start + len);
Miao Xie09348562013-02-07 10:12:07 +00007705
Filipe Manana9c9464c2015-11-04 09:52:04 +00007706 adjust_dio_outstanding_extents(inode, dio_data, len);
Qu Wenruo7cf5b972015-09-08 17:25:55 +08007707 btrfs_free_reserved_data_space(inode, start, len);
chandan50745b02015-08-28 21:10:13 +05307708 WARN_ON(dio_data->reserve < len);
7709 dio_data->reserve -= len;
Filipe Mananaf28a4922015-12-08 19:23:20 +00007710 dio_data->unsubmitted_oe_range_end = start + len;
chandan50745b02015-08-28 21:10:13 +05307711 current->journal_info = dio_data;
Josef Bacikc3473e82012-06-19 10:59:00 -04007712 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04007713
Josef Bacikeb838e72012-07-31 16:28:48 -04007714 /*
7715 * In the case of write we need to clear and unlock the entire range,
7716 * in the case of read we need to unlock only the end area that we
7717 * aren't using if there is any left over space.
7718 */
Liu Bo24c03fa2012-08-22 20:10:38 -06007719 if (lockstart < lockend) {
Miao Xie09348562013-02-07 10:12:07 +00007720 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7721 lockend, unlock_bits, 1, 0,
7722 &cached_state, GFP_NOFS);
Liu Bo24c03fa2012-08-22 20:10:38 -06007723 } else {
Josef Bacikeb838e72012-07-31 16:28:48 -04007724 free_extent_state(cached_state);
Liu Bo24c03fa2012-08-22 20:10:38 -06007725 }
Josef Bacikeb838e72012-07-31 16:28:48 -04007726
Josef Bacik4b46fce2010-05-23 11:00:55 -04007727 free_extent_map(em);
7728
7729 return 0;
Josef Bacikeb838e72012-07-31 16:28:48 -04007730
7731unlock_err:
Josef Bacikeb838e72012-07-31 16:28:48 -04007732 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7733 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
Filipe Manana9c9464c2015-11-04 09:52:04 +00007734err:
chandan50745b02015-08-28 21:10:13 +05307735 if (dio_data)
7736 current->journal_info = dio_data;
Filipe Manana9c9464c2015-11-04 09:52:04 +00007737 /*
7738 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7739 * write less data then expected, so that we don't underflow our inode's
7740 * outstanding extents counter.
7741 */
7742 if (create && dio_data)
7743 adjust_dio_outstanding_extents(inode, dio_data, len);
7744
Josef Bacikeb838e72012-07-31 16:28:48 -04007745 return ret;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007746}
7747
Miao Xie8b110e32014-09-12 18:44:03 +08007748static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7749 int rw, int mirror_num)
7750{
7751 struct btrfs_root *root = BTRFS_I(inode)->root;
7752 int ret;
7753
7754 BUG_ON(rw & REQ_WRITE);
7755
7756 bio_get(bio);
7757
7758 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7759 BTRFS_WQ_ENDIO_DIO_REPAIR);
7760 if (ret)
7761 goto err;
7762
7763 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7764err:
7765 bio_put(bio);
7766 return ret;
7767}
7768
7769static int btrfs_check_dio_repairable(struct inode *inode,
7770 struct bio *failed_bio,
7771 struct io_failure_record *failrec,
7772 int failed_mirror)
7773{
7774 int num_copies;
7775
7776 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7777 failrec->logical, failrec->len);
7778 if (num_copies == 1) {
7779 /*
7780 * we only have a single copy of the data, so don't bother with
7781 * all the retry and error correction code that follows. no
7782 * matter what the error is, it is very likely to persist.
7783 */
7784 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7785 num_copies, failrec->this_mirror, failed_mirror);
7786 return 0;
7787 }
7788
7789 failrec->failed_mirror = failed_mirror;
7790 failrec->this_mirror++;
7791 if (failrec->this_mirror == failed_mirror)
7792 failrec->this_mirror++;
7793
7794 if (failrec->this_mirror > num_copies) {
7795 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7796 num_copies, failrec->this_mirror, failed_mirror);
7797 return 0;
7798 }
7799
7800 return 1;
7801}
7802
7803static int dio_read_error(struct inode *inode, struct bio *failed_bio,
Chandan Rajendra2dabb322016-01-21 15:55:55 +05307804 struct page *page, unsigned int pgoff,
7805 u64 start, u64 end, int failed_mirror,
7806 bio_end_io_t *repair_endio, void *repair_arg)
Miao Xie8b110e32014-09-12 18:44:03 +08007807{
7808 struct io_failure_record *failrec;
7809 struct bio *bio;
7810 int isector;
7811 int read_mode;
7812 int ret;
7813
7814 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7815
7816 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7817 if (ret)
7818 return ret;
7819
7820 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7821 failed_mirror);
7822 if (!ret) {
7823 free_io_failure(inode, failrec);
7824 return -EIO;
7825 }
7826
Chandan Rajendra2dabb322016-01-21 15:55:55 +05307827 if ((failed_bio->bi_vcnt > 1)
7828 || (failed_bio->bi_io_vec->bv_len
7829 > BTRFS_I(inode)->root->sectorsize))
Miao Xie8b110e32014-09-12 18:44:03 +08007830 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7831 else
7832 read_mode = READ_SYNC;
7833
7834 isector = start - btrfs_io_bio(failed_bio)->logical;
7835 isector >>= inode->i_sb->s_blocksize_bits;
7836 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
Chandan Rajendra2dabb322016-01-21 15:55:55 +05307837 pgoff, isector, repair_endio, repair_arg);
Miao Xie8b110e32014-09-12 18:44:03 +08007838 if (!bio) {
7839 free_io_failure(inode, failrec);
7840 return -EIO;
7841 }
7842
7843 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7844 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7845 read_mode, failrec->this_mirror, failrec->in_validation);
7846
7847 ret = submit_dio_repair_bio(inode, bio, read_mode,
7848 failrec->this_mirror);
7849 if (ret) {
7850 free_io_failure(inode, failrec);
7851 bio_put(bio);
7852 }
7853
7854 return ret;
7855}
7856
7857struct btrfs_retry_complete {
7858 struct completion done;
7859 struct inode *inode;
7860 u64 start;
7861 int uptodate;
7862};
7863
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02007864static void btrfs_retry_endio_nocsum(struct bio *bio)
Miao Xie8b110e32014-09-12 18:44:03 +08007865{
7866 struct btrfs_retry_complete *done = bio->bi_private;
Chandan Rajendra2dabb322016-01-21 15:55:55 +05307867 struct inode *inode;
Miao Xie8b110e32014-09-12 18:44:03 +08007868 struct bio_vec *bvec;
7869 int i;
7870
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02007871 if (bio->bi_error)
Miao Xie8b110e32014-09-12 18:44:03 +08007872 goto end;
7873
Chandan Rajendra2dabb322016-01-21 15:55:55 +05307874 ASSERT(bio->bi_vcnt == 1);
7875 inode = bio->bi_io_vec->bv_page->mapping->host;
7876 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
7877
Miao Xie8b110e32014-09-12 18:44:03 +08007878 done->uptodate = 1;
7879 bio_for_each_segment_all(bvec, bio, i)
7880 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7881end:
7882 complete(&done->done);
7883 bio_put(bio);
7884}
7885
7886static int __btrfs_correct_data_nocsum(struct inode *inode,
7887 struct btrfs_io_bio *io_bio)
Josef Bacik4b46fce2010-05-23 11:00:55 -04007888{
Chandan Rajendra2dabb322016-01-21 15:55:55 +05307889 struct btrfs_fs_info *fs_info;
Kent Overstreet2c30c712013-11-07 12:20:26 -08007890 struct bio_vec *bvec;
Miao Xie8b110e32014-09-12 18:44:03 +08007891 struct btrfs_retry_complete done;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007892 u64 start;
Chandan Rajendra2dabb322016-01-21 15:55:55 +05307893 unsigned int pgoff;
7894 u32 sectorsize;
7895 int nr_sectors;
Kent Overstreet2c30c712013-11-07 12:20:26 -08007896 int i;
Miao Xiec1dc0892014-09-12 18:43:56 +08007897 int ret;
Miao Xiedc380ae2014-09-12 18:43:55 +08007898
Chandan Rajendra2dabb322016-01-21 15:55:55 +05307899 fs_info = BTRFS_I(inode)->root->fs_info;
7900 sectorsize = BTRFS_I(inode)->root->sectorsize;
7901
Miao Xiec1dc0892014-09-12 18:43:56 +08007902 start = io_bio->logical;
Miao Xie8b110e32014-09-12 18:44:03 +08007903 done.inode = inode;
7904
7905 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
Chandan Rajendra2dabb322016-01-21 15:55:55 +05307906 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
7907 pgoff = bvec->bv_offset;
7908
7909next_block_or_try_again:
Miao Xie8b110e32014-09-12 18:44:03 +08007910 done.uptodate = 0;
7911 done.start = start;
7912 init_completion(&done.done);
7913
Chandan Rajendra2dabb322016-01-21 15:55:55 +05307914 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
7915 pgoff, start, start + sectorsize - 1,
7916 io_bio->mirror_num,
7917 btrfs_retry_endio_nocsum, &done);
Miao Xie8b110e32014-09-12 18:44:03 +08007918 if (ret)
7919 return ret;
7920
7921 wait_for_completion(&done.done);
7922
7923 if (!done.uptodate) {
7924 /* We might have another mirror, so try again */
Chandan Rajendra2dabb322016-01-21 15:55:55 +05307925 goto next_block_or_try_again;
Miao Xie8b110e32014-09-12 18:44:03 +08007926 }
7927
Chandan Rajendra2dabb322016-01-21 15:55:55 +05307928 start += sectorsize;
7929
7930 if (nr_sectors--) {
7931 pgoff += sectorsize;
7932 goto next_block_or_try_again;
7933 }
Miao Xie8b110e32014-09-12 18:44:03 +08007934 }
7935
7936 return 0;
7937}
7938
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02007939static void btrfs_retry_endio(struct bio *bio)
Miao Xie8b110e32014-09-12 18:44:03 +08007940{
7941 struct btrfs_retry_complete *done = bio->bi_private;
7942 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
Chandan Rajendra2dabb322016-01-21 15:55:55 +05307943 struct inode *inode;
Miao Xie8b110e32014-09-12 18:44:03 +08007944 struct bio_vec *bvec;
Chandan Rajendra2dabb322016-01-21 15:55:55 +05307945 u64 start;
Miao Xie8b110e32014-09-12 18:44:03 +08007946 int uptodate;
7947 int ret;
7948 int i;
7949
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02007950 if (bio->bi_error)
Miao Xie8b110e32014-09-12 18:44:03 +08007951 goto end;
7952
7953 uptodate = 1;
Chandan Rajendra2dabb322016-01-21 15:55:55 +05307954
7955 start = done->start;
7956
7957 ASSERT(bio->bi_vcnt == 1);
7958 inode = bio->bi_io_vec->bv_page->mapping->host;
7959 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
7960
Miao Xie8b110e32014-09-12 18:44:03 +08007961 bio_for_each_segment_all(bvec, bio, i) {
7962 ret = __readpage_endio_check(done->inode, io_bio, i,
Chandan Rajendra2dabb322016-01-21 15:55:55 +05307963 bvec->bv_page, bvec->bv_offset,
7964 done->start, bvec->bv_len);
Miao Xie8b110e32014-09-12 18:44:03 +08007965 if (!ret)
7966 clean_io_failure(done->inode, done->start,
Chandan Rajendra2dabb322016-01-21 15:55:55 +05307967 bvec->bv_page, bvec->bv_offset);
Miao Xie8b110e32014-09-12 18:44:03 +08007968 else
7969 uptodate = 0;
7970 }
7971
7972 done->uptodate = uptodate;
7973end:
7974 complete(&done->done);
7975 bio_put(bio);
7976}
7977
7978static int __btrfs_subio_endio_read(struct inode *inode,
7979 struct btrfs_io_bio *io_bio, int err)
7980{
Chandan Rajendra2dabb322016-01-21 15:55:55 +05307981 struct btrfs_fs_info *fs_info;
Miao Xie8b110e32014-09-12 18:44:03 +08007982 struct bio_vec *bvec;
7983 struct btrfs_retry_complete done;
7984 u64 start;
7985 u64 offset = 0;
Chandan Rajendra2dabb322016-01-21 15:55:55 +05307986 u32 sectorsize;
7987 int nr_sectors;
7988 unsigned int pgoff;
7989 int csum_pos;
Miao Xie8b110e32014-09-12 18:44:03 +08007990 int i;
7991 int ret;
7992
Chandan Rajendra2dabb322016-01-21 15:55:55 +05307993 fs_info = BTRFS_I(inode)->root->fs_info;
7994 sectorsize = BTRFS_I(inode)->root->sectorsize;
7995
Miao Xie8b110e32014-09-12 18:44:03 +08007996 err = 0;
7997 start = io_bio->logical;
7998 done.inode = inode;
7999
Miao Xiec1dc0892014-09-12 18:43:56 +08008000 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308001 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
8002
8003 pgoff = bvec->bv_offset;
8004next_block:
8005 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8006 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8007 bvec->bv_page, pgoff, start,
8008 sectorsize);
Miao Xie8b110e32014-09-12 18:44:03 +08008009 if (likely(!ret))
8010 goto next;
8011try_again:
8012 done.uptodate = 0;
8013 done.start = start;
8014 init_completion(&done.done);
8015
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308016 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
8017 pgoff, start, start + sectorsize - 1,
8018 io_bio->mirror_num,
8019 btrfs_retry_endio, &done);
Miao Xie8b110e32014-09-12 18:44:03 +08008020 if (ret) {
8021 err = ret;
8022 goto next;
8023 }
8024
8025 wait_for_completion(&done.done);
8026
8027 if (!done.uptodate) {
8028 /* We might have another mirror, so try again */
8029 goto try_again;
8030 }
8031next:
Chandan Rajendra2dabb322016-01-21 15:55:55 +05308032 offset += sectorsize;
8033 start += sectorsize;
8034
8035 ASSERT(nr_sectors);
8036
8037 if (--nr_sectors) {
8038 pgoff += sectorsize;
8039 goto next_block;
8040 }
Kent Overstreet2c30c712013-11-07 12:20:26 -08008041 }
Miao Xiec1dc0892014-09-12 18:43:56 +08008042
8043 return err;
8044}
8045
Miao Xie8b110e32014-09-12 18:44:03 +08008046static int btrfs_subio_endio_read(struct inode *inode,
8047 struct btrfs_io_bio *io_bio, int err)
8048{
8049 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8050
8051 if (skip_csum) {
8052 if (unlikely(err))
8053 return __btrfs_correct_data_nocsum(inode, io_bio);
8054 else
8055 return 0;
8056 } else {
8057 return __btrfs_subio_endio_read(inode, io_bio, err);
8058 }
8059}
8060
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02008061static void btrfs_endio_direct_read(struct bio *bio)
Miao Xiec1dc0892014-09-12 18:43:56 +08008062{
8063 struct btrfs_dio_private *dip = bio->bi_private;
8064 struct inode *inode = dip->inode;
8065 struct bio *dio_bio;
8066 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02008067 int err = bio->bi_error;
Miao Xiec1dc0892014-09-12 18:43:56 +08008068
Miao Xie8b110e32014-09-12 18:44:03 +08008069 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8070 err = btrfs_subio_endio_read(inode, io_bio, err);
Miao Xiec1dc0892014-09-12 18:43:56 +08008071
Josef Bacik4b46fce2010-05-23 11:00:55 -04008072 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01008073 dip->logical_offset + dip->bytes - 1);
Chris Mason9be33952013-05-17 18:30:14 -04008074 dio_bio = dip->dio_bio;
Josef Bacik4b46fce2010-05-23 11:00:55 -04008075
Josef Bacik4b46fce2010-05-23 11:00:55 -04008076 kfree(dip);
Josef Bacikc0da7aa2011-03-22 11:05:07 -04008077
Filipe Manana1636d1d2016-02-15 16:20:26 +00008078 dio_bio->bi_error = bio->bi_error;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02008079 dio_end_io(dio_bio, bio->bi_error);
Miao Xie23ea8e52014-09-12 18:43:54 +08008080
8081 if (io_bio->end_io)
8082 io_bio->end_io(io_bio, err);
Chris Mason9be33952013-05-17 18:30:14 -04008083 bio_put(bio);
Josef Bacik4b46fce2010-05-23 11:00:55 -04008084}
8085
Filipe Manana14543772015-11-24 16:23:54 +00008086static void btrfs_endio_direct_write_update_ordered(struct inode *inode,
8087 const u64 offset,
8088 const u64 bytes,
8089 const int uptodate)
Josef Bacik4b46fce2010-05-23 11:00:55 -04008090{
Josef Bacik4b46fce2010-05-23 11:00:55 -04008091 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik4b46fce2010-05-23 11:00:55 -04008092 struct btrfs_ordered_extent *ordered = NULL;
Filipe Manana14543772015-11-24 16:23:54 +00008093 u64 ordered_offset = offset;
8094 u64 ordered_bytes = bytes;
Josef Bacik4b46fce2010-05-23 11:00:55 -04008095 int ret;
8096
Chris Mason163cf092010-11-28 19:56:33 -05008097again:
8098 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8099 &ordered_offset,
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02008100 ordered_bytes,
Filipe Manana14543772015-11-24 16:23:54 +00008101 uptodate);
Josef Bacik4b46fce2010-05-23 11:00:55 -04008102 if (!ret)
Chris Mason163cf092010-11-28 19:56:33 -05008103 goto out_test;
Josef Bacik4b46fce2010-05-23 11:00:55 -04008104
Liu Bo9e0af232014-08-15 23:36:53 +08008105 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8106 finish_ordered_fn, NULL, NULL);
Qu Wenruofccb5d82014-02-28 10:46:10 +08008107 btrfs_queue_work(root->fs_info->endio_write_workers,
8108 &ordered->work);
Chris Mason163cf092010-11-28 19:56:33 -05008109out_test:
8110 /*
8111 * our bio might span multiple ordered extents. If we haven't
8112 * completed the accounting for the whole dio, go back and try again
8113 */
Filipe Manana14543772015-11-24 16:23:54 +00008114 if (ordered_offset < offset + bytes) {
8115 ordered_bytes = offset + bytes - ordered_offset;
Josef Bacik5fd02042012-05-02 14:00:54 -04008116 ordered = NULL;
Chris Mason163cf092010-11-28 19:56:33 -05008117 goto again;
8118 }
Filipe Manana14543772015-11-24 16:23:54 +00008119}
8120
8121static void btrfs_endio_direct_write(struct bio *bio)
8122{
8123 struct btrfs_dio_private *dip = bio->bi_private;
8124 struct bio *dio_bio = dip->dio_bio;
8125
8126 btrfs_endio_direct_write_update_ordered(dip->inode,
8127 dip->logical_offset,
8128 dip->bytes,
8129 !bio->bi_error);
Josef Bacik4b46fce2010-05-23 11:00:55 -04008130
Josef Bacik4b46fce2010-05-23 11:00:55 -04008131 kfree(dip);
Josef Bacikc0da7aa2011-03-22 11:05:07 -04008132
Filipe Manana1636d1d2016-02-15 16:20:26 +00008133 dio_bio->bi_error = bio->bi_error;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02008134 dio_end_io(dio_bio, bio->bi_error);
Chris Mason9be33952013-05-17 18:30:14 -04008135 bio_put(bio);
Josef Bacik4b46fce2010-05-23 11:00:55 -04008136}
8137
Chris Masoneaf25d92010-05-25 09:48:28 -04008138static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
8139 struct bio *bio, int mirror_num,
8140 unsigned long bio_flags, u64 offset)
8141{
8142 int ret;
8143 struct btrfs_root *root = BTRFS_I(inode)->root;
8144 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01008145 BUG_ON(ret); /* -ENOMEM */
Chris Masoneaf25d92010-05-25 09:48:28 -04008146 return 0;
8147}
8148
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02008149static void btrfs_end_dio_bio(struct bio *bio)
Miao Xiee65e1532010-11-22 03:04:43 +00008150{
8151 struct btrfs_dio_private *dip = bio->bi_private;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02008152 int err = bio->bi_error;
Miao Xiee65e1532010-11-22 03:04:43 +00008153
Miao Xie8b110e32014-09-12 18:44:03 +08008154 if (err)
8155 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8156 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
8157 btrfs_ino(dip->inode), bio->bi_rw,
8158 (unsigned long long)bio->bi_iter.bi_sector,
8159 bio->bi_iter.bi_size, err);
8160
8161 if (dip->subio_endio)
8162 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
Miao Xiec1dc0892014-09-12 18:43:56 +08008163
8164 if (err) {
Miao Xiee65e1532010-11-22 03:04:43 +00008165 dip->errors = 1;
8166
8167 /*
8168 * before atomic variable goto zero, we must make sure
8169 * dip->errors is perceived to be set.
8170 */
Peter Zijlstra4e857c52014-03-17 18:06:10 +01008171 smp_mb__before_atomic();
Miao Xiee65e1532010-11-22 03:04:43 +00008172 }
8173
8174 /* if there are more bios still pending for this dio, just exit */
8175 if (!atomic_dec_and_test(&dip->pending_bios))
8176 goto out;
8177
Chris Mason9be33952013-05-17 18:30:14 -04008178 if (dip->errors) {
Miao Xiee65e1532010-11-22 03:04:43 +00008179 bio_io_error(dip->orig_bio);
Chris Mason9be33952013-05-17 18:30:14 -04008180 } else {
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02008181 dip->dio_bio->bi_error = 0;
8182 bio_endio(dip->orig_bio);
Miao Xiee65e1532010-11-22 03:04:43 +00008183 }
8184out:
8185 bio_put(bio);
8186}
8187
8188static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8189 u64 first_sector, gfp_t gfp_flags)
8190{
Chris Masonda2f0f72015-07-02 13:57:22 -07008191 struct bio *bio;
Linus Torvalds22365972015-09-05 15:14:43 -07008192 bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
Chris Masonda2f0f72015-07-02 13:57:22 -07008193 if (bio)
8194 bio_associate_current(bio);
8195 return bio;
Miao Xiee65e1532010-11-22 03:04:43 +00008196}
8197
Miao Xiec1dc0892014-09-12 18:43:56 +08008198static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
8199 struct inode *inode,
8200 struct btrfs_dio_private *dip,
8201 struct bio *bio,
8202 u64 file_offset)
8203{
8204 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8205 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8206 int ret;
8207
8208 /*
8209 * We load all the csum data we need when we submit
8210 * the first bio to reduce the csum tree search and
8211 * contention.
8212 */
8213 if (dip->logical_offset == file_offset) {
8214 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
8215 file_offset);
8216 if (ret)
8217 return ret;
8218 }
8219
8220 if (bio == dip->orig_bio)
8221 return 0;
8222
8223 file_offset -= dip->logical_offset;
8224 file_offset >>= inode->i_sb->s_blocksize_bits;
8225 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8226
8227 return 0;
8228}
8229
Miao Xiee65e1532010-11-22 03:04:43 +00008230static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8231 int rw, u64 file_offset, int skip_sum,
Josef Bacikc3298612012-08-03 16:49:19 -04008232 int async_submit)
Miao Xiee65e1532010-11-22 03:04:43 +00008233{
Miao Xiefacc8a222013-07-25 19:22:34 +08008234 struct btrfs_dio_private *dip = bio->bi_private;
Miao Xiee65e1532010-11-22 03:04:43 +00008235 int write = rw & REQ_WRITE;
8236 struct btrfs_root *root = BTRFS_I(inode)->root;
8237 int ret;
8238
Josef Bacikb812ce22012-11-16 13:56:32 -05008239 if (async_submit)
8240 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8241
Miao Xiee65e1532010-11-22 03:04:43 +00008242 bio_get(bio);
Josef Bacik5fd02042012-05-02 14:00:54 -04008243
8244 if (!write) {
David Sterbabfebd8b2014-07-30 00:25:45 +02008245 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8246 BTRFS_WQ_ENDIO_DATA);
Josef Bacik5fd02042012-05-02 14:00:54 -04008247 if (ret)
8248 goto err;
8249 }
Miao Xiee65e1532010-11-22 03:04:43 +00008250
Josef Bacik1ae39932011-04-06 14:41:34 -04008251 if (skip_sum)
8252 goto map;
8253
8254 if (write && async_submit) {
Miao Xiee65e1532010-11-22 03:04:43 +00008255 ret = btrfs_wq_submit_bio(root->fs_info,
8256 inode, rw, bio, 0, 0,
8257 file_offset,
8258 __btrfs_submit_bio_start_direct_io,
8259 __btrfs_submit_bio_done);
8260 goto err;
Josef Bacik1ae39932011-04-06 14:41:34 -04008261 } else if (write) {
8262 /*
8263 * If we aren't doing async submit, calculate the csum of the
8264 * bio now.
8265 */
8266 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8267 if (ret)
8268 goto err;
Miao Xie23ea8e52014-09-12 18:43:54 +08008269 } else {
Miao Xiec1dc0892014-09-12 18:43:56 +08008270 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8271 file_offset);
Tsutomu Itohc2db1072011-03-01 06:48:31 +00008272 if (ret)
8273 goto err;
8274 }
Josef Bacik1ae39932011-04-06 14:41:34 -04008275map:
8276 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
Miao Xiee65e1532010-11-22 03:04:43 +00008277err:
8278 bio_put(bio);
8279 return ret;
8280}
8281
8282static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
8283 int skip_sum)
8284{
8285 struct inode *inode = dip->inode;
8286 struct btrfs_root *root = BTRFS_I(inode)->root;
Miao Xiee65e1532010-11-22 03:04:43 +00008287 struct bio *bio;
8288 struct bio *orig_bio = dip->orig_bio;
8289 struct bio_vec *bvec = orig_bio->bi_io_vec;
Kent Overstreet4f024f32013-10-11 15:44:27 -07008290 u64 start_sector = orig_bio->bi_iter.bi_sector;
Miao Xiee65e1532010-11-22 03:04:43 +00008291 u64 file_offset = dip->logical_offset;
8292 u64 submit_len = 0;
8293 u64 map_length;
Chandan Rajendra5f4dc8f2016-01-21 15:56:00 +05308294 u32 blocksize = root->sectorsize;
Josef Bacik1ae39932011-04-06 14:41:34 -04008295 int async_submit = 0;
Chandan Rajendra5f4dc8f2016-01-21 15:56:00 +05308296 int nr_sectors;
8297 int ret;
8298 int i;
Miao Xiee65e1532010-11-22 03:04:43 +00008299
Kent Overstreet4f024f32013-10-11 15:44:27 -07008300 map_length = orig_bio->bi_iter.bi_size;
David Woodhouse53b381b2013-01-29 18:40:14 -05008301 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
Miao Xiee65e1532010-11-22 03:04:43 +00008302 &map_length, NULL, 0);
Miao Xie7a5c3c92014-06-17 18:58:59 +08008303 if (ret)
Miao Xiee65e1532010-11-22 03:04:43 +00008304 return -EIO;
Miao Xiefacc8a222013-07-25 19:22:34 +08008305
Kent Overstreet4f024f32013-10-11 15:44:27 -07008306 if (map_length >= orig_bio->bi_iter.bi_size) {
Josef Bacik02f57c72011-04-06 14:25:44 -04008307 bio = orig_bio;
Miao Xiec1dc0892014-09-12 18:43:56 +08008308 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
Josef Bacik02f57c72011-04-06 14:25:44 -04008309 goto submit;
8310 }
8311
David Woodhouse53b381b2013-01-29 18:40:14 -05008312 /* async crcs make it difficult to collect full stripe writes. */
Zhao Leiffe2d202015-01-20 15:11:44 +08008313 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
David Woodhouse53b381b2013-01-29 18:40:14 -05008314 async_submit = 0;
8315 else
8316 async_submit = 1;
8317
Josef Bacik02f57c72011-04-06 14:25:44 -04008318 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8319 if (!bio)
8320 return -ENOMEM;
Miao Xie7a5c3c92014-06-17 18:58:59 +08008321
Josef Bacik02f57c72011-04-06 14:25:44 -04008322 bio->bi_private = dip;
8323 bio->bi_end_io = btrfs_end_dio_bio;
Miao Xiec1dc0892014-09-12 18:43:56 +08008324 btrfs_io_bio(bio)->logical = file_offset;
Josef Bacik02f57c72011-04-06 14:25:44 -04008325 atomic_inc(&dip->pending_bios);
8326
Miao Xiee65e1532010-11-22 03:04:43 +00008327 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
Chandan Rajendra5f4dc8f2016-01-21 15:56:00 +05308328 nr_sectors = BTRFS_BYTES_TO_BLKS(root->fs_info, bvec->bv_len);
8329 i = 0;
8330next_block:
8331 if (unlikely(map_length < submit_len + blocksize ||
8332 bio_add_page(bio, bvec->bv_page, blocksize,
8333 bvec->bv_offset + (i * blocksize)) < blocksize)) {
Miao Xiee65e1532010-11-22 03:04:43 +00008334 /*
8335 * inc the count before we submit the bio so
8336 * we know the end IO handler won't happen before
8337 * we inc the count. Otherwise, the dip might get freed
8338 * before we're done setting it up
8339 */
8340 atomic_inc(&dip->pending_bios);
8341 ret = __btrfs_submit_dio_bio(bio, inode, rw,
8342 file_offset, skip_sum,
Josef Bacikc3298612012-08-03 16:49:19 -04008343 async_submit);
Miao Xiee65e1532010-11-22 03:04:43 +00008344 if (ret) {
8345 bio_put(bio);
8346 atomic_dec(&dip->pending_bios);
8347 goto out_err;
8348 }
8349
Miao Xiee65e1532010-11-22 03:04:43 +00008350 start_sector += submit_len >> 9;
8351 file_offset += submit_len;
8352
8353 submit_len = 0;
Miao Xiee65e1532010-11-22 03:04:43 +00008354
8355 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8356 start_sector, GFP_NOFS);
8357 if (!bio)
8358 goto out_err;
8359 bio->bi_private = dip;
8360 bio->bi_end_io = btrfs_end_dio_bio;
Miao Xiec1dc0892014-09-12 18:43:56 +08008361 btrfs_io_bio(bio)->logical = file_offset;
Miao Xiee65e1532010-11-22 03:04:43 +00008362
Kent Overstreet4f024f32013-10-11 15:44:27 -07008363 map_length = orig_bio->bi_iter.bi_size;
David Woodhouse53b381b2013-01-29 18:40:14 -05008364 ret = btrfs_map_block(root->fs_info, rw,
Stefan Behrens3ec706c2012-11-05 15:46:42 +01008365 start_sector << 9,
Miao Xiee65e1532010-11-22 03:04:43 +00008366 &map_length, NULL, 0);
8367 if (ret) {
8368 bio_put(bio);
8369 goto out_err;
8370 }
Chandan Rajendra5f4dc8f2016-01-21 15:56:00 +05308371
8372 goto next_block;
Miao Xiee65e1532010-11-22 03:04:43 +00008373 } else {
Chandan Rajendra5f4dc8f2016-01-21 15:56:00 +05308374 submit_len += blocksize;
8375 if (--nr_sectors) {
8376 i++;
8377 goto next_block;
8378 }
Miao Xiee65e1532010-11-22 03:04:43 +00008379 bvec++;
8380 }
8381 }
8382
Josef Bacik02f57c72011-04-06 14:25:44 -04008383submit:
Miao Xiee65e1532010-11-22 03:04:43 +00008384 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
Josef Bacikc3298612012-08-03 16:49:19 -04008385 async_submit);
Miao Xiee65e1532010-11-22 03:04:43 +00008386 if (!ret)
8387 return 0;
8388
8389 bio_put(bio);
8390out_err:
8391 dip->errors = 1;
8392 /*
8393 * before atomic variable goto zero, we must
8394 * make sure dip->errors is perceived to be set.
8395 */
Peter Zijlstra4e857c52014-03-17 18:06:10 +01008396 smp_mb__before_atomic();
Miao Xiee65e1532010-11-22 03:04:43 +00008397 if (atomic_dec_and_test(&dip->pending_bios))
8398 bio_io_error(dip->orig_bio);
8399
8400 /* bio_end_io() will handle error, so we needn't return it */
8401 return 0;
8402}
8403
Chris Mason9be33952013-05-17 18:30:14 -04008404static void btrfs_submit_direct(int rw, struct bio *dio_bio,
8405 struct inode *inode, loff_t file_offset)
Josef Bacik4b46fce2010-05-23 11:00:55 -04008406{
Filipe Manana61de7182015-07-01 12:13:10 +01008407 struct btrfs_dio_private *dip = NULL;
8408 struct bio *io_bio = NULL;
Miao Xie23ea8e52014-09-12 18:43:54 +08008409 struct btrfs_io_bio *btrfs_bio;
Josef Bacik4b46fce2010-05-23 11:00:55 -04008410 int skip_sum;
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02008411 int write = rw & REQ_WRITE;
Josef Bacik4b46fce2010-05-23 11:00:55 -04008412 int ret = 0;
8413
8414 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8415
Chris Mason9be33952013-05-17 18:30:14 -04008416 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
Chris Mason9be33952013-05-17 18:30:14 -04008417 if (!io_bio) {
Josef Bacik4b46fce2010-05-23 11:00:55 -04008418 ret = -ENOMEM;
8419 goto free_ordered;
8420 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04008421
Miao Xiec1dc0892014-09-12 18:43:56 +08008422 dip = kzalloc(sizeof(*dip), GFP_NOFS);
Chris Mason9be33952013-05-17 18:30:14 -04008423 if (!dip) {
8424 ret = -ENOMEM;
Filipe Manana61de7182015-07-01 12:13:10 +01008425 goto free_ordered;
Chris Mason9be33952013-05-17 18:30:14 -04008426 }
8427
8428 dip->private = dio_bio->bi_private;
Josef Bacik4b46fce2010-05-23 11:00:55 -04008429 dip->inode = inode;
8430 dip->logical_offset = file_offset;
Kent Overstreet4f024f32013-10-11 15:44:27 -07008431 dip->bytes = dio_bio->bi_iter.bi_size;
8432 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
Chris Mason9be33952013-05-17 18:30:14 -04008433 io_bio->bi_private = dip;
Chris Mason9be33952013-05-17 18:30:14 -04008434 dip->orig_bio = io_bio;
8435 dip->dio_bio = dio_bio;
Miao Xiee65e1532010-11-22 03:04:43 +00008436 atomic_set(&dip->pending_bios, 0);
Miao Xiec1dc0892014-09-12 18:43:56 +08008437 btrfs_bio = btrfs_io_bio(io_bio);
8438 btrfs_bio->logical = file_offset;
Josef Bacik4b46fce2010-05-23 11:00:55 -04008439
Miao Xiec1dc0892014-09-12 18:43:56 +08008440 if (write) {
Chris Mason9be33952013-05-17 18:30:14 -04008441 io_bio->bi_end_io = btrfs_endio_direct_write;
Miao Xiec1dc0892014-09-12 18:43:56 +08008442 } else {
Chris Mason9be33952013-05-17 18:30:14 -04008443 io_bio->bi_end_io = btrfs_endio_direct_read;
Miao Xiec1dc0892014-09-12 18:43:56 +08008444 dip->subio_endio = btrfs_subio_endio_read;
8445 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04008446
Filipe Mananaf28a4922015-12-08 19:23:20 +00008447 /*
8448 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8449 * even if we fail to submit a bio, because in such case we do the
8450 * corresponding error handling below and it must not be done a second
8451 * time by btrfs_direct_IO().
8452 */
8453 if (write) {
8454 struct btrfs_dio_data *dio_data = current->journal_info;
8455
8456 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8457 dip->bytes;
8458 dio_data->unsubmitted_oe_range_start =
8459 dio_data->unsubmitted_oe_range_end;
8460 }
8461
Miao Xiee65e1532010-11-22 03:04:43 +00008462 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8463 if (!ret)
Chris Masoneaf25d92010-05-25 09:48:28 -04008464 return;
Chris Mason9be33952013-05-17 18:30:14 -04008465
Miao Xie23ea8e52014-09-12 18:43:54 +08008466 if (btrfs_bio->end_io)
8467 btrfs_bio->end_io(btrfs_bio, ret);
Chris Mason9be33952013-05-17 18:30:14 -04008468
Josef Bacik4b46fce2010-05-23 11:00:55 -04008469free_ordered:
8470 /*
Filipe Manana61de7182015-07-01 12:13:10 +01008471 * If we arrived here it means either we failed to submit the dip
8472 * or we either failed to clone the dio_bio or failed to allocate the
8473 * dip. If we cloned the dio_bio and allocated the dip, we can just
8474 * call bio_endio against our io_bio so that we get proper resource
8475 * cleanup if we fail to submit the dip, otherwise, we must do the
8476 * same as btrfs_endio_direct_[write|read] because we can't call these
8477 * callbacks - they require an allocated dip and a clone of dio_bio.
Josef Bacik4b46fce2010-05-23 11:00:55 -04008478 */
Filipe Manana61de7182015-07-01 12:13:10 +01008479 if (io_bio && dip) {
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02008480 io_bio->bi_error = -EIO;
8481 bio_endio(io_bio);
Filipe Manana61de7182015-07-01 12:13:10 +01008482 /*
8483 * The end io callbacks free our dip, do the final put on io_bio
8484 * and all the cleanup and final put for dio_bio (through
8485 * dio_end_io()).
8486 */
8487 dip = NULL;
8488 io_bio = NULL;
8489 } else {
Filipe Manana14543772015-11-24 16:23:54 +00008490 if (write)
8491 btrfs_endio_direct_write_update_ordered(inode,
8492 file_offset,
8493 dio_bio->bi_iter.bi_size,
8494 0);
8495 else
Filipe Manana61de7182015-07-01 12:13:10 +01008496 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8497 file_offset + dio_bio->bi_iter.bi_size - 1);
Filipe Manana14543772015-11-24 16:23:54 +00008498
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02008499 dio_bio->bi_error = -EIO;
Filipe Manana61de7182015-07-01 12:13:10 +01008500 /*
8501 * Releases and cleans up our dio_bio, no need to bio_put()
8502 * nor bio_endio()/bio_io_error() against dio_bio.
8503 */
8504 dio_end_io(dio_bio, ret);
Josef Bacik4b46fce2010-05-23 11:00:55 -04008505 }
Filipe Manana61de7182015-07-01 12:13:10 +01008506 if (io_bio)
8507 bio_put(io_bio);
8508 kfree(dip);
Josef Bacik4b46fce2010-05-23 11:00:55 -04008509}
8510
Omar Sandoval6f673762015-03-16 04:33:52 -07008511static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
Al Viro28060d52014-03-22 05:15:17 -04008512 const struct iov_iter *iter, loff_t offset)
Chris Mason5a5f79b2010-05-26 21:33:37 -04008513{
8514 int seg;
Josef Bacika1b75f72011-04-08 15:51:18 +00008515 int i;
Chris Mason5a5f79b2010-05-26 21:33:37 -04008516 unsigned blocksize_mask = root->sectorsize - 1;
8517 ssize_t retval = -EINVAL;
Chris Mason5a5f79b2010-05-26 21:33:37 -04008518
8519 if (offset & blocksize_mask)
8520 goto out;
8521
Al Viro28060d52014-03-22 05:15:17 -04008522 if (iov_iter_alignment(iter) & blocksize_mask)
8523 goto out;
Josef Bacika1b75f72011-04-08 15:51:18 +00008524
Al Viro28060d52014-03-22 05:15:17 -04008525 /* If this is a write we don't need to check anymore */
Omar Sandoval6f673762015-03-16 04:33:52 -07008526 if (iov_iter_rw(iter) == WRITE)
Al Viro28060d52014-03-22 05:15:17 -04008527 return 0;
8528 /*
8529 * Check to make sure we don't have duplicate iov_base's in this
8530 * iovec, if so return EINVAL, otherwise we'll get csum errors
8531 * when reading back.
8532 */
8533 for (seg = 0; seg < iter->nr_segs; seg++) {
8534 for (i = seg + 1; i < iter->nr_segs; i++) {
8535 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
Josef Bacika1b75f72011-04-08 15:51:18 +00008536 goto out;
8537 }
Chris Mason5a5f79b2010-05-26 21:33:37 -04008538 }
8539 retval = 0;
8540out:
8541 return retval;
8542}
Josef Bacikeb838e72012-07-31 16:28:48 -04008543
Omar Sandoval22c61862015-03-16 04:33:53 -07008544static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
8545 loff_t offset)
Chris Mason16432982008-04-10 10:23:21 -04008546{
Josef Bacik4b46fce2010-05-23 11:00:55 -04008547 struct file *file = iocb->ki_filp;
8548 struct inode *inode = file->f_mapping->host;
chandan50745b02015-08-28 21:10:13 +05308549 struct btrfs_root *root = BTRFS_I(inode)->root;
8550 struct btrfs_dio_data dio_data = { 0 };
Miao Xie09348562013-02-07 10:12:07 +00008551 size_t count = 0;
Miao Xie2e60a512013-02-08 07:01:08 +00008552 int flags = 0;
Miao Xie38851cc2013-02-08 07:04:11 +00008553 bool wakeup = true;
8554 bool relock = false;
Miao Xie09348562013-02-07 10:12:07 +00008555 ssize_t ret;
Josef Bacik4b46fce2010-05-23 11:00:55 -04008556
Omar Sandoval6f673762015-03-16 04:33:52 -07008557 if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
Chris Mason5a5f79b2010-05-26 21:33:37 -04008558 return 0;
Chris Mason5a5f79b2010-05-26 21:33:37 -04008559
Jens Axboefe0f07d2015-04-15 17:05:48 -06008560 inode_dio_begin(inode);
Peter Zijlstra4e857c52014-03-17 18:06:10 +01008561 smp_mb__after_atomic();
Miao Xie38851cc2013-02-08 07:04:11 +00008562
Josef Bacik0e267c42013-07-02 10:38:02 -04008563 /*
Miao Xie41bd9ca2014-03-06 13:54:57 +08008564 * The generic stuff only does filemap_write_and_wait_range, which
8565 * isn't enough if we've written compressed pages to this area, so
8566 * we need to flush the dirty pages again to make absolutely sure
8567 * that any outstanding dirty pages are on disk.
Josef Bacik0e267c42013-07-02 10:38:02 -04008568 */
Al Viroa6cbcd42014-03-04 22:38:00 -05008569 count = iov_iter_count(iter);
Miao Xie41bd9ca2014-03-06 13:54:57 +08008570 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8571 &BTRFS_I(inode)->runtime_flags))
Wang Shilong9a025a02014-07-17 11:44:13 +08008572 filemap_fdatawrite_range(inode->i_mapping, offset,
8573 offset + count - 1);
Josef Bacik0e267c42013-07-02 10:38:02 -04008574
Omar Sandoval6f673762015-03-16 04:33:52 -07008575 if (iov_iter_rw(iter) == WRITE) {
Miao Xie38851cc2013-02-08 07:04:11 +00008576 /*
8577 * If the write DIO is beyond the EOF, we need update
8578 * the isize, but it is protected by i_mutex. So we can
8579 * not unlock the i_mutex at this case.
8580 */
8581 if (offset + count <= inode->i_size) {
Al Viro59551022016-01-22 15:40:57 -05008582 inode_unlock(inode);
Miao Xie38851cc2013-02-08 07:04:11 +00008583 relock = true;
8584 }
Qu Wenruo7cf5b972015-09-08 17:25:55 +08008585 ret = btrfs_delalloc_reserve_space(inode, offset, count);
Miao Xie09348562013-02-07 10:12:07 +00008586 if (ret)
Miao Xie38851cc2013-02-08 07:04:11 +00008587 goto out;
chandan50745b02015-08-28 21:10:13 +05308588 dio_data.outstanding_extents = div64_u64(count +
Josef Bacike1cbbfa2015-03-17 10:52:28 -04008589 BTRFS_MAX_EXTENT_SIZE - 1,
8590 BTRFS_MAX_EXTENT_SIZE);
8591
8592 /*
8593 * We need to know how many extents we reserved so that we can
8594 * do the accounting properly if we go over the number we
8595 * originally calculated. Abuse current->journal_info for this.
8596 */
chandan50745b02015-08-28 21:10:13 +05308597 dio_data.reserve = round_up(count, root->sectorsize);
Filipe Mananaf28a4922015-12-08 19:23:20 +00008598 dio_data.unsubmitted_oe_range_start = (u64)offset;
8599 dio_data.unsubmitted_oe_range_end = (u64)offset;
chandan50745b02015-08-28 21:10:13 +05308600 current->journal_info = &dio_data;
David Sterbaee39b432014-09-30 01:33:33 +02008601 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8602 &BTRFS_I(inode)->runtime_flags)) {
Jens Axboefe0f07d2015-04-15 17:05:48 -06008603 inode_dio_end(inode);
Miao Xie38851cc2013-02-08 07:04:11 +00008604 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8605 wakeup = false;
Miao Xie09348562013-02-07 10:12:07 +00008606 }
8607
Omar Sandoval17f8c842015-03-16 04:33:50 -07008608 ret = __blockdev_direct_IO(iocb, inode,
8609 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8610 iter, offset, btrfs_get_blocks_direct, NULL,
8611 btrfs_submit_direct, flags);
Omar Sandoval6f673762015-03-16 04:33:52 -07008612 if (iov_iter_rw(iter) == WRITE) {
Josef Bacike1cbbfa2015-03-17 10:52:28 -04008613 current->journal_info = NULL;
Liu Boddba1bf2015-06-17 16:59:58 +08008614 if (ret < 0 && ret != -EIOCBQUEUED) {
chandan50745b02015-08-28 21:10:13 +05308615 if (dio_data.reserve)
Qu Wenruo7cf5b972015-09-08 17:25:55 +08008616 btrfs_delalloc_release_space(inode, offset,
8617 dio_data.reserve);
Filipe Mananaf28a4922015-12-08 19:23:20 +00008618 /*
8619 * On error we might have left some ordered extents
8620 * without submitting corresponding bios for them, so
8621 * cleanup them up to avoid other tasks getting them
8622 * and waiting for them to complete forever.
8623 */
8624 if (dio_data.unsubmitted_oe_range_start <
8625 dio_data.unsubmitted_oe_range_end)
8626 btrfs_endio_direct_write_update_ordered(inode,
8627 dio_data.unsubmitted_oe_range_start,
8628 dio_data.unsubmitted_oe_range_end -
8629 dio_data.unsubmitted_oe_range_start,
8630 0);
Liu Boddba1bf2015-06-17 16:59:58 +08008631 } else if (ret >= 0 && (size_t)ret < count)
Qu Wenruo7cf5b972015-09-08 17:25:55 +08008632 btrfs_delalloc_release_space(inode, offset,
8633 count - (size_t)ret);
Miao Xie09348562013-02-07 10:12:07 +00008634 }
Miao Xie38851cc2013-02-08 07:04:11 +00008635out:
Miao Xie2e60a512013-02-08 07:01:08 +00008636 if (wakeup)
Jens Axboefe0f07d2015-04-15 17:05:48 -06008637 inode_dio_end(inode);
Miao Xie38851cc2013-02-08 07:04:11 +00008638 if (relock)
Al Viro59551022016-01-22 15:40:57 -05008639 inode_lock(inode);
Miao Xie09348562013-02-07 10:12:07 +00008640
8641 return ret;
Chris Mason16432982008-04-10 10:23:21 -04008642}
8643
Tsutomu Itoh05dadc02012-11-29 05:08:26 +00008644#define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8645
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -05008646static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8647 __u64 start, __u64 len)
8648{
Tsutomu Itoh05dadc02012-11-29 05:08:26 +00008649 int ret;
8650
8651 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8652 if (ret)
8653 return ret;
8654
Chris Masonec29ed52011-02-23 16:23:20 -05008655 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -05008656}
8657
Chris Mason9ebefb182007-06-15 13:50:00 -04008658int btrfs_readpage(struct file *file, struct page *page)
8659{
Chris Masond1310b22008-01-24 16:13:08 -05008660 struct extent_io_tree *tree;
8661 tree = &BTRFS_I(page->mapping->host)->io_tree;
Jan Schmidt8ddc7d92011-06-13 20:02:58 +02008662 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04008663}
Chris Mason1832a6d2007-12-21 16:27:21 -05008664
Chris Mason39279cc2007-06-12 06:35:45 -04008665static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8666{
Chris Masond1310b22008-01-24 16:13:08 -05008667 struct extent_io_tree *tree;
Josef Bacikbe7bd732015-10-22 15:05:09 -04008668 struct inode *inode = page->mapping->host;
8669 int ret;
Chris Masonb888db22007-08-27 16:49:44 -04008670
8671 if (current->flags & PF_MEMALLOC) {
8672 redirty_page_for_writepage(wbc, page);
8673 unlock_page(page);
8674 return 0;
8675 }
Josef Bacikbe7bd732015-10-22 15:05:09 -04008676
8677 /*
8678 * If we are under memory pressure we will call this directly from the
8679 * VM, we need to make sure we have the inode referenced for the ordered
8680 * extent. If not just return like we didn't do anything.
8681 */
8682 if (!igrab(inode)) {
8683 redirty_page_for_writepage(wbc, page);
8684 return AOP_WRITEPAGE_ACTIVATE;
8685 }
Chris Masond1310b22008-01-24 16:13:08 -05008686 tree = &BTRFS_I(page->mapping->host)->io_tree;
Josef Bacikbe7bd732015-10-22 15:05:09 -04008687 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8688 btrfs_add_delayed_iput(inode);
8689 return ret;
Chris Masona52d9a82007-08-27 16:49:44 -04008690}
Chris Mason39279cc2007-06-12 06:35:45 -04008691
Eric Sandeen48a3b632013-04-25 20:41:01 +00008692static int btrfs_writepages(struct address_space *mapping,
8693 struct writeback_control *wbc)
Chris Masonb293f022007-11-01 19:45:34 -04008694{
Chris Masond1310b22008-01-24 16:13:08 -05008695 struct extent_io_tree *tree;
Chris Mason771ed682008-11-06 22:02:51 -05008696
Chris Masond1310b22008-01-24 16:13:08 -05008697 tree = &BTRFS_I(mapping->host)->io_tree;
Chris Masonb293f022007-11-01 19:45:34 -04008698 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8699}
8700
Chris Mason3ab2fb52007-11-08 10:59:22 -05008701static int
8702btrfs_readpages(struct file *file, struct address_space *mapping,
8703 struct list_head *pages, unsigned nr_pages)
8704{
Chris Masond1310b22008-01-24 16:13:08 -05008705 struct extent_io_tree *tree;
8706 tree = &BTRFS_I(mapping->host)->io_tree;
Chris Mason3ab2fb52007-11-08 10:59:22 -05008707 return extent_readpages(tree, mapping, pages, nr_pages,
8708 btrfs_get_extent);
8709}
Chris Masone6dcd2d2008-07-17 12:53:50 -04008710static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
Chris Masona52d9a82007-08-27 16:49:44 -04008711{
Chris Masond1310b22008-01-24 16:13:08 -05008712 struct extent_io_tree *tree;
8713 struct extent_map_tree *map;
Chris Masona52d9a82007-08-27 16:49:44 -04008714 int ret;
Chris Mason39279cc2007-06-12 06:35:45 -04008715
Chris Masond1310b22008-01-24 16:13:08 -05008716 tree = &BTRFS_I(page->mapping->host)->io_tree;
8717 map = &BTRFS_I(page->mapping->host)->extent_tree;
Chris Mason70dec802008-01-29 09:59:12 -05008718 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
Chris Masona52d9a82007-08-27 16:49:44 -04008719 if (ret == 1) {
8720 ClearPagePrivate(page);
8721 set_page_private(page, 0);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03008722 put_page(page);
Chris Masona52d9a82007-08-27 16:49:44 -04008723 }
8724 return ret;
8725}
Chris Mason39279cc2007-06-12 06:35:45 -04008726
Chris Masone6dcd2d2008-07-17 12:53:50 -04008727static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8728{
Chris Mason98509cf2008-09-11 15:51:43 -04008729 if (PageWriteback(page) || PageDirty(page))
8730 return 0;
Yan Zhengb335b002009-02-12 10:06:04 -05008731 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04008732}
8733
Lukas Czernerd47992f2013-05-21 23:17:23 -04008734static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8735 unsigned int length)
Chris Masona52d9a82007-08-27 16:49:44 -04008736{
Josef Bacik5fd02042012-05-02 14:00:54 -04008737 struct inode *inode = page->mapping->host;
Chris Masond1310b22008-01-24 16:13:08 -05008738 struct extent_io_tree *tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -04008739 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00008740 struct extent_state *cached_state = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04008741 u64 page_start = page_offset(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03008742 u64 page_end = page_start + PAGE_SIZE - 1;
Chandan Rajendradbfdb6d2016-01-21 15:55:58 +05308743 u64 start;
8744 u64 end;
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00008745 int inode_evicting = inode->i_state & I_FREEING;
Chris Masona52d9a82007-08-27 16:49:44 -04008746
Chris Mason8b62b722009-09-02 16:53:46 -04008747 /*
8748 * we have the page locked, so new writeback can't start,
8749 * and the dirty bit won't be cleared while we are here.
8750 *
8751 * Wait for IO on this page so that we can safely clear
8752 * the PagePrivate2 bit and do ordered accounting
8753 */
Chris Masone6dcd2d2008-07-17 12:53:50 -04008754 wait_on_page_writeback(page);
Chris Mason8b62b722009-09-02 16:53:46 -04008755
Josef Bacik5fd02042012-05-02 14:00:54 -04008756 tree = &BTRFS_I(inode)->io_tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -04008757 if (offset) {
8758 btrfs_releasepage(page, GFP_NOFS);
8759 return;
8760 }
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00008761
8762 if (!inode_evicting)
David Sterbaff13db42015-12-03 14:30:40 +01008763 lock_extent_bits(tree, page_start, page_end, &cached_state);
Chandan Rajendradbfdb6d2016-01-21 15:55:58 +05308764again:
8765 start = page_start;
8766 ordered = btrfs_lookup_ordered_range(inode, start,
8767 page_end - start + 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04008768 if (ordered) {
Chandan Rajendradbfdb6d2016-01-21 15:55:58 +05308769 end = min(page_end, ordered->file_offset + ordered->len - 1);
Chris Masoneb84ae02008-07-17 13:53:27 -04008770 /*
8771 * IO on this page will never be started, so we need
8772 * to account for any ordered extents now
8773 */
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00008774 if (!inode_evicting)
Chandan Rajendradbfdb6d2016-01-21 15:55:58 +05308775 clear_extent_bit(tree, start, end,
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00008776 EXTENT_DIRTY | EXTENT_DELALLOC |
8777 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8778 EXTENT_DEFRAG, 1, 0, &cached_state,
8779 GFP_NOFS);
Chris Mason8b62b722009-09-02 16:53:46 -04008780 /*
8781 * whoever cleared the private bit is responsible
8782 * for the finish_ordered_io
8783 */
Josef Bacik77cef2e2013-08-29 13:57:21 -04008784 if (TestClearPagePrivate2(page)) {
8785 struct btrfs_ordered_inode_tree *tree;
8786 u64 new_len;
8787
8788 tree = &BTRFS_I(inode)->ordered_tree;
8789
8790 spin_lock_irq(&tree->lock);
8791 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
Chandan Rajendradbfdb6d2016-01-21 15:55:58 +05308792 new_len = start - ordered->file_offset;
Josef Bacik77cef2e2013-08-29 13:57:21 -04008793 if (new_len < ordered->truncated_len)
8794 ordered->truncated_len = new_len;
8795 spin_unlock_irq(&tree->lock);
8796
8797 if (btrfs_dec_test_ordered_pending(inode, &ordered,
Chandan Rajendradbfdb6d2016-01-21 15:55:58 +05308798 start,
8799 end - start + 1, 1))
Josef Bacik77cef2e2013-08-29 13:57:21 -04008800 btrfs_finish_ordered_io(ordered);
Chris Mason8b62b722009-09-02 16:53:46 -04008801 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04008802 btrfs_put_ordered_extent(ordered);
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00008803 if (!inode_evicting) {
8804 cached_state = NULL;
Chandan Rajendradbfdb6d2016-01-21 15:55:58 +05308805 lock_extent_bits(tree, start, end,
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00008806 &cached_state);
8807 }
Chandan Rajendradbfdb6d2016-01-21 15:55:58 +05308808
8809 start = end + 1;
8810 if (start < page_end)
8811 goto again;
Chris Masone6dcd2d2008-07-17 12:53:50 -04008812 }
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00008813
Qu Wenruob9d0b382015-09-29 10:35:16 +08008814 /*
8815 * Qgroup reserved space handler
8816 * Page here will be either
8817 * 1) Already written to disk
8818 * In this case, its reserved space is released from data rsv map
8819 * and will be freed by delayed_ref handler finally.
8820 * So even we call qgroup_free_data(), it won't decrease reserved
8821 * space.
8822 * 2) Not written to disk
8823 * This means the reserved space should be freed here.
8824 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03008825 btrfs_qgroup_free_data(inode, page_start, PAGE_SIZE);
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00008826 if (!inode_evicting) {
8827 clear_extent_bit(tree, page_start, page_end,
8828 EXTENT_LOCKED | EXTENT_DIRTY |
8829 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8830 EXTENT_DEFRAG, 1, 1,
8831 &cached_state, GFP_NOFS);
8832
8833 __btrfs_releasepage(page, GFP_NOFS);
8834 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04008835
Chris Mason4a096752008-07-21 10:29:44 -04008836 ClearPageChecked(page);
Chris Mason9ad6b7b2008-04-18 16:11:30 -04008837 if (PagePrivate(page)) {
Chris Mason9ad6b7b2008-04-18 16:11:30 -04008838 ClearPagePrivate(page);
8839 set_page_private(page, 0);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03008840 put_page(page);
Chris Mason9ad6b7b2008-04-18 16:11:30 -04008841 }
Chris Mason39279cc2007-06-12 06:35:45 -04008842}
8843
Chris Mason9ebefb182007-06-15 13:50:00 -04008844/*
8845 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8846 * called from a page fault handler when a page is first dirtied. Hence we must
8847 * be careful to check for EOF conditions here. We set the page up correctly
8848 * for a written page which means we get ENOSPC checking when writing into
8849 * holes and correct delalloc and unwritten extent mapping on filesystems that
8850 * support these features.
8851 *
8852 * We are not allowed to take the i_mutex here so we have to play games to
8853 * protect against truncate races as the page could now be beyond EOF. Because
8854 * vmtruncate() writes the inode size before removing pages, once we have the
8855 * page lock we can determine safely if the page is beyond EOF. If it is not
8856 * beyond EOF, then the page is guaranteed safe against truncation until we
8857 * unlock the page.
8858 */
Nick Pigginc2ec1752009-03-31 15:23:21 -07008859int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Chris Mason9ebefb182007-06-15 13:50:00 -04008860{
Nick Pigginc2ec1752009-03-31 15:23:21 -07008861 struct page *page = vmf->page;
Al Viro496ad9a2013-01-23 17:07:38 -05008862 struct inode *inode = file_inode(vma->vm_file);
Chris Mason1832a6d2007-12-21 16:27:21 -05008863 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masone6dcd2d2008-07-17 12:53:50 -04008864 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8865 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00008866 struct extent_state *cached_state = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04008867 char *kaddr;
8868 unsigned long zero_start;
Chris Mason9ebefb182007-06-15 13:50:00 -04008869 loff_t size;
Chris Mason1832a6d2007-12-21 16:27:21 -05008870 int ret;
Chris Mason9998eb72012-01-25 13:47:40 -05008871 int reserved = 0;
Chandan Rajendrad0b7da882016-01-21 15:55:57 +05308872 u64 reserved_space;
Chris Masona52d9a82007-08-27 16:49:44 -04008873 u64 page_start;
Chris Masone6dcd2d2008-07-17 12:53:50 -04008874 u64 page_end;
Chandan Rajendrad0b7da882016-01-21 15:55:57 +05308875 u64 end;
8876
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03008877 reserved_space = PAGE_SIZE;
Chris Mason9ebefb182007-06-15 13:50:00 -04008878
Jan Karab2b5ef52012-06-12 16:20:45 +02008879 sb_start_pagefault(inode->i_sb);
Qu Wenruodf480632015-09-08 17:25:54 +08008880 page_start = page_offset(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03008881 page_end = page_start + PAGE_SIZE - 1;
Chandan Rajendrad0b7da882016-01-21 15:55:57 +05308882 end = page_end;
Qu Wenruodf480632015-09-08 17:25:54 +08008883
Chandan Rajendrad0b7da882016-01-21 15:55:57 +05308884 /*
8885 * Reserving delalloc space after obtaining the page lock can lead to
8886 * deadlock. For example, if a dirty page is locked by this function
8887 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8888 * dirty page write out, then the btrfs_writepage() function could
8889 * end up waiting indefinitely to get a lock on the page currently
8890 * being processed by btrfs_page_mkwrite() function.
8891 */
Qu Wenruo7cf5b972015-09-08 17:25:55 +08008892 ret = btrfs_delalloc_reserve_space(inode, page_start,
Chandan Rajendrad0b7da882016-01-21 15:55:57 +05308893 reserved_space);
Chris Mason9998eb72012-01-25 13:47:40 -05008894 if (!ret) {
Josef Bacike41f9412012-03-26 09:46:47 -04008895 ret = file_update_time(vma->vm_file);
Chris Mason9998eb72012-01-25 13:47:40 -05008896 reserved = 1;
8897 }
Nick Piggin56a76f82009-03-31 15:23:23 -07008898 if (ret) {
8899 if (ret == -ENOMEM)
8900 ret = VM_FAULT_OOM;
8901 else /* -ENOSPC, -EIO, etc */
8902 ret = VM_FAULT_SIGBUS;
Chris Mason9998eb72012-01-25 13:47:40 -05008903 if (reserved)
8904 goto out;
8905 goto out_noreserve;
Nick Piggin56a76f82009-03-31 15:23:23 -07008906 }
Chris Mason1832a6d2007-12-21 16:27:21 -05008907
Nick Piggin56a76f82009-03-31 15:23:23 -07008908 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
Chris Masone6dcd2d2008-07-17 12:53:50 -04008909again:
Chris Mason9ebefb182007-06-15 13:50:00 -04008910 lock_page(page);
Chris Mason9ebefb182007-06-15 13:50:00 -04008911 size = i_size_read(inode);
Chris Masona52d9a82007-08-27 16:49:44 -04008912
Chris Mason9ebefb182007-06-15 13:50:00 -04008913 if ((page->mapping != inode->i_mapping) ||
Chris Masone6dcd2d2008-07-17 12:53:50 -04008914 (page_start >= size)) {
Chris Mason9ebefb182007-06-15 13:50:00 -04008915 /* page got truncated out from underneath us */
8916 goto out_unlock;
8917 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04008918 wait_on_page_writeback(page);
8919
David Sterbaff13db42015-12-03 14:30:40 +01008920 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
Chris Masone6dcd2d2008-07-17 12:53:50 -04008921 set_page_extent_mapped(page);
8922
Chris Masoneb84ae02008-07-17 13:53:27 -04008923 /*
8924 * we can't set the delalloc bits if there are pending ordered
8925 * extents. Drop our locks and wait for them to finish
8926 */
Chandan Rajendrad0b7da882016-01-21 15:55:57 +05308927 ordered = btrfs_lookup_ordered_range(inode, page_start, page_end);
Chris Masone6dcd2d2008-07-17 12:53:50 -04008928 if (ordered) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00008929 unlock_extent_cached(io_tree, page_start, page_end,
8930 &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04008931 unlock_page(page);
Chris Masoneb84ae02008-07-17 13:53:27 -04008932 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04008933 btrfs_put_ordered_extent(ordered);
8934 goto again;
8935 }
8936
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03008937 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
Chandan Rajendrad0b7da882016-01-21 15:55:57 +05308938 reserved_space = round_up(size - page_start, root->sectorsize);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03008939 if (reserved_space < PAGE_SIZE) {
Chandan Rajendrad0b7da882016-01-21 15:55:57 +05308940 end = page_start + reserved_space - 1;
8941 spin_lock(&BTRFS_I(inode)->lock);
8942 BTRFS_I(inode)->outstanding_extents++;
8943 spin_unlock(&BTRFS_I(inode)->lock);
8944 btrfs_delalloc_release_space(inode, page_start,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03008945 PAGE_SIZE - reserved_space);
Chandan Rajendrad0b7da882016-01-21 15:55:57 +05308946 }
8947 }
8948
Josef Bacikfbf19082009-10-01 17:10:23 -04008949 /*
8950 * XXX - page_mkwrite gets called every time the page is dirtied, even
8951 * if it was already dirty, so for space accounting reasons we need to
8952 * clear any delalloc bits for the range we are fixing to save. There
8953 * is probably a better way to do this, but for now keep consistent with
8954 * prepare_pages in the normal write path.
8955 */
Chandan Rajendrad0b7da882016-01-21 15:55:57 +05308956 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
Liu Bo9e8a4a82012-09-05 19:10:51 -06008957 EXTENT_DIRTY | EXTENT_DELALLOC |
8958 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
Josef Bacik2ac55d42010-02-03 19:33:23 +00008959 0, 0, &cached_state, GFP_NOFS);
Josef Bacikfbf19082009-10-01 17:10:23 -04008960
Chandan Rajendrad0b7da882016-01-21 15:55:57 +05308961 ret = btrfs_set_extent_delalloc(inode, page_start, end,
Josef Bacik2ac55d42010-02-03 19:33:23 +00008962 &cached_state);
Josef Bacik9ed74f22009-09-11 16:12:44 -04008963 if (ret) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00008964 unlock_extent_cached(io_tree, page_start, page_end,
8965 &cached_state, GFP_NOFS);
Josef Bacik9ed74f22009-09-11 16:12:44 -04008966 ret = VM_FAULT_SIGBUS;
8967 goto out_unlock;
8968 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04008969 ret = 0;
Chris Mason9ebefb182007-06-15 13:50:00 -04008970
8971 /* page is wholly or partially inside EOF */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03008972 if (page_start + PAGE_SIZE > size)
8973 zero_start = size & ~PAGE_MASK;
Chris Mason9ebefb182007-06-15 13:50:00 -04008974 else
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03008975 zero_start = PAGE_SIZE;
Chris Mason9ebefb182007-06-15 13:50:00 -04008976
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03008977 if (zero_start != PAGE_SIZE) {
Chris Masone6dcd2d2008-07-17 12:53:50 -04008978 kaddr = kmap(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03008979 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
Chris Masone6dcd2d2008-07-17 12:53:50 -04008980 flush_dcache_page(page);
8981 kunmap(page);
8982 }
Chris Mason247e7432008-07-17 12:53:51 -04008983 ClearPageChecked(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04008984 set_page_dirty(page);
Chris Mason50a9b212009-09-11 12:33:12 -04008985 SetPageUptodate(page);
Chris Mason5a3f23d2009-03-31 13:27:11 -04008986
Chris Mason257c62e2009-10-13 13:21:08 -04008987 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8988 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
Liu Bo46d8bc32012-08-29 01:07:55 -06008989 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
Chris Mason257c62e2009-10-13 13:21:08 -04008990
Josef Bacik2ac55d42010-02-03 19:33:23 +00008991 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
Chris Mason9ebefb182007-06-15 13:50:00 -04008992
8993out_unlock:
Jan Karab2b5ef52012-06-12 16:20:45 +02008994 if (!ret) {
8995 sb_end_pagefault(inode->i_sb);
Chris Mason50a9b212009-09-11 12:33:12 -04008996 return VM_FAULT_LOCKED;
Jan Karab2b5ef52012-06-12 16:20:45 +02008997 }
Chris Mason9ebefb182007-06-15 13:50:00 -04008998 unlock_page(page);
Chris Mason1832a6d2007-12-21 16:27:21 -05008999out:
Chandan Rajendrad0b7da882016-01-21 15:55:57 +05309000 btrfs_delalloc_release_space(inode, page_start, reserved_space);
Chris Mason9998eb72012-01-25 13:47:40 -05009001out_noreserve:
Jan Karab2b5ef52012-06-12 16:20:45 +02009002 sb_end_pagefault(inode->i_sb);
Chris Mason9ebefb182007-06-15 13:50:00 -04009003 return ret;
9004}
9005
Josef Bacika41ad392011-01-31 15:30:16 -05009006static int btrfs_truncate(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04009007{
9008 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacikfcb80c22011-05-03 10:40:22 -04009009 struct btrfs_block_rsv *rsv;
Josef Bacika71754f2013-06-17 17:14:39 -04009010 int ret = 0;
Josef Bacik3893e332011-01-31 16:03:11 -05009011 int err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04009012 struct btrfs_trans_handle *trans;
Chris Masondbe674a2008-07-17 12:54:05 -04009013 u64 mask = root->sectorsize - 1;
Josef Bacik07127182011-08-19 10:29:59 -04009014 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
Chris Mason39279cc2007-06-12 06:35:45 -04009015
Josef Bacik0ef8b722013-10-25 16:13:35 -04009016 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9017 (u64)-1);
9018 if (ret)
9019 return ret;
Chris Mason39279cc2007-06-12 06:35:45 -04009020
Josef Bacikfcb80c22011-05-03 10:40:22 -04009021 /*
9022 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
9023 * 3 things going on here
9024 *
9025 * 1) We need to reserve space for our orphan item and the space to
9026 * delete our orphan item. Lord knows we don't want to have a dangling
9027 * orphan item because we didn't reserve space to remove it.
9028 *
9029 * 2) We need to reserve space to update our inode.
9030 *
9031 * 3) We need to have something to cache all the space that is going to
9032 * be free'd up by the truncate operation, but also have some slack
9033 * space reserved in case it uses space during the truncate (thank you
9034 * very much snapshotting).
9035 *
9036 * And we need these to all be seperate. The fact is we can use alot of
9037 * space doing the truncate, and we have no earthly idea how much space
9038 * we will use, so we need the truncate reservation to be seperate so it
9039 * doesn't end up using space reserved for updating the inode or
9040 * removing the orphan item. We also need to be able to stop the
9041 * transaction and start a new one, which means we need to be able to
9042 * update the inode several times, and we have no idea of knowing how
9043 * many times that will be, so we can't just reserve 1 item for the
9044 * entirety of the opration, so that has to be done seperately as well.
9045 * Then there is the orphan item, which does indeed need to be held on
9046 * to for the whole operation, and we need nobody to touch this reserved
9047 * space except the orphan code.
9048 *
9049 * So that leaves us with
9050 *
9051 * 1) root->orphan_block_rsv - for the orphan deletion.
9052 * 2) rsv - for the truncate reservation, which we will steal from the
9053 * transaction reservation.
9054 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9055 * updating the inode.
9056 */
Miao Xie66d8f3d2012-09-06 04:02:28 -06009057 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
Josef Bacikfcb80c22011-05-03 10:40:22 -04009058 if (!rsv)
9059 return -ENOMEM;
Josef Bacik4a338542011-08-29 11:01:31 -04009060 rsv->size = min_size;
Josef Bacikca7e70f2012-08-27 17:48:15 -04009061 rsv->failfast = 1;
Josef Bacikf0cd8462011-03-04 14:37:08 -05009062
Josef Bacik907cbce2011-08-08 13:46:15 -04009063 /*
Josef Bacik07127182011-08-19 10:29:59 -04009064 * 1 for the truncate slack space
Josef Bacik907cbce2011-08-08 13:46:15 -04009065 * 1 for updating the inode.
9066 */
Josef Bacikf3fe8202013-01-07 17:03:21 -05009067 trans = btrfs_start_transaction(root, 2);
Josef Bacikfcb80c22011-05-03 10:40:22 -04009068 if (IS_ERR(trans)) {
9069 err = PTR_ERR(trans);
9070 goto out;
9071 }
Josef Bacikf0cd8462011-03-04 14:37:08 -05009072
Josef Bacik907cbce2011-08-08 13:46:15 -04009073 /* Migrate the slack space for the truncate to our reserve */
9074 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
9075 min_size);
Josef Bacikfcb80c22011-05-03 10:40:22 -04009076 BUG_ON(ret);
Josef Bacikf0cd8462011-03-04 14:37:08 -05009077
Chris Mason5a3f23d2009-03-31 13:27:11 -04009078 /*
Josef Bacik5dc562c2012-08-17 13:14:17 -04009079 * So if we truncate and then write and fsync we normally would just
9080 * write the extents that changed, which is a problem if we need to
9081 * first truncate that entire inode. So set this flag so we write out
9082 * all of the extents in the inode to the sync log so we're completely
9083 * safe.
9084 */
9085 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
Josef Bacikca7e70f2012-08-27 17:48:15 -04009086 trans->block_rsv = rsv;
Josef Bacik5dc562c2012-08-17 13:14:17 -04009087
Yan, Zheng80825102009-11-12 09:35:36 +00009088 while (1) {
9089 ret = btrfs_truncate_inode_items(trans, root, inode,
9090 inode->i_size,
9091 BTRFS_EXTENT_DATA_KEY);
Chris Mason28ed1342014-12-17 09:41:04 -08009092 if (ret != -ENOSPC && ret != -EAGAIN) {
Josef Bacik3893e332011-01-31 16:03:11 -05009093 err = ret;
Yan, Zheng80825102009-11-12 09:35:36 +00009094 break;
Josef Bacik3893e332011-01-31 16:03:11 -05009095 }
Chris Mason39279cc2007-06-12 06:35:45 -04009096
Josef Bacikfcb80c22011-05-03 10:40:22 -04009097 trans->block_rsv = &root->fs_info->trans_block_rsv;
Yan, Zheng80825102009-11-12 09:35:36 +00009098 ret = btrfs_update_inode(trans, root, inode);
Josef Bacik3893e332011-01-31 16:03:11 -05009099 if (ret) {
9100 err = ret;
9101 break;
9102 }
Josef Bacikca7e70f2012-08-27 17:48:15 -04009103
Yan, Zheng80825102009-11-12 09:35:36 +00009104 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00009105 btrfs_btree_balance_dirty(root);
Josef Bacikca7e70f2012-08-27 17:48:15 -04009106
9107 trans = btrfs_start_transaction(root, 2);
9108 if (IS_ERR(trans)) {
9109 ret = err = PTR_ERR(trans);
9110 trans = NULL;
9111 break;
9112 }
9113
9114 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
9115 rsv, min_size);
9116 BUG_ON(ret); /* shouldn't happen */
9117 trans->block_rsv = rsv;
Yan, Zheng80825102009-11-12 09:35:36 +00009118 }
9119
9120 if (ret == 0 && inode->i_nlink > 0) {
Josef Bacikfcb80c22011-05-03 10:40:22 -04009121 trans->block_rsv = root->orphan_block_rsv;
Yan, Zheng80825102009-11-12 09:35:36 +00009122 ret = btrfs_orphan_del(trans, inode);
Josef Bacik3893e332011-01-31 16:03:11 -05009123 if (ret)
9124 err = ret;
Yan, Zheng80825102009-11-12 09:35:36 +00009125 }
9126
Chris Mason917c16b2011-11-08 14:49:59 -05009127 if (trans) {
9128 trans->block_rsv = &root->fs_info->trans_block_rsv;
9129 ret = btrfs_update_inode(trans, root, inode);
9130 if (ret && !err)
9131 err = ret;
Josef Bacik7b128762008-07-24 12:17:14 -04009132
Josef Bacik7ad85bb2012-01-12 19:10:12 -05009133 ret = btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00009134 btrfs_btree_balance_dirty(root);
Chris Mason917c16b2011-11-08 14:49:59 -05009135 }
Josef Bacikfcb80c22011-05-03 10:40:22 -04009136
9137out:
9138 btrfs_free_block_rsv(root, rsv);
9139
Josef Bacik3893e332011-01-31 16:03:11 -05009140 if (ret && !err)
9141 err = ret;
Josef Bacika41ad392011-01-31 15:30:16 -05009142
Josef Bacik3893e332011-01-31 16:03:11 -05009143 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04009144}
9145
Sven Wegener3b963622008-06-09 21:57:42 -04009146/*
Chris Masond352ac62008-09-29 15:18:18 -04009147 * create a new subvolume directory/inode (helper for the ioctl).
9148 */
Yan Zhengd2fb3432008-12-11 16:30:39 -05009149int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
Filipe David Borba Manana63541922014-01-07 11:47:46 +00009150 struct btrfs_root *new_root,
9151 struct btrfs_root *parent_root,
9152 u64 new_dirid)
Chris Mason39279cc2007-06-12 06:35:45 -04009153{
Chris Mason39279cc2007-06-12 06:35:45 -04009154 struct inode *inode;
Yan, Zheng76dda932009-09-21 16:00:26 -04009155 int err;
Chris Mason00e4e6b2008-08-05 11:18:09 -04009156 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04009157
Florian Albrechtskirchinger12fc9d02012-02-10 22:15:54 +01009158 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9159 new_dirid, new_dirid,
9160 S_IFDIR | (~current_umask() & S_IRWXUGO),
9161 &index);
Chris Mason54aa1f42007-06-22 14:16:25 -04009162 if (IS_ERR(inode))
Christoph Hellwigf46b5a62008-06-11 21:53:53 -04009163 return PTR_ERR(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04009164 inode->i_op = &btrfs_dir_inode_operations;
9165 inode->i_fop = &btrfs_dir_file_operations;
9166
Miklos Szeredibfe86842011-10-28 14:13:29 +02009167 set_nlink(inode, 1);
Chris Masondbe674a2008-07-17 12:54:05 -04009168 btrfs_i_size_write(inode, 0);
Chris Masonb0d5d102014-09-08 13:08:51 -07009169 unlock_new_inode(inode);
Sven Wegener3b963622008-06-09 21:57:42 -04009170
Filipe David Borba Manana63541922014-01-07 11:47:46 +00009171 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9172 if (err)
9173 btrfs_err(new_root->fs_info,
David Sterba351fd352014-05-15 16:48:20 +02009174 "error inheriting subvolume %llu properties: %d",
Filipe David Borba Manana63541922014-01-07 11:47:46 +00009175 new_root->root_key.objectid, err);
9176
Yan, Zheng76dda932009-09-21 16:00:26 -04009177 err = btrfs_update_inode(trans, new_root, inode);
Christoph Hellwigcb8e7092008-10-09 13:39:39 -04009178
Yan, Zheng76dda932009-09-21 16:00:26 -04009179 iput(inode);
Mark Fashehce598972011-07-26 11:32:23 -07009180 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04009181}
9182
Chris Mason39279cc2007-06-12 06:35:45 -04009183struct inode *btrfs_alloc_inode(struct super_block *sb)
9184{
9185 struct btrfs_inode *ei;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009186 struct inode *inode;
Chris Mason39279cc2007-06-12 06:35:45 -04009187
9188 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9189 if (!ei)
9190 return NULL;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009191
9192 ei->root = NULL;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009193 ei->generation = 0;
Josef Bacik15ee9bc2007-08-10 16:22:09 -04009194 ei->last_trans = 0;
Chris Mason257c62e2009-10-13 13:21:08 -04009195 ei->last_sub_trans = 0;
Chris Masone02119d2008-09-05 16:13:11 -04009196 ei->logged_trans = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009197 ei->delalloc_bytes = 0;
Wang Shilong47059d92014-07-03 18:22:07 +08009198 ei->defrag_bytes = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009199 ei->disk_i_size = 0;
9200 ei->flags = 0;
Josef Bacik7709cde2011-08-04 10:25:02 -04009201 ei->csum_bytes = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009202 ei->index_cnt = (u64)-1;
Miao Xie67de1172013-12-26 13:07:06 +08009203 ei->dir_index = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009204 ei->last_unlink_trans = 0;
Liu Bo46d8bc32012-08-29 01:07:55 -06009205 ei->last_log_commit = 0;
David Sterba8089fe62015-11-19 14:15:51 +01009206 ei->delayed_iput_count = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009207
Josef Bacik9e0baf62011-07-15 15:16:44 +00009208 spin_lock_init(&ei->lock);
9209 ei->outstanding_extents = 0;
9210 ei->reserved_extents = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009211
Josef Bacik72ac3c02012-05-23 14:13:11 -04009212 ei->runtime_flags = 0;
Li Zefan261507a02010-12-17 14:21:50 +08009213 ei->force_compress = BTRFS_COMPRESS_NONE;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009214
Miao Xie16cdcec2011-04-22 18:12:22 +08009215 ei->delayed_node = NULL;
9216
chandan r9cc97d62012-07-04 12:48:07 +05309217 ei->i_otime.tv_sec = 0;
9218 ei->i_otime.tv_nsec = 0;
9219
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009220 inode = &ei->vfs_inode;
David Sterbaa8067e02011-04-21 00:34:43 +02009221 extent_map_tree_init(&ei->extent_tree);
David Sterbaf993c882011-04-20 23:35:57 +02009222 extent_io_tree_init(&ei->io_tree, &inode->i_data);
9223 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
Josef Bacik0b32f4b2012-03-13 09:38:00 -04009224 ei->io_tree.track_uptodate = 1;
9225 ei->io_failure_tree.track_uptodate = 1;
Josef Bacikb812ce22012-11-16 13:56:32 -05009226 atomic_set(&ei->sync_writers, 0);
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009227 mutex_init(&ei->log_mutex);
Josef Bacikf2486792012-01-13 12:09:22 -05009228 mutex_init(&ei->delalloc_mutex);
Chris Masone6dcd2d2008-07-17 12:53:50 -04009229 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009230 INIT_LIST_HEAD(&ei->delalloc_inodes);
David Sterba8089fe62015-11-19 14:15:51 +01009231 INIT_LIST_HEAD(&ei->delayed_iput);
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04009232 RB_CLEAR_NODE(&ei->rb_node);
9233
9234 return inode;
Chris Mason39279cc2007-06-12 06:35:45 -04009235}
9236
Josef Bacikaaedb552013-10-11 14:44:09 -04009237#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9238void btrfs_test_destroy_inode(struct inode *inode)
9239{
9240 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9241 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9242}
9243#endif
9244
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11009245static void btrfs_i_callback(struct rcu_head *head)
9246{
9247 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11009248 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9249}
9250
Chris Mason39279cc2007-06-12 06:35:45 -04009251void btrfs_destroy_inode(struct inode *inode)
9252{
Chris Masone6dcd2d2008-07-17 12:53:50 -04009253 struct btrfs_ordered_extent *ordered;
Chris Mason5a3f23d2009-03-31 13:27:11 -04009254 struct btrfs_root *root = BTRFS_I(inode)->root;
9255
Al Virob3d9b7a2012-06-09 13:51:19 -04009256 WARN_ON(!hlist_empty(&inode->i_dentry));
Chris Mason39279cc2007-06-12 06:35:45 -04009257 WARN_ON(inode->i_data.nrpages);
Josef Bacik9e0baf62011-07-15 15:16:44 +00009258 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9259 WARN_ON(BTRFS_I(inode)->reserved_extents);
Josef Bacik7709cde2011-08-04 10:25:02 -04009260 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9261 WARN_ON(BTRFS_I(inode)->csum_bytes);
Wang Shilong47059d92014-07-03 18:22:07 +08009262 WARN_ON(BTRFS_I(inode)->defrag_bytes);
Chris Mason39279cc2007-06-12 06:35:45 -04009263
Chris Mason5a3f23d2009-03-31 13:27:11 -04009264 /*
Josef Bacika6dbd422009-11-11 15:53:34 -05009265 * This can happen where we create an inode, but somebody else also
9266 * created the same inode and we need to destroy the one we already
9267 * created.
9268 */
9269 if (!root)
9270 goto free;
9271
Josef Bacik8a35d952012-05-23 14:26:42 -04009272 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9273 &BTRFS_I(inode)->runtime_flags)) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00009274 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02009275 btrfs_ino(inode));
Josef Bacik8a35d952012-05-23 14:26:42 -04009276 atomic_dec(&root->orphan_inodes);
Josef Bacik7b128762008-07-24 12:17:14 -04009277 }
Josef Bacik7b128762008-07-24 12:17:14 -04009278
Chris Masond3977122009-01-05 21:25:51 -05009279 while (1) {
Chris Masone6dcd2d2008-07-17 12:53:50 -04009280 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9281 if (!ordered)
9282 break;
9283 else {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00009284 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02009285 ordered->file_offset, ordered->len);
Chris Masone6dcd2d2008-07-17 12:53:50 -04009286 btrfs_remove_ordered_extent(inode, ordered);
9287 btrfs_put_ordered_extent(ordered);
9288 btrfs_put_ordered_extent(ordered);
9289 }
9290 }
Qu Wenruo56fa9d02015-10-13 09:53:10 +08009291 btrfs_qgroup_check_reserved_leak(inode);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04009292 inode_tree_del(inode);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04009293 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
Josef Bacika6dbd422009-11-11 15:53:34 -05009294free:
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11009295 call_rcu(&inode->i_rcu, btrfs_i_callback);
Chris Mason39279cc2007-06-12 06:35:45 -04009296}
9297
Al Viro45321ac2010-06-07 13:43:19 -04009298int btrfs_drop_inode(struct inode *inode)
Yan, Zheng76dda932009-09-21 16:00:26 -04009299{
9300 struct btrfs_root *root = BTRFS_I(inode)->root;
Al Viro45321ac2010-06-07 13:43:19 -04009301
Naohiro Aota6379ef92013-06-06 09:56:34 +00009302 if (root == NULL)
9303 return 1;
9304
Liu Bofa6ac872013-02-20 14:10:23 +00009305 /* the snap/subvol tree is on deleting */
Stefan Behrens69e9c6c2013-09-05 16:58:43 +02009306 if (btrfs_root_refs(&root->root_item) == 0)
Al Viro45321ac2010-06-07 13:43:19 -04009307 return 1;
Yan, Zheng76dda932009-09-21 16:00:26 -04009308 else
Al Viro45321ac2010-06-07 13:43:19 -04009309 return generic_drop_inode(inode);
Yan, Zheng76dda932009-09-21 16:00:26 -04009310}
9311
Sven Wegener0ee0fda2008-07-30 16:54:26 -04009312static void init_once(void *foo)
Chris Mason39279cc2007-06-12 06:35:45 -04009313{
9314 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9315
9316 inode_init_once(&ei->vfs_inode);
9317}
9318
9319void btrfs_destroy_cachep(void)
9320{
Kirill A. Shutemov8c0a8532012-09-26 11:33:07 +10009321 /*
9322 * Make sure all delayed rcu free inodes are flushed before we
9323 * destroy cache.
9324 */
9325 rcu_barrier();
Kinglong Mee5598e902016-01-29 21:36:35 +08009326 kmem_cache_destroy(btrfs_inode_cachep);
9327 kmem_cache_destroy(btrfs_trans_handle_cachep);
9328 kmem_cache_destroy(btrfs_transaction_cachep);
9329 kmem_cache_destroy(btrfs_path_cachep);
9330 kmem_cache_destroy(btrfs_free_space_cachep);
Chris Mason39279cc2007-06-12 06:35:45 -04009331}
9332
9333int btrfs_init_cachep(void)
9334{
David Sterba837e1972012-09-07 03:00:48 -06009335 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02009336 sizeof(struct btrfs_inode), 0,
Vladimir Davydov5d097052016-01-14 15:18:21 -08009337 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9338 init_once);
Chris Mason39279cc2007-06-12 06:35:45 -04009339 if (!btrfs_inode_cachep)
9340 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02009341
David Sterba837e1972012-09-07 03:00:48 -06009342 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02009343 sizeof(struct btrfs_trans_handle), 0,
9344 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04009345 if (!btrfs_trans_handle_cachep)
9346 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02009347
David Sterba837e1972012-09-07 03:00:48 -06009348 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02009349 sizeof(struct btrfs_transaction), 0,
9350 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04009351 if (!btrfs_transaction_cachep)
9352 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02009353
David Sterba837e1972012-09-07 03:00:48 -06009354 btrfs_path_cachep = kmem_cache_create("btrfs_path",
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02009355 sizeof(struct btrfs_path), 0,
9356 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04009357 if (!btrfs_path_cachep)
9358 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02009359
David Sterba837e1972012-09-07 03:00:48 -06009360 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
Josef Bacikdc89e982011-01-28 17:05:48 -05009361 sizeof(struct btrfs_free_space), 0,
9362 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9363 if (!btrfs_free_space_cachep)
9364 goto fail;
9365
Chris Mason39279cc2007-06-12 06:35:45 -04009366 return 0;
9367fail:
9368 btrfs_destroy_cachep();
9369 return -ENOMEM;
9370}
9371
9372static int btrfs_getattr(struct vfsmount *mnt,
9373 struct dentry *dentry, struct kstat *stat)
9374{
Miao Xiedf0af1a2013-01-29 10:11:59 +00009375 u64 delalloc_bytes;
David Howells2b0143b2015-03-17 22:25:59 +00009376 struct inode *inode = d_inode(dentry);
David Sterbafadc0d82011-11-20 07:33:38 -05009377 u32 blocksize = inode->i_sb->s_blocksize;
9378
Chris Mason39279cc2007-06-12 06:35:45 -04009379 generic_fillattr(inode, stat);
Al Viro0ee5dc62011-07-07 15:44:25 -04009380 stat->dev = BTRFS_I(inode)->root->anon_dev;
Miao Xiedf0af1a2013-01-29 10:11:59 +00009381
9382 spin_lock(&BTRFS_I(inode)->lock);
9383 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9384 spin_unlock(&BTRFS_I(inode)->lock);
David Sterbafadc0d82011-11-20 07:33:38 -05009385 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
Miao Xiedf0af1a2013-01-29 10:11:59 +00009386 ALIGN(delalloc_bytes, blocksize)) >> 9;
Chris Mason39279cc2007-06-12 06:35:45 -04009387 return 0;
9388}
9389
Chris Masond3977122009-01-05 21:25:51 -05009390static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9391 struct inode *new_dir, struct dentry *new_dentry)
Chris Mason39279cc2007-06-12 06:35:45 -04009392{
9393 struct btrfs_trans_handle *trans;
9394 struct btrfs_root *root = BTRFS_I(old_dir)->root;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04009395 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
David Howells2b0143b2015-03-17 22:25:59 +00009396 struct inode *new_inode = d_inode(new_dentry);
9397 struct inode *old_inode = d_inode(old_dentry);
Chris Mason00e4e6b2008-08-05 11:18:09 -04009398 u64 index = 0;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04009399 u64 root_objectid;
Chris Mason39279cc2007-06-12 06:35:45 -04009400 int ret;
Li Zefan33345d012011-04-20 10:31:50 +08009401 u64 old_ino = btrfs_ino(old_inode);
Chris Mason39279cc2007-06-12 06:35:45 -04009402
Li Zefan33345d012011-04-20 10:31:50 +08009403 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
Yan, Zhengf679a842009-09-24 09:17:31 -04009404 return -EPERM;
9405
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04009406 /* we only allow rename subvolume link between subvolumes */
Li Zefan33345d012011-04-20 10:31:50 +08009407 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
Chris Mason3394e162008-11-17 20:42:26 -05009408 return -EXDEV;
9409
Li Zefan33345d012011-04-20 10:31:50 +08009410 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9411 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04009412 return -ENOTEMPTY;
9413
Chris Mason39279cc2007-06-12 06:35:45 -04009414 if (S_ISDIR(old_inode->i_mode) && new_inode &&
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04009415 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
Chris Mason39279cc2007-06-12 06:35:45 -04009416 return -ENOTEMPTY;
Chris Mason9c520572012-12-17 14:26:57 -05009417
9418
9419 /* check for collisions, even if the name isn't there */
Josef Bacik4871c152013-10-09 12:24:04 -04009420 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
Chris Mason9c520572012-12-17 14:26:57 -05009421 new_dentry->d_name.name,
9422 new_dentry->d_name.len);
9423
9424 if (ret) {
9425 if (ret == -EEXIST) {
9426 /* we shouldn't get
9427 * eexist without a new_inode */
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05309428 if (WARN_ON(!new_inode)) {
Chris Mason9c520572012-12-17 14:26:57 -05009429 return ret;
9430 }
9431 } else {
9432 /* maybe -EOVERFLOW */
9433 return ret;
9434 }
9435 }
9436 ret = 0;
9437
Chris Mason5a3f23d2009-03-31 13:27:11 -04009438 /*
Chris Mason8d875f92014-08-12 10:47:42 -07009439 * we're using rename to replace one file with another. Start IO on it
9440 * now so we don't add too much work to the end of the transaction
Chris Mason5a3f23d2009-03-31 13:27:11 -04009441 */
Chris Mason8d875f92014-08-12 10:47:42 -07009442 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
Chris Mason5a3f23d2009-03-31 13:27:11 -04009443 filemap_flush(old_inode->i_mapping);
9444
Yan, Zheng76dda932009-09-21 16:00:26 -04009445 /* close the racy window with snapshot create/destroy ioctl */
Li Zefan33345d012011-04-20 10:31:50 +08009446 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
Yan, Zheng76dda932009-09-21 16:00:26 -04009447 down_read(&root->fs_info->subvol_sem);
Yan, Zhenga22285a2010-05-16 10:48:46 -04009448 /*
9449 * We want to reserve the absolute worst case amount of items. So if
9450 * both inodes are subvols and we need to unlink them then that would
9451 * require 4 item modifications, but if they are both normal inodes it
9452 * would require 5 item modifications, so we'll assume their normal
9453 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9454 * should cover the worst case number of items we'll modify.
9455 */
Josef Bacik6e137ed2013-03-26 15:26:55 -04009456 trans = btrfs_start_transaction(root, 11);
Johann Lombardib44c59a2011-03-31 13:23:47 +00009457 if (IS_ERR(trans)) {
9458 ret = PTR_ERR(trans);
9459 goto out_notrans;
9460 }
Chris Mason5f39d392007-10-15 16:14:19 -04009461
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04009462 if (dest != root)
9463 btrfs_record_root_in_trans(trans, dest);
9464
Yan, Zhenga5719522009-09-24 09:17:31 -04009465 ret = btrfs_set_inode_index(new_dir, &index);
9466 if (ret)
9467 goto out_fail;
Chris Mason5a3f23d2009-03-31 13:27:11 -04009468
Miao Xie67de1172013-12-26 13:07:06 +08009469 BTRFS_I(old_inode)->dir_index = 0ULL;
Li Zefan33345d012011-04-20 10:31:50 +08009470 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04009471 /* force full log commit if subvolume involved. */
Miao Xie995946d2014-04-02 19:51:06 +08009472 btrfs_set_log_full_commit(root->fs_info, trans);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04009473 } else {
Yan, Zhenga5719522009-09-24 09:17:31 -04009474 ret = btrfs_insert_inode_ref(trans, dest,
9475 new_dentry->d_name.name,
9476 new_dentry->d_name.len,
Li Zefan33345d012011-04-20 10:31:50 +08009477 old_ino,
9478 btrfs_ino(new_dir), index);
Yan, Zhenga5719522009-09-24 09:17:31 -04009479 if (ret)
9480 goto out_fail;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04009481 /*
9482 * this is an ugly little race, but the rename is required
9483 * to make sure that if we crash, the inode is either at the
9484 * old name or the new one. pinning the log transaction lets
9485 * us make sure we don't allow a log commit to come in after
9486 * we unlink the name but before we add the new name back in.
9487 */
9488 btrfs_pin_log_trans(root);
9489 }
Chris Mason5a3f23d2009-03-31 13:27:11 -04009490
Josef Bacik0c4d2d92012-04-05 15:03:02 -04009491 inode_inc_iversion(old_dir);
9492 inode_inc_iversion(new_dir);
9493 inode_inc_iversion(old_inode);
Deepa Dinamani04b285f2016-02-06 23:57:21 -08009494 old_dir->i_ctime = old_dir->i_mtime =
9495 new_dir->i_ctime = new_dir->i_mtime =
9496 old_inode->i_ctime = current_fs_time(old_dir->i_sb);
Chris Mason5f39d392007-10-15 16:14:19 -04009497
Chris Mason12fcfd22009-03-24 10:24:20 -04009498 if (old_dentry->d_parent != new_dentry->d_parent)
9499 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9500
Li Zefan33345d012011-04-20 10:31:50 +08009501 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04009502 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9503 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9504 old_dentry->d_name.name,
9505 old_dentry->d_name.len);
9506 } else {
Al Viro92986792011-03-04 17:14:37 +00009507 ret = __btrfs_unlink_inode(trans, root, old_dir,
David Howells2b0143b2015-03-17 22:25:59 +00009508 d_inode(old_dentry),
Al Viro92986792011-03-04 17:14:37 +00009509 old_dentry->d_name.name,
9510 old_dentry->d_name.len);
9511 if (!ret)
9512 ret = btrfs_update_inode(trans, root, old_inode);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04009513 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01009514 if (ret) {
9515 btrfs_abort_transaction(trans, root, ret);
9516 goto out_fail;
9517 }
Chris Mason39279cc2007-06-12 06:35:45 -04009518
9519 if (new_inode) {
Josef Bacik0c4d2d92012-04-05 15:03:02 -04009520 inode_inc_iversion(new_inode);
Deepa Dinamani04b285f2016-02-06 23:57:21 -08009521 new_inode->i_ctime = current_fs_time(new_inode->i_sb);
Li Zefan33345d012011-04-20 10:31:50 +08009522 if (unlikely(btrfs_ino(new_inode) ==
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04009523 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9524 root_objectid = BTRFS_I(new_inode)->location.objectid;
9525 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9526 root_objectid,
9527 new_dentry->d_name.name,
9528 new_dentry->d_name.len);
9529 BUG_ON(new_inode->i_nlink == 0);
9530 } else {
9531 ret = btrfs_unlink_inode(trans, dest, new_dir,
David Howells2b0143b2015-03-17 22:25:59 +00009532 d_inode(new_dentry),
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04009533 new_dentry->d_name.name,
9534 new_dentry->d_name.len);
9535 }
Josef Bacik4ef31a42013-08-13 14:10:08 -04009536 if (!ret && new_inode->i_nlink == 0)
David Howells2b0143b2015-03-17 22:25:59 +00009537 ret = btrfs_orphan_add(trans, d_inode(new_dentry));
Jeff Mahoney79787ea2012-03-12 16:03:00 +01009538 if (ret) {
9539 btrfs_abort_transaction(trans, root, ret);
9540 goto out_fail;
9541 }
Chris Mason39279cc2007-06-12 06:35:45 -04009542 }
Josef Bacikaec74772008-07-24 12:12:38 -04009543
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04009544 ret = btrfs_add_link(trans, new_dir, old_inode,
9545 new_dentry->d_name.name,
Yan, Zhenga5719522009-09-24 09:17:31 -04009546 new_dentry->d_name.len, 0, index);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01009547 if (ret) {
9548 btrfs_abort_transaction(trans, root, ret);
9549 goto out_fail;
9550 }
Chris Mason39279cc2007-06-12 06:35:45 -04009551
Miao Xie67de1172013-12-26 13:07:06 +08009552 if (old_inode->i_nlink == 1)
9553 BTRFS_I(old_inode)->dir_index = index;
9554
Li Zefan33345d012011-04-20 10:31:50 +08009555 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
Al Viro10d9f302011-07-16 23:09:10 -04009556 struct dentry *parent = new_dentry->d_parent;
Josef Bacik6a912212010-11-20 09:48:00 +00009557 btrfs_log_new_name(trans, old_inode, old_dir, parent);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04009558 btrfs_end_log_trans(root);
9559 }
Chris Mason39279cc2007-06-12 06:35:45 -04009560out_fail:
Josef Bacik7ad85bb2012-01-12 19:10:12 -05009561 btrfs_end_transaction(trans, root);
Johann Lombardib44c59a2011-03-31 13:23:47 +00009562out_notrans:
Li Zefan33345d012011-04-20 10:31:50 +08009563 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
Yan, Zheng76dda932009-09-21 16:00:26 -04009564 up_read(&root->fs_info->subvol_sem);
Josef Bacik9ed74f22009-09-11 16:12:44 -04009565
Chris Mason39279cc2007-06-12 06:35:45 -04009566 return ret;
9567}
9568
Miklos Szeredi80ace852014-07-23 15:15:32 +02009569static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9570 struct inode *new_dir, struct dentry *new_dentry,
9571 unsigned int flags)
9572{
9573 if (flags & ~RENAME_NOREPLACE)
9574 return -EINVAL;
9575
9576 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
9577}
9578
Miao Xie8ccf6f192012-10-25 09:28:04 +00009579static void btrfs_run_delalloc_work(struct btrfs_work *work)
9580{
9581 struct btrfs_delalloc_work *delalloc_work;
Josef Bacik9f23e282013-10-28 15:03:41 -04009582 struct inode *inode;
Miao Xie8ccf6f192012-10-25 09:28:04 +00009583
9584 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9585 work);
Josef Bacik9f23e282013-10-28 15:03:41 -04009586 inode = delalloc_work->inode;
David Sterba30424602015-11-27 19:27:11 +01009587 filemap_flush(inode->i_mapping);
9588 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9589 &BTRFS_I(inode)->runtime_flags))
Josef Bacik9f23e282013-10-28 15:03:41 -04009590 filemap_flush(inode->i_mapping);
Miao Xie8ccf6f192012-10-25 09:28:04 +00009591
9592 if (delalloc_work->delay_iput)
Josef Bacik9f23e282013-10-28 15:03:41 -04009593 btrfs_add_delayed_iput(inode);
Miao Xie8ccf6f192012-10-25 09:28:04 +00009594 else
Josef Bacik9f23e282013-10-28 15:03:41 -04009595 iput(inode);
Miao Xie8ccf6f192012-10-25 09:28:04 +00009596 complete(&delalloc_work->completion);
9597}
9598
9599struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
David Sterba651d4942015-11-27 19:24:16 +01009600 int delay_iput)
Miao Xie8ccf6f192012-10-25 09:28:04 +00009601{
9602 struct btrfs_delalloc_work *work;
9603
David Sterba100d5702015-12-08 14:39:32 +01009604 work = kmalloc(sizeof(*work), GFP_NOFS);
Miao Xie8ccf6f192012-10-25 09:28:04 +00009605 if (!work)
9606 return NULL;
9607
9608 init_completion(&work->completion);
9609 INIT_LIST_HEAD(&work->list);
9610 work->inode = inode;
Miao Xie8ccf6f192012-10-25 09:28:04 +00009611 work->delay_iput = delay_iput;
Liu Bo9e0af232014-08-15 23:36:53 +08009612 WARN_ON_ONCE(!inode);
9613 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9614 btrfs_run_delalloc_work, NULL, NULL);
Miao Xie8ccf6f192012-10-25 09:28:04 +00009615
9616 return work;
9617}
9618
9619void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9620{
9621 wait_for_completion(&work->completion);
David Sterba100d5702015-12-08 14:39:32 +01009622 kfree(work);
Miao Xie8ccf6f192012-10-25 09:28:04 +00009623}
9624
Chris Masond352ac62008-09-29 15:18:18 -04009625/*
9626 * some fairly slow code that needs optimization. This walks the list
9627 * of all the inodes with pending delalloc and forces them to disk.
9628 */
Miao Xie6c255e62014-03-06 13:55:01 +08009629static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9630 int nr)
Chris Masonea8c2812008-08-04 23:17:27 -04009631{
Chris Masonea8c2812008-08-04 23:17:27 -04009632 struct btrfs_inode *binode;
Zheng Yan5b21f2e2008-09-26 10:05:38 -04009633 struct inode *inode;
Miao Xie8ccf6f192012-10-25 09:28:04 +00009634 struct btrfs_delalloc_work *work, *next;
9635 struct list_head works;
Miao Xie1eafa6c2013-01-22 10:49:00 +00009636 struct list_head splice;
Miao Xie8ccf6f192012-10-25 09:28:04 +00009637 int ret = 0;
Chris Masonea8c2812008-08-04 23:17:27 -04009638
Miao Xie8ccf6f192012-10-25 09:28:04 +00009639 INIT_LIST_HEAD(&works);
Miao Xie1eafa6c2013-01-22 10:49:00 +00009640 INIT_LIST_HEAD(&splice);
Miao Xie63607cc2013-01-22 10:50:35 +00009641
Miao Xie573bfb72014-03-06 13:55:03 +08009642 mutex_lock(&root->delalloc_mutex);
Miao Xieeb73c1b2013-05-15 07:48:22 +00009643 spin_lock(&root->delalloc_lock);
9644 list_splice_init(&root->delalloc_inodes, &splice);
Miao Xie1eafa6c2013-01-22 10:49:00 +00009645 while (!list_empty(&splice)) {
9646 binode = list_entry(splice.next, struct btrfs_inode,
Chris Masonea8c2812008-08-04 23:17:27 -04009647 delalloc_inodes);
Miao Xie1eafa6c2013-01-22 10:49:00 +00009648
Miao Xieeb73c1b2013-05-15 07:48:22 +00009649 list_move_tail(&binode->delalloc_inodes,
9650 &root->delalloc_inodes);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04009651 inode = igrab(&binode->vfs_inode);
Miao Xiedf0af1a2013-01-29 10:11:59 +00009652 if (!inode) {
Miao Xieeb73c1b2013-05-15 07:48:22 +00009653 cond_resched_lock(&root->delalloc_lock);
Miao Xie1eafa6c2013-01-22 10:49:00 +00009654 continue;
Miao Xiedf0af1a2013-01-29 10:11:59 +00009655 }
Miao Xieeb73c1b2013-05-15 07:48:22 +00009656 spin_unlock(&root->delalloc_lock);
Miao Xie1eafa6c2013-01-22 10:49:00 +00009657
David Sterba651d4942015-11-27 19:24:16 +01009658 work = btrfs_alloc_delalloc_work(inode, delay_iput);
David Sterba5d99a9982014-09-29 19:20:37 +02009659 if (!work) {
Josef Bacikf4ab9ea2013-09-17 11:25:44 -04009660 if (delay_iput)
9661 btrfs_add_delayed_iput(inode);
9662 else
9663 iput(inode);
Miao Xie1eafa6c2013-01-22 10:49:00 +00009664 ret = -ENOMEM;
Wang Shilonga1ecaab2014-04-02 19:53:32 +08009665 goto out;
Zheng Yan5b21f2e2008-09-26 10:05:38 -04009666 }
Miao Xie1eafa6c2013-01-22 10:49:00 +00009667 list_add_tail(&work->list, &works);
Qu Wenruoa44903a2014-02-28 10:46:09 +08009668 btrfs_queue_work(root->fs_info->flush_workers,
9669 &work->work);
Miao Xie6c255e62014-03-06 13:55:01 +08009670 ret++;
9671 if (nr != -1 && ret >= nr)
Wang Shilonga1ecaab2014-04-02 19:53:32 +08009672 goto out;
Zheng Yan5b21f2e2008-09-26 10:05:38 -04009673 cond_resched();
Miao Xieeb73c1b2013-05-15 07:48:22 +00009674 spin_lock(&root->delalloc_lock);
Chris Masonea8c2812008-08-04 23:17:27 -04009675 }
Miao Xieeb73c1b2013-05-15 07:48:22 +00009676 spin_unlock(&root->delalloc_lock);
Chris Mason8c8bee12008-09-29 11:19:10 -04009677
Wang Shilonga1ecaab2014-04-02 19:53:32 +08009678out:
Miao Xie1eafa6c2013-01-22 10:49:00 +00009679 list_for_each_entry_safe(work, next, &works, list) {
9680 list_del_init(&work->list);
9681 btrfs_wait_and_free_delalloc_work(work);
9682 }
9683
Miao Xieeb73c1b2013-05-15 07:48:22 +00009684 if (!list_empty_careful(&splice)) {
9685 spin_lock(&root->delalloc_lock);
9686 list_splice_tail(&splice, &root->delalloc_inodes);
9687 spin_unlock(&root->delalloc_lock);
9688 }
Miao Xie573bfb72014-03-06 13:55:03 +08009689 mutex_unlock(&root->delalloc_mutex);
Miao Xieeb73c1b2013-05-15 07:48:22 +00009690 return ret;
9691}
9692
9693int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9694{
9695 int ret;
9696
Wang Shilong2c21b4d2014-01-14 19:42:20 +08009697 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
Miao Xieeb73c1b2013-05-15 07:48:22 +00009698 return -EROFS;
9699
Miao Xie6c255e62014-03-06 13:55:01 +08009700 ret = __start_delalloc_inodes(root, delay_iput, -1);
9701 if (ret > 0)
9702 ret = 0;
Miao Xieeb73c1b2013-05-15 07:48:22 +00009703 /*
9704 * the filemap_flush will queue IO into the worker threads, but
Chris Mason8c8bee12008-09-29 11:19:10 -04009705 * we have to make sure the IO is actually started and that
9706 * ordered extents get created before we return
9707 */
9708 atomic_inc(&root->fs_info->async_submit_draining);
Chris Masond3977122009-01-05 21:25:51 -05009709 while (atomic_read(&root->fs_info->nr_async_submits) ||
Chris Mason771ed682008-11-06 22:02:51 -05009710 atomic_read(&root->fs_info->async_delalloc_pages)) {
Chris Mason8c8bee12008-09-29 11:19:10 -04009711 wait_event(root->fs_info->async_submit_wait,
Chris Mason771ed682008-11-06 22:02:51 -05009712 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9713 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
Chris Mason8c8bee12008-09-29 11:19:10 -04009714 }
9715 atomic_dec(&root->fs_info->async_submit_draining);
Miao Xieeb73c1b2013-05-15 07:48:22 +00009716 return ret;
9717}
9718
Miao Xie6c255e62014-03-06 13:55:01 +08009719int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9720 int nr)
Miao Xieeb73c1b2013-05-15 07:48:22 +00009721{
9722 struct btrfs_root *root;
9723 struct list_head splice;
9724 int ret;
9725
Wang Shilong2c21b4d2014-01-14 19:42:20 +08009726 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
Miao Xieeb73c1b2013-05-15 07:48:22 +00009727 return -EROFS;
9728
9729 INIT_LIST_HEAD(&splice);
9730
Miao Xie573bfb72014-03-06 13:55:03 +08009731 mutex_lock(&fs_info->delalloc_root_mutex);
Miao Xieeb73c1b2013-05-15 07:48:22 +00009732 spin_lock(&fs_info->delalloc_root_lock);
9733 list_splice_init(&fs_info->delalloc_roots, &splice);
Miao Xie6c255e62014-03-06 13:55:01 +08009734 while (!list_empty(&splice) && nr) {
Miao Xieeb73c1b2013-05-15 07:48:22 +00009735 root = list_first_entry(&splice, struct btrfs_root,
9736 delalloc_root);
9737 root = btrfs_grab_fs_root(root);
9738 BUG_ON(!root);
9739 list_move_tail(&root->delalloc_root,
9740 &fs_info->delalloc_roots);
9741 spin_unlock(&fs_info->delalloc_root_lock);
9742
Miao Xie6c255e62014-03-06 13:55:01 +08009743 ret = __start_delalloc_inodes(root, delay_iput, nr);
Miao Xieeb73c1b2013-05-15 07:48:22 +00009744 btrfs_put_fs_root(root);
Miao Xie6c255e62014-03-06 13:55:01 +08009745 if (ret < 0)
Miao Xieeb73c1b2013-05-15 07:48:22 +00009746 goto out;
9747
Miao Xie6c255e62014-03-06 13:55:01 +08009748 if (nr != -1) {
9749 nr -= ret;
9750 WARN_ON(nr < 0);
9751 }
Miao Xieeb73c1b2013-05-15 07:48:22 +00009752 spin_lock(&fs_info->delalloc_root_lock);
9753 }
9754 spin_unlock(&fs_info->delalloc_root_lock);
9755
Miao Xie6c255e62014-03-06 13:55:01 +08009756 ret = 0;
Miao Xieeb73c1b2013-05-15 07:48:22 +00009757 atomic_inc(&fs_info->async_submit_draining);
9758 while (atomic_read(&fs_info->nr_async_submits) ||
9759 atomic_read(&fs_info->async_delalloc_pages)) {
9760 wait_event(fs_info->async_submit_wait,
9761 (atomic_read(&fs_info->nr_async_submits) == 0 &&
9762 atomic_read(&fs_info->async_delalloc_pages) == 0));
9763 }
9764 atomic_dec(&fs_info->async_submit_draining);
Miao Xie8ccf6f192012-10-25 09:28:04 +00009765out:
Miao Xie1eafa6c2013-01-22 10:49:00 +00009766 if (!list_empty_careful(&splice)) {
Miao Xieeb73c1b2013-05-15 07:48:22 +00009767 spin_lock(&fs_info->delalloc_root_lock);
9768 list_splice_tail(&splice, &fs_info->delalloc_roots);
9769 spin_unlock(&fs_info->delalloc_root_lock);
Miao Xie1eafa6c2013-01-22 10:49:00 +00009770 }
Miao Xie573bfb72014-03-06 13:55:03 +08009771 mutex_unlock(&fs_info->delalloc_root_mutex);
Miao Xie8ccf6f192012-10-25 09:28:04 +00009772 return ret;
Chris Masonea8c2812008-08-04 23:17:27 -04009773}
9774
Chris Mason39279cc2007-06-12 06:35:45 -04009775static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9776 const char *symname)
9777{
9778 struct btrfs_trans_handle *trans;
9779 struct btrfs_root *root = BTRFS_I(dir)->root;
9780 struct btrfs_path *path;
9781 struct btrfs_key key;
Chris Mason1832a6d2007-12-21 16:27:21 -05009782 struct inode *inode = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04009783 int err;
9784 int drop_inode = 0;
9785 u64 objectid;
Dulshani Gunawardhana67871252013-10-31 10:33:04 +05309786 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04009787 int name_len;
9788 int datasize;
Chris Mason5f39d392007-10-15 16:14:19 -04009789 unsigned long ptr;
Chris Mason39279cc2007-06-12 06:35:45 -04009790 struct btrfs_file_extent_item *ei;
Chris Mason5f39d392007-10-15 16:14:19 -04009791 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04009792
Filipe David Borba Mananaf06becc2013-09-16 09:53:28 +01009793 name_len = strlen(symname);
Chris Mason39279cc2007-06-12 06:35:45 -04009794 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9795 return -ENAMETOOLONG;
Chris Mason1832a6d2007-12-21 16:27:21 -05009796
Josef Bacik9ed74f22009-09-11 16:12:44 -04009797 /*
9798 * 2 items for inode item and ref
9799 * 2 items for dir items
Filipe Manana9269d122015-12-31 18:16:29 +00009800 * 1 item for updating parent inode item
9801 * 1 item for the inline extent item
Josef Bacik9ed74f22009-09-11 16:12:44 -04009802 * 1 item for xattr if selinux is on
9803 */
Filipe Manana9269d122015-12-31 18:16:29 +00009804 trans = btrfs_start_transaction(root, 7);
Yan, Zhenga22285a2010-05-16 10:48:46 -04009805 if (IS_ERR(trans))
9806 return PTR_ERR(trans);
Chris Mason1832a6d2007-12-21 16:27:21 -05009807
Li Zefan581bb052011-04-20 10:06:11 +08009808 err = btrfs_find_free_ino(root, &objectid);
9809 if (err)
9810 goto out_unlock;
9811
Josef Bacikaec74772008-07-24 12:12:38 -04009812 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Li Zefan33345d012011-04-20 10:31:50 +08009813 dentry->d_name.len, btrfs_ino(dir), objectid,
Josef Bacikd82a6f12011-05-11 15:26:06 -04009814 S_IFLNK|S_IRWXUGO, &index);
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04009815 if (IS_ERR(inode)) {
9816 err = PTR_ERR(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04009817 goto out_unlock;
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04009818 }
Chris Mason39279cc2007-06-12 06:35:45 -04009819
Casey Schauflerad19db72011-12-15 10:09:07 -05009820 /*
9821 * If the active LSM wants to access the inode during
9822 * d_instantiate it needs these. Smack checks to see
9823 * if the filesystem supports xattrs by looking at the
9824 * ops vector.
9825 */
9826 inode->i_fop = &btrfs_file_operations;
9827 inode->i_op = &btrfs_file_inode_operations;
Chris Masonb0d5d102014-09-08 13:08:51 -07009828 inode->i_mapping->a_ops = &btrfs_aops;
Chris Masonb0d5d102014-09-08 13:08:51 -07009829 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9830
9831 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9832 if (err)
9833 goto out_unlock_inode;
Casey Schauflerad19db72011-12-15 10:09:07 -05009834
Chris Mason39279cc2007-06-12 06:35:45 -04009835 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07009836 if (!path) {
9837 err = -ENOMEM;
Chris Masonb0d5d102014-09-08 13:08:51 -07009838 goto out_unlock_inode;
Mark Fashehd8926bb2011-07-13 10:38:47 -07009839 }
Li Zefan33345d012011-04-20 10:31:50 +08009840 key.objectid = btrfs_ino(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04009841 key.offset = 0;
David Sterba962a2982014-06-04 18:41:45 +02009842 key.type = BTRFS_EXTENT_DATA_KEY;
Chris Mason39279cc2007-06-12 06:35:45 -04009843 datasize = btrfs_file_extent_calc_inline_size(name_len);
9844 err = btrfs_insert_empty_item(trans, root, path, &key,
9845 datasize);
Chris Mason54aa1f42007-06-22 14:16:25 -04009846 if (err) {
Julia Lawallb0839162011-05-14 07:10:51 +00009847 btrfs_free_path(path);
Chris Masonb0d5d102014-09-08 13:08:51 -07009848 goto out_unlock_inode;
Chris Mason54aa1f42007-06-22 14:16:25 -04009849 }
Chris Mason5f39d392007-10-15 16:14:19 -04009850 leaf = path->nodes[0];
9851 ei = btrfs_item_ptr(leaf, path->slots[0],
9852 struct btrfs_file_extent_item);
9853 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9854 btrfs_set_file_extent_type(leaf, ei,
Chris Mason39279cc2007-06-12 06:35:45 -04009855 BTRFS_FILE_EXTENT_INLINE);
Chris Masonc8b97812008-10-29 14:49:59 -04009856 btrfs_set_file_extent_encryption(leaf, ei, 0);
9857 btrfs_set_file_extent_compression(leaf, ei, 0);
9858 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9859 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9860
Chris Mason39279cc2007-06-12 06:35:45 -04009861 ptr = btrfs_file_extent_inline_start(ei);
Chris Mason5f39d392007-10-15 16:14:19 -04009862 write_extent_buffer(leaf, symname, ptr, name_len);
9863 btrfs_mark_buffer_dirty(leaf);
Chris Mason39279cc2007-06-12 06:35:45 -04009864 btrfs_free_path(path);
Chris Mason5f39d392007-10-15 16:14:19 -04009865
Chris Mason39279cc2007-06-12 06:35:45 -04009866 inode->i_op = &btrfs_symlink_inode_operations;
Al Viro21fc61c2015-11-17 01:07:57 -05009867 inode_nohighmem(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04009868 inode->i_mapping->a_ops = &btrfs_symlink_aops;
Yan Zhengd899e052008-10-30 14:25:28 -04009869 inode_set_bytes(inode, name_len);
Filipe David Borba Mananaf06becc2013-09-16 09:53:28 +01009870 btrfs_i_size_write(inode, name_len);
Chris Mason54aa1f42007-06-22 14:16:25 -04009871 err = btrfs_update_inode(trans, root, inode);
Filipe Mananad50866d2015-12-31 18:08:24 +00009872 /*
9873 * Last step, add directory indexes for our symlink inode. This is the
9874 * last step to avoid extra cleanup of these indexes if an error happens
9875 * elsewhere above.
9876 */
9877 if (!err)
9878 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
Chris Masonb0d5d102014-09-08 13:08:51 -07009879 if (err) {
Chris Mason54aa1f42007-06-22 14:16:25 -04009880 drop_inode = 1;
Chris Masonb0d5d102014-09-08 13:08:51 -07009881 goto out_unlock_inode;
9882 }
9883
9884 unlock_new_inode(inode);
9885 d_instantiate(dentry, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04009886
9887out_unlock:
Josef Bacik7ad85bb2012-01-12 19:10:12 -05009888 btrfs_end_transaction(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04009889 if (drop_inode) {
9890 inode_dec_link_count(inode);
9891 iput(inode);
9892 }
Liu Bob53d3f52012-11-14 14:34:34 +00009893 btrfs_btree_balance_dirty(root);
Chris Mason39279cc2007-06-12 06:35:45 -04009894 return err;
Chris Masonb0d5d102014-09-08 13:08:51 -07009895
9896out_unlock_inode:
9897 drop_inode = 1;
9898 unlock_new_inode(inode);
9899 goto out_unlock;
Chris Mason39279cc2007-06-12 06:35:45 -04009900}
Chris Mason16432982008-04-10 10:23:21 -04009901
Josef Bacik0af3d002010-06-21 14:48:16 -04009902static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9903 u64 start, u64 num_bytes, u64 min_size,
9904 loff_t actual_len, u64 *alloc_hint,
9905 struct btrfs_trans_handle *trans)
Yan Zhengd899e052008-10-30 14:25:28 -04009906{
Josef Bacik5dc562c2012-08-17 13:14:17 -04009907 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9908 struct extent_map *em;
Yan Zhengd899e052008-10-30 14:25:28 -04009909 struct btrfs_root *root = BTRFS_I(inode)->root;
9910 struct btrfs_key ins;
Yan Zhengd899e052008-10-30 14:25:28 -04009911 u64 cur_offset = start;
Josef Bacik55a61d12010-11-22 18:50:32 +00009912 u64 i_size;
Chris Mason154ea282013-03-05 11:11:26 -05009913 u64 cur_bytes;
Josef Bacik0b670dc2015-09-23 17:11:16 -04009914 u64 last_alloc = (u64)-1;
Yan Zhengd899e052008-10-30 14:25:28 -04009915 int ret = 0;
Josef Bacik0af3d002010-06-21 14:48:16 -04009916 bool own_trans = true;
Yan Zhengd899e052008-10-30 14:25:28 -04009917
Josef Bacik0af3d002010-06-21 14:48:16 -04009918 if (trans)
9919 own_trans = false;
Yan Zhengd899e052008-10-30 14:25:28 -04009920 while (num_bytes > 0) {
Josef Bacik0af3d002010-06-21 14:48:16 -04009921 if (own_trans) {
9922 trans = btrfs_start_transaction(root, 3);
9923 if (IS_ERR(trans)) {
9924 ret = PTR_ERR(trans);
9925 break;
9926 }
Yan Zhengd899e052008-10-30 14:25:28 -04009927 }
Yan, Zheng5a303d52009-11-12 09:34:52 +00009928
Byongho Leeee221842015-12-15 01:42:10 +09009929 cur_bytes = min_t(u64, num_bytes, SZ_256M);
Chris Mason154ea282013-03-05 11:11:26 -05009930 cur_bytes = max(cur_bytes, min_size);
Josef Bacik0b670dc2015-09-23 17:11:16 -04009931 /*
9932 * If we are severely fragmented we could end up with really
9933 * small allocations, so if the allocator is returning small
9934 * chunks lets make its job easier by only searching for those
9935 * sized chunks.
9936 */
9937 cur_bytes = min(cur_bytes, last_alloc);
Josef Bacik00361582013-08-14 14:02:47 -04009938 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
Miao Xiee570fd22014-06-19 10:42:50 +08009939 *alloc_hint, &ins, 1, 0);
Yan, Zheng5a303d52009-11-12 09:34:52 +00009940 if (ret) {
Josef Bacik0af3d002010-06-21 14:48:16 -04009941 if (own_trans)
9942 btrfs_end_transaction(trans, root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04009943 break;
Yan, Zheng5a303d52009-11-12 09:34:52 +00009944 }
9945
Josef Bacik0b670dc2015-09-23 17:11:16 -04009946 last_alloc = ins.offset;
Yan Zhengd899e052008-10-30 14:25:28 -04009947 ret = insert_reserved_file_extent(trans, inode,
9948 cur_offset, ins.objectid,
9949 ins.offset, ins.offset,
Yan, Zheng920bbbf2009-11-12 09:34:08 +00009950 ins.offset, 0, 0, 0,
Yan Zhengd899e052008-10-30 14:25:28 -04009951 BTRFS_FILE_EXTENT_PREALLOC);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01009952 if (ret) {
Josef Bacik857cc2f2013-10-07 15:21:08 -04009953 btrfs_free_reserved_extent(root, ins.objectid,
Miao Xiee570fd22014-06-19 10:42:50 +08009954 ins.offset, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01009955 btrfs_abort_transaction(trans, root, ret);
9956 if (own_trans)
9957 btrfs_end_transaction(trans, root);
9958 break;
9959 }
Dongsheng Yang31193212014-12-12 16:44:35 +08009960
Chris Masona1ed8352009-09-11 12:27:37 -04009961 btrfs_drop_extent_cache(inode, cur_offset,
9962 cur_offset + ins.offset -1, 0);
Yan, Zheng5a303d52009-11-12 09:34:52 +00009963
Josef Bacik5dc562c2012-08-17 13:14:17 -04009964 em = alloc_extent_map();
9965 if (!em) {
9966 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9967 &BTRFS_I(inode)->runtime_flags);
9968 goto next;
9969 }
9970
9971 em->start = cur_offset;
9972 em->orig_start = cur_offset;
9973 em->len = ins.offset;
9974 em->block_start = ins.objectid;
9975 em->block_len = ins.offset;
Josef Bacikb4939682012-12-03 10:31:19 -05009976 em->orig_block_len = ins.offset;
Josef Bacikcc95bef2013-04-04 14:31:27 -04009977 em->ram_bytes = ins.offset;
Josef Bacik5dc562c2012-08-17 13:14:17 -04009978 em->bdev = root->fs_info->fs_devices->latest_bdev;
9979 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9980 em->generation = trans->transid;
9981
9982 while (1) {
9983 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -04009984 ret = add_extent_mapping(em_tree, em, 1);
Josef Bacik5dc562c2012-08-17 13:14:17 -04009985 write_unlock(&em_tree->lock);
9986 if (ret != -EEXIST)
9987 break;
9988 btrfs_drop_extent_cache(inode, cur_offset,
9989 cur_offset + ins.offset - 1,
9990 0);
9991 }
9992 free_extent_map(em);
9993next:
Yan Zhengd899e052008-10-30 14:25:28 -04009994 num_bytes -= ins.offset;
9995 cur_offset += ins.offset;
Yan, Zhengefa56462010-05-16 10:49:59 -04009996 *alloc_hint = ins.objectid + ins.offset;
Yan, Zheng5a303d52009-11-12 09:34:52 +00009997
Josef Bacik0c4d2d92012-04-05 15:03:02 -04009998 inode_inc_iversion(inode);
Deepa Dinamani04b285f2016-02-06 23:57:21 -08009999 inode->i_ctime = current_fs_time(inode->i_sb);
Christoph Hellwig6cbff002009-04-17 10:37:41 +020010000 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
Yan Zhengd899e052008-10-30 14:25:28 -040010001 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
Yan, Zhengefa56462010-05-16 10:49:59 -040010002 (actual_len > inode->i_size) &&
10003 (cur_offset > inode->i_size)) {
Aneesh Kumar K.Vd1ea6a62010-01-20 07:28:54 +000010004 if (cur_offset > actual_len)
Josef Bacik55a61d12010-11-22 18:50:32 +000010005 i_size = actual_len;
Aneesh Kumar K.Vd1ea6a62010-01-20 07:28:54 +000010006 else
Josef Bacik55a61d12010-11-22 18:50:32 +000010007 i_size = cur_offset;
10008 i_size_write(inode, i_size);
10009 btrfs_ordered_update_i_size(inode, i_size, NULL);
Yan, Zheng5a303d52009-11-12 09:34:52 +000010010 }
10011
Yan Zhengd899e052008-10-30 14:25:28 -040010012 ret = btrfs_update_inode(trans, root, inode);
Jeff Mahoney79787ea2012-03-12 16:03:00 +010010013
10014 if (ret) {
10015 btrfs_abort_transaction(trans, root, ret);
10016 if (own_trans)
10017 btrfs_end_transaction(trans, root);
10018 break;
10019 }
Yan Zhengd899e052008-10-30 14:25:28 -040010020
Josef Bacik0af3d002010-06-21 14:48:16 -040010021 if (own_trans)
10022 btrfs_end_transaction(trans, root);
Yan, Zheng5a303d52009-11-12 09:34:52 +000010023 }
Yan Zhengd899e052008-10-30 14:25:28 -040010024 return ret;
10025}
10026
Josef Bacik0af3d002010-06-21 14:48:16 -040010027int btrfs_prealloc_file_range(struct inode *inode, int mode,
10028 u64 start, u64 num_bytes, u64 min_size,
10029 loff_t actual_len, u64 *alloc_hint)
10030{
10031 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10032 min_size, actual_len, alloc_hint,
10033 NULL);
10034}
10035
10036int btrfs_prealloc_file_range_trans(struct inode *inode,
10037 struct btrfs_trans_handle *trans, int mode,
10038 u64 start, u64 num_bytes, u64 min_size,
10039 loff_t actual_len, u64 *alloc_hint)
10040{
10041 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10042 min_size, actual_len, alloc_hint, trans);
10043}
10044
Chris Masone6dcd2d2008-07-17 12:53:50 -040010045static int btrfs_set_page_dirty(struct page *page)
10046{
Chris Masone6dcd2d2008-07-17 12:53:50 -040010047 return __set_page_dirty_nobuffers(page);
10048}
10049
Al Viro10556cb2011-06-20 19:28:19 -040010050static int btrfs_permission(struct inode *inode, int mask)
Yanfdebe2b2008-01-14 13:26:08 -050010051{
Li Zefanb83cc962010-12-20 16:04:08 +080010052 struct btrfs_root *root = BTRFS_I(inode)->root;
Jeff Mahoneycb6db4e2011-08-15 17:27:21 +000010053 umode_t mode = inode->i_mode;
Li Zefanb83cc962010-12-20 16:04:08 +080010054
Jeff Mahoneycb6db4e2011-08-15 17:27:21 +000010055 if (mask & MAY_WRITE &&
10056 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10057 if (btrfs_root_readonly(root))
10058 return -EROFS;
10059 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10060 return -EACCES;
10061 }
Al Viro2830ba72011-06-20 19:16:29 -040010062 return generic_permission(inode, mask);
Yanfdebe2b2008-01-14 13:26:08 -050010063}
Chris Mason39279cc2007-06-12 06:35:45 -040010064
Filipe Mananaef3b9af2014-04-27 20:40:45 +010010065static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10066{
10067 struct btrfs_trans_handle *trans;
10068 struct btrfs_root *root = BTRFS_I(dir)->root;
10069 struct inode *inode = NULL;
10070 u64 objectid;
10071 u64 index;
10072 int ret = 0;
10073
10074 /*
10075 * 5 units required for adding orphan entry
10076 */
10077 trans = btrfs_start_transaction(root, 5);
10078 if (IS_ERR(trans))
10079 return PTR_ERR(trans);
10080
10081 ret = btrfs_find_free_ino(root, &objectid);
10082 if (ret)
10083 goto out;
10084
10085 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10086 btrfs_ino(dir), objectid, mode, &index);
10087 if (IS_ERR(inode)) {
10088 ret = PTR_ERR(inode);
10089 inode = NULL;
10090 goto out;
10091 }
10092
Filipe Mananaef3b9af2014-04-27 20:40:45 +010010093 inode->i_fop = &btrfs_file_operations;
10094 inode->i_op = &btrfs_file_inode_operations;
10095
10096 inode->i_mapping->a_ops = &btrfs_aops;
Filipe Mananaef3b9af2014-04-27 20:40:45 +010010097 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10098
Chris Masonb0d5d102014-09-08 13:08:51 -070010099 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10100 if (ret)
10101 goto out_inode;
10102
10103 ret = btrfs_update_inode(trans, root, inode);
10104 if (ret)
10105 goto out_inode;
Filipe Mananaef3b9af2014-04-27 20:40:45 +010010106 ret = btrfs_orphan_add(trans, inode);
10107 if (ret)
Chris Masonb0d5d102014-09-08 13:08:51 -070010108 goto out_inode;
Filipe Mananaef3b9af2014-04-27 20:40:45 +010010109
Filipe Manana5762b5c2014-08-01 00:10:32 +010010110 /*
10111 * We set number of links to 0 in btrfs_new_inode(), and here we set
10112 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10113 * through:
10114 *
10115 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10116 */
10117 set_nlink(inode, 1);
Chris Masonb0d5d102014-09-08 13:08:51 -070010118 unlock_new_inode(inode);
Filipe Mananaef3b9af2014-04-27 20:40:45 +010010119 d_tmpfile(dentry, inode);
10120 mark_inode_dirty(inode);
10121
10122out:
10123 btrfs_end_transaction(trans, root);
10124 if (ret)
10125 iput(inode);
10126 btrfs_balance_delayed_items(root);
10127 btrfs_btree_balance_dirty(root);
Filipe Mananaef3b9af2014-04-27 20:40:45 +010010128 return ret;
Chris Masonb0d5d102014-09-08 13:08:51 -070010129
10130out_inode:
10131 unlock_new_inode(inode);
10132 goto out;
10133
Filipe Mananaef3b9af2014-04-27 20:40:45 +010010134}
10135
Filipe Mananab38ef712014-11-13 17:01:45 +000010136/* Inspired by filemap_check_errors() */
10137int btrfs_inode_check_errors(struct inode *inode)
10138{
10139 int ret = 0;
10140
10141 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
10142 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
10143 ret = -ENOSPC;
10144 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
10145 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
10146 ret = -EIO;
10147
10148 return ret;
10149}
10150
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -070010151static const struct inode_operations btrfs_dir_inode_operations = {
Chris Mason3394e162008-11-17 20:42:26 -050010152 .getattr = btrfs_getattr,
Chris Mason39279cc2007-06-12 06:35:45 -040010153 .lookup = btrfs_lookup,
10154 .create = btrfs_create,
10155 .unlink = btrfs_unlink,
10156 .link = btrfs_link,
10157 .mkdir = btrfs_mkdir,
10158 .rmdir = btrfs_rmdir,
Miklos Szeredi80ace852014-07-23 15:15:32 +020010159 .rename2 = btrfs_rename2,
Chris Mason39279cc2007-06-12 06:35:45 -040010160 .symlink = btrfs_symlink,
10161 .setattr = btrfs_setattr,
Josef Bacik618e21d2007-07-11 10:18:17 -040010162 .mknod = btrfs_mknod,
Christoph Hellwig95819c02008-08-28 06:21:17 -040010163 .setxattr = btrfs_setxattr,
Andreas Gruenbacher9172abb2015-12-02 14:44:37 +010010164 .getxattr = generic_getxattr,
Josef Bacik5103e942007-11-16 11:45:54 -050010165 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -040010166 .removexattr = btrfs_removexattr,
Yanfdebe2b2008-01-14 13:26:08 -050010167 .permission = btrfs_permission,
Christoph Hellwig4e34e712011-07-23 17:37:31 +020010168 .get_acl = btrfs_get_acl,
Christoph Hellwig996a7102013-12-20 05:16:43 -080010169 .set_acl = btrfs_set_acl,
Guangyu Sun93fd63c2013-09-16 10:42:03 -070010170 .update_time = btrfs_update_time,
Filipe Mananaef3b9af2014-04-27 20:40:45 +010010171 .tmpfile = btrfs_tmpfile,
Chris Mason39279cc2007-06-12 06:35:45 -040010172};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -070010173static const struct inode_operations btrfs_dir_ro_inode_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -040010174 .lookup = btrfs_lookup,
Yanfdebe2b2008-01-14 13:26:08 -050010175 .permission = btrfs_permission,
Christoph Hellwig4e34e712011-07-23 17:37:31 +020010176 .get_acl = btrfs_get_acl,
Christoph Hellwig996a7102013-12-20 05:16:43 -080010177 .set_acl = btrfs_set_acl,
Guangyu Sun93fd63c2013-09-16 10:42:03 -070010178 .update_time = btrfs_update_time,
Chris Mason39279cc2007-06-12 06:35:45 -040010179};
Yan, Zheng76dda932009-09-21 16:00:26 -040010180
Alexey Dobriyan828c0952009-10-01 15:43:56 -070010181static const struct file_operations btrfs_dir_file_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -040010182 .llseek = generic_file_llseek,
10183 .read = generic_read_dir,
Al Viro9cdda8d2013-05-22 16:48:09 -040010184 .iterate = btrfs_real_readdir,
Christoph Hellwig34287aa2007-09-14 10:22:47 -040010185 .unlocked_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -040010186#ifdef CONFIG_COMPAT
Christoph Hellwig34287aa2007-09-14 10:22:47 -040010187 .compat_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -040010188#endif
Sage Weil6bf13c02008-06-10 10:07:39 -040010189 .release = btrfs_release_file,
Chris Masone02119d2008-09-05 16:13:11 -040010190 .fsync = btrfs_sync_file,
Chris Mason39279cc2007-06-12 06:35:45 -040010191};
10192
David Sterba20e55062015-11-19 11:42:28 +010010193static const struct extent_io_ops btrfs_extent_io_ops = {
Chris Mason07157aa2007-08-30 08:50:51 -040010194 .fill_delalloc = run_delalloc_range,
Chris Mason065631f2008-02-20 12:07:25 -050010195 .submit_bio_hook = btrfs_submit_bio_hook,
Chris Mason239b14b2008-03-24 15:02:07 -040010196 .merge_bio_hook = btrfs_merge_bio_hook,
Chris Mason07157aa2007-08-30 08:50:51 -040010197 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
Chris Masone6dcd2d2008-07-17 12:53:50 -040010198 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
Chris Mason247e7432008-07-17 12:53:51 -040010199 .writepage_start_hook = btrfs_writepage_start_hook,
Chris Masonb0c68f82008-01-31 11:05:37 -050010200 .set_bit_hook = btrfs_set_bit_hook,
10201 .clear_bit_hook = btrfs_clear_bit_hook,
Josef Bacik9ed74f22009-09-11 16:12:44 -040010202 .merge_extent_hook = btrfs_merge_extent_hook,
10203 .split_extent_hook = btrfs_split_extent_hook,
Chris Mason07157aa2007-08-30 08:50:51 -040010204};
10205
Chris Mason35054392009-01-21 13:11:13 -050010206/*
10207 * btrfs doesn't support the bmap operation because swapfiles
10208 * use bmap to make a mapping of extents in the file. They assume
10209 * these extents won't change over the life of the file and they
10210 * use the bmap result to do IO directly to the drive.
10211 *
10212 * the btrfs bmap call would return logical addresses that aren't
10213 * suitable for IO and they also will change frequently as COW
10214 * operations happen. So, swapfile + btrfs == corruption.
10215 *
10216 * For now we're avoiding this by dropping bmap.
10217 */
Alexey Dobriyan7f094102009-09-21 17:01:10 -070010218static const struct address_space_operations btrfs_aops = {
Chris Mason39279cc2007-06-12 06:35:45 -040010219 .readpage = btrfs_readpage,
10220 .writepage = btrfs_writepage,
Chris Masonb293f022007-11-01 19:45:34 -040010221 .writepages = btrfs_writepages,
Chris Mason3ab2fb52007-11-08 10:59:22 -050010222 .readpages = btrfs_readpages,
Chris Mason16432982008-04-10 10:23:21 -040010223 .direct_IO = btrfs_direct_IO,
Chris Masona52d9a82007-08-27 16:49:44 -040010224 .invalidatepage = btrfs_invalidatepage,
10225 .releasepage = btrfs_releasepage,
Chris Masone6dcd2d2008-07-17 12:53:50 -040010226 .set_page_dirty = btrfs_set_page_dirty,
Andi Kleen465fdd92009-09-16 11:50:18 +020010227 .error_remove_page = generic_error_remove_page,
Chris Mason39279cc2007-06-12 06:35:45 -040010228};
10229
Alexey Dobriyan7f094102009-09-21 17:01:10 -070010230static const struct address_space_operations btrfs_symlink_aops = {
Chris Mason39279cc2007-06-12 06:35:45 -040010231 .readpage = btrfs_readpage,
10232 .writepage = btrfs_writepage,
Chris Mason2bf5a722007-08-30 11:54:02 -040010233 .invalidatepage = btrfs_invalidatepage,
10234 .releasepage = btrfs_releasepage,
Chris Mason39279cc2007-06-12 06:35:45 -040010235};
10236
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -070010237static const struct inode_operations btrfs_file_inode_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -040010238 .getattr = btrfs_getattr,
10239 .setattr = btrfs_setattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -040010240 .setxattr = btrfs_setxattr,
Andreas Gruenbacher9172abb2015-12-02 14:44:37 +010010241 .getxattr = generic_getxattr,
Josef Bacik5103e942007-11-16 11:45:54 -050010242 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -040010243 .removexattr = btrfs_removexattr,
Yanfdebe2b2008-01-14 13:26:08 -050010244 .permission = btrfs_permission,
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -050010245 .fiemap = btrfs_fiemap,
Christoph Hellwig4e34e712011-07-23 17:37:31 +020010246 .get_acl = btrfs_get_acl,
Christoph Hellwig996a7102013-12-20 05:16:43 -080010247 .set_acl = btrfs_set_acl,
Josef Bacike41f9412012-03-26 09:46:47 -040010248 .update_time = btrfs_update_time,
Chris Mason39279cc2007-06-12 06:35:45 -040010249};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -070010250static const struct inode_operations btrfs_special_inode_operations = {
Josef Bacik618e21d2007-07-11 10:18:17 -040010251 .getattr = btrfs_getattr,
10252 .setattr = btrfs_setattr,
Yanfdebe2b2008-01-14 13:26:08 -050010253 .permission = btrfs_permission,
Christoph Hellwig95819c02008-08-28 06:21:17 -040010254 .setxattr = btrfs_setxattr,
Andreas Gruenbacher9172abb2015-12-02 14:44:37 +010010255 .getxattr = generic_getxattr,
Josef Bacik33268ea2008-07-24 12:16:36 -040010256 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -040010257 .removexattr = btrfs_removexattr,
Christoph Hellwig4e34e712011-07-23 17:37:31 +020010258 .get_acl = btrfs_get_acl,
Christoph Hellwig996a7102013-12-20 05:16:43 -080010259 .set_acl = btrfs_set_acl,
Josef Bacike41f9412012-03-26 09:46:47 -040010260 .update_time = btrfs_update_time,
Josef Bacik618e21d2007-07-11 10:18:17 -040010261};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -070010262static const struct inode_operations btrfs_symlink_inode_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -040010263 .readlink = generic_readlink,
Al Viro6b255392015-11-17 10:20:54 -050010264 .get_link = page_get_link,
Li Zefanf2095612010-11-19 02:05:24 +000010265 .getattr = btrfs_getattr,
Josef Bacik22c44fe2011-11-30 10:45:38 -050010266 .setattr = btrfs_setattr,
Yanfdebe2b2008-01-14 13:26:08 -050010267 .permission = btrfs_permission,
Jim Owens0279b4c2009-02-04 09:29:13 -050010268 .setxattr = btrfs_setxattr,
Andreas Gruenbacher9172abb2015-12-02 14:44:37 +010010269 .getxattr = generic_getxattr,
Jim Owens0279b4c2009-02-04 09:29:13 -050010270 .listxattr = btrfs_listxattr,
10271 .removexattr = btrfs_removexattr,
Josef Bacike41f9412012-03-26 09:46:47 -040010272 .update_time = btrfs_update_time,
Chris Mason39279cc2007-06-12 06:35:45 -040010273};
Yan, Zheng76dda932009-09-21 16:00:26 -040010274
Alexey Dobriyan82d339d2009-10-09 09:54:36 -040010275const struct dentry_operations btrfs_dentry_operations = {
Yan, Zheng76dda932009-09-21 16:00:26 -040010276 .d_delete = btrfs_dentry_delete,
Josef Bacikb4aff1f2011-06-28 16:18:59 -040010277 .d_release = btrfs_dentry_release,
Yan, Zheng76dda932009-09-21 16:00:26 -040010278};