blob: 54bcf639d1cf1f48df951477b16e6266402f7bc6 [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>
Kent Overstreeta27bb332013-05-07 16:19:08 -070035#include <linux/aio.h>
Chris Mason9ebefb182007-06-15 13:50:00 -040036#include <linux/bit_spinlock.h>
Josef Bacik5103e942007-11-16 11:45:54 -050037#include <linux/xattr.h>
Josef Bacik33268ea2008-07-24 12:16:36 -040038#include <linux/posix_acl.h>
Yan Zhengd899e052008-10-30 14:25:28 -040039#include <linux/falloc.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090040#include <linux/slab.h>
David Sterba7a36dde2011-05-06 15:33:15 +020041#include <linux/ratelimit.h>
Josef Bacik22c44fe2011-11-30 10:45:38 -050042#include <linux/mount.h>
Filipe Brandenburger55e301f2013-01-29 06:04:50 +000043#include <linux/btrfs.h>
David Woodhouse53b381b2013-01-29 18:40:14 -050044#include <linux/blkdev.h>
Josef Bacikf23b5a52013-06-19 10:16:26 -040045#include <linux/posix_acl_xattr.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"
Chris Mason39279cc2007-06-12 06:35:45 -040062
63struct btrfs_iget_args {
Chris Mason90d3e592014-01-09 17:28:00 -080064 struct btrfs_key *location;
Chris Mason39279cc2007-06-12 06:35:45 -040065 struct btrfs_root *root;
66};
67
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -070068static const struct inode_operations btrfs_dir_inode_operations;
69static const struct inode_operations btrfs_symlink_inode_operations;
70static const struct inode_operations btrfs_dir_ro_inode_operations;
71static const struct inode_operations btrfs_special_inode_operations;
72static const struct inode_operations btrfs_file_inode_operations;
Alexey Dobriyan7f094102009-09-21 17:01:10 -070073static const struct address_space_operations btrfs_aops;
74static const struct address_space_operations btrfs_symlink_aops;
Alexey Dobriyan828c0952009-10-01 15:43:56 -070075static const struct file_operations btrfs_dir_file_operations;
Chris Masond1310b22008-01-24 16:13:08 -050076static struct extent_io_ops btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -040077
78static struct kmem_cache *btrfs_inode_cachep;
Miao Xie8ccf6f192012-10-25 09:28:04 +000079static struct kmem_cache *btrfs_delalloc_work_cachep;
Chris Mason39279cc2007-06-12 06:35:45 -040080struct kmem_cache *btrfs_trans_handle_cachep;
81struct kmem_cache *btrfs_transaction_cachep;
Chris Mason39279cc2007-06-12 06:35:45 -040082struct kmem_cache *btrfs_path_cachep;
Josef Bacikdc89e982011-01-28 17:05:48 -050083struct kmem_cache *btrfs_free_space_cachep;
Chris Mason39279cc2007-06-12 06:35:45 -040084
85#define S_SHIFT 12
86static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
87 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
88 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
89 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
90 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
91 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
92 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
93 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
94};
95
Eric Sandeen3972f262013-01-12 02:57:22 +000096static int btrfs_setsize(struct inode *inode, struct iattr *attr);
Josef Bacika41ad392011-01-31 15:30:16 -050097static int btrfs_truncate(struct inode *inode);
Josef Bacik5fd02042012-05-02 14:00:54 -040098static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
Chris Mason771ed682008-11-06 22:02:51 -050099static noinline int cow_file_range(struct inode *inode,
100 struct page *locked_page,
101 u64 start, u64 end, int *page_started,
102 unsigned long *nr_written, int unlock);
Josef Bacik70c8a912012-10-11 16:54:30 -0400103static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
104 u64 len, u64 orig_start,
105 u64 block_start, u64 block_len,
Josef Bacikcc95bef2013-04-04 14:31:27 -0400106 u64 orig_block_len, u64 ram_bytes,
107 int type);
Josef Bacik7b128762008-07-24 12:17:14 -0400108
Eric Sandeen48a3b632013-04-25 20:41:01 +0000109static int btrfs_dirty_inode(struct inode *inode);
Chris Mason39279cc2007-06-12 06:35:45 -0400110
Yan, Zhengf34f57a2009-11-12 09:35:27 +0000111static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
Eric Paris2a7dba32011-02-01 11:05:39 -0500112 struct inode *inode, struct inode *dir,
113 const struct qstr *qstr)
Jim Owens0279b4c2009-02-04 09:29:13 -0500114{
115 int err;
116
Yan, Zhengf34f57a2009-11-12 09:35:27 +0000117 err = btrfs_init_acl(trans, inode, dir);
Jim Owens0279b4c2009-02-04 09:29:13 -0500118 if (!err)
Eric Paris2a7dba32011-02-01 11:05:39 -0500119 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
Jim Owens0279b4c2009-02-04 09:29:13 -0500120 return err;
121}
122
Chris Masond352ac62008-09-29 15:18:18 -0400123/*
Chris Masonc8b97812008-10-29 14:49:59 -0400124 * this does all the hard work for inserting an inline extent into
125 * the btree. The caller should have done a btrfs_drop_extents so that
126 * no overlapping inline items exist in the btree
127 */
Chris Mason40f76582014-05-21 13:35:51 -0700128static int insert_inline_extent(struct btrfs_trans_handle *trans,
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000129 struct btrfs_path *path, int extent_inserted,
Chris Masonc8b97812008-10-29 14:49:59 -0400130 struct btrfs_root *root, struct inode *inode,
131 u64 start, size_t size, size_t compressed_size,
Li Zefanfe3f5662011-03-28 08:30:38 +0000132 int compress_type,
Chris Masonc8b97812008-10-29 14:49:59 -0400133 struct page **compressed_pages)
134{
Chris Masonc8b97812008-10-29 14:49:59 -0400135 struct extent_buffer *leaf;
136 struct page *page = NULL;
137 char *kaddr;
138 unsigned long ptr;
139 struct btrfs_file_extent_item *ei;
140 int err = 0;
141 int ret;
142 size_t cur_size = size;
Chris Masonc8b97812008-10-29 14:49:59 -0400143 unsigned long offset;
Chris Masonc8b97812008-10-29 14:49:59 -0400144
Li Zefanfe3f5662011-03-28 08:30:38 +0000145 if (compressed_size && compressed_pages)
Chris Masonc8b97812008-10-29 14:49:59 -0400146 cur_size = compressed_size;
Chris Masonc8b97812008-10-29 14:49:59 -0400147
Chris Masonc8b97812008-10-29 14:49:59 -0400148 inode_add_bytes(inode, size);
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000149
150 if (!extent_inserted) {
151 struct btrfs_key key;
152 size_t datasize;
153
154 key.objectid = btrfs_ino(inode);
155 key.offset = start;
David Sterba962a2982014-06-04 18:41:45 +0200156 key.type = BTRFS_EXTENT_DATA_KEY;
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000157
158 datasize = btrfs_file_extent_calc_inline_size(cur_size);
159 path->leave_spinning = 1;
160 ret = btrfs_insert_empty_item(trans, root, path, &key,
161 datasize);
162 if (ret) {
163 err = ret;
164 goto fail;
165 }
Chris Masonc8b97812008-10-29 14:49:59 -0400166 }
167 leaf = path->nodes[0];
168 ei = btrfs_item_ptr(leaf, path->slots[0],
169 struct btrfs_file_extent_item);
170 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
171 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
172 btrfs_set_file_extent_encryption(leaf, ei, 0);
173 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
174 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
175 ptr = btrfs_file_extent_inline_start(ei);
176
Li Zefan261507a02010-12-17 14:21:50 +0800177 if (compress_type != BTRFS_COMPRESS_NONE) {
Chris Masonc8b97812008-10-29 14:49:59 -0400178 struct page *cpage;
179 int i = 0;
Chris Masond3977122009-01-05 21:25:51 -0500180 while (compressed_size > 0) {
Chris Masonc8b97812008-10-29 14:49:59 -0400181 cpage = compressed_pages[i];
Chris Mason5b050f02008-11-11 09:34:41 -0500182 cur_size = min_t(unsigned long, compressed_size,
Chris Masonc8b97812008-10-29 14:49:59 -0400183 PAGE_CACHE_SIZE);
184
Cong Wang7ac687d2011-11-25 23:14:28 +0800185 kaddr = kmap_atomic(cpage);
Chris Masonc8b97812008-10-29 14:49:59 -0400186 write_extent_buffer(leaf, kaddr, ptr, cur_size);
Cong Wang7ac687d2011-11-25 23:14:28 +0800187 kunmap_atomic(kaddr);
Chris Masonc8b97812008-10-29 14:49:59 -0400188
189 i++;
190 ptr += cur_size;
191 compressed_size -= cur_size;
192 }
193 btrfs_set_file_extent_compression(leaf, ei,
Li Zefan261507a02010-12-17 14:21:50 +0800194 compress_type);
Chris Masonc8b97812008-10-29 14:49:59 -0400195 } else {
196 page = find_get_page(inode->i_mapping,
197 start >> PAGE_CACHE_SHIFT);
198 btrfs_set_file_extent_compression(leaf, ei, 0);
Cong Wang7ac687d2011-11-25 23:14:28 +0800199 kaddr = kmap_atomic(page);
Chris Masonc8b97812008-10-29 14:49:59 -0400200 offset = start & (PAGE_CACHE_SIZE - 1);
201 write_extent_buffer(leaf, kaddr + offset, ptr, size);
Cong Wang7ac687d2011-11-25 23:14:28 +0800202 kunmap_atomic(kaddr);
Chris Masonc8b97812008-10-29 14:49:59 -0400203 page_cache_release(page);
204 }
205 btrfs_mark_buffer_dirty(leaf);
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000206 btrfs_release_path(path);
Chris Masonc8b97812008-10-29 14:49:59 -0400207
Yan, Zhengc2167752009-11-12 09:34:21 +0000208 /*
209 * we're an inline extent, so nobody can
210 * extend the file past i_size without locking
211 * a page we already have locked.
212 *
213 * We must do any isize and inode updates
214 * before we unlock the pages. Otherwise we
215 * could end up racing with unlink.
216 */
Chris Masonc8b97812008-10-29 14:49:59 -0400217 BTRFS_I(inode)->disk_i_size = inode->i_size;
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100218 ret = btrfs_update_inode(trans, root, inode);
Yan, Zhengc2167752009-11-12 09:34:21 +0000219
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100220 return ret;
Chris Masonc8b97812008-10-29 14:49:59 -0400221fail:
Chris Masonc8b97812008-10-29 14:49:59 -0400222 return err;
223}
224
225
226/*
227 * conditionally insert an inline extent into the file. This
228 * does the checks required to make sure the data is small enough
229 * to fit as an inline extent.
230 */
Josef Bacik00361582013-08-14 14:02:47 -0400231static noinline int cow_file_range_inline(struct btrfs_root *root,
232 struct inode *inode, u64 start,
233 u64 end, size_t compressed_size,
234 int compress_type,
235 struct page **compressed_pages)
Chris Masonc8b97812008-10-29 14:49:59 -0400236{
Josef Bacik00361582013-08-14 14:02:47 -0400237 struct btrfs_trans_handle *trans;
Chris Masonc8b97812008-10-29 14:49:59 -0400238 u64 isize = i_size_read(inode);
239 u64 actual_end = min(end + 1, isize);
240 u64 inline_len = actual_end - start;
Qu Wenruofda28322013-02-26 08:10:22 +0000241 u64 aligned_end = ALIGN(end, root->sectorsize);
Chris Masonc8b97812008-10-29 14:49:59 -0400242 u64 data_len = inline_len;
243 int ret;
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000244 struct btrfs_path *path;
245 int extent_inserted = 0;
246 u32 extent_item_size;
Chris Masonc8b97812008-10-29 14:49:59 -0400247
248 if (compressed_size)
249 data_len = compressed_size;
250
251 if (start > 0 ||
Wang Shilong354877b2014-07-17 11:44:11 +0800252 actual_end > PAGE_CACHE_SIZE ||
253 data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
Chris Masonc8b97812008-10-29 14:49:59 -0400254 (!compressed_size &&
255 (actual_end & (root->sectorsize - 1)) == 0) ||
256 end + 1 < isize ||
257 data_len > root->fs_info->max_inline) {
258 return 1;
259 }
260
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000261 path = btrfs_alloc_path();
262 if (!path)
263 return -ENOMEM;
264
Josef Bacik00361582013-08-14 14:02:47 -0400265 trans = btrfs_join_transaction(root);
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000266 if (IS_ERR(trans)) {
267 btrfs_free_path(path);
Josef Bacik00361582013-08-14 14:02:47 -0400268 return PTR_ERR(trans);
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000269 }
Josef Bacik00361582013-08-14 14:02:47 -0400270 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
271
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000272 if (compressed_size && compressed_pages)
273 extent_item_size = btrfs_file_extent_calc_inline_size(
274 compressed_size);
275 else
276 extent_item_size = btrfs_file_extent_calc_inline_size(
277 inline_len);
278
279 ret = __btrfs_drop_extents(trans, root, inode, path,
280 start, aligned_end, NULL,
281 1, 1, extent_item_size, &extent_inserted);
Josef Bacik00361582013-08-14 14:02:47 -0400282 if (ret) {
283 btrfs_abort_transaction(trans, root, ret);
284 goto out;
285 }
Chris Masonc8b97812008-10-29 14:49:59 -0400286
287 if (isize > actual_end)
288 inline_len = min_t(u64, isize, actual_end);
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000289 ret = insert_inline_extent(trans, path, extent_inserted,
290 root, inode, start,
Chris Masonc8b97812008-10-29 14:49:59 -0400291 inline_len, compressed_size,
Li Zefanfe3f5662011-03-28 08:30:38 +0000292 compress_type, compressed_pages);
Josef Bacik2adcac12012-05-23 16:10:14 -0400293 if (ret && ret != -ENOSPC) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100294 btrfs_abort_transaction(trans, root, ret);
Josef Bacik00361582013-08-14 14:02:47 -0400295 goto out;
Josef Bacik2adcac12012-05-23 16:10:14 -0400296 } else if (ret == -ENOSPC) {
Josef Bacik00361582013-08-14 14:02:47 -0400297 ret = 1;
298 goto out;
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100299 }
Josef Bacik2adcac12012-05-23 16:10:14 -0400300
Josef Bacikbdc20e62013-02-28 13:23:38 -0500301 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -0400302 btrfs_delalloc_release_metadata(inode, end + 1 - start);
Chris Masona1ed8352009-09-11 12:27:37 -0400303 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
Josef Bacik00361582013-08-14 14:02:47 -0400304out:
Filipe David Borba Manana1acae572014-01-07 11:42:27 +0000305 btrfs_free_path(path);
Josef Bacik00361582013-08-14 14:02:47 -0400306 btrfs_end_transaction(trans, root);
307 return ret;
Chris Masonc8b97812008-10-29 14:49:59 -0400308}
309
Chris Mason771ed682008-11-06 22:02:51 -0500310struct async_extent {
311 u64 start;
312 u64 ram_size;
313 u64 compressed_size;
314 struct page **pages;
315 unsigned long nr_pages;
Li Zefan261507a02010-12-17 14:21:50 +0800316 int compress_type;
Chris Mason771ed682008-11-06 22:02:51 -0500317 struct list_head list;
318};
319
320struct async_cow {
321 struct inode *inode;
322 struct btrfs_root *root;
323 struct page *locked_page;
324 u64 start;
325 u64 end;
326 struct list_head extents;
327 struct btrfs_work work;
328};
329
330static noinline int add_async_extent(struct async_cow *cow,
331 u64 start, u64 ram_size,
332 u64 compressed_size,
333 struct page **pages,
Li Zefan261507a02010-12-17 14:21:50 +0800334 unsigned long nr_pages,
335 int compress_type)
Chris Mason771ed682008-11-06 22:02:51 -0500336{
337 struct async_extent *async_extent;
338
339 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100340 BUG_ON(!async_extent); /* -ENOMEM */
Chris Mason771ed682008-11-06 22:02:51 -0500341 async_extent->start = start;
342 async_extent->ram_size = ram_size;
343 async_extent->compressed_size = compressed_size;
344 async_extent->pages = pages;
345 async_extent->nr_pages = nr_pages;
Li Zefan261507a02010-12-17 14:21:50 +0800346 async_extent->compress_type = compress_type;
Chris Mason771ed682008-11-06 22:02:51 -0500347 list_add_tail(&async_extent->list, &cow->extents);
348 return 0;
349}
350
Wang Shilongf79707b2014-07-17 11:44:09 +0800351static inline int inode_need_compress(struct inode *inode)
352{
353 struct btrfs_root *root = BTRFS_I(inode)->root;
354
355 /* force compress */
356 if (btrfs_test_opt(root, FORCE_COMPRESS))
357 return 1;
358 /* bad compression ratios */
359 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
360 return 0;
361 if (btrfs_test_opt(root, COMPRESS) ||
362 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
363 BTRFS_I(inode)->force_compress)
364 return 1;
365 return 0;
366}
367
Chris Masonc8b97812008-10-29 14:49:59 -0400368/*
Chris Mason771ed682008-11-06 22:02:51 -0500369 * we create compressed extents in two phases. The first
370 * phase compresses a range of pages that have already been
371 * locked (both pages and state bits are locked).
Chris Masonc8b97812008-10-29 14:49:59 -0400372 *
Chris Mason771ed682008-11-06 22:02:51 -0500373 * This is done inside an ordered work queue, and the compression
374 * is spread across many cpus. The actual IO submission is step
375 * two, and the ordered work queue takes care of making sure that
376 * happens in the same order things were put onto the queue by
377 * writepages and friends.
Chris Masonc8b97812008-10-29 14:49:59 -0400378 *
Chris Mason771ed682008-11-06 22:02:51 -0500379 * If this code finds it can't get good compression, it puts an
380 * entry onto the work queue to write the uncompressed bytes. This
381 * makes sure that both compressed inodes and uncompressed inodes
Artem Bityutskiyb2570312012-07-25 18:12:06 +0300382 * are written in the same order that the flusher thread sent them
383 * down.
Chris Masond352ac62008-09-29 15:18:18 -0400384 */
Filipe Mananac44f6492014-10-09 21:15:44 +0100385static noinline void compress_file_range(struct inode *inode,
Chris Mason771ed682008-11-06 22:02:51 -0500386 struct page *locked_page,
387 u64 start, u64 end,
388 struct async_cow *async_cow,
389 int *num_added)
Chris Masonb888db22007-08-27 16:49:44 -0400390{
391 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masondb945352007-10-15 16:15:53 -0400392 u64 num_bytes;
Chris Masondb945352007-10-15 16:15:53 -0400393 u64 blocksize = root->sectorsize;
Chris Masonc8b97812008-10-29 14:49:59 -0400394 u64 actual_end;
Chris Mason42dc7ba2008-12-15 11:44:56 -0500395 u64 isize = i_size_read(inode);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400396 int ret = 0;
Chris Masonc8b97812008-10-29 14:49:59 -0400397 struct page **pages = NULL;
398 unsigned long nr_pages;
399 unsigned long nr_pages_ret = 0;
400 unsigned long total_compressed = 0;
401 unsigned long total_in = 0;
402 unsigned long max_compressed = 128 * 1024;
Chris Mason771ed682008-11-06 22:02:51 -0500403 unsigned long max_uncompressed = 128 * 1024;
Chris Masonc8b97812008-10-29 14:49:59 -0400404 int i;
405 int will_compress;
Li Zefan261507a02010-12-17 14:21:50 +0800406 int compress_type = root->fs_info->compress_type;
Chris Mason4adaa612013-03-26 13:07:00 -0400407 int redirty = 0;
Chris Masonb888db22007-08-27 16:49:44 -0400408
Liu Bo4cb13e52012-03-29 09:57:45 -0400409 /* if this is a small write inside eof, kick off a defrag */
410 if ((end - start + 1) < 16 * 1024 &&
411 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
Chris Mason4cb53002011-05-24 15:35:30 -0400412 btrfs_add_inode_defrag(NULL, inode);
413
Chris Mason42dc7ba2008-12-15 11:44:56 -0500414 actual_end = min_t(u64, isize, end + 1);
Chris Masonc8b97812008-10-29 14:49:59 -0400415again:
416 will_compress = 0;
417 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
418 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
419
Chris Masonf03d9301f2009-02-04 09:31:06 -0500420 /*
421 * we don't want to send crud past the end of i_size through
422 * compression, that's just a waste of CPU time. So, if the
423 * end of the file is before the start of our current
424 * requested range of bytes, we bail out to the uncompressed
425 * cleanup code that can deal with all of this.
426 *
427 * It isn't really the fastest way to fix things, but this is a
428 * very uncommon corner.
429 */
430 if (actual_end <= start)
431 goto cleanup_and_bail_uncompressed;
432
Chris Masonc8b97812008-10-29 14:49:59 -0400433 total_compressed = actual_end - start;
434
Shilong Wang4bcbb332014-10-07 18:44:35 -0400435 /*
436 * skip compression for a small file range(<=blocksize) that
437 * isn't an inline extent, since it dosen't save disk space at all.
438 */
439 if (total_compressed <= blocksize &&
440 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
441 goto cleanup_and_bail_uncompressed;
442
Chris Masonc8b97812008-10-29 14:49:59 -0400443 /* we want to make sure that amount of ram required to uncompress
444 * an extent is reasonable, so we limit the total size in ram
Chris Mason771ed682008-11-06 22:02:51 -0500445 * of a compressed extent to 128k. This is a crucial number
446 * because it also controls how easily we can spread reads across
447 * cpus for decompression.
448 *
449 * We also want to make sure the amount of IO required to do
450 * a random read is reasonably small, so we limit the size of
451 * a compressed extent to 128k.
Chris Masonc8b97812008-10-29 14:49:59 -0400452 */
453 total_compressed = min(total_compressed, max_uncompressed);
Qu Wenruofda28322013-02-26 08:10:22 +0000454 num_bytes = ALIGN(end - start + 1, blocksize);
Chris Masonbe20aa92007-12-17 20:14:01 -0500455 num_bytes = max(blocksize, num_bytes);
Chris Masonc8b97812008-10-29 14:49:59 -0400456 total_in = 0;
457 ret = 0;
Chris Masondb945352007-10-15 16:15:53 -0400458
Chris Mason771ed682008-11-06 22:02:51 -0500459 /*
460 * we do compression for mount -o compress and when the
461 * inode has not been flagged as nocompress. This flag can
462 * change at any time if we discover bad compression ratios.
Chris Masonc8b97812008-10-29 14:49:59 -0400463 */
Wang Shilongf79707b2014-07-17 11:44:09 +0800464 if (inode_need_compress(inode)) {
Chris Masonc8b97812008-10-29 14:49:59 -0400465 WARN_ON(pages);
Chris Masoncfbc2462008-10-30 13:22:14 -0400466 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
Li Zefan560f7d72011-09-08 10:22:01 +0800467 if (!pages) {
468 /* just bail out to the uncompressed code */
469 goto cont;
470 }
Chris Mason179e29e2007-11-01 11:28:41 -0400471
Li Zefan261507a02010-12-17 14:21:50 +0800472 if (BTRFS_I(inode)->force_compress)
473 compress_type = BTRFS_I(inode)->force_compress;
474
Chris Mason4adaa612013-03-26 13:07:00 -0400475 /*
476 * we need to call clear_page_dirty_for_io on each
477 * page in the range. Otherwise applications with the file
478 * mmap'd can wander in and change the page contents while
479 * we are compressing them.
480 *
481 * If the compression fails for any reason, we set the pages
482 * dirty again later on.
483 */
484 extent_range_clear_dirty_for_io(inode, start, end);
485 redirty = 1;
Li Zefan261507a02010-12-17 14:21:50 +0800486 ret = btrfs_compress_pages(compress_type,
487 inode->i_mapping, start,
488 total_compressed, pages,
489 nr_pages, &nr_pages_ret,
490 &total_in,
491 &total_compressed,
492 max_compressed);
Chris Masonc8b97812008-10-29 14:49:59 -0400493
494 if (!ret) {
495 unsigned long offset = total_compressed &
496 (PAGE_CACHE_SIZE - 1);
497 struct page *page = pages[nr_pages_ret - 1];
498 char *kaddr;
499
500 /* zero the tail end of the last page, we might be
501 * sending it down to disk
502 */
503 if (offset) {
Cong Wang7ac687d2011-11-25 23:14:28 +0800504 kaddr = kmap_atomic(page);
Chris Masonc8b97812008-10-29 14:49:59 -0400505 memset(kaddr + offset, 0,
506 PAGE_CACHE_SIZE - offset);
Cong Wang7ac687d2011-11-25 23:14:28 +0800507 kunmap_atomic(kaddr);
Chris Masonc8b97812008-10-29 14:49:59 -0400508 }
509 will_compress = 1;
510 }
511 }
Li Zefan560f7d72011-09-08 10:22:01 +0800512cont:
Chris Masonc8b97812008-10-29 14:49:59 -0400513 if (start == 0) {
514 /* lets try to make an inline extent */
Chris Mason771ed682008-11-06 22:02:51 -0500515 if (ret || total_in < (actual_end - start)) {
Chris Masonc8b97812008-10-29 14:49:59 -0400516 /* we didn't compress the entire range, try
Chris Mason771ed682008-11-06 22:02:51 -0500517 * to make an uncompressed inline extent.
Chris Masonc8b97812008-10-29 14:49:59 -0400518 */
Josef Bacik00361582013-08-14 14:02:47 -0400519 ret = cow_file_range_inline(root, inode, start, end,
520 0, 0, NULL);
Chris Masonc8b97812008-10-29 14:49:59 -0400521 } else {
Chris Mason771ed682008-11-06 22:02:51 -0500522 /* try making a compressed inline extent */
Josef Bacik00361582013-08-14 14:02:47 -0400523 ret = cow_file_range_inline(root, inode, start, end,
Li Zefanfe3f5662011-03-28 08:30:38 +0000524 total_compressed,
525 compress_type, pages);
Chris Masonc8b97812008-10-29 14:49:59 -0400526 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100527 if (ret <= 0) {
Josef Bacik151a41b2013-07-29 13:22:24 -0400528 unsigned long clear_flags = EXTENT_DELALLOC |
529 EXTENT_DEFRAG;
Filipe Mananae6eb4312014-10-10 10:45:12 +0100530 unsigned long page_error_op;
531
Josef Bacik151a41b2013-07-29 13:22:24 -0400532 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
Filipe Mananae6eb4312014-10-10 10:45:12 +0100533 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
Josef Bacik151a41b2013-07-29 13:22:24 -0400534
Chris Mason771ed682008-11-06 22:02:51 -0500535 /*
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100536 * inline extent creation worked or returned error,
537 * we don't need to create any more async work items.
538 * Unlock and free up our temp pages.
Chris Mason771ed682008-11-06 22:02:51 -0500539 */
Josef Bacikc2790a22013-07-29 11:20:47 -0400540 extent_clear_unlock_delalloc(inode, start, end, NULL,
Josef Bacik151a41b2013-07-29 13:22:24 -0400541 clear_flags, PAGE_UNLOCK |
Josef Bacikc2790a22013-07-29 11:20:47 -0400542 PAGE_CLEAR_DIRTY |
543 PAGE_SET_WRITEBACK |
Filipe Mananae6eb4312014-10-10 10:45:12 +0100544 page_error_op |
Josef Bacikc2790a22013-07-29 11:20:47 -0400545 PAGE_END_WRITEBACK);
Chris Masonc8b97812008-10-29 14:49:59 -0400546 goto free_pages_out;
547 }
548 }
549
550 if (will_compress) {
551 /*
552 * we aren't doing an inline extent round the compressed size
553 * up to a block size boundary so the allocator does sane
554 * things
555 */
Qu Wenruofda28322013-02-26 08:10:22 +0000556 total_compressed = ALIGN(total_compressed, blocksize);
Chris Masonc8b97812008-10-29 14:49:59 -0400557
558 /*
559 * one last check to make sure the compression is really a
560 * win, compare the page count read with the blocks on disk
561 */
Qu Wenruofda28322013-02-26 08:10:22 +0000562 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
Chris Masonc8b97812008-10-29 14:49:59 -0400563 if (total_compressed >= total_in) {
564 will_compress = 0;
565 } else {
Chris Masonc8b97812008-10-29 14:49:59 -0400566 num_bytes = total_in;
567 }
568 }
569 if (!will_compress && pages) {
570 /*
571 * the compression code ran but failed to make things smaller,
572 * free any pages it allocated and our page pointer array
573 */
574 for (i = 0; i < nr_pages_ret; i++) {
Chris Mason70b99e62008-10-31 12:46:39 -0400575 WARN_ON(pages[i]->mapping);
Chris Masonc8b97812008-10-29 14:49:59 -0400576 page_cache_release(pages[i]);
577 }
578 kfree(pages);
579 pages = NULL;
580 total_compressed = 0;
581 nr_pages_ret = 0;
582
583 /* flag the file so we don't compress in the future */
Chris Mason1e701a32010-03-11 09:42:04 -0500584 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
585 !(BTRFS_I(inode)->force_compress)) {
Chris Masona555f812010-01-28 16:18:15 -0500586 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
Chris Mason1e701a32010-03-11 09:42:04 -0500587 }
Chris Masonc8b97812008-10-29 14:49:59 -0400588 }
Chris Mason771ed682008-11-06 22:02:51 -0500589 if (will_compress) {
590 *num_added += 1;
591
592 /* the async work queues will take care of doing actual
593 * allocation on disk for these compressed pages,
594 * and will submit them to the elevator.
595 */
596 add_async_extent(async_cow, start, num_bytes,
Li Zefan261507a02010-12-17 14:21:50 +0800597 total_compressed, pages, nr_pages_ret,
598 compress_type);
Chris Mason771ed682008-11-06 22:02:51 -0500599
Yan, Zheng24ae6362010-12-06 07:02:36 +0000600 if (start + num_bytes < end) {
Chris Mason771ed682008-11-06 22:02:51 -0500601 start += num_bytes;
602 pages = NULL;
603 cond_resched();
604 goto again;
605 }
606 } else {
Chris Masonf03d9301f2009-02-04 09:31:06 -0500607cleanup_and_bail_uncompressed:
Chris Mason771ed682008-11-06 22:02:51 -0500608 /*
609 * No compression, but we still need to write the pages in
610 * the file we've been given so far. redirty the locked
611 * page if it corresponds to our extent and set things up
612 * for the async work queue to run cow_file_range to do
613 * the normal delalloc dance
614 */
615 if (page_offset(locked_page) >= start &&
616 page_offset(locked_page) <= end) {
617 __set_page_dirty_nobuffers(locked_page);
618 /* unlocked later on in the async handlers */
619 }
Chris Mason4adaa612013-03-26 13:07:00 -0400620 if (redirty)
621 extent_range_redirty_for_io(inode, start, end);
Li Zefan261507a02010-12-17 14:21:50 +0800622 add_async_extent(async_cow, start, end - start + 1,
623 0, NULL, 0, BTRFS_COMPRESS_NONE);
Chris Mason771ed682008-11-06 22:02:51 -0500624 *num_added += 1;
625 }
626
Filipe Mananac44f6492014-10-09 21:15:44 +0100627 return;
Chris Mason771ed682008-11-06 22:02:51 -0500628
629free_pages_out:
630 for (i = 0; i < nr_pages_ret; i++) {
631 WARN_ON(pages[i]->mapping);
632 page_cache_release(pages[i]);
633 }
Chris Masond3977122009-01-05 21:25:51 -0500634 kfree(pages);
Chris Mason771ed682008-11-06 22:02:51 -0500635}
Chris Mason771ed682008-11-06 22:02:51 -0500636
Filipe Manana40ae8372014-10-06 22:14:24 +0100637static void free_async_extent_pages(struct async_extent *async_extent)
638{
639 int i;
640
641 if (!async_extent->pages)
642 return;
643
644 for (i = 0; i < async_extent->nr_pages; i++) {
645 WARN_ON(async_extent->pages[i]->mapping);
646 page_cache_release(async_extent->pages[i]);
647 }
648 kfree(async_extent->pages);
649 async_extent->nr_pages = 0;
650 async_extent->pages = NULL;
Chris Mason771ed682008-11-06 22:02:51 -0500651}
652
653/*
654 * phase two of compressed writeback. This is the ordered portion
655 * of the code, which only gets called in the order the work was
656 * queued. We walk all the async extents created by compress_file_range
657 * and send them down to the disk.
658 */
Filipe Mananadec8f172014-10-06 22:14:26 +0100659static noinline void submit_compressed_extents(struct inode *inode,
Chris Mason771ed682008-11-06 22:02:51 -0500660 struct async_cow *async_cow)
661{
662 struct async_extent *async_extent;
663 u64 alloc_hint = 0;
Chris Mason771ed682008-11-06 22:02:51 -0500664 struct btrfs_key ins;
665 struct extent_map *em;
666 struct btrfs_root *root = BTRFS_I(inode)->root;
667 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
668 struct extent_io_tree *io_tree;
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500669 int ret = 0;
Chris Mason771ed682008-11-06 22:02:51 -0500670
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500671again:
Chris Masond3977122009-01-05 21:25:51 -0500672 while (!list_empty(&async_cow->extents)) {
Chris Mason771ed682008-11-06 22:02:51 -0500673 async_extent = list_entry(async_cow->extents.next,
674 struct async_extent, list);
675 list_del(&async_extent->list);
676
677 io_tree = &BTRFS_I(inode)->io_tree;
678
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500679retry:
Chris Mason771ed682008-11-06 22:02:51 -0500680 /* did the compression code fall back to uncompressed IO? */
681 if (!async_extent->pages) {
682 int page_started = 0;
683 unsigned long nr_written = 0;
684
685 lock_extent(io_tree, async_extent->start,
Josef Bacik2ac55d42010-02-03 19:33:23 +0000686 async_extent->start +
Jeff Mahoneyd0082372012-03-01 14:57:19 +0100687 async_extent->ram_size - 1);
Chris Mason771ed682008-11-06 22:02:51 -0500688
689 /* allocate blocks */
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500690 ret = cow_file_range(inode, async_cow->locked_page,
691 async_extent->start,
692 async_extent->start +
693 async_extent->ram_size - 1,
694 &page_started, &nr_written, 0);
Chris Mason771ed682008-11-06 22:02:51 -0500695
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100696 /* JDM XXX */
697
Chris Mason771ed682008-11-06 22:02:51 -0500698 /*
699 * if page_started, cow_file_range inserted an
700 * inline extent and took care of all the unlocking
701 * and IO for us. Otherwise, we need to submit
702 * all those pages down to the drive.
703 */
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500704 if (!page_started && !ret)
Chris Mason771ed682008-11-06 22:02:51 -0500705 extent_write_locked_range(io_tree,
706 inode, async_extent->start,
Chris Masond3977122009-01-05 21:25:51 -0500707 async_extent->start +
Chris Mason771ed682008-11-06 22:02:51 -0500708 async_extent->ram_size - 1,
709 btrfs_get_extent,
710 WB_SYNC_ALL);
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500711 else if (ret)
712 unlock_page(async_cow->locked_page);
Chris Mason771ed682008-11-06 22:02:51 -0500713 kfree(async_extent);
714 cond_resched();
715 continue;
716 }
717
718 lock_extent(io_tree, async_extent->start,
Jeff Mahoneyd0082372012-03-01 14:57:19 +0100719 async_extent->start + async_extent->ram_size - 1);
Chris Mason771ed682008-11-06 22:02:51 -0500720
Josef Bacik00361582013-08-14 14:02:47 -0400721 ret = btrfs_reserve_extent(root,
Chris Mason771ed682008-11-06 22:02:51 -0500722 async_extent->compressed_size,
723 async_extent->compressed_size,
Miao Xiee570fd22014-06-19 10:42:50 +0800724 0, alloc_hint, &ins, 1, 1);
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500725 if (ret) {
Filipe Manana40ae8372014-10-06 22:14:24 +0100726 free_async_extent_pages(async_extent);
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500727
Josef Bacikfdf8e2e2013-06-14 16:58:23 -0400728 if (ret == -ENOSPC) {
729 unlock_extent(io_tree, async_extent->start,
730 async_extent->start +
731 async_extent->ram_size - 1);
Liu Boce620032014-07-24 22:48:05 +0800732
733 /*
734 * we need to redirty the pages if we decide to
735 * fallback to uncompressed IO, otherwise we
736 * will not submit these pages down to lower
737 * layers.
738 */
739 extent_range_redirty_for_io(inode,
740 async_extent->start,
741 async_extent->start +
742 async_extent->ram_size - 1);
743
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100744 goto retry;
Josef Bacikfdf8e2e2013-06-14 16:58:23 -0400745 }
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500746 goto out_free;
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500747 }
748
Yan, Zhengc2167752009-11-12 09:34:21 +0000749 /*
750 * here we're doing allocation and writeback of the
751 * compressed pages
752 */
753 btrfs_drop_extent_cache(inode, async_extent->start,
754 async_extent->start +
755 async_extent->ram_size - 1, 0);
756
David Sterba172ddd62011-04-21 00:48:27 +0200757 em = alloc_extent_map();
Liu Bob9aa55b2013-05-14 02:12:15 +0000758 if (!em) {
759 ret = -ENOMEM;
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500760 goto out_free_reserve;
Liu Bob9aa55b2013-05-14 02:12:15 +0000761 }
Chris Mason771ed682008-11-06 22:02:51 -0500762 em->start = async_extent->start;
763 em->len = async_extent->ram_size;
Chris Mason445a6942008-11-10 11:53:33 -0500764 em->orig_start = em->start;
Josef Bacik2ab28f32012-10-12 15:27:49 -0400765 em->mod_start = em->start;
766 em->mod_len = em->len;
Chris Mason771ed682008-11-06 22:02:51 -0500767
768 em->block_start = ins.objectid;
769 em->block_len = ins.offset;
Josef Bacikb4939682012-12-03 10:31:19 -0500770 em->orig_block_len = ins.offset;
Josef Bacikcc95bef2013-04-04 14:31:27 -0400771 em->ram_bytes = async_extent->ram_size;
Chris Mason771ed682008-11-06 22:02:51 -0500772 em->bdev = root->fs_info->fs_devices->latest_bdev;
Li Zefan261507a02010-12-17 14:21:50 +0800773 em->compress_type = async_extent->compress_type;
Chris Mason771ed682008-11-06 22:02:51 -0500774 set_bit(EXTENT_FLAG_PINNED, &em->flags);
775 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
Josef Bacik70c8a912012-10-11 16:54:30 -0400776 em->generation = -1;
Chris Mason771ed682008-11-06 22:02:51 -0500777
Chris Masond3977122009-01-05 21:25:51 -0500778 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -0400779 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -0400780 ret = add_extent_mapping(em_tree, em, 1);
Chris Mason890871b2009-09-02 16:24:52 -0400781 write_unlock(&em_tree->lock);
Chris Mason771ed682008-11-06 22:02:51 -0500782 if (ret != -EEXIST) {
783 free_extent_map(em);
784 break;
785 }
786 btrfs_drop_extent_cache(inode, async_extent->start,
787 async_extent->start +
788 async_extent->ram_size - 1, 0);
789 }
790
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500791 if (ret)
792 goto out_free_reserve;
793
Li Zefan261507a02010-12-17 14:21:50 +0800794 ret = btrfs_add_ordered_extent_compress(inode,
795 async_extent->start,
796 ins.objectid,
797 async_extent->ram_size,
798 ins.offset,
799 BTRFS_ORDERED_COMPRESSED,
800 async_extent->compress_type);
Filipe Mananad9f85962014-08-25 10:43:00 +0100801 if (ret) {
802 btrfs_drop_extent_cache(inode, async_extent->start,
803 async_extent->start +
804 async_extent->ram_size - 1, 0);
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500805 goto out_free_reserve;
Filipe Mananad9f85962014-08-25 10:43:00 +0100806 }
Chris Mason771ed682008-11-06 22:02:51 -0500807
Chris Mason771ed682008-11-06 22:02:51 -0500808 /*
809 * clear dirty, set writeback and unlock the pages.
810 */
Josef Bacikc2790a22013-07-29 11:20:47 -0400811 extent_clear_unlock_delalloc(inode, async_extent->start,
Chris Masona791e352009-10-08 11:27:10 -0400812 async_extent->start +
813 async_extent->ram_size - 1,
Josef Bacik151a41b2013-07-29 13:22:24 -0400814 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
815 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
Josef Bacikc2790a22013-07-29 11:20:47 -0400816 PAGE_SET_WRITEBACK);
Chris Mason771ed682008-11-06 22:02:51 -0500817 ret = btrfs_submit_compressed_write(inode,
Chris Masond3977122009-01-05 21:25:51 -0500818 async_extent->start,
819 async_extent->ram_size,
820 ins.objectid,
821 ins.offset, async_extent->pages,
822 async_extent->nr_pages);
Filipe Mananafce2a4e2014-10-06 22:14:23 +0100823 if (ret) {
824 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
825 struct page *p = async_extent->pages[0];
826 const u64 start = async_extent->start;
827 const u64 end = start + async_extent->ram_size - 1;
828
829 p->mapping = inode->i_mapping;
830 tree->ops->writepage_end_io_hook(p, start, end,
831 NULL, 0);
832 p->mapping = NULL;
833 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
834 PAGE_END_WRITEBACK |
835 PAGE_SET_ERROR);
Filipe Manana40ae8372014-10-06 22:14:24 +0100836 free_async_extent_pages(async_extent);
Filipe Mananafce2a4e2014-10-06 22:14:23 +0100837 }
Chris Mason771ed682008-11-06 22:02:51 -0500838 alloc_hint = ins.objectid + ins.offset;
839 kfree(async_extent);
840 cond_resched();
841 }
Filipe Mananadec8f172014-10-06 22:14:26 +0100842 return;
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500843out_free_reserve:
Miao Xiee570fd22014-06-19 10:42:50 +0800844 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100845out_free:
Josef Bacikc2790a22013-07-29 11:20:47 -0400846 extent_clear_unlock_delalloc(inode, async_extent->start,
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500847 async_extent->start +
848 async_extent->ram_size - 1,
Josef Bacikc2790a22013-07-29 11:20:47 -0400849 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
Josef Bacik151a41b2013-07-29 13:22:24 -0400850 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
851 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
Filipe Manana704de492014-10-06 22:14:22 +0100852 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
853 PAGE_SET_ERROR);
Filipe Manana40ae8372014-10-06 22:14:24 +0100854 free_async_extent_pages(async_extent);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100855 kfree(async_extent);
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500856 goto again;
Chris Mason771ed682008-11-06 22:02:51 -0500857}
858
Josef Bacik4b46fce2010-05-23 11:00:55 -0400859static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
860 u64 num_bytes)
861{
862 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
863 struct extent_map *em;
864 u64 alloc_hint = 0;
865
866 read_lock(&em_tree->lock);
867 em = search_extent_mapping(em_tree, start, num_bytes);
868 if (em) {
869 /*
870 * if block start isn't an actual block number then find the
871 * first block in this inode and use that as a hint. If that
872 * block is also bogus then just don't worry about it.
873 */
874 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
875 free_extent_map(em);
876 em = search_extent_mapping(em_tree, 0, 0);
877 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
878 alloc_hint = em->block_start;
879 if (em)
880 free_extent_map(em);
881 } else {
882 alloc_hint = em->block_start;
883 free_extent_map(em);
884 }
885 }
886 read_unlock(&em_tree->lock);
887
888 return alloc_hint;
889}
890
Chris Mason771ed682008-11-06 22:02:51 -0500891/*
892 * when extent_io.c finds a delayed allocation range in the file,
893 * the call backs end up in this code. The basic idea is to
894 * allocate extents on disk for the range, and create ordered data structs
895 * in ram to track those extents.
896 *
897 * locked_page is the page that writepage had locked already. We use
898 * it to make sure we don't do extra locks or unlocks.
899 *
900 * *page_started is set to one if we unlock locked_page and do everything
901 * required to start IO on it. It may be clean and already done with
902 * IO when we return.
903 */
Josef Bacik00361582013-08-14 14:02:47 -0400904static noinline int cow_file_range(struct inode *inode,
905 struct page *locked_page,
906 u64 start, u64 end, int *page_started,
907 unsigned long *nr_written,
908 int unlock)
Chris Mason771ed682008-11-06 22:02:51 -0500909{
Josef Bacik00361582013-08-14 14:02:47 -0400910 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason771ed682008-11-06 22:02:51 -0500911 u64 alloc_hint = 0;
912 u64 num_bytes;
913 unsigned long ram_size;
914 u64 disk_num_bytes;
915 u64 cur_alloc_size;
916 u64 blocksize = root->sectorsize;
Chris Mason771ed682008-11-06 22:02:51 -0500917 struct btrfs_key ins;
918 struct extent_map *em;
919 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
920 int ret = 0;
921
Josef Bacik02ecd2c2013-10-25 16:19:08 -0400922 if (btrfs_is_free_space_inode(inode)) {
923 WARN_ON_ONCE(1);
Josef Bacik29bce2f2014-02-07 12:21:23 -0500924 ret = -EINVAL;
925 goto out_unlock;
Josef Bacik02ecd2c2013-10-25 16:19:08 -0400926 }
Chris Mason771ed682008-11-06 22:02:51 -0500927
Qu Wenruofda28322013-02-26 08:10:22 +0000928 num_bytes = ALIGN(end - start + 1, blocksize);
Chris Mason771ed682008-11-06 22:02:51 -0500929 num_bytes = max(blocksize, num_bytes);
930 disk_num_bytes = num_bytes;
Chris Mason771ed682008-11-06 22:02:51 -0500931
Chris Mason4cb53002011-05-24 15:35:30 -0400932 /* if this is a small write inside eof, kick off defrag */
Liu Bo4cb13e52012-03-29 09:57:45 -0400933 if (num_bytes < 64 * 1024 &&
934 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
Josef Bacik00361582013-08-14 14:02:47 -0400935 btrfs_add_inode_defrag(NULL, inode);
Chris Mason4cb53002011-05-24 15:35:30 -0400936
Chris Mason771ed682008-11-06 22:02:51 -0500937 if (start == 0) {
938 /* lets try to make an inline extent */
Josef Bacik00361582013-08-14 14:02:47 -0400939 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
940 NULL);
Chris Mason771ed682008-11-06 22:02:51 -0500941 if (ret == 0) {
Josef Bacikc2790a22013-07-29 11:20:47 -0400942 extent_clear_unlock_delalloc(inode, start, end, NULL,
943 EXTENT_LOCKED | EXTENT_DELALLOC |
Josef Bacik151a41b2013-07-29 13:22:24 -0400944 EXTENT_DEFRAG, PAGE_UNLOCK |
Josef Bacikc2790a22013-07-29 11:20:47 -0400945 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
946 PAGE_END_WRITEBACK);
Yan, Zhengc2167752009-11-12 09:34:21 +0000947
Chris Mason771ed682008-11-06 22:02:51 -0500948 *nr_written = *nr_written +
949 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
950 *page_started = 1;
Chris Mason771ed682008-11-06 22:02:51 -0500951 goto out;
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100952 } else if (ret < 0) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100953 goto out_unlock;
Chris Mason771ed682008-11-06 22:02:51 -0500954 }
955 }
Chris Masonc8b97812008-10-29 14:49:59 -0400956
957 BUG_ON(disk_num_bytes >
David Sterba6c417612011-04-13 15:41:04 +0200958 btrfs_super_total_bytes(root->fs_info->super_copy));
Chris Masonc8b97812008-10-29 14:49:59 -0400959
Josef Bacik4b46fce2010-05-23 11:00:55 -0400960 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400961 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
Chris Mason3b951512008-04-17 11:29:12 -0400962
Chris Masond3977122009-01-05 21:25:51 -0500963 while (disk_num_bytes > 0) {
Chris Masona791e352009-10-08 11:27:10 -0400964 unsigned long op;
965
Josef Bacik287a0ab2010-03-19 18:07:23 +0000966 cur_alloc_size = disk_num_bytes;
Josef Bacik00361582013-08-14 14:02:47 -0400967 ret = btrfs_reserve_extent(root, cur_alloc_size,
Chris Mason771ed682008-11-06 22:02:51 -0500968 root->sectorsize, 0, alloc_hint,
Miao Xiee570fd22014-06-19 10:42:50 +0800969 &ins, 1, 1);
Josef Bacik00361582013-08-14 14:02:47 -0400970 if (ret < 0)
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100971 goto out_unlock;
Chris Masond3977122009-01-05 21:25:51 -0500972
David Sterba172ddd62011-04-21 00:48:27 +0200973 em = alloc_extent_map();
Liu Bob9aa55b2013-05-14 02:12:15 +0000974 if (!em) {
975 ret = -ENOMEM;
Liu Boace68ba2013-04-22 10:53:47 +0000976 goto out_reserve;
Liu Bob9aa55b2013-05-14 02:12:15 +0000977 }
Chris Masone6dcd2d2008-07-17 12:53:50 -0400978 em->start = start;
Chris Mason445a6942008-11-10 11:53:33 -0500979 em->orig_start = em->start;
Chris Mason771ed682008-11-06 22:02:51 -0500980 ram_size = ins.offset;
981 em->len = ins.offset;
Josef Bacik2ab28f32012-10-12 15:27:49 -0400982 em->mod_start = em->start;
983 em->mod_len = em->len;
Chris Masonc8b97812008-10-29 14:49:59 -0400984
Chris Masone6dcd2d2008-07-17 12:53:50 -0400985 em->block_start = ins.objectid;
Chris Masonc8b97812008-10-29 14:49:59 -0400986 em->block_len = ins.offset;
Josef Bacikb4939682012-12-03 10:31:19 -0500987 em->orig_block_len = ins.offset;
Josef Bacikcc95bef2013-04-04 14:31:27 -0400988 em->ram_bytes = ram_size;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400989 em->bdev = root->fs_info->fs_devices->latest_bdev;
Chris Mason7f3c74f2008-07-18 12:01:11 -0400990 set_bit(EXTENT_FLAG_PINNED, &em->flags);
Josef Bacik70c8a912012-10-11 16:54:30 -0400991 em->generation = -1;
Chris Masonc8b97812008-10-29 14:49:59 -0400992
Chris Masond3977122009-01-05 21:25:51 -0500993 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -0400994 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -0400995 ret = add_extent_mapping(em_tree, em, 1);
Chris Mason890871b2009-09-02 16:24:52 -0400996 write_unlock(&em_tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400997 if (ret != -EEXIST) {
998 free_extent_map(em);
999 break;
1000 }
1001 btrfs_drop_extent_cache(inode, start,
Chris Masonc8b97812008-10-29 14:49:59 -04001002 start + ram_size - 1, 0);
Chris Masone6dcd2d2008-07-17 12:53:50 -04001003 }
Liu Boace68ba2013-04-22 10:53:47 +00001004 if (ret)
1005 goto out_reserve;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001006
Chris Mason98d20f62008-04-14 09:46:10 -04001007 cur_alloc_size = ins.offset;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001008 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
Chris Mason771ed682008-11-06 22:02:51 -05001009 ram_size, cur_alloc_size, 0);
Liu Boace68ba2013-04-22 10:53:47 +00001010 if (ret)
Filipe Mananad9f85962014-08-25 10:43:00 +01001011 goto out_drop_extent_cache;
Chris Masonc8b97812008-10-29 14:49:59 -04001012
Yan Zheng17d217f2008-12-12 10:03:38 -05001013 if (root->root_key.objectid ==
1014 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1015 ret = btrfs_reloc_clone_csums(inode, start,
1016 cur_alloc_size);
Josef Bacik00361582013-08-14 14:02:47 -04001017 if (ret)
Filipe Mananad9f85962014-08-25 10:43:00 +01001018 goto out_drop_extent_cache;
Yan Zheng17d217f2008-12-12 10:03:38 -05001019 }
1020
Chris Masond3977122009-01-05 21:25:51 -05001021 if (disk_num_bytes < cur_alloc_size)
Chris Mason3b951512008-04-17 11:29:12 -04001022 break;
Chris Masond3977122009-01-05 21:25:51 -05001023
Chris Masonc8b97812008-10-29 14:49:59 -04001024 /* we're not doing compressed IO, don't unlock the first
1025 * page (which the caller expects to stay locked), don't
1026 * clear any dirty bits and don't set any writeback bits
Chris Mason8b62b722009-09-02 16:53:46 -04001027 *
1028 * Do set the Private2 bit so we know this page was properly
1029 * setup for writepage
Chris Masonc8b97812008-10-29 14:49:59 -04001030 */
Josef Bacikc2790a22013-07-29 11:20:47 -04001031 op = unlock ? PAGE_UNLOCK : 0;
1032 op |= PAGE_SET_PRIVATE2;
Chris Masona791e352009-10-08 11:27:10 -04001033
Josef Bacikc2790a22013-07-29 11:20:47 -04001034 extent_clear_unlock_delalloc(inode, start,
1035 start + ram_size - 1, locked_page,
1036 EXTENT_LOCKED | EXTENT_DELALLOC,
1037 op);
Chris Masonc8b97812008-10-29 14:49:59 -04001038 disk_num_bytes -= cur_alloc_size;
Chris Masonc59f8952007-12-17 20:14:04 -05001039 num_bytes -= cur_alloc_size;
1040 alloc_hint = ins.objectid + ins.offset;
1041 start += cur_alloc_size;
Chris Masonb888db22007-08-27 16:49:44 -04001042 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001043out:
Chris Masonbe20aa92007-12-17 20:14:01 -05001044 return ret;
Miao Xieb7d5b0a2012-11-01 07:32:18 +00001045
Filipe Mananad9f85962014-08-25 10:43:00 +01001046out_drop_extent_cache:
1047 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
Liu Boace68ba2013-04-22 10:53:47 +00001048out_reserve:
Miao Xiee570fd22014-06-19 10:42:50 +08001049 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001050out_unlock:
Josef Bacikc2790a22013-07-29 11:20:47 -04001051 extent_clear_unlock_delalloc(inode, start, end, locked_page,
Josef Bacik151a41b2013-07-29 13:22:24 -04001052 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1053 EXTENT_DELALLOC | EXTENT_DEFRAG,
1054 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1055 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001056 goto out;
Chris Mason771ed682008-11-06 22:02:51 -05001057}
Chris Masonc8b97812008-10-29 14:49:59 -04001058
Chris Mason771ed682008-11-06 22:02:51 -05001059/*
1060 * work queue call back to started compression on a file and pages
1061 */
1062static noinline void async_cow_start(struct btrfs_work *work)
1063{
1064 struct async_cow *async_cow;
1065 int num_added = 0;
1066 async_cow = container_of(work, struct async_cow, work);
1067
1068 compress_file_range(async_cow->inode, async_cow->locked_page,
1069 async_cow->start, async_cow->end, async_cow,
1070 &num_added);
Josef Bacik8180ef82012-06-08 15:16:12 -04001071 if (num_added == 0) {
Josef Bacikcb77fcd2012-06-15 12:19:48 -06001072 btrfs_add_delayed_iput(async_cow->inode);
Chris Mason771ed682008-11-06 22:02:51 -05001073 async_cow->inode = NULL;
Josef Bacik8180ef82012-06-08 15:16:12 -04001074 }
Chris Mason771ed682008-11-06 22:02:51 -05001075}
1076
1077/*
1078 * work queue call back to submit previously compressed pages
1079 */
1080static noinline void async_cow_submit(struct btrfs_work *work)
1081{
1082 struct async_cow *async_cow;
1083 struct btrfs_root *root;
1084 unsigned long nr_pages;
1085
1086 async_cow = container_of(work, struct async_cow, work);
1087
1088 root = async_cow->root;
1089 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1090 PAGE_CACHE_SHIFT;
1091
Josef Bacik66657b32012-08-01 15:36:24 -04001092 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
Liu Bo287082b2012-06-28 04:02:24 -06001093 5 * 1024 * 1024 &&
Chris Mason771ed682008-11-06 22:02:51 -05001094 waitqueue_active(&root->fs_info->async_submit_wait))
1095 wake_up(&root->fs_info->async_submit_wait);
1096
Chris Masond3977122009-01-05 21:25:51 -05001097 if (async_cow->inode)
Chris Mason771ed682008-11-06 22:02:51 -05001098 submit_compressed_extents(async_cow->inode, async_cow);
Chris Mason771ed682008-11-06 22:02:51 -05001099}
Chris Masonc8b97812008-10-29 14:49:59 -04001100
Chris Mason771ed682008-11-06 22:02:51 -05001101static noinline void async_cow_free(struct btrfs_work *work)
1102{
1103 struct async_cow *async_cow;
1104 async_cow = container_of(work, struct async_cow, work);
Josef Bacik8180ef82012-06-08 15:16:12 -04001105 if (async_cow->inode)
Josef Bacikcb77fcd2012-06-15 12:19:48 -06001106 btrfs_add_delayed_iput(async_cow->inode);
Chris Mason771ed682008-11-06 22:02:51 -05001107 kfree(async_cow);
1108}
1109
1110static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1111 u64 start, u64 end, int *page_started,
1112 unsigned long *nr_written)
1113{
1114 struct async_cow *async_cow;
1115 struct btrfs_root *root = BTRFS_I(inode)->root;
1116 unsigned long nr_pages;
1117 u64 cur_end;
Liu Bo287082b2012-06-28 04:02:24 -06001118 int limit = 10 * 1024 * 1024;
Chris Mason771ed682008-11-06 22:02:51 -05001119
Chris Masona3429ab2009-10-08 12:30:20 -04001120 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1121 1, 0, NULL, GFP_NOFS);
Chris Masond3977122009-01-05 21:25:51 -05001122 while (start < end) {
Chris Mason771ed682008-11-06 22:02:51 -05001123 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001124 BUG_ON(!async_cow); /* -ENOMEM */
Josef Bacik8180ef82012-06-08 15:16:12 -04001125 async_cow->inode = igrab(inode);
Chris Mason771ed682008-11-06 22:02:51 -05001126 async_cow->root = root;
1127 async_cow->locked_page = locked_page;
1128 async_cow->start = start;
1129
Wang Shilongf79707b2014-07-17 11:44:09 +08001130 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1131 !btrfs_test_opt(root, FORCE_COMPRESS))
Chris Mason771ed682008-11-06 22:02:51 -05001132 cur_end = end;
1133 else
1134 cur_end = min(end, start + 512 * 1024 - 1);
1135
1136 async_cow->end = cur_end;
1137 INIT_LIST_HEAD(&async_cow->extents);
1138
Liu Bo9e0af232014-08-15 23:36:53 +08001139 btrfs_init_work(&async_cow->work,
1140 btrfs_delalloc_helper,
1141 async_cow_start, async_cow_submit,
1142 async_cow_free);
Chris Mason771ed682008-11-06 22:02:51 -05001143
Chris Mason771ed682008-11-06 22:02:51 -05001144 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1145 PAGE_CACHE_SHIFT;
1146 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1147
Qu Wenruoafe3d242014-02-28 10:46:07 +08001148 btrfs_queue_work(root->fs_info->delalloc_workers,
1149 &async_cow->work);
Chris Mason771ed682008-11-06 22:02:51 -05001150
1151 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1152 wait_event(root->fs_info->async_submit_wait,
1153 (atomic_read(&root->fs_info->async_delalloc_pages) <
1154 limit));
1155 }
1156
Chris Masond3977122009-01-05 21:25:51 -05001157 while (atomic_read(&root->fs_info->async_submit_draining) &&
Chris Mason771ed682008-11-06 22:02:51 -05001158 atomic_read(&root->fs_info->async_delalloc_pages)) {
1159 wait_event(root->fs_info->async_submit_wait,
1160 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1161 0));
1162 }
1163
1164 *nr_written += nr_pages;
1165 start = cur_end + 1;
1166 }
1167 *page_started = 1;
1168 return 0;
Chris Masonbe20aa92007-12-17 20:14:01 -05001169}
1170
Chris Masond3977122009-01-05 21:25:51 -05001171static noinline int csum_exist_in_range(struct btrfs_root *root,
Yan Zheng17d217f2008-12-12 10:03:38 -05001172 u64 bytenr, u64 num_bytes)
1173{
1174 int ret;
1175 struct btrfs_ordered_sum *sums;
1176 LIST_HEAD(list);
1177
Yan Zheng07d400a2009-01-06 11:42:00 -05001178 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
Arne Jansena2de7332011-03-08 14:14:00 +01001179 bytenr + num_bytes - 1, &list, 0);
Yan Zheng17d217f2008-12-12 10:03:38 -05001180 if (ret == 0 && list_empty(&list))
1181 return 0;
1182
1183 while (!list_empty(&list)) {
1184 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1185 list_del(&sums->list);
1186 kfree(sums);
1187 }
1188 return 1;
1189}
1190
Chris Masond352ac62008-09-29 15:18:18 -04001191/*
1192 * when nowcow writeback call back. This checks for snapshots or COW copies
1193 * of the extents that exist in the file, and COWs the file as required.
1194 *
1195 * If no cow copies or snapshots exist, we write directly to the existing
1196 * blocks on disk
1197 */
Chris Mason7f366cf2009-03-12 20:12:45 -04001198static noinline int run_delalloc_nocow(struct inode *inode,
1199 struct page *locked_page,
Chris Mason771ed682008-11-06 22:02:51 -05001200 u64 start, u64 end, int *page_started, int force,
1201 unsigned long *nr_written)
Chris Masonbe20aa92007-12-17 20:14:01 -05001202{
Chris Masonbe20aa92007-12-17 20:14:01 -05001203 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001204 struct btrfs_trans_handle *trans;
Chris Masonbe20aa92007-12-17 20:14:01 -05001205 struct extent_buffer *leaf;
Chris Masonbe20aa92007-12-17 20:14:01 -05001206 struct btrfs_path *path;
Yan Zheng80ff3852008-10-30 14:20:02 -04001207 struct btrfs_file_extent_item *fi;
Chris Masonbe20aa92007-12-17 20:14:01 -05001208 struct btrfs_key found_key;
Yan Zheng80ff3852008-10-30 14:20:02 -04001209 u64 cow_start;
1210 u64 cur_offset;
1211 u64 extent_end;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001212 u64 extent_offset;
Yan Zheng80ff3852008-10-30 14:20:02 -04001213 u64 disk_bytenr;
1214 u64 num_bytes;
Josef Bacikb4939682012-12-03 10:31:19 -05001215 u64 disk_num_bytes;
Josef Bacikcc95bef2013-04-04 14:31:27 -04001216 u64 ram_bytes;
Yan Zheng80ff3852008-10-30 14:20:02 -04001217 int extent_type;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001218 int ret, err;
Yan Zhengd899e052008-10-30 14:25:28 -04001219 int type;
Yan Zheng80ff3852008-10-30 14:20:02 -04001220 int nocow;
1221 int check_prev = 1;
Li Zefan82d59022011-04-20 10:33:24 +08001222 bool nolock;
Li Zefan33345d012011-04-20 10:31:50 +08001223 u64 ino = btrfs_ino(inode);
Chris Masonbe20aa92007-12-17 20:14:01 -05001224
1225 path = btrfs_alloc_path();
Josef Bacik17ca04a2012-05-31 15:58:55 -04001226 if (!path) {
Josef Bacikc2790a22013-07-29 11:20:47 -04001227 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1228 EXTENT_LOCKED | EXTENT_DELALLOC |
Josef Bacik151a41b2013-07-29 13:22:24 -04001229 EXTENT_DO_ACCOUNTING |
1230 EXTENT_DEFRAG, PAGE_UNLOCK |
Josef Bacikc2790a22013-07-29 11:20:47 -04001231 PAGE_CLEAR_DIRTY |
1232 PAGE_SET_WRITEBACK |
1233 PAGE_END_WRITEBACK);
Mark Fashehd8926bb2011-07-13 10:38:47 -07001234 return -ENOMEM;
Josef Bacik17ca04a2012-05-31 15:58:55 -04001235 }
Li Zefan82d59022011-04-20 10:33:24 +08001236
Liu Bo83eea1f2012-07-10 05:28:39 -06001237 nolock = btrfs_is_free_space_inode(inode);
Li Zefan82d59022011-04-20 10:33:24 +08001238
1239 if (nolock)
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001240 trans = btrfs_join_transaction_nolock(root);
Li Zefan82d59022011-04-20 10:33:24 +08001241 else
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001242 trans = btrfs_join_transaction(root);
Chris Masonff5714c2011-05-28 07:00:39 -04001243
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001244 if (IS_ERR(trans)) {
Josef Bacikc2790a22013-07-29 11:20:47 -04001245 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1246 EXTENT_LOCKED | EXTENT_DELALLOC |
Josef Bacik151a41b2013-07-29 13:22:24 -04001247 EXTENT_DO_ACCOUNTING |
1248 EXTENT_DEFRAG, PAGE_UNLOCK |
Josef Bacikc2790a22013-07-29 11:20:47 -04001249 PAGE_CLEAR_DIRTY |
1250 PAGE_SET_WRITEBACK |
1251 PAGE_END_WRITEBACK);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001252 btrfs_free_path(path);
1253 return PTR_ERR(trans);
1254 }
1255
Josef Bacik74b21072011-04-13 12:02:53 -04001256 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
Chris Masonbe20aa92007-12-17 20:14:01 -05001257
Yan Zheng80ff3852008-10-30 14:20:02 -04001258 cow_start = (u64)-1;
1259 cur_offset = start;
1260 while (1) {
Li Zefan33345d012011-04-20 10:31:50 +08001261 ret = btrfs_lookup_file_extent(trans, root, path, ino,
Yan Zheng80ff3852008-10-30 14:20:02 -04001262 cur_offset, 0);
Josef Bacikd788a342013-10-25 16:55:08 -04001263 if (ret < 0)
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001264 goto error;
Yan Zheng80ff3852008-10-30 14:20:02 -04001265 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1266 leaf = path->nodes[0];
1267 btrfs_item_key_to_cpu(leaf, &found_key,
1268 path->slots[0] - 1);
Li Zefan33345d012011-04-20 10:31:50 +08001269 if (found_key.objectid == ino &&
Yan Zheng80ff3852008-10-30 14:20:02 -04001270 found_key.type == BTRFS_EXTENT_DATA_KEY)
1271 path->slots[0]--;
1272 }
1273 check_prev = 0;
1274next_slot:
1275 leaf = path->nodes[0];
1276 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1277 ret = btrfs_next_leaf(root, path);
Josef Bacikd788a342013-10-25 16:55:08 -04001278 if (ret < 0)
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001279 goto error;
Yan Zheng80ff3852008-10-30 14:20:02 -04001280 if (ret > 0)
1281 break;
1282 leaf = path->nodes[0];
1283 }
Chris Masonbe20aa92007-12-17 20:14:01 -05001284
Yan Zheng80ff3852008-10-30 14:20:02 -04001285 nocow = 0;
1286 disk_bytenr = 0;
Yan Zheng17d217f2008-12-12 10:03:38 -05001287 num_bytes = 0;
Yan Zheng80ff3852008-10-30 14:20:02 -04001288 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
Chris Masonbe20aa92007-12-17 20:14:01 -05001289
Li Zefan33345d012011-04-20 10:31:50 +08001290 if (found_key.objectid > ino ||
Yan Zheng80ff3852008-10-30 14:20:02 -04001291 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1292 found_key.offset > end)
1293 break;
Chris Masonbe20aa92007-12-17 20:14:01 -05001294
Yan Zheng80ff3852008-10-30 14:20:02 -04001295 if (found_key.offset > cur_offset) {
1296 extent_end = found_key.offset;
Chris Masone9061e22009-10-09 09:57:45 -04001297 extent_type = 0;
Yan Zheng80ff3852008-10-30 14:20:02 -04001298 goto out_check;
1299 }
Chris Masonc31f8832008-01-08 15:46:31 -05001300
Yan Zheng80ff3852008-10-30 14:20:02 -04001301 fi = btrfs_item_ptr(leaf, path->slots[0],
1302 struct btrfs_file_extent_item);
1303 extent_type = btrfs_file_extent_type(leaf, fi);
Chris Masonbe20aa92007-12-17 20:14:01 -05001304
Josef Bacikcc95bef2013-04-04 14:31:27 -04001305 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
Yan Zhengd899e052008-10-30 14:25:28 -04001306 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1307 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng80ff3852008-10-30 14:20:02 -04001308 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001309 extent_offset = btrfs_file_extent_offset(leaf, fi);
Yan Zheng80ff3852008-10-30 14:20:02 -04001310 extent_end = found_key.offset +
1311 btrfs_file_extent_num_bytes(leaf, fi);
Josef Bacikb4939682012-12-03 10:31:19 -05001312 disk_num_bytes =
1313 btrfs_file_extent_disk_num_bytes(leaf, fi);
Yan Zheng80ff3852008-10-30 14:20:02 -04001314 if (extent_end <= start) {
1315 path->slots[0]++;
1316 goto next_slot;
1317 }
Yan Zheng17d217f2008-12-12 10:03:38 -05001318 if (disk_bytenr == 0)
1319 goto out_check;
Yan Zheng80ff3852008-10-30 14:20:02 -04001320 if (btrfs_file_extent_compression(leaf, fi) ||
1321 btrfs_file_extent_encryption(leaf, fi) ||
1322 btrfs_file_extent_other_encoding(leaf, fi))
1323 goto out_check;
Yan Zhengd899e052008-10-30 14:25:28 -04001324 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1325 goto out_check;
Yan Zhengd2fb3432008-12-11 16:30:39 -05001326 if (btrfs_extent_readonly(root, disk_bytenr))
Yan Zheng80ff3852008-10-30 14:20:02 -04001327 goto out_check;
Li Zefan33345d012011-04-20 10:31:50 +08001328 if (btrfs_cross_ref_exist(trans, root, ino,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001329 found_key.offset -
1330 extent_offset, disk_bytenr))
Yan Zheng17d217f2008-12-12 10:03:38 -05001331 goto out_check;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001332 disk_bytenr += extent_offset;
Yan Zheng17d217f2008-12-12 10:03:38 -05001333 disk_bytenr += cur_offset - found_key.offset;
1334 num_bytes = min(end + 1, extent_end) - cur_offset;
1335 /*
Wang Shilonge9894fd2014-03-27 11:12:25 +08001336 * if there are pending snapshots for this root,
1337 * we fall into common COW way.
1338 */
1339 if (!nolock) {
Filipe Manana9ea24bb2014-10-29 11:57:59 +00001340 err = btrfs_start_write_no_snapshoting(root);
Wang Shilonge9894fd2014-03-27 11:12:25 +08001341 if (!err)
1342 goto out_check;
1343 }
1344 /*
Yan Zheng17d217f2008-12-12 10:03:38 -05001345 * force cow if csum exists in the range.
1346 * this ensure that csum for a given extent are
1347 * either valid or do not exist.
1348 */
1349 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1350 goto out_check;
Yan Zheng80ff3852008-10-30 14:20:02 -04001351 nocow = 1;
1352 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1353 extent_end = found_key.offset +
Chris Mason514ac8a2014-01-03 21:07:00 -08001354 btrfs_file_extent_inline_len(leaf,
1355 path->slots[0], fi);
Yan Zheng80ff3852008-10-30 14:20:02 -04001356 extent_end = ALIGN(extent_end, root->sectorsize);
1357 } else {
1358 BUG_ON(1);
1359 }
1360out_check:
1361 if (extent_end <= start) {
1362 path->slots[0]++;
Wang Shilonge9894fd2014-03-27 11:12:25 +08001363 if (!nolock && nocow)
Filipe Manana9ea24bb2014-10-29 11:57:59 +00001364 btrfs_end_write_no_snapshoting(root);
Yan Zheng80ff3852008-10-30 14:20:02 -04001365 goto next_slot;
1366 }
1367 if (!nocow) {
1368 if (cow_start == (u64)-1)
1369 cow_start = cur_offset;
1370 cur_offset = extent_end;
1371 if (cur_offset > end)
1372 break;
1373 path->slots[0]++;
1374 goto next_slot;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001375 }
Chris Masonbe20aa92007-12-17 20:14:01 -05001376
David Sterbab3b4aa72011-04-21 01:20:15 +02001377 btrfs_release_path(path);
Yan Zheng80ff3852008-10-30 14:20:02 -04001378 if (cow_start != (u64)-1) {
Josef Bacik00361582013-08-14 14:02:47 -04001379 ret = cow_file_range(inode, locked_page,
1380 cow_start, found_key.offset - 1,
1381 page_started, nr_written, 1);
Wang Shilonge9894fd2014-03-27 11:12:25 +08001382 if (ret) {
1383 if (!nolock && nocow)
Filipe Manana9ea24bb2014-10-29 11:57:59 +00001384 btrfs_end_write_no_snapshoting(root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001385 goto error;
Wang Shilonge9894fd2014-03-27 11:12:25 +08001386 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001387 cow_start = (u64)-1;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001388 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001389
Yan Zhengd899e052008-10-30 14:25:28 -04001390 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1391 struct extent_map *em;
1392 struct extent_map_tree *em_tree;
1393 em_tree = &BTRFS_I(inode)->extent_tree;
David Sterba172ddd62011-04-21 00:48:27 +02001394 em = alloc_extent_map();
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001395 BUG_ON(!em); /* -ENOMEM */
Yan Zhengd899e052008-10-30 14:25:28 -04001396 em->start = cur_offset;
Josef Bacik70c8a912012-10-11 16:54:30 -04001397 em->orig_start = found_key.offset - extent_offset;
Yan Zhengd899e052008-10-30 14:25:28 -04001398 em->len = num_bytes;
1399 em->block_len = num_bytes;
1400 em->block_start = disk_bytenr;
Josef Bacikb4939682012-12-03 10:31:19 -05001401 em->orig_block_len = disk_num_bytes;
Josef Bacikcc95bef2013-04-04 14:31:27 -04001402 em->ram_bytes = ram_bytes;
Yan Zhengd899e052008-10-30 14:25:28 -04001403 em->bdev = root->fs_info->fs_devices->latest_bdev;
Josef Bacik2ab28f32012-10-12 15:27:49 -04001404 em->mod_start = em->start;
1405 em->mod_len = em->len;
Yan Zhengd899e052008-10-30 14:25:28 -04001406 set_bit(EXTENT_FLAG_PINNED, &em->flags);
Josef Bacikb11e2342012-12-03 10:58:15 -05001407 set_bit(EXTENT_FLAG_FILLING, &em->flags);
Josef Bacik70c8a912012-10-11 16:54:30 -04001408 em->generation = -1;
Yan Zhengd899e052008-10-30 14:25:28 -04001409 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -04001410 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -04001411 ret = add_extent_mapping(em_tree, em, 1);
Chris Mason890871b2009-09-02 16:24:52 -04001412 write_unlock(&em_tree->lock);
Yan Zhengd899e052008-10-30 14:25:28 -04001413 if (ret != -EEXIST) {
1414 free_extent_map(em);
1415 break;
1416 }
1417 btrfs_drop_extent_cache(inode, em->start,
1418 em->start + em->len - 1, 0);
1419 }
1420 type = BTRFS_ORDERED_PREALLOC;
1421 } else {
1422 type = BTRFS_ORDERED_NOCOW;
1423 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001424
1425 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
Yan Zhengd899e052008-10-30 14:25:28 -04001426 num_bytes, num_bytes, type);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001427 BUG_ON(ret); /* -ENOMEM */
Chris Mason771ed682008-11-06 22:02:51 -05001428
Yan, Zhengefa56462010-05-16 10:49:59 -04001429 if (root->root_key.objectid ==
1430 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1431 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1432 num_bytes);
Wang Shilonge9894fd2014-03-27 11:12:25 +08001433 if (ret) {
1434 if (!nolock && nocow)
Filipe Manana9ea24bb2014-10-29 11:57:59 +00001435 btrfs_end_write_no_snapshoting(root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001436 goto error;
Wang Shilonge9894fd2014-03-27 11:12:25 +08001437 }
Yan, Zhengefa56462010-05-16 10:49:59 -04001438 }
1439
Josef Bacikc2790a22013-07-29 11:20:47 -04001440 extent_clear_unlock_delalloc(inode, cur_offset,
1441 cur_offset + num_bytes - 1,
1442 locked_page, EXTENT_LOCKED |
1443 EXTENT_DELALLOC, PAGE_UNLOCK |
1444 PAGE_SET_PRIVATE2);
Wang Shilonge9894fd2014-03-27 11:12:25 +08001445 if (!nolock && nocow)
Filipe Manana9ea24bb2014-10-29 11:57:59 +00001446 btrfs_end_write_no_snapshoting(root);
Yan Zheng80ff3852008-10-30 14:20:02 -04001447 cur_offset = extent_end;
1448 if (cur_offset > end)
1449 break;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001450 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001451 btrfs_release_path(path);
Yan Zheng80ff3852008-10-30 14:20:02 -04001452
Josef Bacik17ca04a2012-05-31 15:58:55 -04001453 if (cur_offset <= end && cow_start == (u64)-1) {
Yan Zheng80ff3852008-10-30 14:20:02 -04001454 cow_start = cur_offset;
Josef Bacik17ca04a2012-05-31 15:58:55 -04001455 cur_offset = end;
1456 }
1457
Yan Zheng80ff3852008-10-30 14:20:02 -04001458 if (cow_start != (u64)-1) {
Josef Bacik00361582013-08-14 14:02:47 -04001459 ret = cow_file_range(inode, locked_page, cow_start, end,
1460 page_started, nr_written, 1);
Josef Bacikd788a342013-10-25 16:55:08 -04001461 if (ret)
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001462 goto error;
Yan Zheng80ff3852008-10-30 14:20:02 -04001463 }
1464
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001465error:
Miao Xiea698d0752012-09-20 01:51:59 -06001466 err = btrfs_end_transaction(trans, root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001467 if (!ret)
1468 ret = err;
1469
Josef Bacik17ca04a2012-05-31 15:58:55 -04001470 if (ret && cur_offset < end)
Josef Bacikc2790a22013-07-29 11:20:47 -04001471 extent_clear_unlock_delalloc(inode, cur_offset, end,
1472 locked_page, EXTENT_LOCKED |
Josef Bacik151a41b2013-07-29 13:22:24 -04001473 EXTENT_DELALLOC | EXTENT_DEFRAG |
1474 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1475 PAGE_CLEAR_DIRTY |
Josef Bacikc2790a22013-07-29 11:20:47 -04001476 PAGE_SET_WRITEBACK |
1477 PAGE_END_WRITEBACK);
Yan Zheng7ea394f2008-08-05 13:05:02 -04001478 btrfs_free_path(path);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001479 return ret;
Chris Masonbe20aa92007-12-17 20:14:01 -05001480}
1481
Wang Shilong47059d92014-07-03 18:22:07 +08001482static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1483{
1484
1485 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1486 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1487 return 0;
1488
1489 /*
1490 * @defrag_bytes is a hint value, no spinlock held here,
1491 * if is not zero, it means the file is defragging.
1492 * Force cow if given extent needs to be defragged.
1493 */
1494 if (BTRFS_I(inode)->defrag_bytes &&
1495 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1496 EXTENT_DEFRAG, 0, NULL))
1497 return 1;
1498
1499 return 0;
1500}
1501
Chris Masond352ac62008-09-29 15:18:18 -04001502/*
1503 * extent_io.c call back to do delayed allocation processing
1504 */
Chris Masonc8b97812008-10-29 14:49:59 -04001505static int run_delalloc_range(struct inode *inode, struct page *locked_page,
Chris Mason771ed682008-11-06 22:02:51 -05001506 u64 start, u64 end, int *page_started,
1507 unsigned long *nr_written)
Chris Masonbe20aa92007-12-17 20:14:01 -05001508{
Chris Masonbe20aa92007-12-17 20:14:01 -05001509 int ret;
Wang Shilong47059d92014-07-03 18:22:07 +08001510 int force_cow = need_force_cow(inode, start, end);
Chris Masona2135012008-06-25 16:01:30 -04001511
Wang Shilong47059d92014-07-03 18:22:07 +08001512 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
Chris Masonc8b97812008-10-29 14:49:59 -04001513 ret = run_delalloc_nocow(inode, locked_page, start, end,
Chris Masond3977122009-01-05 21:25:51 -05001514 page_started, 1, nr_written);
Wang Shilong47059d92014-07-03 18:22:07 +08001515 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
Yan Zhengd899e052008-10-30 14:25:28 -04001516 ret = run_delalloc_nocow(inode, locked_page, start, end,
Chris Masond3977122009-01-05 21:25:51 -05001517 page_started, 0, nr_written);
Wang Shilong78160302014-07-17 11:44:10 +08001518 } else if (!inode_need_compress(inode)) {
Chris Mason7f366cf2009-03-12 20:12:45 -04001519 ret = cow_file_range(inode, locked_page, start, end,
1520 page_started, nr_written, 1);
Josef Bacik7ddf5a42012-06-08 15:26:47 -04001521 } else {
1522 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1523 &BTRFS_I(inode)->runtime_flags);
Chris Mason771ed682008-11-06 22:02:51 -05001524 ret = cow_file_range_async(inode, locked_page, start, end,
Chris Masond3977122009-01-05 21:25:51 -05001525 page_started, nr_written);
Josef Bacik7ddf5a42012-06-08 15:26:47 -04001526 }
Chris Masonb888db22007-08-27 16:49:44 -04001527 return ret;
1528}
1529
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001530static void btrfs_split_extent_hook(struct inode *inode,
1531 struct extent_state *orig, u64 split)
Josef Bacik9ed74f22009-09-11 16:12:44 -04001532{
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001533 /* not delalloc, ignore it */
Josef Bacik9ed74f22009-09-11 16:12:44 -04001534 if (!(orig->state & EXTENT_DELALLOC))
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001535 return;
Josef Bacik9ed74f22009-09-11 16:12:44 -04001536
Josef Bacik9e0baf62011-07-15 15:16:44 +00001537 spin_lock(&BTRFS_I(inode)->lock);
1538 BTRFS_I(inode)->outstanding_extents++;
1539 spin_unlock(&BTRFS_I(inode)->lock);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001540}
1541
1542/*
1543 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1544 * extents so we can keep track of new extents that are just merged onto old
1545 * extents, such as when we are doing sequential writes, so we can properly
1546 * account for the metadata space we'll need.
1547 */
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001548static void btrfs_merge_extent_hook(struct inode *inode,
1549 struct extent_state *new,
1550 struct extent_state *other)
Josef Bacik9ed74f22009-09-11 16:12:44 -04001551{
Josef Bacik9ed74f22009-09-11 16:12:44 -04001552 /* not delalloc, ignore it */
1553 if (!(other->state & EXTENT_DELALLOC))
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001554 return;
Josef Bacik9ed74f22009-09-11 16:12:44 -04001555
Josef Bacik9e0baf62011-07-15 15:16:44 +00001556 spin_lock(&BTRFS_I(inode)->lock);
1557 BTRFS_I(inode)->outstanding_extents--;
1558 spin_unlock(&BTRFS_I(inode)->lock);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001559}
1560
Miao Xieeb73c1b2013-05-15 07:48:22 +00001561static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1562 struct inode *inode)
1563{
1564 spin_lock(&root->delalloc_lock);
1565 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1566 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1567 &root->delalloc_inodes);
1568 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1569 &BTRFS_I(inode)->runtime_flags);
1570 root->nr_delalloc_inodes++;
1571 if (root->nr_delalloc_inodes == 1) {
1572 spin_lock(&root->fs_info->delalloc_root_lock);
1573 BUG_ON(!list_empty(&root->delalloc_root));
1574 list_add_tail(&root->delalloc_root,
1575 &root->fs_info->delalloc_roots);
1576 spin_unlock(&root->fs_info->delalloc_root_lock);
1577 }
1578 }
1579 spin_unlock(&root->delalloc_lock);
1580}
1581
1582static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1583 struct inode *inode)
1584{
1585 spin_lock(&root->delalloc_lock);
1586 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1587 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1588 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1589 &BTRFS_I(inode)->runtime_flags);
1590 root->nr_delalloc_inodes--;
1591 if (!root->nr_delalloc_inodes) {
1592 spin_lock(&root->fs_info->delalloc_root_lock);
1593 BUG_ON(list_empty(&root->delalloc_root));
1594 list_del_init(&root->delalloc_root);
1595 spin_unlock(&root->fs_info->delalloc_root_lock);
1596 }
1597 }
1598 spin_unlock(&root->delalloc_lock);
1599}
1600
Chris Masond352ac62008-09-29 15:18:18 -04001601/*
1602 * extent_io.c set_bit_hook, used to track delayed allocation
1603 * bytes in this file, and to maintain the list of inodes that
1604 * have pending delalloc work to be done.
1605 */
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001606static void btrfs_set_bit_hook(struct inode *inode,
David Sterba41074882013-04-29 13:38:46 +00001607 struct extent_state *state, unsigned long *bits)
Chris Mason291d6732008-01-29 15:55:23 -05001608{
Josef Bacik9ed74f22009-09-11 16:12:44 -04001609
Wang Shilong47059d92014-07-03 18:22:07 +08001610 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1611 WARN_ON(1);
Chris Mason75eff682008-12-15 15:54:40 -05001612 /*
1613 * set_bit and clear bit hooks normally require _irqsave/restore
Sergei Trofimovich27160b62011-05-20 20:20:32 +00001614 * but in this case, we are only testing for the DELALLOC
Chris Mason75eff682008-12-15 15:54:40 -05001615 * bit, which is only set or cleared with irqs on
1616 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001617 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
Chris Mason291d6732008-01-29 15:55:23 -05001618 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001619 u64 len = state->end + 1 - state->start;
Liu Bo83eea1f2012-07-10 05:28:39 -06001620 bool do_list = !btrfs_is_free_space_inode(inode);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001621
Josef Bacik9e0baf62011-07-15 15:16:44 +00001622 if (*bits & EXTENT_FIRST_DELALLOC) {
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001623 *bits &= ~EXTENT_FIRST_DELALLOC;
Josef Bacik9e0baf62011-07-15 15:16:44 +00001624 } else {
1625 spin_lock(&BTRFS_I(inode)->lock);
1626 BTRFS_I(inode)->outstanding_extents++;
1627 spin_unlock(&BTRFS_I(inode)->lock);
1628 }
Josef Bacik287a0ab2010-03-19 18:07:23 +00001629
Miao Xie963d6782013-01-29 10:10:51 +00001630 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1631 root->fs_info->delalloc_batch);
Miao Xiedf0af1a2013-01-29 10:11:59 +00001632 spin_lock(&BTRFS_I(inode)->lock);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001633 BTRFS_I(inode)->delalloc_bytes += len;
Wang Shilong47059d92014-07-03 18:22:07 +08001634 if (*bits & EXTENT_DEFRAG)
1635 BTRFS_I(inode)->defrag_bytes += len;
Miao Xiedf0af1a2013-01-29 10:11:59 +00001636 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
Miao Xieeb73c1b2013-05-15 07:48:22 +00001637 &BTRFS_I(inode)->runtime_flags))
1638 btrfs_add_delalloc_inodes(root, inode);
Miao Xiedf0af1a2013-01-29 10:11:59 +00001639 spin_unlock(&BTRFS_I(inode)->lock);
Chris Mason291d6732008-01-29 15:55:23 -05001640 }
Chris Mason291d6732008-01-29 15:55:23 -05001641}
1642
Chris Masond352ac62008-09-29 15:18:18 -04001643/*
1644 * extent_io.c clear_bit_hook, see set_bit_hook for why
1645 */
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001646static void btrfs_clear_bit_hook(struct inode *inode,
David Sterba41074882013-04-29 13:38:46 +00001647 struct extent_state *state,
1648 unsigned long *bits)
Chris Mason291d6732008-01-29 15:55:23 -05001649{
Wang Shilong47059d92014-07-03 18:22:07 +08001650 u64 len = state->end + 1 - state->start;
1651
1652 spin_lock(&BTRFS_I(inode)->lock);
1653 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1654 BTRFS_I(inode)->defrag_bytes -= len;
1655 spin_unlock(&BTRFS_I(inode)->lock);
1656
Chris Mason75eff682008-12-15 15:54:40 -05001657 /*
1658 * set_bit and clear bit hooks normally require _irqsave/restore
Sergei Trofimovich27160b62011-05-20 20:20:32 +00001659 * but in this case, we are only testing for the DELALLOC
Chris Mason75eff682008-12-15 15:54:40 -05001660 * bit, which is only set or cleared with irqs on
1661 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001662 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
Chris Mason291d6732008-01-29 15:55:23 -05001663 struct btrfs_root *root = BTRFS_I(inode)->root;
Liu Bo83eea1f2012-07-10 05:28:39 -06001664 bool do_list = !btrfs_is_free_space_inode(inode);
Chris Masonbcbfce82008-04-22 13:26:47 -04001665
Josef Bacik9e0baf62011-07-15 15:16:44 +00001666 if (*bits & EXTENT_FIRST_DELALLOC) {
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001667 *bits &= ~EXTENT_FIRST_DELALLOC;
Josef Bacik9e0baf62011-07-15 15:16:44 +00001668 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1669 spin_lock(&BTRFS_I(inode)->lock);
1670 BTRFS_I(inode)->outstanding_extents--;
1671 spin_unlock(&BTRFS_I(inode)->lock);
1672 }
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001673
Josef Bacikb6d08f02013-09-27 14:57:43 -04001674 /*
1675 * We don't reserve metadata space for space cache inodes so we
1676 * don't need to call dellalloc_release_metadata if there is an
1677 * error.
1678 */
1679 if (*bits & EXTENT_DO_ACCOUNTING &&
1680 root != root->fs_info->tree_root)
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001681 btrfs_delalloc_release_metadata(inode, len);
1682
Josef Bacik0cb59c92010-07-02 12:14:14 -04001683 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
Josef Bacik7ee9e442013-06-21 16:37:03 -04001684 && do_list && !(state->state & EXTENT_NORESERVE))
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001685 btrfs_free_reserved_data_space(inode, len);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001686
Miao Xie963d6782013-01-29 10:10:51 +00001687 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1688 root->fs_info->delalloc_batch);
Miao Xiedf0af1a2013-01-29 10:11:59 +00001689 spin_lock(&BTRFS_I(inode)->lock);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001690 BTRFS_I(inode)->delalloc_bytes -= len;
Josef Bacik0cb59c92010-07-02 12:14:14 -04001691 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
Miao Xiedf0af1a2013-01-29 10:11:59 +00001692 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
Miao Xieeb73c1b2013-05-15 07:48:22 +00001693 &BTRFS_I(inode)->runtime_flags))
1694 btrfs_del_delalloc_inode(root, inode);
Miao Xiedf0af1a2013-01-29 10:11:59 +00001695 spin_unlock(&BTRFS_I(inode)->lock);
Chris Mason291d6732008-01-29 15:55:23 -05001696 }
Chris Mason291d6732008-01-29 15:55:23 -05001697}
1698
Chris Masond352ac62008-09-29 15:18:18 -04001699/*
1700 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1701 * we don't create bios that span stripes or chunks
1702 */
David Woodhouse64a16702009-07-15 23:29:37 +01001703int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
Chris Masonc8b97812008-10-29 14:49:59 -04001704 size_t size, struct bio *bio,
1705 unsigned long bio_flags)
Chris Mason239b14b2008-03-24 15:02:07 -04001706{
1707 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001708 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
Chris Mason239b14b2008-03-24 15:02:07 -04001709 u64 length = 0;
1710 u64 map_length;
Chris Mason239b14b2008-03-24 15:02:07 -04001711 int ret;
1712
Chris Mason771ed682008-11-06 22:02:51 -05001713 if (bio_flags & EXTENT_BIO_COMPRESSED)
1714 return 0;
1715
Kent Overstreet4f024f32013-10-11 15:44:27 -07001716 length = bio->bi_iter.bi_size;
Chris Mason239b14b2008-03-24 15:02:07 -04001717 map_length = length;
David Woodhouse64a16702009-07-15 23:29:37 +01001718 ret = btrfs_map_block(root->fs_info, rw, logical,
Chris Masonf1885912008-04-09 16:28:12 -04001719 &map_length, NULL, 0);
Stefan Behrens3ec706c2012-11-05 15:46:42 +01001720 /* Will always return 0 with map_multi == NULL */
Jeff Mahoney3444a972011-10-03 23:23:13 -04001721 BUG_ON(ret < 0);
Chris Masond3977122009-01-05 21:25:51 -05001722 if (map_length < length + size)
Chris Mason239b14b2008-03-24 15:02:07 -04001723 return 1;
Jeff Mahoney3444a972011-10-03 23:23:13 -04001724 return 0;
Chris Mason239b14b2008-03-24 15:02:07 -04001725}
1726
Chris Masond352ac62008-09-29 15:18:18 -04001727/*
1728 * in order to insert checksums into the metadata in large chunks,
1729 * we wait until bio submission time. All the pages in the bio are
1730 * checksummed and sums are attached onto the ordered extent record.
1731 *
1732 * At IO completion time the cums attached on the ordered extent record
1733 * are inserted into the btree
1734 */
Chris Masond3977122009-01-05 21:25:51 -05001735static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1736 struct bio *bio, int mirror_num,
Chris Masoneaf25d92010-05-25 09:48:28 -04001737 unsigned long bio_flags,
1738 u64 bio_offset)
Chris Mason065631f2008-02-20 12:07:25 -05001739{
Chris Mason065631f2008-02-20 12:07:25 -05001740 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason065631f2008-02-20 12:07:25 -05001741 int ret = 0;
Chris Masone0156402008-04-16 11:15:20 -04001742
Chris Masond20f7042008-12-08 16:58:54 -05001743 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001744 BUG_ON(ret); /* -ENOMEM */
Chris Mason4a69a412008-11-06 22:03:00 -05001745 return 0;
1746}
Chris Masone0156402008-04-16 11:15:20 -04001747
Chris Mason4a69a412008-11-06 22:03:00 -05001748/*
1749 * in order to insert checksums into the metadata in large chunks,
1750 * we wait until bio submission time. All the pages in the bio are
1751 * checksummed and sums are attached onto the ordered extent record.
1752 *
1753 * At IO completion time the cums attached on the ordered extent record
1754 * are inserted into the btree
1755 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001756static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
Chris Masoneaf25d92010-05-25 09:48:28 -04001757 int mirror_num, unsigned long bio_flags,
1758 u64 bio_offset)
Chris Mason4a69a412008-11-06 22:03:00 -05001759{
1760 struct btrfs_root *root = BTRFS_I(inode)->root;
Stefan Behrens61891922012-11-05 18:51:52 +01001761 int ret;
1762
1763 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1764 if (ret)
1765 bio_endio(bio, ret);
1766 return ret;
Chris Mason44b8bd72008-04-16 11:14:51 -04001767}
1768
Chris Masond352ac62008-09-29 15:18:18 -04001769/*
Chris Masoncad321a2008-12-17 14:51:42 -05001770 * extent_io.c submission hook. This does the right thing for csum calculation
1771 * on write, or reading the csums from the tree before a read
Chris Masond352ac62008-09-29 15:18:18 -04001772 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001773static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
Chris Masoneaf25d92010-05-25 09:48:28 -04001774 int mirror_num, unsigned long bio_flags,
1775 u64 bio_offset)
Chris Mason44b8bd72008-04-16 11:14:51 -04001776{
1777 struct btrfs_root *root = BTRFS_I(inode)->root;
1778 int ret = 0;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001779 int skip_sum;
Jeff Mahoney04173412011-10-03 23:23:12 -04001780 int metadata = 0;
Josef Bacikb812ce22012-11-16 13:56:32 -05001781 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
Chris Mason44b8bd72008-04-16 11:14:51 -04001782
Christoph Hellwig6cbff002009-04-17 10:37:41 +02001783 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
Chris Masoncad321a2008-12-17 14:51:42 -05001784
Liu Bo83eea1f2012-07-10 05:28:39 -06001785 if (btrfs_is_free_space_inode(inode))
Jeff Mahoney04173412011-10-03 23:23:12 -04001786 metadata = 2;
1787
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001788 if (!(rw & REQ_WRITE)) {
Josef Bacik5fd02042012-05-02 14:00:54 -04001789 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1790 if (ret)
Stefan Behrens61891922012-11-05 18:51:52 +01001791 goto out;
Josef Bacik5fd02042012-05-02 14:00:54 -04001792
Chris Masond20f7042008-12-08 16:58:54 -05001793 if (bio_flags & EXTENT_BIO_COMPRESSED) {
Stefan Behrens61891922012-11-05 18:51:52 +01001794 ret = btrfs_submit_compressed_read(inode, bio,
1795 mirror_num,
1796 bio_flags);
1797 goto out;
Tsutomu Itohc2db1072011-03-01 06:48:31 +00001798 } else if (!skip_sum) {
1799 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1800 if (ret)
Stefan Behrens61891922012-11-05 18:51:52 +01001801 goto out;
Tsutomu Itohc2db1072011-03-01 06:48:31 +00001802 }
Chris Mason4d1b5fb2008-08-20 09:44:52 -04001803 goto mapit;
Josef Bacikb812ce22012-11-16 13:56:32 -05001804 } else if (async && !skip_sum) {
Yan Zheng17d217f2008-12-12 10:03:38 -05001805 /* csum items have already been cloned */
1806 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1807 goto mapit;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001808 /* we're doing a write, do the async checksumming */
Stefan Behrens61891922012-11-05 18:51:52 +01001809 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
Chris Mason44b8bd72008-04-16 11:14:51 -04001810 inode, rw, bio, mirror_num,
Chris Masoneaf25d92010-05-25 09:48:28 -04001811 bio_flags, bio_offset,
1812 __btrfs_submit_bio_start,
Chris Mason4a69a412008-11-06 22:03:00 -05001813 __btrfs_submit_bio_done);
Stefan Behrens61891922012-11-05 18:51:52 +01001814 goto out;
Josef Bacikb812ce22012-11-16 13:56:32 -05001815 } else if (!skip_sum) {
1816 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1817 if (ret)
1818 goto out;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001819 }
1820
Chris Mason0b86a832008-03-24 15:01:56 -04001821mapit:
Stefan Behrens61891922012-11-05 18:51:52 +01001822 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1823
1824out:
1825 if (ret < 0)
1826 bio_endio(bio, ret);
1827 return ret;
Chris Mason065631f2008-02-20 12:07:25 -05001828}
Chris Mason6885f302008-02-20 16:11:05 -05001829
Chris Masond352ac62008-09-29 15:18:18 -04001830/*
1831 * given a list of ordered sums record them in the inode. This happens
1832 * at IO completion time based on sums calculated at bio submission time.
1833 */
Chris Masonba1da2f2008-07-17 12:54:15 -04001834static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
Chris Masone6dcd2d2008-07-17 12:53:50 -04001835 struct inode *inode, u64 file_offset,
1836 struct list_head *list)
1837{
Chris Masone6dcd2d2008-07-17 12:53:50 -04001838 struct btrfs_ordered_sum *sum;
1839
Qinghuang Fengc6e30872009-01-21 10:59:08 -05001840 list_for_each_entry(sum, list, list) {
Miao Xie39847c42013-03-28 08:08:20 +00001841 trans->adding_csums = 1;
Chris Masond20f7042008-12-08 16:58:54 -05001842 btrfs_csum_file_blocks(trans,
1843 BTRFS_I(inode)->root->fs_info->csum_root, sum);
Miao Xie39847c42013-03-28 08:08:20 +00001844 trans->adding_csums = 0;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001845 }
1846 return 0;
1847}
1848
Josef Bacik2ac55d42010-02-03 19:33:23 +00001849int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1850 struct extent_state **cached_state)
Chris Masonea8c2812008-08-04 23:17:27 -04001851{
Julia Lawall6c1500f2012-11-03 20:30:18 +00001852 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
Chris Masonea8c2812008-08-04 23:17:27 -04001853 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
Josef Bacik2ac55d42010-02-03 19:33:23 +00001854 cached_state, GFP_NOFS);
Chris Masonea8c2812008-08-04 23:17:27 -04001855}
1856
Chris Masond352ac62008-09-29 15:18:18 -04001857/* see btrfs_writepage_start_hook for details on why this is required */
Chris Mason247e7432008-07-17 12:53:51 -04001858struct btrfs_writepage_fixup {
1859 struct page *page;
1860 struct btrfs_work work;
1861};
1862
Christoph Hellwigb2950862008-12-02 09:54:17 -05001863static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
Chris Mason247e7432008-07-17 12:53:51 -04001864{
1865 struct btrfs_writepage_fixup *fixup;
1866 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00001867 struct extent_state *cached_state = NULL;
Chris Mason247e7432008-07-17 12:53:51 -04001868 struct page *page;
1869 struct inode *inode;
1870 u64 page_start;
1871 u64 page_end;
Jeff Mahoney87826df2012-02-15 16:23:57 +01001872 int ret;
Chris Mason247e7432008-07-17 12:53:51 -04001873
1874 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1875 page = fixup->page;
Chris Mason4a096752008-07-21 10:29:44 -04001876again:
Chris Mason247e7432008-07-17 12:53:51 -04001877 lock_page(page);
1878 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1879 ClearPageChecked(page);
1880 goto out_page;
1881 }
1882
1883 inode = page->mapping->host;
1884 page_start = page_offset(page);
1885 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1886
Josef Bacik2ac55d42010-02-03 19:33:23 +00001887 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01001888 &cached_state);
Chris Mason4a096752008-07-21 10:29:44 -04001889
1890 /* already ordered? We're done */
Chris Mason8b62b722009-09-02 16:53:46 -04001891 if (PagePrivate2(page))
Chris Mason247e7432008-07-17 12:53:51 -04001892 goto out;
Chris Mason4a096752008-07-21 10:29:44 -04001893
1894 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1895 if (ordered) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00001896 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1897 page_end, &cached_state, GFP_NOFS);
Chris Mason4a096752008-07-21 10:29:44 -04001898 unlock_page(page);
1899 btrfs_start_ordered_extent(inode, ordered, 1);
Jeff Mahoney87826df2012-02-15 16:23:57 +01001900 btrfs_put_ordered_extent(ordered);
Chris Mason4a096752008-07-21 10:29:44 -04001901 goto again;
1902 }
Chris Mason247e7432008-07-17 12:53:51 -04001903
Jeff Mahoney87826df2012-02-15 16:23:57 +01001904 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1905 if (ret) {
1906 mapping_set_error(page->mapping, ret);
1907 end_extent_writepage(page, ret, page_start, page_end);
1908 ClearPageChecked(page);
1909 goto out;
1910 }
1911
Josef Bacik2ac55d42010-02-03 19:33:23 +00001912 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
Chris Mason247e7432008-07-17 12:53:51 -04001913 ClearPageChecked(page);
Jeff Mahoney87826df2012-02-15 16:23:57 +01001914 set_page_dirty(page);
Chris Mason247e7432008-07-17 12:53:51 -04001915out:
Josef Bacik2ac55d42010-02-03 19:33:23 +00001916 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1917 &cached_state, GFP_NOFS);
Chris Mason247e7432008-07-17 12:53:51 -04001918out_page:
1919 unlock_page(page);
1920 page_cache_release(page);
Miao Xieb897abe2011-01-26 16:19:22 +08001921 kfree(fixup);
Chris Mason247e7432008-07-17 12:53:51 -04001922}
1923
1924/*
1925 * There are a few paths in the higher layers of the kernel that directly
1926 * set the page dirty bit without asking the filesystem if it is a
1927 * good idea. This causes problems because we want to make sure COW
1928 * properly happens and the data=ordered rules are followed.
1929 *
Chris Masonc8b97812008-10-29 14:49:59 -04001930 * In our case any range that doesn't have the ORDERED bit set
Chris Mason247e7432008-07-17 12:53:51 -04001931 * hasn't been properly setup for IO. We kick off an async process
1932 * to fix it up. The async helper will wait for ordered extents, set
1933 * the delalloc bit and make it safe to write the page.
1934 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001935static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
Chris Mason247e7432008-07-17 12:53:51 -04001936{
1937 struct inode *inode = page->mapping->host;
1938 struct btrfs_writepage_fixup *fixup;
1939 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason247e7432008-07-17 12:53:51 -04001940
Chris Mason8b62b722009-09-02 16:53:46 -04001941 /* this page is properly in the ordered list */
1942 if (TestClearPagePrivate2(page))
Chris Mason247e7432008-07-17 12:53:51 -04001943 return 0;
1944
1945 if (PageChecked(page))
1946 return -EAGAIN;
1947
1948 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1949 if (!fixup)
1950 return -EAGAIN;
Chris Masonf4219502008-07-22 11:18:09 -04001951
Chris Mason247e7432008-07-17 12:53:51 -04001952 SetPageChecked(page);
1953 page_cache_get(page);
Liu Bo9e0af232014-08-15 23:36:53 +08001954 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
1955 btrfs_writepage_fixup_worker, NULL, NULL);
Chris Mason247e7432008-07-17 12:53:51 -04001956 fixup->page = page;
Qu Wenruodc6e3202014-02-28 10:46:14 +08001957 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
Jeff Mahoney87826df2012-02-15 16:23:57 +01001958 return -EBUSY;
Chris Mason247e7432008-07-17 12:53:51 -04001959}
1960
Yan Zhengd899e052008-10-30 14:25:28 -04001961static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1962 struct inode *inode, u64 file_pos,
1963 u64 disk_bytenr, u64 disk_num_bytes,
1964 u64 num_bytes, u64 ram_bytes,
1965 u8 compression, u8 encryption,
1966 u16 other_encoding, int extent_type)
1967{
1968 struct btrfs_root *root = BTRFS_I(inode)->root;
1969 struct btrfs_file_extent_item *fi;
1970 struct btrfs_path *path;
1971 struct extent_buffer *leaf;
1972 struct btrfs_key ins;
Filipe David Borba Manana1acae572014-01-07 11:42:27 +00001973 int extent_inserted = 0;
Yan Zhengd899e052008-10-30 14:25:28 -04001974 int ret;
1975
1976 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07001977 if (!path)
1978 return -ENOMEM;
Yan Zhengd899e052008-10-30 14:25:28 -04001979
Chris Masona1ed8352009-09-11 12:27:37 -04001980 /*
1981 * we may be replacing one extent in the tree with another.
1982 * The new extent is pinned in the extent map, and we don't want
1983 * to drop it from the cache until it is completely in the btree.
1984 *
1985 * So, tell btrfs_drop_extents to leave this extent in the cache.
1986 * the caller is expected to unpin it and allow it to be merged
1987 * with the others.
1988 */
Filipe David Borba Manana1acae572014-01-07 11:42:27 +00001989 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
1990 file_pos + num_bytes, NULL, 0,
1991 1, sizeof(*fi), &extent_inserted);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001992 if (ret)
1993 goto out;
Yan Zhengd899e052008-10-30 14:25:28 -04001994
Filipe David Borba Manana1acae572014-01-07 11:42:27 +00001995 if (!extent_inserted) {
1996 ins.objectid = btrfs_ino(inode);
1997 ins.offset = file_pos;
1998 ins.type = BTRFS_EXTENT_DATA_KEY;
1999
2000 path->leave_spinning = 1;
2001 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2002 sizeof(*fi));
2003 if (ret)
2004 goto out;
2005 }
Yan Zhengd899e052008-10-30 14:25:28 -04002006 leaf = path->nodes[0];
2007 fi = btrfs_item_ptr(leaf, path->slots[0],
2008 struct btrfs_file_extent_item);
2009 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2010 btrfs_set_file_extent_type(leaf, fi, extent_type);
2011 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2012 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2013 btrfs_set_file_extent_offset(leaf, fi, 0);
2014 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2015 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2016 btrfs_set_file_extent_compression(leaf, fi, compression);
2017 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2018 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
Chris Masonb9473432009-03-13 11:00:37 -04002019
Yan Zhengd899e052008-10-30 14:25:28 -04002020 btrfs_mark_buffer_dirty(leaf);
Josef Bacikce195332012-09-25 15:26:16 -04002021 btrfs_release_path(path);
Yan Zhengd899e052008-10-30 14:25:28 -04002022
2023 inode_add_bytes(inode, num_bytes);
Yan Zhengd899e052008-10-30 14:25:28 -04002024
2025 ins.objectid = disk_bytenr;
2026 ins.offset = disk_num_bytes;
2027 ins.type = BTRFS_EXTENT_ITEM_KEY;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002028 ret = btrfs_alloc_reserved_file_extent(trans, root,
2029 root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +08002030 btrfs_ino(inode), file_pos, &ins);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002031out:
Yan Zhengd899e052008-10-30 14:25:28 -04002032 btrfs_free_path(path);
Chris Masonb9473432009-03-13 11:00:37 -04002033
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002034 return ret;
Yan Zhengd899e052008-10-30 14:25:28 -04002035}
2036
Liu Bo38c227d2013-01-29 03:18:40 +00002037/* snapshot-aware defrag */
2038struct sa_defrag_extent_backref {
2039 struct rb_node node;
2040 struct old_sa_defrag_extent *old;
2041 u64 root_id;
2042 u64 inum;
2043 u64 file_pos;
2044 u64 extent_offset;
2045 u64 num_bytes;
2046 u64 generation;
2047};
2048
2049struct old_sa_defrag_extent {
2050 struct list_head list;
2051 struct new_sa_defrag_extent *new;
2052
2053 u64 extent_offset;
2054 u64 bytenr;
2055 u64 offset;
2056 u64 len;
2057 int count;
2058};
2059
2060struct new_sa_defrag_extent {
2061 struct rb_root root;
2062 struct list_head head;
2063 struct btrfs_path *path;
2064 struct inode *inode;
2065 u64 file_pos;
2066 u64 len;
2067 u64 bytenr;
2068 u64 disk_len;
2069 u8 compress_type;
2070};
2071
2072static int backref_comp(struct sa_defrag_extent_backref *b1,
2073 struct sa_defrag_extent_backref *b2)
2074{
2075 if (b1->root_id < b2->root_id)
2076 return -1;
2077 else if (b1->root_id > b2->root_id)
2078 return 1;
2079
2080 if (b1->inum < b2->inum)
2081 return -1;
2082 else if (b1->inum > b2->inum)
2083 return 1;
2084
2085 if (b1->file_pos < b2->file_pos)
2086 return -1;
2087 else if (b1->file_pos > b2->file_pos)
2088 return 1;
2089
2090 /*
2091 * [------------------------------] ===> (a range of space)
2092 * |<--->| |<---->| =============> (fs/file tree A)
2093 * |<---------------------------->| ===> (fs/file tree B)
2094 *
2095 * A range of space can refer to two file extents in one tree while
2096 * refer to only one file extent in another tree.
2097 *
2098 * So we may process a disk offset more than one time(two extents in A)
2099 * and locate at the same extent(one extent in B), then insert two same
2100 * backrefs(both refer to the extent in B).
2101 */
2102 return 0;
2103}
2104
2105static void backref_insert(struct rb_root *root,
2106 struct sa_defrag_extent_backref *backref)
2107{
2108 struct rb_node **p = &root->rb_node;
2109 struct rb_node *parent = NULL;
2110 struct sa_defrag_extent_backref *entry;
2111 int ret;
2112
2113 while (*p) {
2114 parent = *p;
2115 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2116
2117 ret = backref_comp(backref, entry);
2118 if (ret < 0)
2119 p = &(*p)->rb_left;
2120 else
2121 p = &(*p)->rb_right;
2122 }
2123
2124 rb_link_node(&backref->node, parent, p);
2125 rb_insert_color(&backref->node, root);
2126}
2127
2128/*
2129 * Note the backref might has changed, and in this case we just return 0.
2130 */
2131static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2132 void *ctx)
2133{
2134 struct btrfs_file_extent_item *extent;
2135 struct btrfs_fs_info *fs_info;
2136 struct old_sa_defrag_extent *old = ctx;
2137 struct new_sa_defrag_extent *new = old->new;
2138 struct btrfs_path *path = new->path;
2139 struct btrfs_key key;
2140 struct btrfs_root *root;
2141 struct sa_defrag_extent_backref *backref;
2142 struct extent_buffer *leaf;
2143 struct inode *inode = new->inode;
2144 int slot;
2145 int ret;
2146 u64 extent_offset;
2147 u64 num_bytes;
2148
2149 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2150 inum == btrfs_ino(inode))
2151 return 0;
2152
2153 key.objectid = root_id;
2154 key.type = BTRFS_ROOT_ITEM_KEY;
2155 key.offset = (u64)-1;
2156
2157 fs_info = BTRFS_I(inode)->root->fs_info;
2158 root = btrfs_read_fs_root_no_name(fs_info, &key);
2159 if (IS_ERR(root)) {
2160 if (PTR_ERR(root) == -ENOENT)
2161 return 0;
2162 WARN_ON(1);
2163 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2164 inum, offset, root_id);
2165 return PTR_ERR(root);
2166 }
2167
2168 key.objectid = inum;
2169 key.type = BTRFS_EXTENT_DATA_KEY;
2170 if (offset > (u64)-1 << 32)
2171 key.offset = 0;
2172 else
2173 key.offset = offset;
2174
2175 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05302176 if (WARN_ON(ret < 0))
Liu Bo38c227d2013-01-29 03:18:40 +00002177 return ret;
Josef Bacik50f13192013-07-22 12:50:37 -04002178 ret = 0;
Liu Bo38c227d2013-01-29 03:18:40 +00002179
2180 while (1) {
2181 cond_resched();
2182
2183 leaf = path->nodes[0];
2184 slot = path->slots[0];
2185
2186 if (slot >= btrfs_header_nritems(leaf)) {
2187 ret = btrfs_next_leaf(root, path);
2188 if (ret < 0) {
2189 goto out;
2190 } else if (ret > 0) {
2191 ret = 0;
2192 goto out;
2193 }
2194 continue;
2195 }
2196
2197 path->slots[0]++;
2198
2199 btrfs_item_key_to_cpu(leaf, &key, slot);
2200
2201 if (key.objectid > inum)
2202 goto out;
2203
2204 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2205 continue;
2206
2207 extent = btrfs_item_ptr(leaf, slot,
2208 struct btrfs_file_extent_item);
2209
2210 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2211 continue;
2212
Liu Boe68afa42013-07-01 22:13:26 +08002213 /*
2214 * 'offset' refers to the exact key.offset,
2215 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2216 * (key.offset - extent_offset).
2217 */
2218 if (key.offset != offset)
Liu Bo38c227d2013-01-29 03:18:40 +00002219 continue;
2220
Liu Boe68afa42013-07-01 22:13:26 +08002221 extent_offset = btrfs_file_extent_offset(leaf, extent);
Liu Bo38c227d2013-01-29 03:18:40 +00002222 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
Liu Boe68afa42013-07-01 22:13:26 +08002223
Liu Bo38c227d2013-01-29 03:18:40 +00002224 if (extent_offset >= old->extent_offset + old->offset +
2225 old->len || extent_offset + num_bytes <=
2226 old->extent_offset + old->offset)
2227 continue;
Liu Bo38c227d2013-01-29 03:18:40 +00002228 break;
2229 }
2230
2231 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2232 if (!backref) {
2233 ret = -ENOENT;
2234 goto out;
2235 }
2236
2237 backref->root_id = root_id;
2238 backref->inum = inum;
Liu Boe68afa42013-07-01 22:13:26 +08002239 backref->file_pos = offset;
Liu Bo38c227d2013-01-29 03:18:40 +00002240 backref->num_bytes = num_bytes;
2241 backref->extent_offset = extent_offset;
2242 backref->generation = btrfs_file_extent_generation(leaf, extent);
2243 backref->old = old;
2244 backref_insert(&new->root, backref);
2245 old->count++;
2246out:
2247 btrfs_release_path(path);
2248 WARN_ON(ret);
2249 return ret;
2250}
2251
2252static noinline bool record_extent_backrefs(struct btrfs_path *path,
2253 struct new_sa_defrag_extent *new)
2254{
2255 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2256 struct old_sa_defrag_extent *old, *tmp;
2257 int ret;
2258
2259 new->path = path;
2260
2261 list_for_each_entry_safe(old, tmp, &new->head, list) {
Liu Boe68afa42013-07-01 22:13:26 +08002262 ret = iterate_inodes_from_logical(old->bytenr +
2263 old->extent_offset, fs_info,
Liu Bo38c227d2013-01-29 03:18:40 +00002264 path, record_one_backref,
2265 old);
Josef Bacik4724b102013-11-05 11:11:40 -05002266 if (ret < 0 && ret != -ENOENT)
2267 return false;
Liu Bo38c227d2013-01-29 03:18:40 +00002268
2269 /* no backref to be processed for this extent */
2270 if (!old->count) {
2271 list_del(&old->list);
2272 kfree(old);
2273 }
2274 }
2275
2276 if (list_empty(&new->head))
2277 return false;
2278
2279 return true;
2280}
2281
2282static int relink_is_mergable(struct extent_buffer *leaf,
2283 struct btrfs_file_extent_item *fi,
Liu Bo116e0022013-08-02 16:30:40 +08002284 struct new_sa_defrag_extent *new)
Liu Bo38c227d2013-01-29 03:18:40 +00002285{
Liu Bo116e0022013-08-02 16:30:40 +08002286 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
Liu Bo38c227d2013-01-29 03:18:40 +00002287 return 0;
2288
2289 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2290 return 0;
2291
Liu Bo116e0022013-08-02 16:30:40 +08002292 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2293 return 0;
2294
2295 if (btrfs_file_extent_encryption(leaf, fi) ||
Liu Bo38c227d2013-01-29 03:18:40 +00002296 btrfs_file_extent_other_encoding(leaf, fi))
2297 return 0;
2298
2299 return 1;
2300}
2301
2302/*
2303 * Note the backref might has changed, and in this case we just return 0.
2304 */
2305static noinline int relink_extent_backref(struct btrfs_path *path,
2306 struct sa_defrag_extent_backref *prev,
2307 struct sa_defrag_extent_backref *backref)
2308{
2309 struct btrfs_file_extent_item *extent;
2310 struct btrfs_file_extent_item *item;
2311 struct btrfs_ordered_extent *ordered;
2312 struct btrfs_trans_handle *trans;
2313 struct btrfs_fs_info *fs_info;
2314 struct btrfs_root *root;
2315 struct btrfs_key key;
2316 struct extent_buffer *leaf;
2317 struct old_sa_defrag_extent *old = backref->old;
2318 struct new_sa_defrag_extent *new = old->new;
2319 struct inode *src_inode = new->inode;
2320 struct inode *inode;
2321 struct extent_state *cached = NULL;
2322 int ret = 0;
2323 u64 start;
2324 u64 len;
2325 u64 lock_start;
2326 u64 lock_end;
2327 bool merge = false;
2328 int index;
2329
2330 if (prev && prev->root_id == backref->root_id &&
2331 prev->inum == backref->inum &&
2332 prev->file_pos + prev->num_bytes == backref->file_pos)
2333 merge = true;
2334
2335 /* step 1: get root */
2336 key.objectid = backref->root_id;
2337 key.type = BTRFS_ROOT_ITEM_KEY;
2338 key.offset = (u64)-1;
2339
2340 fs_info = BTRFS_I(src_inode)->root->fs_info;
2341 index = srcu_read_lock(&fs_info->subvol_srcu);
2342
2343 root = btrfs_read_fs_root_no_name(fs_info, &key);
2344 if (IS_ERR(root)) {
2345 srcu_read_unlock(&fs_info->subvol_srcu, index);
2346 if (PTR_ERR(root) == -ENOENT)
2347 return 0;
2348 return PTR_ERR(root);
2349 }
Liu Bo38c227d2013-01-29 03:18:40 +00002350
Wang Shilongbcbba5e2014-02-08 23:46:35 +08002351 if (btrfs_root_readonly(root)) {
2352 srcu_read_unlock(&fs_info->subvol_srcu, index);
2353 return 0;
2354 }
2355
Liu Bo38c227d2013-01-29 03:18:40 +00002356 /* step 2: get inode */
2357 key.objectid = backref->inum;
2358 key.type = BTRFS_INODE_ITEM_KEY;
2359 key.offset = 0;
2360
2361 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2362 if (IS_ERR(inode)) {
2363 srcu_read_unlock(&fs_info->subvol_srcu, index);
2364 return 0;
2365 }
2366
2367 srcu_read_unlock(&fs_info->subvol_srcu, index);
2368
2369 /* step 3: relink backref */
2370 lock_start = backref->file_pos;
2371 lock_end = backref->file_pos + backref->num_bytes - 1;
2372 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2373 0, &cached);
2374
2375 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2376 if (ordered) {
2377 btrfs_put_ordered_extent(ordered);
2378 goto out_unlock;
2379 }
2380
2381 trans = btrfs_join_transaction(root);
2382 if (IS_ERR(trans)) {
2383 ret = PTR_ERR(trans);
2384 goto out_unlock;
2385 }
2386
2387 key.objectid = backref->inum;
2388 key.type = BTRFS_EXTENT_DATA_KEY;
2389 key.offset = backref->file_pos;
2390
2391 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2392 if (ret < 0) {
2393 goto out_free_path;
2394 } else if (ret > 0) {
2395 ret = 0;
2396 goto out_free_path;
2397 }
2398
2399 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2400 struct btrfs_file_extent_item);
2401
2402 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2403 backref->generation)
2404 goto out_free_path;
2405
2406 btrfs_release_path(path);
2407
2408 start = backref->file_pos;
2409 if (backref->extent_offset < old->extent_offset + old->offset)
2410 start += old->extent_offset + old->offset -
2411 backref->extent_offset;
2412
2413 len = min(backref->extent_offset + backref->num_bytes,
2414 old->extent_offset + old->offset + old->len);
2415 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2416
2417 ret = btrfs_drop_extents(trans, root, inode, start,
2418 start + len, 1);
2419 if (ret)
2420 goto out_free_path;
2421again:
2422 key.objectid = btrfs_ino(inode);
2423 key.type = BTRFS_EXTENT_DATA_KEY;
2424 key.offset = start;
2425
Liu Boa09a0a72013-03-11 09:20:58 +00002426 path->leave_spinning = 1;
Liu Bo38c227d2013-01-29 03:18:40 +00002427 if (merge) {
2428 struct btrfs_file_extent_item *fi;
2429 u64 extent_len;
2430 struct btrfs_key found_key;
2431
Gui Hecheng3c9665d2014-01-23 13:41:09 +08002432 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
Liu Bo38c227d2013-01-29 03:18:40 +00002433 if (ret < 0)
2434 goto out_free_path;
2435
2436 path->slots[0]--;
2437 leaf = path->nodes[0];
2438 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2439
2440 fi = btrfs_item_ptr(leaf, path->slots[0],
2441 struct btrfs_file_extent_item);
2442 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2443
Liu Bo116e0022013-08-02 16:30:40 +08002444 if (extent_len + found_key.offset == start &&
2445 relink_is_mergable(leaf, fi, new)) {
Liu Bo38c227d2013-01-29 03:18:40 +00002446 btrfs_set_file_extent_num_bytes(leaf, fi,
2447 extent_len + len);
2448 btrfs_mark_buffer_dirty(leaf);
2449 inode_add_bytes(inode, len);
2450
2451 ret = 1;
2452 goto out_free_path;
2453 } else {
2454 merge = false;
2455 btrfs_release_path(path);
2456 goto again;
2457 }
2458 }
2459
2460 ret = btrfs_insert_empty_item(trans, root, path, &key,
2461 sizeof(*extent));
2462 if (ret) {
2463 btrfs_abort_transaction(trans, root, ret);
2464 goto out_free_path;
2465 }
2466
2467 leaf = path->nodes[0];
2468 item = btrfs_item_ptr(leaf, path->slots[0],
2469 struct btrfs_file_extent_item);
2470 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2471 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2472 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2473 btrfs_set_file_extent_num_bytes(leaf, item, len);
2474 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2475 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2476 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2477 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2478 btrfs_set_file_extent_encryption(leaf, item, 0);
2479 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2480
2481 btrfs_mark_buffer_dirty(leaf);
2482 inode_add_bytes(inode, len);
Liu Boa09a0a72013-03-11 09:20:58 +00002483 btrfs_release_path(path);
Liu Bo38c227d2013-01-29 03:18:40 +00002484
2485 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2486 new->disk_len, 0,
2487 backref->root_id, backref->inum,
2488 new->file_pos, 0); /* start - extent_offset */
2489 if (ret) {
2490 btrfs_abort_transaction(trans, root, ret);
2491 goto out_free_path;
2492 }
2493
2494 ret = 1;
2495out_free_path:
2496 btrfs_release_path(path);
Liu Boa09a0a72013-03-11 09:20:58 +00002497 path->leave_spinning = 0;
Liu Bo38c227d2013-01-29 03:18:40 +00002498 btrfs_end_transaction(trans, root);
2499out_unlock:
2500 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2501 &cached, GFP_NOFS);
2502 iput(inode);
2503 return ret;
2504}
2505
Liu Bo6f519562013-10-29 10:45:05 +08002506static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2507{
2508 struct old_sa_defrag_extent *old, *tmp;
2509
2510 if (!new)
2511 return;
2512
2513 list_for_each_entry_safe(old, tmp, &new->head, list) {
2514 list_del(&old->list);
2515 kfree(old);
2516 }
2517 kfree(new);
2518}
2519
Liu Bo38c227d2013-01-29 03:18:40 +00002520static void relink_file_extents(struct new_sa_defrag_extent *new)
2521{
2522 struct btrfs_path *path;
Liu Bo38c227d2013-01-29 03:18:40 +00002523 struct sa_defrag_extent_backref *backref;
2524 struct sa_defrag_extent_backref *prev = NULL;
2525 struct inode *inode;
2526 struct btrfs_root *root;
2527 struct rb_node *node;
2528 int ret;
2529
2530 inode = new->inode;
2531 root = BTRFS_I(inode)->root;
2532
2533 path = btrfs_alloc_path();
2534 if (!path)
2535 return;
2536
2537 if (!record_extent_backrefs(path, new)) {
2538 btrfs_free_path(path);
2539 goto out;
2540 }
2541 btrfs_release_path(path);
2542
2543 while (1) {
2544 node = rb_first(&new->root);
2545 if (!node)
2546 break;
2547 rb_erase(node, &new->root);
2548
2549 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2550
2551 ret = relink_extent_backref(path, prev, backref);
2552 WARN_ON(ret < 0);
2553
2554 kfree(prev);
2555
2556 if (ret == 1)
2557 prev = backref;
2558 else
2559 prev = NULL;
2560 cond_resched();
2561 }
2562 kfree(prev);
2563
2564 btrfs_free_path(path);
Liu Bo38c227d2013-01-29 03:18:40 +00002565out:
Liu Bo6f519562013-10-29 10:45:05 +08002566 free_sa_defrag_extent(new);
2567
Liu Bo38c227d2013-01-29 03:18:40 +00002568 atomic_dec(&root->fs_info->defrag_running);
2569 wake_up(&root->fs_info->transaction_wait);
Liu Bo38c227d2013-01-29 03:18:40 +00002570}
2571
2572static struct new_sa_defrag_extent *
2573record_old_file_extents(struct inode *inode,
2574 struct btrfs_ordered_extent *ordered)
2575{
2576 struct btrfs_root *root = BTRFS_I(inode)->root;
2577 struct btrfs_path *path;
2578 struct btrfs_key key;
Liu Bo6f519562013-10-29 10:45:05 +08002579 struct old_sa_defrag_extent *old;
Liu Bo38c227d2013-01-29 03:18:40 +00002580 struct new_sa_defrag_extent *new;
2581 int ret;
2582
2583 new = kmalloc(sizeof(*new), GFP_NOFS);
2584 if (!new)
2585 return NULL;
2586
2587 new->inode = inode;
2588 new->file_pos = ordered->file_offset;
2589 new->len = ordered->len;
2590 new->bytenr = ordered->start;
2591 new->disk_len = ordered->disk_len;
2592 new->compress_type = ordered->compress_type;
2593 new->root = RB_ROOT;
2594 INIT_LIST_HEAD(&new->head);
2595
2596 path = btrfs_alloc_path();
2597 if (!path)
2598 goto out_kfree;
2599
2600 key.objectid = btrfs_ino(inode);
2601 key.type = BTRFS_EXTENT_DATA_KEY;
2602 key.offset = new->file_pos;
2603
2604 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2605 if (ret < 0)
2606 goto out_free_path;
2607 if (ret > 0 && path->slots[0] > 0)
2608 path->slots[0]--;
2609
2610 /* find out all the old extents for the file range */
2611 while (1) {
2612 struct btrfs_file_extent_item *extent;
2613 struct extent_buffer *l;
2614 int slot;
2615 u64 num_bytes;
2616 u64 offset;
2617 u64 end;
2618 u64 disk_bytenr;
2619 u64 extent_offset;
2620
2621 l = path->nodes[0];
2622 slot = path->slots[0];
2623
2624 if (slot >= btrfs_header_nritems(l)) {
2625 ret = btrfs_next_leaf(root, path);
2626 if (ret < 0)
Liu Bo6f519562013-10-29 10:45:05 +08002627 goto out_free_path;
Liu Bo38c227d2013-01-29 03:18:40 +00002628 else if (ret > 0)
2629 break;
2630 continue;
2631 }
2632
2633 btrfs_item_key_to_cpu(l, &key, slot);
2634
2635 if (key.objectid != btrfs_ino(inode))
2636 break;
2637 if (key.type != BTRFS_EXTENT_DATA_KEY)
2638 break;
2639 if (key.offset >= new->file_pos + new->len)
2640 break;
2641
2642 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2643
2644 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2645 if (key.offset + num_bytes < new->file_pos)
2646 goto next;
2647
2648 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2649 if (!disk_bytenr)
2650 goto next;
2651
2652 extent_offset = btrfs_file_extent_offset(l, extent);
2653
2654 old = kmalloc(sizeof(*old), GFP_NOFS);
2655 if (!old)
Liu Bo6f519562013-10-29 10:45:05 +08002656 goto out_free_path;
Liu Bo38c227d2013-01-29 03:18:40 +00002657
2658 offset = max(new->file_pos, key.offset);
2659 end = min(new->file_pos + new->len, key.offset + num_bytes);
2660
2661 old->bytenr = disk_bytenr;
2662 old->extent_offset = extent_offset;
2663 old->offset = offset - key.offset;
2664 old->len = end - offset;
2665 old->new = new;
2666 old->count = 0;
2667 list_add_tail(&old->list, &new->head);
2668next:
2669 path->slots[0]++;
2670 cond_resched();
2671 }
2672
2673 btrfs_free_path(path);
2674 atomic_inc(&root->fs_info->defrag_running);
2675
2676 return new;
2677
Liu Bo38c227d2013-01-29 03:18:40 +00002678out_free_path:
2679 btrfs_free_path(path);
2680out_kfree:
Liu Bo6f519562013-10-29 10:45:05 +08002681 free_sa_defrag_extent(new);
Liu Bo38c227d2013-01-29 03:18:40 +00002682 return NULL;
2683}
2684
Miao Xiee570fd22014-06-19 10:42:50 +08002685static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2686 u64 start, u64 len)
2687{
2688 struct btrfs_block_group_cache *cache;
2689
2690 cache = btrfs_lookup_block_group(root->fs_info, start);
2691 ASSERT(cache);
2692
2693 spin_lock(&cache->lock);
2694 cache->delalloc_bytes -= len;
2695 spin_unlock(&cache->lock);
2696
2697 btrfs_put_block_group(cache);
2698}
2699
Chris Masond352ac62008-09-29 15:18:18 -04002700/* as ordered data IO finishes, this gets called so we can finish
2701 * an ordered extent if the range of bytes in the file it covers are
2702 * fully written.
2703 */
Josef Bacik5fd02042012-05-02 14:00:54 -04002704static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
Chris Masone6dcd2d2008-07-17 12:53:50 -04002705{
Josef Bacik5fd02042012-05-02 14:00:54 -04002706 struct inode *inode = ordered_extent->inode;
Chris Masone6dcd2d2008-07-17 12:53:50 -04002707 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04002708 struct btrfs_trans_handle *trans = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04002709 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Josef Bacik2ac55d42010-02-03 19:33:23 +00002710 struct extent_state *cached_state = NULL;
Liu Bo38c227d2013-01-29 03:18:40 +00002711 struct new_sa_defrag_extent *new = NULL;
Li Zefan261507a02010-12-17 14:21:50 +08002712 int compress_type = 0;
Josef Bacik77cef2e2013-08-29 13:57:21 -04002713 int ret = 0;
2714 u64 logical_len = ordered_extent->len;
Li Zefan82d59022011-04-20 10:33:24 +08002715 bool nolock;
Josef Bacik77cef2e2013-08-29 13:57:21 -04002716 bool truncated = false;
Chris Masone6dcd2d2008-07-17 12:53:50 -04002717
Liu Bo83eea1f2012-07-10 05:28:39 -06002718 nolock = btrfs_is_free_space_inode(inode);
Josef Bacik0cb59c92010-07-02 12:14:14 -04002719
Josef Bacik5fd02042012-05-02 14:00:54 -04002720 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2721 ret = -EIO;
2722 goto out;
2723 }
2724
Miao Xief6124962014-09-12 18:44:04 +08002725 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2726 ordered_extent->file_offset +
2727 ordered_extent->len - 1);
2728
Josef Bacik77cef2e2013-08-29 13:57:21 -04002729 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2730 truncated = true;
2731 logical_len = ordered_extent->truncated_len;
2732 /* Truncated the entire extent, don't bother adding */
2733 if (!logical_len)
2734 goto out;
2735 }
2736
Yan, Zhengc2167752009-11-12 09:34:21 +00002737 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002738 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
Josef Bacik6c760c02012-11-09 10:53:21 -05002739 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2740 if (nolock)
2741 trans = btrfs_join_transaction_nolock(root);
2742 else
2743 trans = btrfs_join_transaction(root);
2744 if (IS_ERR(trans)) {
2745 ret = PTR_ERR(trans);
2746 trans = NULL;
2747 goto out;
Yan, Zhengc2167752009-11-12 09:34:21 +00002748 }
Josef Bacik6c760c02012-11-09 10:53:21 -05002749 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2750 ret = btrfs_update_inode_fallback(trans, root, inode);
2751 if (ret) /* -ENOMEM or corruption */
2752 btrfs_abort_transaction(trans, root, ret);
Yan, Zhengc2167752009-11-12 09:34:21 +00002753 goto out;
2754 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04002755
Josef Bacik2ac55d42010-02-03 19:33:23 +00002756 lock_extent_bits(io_tree, ordered_extent->file_offset,
2757 ordered_extent->file_offset + ordered_extent->len - 1,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01002758 0, &cached_state);
Chris Masone6dcd2d2008-07-17 12:53:50 -04002759
Liu Bo38c227d2013-01-29 03:18:40 +00002760 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2761 ordered_extent->file_offset + ordered_extent->len - 1,
2762 EXTENT_DEFRAG, 1, cached_state);
2763 if (ret) {
2764 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
Josef Bacik8101c8d2014-01-29 16:05:30 -05002765 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
Liu Bo38c227d2013-01-29 03:18:40 +00002766 /* the inode is shared */
2767 new = record_old_file_extents(inode, ordered_extent);
2768
2769 clear_extent_bit(io_tree, ordered_extent->file_offset,
2770 ordered_extent->file_offset + ordered_extent->len - 1,
2771 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2772 }
2773
Josef Bacik0cb59c92010-07-02 12:14:14 -04002774 if (nolock)
Josef Bacik7a7eaa42011-04-13 12:54:33 -04002775 trans = btrfs_join_transaction_nolock(root);
Josef Bacik0cb59c92010-07-02 12:14:14 -04002776 else
Josef Bacik7a7eaa42011-04-13 12:54:33 -04002777 trans = btrfs_join_transaction(root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002778 if (IS_ERR(trans)) {
2779 ret = PTR_ERR(trans);
2780 trans = NULL;
2781 goto out_unlock;
2782 }
Chris Masona79b7d42014-05-22 16:18:52 -07002783
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04002784 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
Yan, Zhengc2167752009-11-12 09:34:21 +00002785
Chris Masonc8b97812008-10-29 14:49:59 -04002786 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
Li Zefan261507a02010-12-17 14:21:50 +08002787 compress_type = ordered_extent->compress_type;
Yan Zhengd899e052008-10-30 14:25:28 -04002788 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
Li Zefan261507a02010-12-17 14:21:50 +08002789 BUG_ON(compress_type);
Yan, Zheng920bbbf2009-11-12 09:34:08 +00002790 ret = btrfs_mark_extent_written(trans, inode,
Yan Zhengd899e052008-10-30 14:25:28 -04002791 ordered_extent->file_offset,
2792 ordered_extent->file_offset +
Josef Bacik77cef2e2013-08-29 13:57:21 -04002793 logical_len);
Yan Zhengd899e052008-10-30 14:25:28 -04002794 } else {
Josef Bacik0af3d002010-06-21 14:48:16 -04002795 BUG_ON(root == root->fs_info->tree_root);
Yan Zhengd899e052008-10-30 14:25:28 -04002796 ret = insert_reserved_file_extent(trans, inode,
2797 ordered_extent->file_offset,
2798 ordered_extent->start,
2799 ordered_extent->disk_len,
Josef Bacik77cef2e2013-08-29 13:57:21 -04002800 logical_len, logical_len,
Li Zefan261507a02010-12-17 14:21:50 +08002801 compress_type, 0, 0,
Yan Zhengd899e052008-10-30 14:25:28 -04002802 BTRFS_FILE_EXTENT_REG);
Miao Xiee570fd22014-06-19 10:42:50 +08002803 if (!ret)
2804 btrfs_release_delalloc_bytes(root,
2805 ordered_extent->start,
2806 ordered_extent->disk_len);
Yan Zhengd899e052008-10-30 14:25:28 -04002807 }
Josef Bacik5dc562c2012-08-17 13:14:17 -04002808 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2809 ordered_extent->file_offset, ordered_extent->len,
2810 trans->transid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002811 if (ret < 0) {
2812 btrfs_abort_transaction(trans, root, ret);
Josef Bacik5fd02042012-05-02 14:00:54 -04002813 goto out_unlock;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002814 }
Josef Bacik2ac55d42010-02-03 19:33:23 +00002815
Chris Masone6dcd2d2008-07-17 12:53:50 -04002816 add_pending_csums(trans, inode, ordered_extent->file_offset,
2817 &ordered_extent->list);
2818
Josef Bacik6c760c02012-11-09 10:53:21 -05002819 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2820 ret = btrfs_update_inode_fallback(trans, root, inode);
2821 if (ret) { /* -ENOMEM or corruption */
2822 btrfs_abort_transaction(trans, root, ret);
2823 goto out_unlock;
Josef Bacik1ef30be2011-04-05 19:25:36 -04002824 }
2825 ret = 0;
Josef Bacik5fd02042012-05-02 14:00:54 -04002826out_unlock:
2827 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2828 ordered_extent->file_offset +
2829 ordered_extent->len - 1, &cached_state, GFP_NOFS);
Yan, Zhengc2167752009-11-12 09:34:21 +00002830out:
Josef Bacik5b0e95b2011-10-06 08:58:24 -04002831 if (root != root->fs_info->tree_root)
Josef Bacik0cb59c92010-07-02 12:14:14 -04002832 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
Miao Xiea698d0752012-09-20 01:51:59 -06002833 if (trans)
2834 btrfs_end_transaction(trans, root);
Josef Bacik0cb59c92010-07-02 12:14:14 -04002835
Josef Bacik77cef2e2013-08-29 13:57:21 -04002836 if (ret || truncated) {
2837 u64 start, end;
2838
2839 if (truncated)
2840 start = ordered_extent->file_offset + logical_len;
2841 else
2842 start = ordered_extent->file_offset;
2843 end = ordered_extent->file_offset + ordered_extent->len - 1;
2844 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2845
2846 /* Drop the cache for the part of the extent we didn't write. */
2847 btrfs_drop_extent_cache(inode, start, end, 0);
Josef Bacik5fd02042012-05-02 14:00:54 -04002848
Josef Bacik0bec9ef2013-01-31 14:58:00 -05002849 /*
2850 * If the ordered extent had an IOERR or something else went
2851 * wrong we need to return the space for this ordered extent
Josef Bacik77cef2e2013-08-29 13:57:21 -04002852 * back to the allocator. We only free the extent in the
2853 * truncated case if we didn't write out the extent at all.
Josef Bacik0bec9ef2013-01-31 14:58:00 -05002854 */
Josef Bacik77cef2e2013-08-29 13:57:21 -04002855 if ((ret || !logical_len) &&
2856 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
Josef Bacik0bec9ef2013-01-31 14:58:00 -05002857 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2858 btrfs_free_reserved_extent(root, ordered_extent->start,
Miao Xiee570fd22014-06-19 10:42:50 +08002859 ordered_extent->disk_len, 1);
Josef Bacik0bec9ef2013-01-31 14:58:00 -05002860 }
2861
2862
Josef Bacik5fd02042012-05-02 14:00:54 -04002863 /*
Liu Bo8bad3c02012-06-18 12:14:23 +08002864 * This needs to be done to make sure anybody waiting knows we are done
2865 * updating everything for this ordered extent.
Josef Bacik5fd02042012-05-02 14:00:54 -04002866 */
2867 btrfs_remove_ordered_extent(inode, ordered_extent);
2868
Liu Bo38c227d2013-01-29 03:18:40 +00002869 /* for snapshot-aware defrag */
Liu Bo6f519562013-10-29 10:45:05 +08002870 if (new) {
2871 if (ret) {
2872 free_sa_defrag_extent(new);
2873 atomic_dec(&root->fs_info->defrag_running);
2874 } else {
2875 relink_file_extents(new);
2876 }
2877 }
Liu Bo38c227d2013-01-29 03:18:40 +00002878
Chris Masone6dcd2d2008-07-17 12:53:50 -04002879 /* once for us */
2880 btrfs_put_ordered_extent(ordered_extent);
2881 /* once for the tree */
2882 btrfs_put_ordered_extent(ordered_extent);
2883
Josef Bacik5fd02042012-05-02 14:00:54 -04002884 return ret;
2885}
2886
2887static void finish_ordered_fn(struct btrfs_work *work)
2888{
2889 struct btrfs_ordered_extent *ordered_extent;
2890 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2891 btrfs_finish_ordered_io(ordered_extent);
Chris Masone6dcd2d2008-07-17 12:53:50 -04002892}
2893
Christoph Hellwigb2950862008-12-02 09:54:17 -05002894static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
Chris Mason211f90e2008-07-18 11:56:15 -04002895 struct extent_state *state, int uptodate)
2896{
Josef Bacik5fd02042012-05-02 14:00:54 -04002897 struct inode *inode = page->mapping->host;
2898 struct btrfs_root *root = BTRFS_I(inode)->root;
2899 struct btrfs_ordered_extent *ordered_extent = NULL;
Liu Bo9e0af232014-08-15 23:36:53 +08002900 struct btrfs_workqueue *wq;
2901 btrfs_work_func_t func;
Josef Bacik5fd02042012-05-02 14:00:54 -04002902
liubo1abe9b82011-03-24 11:18:59 +00002903 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2904
Chris Mason8b62b722009-09-02 16:53:46 -04002905 ClearPagePrivate2(page);
Josef Bacik5fd02042012-05-02 14:00:54 -04002906 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2907 end - start + 1, uptodate))
2908 return 0;
2909
Liu Bo9e0af232014-08-15 23:36:53 +08002910 if (btrfs_is_free_space_inode(inode)) {
2911 wq = root->fs_info->endio_freespace_worker;
2912 func = btrfs_freespace_write_helper;
2913 } else {
2914 wq = root->fs_info->endio_write_workers;
2915 func = btrfs_endio_write_helper;
2916 }
Josef Bacik5fd02042012-05-02 14:00:54 -04002917
Liu Bo9e0af232014-08-15 23:36:53 +08002918 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
2919 NULL);
2920 btrfs_queue_work(wq, &ordered_extent->work);
Josef Bacik5fd02042012-05-02 14:00:54 -04002921
2922 return 0;
Chris Mason211f90e2008-07-18 11:56:15 -04002923}
2924
Miao Xiedc380ae2014-09-12 18:43:55 +08002925static int __readpage_endio_check(struct inode *inode,
2926 struct btrfs_io_bio *io_bio,
2927 int icsum, struct page *page,
2928 int pgoff, u64 start, size_t len)
2929{
2930 char *kaddr;
2931 u32 csum_expected;
2932 u32 csum = ~(u32)0;
2933 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2934 DEFAULT_RATELIMIT_BURST);
2935
2936 csum_expected = *(((u32 *)io_bio->csum) + icsum);
2937
2938 kaddr = kmap_atomic(page);
2939 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
2940 btrfs_csum_final(csum, (char *)&csum);
2941 if (csum != csum_expected)
2942 goto zeroit;
2943
2944 kunmap_atomic(kaddr);
2945 return 0;
2946zeroit:
2947 if (__ratelimit(&_rs))
2948 btrfs_info(BTRFS_I(inode)->root->fs_info,
2949 "csum failed ino %llu off %llu csum %u expected csum %u",
2950 btrfs_ino(inode), start, csum, csum_expected);
2951 memset(kaddr + pgoff, 1, len);
2952 flush_dcache_page(page);
2953 kunmap_atomic(kaddr);
2954 if (csum_expected == 0)
2955 return 0;
2956 return -EIO;
2957}
2958
Chris Masond352ac62008-09-29 15:18:18 -04002959/*
Chris Masond352ac62008-09-29 15:18:18 -04002960 * when reads are done, we need to check csums to verify the data is correct
Jan Schmidt4a54c8c2011-07-22 15:41:52 +02002961 * if there's a match, we allow the bio to finish. If not, the code in
2962 * extent_io.c will try to find good copies for us.
Chris Masond352ac62008-09-29 15:18:18 -04002963 */
Miao Xiefacc8a222013-07-25 19:22:34 +08002964static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2965 u64 phy_offset, struct page *page,
2966 u64 start, u64 end, int mirror)
Chris Mason07157aa2007-08-30 08:50:51 -04002967{
Miao Xie4eee4fa2012-12-21 09:17:45 +00002968 size_t offset = start - page_offset(page);
Chris Mason07157aa2007-08-30 08:50:51 -04002969 struct inode *inode = page->mapping->host;
Chris Masond1310b22008-01-24 16:13:08 -05002970 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Chris Masonff79f812007-10-15 16:22:25 -04002971 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masond1310b22008-01-24 16:13:08 -05002972
Chris Masond20f7042008-12-08 16:58:54 -05002973 if (PageChecked(page)) {
2974 ClearPageChecked(page);
Miao Xiedc380ae2014-09-12 18:43:55 +08002975 return 0;
Chris Masond20f7042008-12-08 16:58:54 -05002976 }
Christoph Hellwig6cbff002009-04-17 10:37:41 +02002977
2978 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
Miao Xiedc380ae2014-09-12 18:43:55 +08002979 return 0;
Chris Masond20f7042008-12-08 16:58:54 -05002980
Yan Zheng17d217f2008-12-12 10:03:38 -05002981 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
Chris Mason9655d292009-09-02 15:22:30 -04002982 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
Yan Zheng17d217f2008-12-12 10:03:38 -05002983 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2984 GFP_NOFS);
2985 return 0;
2986 }
2987
Miao Xiefacc8a222013-07-25 19:22:34 +08002988 phy_offset >>= inode->i_sb->s_blocksize_bits;
Miao Xiedc380ae2014-09-12 18:43:55 +08002989 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
2990 start, (size_t)(end - start + 1));
Chris Mason07157aa2007-08-30 08:50:51 -04002991}
Chris Masonb888db22007-08-27 16:49:44 -04002992
Yan, Zheng24bbcf02009-11-12 09:36:34 +00002993struct delayed_iput {
2994 struct list_head list;
2995 struct inode *inode;
2996};
2997
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002998/* JDM: If this is fs-wide, why can't we add a pointer to
2999 * btrfs_inode instead and avoid the allocation? */
Yan, Zheng24bbcf02009-11-12 09:36:34 +00003000void btrfs_add_delayed_iput(struct inode *inode)
3001{
3002 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3003 struct delayed_iput *delayed;
3004
3005 if (atomic_add_unless(&inode->i_count, -1, 1))
3006 return;
3007
3008 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
3009 delayed->inode = inode;
3010
3011 spin_lock(&fs_info->delayed_iput_lock);
3012 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
3013 spin_unlock(&fs_info->delayed_iput_lock);
3014}
3015
3016void btrfs_run_delayed_iputs(struct btrfs_root *root)
3017{
3018 LIST_HEAD(list);
3019 struct btrfs_fs_info *fs_info = root->fs_info;
3020 struct delayed_iput *delayed;
3021 int empty;
3022
3023 spin_lock(&fs_info->delayed_iput_lock);
3024 empty = list_empty(&fs_info->delayed_iputs);
3025 spin_unlock(&fs_info->delayed_iput_lock);
3026 if (empty)
3027 return;
3028
Yan, Zheng24bbcf02009-11-12 09:36:34 +00003029 spin_lock(&fs_info->delayed_iput_lock);
3030 list_splice_init(&fs_info->delayed_iputs, &list);
3031 spin_unlock(&fs_info->delayed_iput_lock);
3032
3033 while (!list_empty(&list)) {
3034 delayed = list_entry(list.next, struct delayed_iput, list);
3035 list_del(&delayed->list);
3036 iput(delayed->inode);
3037 kfree(delayed);
3038 }
Yan, Zheng24bbcf02009-11-12 09:36:34 +00003039}
3040
Yan, Zhengd68fc572010-05-16 10:49:58 -04003041/*
Justin P. Mattock42b2aa82011-11-28 20:31:00 -08003042 * This is called in transaction commit time. If there are no orphan
Yan, Zhengd68fc572010-05-16 10:49:58 -04003043 * files in the subvolume, it removes orphan item and frees block_rsv
3044 * structure.
3045 */
3046void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3047 struct btrfs_root *root)
3048{
Josef Bacik90290e12011-12-02 15:44:12 -05003049 struct btrfs_block_rsv *block_rsv;
Yan, Zhengd68fc572010-05-16 10:49:58 -04003050 int ret;
3051
Josef Bacik8a35d952012-05-23 14:26:42 -04003052 if (atomic_read(&root->orphan_inodes) ||
Yan, Zhengd68fc572010-05-16 10:49:58 -04003053 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3054 return;
3055
Josef Bacik90290e12011-12-02 15:44:12 -05003056 spin_lock(&root->orphan_lock);
Josef Bacik8a35d952012-05-23 14:26:42 -04003057 if (atomic_read(&root->orphan_inodes)) {
Josef Bacik90290e12011-12-02 15:44:12 -05003058 spin_unlock(&root->orphan_lock);
3059 return;
3060 }
3061
3062 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3063 spin_unlock(&root->orphan_lock);
3064 return;
3065 }
3066
3067 block_rsv = root->orphan_block_rsv;
3068 root->orphan_block_rsv = NULL;
3069 spin_unlock(&root->orphan_lock);
3070
Miao Xie27cdeb72014-04-02 19:51:05 +08003071 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
Yan, Zhengd68fc572010-05-16 10:49:58 -04003072 btrfs_root_refs(&root->root_item) > 0) {
3073 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3074 root->root_key.objectid);
Josef Bacik4ef31a42013-08-13 14:10:08 -04003075 if (ret)
3076 btrfs_abort_transaction(trans, root, ret);
3077 else
Miao Xie27cdeb72014-04-02 19:51:05 +08003078 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3079 &root->state);
Yan, Zhengd68fc572010-05-16 10:49:58 -04003080 }
3081
Josef Bacik90290e12011-12-02 15:44:12 -05003082 if (block_rsv) {
3083 WARN_ON(block_rsv->size > 0);
3084 btrfs_free_block_rsv(root, block_rsv);
Yan, Zhengd68fc572010-05-16 10:49:58 -04003085 }
3086}
3087
3088/*
Josef Bacik7b128762008-07-24 12:17:14 -04003089 * This creates an orphan entry for the given inode in case something goes
3090 * wrong in the middle of an unlink/truncate.
Yan, Zhengd68fc572010-05-16 10:49:58 -04003091 *
3092 * NOTE: caller of this function should reserve 5 units of metadata for
3093 * this function.
Josef Bacik7b128762008-07-24 12:17:14 -04003094 */
3095int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3096{
3097 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zhengd68fc572010-05-16 10:49:58 -04003098 struct btrfs_block_rsv *block_rsv = NULL;
3099 int reserve = 0;
3100 int insert = 0;
3101 int ret;
Josef Bacik7b128762008-07-24 12:17:14 -04003102
Yan, Zhengd68fc572010-05-16 10:49:58 -04003103 if (!root->orphan_block_rsv) {
Miao Xie66d8f3d2012-09-06 04:02:28 -06003104 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
Tsutomu Itohb5324022011-07-19 07:27:20 +00003105 if (!block_rsv)
3106 return -ENOMEM;
Josef Bacik7b128762008-07-24 12:17:14 -04003107 }
3108
Yan, Zhengd68fc572010-05-16 10:49:58 -04003109 spin_lock(&root->orphan_lock);
3110 if (!root->orphan_block_rsv) {
3111 root->orphan_block_rsv = block_rsv;
3112 } else if (block_rsv) {
3113 btrfs_free_block_rsv(root, block_rsv);
3114 block_rsv = NULL;
3115 }
Josef Bacik7b128762008-07-24 12:17:14 -04003116
Josef Bacik8a35d952012-05-23 14:26:42 -04003117 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3118 &BTRFS_I(inode)->runtime_flags)) {
Yan, Zhengd68fc572010-05-16 10:49:58 -04003119#if 0
3120 /*
3121 * For proper ENOSPC handling, we should do orphan
3122 * cleanup when mounting. But this introduces backward
3123 * compatibility issue.
3124 */
3125 if (!xchg(&root->orphan_item_inserted, 1))
3126 insert = 2;
3127 else
3128 insert = 1;
3129#endif
3130 insert = 1;
Miao Xie321f0e72012-08-28 22:13:02 -06003131 atomic_inc(&root->orphan_inodes);
Yan, Zhengd68fc572010-05-16 10:49:58 -04003132 }
Josef Bacik7b128762008-07-24 12:17:14 -04003133
Josef Bacik72ac3c02012-05-23 14:13:11 -04003134 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3135 &BTRFS_I(inode)->runtime_flags))
Yan, Zhengd68fc572010-05-16 10:49:58 -04003136 reserve = 1;
Yan, Zhengd68fc572010-05-16 10:49:58 -04003137 spin_unlock(&root->orphan_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04003138
Yan, Zhengd68fc572010-05-16 10:49:58 -04003139 /* grab metadata reservation from transaction handle */
3140 if (reserve) {
3141 ret = btrfs_orphan_reserve_metadata(trans, inode);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003142 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
Yan, Zhengd68fc572010-05-16 10:49:58 -04003143 }
3144
3145 /* insert an orphan item to track this unlinked/truncated file */
3146 if (insert >= 1) {
Li Zefan33345d012011-04-20 10:31:50 +08003147 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
Josef Bacik4ef31a42013-08-13 14:10:08 -04003148 if (ret) {
Filipe David Borba Manana703c88e2013-09-22 21:54:55 +01003149 atomic_dec(&root->orphan_inodes);
Josef Bacik4ef31a42013-08-13 14:10:08 -04003150 if (reserve) {
3151 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3152 &BTRFS_I(inode)->runtime_flags);
3153 btrfs_orphan_release_metadata(inode);
3154 }
3155 if (ret != -EEXIST) {
Josef Bacike8e7cff2013-08-21 15:54:00 -04003156 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3157 &BTRFS_I(inode)->runtime_flags);
Josef Bacik4ef31a42013-08-13 14:10:08 -04003158 btrfs_abort_transaction(trans, root, ret);
3159 return ret;
3160 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003161 }
3162 ret = 0;
Yan, Zhengd68fc572010-05-16 10:49:58 -04003163 }
3164
3165 /* insert an orphan item to track subvolume contains orphan files */
3166 if (insert >= 2) {
3167 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3168 root->root_key.objectid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003169 if (ret && ret != -EEXIST) {
3170 btrfs_abort_transaction(trans, root, ret);
3171 return ret;
3172 }
Yan, Zhengd68fc572010-05-16 10:49:58 -04003173 }
3174 return 0;
Josef Bacik7b128762008-07-24 12:17:14 -04003175}
3176
3177/*
3178 * We have done the truncate/delete so we can go ahead and remove the orphan
3179 * item for this particular inode.
3180 */
Eric Sandeen48a3b632013-04-25 20:41:01 +00003181static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3182 struct inode *inode)
Josef Bacik7b128762008-07-24 12:17:14 -04003183{
3184 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zhengd68fc572010-05-16 10:49:58 -04003185 int delete_item = 0;
3186 int release_rsv = 0;
Josef Bacik7b128762008-07-24 12:17:14 -04003187 int ret = 0;
3188
Yan, Zhengd68fc572010-05-16 10:49:58 -04003189 spin_lock(&root->orphan_lock);
Josef Bacik8a35d952012-05-23 14:26:42 -04003190 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3191 &BTRFS_I(inode)->runtime_flags))
Yan, Zhengd68fc572010-05-16 10:49:58 -04003192 delete_item = 1;
Josef Bacik7b128762008-07-24 12:17:14 -04003193
Josef Bacik72ac3c02012-05-23 14:13:11 -04003194 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3195 &BTRFS_I(inode)->runtime_flags))
Yan, Zhengd68fc572010-05-16 10:49:58 -04003196 release_rsv = 1;
Yan, Zhengd68fc572010-05-16 10:49:58 -04003197 spin_unlock(&root->orphan_lock);
3198
Filipe David Borba Manana703c88e2013-09-22 21:54:55 +01003199 if (delete_item) {
Josef Bacik8a35d952012-05-23 14:26:42 -04003200 atomic_dec(&root->orphan_inodes);
Filipe David Borba Manana703c88e2013-09-22 21:54:55 +01003201 if (trans)
3202 ret = btrfs_del_orphan_item(trans, root,
3203 btrfs_ino(inode));
Josef Bacik8a35d952012-05-23 14:26:42 -04003204 }
Josef Bacik7b128762008-07-24 12:17:14 -04003205
Filipe David Borba Manana703c88e2013-09-22 21:54:55 +01003206 if (release_rsv)
3207 btrfs_orphan_release_metadata(inode);
3208
Josef Bacik4ef31a42013-08-13 14:10:08 -04003209 return ret;
Josef Bacik7b128762008-07-24 12:17:14 -04003210}
3211
3212/*
3213 * this cleans up any orphans that may be left on the list from the last use
3214 * of this root.
3215 */
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003216int btrfs_orphan_cleanup(struct btrfs_root *root)
Josef Bacik7b128762008-07-24 12:17:14 -04003217{
3218 struct btrfs_path *path;
3219 struct extent_buffer *leaf;
Josef Bacik7b128762008-07-24 12:17:14 -04003220 struct btrfs_key key, found_key;
3221 struct btrfs_trans_handle *trans;
3222 struct inode *inode;
Josef Bacik8f6d7f42011-09-26 15:55:20 -04003223 u64 last_objectid = 0;
Josef Bacik7b128762008-07-24 12:17:14 -04003224 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3225
Yan, Zhengd68fc572010-05-16 10:49:58 -04003226 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003227 return 0;
Yan, Zhengc71bf092009-11-12 09:34:40 +00003228
3229 path = btrfs_alloc_path();
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003230 if (!path) {
3231 ret = -ENOMEM;
3232 goto out;
3233 }
Josef Bacik7b128762008-07-24 12:17:14 -04003234 path->reada = -1;
3235
3236 key.objectid = BTRFS_ORPHAN_OBJECTID;
David Sterba962a2982014-06-04 18:41:45 +02003237 key.type = BTRFS_ORPHAN_ITEM_KEY;
Josef Bacik7b128762008-07-24 12:17:14 -04003238 key.offset = (u64)-1;
3239
Josef Bacik7b128762008-07-24 12:17:14 -04003240 while (1) {
3241 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003242 if (ret < 0)
3243 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04003244
3245 /*
3246 * if ret == 0 means we found what we were searching for, which
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003247 * is weird, but possible, so only screw with path if we didn't
Josef Bacik7b128762008-07-24 12:17:14 -04003248 * find the key and see if we have stuff that matches
3249 */
3250 if (ret > 0) {
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003251 ret = 0;
Josef Bacik7b128762008-07-24 12:17:14 -04003252 if (path->slots[0] == 0)
3253 break;
3254 path->slots[0]--;
3255 }
3256
3257 /* pull out the item */
3258 leaf = path->nodes[0];
Josef Bacik7b128762008-07-24 12:17:14 -04003259 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3260
3261 /* make sure the item matches what we want */
3262 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3263 break;
David Sterba962a2982014-06-04 18:41:45 +02003264 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
Josef Bacik7b128762008-07-24 12:17:14 -04003265 break;
3266
3267 /* release the path since we're done with it */
David Sterbab3b4aa72011-04-21 01:20:15 +02003268 btrfs_release_path(path);
Josef Bacik7b128762008-07-24 12:17:14 -04003269
3270 /*
3271 * this is where we are basically btrfs_lookup, without the
3272 * crossing root thing. we store the inode number in the
3273 * offset of the orphan item.
3274 */
Josef Bacik8f6d7f42011-09-26 15:55:20 -04003275
3276 if (found_key.offset == last_objectid) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00003277 btrfs_err(root->fs_info,
3278 "Error removing orphan entry, stopping orphan cleanup");
Josef Bacik8f6d7f42011-09-26 15:55:20 -04003279 ret = -EINVAL;
3280 goto out;
3281 }
3282
3283 last_objectid = found_key.offset;
3284
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003285 found_key.objectid = found_key.offset;
3286 found_key.type = BTRFS_INODE_ITEM_KEY;
3287 found_key.offset = 0;
Josef Bacik73f73412009-12-04 17:38:27 +00003288 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
Rusty Russell8c6ffba2013-07-15 11:20:32 +09303289 ret = PTR_ERR_OR_ZERO(inode);
Josef Bacika8c9e572011-09-21 16:55:59 -04003290 if (ret && ret != -ESTALE)
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003291 goto out;
Josef Bacika8c9e572011-09-21 16:55:59 -04003292
Arne Jansenf8e9e0b2011-12-14 20:12:02 -05003293 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3294 struct btrfs_root *dead_root;
3295 struct btrfs_fs_info *fs_info = root->fs_info;
3296 int is_dead_root = 0;
3297
3298 /*
3299 * this is an orphan in the tree root. Currently these
3300 * could come from 2 sources:
3301 * a) a snapshot deletion in progress
3302 * b) a free space cache inode
3303 * We need to distinguish those two, as the snapshot
3304 * orphan must not get deleted.
3305 * find_dead_roots already ran before us, so if this
3306 * is a snapshot deletion, we should find the root
3307 * in the dead_roots list
3308 */
3309 spin_lock(&fs_info->trans_lock);
3310 list_for_each_entry(dead_root, &fs_info->dead_roots,
3311 root_list) {
3312 if (dead_root->root_key.objectid ==
3313 found_key.objectid) {
3314 is_dead_root = 1;
3315 break;
3316 }
3317 }
3318 spin_unlock(&fs_info->trans_lock);
3319 if (is_dead_root) {
3320 /* prevent this orphan from being found again */
3321 key.offset = found_key.objectid - 1;
3322 continue;
3323 }
3324 }
Josef Bacika8c9e572011-09-21 16:55:59 -04003325 /*
3326 * Inode is already gone but the orphan item is still there,
3327 * kill the orphan item.
3328 */
3329 if (ret == -ESTALE) {
3330 trans = btrfs_start_transaction(root, 1);
3331 if (IS_ERR(trans)) {
3332 ret = PTR_ERR(trans);
3333 goto out;
3334 }
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00003335 btrfs_debug(root->fs_info, "auto deleting %Lu",
3336 found_key.objectid);
Josef Bacika8c9e572011-09-21 16:55:59 -04003337 ret = btrfs_del_orphan_item(trans, root,
3338 found_key.objectid);
Josef Bacika8c9e572011-09-21 16:55:59 -04003339 btrfs_end_transaction(trans, root);
Josef Bacik4ef31a42013-08-13 14:10:08 -04003340 if (ret)
3341 goto out;
Josef Bacika8c9e572011-09-21 16:55:59 -04003342 continue;
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003343 }
Josef Bacik7b128762008-07-24 12:17:14 -04003344
Josef Bacik7b128762008-07-24 12:17:14 -04003345 /*
3346 * add this inode to the orphan list so btrfs_orphan_del does
3347 * the proper thing when we hit it
3348 */
Josef Bacik8a35d952012-05-23 14:26:42 -04003349 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3350 &BTRFS_I(inode)->runtime_flags);
Josef Bacik925396e2013-02-01 15:57:47 -05003351 atomic_inc(&root->orphan_inodes);
Josef Bacik7b128762008-07-24 12:17:14 -04003352
Josef Bacik7b128762008-07-24 12:17:14 -04003353 /* if we have links, this was a truncate, lets do that */
3354 if (inode->i_nlink) {
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05303355 if (WARN_ON(!S_ISREG(inode->i_mode))) {
Josef Bacika41ad392011-01-31 15:30:16 -05003356 iput(inode);
3357 continue;
3358 }
Josef Bacik7b128762008-07-24 12:17:14 -04003359 nr_truncate++;
Josef Bacikf3fe8202013-01-07 17:03:21 -05003360
3361 /* 1 for the orphan item deletion. */
3362 trans = btrfs_start_transaction(root, 1);
3363 if (IS_ERR(trans)) {
Josef Bacikc69b26b2013-06-03 16:51:23 -04003364 iput(inode);
Josef Bacikf3fe8202013-01-07 17:03:21 -05003365 ret = PTR_ERR(trans);
3366 goto out;
3367 }
3368 ret = btrfs_orphan_add(trans, inode);
3369 btrfs_end_transaction(trans, root);
Josef Bacikc69b26b2013-06-03 16:51:23 -04003370 if (ret) {
3371 iput(inode);
Josef Bacikf3fe8202013-01-07 17:03:21 -05003372 goto out;
Josef Bacikc69b26b2013-06-03 16:51:23 -04003373 }
Josef Bacikf3fe8202013-01-07 17:03:21 -05003374
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003375 ret = btrfs_truncate(inode);
Josef Bacik4a7d0f62013-02-07 16:27:28 -05003376 if (ret)
3377 btrfs_orphan_del(NULL, inode);
Josef Bacik7b128762008-07-24 12:17:14 -04003378 } else {
3379 nr_unlink++;
3380 }
3381
3382 /* this will do delete_inode and everything for us */
3383 iput(inode);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003384 if (ret)
3385 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04003386 }
Miao Xie3254c872011-11-10 20:45:05 -05003387 /* release the path since we're done with it */
3388 btrfs_release_path(path);
3389
Yan, Zhengd68fc572010-05-16 10:49:58 -04003390 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3391
3392 if (root->orphan_block_rsv)
3393 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3394 (u64)-1);
3395
Miao Xie27cdeb72014-04-02 19:51:05 +08003396 if (root->orphan_block_rsv ||
3397 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
Josef Bacik7a7eaa42011-04-13 12:54:33 -04003398 trans = btrfs_join_transaction(root);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003399 if (!IS_ERR(trans))
3400 btrfs_end_transaction(trans, root);
Yan, Zhengd68fc572010-05-16 10:49:58 -04003401 }
Josef Bacik7b128762008-07-24 12:17:14 -04003402
3403 if (nr_unlink)
David Sterba4884b472013-03-20 13:31:27 +00003404 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
Josef Bacik7b128762008-07-24 12:17:14 -04003405 if (nr_truncate)
David Sterba4884b472013-03-20 13:31:27 +00003406 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003407
3408out:
3409 if (ret)
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00003410 btrfs_crit(root->fs_info,
3411 "could not do orphan cleanup %d", ret);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003412 btrfs_free_path(path);
3413 return ret;
Josef Bacik7b128762008-07-24 12:17:14 -04003414}
3415
Chris Masond352ac62008-09-29 15:18:18 -04003416/*
Chris Mason46a53cc2009-04-27 11:47:50 -04003417 * very simple check to peek ahead in the leaf looking for xattrs. If we
3418 * don't find any xattrs, we know there can't be any acls.
3419 *
3420 * slot is the slot the inode is in, objectid is the objectid of the inode
3421 */
3422static noinline int acls_after_inode_item(struct extent_buffer *leaf,
Filipe David Borba Manana63541922014-01-07 11:47:46 +00003423 int slot, u64 objectid,
3424 int *first_xattr_slot)
Chris Mason46a53cc2009-04-27 11:47:50 -04003425{
3426 u32 nritems = btrfs_header_nritems(leaf);
3427 struct btrfs_key found_key;
Josef Bacikf23b5a52013-06-19 10:16:26 -04003428 static u64 xattr_access = 0;
3429 static u64 xattr_default = 0;
Chris Mason46a53cc2009-04-27 11:47:50 -04003430 int scanned = 0;
3431
Josef Bacikf23b5a52013-06-19 10:16:26 -04003432 if (!xattr_access) {
3433 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3434 strlen(POSIX_ACL_XATTR_ACCESS));
3435 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3436 strlen(POSIX_ACL_XATTR_DEFAULT));
3437 }
3438
Chris Mason46a53cc2009-04-27 11:47:50 -04003439 slot++;
Filipe David Borba Manana63541922014-01-07 11:47:46 +00003440 *first_xattr_slot = -1;
Chris Mason46a53cc2009-04-27 11:47:50 -04003441 while (slot < nritems) {
3442 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3443
3444 /* we found a different objectid, there must not be acls */
3445 if (found_key.objectid != objectid)
3446 return 0;
3447
3448 /* we found an xattr, assume we've got an acl */
Josef Bacikf23b5a52013-06-19 10:16:26 -04003449 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
Filipe David Borba Manana63541922014-01-07 11:47:46 +00003450 if (*first_xattr_slot == -1)
3451 *first_xattr_slot = slot;
Josef Bacikf23b5a52013-06-19 10:16:26 -04003452 if (found_key.offset == xattr_access ||
3453 found_key.offset == xattr_default)
3454 return 1;
3455 }
Chris Mason46a53cc2009-04-27 11:47:50 -04003456
3457 /*
3458 * we found a key greater than an xattr key, there can't
3459 * be any acls later on
3460 */
3461 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3462 return 0;
3463
3464 slot++;
3465 scanned++;
3466
3467 /*
3468 * it goes inode, inode backrefs, xattrs, extents,
3469 * so if there are a ton of hard links to an inode there can
3470 * be a lot of backrefs. Don't waste time searching too hard,
3471 * this is just an optimization
3472 */
3473 if (scanned >= 8)
3474 break;
3475 }
3476 /* we hit the end of the leaf before we found an xattr or
3477 * something larger than an xattr. We have to assume the inode
3478 * has acls
3479 */
Filipe David Borba Manana63541922014-01-07 11:47:46 +00003480 if (*first_xattr_slot == -1)
3481 *first_xattr_slot = slot;
Chris Mason46a53cc2009-04-27 11:47:50 -04003482 return 1;
3483}
3484
3485/*
Chris Masond352ac62008-09-29 15:18:18 -04003486 * read an inode from the btree into the in-memory inode
3487 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003488static void btrfs_read_locked_inode(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04003489{
3490 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04003491 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04003492 struct btrfs_inode_item *inode_item;
Chris Mason0b86a832008-03-24 15:01:56 -04003493 struct btrfs_timespec *tspec;
Chris Mason39279cc2007-06-12 06:35:45 -04003494 struct btrfs_root *root = BTRFS_I(inode)->root;
3495 struct btrfs_key location;
Miao Xie67de1172013-12-26 13:07:06 +08003496 unsigned long ptr;
Chris Mason46a53cc2009-04-27 11:47:50 -04003497 int maybe_acls;
Josef Bacik618e21d2007-07-11 10:18:17 -04003498 u32 rdev;
Chris Mason39279cc2007-06-12 06:35:45 -04003499 int ret;
Miao Xie2f7e33d2011-06-23 07:27:13 +00003500 bool filled = false;
Filipe David Borba Manana63541922014-01-07 11:47:46 +00003501 int first_xattr_slot;
Miao Xie2f7e33d2011-06-23 07:27:13 +00003502
3503 ret = btrfs_fill_inode(inode, &rdev);
3504 if (!ret)
3505 filled = true;
Chris Mason39279cc2007-06-12 06:35:45 -04003506
3507 path = btrfs_alloc_path();
Mark Fasheh1748f842011-07-12 11:25:31 -07003508 if (!path)
3509 goto make_bad;
3510
Chris Mason39279cc2007-06-12 06:35:45 -04003511 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
Chris Masondc17ff82008-01-08 15:46:30 -05003512
Chris Mason39279cc2007-06-12 06:35:45 -04003513 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
Chris Mason5f39d392007-10-15 16:14:19 -04003514 if (ret)
Chris Mason39279cc2007-06-12 06:35:45 -04003515 goto make_bad;
Chris Mason39279cc2007-06-12 06:35:45 -04003516
Chris Mason5f39d392007-10-15 16:14:19 -04003517 leaf = path->nodes[0];
Miao Xie2f7e33d2011-06-23 07:27:13 +00003518
3519 if (filled)
Miao Xie67de1172013-12-26 13:07:06 +08003520 goto cache_index;
Miao Xie2f7e33d2011-06-23 07:27:13 +00003521
Chris Mason5f39d392007-10-15 16:14:19 -04003522 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3523 struct btrfs_inode_item);
Chris Mason5f39d392007-10-15 16:14:19 -04003524 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
Miklos Szeredibfe86842011-10-28 14:13:29 +02003525 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
Eric W. Biederman2f2f43d2012-02-10 11:05:07 -08003526 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3527 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
Chris Masondbe674a2008-07-17 12:54:05 -04003528 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
Chris Mason5f39d392007-10-15 16:14:19 -04003529
3530 tspec = btrfs_inode_atime(inode_item);
3531 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3532 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3533
3534 tspec = btrfs_inode_mtime(inode_item);
3535 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3536 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3537
3538 tspec = btrfs_inode_ctime(inode_item);
3539 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3540 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3541
Yan Zhenga76a3cd2008-10-09 11:46:29 -04003542 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
Chris Masone02119d2008-09-05 16:13:11 -04003543 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
Josef Bacik5dc562c2012-08-17 13:14:17 -04003544 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3545
3546 /*
3547 * If we were modified in the current generation and evicted from memory
3548 * and then re-read we need to do a full sync since we don't have any
3549 * idea about which extents were modified before we were evicted from
3550 * cache.
3551 */
3552 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3553 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3554 &BTRFS_I(inode)->runtime_flags);
3555
Josef Bacik0c4d2d92012-04-05 15:03:02 -04003556 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
Chris Masone02119d2008-09-05 16:13:11 -04003557 inode->i_generation = BTRFS_I(inode)->generation;
Josef Bacik618e21d2007-07-11 10:18:17 -04003558 inode->i_rdev = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04003559 rdev = btrfs_inode_rdev(leaf, inode_item);
3560
Josef Bacikaec74772008-07-24 12:12:38 -04003561 BTRFS_I(inode)->index_cnt = (u64)-1;
Yan Zhengd2fb3432008-12-11 16:30:39 -05003562 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
Miao Xie67de1172013-12-26 13:07:06 +08003563
3564cache_index:
3565 path->slots[0]++;
3566 if (inode->i_nlink != 1 ||
3567 path->slots[0] >= btrfs_header_nritems(leaf))
3568 goto cache_acl;
3569
3570 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3571 if (location.objectid != btrfs_ino(inode))
3572 goto cache_acl;
3573
3574 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3575 if (location.type == BTRFS_INODE_REF_KEY) {
3576 struct btrfs_inode_ref *ref;
3577
3578 ref = (struct btrfs_inode_ref *)ptr;
3579 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3580 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3581 struct btrfs_inode_extref *extref;
3582
3583 extref = (struct btrfs_inode_extref *)ptr;
3584 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3585 extref);
3586 }
Miao Xie2f7e33d2011-06-23 07:27:13 +00003587cache_acl:
Chris Mason46a53cc2009-04-27 11:47:50 -04003588 /*
3589 * try to precache a NULL acl entry for files that don't have
3590 * any xattrs or acls
3591 */
Li Zefan33345d012011-04-20 10:31:50 +08003592 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
Filipe David Borba Manana63541922014-01-07 11:47:46 +00003593 btrfs_ino(inode), &first_xattr_slot);
3594 if (first_xattr_slot != -1) {
3595 path->slots[0] = first_xattr_slot;
3596 ret = btrfs_load_inode_props(inode, path);
3597 if (ret)
3598 btrfs_err(root->fs_info,
David Sterba351fd352014-05-15 16:48:20 +02003599 "error loading props for ino %llu (root %llu): %d",
Filipe David Borba Manana63541922014-01-07 11:47:46 +00003600 btrfs_ino(inode),
3601 root->root_key.objectid, ret);
3602 }
3603 btrfs_free_path(path);
3604
Al Viro72c04902009-06-24 16:58:48 -04003605 if (!maybe_acls)
3606 cache_no_acl(inode);
Chris Mason46a53cc2009-04-27 11:47:50 -04003607
Chris Mason39279cc2007-06-12 06:35:45 -04003608 switch (inode->i_mode & S_IFMT) {
Chris Mason39279cc2007-06-12 06:35:45 -04003609 case S_IFREG:
3610 inode->i_mapping->a_ops = &btrfs_aops;
Chris Masond1310b22008-01-24 16:13:08 -05003611 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -04003612 inode->i_fop = &btrfs_file_operations;
3613 inode->i_op = &btrfs_file_inode_operations;
3614 break;
3615 case S_IFDIR:
3616 inode->i_fop = &btrfs_dir_file_operations;
3617 if (root == root->fs_info->tree_root)
3618 inode->i_op = &btrfs_dir_ro_inode_operations;
3619 else
3620 inode->i_op = &btrfs_dir_inode_operations;
3621 break;
3622 case S_IFLNK:
3623 inode->i_op = &btrfs_symlink_inode_operations;
3624 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3625 break;
Josef Bacik618e21d2007-07-11 10:18:17 -04003626 default:
Jim Owens0279b4c2009-02-04 09:29:13 -05003627 inode->i_op = &btrfs_special_inode_operations;
Josef Bacik618e21d2007-07-11 10:18:17 -04003628 init_special_inode(inode, inode->i_mode, rdev);
3629 break;
Chris Mason39279cc2007-06-12 06:35:45 -04003630 }
Christoph Hellwig6cbff002009-04-17 10:37:41 +02003631
3632 btrfs_update_iflags(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04003633 return;
3634
3635make_bad:
Chris Mason39279cc2007-06-12 06:35:45 -04003636 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04003637 make_bad_inode(inode);
3638}
3639
Chris Masond352ac62008-09-29 15:18:18 -04003640/*
3641 * given a leaf and an inode, copy the inode fields into the leaf
3642 */
Chris Masone02119d2008-09-05 16:13:11 -04003643static void fill_inode_item(struct btrfs_trans_handle *trans,
3644 struct extent_buffer *leaf,
Chris Mason5f39d392007-10-15 16:14:19 -04003645 struct btrfs_inode_item *item,
Chris Mason39279cc2007-06-12 06:35:45 -04003646 struct inode *inode)
3647{
Liu Bo51fab692012-12-27 09:01:21 +00003648 struct btrfs_map_token token;
Chris Mason5f39d392007-10-15 16:14:19 -04003649
Liu Bo51fab692012-12-27 09:01:21 +00003650 btrfs_init_map_token(&token);
Chris Mason5f39d392007-10-15 16:14:19 -04003651
Liu Bo51fab692012-12-27 09:01:21 +00003652 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3653 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3654 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3655 &token);
3656 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3657 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
Chris Mason5f39d392007-10-15 16:14:19 -04003658
Liu Bo51fab692012-12-27 09:01:21 +00003659 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3660 inode->i_atime.tv_sec, &token);
3661 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3662 inode->i_atime.tv_nsec, &token);
Chris Mason5f39d392007-10-15 16:14:19 -04003663
Liu Bo51fab692012-12-27 09:01:21 +00003664 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3665 inode->i_mtime.tv_sec, &token);
3666 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3667 inode->i_mtime.tv_nsec, &token);
3668
3669 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3670 inode->i_ctime.tv_sec, &token);
3671 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3672 inode->i_ctime.tv_nsec, &token);
3673
3674 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3675 &token);
3676 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3677 &token);
3678 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3679 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3680 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3681 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3682 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
Chris Mason39279cc2007-06-12 06:35:45 -04003683}
3684
Chris Masond352ac62008-09-29 15:18:18 -04003685/*
3686 * copy everything in the in-memory inode into the btree.
3687 */
Chris Mason21151332011-11-10 20:39:08 -05003688static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
Chris Masond3977122009-01-05 21:25:51 -05003689 struct btrfs_root *root, struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04003690{
3691 struct btrfs_inode_item *inode_item;
3692 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04003693 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04003694 int ret;
3695
3696 path = btrfs_alloc_path();
Miao Xie16cdcec2011-04-22 18:12:22 +08003697 if (!path)
3698 return -ENOMEM;
3699
Chris Masonb9473432009-03-13 11:00:37 -04003700 path->leave_spinning = 1;
Miao Xie16cdcec2011-04-22 18:12:22 +08003701 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3702 1);
Chris Mason39279cc2007-06-12 06:35:45 -04003703 if (ret) {
3704 if (ret > 0)
3705 ret = -ENOENT;
3706 goto failed;
3707 }
3708
Chris Mason5f39d392007-10-15 16:14:19 -04003709 leaf = path->nodes[0];
3710 inode_item = btrfs_item_ptr(leaf, path->slots[0],
Miao Xie16cdcec2011-04-22 18:12:22 +08003711 struct btrfs_inode_item);
Chris Mason39279cc2007-06-12 06:35:45 -04003712
Chris Masone02119d2008-09-05 16:13:11 -04003713 fill_inode_item(trans, leaf, inode_item, inode);
Chris Mason5f39d392007-10-15 16:14:19 -04003714 btrfs_mark_buffer_dirty(leaf);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04003715 btrfs_set_inode_last_trans(trans, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04003716 ret = 0;
3717failed:
Chris Mason39279cc2007-06-12 06:35:45 -04003718 btrfs_free_path(path);
3719 return ret;
3720}
3721
Chris Masond352ac62008-09-29 15:18:18 -04003722/*
Chris Mason21151332011-11-10 20:39:08 -05003723 * copy everything in the in-memory inode into the btree.
3724 */
3725noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3726 struct btrfs_root *root, struct inode *inode)
3727{
3728 int ret;
3729
3730 /*
3731 * If the inode is a free space inode, we can deadlock during commit
3732 * if we put it into the delayed code.
3733 *
3734 * The data relocation inode should also be directly updated
3735 * without delay
3736 */
Liu Bo83eea1f2012-07-10 05:28:39 -06003737 if (!btrfs_is_free_space_inode(inode)
Josef Bacik1d52c782014-09-18 11:30:44 -04003738 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3739 && !root->fs_info->log_root_recovering) {
Alexander Block8ea05e32012-07-25 17:35:53 +02003740 btrfs_update_root_times(trans, root);
3741
Chris Mason21151332011-11-10 20:39:08 -05003742 ret = btrfs_delayed_update_inode(trans, root, inode);
3743 if (!ret)
3744 btrfs_set_inode_last_trans(trans, inode);
3745 return ret;
3746 }
3747
3748 return btrfs_update_inode_item(trans, root, inode);
3749}
3750
Josef Bacikbe6aef62012-10-22 15:43:12 -04003751noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3752 struct btrfs_root *root,
3753 struct inode *inode)
Chris Mason21151332011-11-10 20:39:08 -05003754{
3755 int ret;
3756
3757 ret = btrfs_update_inode(trans, root, inode);
3758 if (ret == -ENOSPC)
3759 return btrfs_update_inode_item(trans, root, inode);
3760 return ret;
3761}
3762
3763/*
Chris Masond352ac62008-09-29 15:18:18 -04003764 * unlink helper that gets used here in inode.c and in the tree logging
3765 * recovery code. It remove a link in a directory with a given name, and
3766 * also drops the back refs in the inode to the directory
3767 */
Al Viro92986792011-03-04 17:14:37 +00003768static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3769 struct btrfs_root *root,
3770 struct inode *dir, struct inode *inode,
3771 const char *name, int name_len)
Chris Mason39279cc2007-06-12 06:35:45 -04003772{
3773 struct btrfs_path *path;
Chris Mason39279cc2007-06-12 06:35:45 -04003774 int ret = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04003775 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04003776 struct btrfs_dir_item *di;
Chris Mason5f39d392007-10-15 16:14:19 -04003777 struct btrfs_key key;
Josef Bacikaec74772008-07-24 12:12:38 -04003778 u64 index;
Li Zefan33345d012011-04-20 10:31:50 +08003779 u64 ino = btrfs_ino(inode);
3780 u64 dir_ino = btrfs_ino(dir);
Chris Mason39279cc2007-06-12 06:35:45 -04003781
3782 path = btrfs_alloc_path();
Chris Mason54aa1f42007-06-22 14:16:25 -04003783 if (!path) {
3784 ret = -ENOMEM;
Tsutomu Itoh554233a2011-02-03 03:16:25 +00003785 goto out;
Chris Mason54aa1f42007-06-22 14:16:25 -04003786 }
3787
Chris Masonb9473432009-03-13 11:00:37 -04003788 path->leave_spinning = 1;
Li Zefan33345d012011-04-20 10:31:50 +08003789 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
Chris Mason39279cc2007-06-12 06:35:45 -04003790 name, name_len, -1);
3791 if (IS_ERR(di)) {
3792 ret = PTR_ERR(di);
3793 goto err;
3794 }
3795 if (!di) {
3796 ret = -ENOENT;
3797 goto err;
3798 }
Chris Mason5f39d392007-10-15 16:14:19 -04003799 leaf = path->nodes[0];
3800 btrfs_dir_item_key_to_cpu(leaf, di, &key);
Chris Mason39279cc2007-06-12 06:35:45 -04003801 ret = btrfs_delete_one_dir_name(trans, root, path, di);
Chris Mason54aa1f42007-06-22 14:16:25 -04003802 if (ret)
3803 goto err;
David Sterbab3b4aa72011-04-21 01:20:15 +02003804 btrfs_release_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04003805
Miao Xie67de1172013-12-26 13:07:06 +08003806 /*
3807 * If we don't have dir index, we have to get it by looking up
3808 * the inode ref, since we get the inode ref, remove it directly,
3809 * it is unnecessary to do delayed deletion.
3810 *
3811 * But if we have dir index, needn't search inode ref to get it.
3812 * Since the inode ref is close to the inode item, it is better
3813 * that we delay to delete it, and just do this deletion when
3814 * we update the inode item.
3815 */
3816 if (BTRFS_I(inode)->dir_index) {
3817 ret = btrfs_delayed_delete_inode_ref(inode);
3818 if (!ret) {
3819 index = BTRFS_I(inode)->dir_index;
3820 goto skip_backref;
3821 }
3822 }
3823
Li Zefan33345d012011-04-20 10:31:50 +08003824 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3825 dir_ino, &index);
Josef Bacikaec74772008-07-24 12:12:38 -04003826 if (ret) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00003827 btrfs_info(root->fs_info,
3828 "failed to delete reference to %.*s, inode %llu parent %llu",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02003829 name_len, name, ino, dir_ino);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003830 btrfs_abort_transaction(trans, root, ret);
Josef Bacikaec74772008-07-24 12:12:38 -04003831 goto err;
3832 }
Miao Xie67de1172013-12-26 13:07:06 +08003833skip_backref:
Miao Xie16cdcec2011-04-22 18:12:22 +08003834 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003835 if (ret) {
3836 btrfs_abort_transaction(trans, root, ret);
Chris Mason39279cc2007-06-12 06:35:45 -04003837 goto err;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003838 }
Chris Mason39279cc2007-06-12 06:35:45 -04003839
Chris Masone02119d2008-09-05 16:13:11 -04003840 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
Li Zefan33345d012011-04-20 10:31:50 +08003841 inode, dir_ino);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003842 if (ret != 0 && ret != -ENOENT) {
3843 btrfs_abort_transaction(trans, root, ret);
3844 goto err;
3845 }
Chris Masone02119d2008-09-05 16:13:11 -04003846
3847 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3848 dir, index);
Chris Mason6418c962010-10-30 07:34:24 -04003849 if (ret == -ENOENT)
3850 ret = 0;
Zach Brownd4e39912013-04-02 21:02:16 +00003851 else if (ret)
3852 btrfs_abort_transaction(trans, root, ret);
Chris Mason39279cc2007-06-12 06:35:45 -04003853err:
3854 btrfs_free_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003855 if (ret)
3856 goto out;
3857
3858 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
Josef Bacik0c4d2d92012-04-05 15:03:02 -04003859 inode_inc_iversion(inode);
3860 inode_inc_iversion(dir);
Chris Masone02119d2008-09-05 16:13:11 -04003861 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
Tsutomu Itohb9959292012-06-25 21:25:22 -06003862 ret = btrfs_update_inode(trans, root, dir);
Chris Masone02119d2008-09-05 16:13:11 -04003863out:
Chris Mason39279cc2007-06-12 06:35:45 -04003864 return ret;
3865}
3866
Al Viro92986792011-03-04 17:14:37 +00003867int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3868 struct btrfs_root *root,
3869 struct inode *dir, struct inode *inode,
3870 const char *name, int name_len)
3871{
3872 int ret;
3873 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3874 if (!ret) {
Zach Brown8b558c52013-10-16 12:10:34 -07003875 drop_nlink(inode);
Al Viro92986792011-03-04 17:14:37 +00003876 ret = btrfs_update_inode(trans, root, inode);
3877 }
3878 return ret;
3879}
Yan, Zhenga22285a2010-05-16 10:48:46 -04003880
3881/*
3882 * helper to start transaction for unlink and rmdir.
3883 *
Josef Bacikd52be812013-05-29 14:54:47 -04003884 * unlink and rmdir are special in btrfs, they do not always free space, so
3885 * if we cannot make our reservations the normal way try and see if there is
3886 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3887 * allow the unlink to occur.
Yan, Zhenga22285a2010-05-16 10:48:46 -04003888 */
Josef Bacikd52be812013-05-29 14:54:47 -04003889static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
Yan, Zhenga22285a2010-05-16 10:48:46 -04003890{
3891 struct btrfs_trans_handle *trans;
3892 struct btrfs_root *root = BTRFS_I(dir)->root;
Yan, Zhenga22285a2010-05-16 10:48:46 -04003893 int ret;
3894
Josef Bacike70bea52011-10-11 14:18:24 -04003895 /*
3896 * 1 for the possible orphan item
3897 * 1 for the dir item
3898 * 1 for the dir index
3899 * 1 for the inode ref
Josef Bacike70bea52011-10-11 14:18:24 -04003900 * 1 for the inode
3901 */
Josef Bacik6e137ed2013-03-26 15:26:55 -04003902 trans = btrfs_start_transaction(root, 5);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003903 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3904 return trans;
3905
Josef Bacikd52be812013-05-29 14:54:47 -04003906 if (PTR_ERR(trans) == -ENOSPC) {
3907 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003908
Josef Bacikd52be812013-05-29 14:54:47 -04003909 trans = btrfs_start_transaction(root, 0);
3910 if (IS_ERR(trans))
3911 return trans;
3912 ret = btrfs_cond_migrate_bytes(root->fs_info,
3913 &root->fs_info->trans_block_rsv,
3914 num_bytes, 5);
3915 if (ret) {
3916 btrfs_end_transaction(trans, root);
3917 return ERR_PTR(ret);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003918 }
Josef Bacik5a77d762011-11-01 14:32:23 -04003919 trans->block_rsv = &root->fs_info->trans_block_rsv;
Josef Bacikd52be812013-05-29 14:54:47 -04003920 trans->bytes_reserved = num_bytes;
Yan, Zhenga22285a2010-05-16 10:48:46 -04003921 }
Josef Bacikd52be812013-05-29 14:54:47 -04003922 return trans;
Yan, Zhenga22285a2010-05-16 10:48:46 -04003923}
3924
Chris Mason39279cc2007-06-12 06:35:45 -04003925static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3926{
Yan, Zhenga22285a2010-05-16 10:48:46 -04003927 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04003928 struct btrfs_trans_handle *trans;
Josef Bacik7b128762008-07-24 12:17:14 -04003929 struct inode *inode = dentry->d_inode;
Chris Mason39279cc2007-06-12 06:35:45 -04003930 int ret;
3931
Josef Bacikd52be812013-05-29 14:54:47 -04003932 trans = __unlink_start_trans(dir);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003933 if (IS_ERR(trans))
Josef Bacik5df6a9f2009-11-10 21:23:48 -05003934 return PTR_ERR(trans);
Chris Mason5f39d392007-10-15 16:14:19 -04003935
Chris Mason12fcfd22009-03-24 10:24:20 -04003936 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3937
Chris Masone02119d2008-09-05 16:13:11 -04003938 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3939 dentry->d_name.name, dentry->d_name.len);
Tsutomu Itohb5324022011-07-19 07:27:20 +00003940 if (ret)
3941 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04003942
Yan, Zhenga22285a2010-05-16 10:48:46 -04003943 if (inode->i_nlink == 0) {
Josef Bacik7b128762008-07-24 12:17:14 -04003944 ret = btrfs_orphan_add(trans, inode);
Tsutomu Itohb5324022011-07-19 07:27:20 +00003945 if (ret)
3946 goto out;
Yan, Zhenga22285a2010-05-16 10:48:46 -04003947 }
Josef Bacik7b128762008-07-24 12:17:14 -04003948
Tsutomu Itohb5324022011-07-19 07:27:20 +00003949out:
Josef Bacikd52be812013-05-29 14:54:47 -04003950 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00003951 btrfs_btree_balance_dirty(root);
Chris Mason39279cc2007-06-12 06:35:45 -04003952 return ret;
3953}
3954
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003955int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3956 struct btrfs_root *root,
3957 struct inode *dir, u64 objectid,
3958 const char *name, int name_len)
3959{
3960 struct btrfs_path *path;
3961 struct extent_buffer *leaf;
3962 struct btrfs_dir_item *di;
3963 struct btrfs_key key;
3964 u64 index;
3965 int ret;
Li Zefan33345d012011-04-20 10:31:50 +08003966 u64 dir_ino = btrfs_ino(dir);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003967
3968 path = btrfs_alloc_path();
3969 if (!path)
3970 return -ENOMEM;
3971
Li Zefan33345d012011-04-20 10:31:50 +08003972 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003973 name, name_len, -1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003974 if (IS_ERR_OR_NULL(di)) {
3975 if (!di)
3976 ret = -ENOENT;
3977 else
3978 ret = PTR_ERR(di);
3979 goto out;
3980 }
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003981
3982 leaf = path->nodes[0];
3983 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3984 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3985 ret = btrfs_delete_one_dir_name(trans, root, path, di);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003986 if (ret) {
3987 btrfs_abort_transaction(trans, root, ret);
3988 goto out;
3989 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003990 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003991
3992 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3993 objectid, root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +08003994 dir_ino, &index, name, name_len);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003995 if (ret < 0) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003996 if (ret != -ENOENT) {
3997 btrfs_abort_transaction(trans, root, ret);
3998 goto out;
3999 }
Li Zefan33345d012011-04-20 10:31:50 +08004000 di = btrfs_search_dir_index_item(root, path, dir_ino,
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004001 name, name_len);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004002 if (IS_ERR_OR_NULL(di)) {
4003 if (!di)
4004 ret = -ENOENT;
4005 else
4006 ret = PTR_ERR(di);
4007 btrfs_abort_transaction(trans, root, ret);
4008 goto out;
4009 }
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004010
4011 leaf = path->nodes[0];
4012 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02004013 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004014 index = key.offset;
4015 }
David Sterbab3b4aa72011-04-21 01:20:15 +02004016 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004017
Miao Xie16cdcec2011-04-22 18:12:22 +08004018 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004019 if (ret) {
4020 btrfs_abort_transaction(trans, root, ret);
4021 goto out;
4022 }
Miao Xie16cdcec2011-04-22 18:12:22 +08004023
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004024 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
Josef Bacik0c4d2d92012-04-05 15:03:02 -04004025 inode_inc_iversion(dir);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004026 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
Josef Bacik5a24e842012-08-08 10:12:59 -06004027 ret = btrfs_update_inode_fallback(trans, root, dir);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004028 if (ret)
4029 btrfs_abort_transaction(trans, root, ret);
4030out:
Josef Bacik71d7aed2011-06-14 14:24:32 -04004031 btrfs_free_path(path);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004032 return ret;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004033}
4034
Chris Mason39279cc2007-06-12 06:35:45 -04004035static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4036{
4037 struct inode *inode = dentry->d_inode;
Chris Mason1832a6d2007-12-21 16:27:21 -05004038 int err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004039 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04004040 struct btrfs_trans_handle *trans;
Chris Mason39279cc2007-06-12 06:35:45 -04004041
David Sterbab3ae2442012-09-13 16:04:34 -06004042 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
Yan134d4512007-10-25 15:49:25 -04004043 return -ENOTEMPTY;
David Sterbab3ae2442012-09-13 16:04:34 -06004044 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4045 return -EPERM;
Yan134d4512007-10-25 15:49:25 -04004046
Josef Bacikd52be812013-05-29 14:54:47 -04004047 trans = __unlink_start_trans(dir);
Yan, Zhenga22285a2010-05-16 10:48:46 -04004048 if (IS_ERR(trans))
Josef Bacik5df6a9f2009-11-10 21:23:48 -05004049 return PTR_ERR(trans);
Josef Bacik5df6a9f2009-11-10 21:23:48 -05004050
Li Zefan33345d012011-04-20 10:31:50 +08004051 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004052 err = btrfs_unlink_subvol(trans, root, dir,
4053 BTRFS_I(inode)->location.objectid,
4054 dentry->d_name.name,
4055 dentry->d_name.len);
4056 goto out;
4057 }
4058
Josef Bacik7b128762008-07-24 12:17:14 -04004059 err = btrfs_orphan_add(trans, inode);
4060 if (err)
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004061 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04004062
Chris Mason39279cc2007-06-12 06:35:45 -04004063 /* now the directory is empty */
Chris Masone02119d2008-09-05 16:13:11 -04004064 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
4065 dentry->d_name.name, dentry->d_name.len);
Chris Masond3977122009-01-05 21:25:51 -05004066 if (!err)
Chris Masondbe674a2008-07-17 12:54:05 -04004067 btrfs_i_size_write(inode, 0);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004068out:
Josef Bacikd52be812013-05-29 14:54:47 -04004069 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00004070 btrfs_btree_balance_dirty(root);
Chris Mason39544012007-12-12 14:38:19 -05004071
Chris Mason39279cc2007-06-12 06:35:45 -04004072 return err;
4073}
4074
Chris Mason323ac952008-10-01 19:05:46 -04004075/*
Chris Mason39279cc2007-06-12 06:35:45 -04004076 * this can truncate away extent items, csum items and directory items.
4077 * It starts at a high offset and removes keys until it can't find
Chris Masond352ac62008-09-29 15:18:18 -04004078 * any higher than new_size
Chris Mason39279cc2007-06-12 06:35:45 -04004079 *
4080 * csum items that cross the new i_size are truncated to the new size
4081 * as well.
Josef Bacik7b128762008-07-24 12:17:14 -04004082 *
4083 * min_type is the minimum key type to truncate down to. If set to 0, this
4084 * will kill all the items on this inode, including the INODE_ITEM_KEY.
Chris Mason39279cc2007-06-12 06:35:45 -04004085 */
Yan, Zheng80825102009-11-12 09:35:36 +00004086int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4087 struct btrfs_root *root,
4088 struct inode *inode,
4089 u64 new_size, u32 min_type)
Chris Mason39279cc2007-06-12 06:35:45 -04004090{
Chris Mason39279cc2007-06-12 06:35:45 -04004091 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04004092 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04004093 struct btrfs_file_extent_item *fi;
Yan, Zheng80825102009-11-12 09:35:36 +00004094 struct btrfs_key key;
4095 struct btrfs_key found_key;
Chris Mason39279cc2007-06-12 06:35:45 -04004096 u64 extent_start = 0;
Chris Masondb945352007-10-15 16:15:53 -04004097 u64 extent_num_bytes = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004098 u64 extent_offset = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004099 u64 item_end = 0;
Josef Bacik7f4f6e02013-08-29 16:43:28 -04004100 u64 last_size = (u64)-1;
Yan, Zheng80825102009-11-12 09:35:36 +00004101 u32 found_type = (u8)-1;
Chris Mason39279cc2007-06-12 06:35:45 -04004102 int found_extent;
4103 int del_item;
Chris Mason85e21ba2008-01-29 15:11:36 -05004104 int pending_del_nr = 0;
4105 int pending_del_slot = 0;
Chris Mason179e29e2007-11-01 11:28:41 -04004106 int extent_type = -1;
Yan, Zheng80825102009-11-12 09:35:36 +00004107 int ret;
4108 int err = 0;
Li Zefan33345d012011-04-20 10:31:50 +08004109 u64 ino = btrfs_ino(inode);
Yan, Zheng80825102009-11-12 09:35:36 +00004110
4111 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
Chris Mason39279cc2007-06-12 06:35:45 -04004112
Mark Fasheh0eb0e192011-07-12 16:44:10 -07004113 path = btrfs_alloc_path();
4114 if (!path)
4115 return -ENOMEM;
4116 path->reada = -1;
4117
Josef Bacik5dc562c2012-08-17 13:14:17 -04004118 /*
4119 * We want to drop from the next block forward in case this new size is
4120 * not block aligned since we will be keeping the last block of the
4121 * extent just the way it is.
4122 */
Miao Xie27cdeb72014-04-02 19:51:05 +08004123 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4124 root == root->fs_info->tree_root)
Qu Wenruofda28322013-02-26 08:10:22 +00004125 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4126 root->sectorsize), (u64)-1, 0);
Yan, Zheng80825102009-11-12 09:35:36 +00004127
Miao Xie16cdcec2011-04-22 18:12:22 +08004128 /*
4129 * This function is also used to drop the items in the log tree before
4130 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4131 * it is used to drop the loged items. So we shouldn't kill the delayed
4132 * items.
4133 */
4134 if (min_type == 0 && root == BTRFS_I(inode)->root)
4135 btrfs_kill_delayed_inode_items(inode);
4136
Li Zefan33345d012011-04-20 10:31:50 +08004137 key.objectid = ino;
Chris Mason39279cc2007-06-12 06:35:45 -04004138 key.offset = (u64)-1;
Chris Mason5f39d392007-10-15 16:14:19 -04004139 key.type = (u8)-1;
4140
Chris Mason85e21ba2008-01-29 15:11:36 -05004141search_again:
Chris Masonb9473432009-03-13 11:00:37 -04004142 path->leave_spinning = 1;
Chris Mason85e21ba2008-01-29 15:11:36 -05004143 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
Yan, Zheng80825102009-11-12 09:35:36 +00004144 if (ret < 0) {
4145 err = ret;
4146 goto out;
4147 }
Chris Masond3977122009-01-05 21:25:51 -05004148
Chris Mason85e21ba2008-01-29 15:11:36 -05004149 if (ret > 0) {
Chris Masone02119d2008-09-05 16:13:11 -04004150 /* there are no items in the tree for us to truncate, we're
4151 * done
4152 */
Yan, Zheng80825102009-11-12 09:35:36 +00004153 if (path->slots[0] == 0)
4154 goto out;
Chris Mason85e21ba2008-01-29 15:11:36 -05004155 path->slots[0]--;
4156 }
4157
Chris Masond3977122009-01-05 21:25:51 -05004158 while (1) {
Chris Mason39279cc2007-06-12 06:35:45 -04004159 fi = NULL;
Chris Mason5f39d392007-10-15 16:14:19 -04004160 leaf = path->nodes[0];
4161 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
David Sterba962a2982014-06-04 18:41:45 +02004162 found_type = found_key.type;
Chris Mason39279cc2007-06-12 06:35:45 -04004163
Li Zefan33345d012011-04-20 10:31:50 +08004164 if (found_key.objectid != ino)
Chris Mason39279cc2007-06-12 06:35:45 -04004165 break;
Chris Mason5f39d392007-10-15 16:14:19 -04004166
Chris Mason85e21ba2008-01-29 15:11:36 -05004167 if (found_type < min_type)
Chris Mason39279cc2007-06-12 06:35:45 -04004168 break;
4169
Chris Mason5f39d392007-10-15 16:14:19 -04004170 item_end = found_key.offset;
Chris Mason39279cc2007-06-12 06:35:45 -04004171 if (found_type == BTRFS_EXTENT_DATA_KEY) {
Chris Mason5f39d392007-10-15 16:14:19 -04004172 fi = btrfs_item_ptr(leaf, path->slots[0],
Chris Mason39279cc2007-06-12 06:35:45 -04004173 struct btrfs_file_extent_item);
Chris Mason179e29e2007-11-01 11:28:41 -04004174 extent_type = btrfs_file_extent_type(leaf, fi);
4175 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
Chris Mason5f39d392007-10-15 16:14:19 -04004176 item_end +=
Chris Masondb945352007-10-15 16:15:53 -04004177 btrfs_file_extent_num_bytes(leaf, fi);
Chris Mason179e29e2007-11-01 11:28:41 -04004178 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason179e29e2007-11-01 11:28:41 -04004179 item_end += btrfs_file_extent_inline_len(leaf,
Chris Mason514ac8a2014-01-03 21:07:00 -08004180 path->slots[0], fi);
Chris Mason39279cc2007-06-12 06:35:45 -04004181 }
Yan008630c2007-11-07 13:31:09 -05004182 item_end--;
Chris Mason39279cc2007-06-12 06:35:45 -04004183 }
Yan, Zheng80825102009-11-12 09:35:36 +00004184 if (found_type > min_type) {
Chris Mason39279cc2007-06-12 06:35:45 -04004185 del_item = 1;
Yan, Zheng80825102009-11-12 09:35:36 +00004186 } else {
4187 if (item_end < new_size)
4188 break;
4189 if (found_key.offset >= new_size)
4190 del_item = 1;
4191 else
4192 del_item = 0;
4193 }
Chris Mason39279cc2007-06-12 06:35:45 -04004194 found_extent = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004195 /* FIXME, shrink the extent if the ref count is only 1 */
Chris Mason179e29e2007-11-01 11:28:41 -04004196 if (found_type != BTRFS_EXTENT_DATA_KEY)
4197 goto delete;
4198
Josef Bacik7f4f6e02013-08-29 16:43:28 -04004199 if (del_item)
4200 last_size = found_key.offset;
4201 else
4202 last_size = new_size;
4203
Chris Mason179e29e2007-11-01 11:28:41 -04004204 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
Chris Mason39279cc2007-06-12 06:35:45 -04004205 u64 num_dec;
Chris Masondb945352007-10-15 16:15:53 -04004206 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
Miao Xief70a9a6b2012-01-12 19:10:12 -05004207 if (!del_item) {
Chris Masondb945352007-10-15 16:15:53 -04004208 u64 orig_num_bytes =
4209 btrfs_file_extent_num_bytes(leaf, fi);
Qu Wenruofda28322013-02-26 08:10:22 +00004210 extent_num_bytes = ALIGN(new_size -
4211 found_key.offset,
4212 root->sectorsize);
Chris Masondb945352007-10-15 16:15:53 -04004213 btrfs_set_file_extent_num_bytes(leaf, fi,
4214 extent_num_bytes);
4215 num_dec = (orig_num_bytes -
Chris Mason90692182008-02-08 13:49:28 -05004216 extent_num_bytes);
Miao Xie27cdeb72014-04-02 19:51:05 +08004217 if (test_bit(BTRFS_ROOT_REF_COWS,
4218 &root->state) &&
4219 extent_start != 0)
Yan Zhenga76a3cd2008-10-09 11:46:29 -04004220 inode_sub_bytes(inode, num_dec);
Chris Mason5f39d392007-10-15 16:14:19 -04004221 btrfs_mark_buffer_dirty(leaf);
Chris Mason39279cc2007-06-12 06:35:45 -04004222 } else {
Chris Masondb945352007-10-15 16:15:53 -04004223 extent_num_bytes =
4224 btrfs_file_extent_disk_num_bytes(leaf,
4225 fi);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004226 extent_offset = found_key.offset -
4227 btrfs_file_extent_offset(leaf, fi);
4228
Chris Mason39279cc2007-06-12 06:35:45 -04004229 /* FIXME blocksize != 4096 */
Chris Mason90692182008-02-08 13:49:28 -05004230 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
Chris Mason39279cc2007-06-12 06:35:45 -04004231 if (extent_start != 0) {
4232 found_extent = 1;
Miao Xie27cdeb72014-04-02 19:51:05 +08004233 if (test_bit(BTRFS_ROOT_REF_COWS,
4234 &root->state))
Yan Zhenga76a3cd2008-10-09 11:46:29 -04004235 inode_sub_bytes(inode, num_dec);
Chris Mason39279cc2007-06-12 06:35:45 -04004236 }
4237 }
Chris Mason90692182008-02-08 13:49:28 -05004238 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Masonc8b97812008-10-29 14:49:59 -04004239 /*
4240 * we can't truncate inline items that have had
4241 * special encodings
4242 */
4243 if (!del_item &&
4244 btrfs_file_extent_compression(leaf, fi) == 0 &&
4245 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4246 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
Chris Masone02119d2008-09-05 16:13:11 -04004247 u32 size = new_size - found_key.offset;
4248
Miao Xie27cdeb72014-04-02 19:51:05 +08004249 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
Yan Zhenga76a3cd2008-10-09 11:46:29 -04004250 inode_sub_bytes(inode, item_end + 1 -
4251 new_size);
Chris Mason514ac8a2014-01-03 21:07:00 -08004252
4253 /*
4254 * update the ram bytes to properly reflect
4255 * the new size of our item
4256 */
4257 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
Chris Masone02119d2008-09-05 16:13:11 -04004258 size =
4259 btrfs_file_extent_calc_inline_size(size);
Tsutomu Itohafe5fea2013-04-16 05:18:22 +00004260 btrfs_truncate_item(root, path, size, 1);
Miao Xie27cdeb72014-04-02 19:51:05 +08004261 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4262 &root->state)) {
Yan Zhenga76a3cd2008-10-09 11:46:29 -04004263 inode_sub_bytes(inode, item_end + 1 -
4264 found_key.offset);
Chris Mason90692182008-02-08 13:49:28 -05004265 }
Chris Mason39279cc2007-06-12 06:35:45 -04004266 }
Chris Mason179e29e2007-11-01 11:28:41 -04004267delete:
Chris Mason39279cc2007-06-12 06:35:45 -04004268 if (del_item) {
Chris Mason85e21ba2008-01-29 15:11:36 -05004269 if (!pending_del_nr) {
4270 /* no pending yet, add ourselves */
4271 pending_del_slot = path->slots[0];
4272 pending_del_nr = 1;
4273 } else if (pending_del_nr &&
4274 path->slots[0] + 1 == pending_del_slot) {
4275 /* hop on the pending chunk */
4276 pending_del_nr++;
4277 pending_del_slot = path->slots[0];
4278 } else {
Chris Masond3977122009-01-05 21:25:51 -05004279 BUG();
Chris Mason85e21ba2008-01-29 15:11:36 -05004280 }
Chris Mason39279cc2007-06-12 06:35:45 -04004281 } else {
4282 break;
4283 }
Miao Xie27cdeb72014-04-02 19:51:05 +08004284 if (found_extent &&
4285 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4286 root == root->fs_info->tree_root)) {
Chris Masonb9473432009-03-13 11:00:37 -04004287 btrfs_set_path_blocking(path);
Chris Mason39279cc2007-06-12 06:35:45 -04004288 ret = btrfs_free_extent(trans, root, extent_start,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004289 extent_num_bytes, 0,
4290 btrfs_header_owner(leaf),
Arne Jansen66d7e7f2011-09-12 15:26:38 +02004291 ino, extent_offset, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04004292 BUG_ON(ret);
4293 }
Chris Mason85e21ba2008-01-29 15:11:36 -05004294
Yan, Zheng80825102009-11-12 09:35:36 +00004295 if (found_type == BTRFS_INODE_ITEM_KEY)
4296 break;
4297
4298 if (path->slots[0] == 0 ||
4299 path->slots[0] != pending_del_slot) {
Yan, Zheng80825102009-11-12 09:35:36 +00004300 if (pending_del_nr) {
4301 ret = btrfs_del_items(trans, root, path,
4302 pending_del_slot,
4303 pending_del_nr);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004304 if (ret) {
4305 btrfs_abort_transaction(trans,
4306 root, ret);
4307 goto error;
4308 }
Yan, Zheng80825102009-11-12 09:35:36 +00004309 pending_del_nr = 0;
4310 }
David Sterbab3b4aa72011-04-21 01:20:15 +02004311 btrfs_release_path(path);
Chris Mason85e21ba2008-01-29 15:11:36 -05004312 goto search_again;
Yan, Zheng80825102009-11-12 09:35:36 +00004313 } else {
4314 path->slots[0]--;
Chris Mason85e21ba2008-01-29 15:11:36 -05004315 }
Chris Mason39279cc2007-06-12 06:35:45 -04004316 }
Yan, Zheng80825102009-11-12 09:35:36 +00004317out:
Chris Mason85e21ba2008-01-29 15:11:36 -05004318 if (pending_del_nr) {
4319 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4320 pending_del_nr);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004321 if (ret)
4322 btrfs_abort_transaction(trans, root, ret);
Chris Mason85e21ba2008-01-29 15:11:36 -05004323 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004324error:
Filipe Mananadac57052014-08-29 20:54:26 +01004325 if (last_size != (u64)-1 &&
4326 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
Josef Bacik7f4f6e02013-08-29 16:43:28 -04004327 btrfs_ordered_update_i_size(inode, last_size, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04004328 btrfs_free_path(path);
Yan, Zheng80825102009-11-12 09:35:36 +00004329 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04004330}
4331
Chris Masona52d9a82007-08-27 16:49:44 -04004332/*
Josef Bacik2aaa6652012-08-29 14:27:18 -04004333 * btrfs_truncate_page - read, zero a chunk and write a page
4334 * @inode - inode that we're zeroing
4335 * @from - the offset to start zeroing
4336 * @len - the length to zero, 0 to zero the entire range respective to the
4337 * offset
4338 * @front - zero up to the offset instead of from the offset on
4339 *
4340 * This will find the page for the "from" offset and cow the page and zero the
4341 * part we want to zero. This is used with truncate and hole punching.
Chris Masona52d9a82007-08-27 16:49:44 -04004342 */
Josef Bacik2aaa6652012-08-29 14:27:18 -04004343int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4344 int front)
Chris Masona52d9a82007-08-27 16:49:44 -04004345{
Josef Bacik2aaa6652012-08-29 14:27:18 -04004346 struct address_space *mapping = inode->i_mapping;
Chris Masondb945352007-10-15 16:15:53 -04004347 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004348 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4349 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00004350 struct extent_state *cached_state = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004351 char *kaddr;
Chris Masondb945352007-10-15 16:15:53 -04004352 u32 blocksize = root->sectorsize;
Chris Masona52d9a82007-08-27 16:49:44 -04004353 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4354 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4355 struct page *page;
Josef Bacik3b16a4e2011-09-21 15:05:58 -04004356 gfp_t mask = btrfs_alloc_write_mask(mapping);
Chris Masona52d9a82007-08-27 16:49:44 -04004357 int ret = 0;
4358 u64 page_start;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004359 u64 page_end;
Chris Masona52d9a82007-08-27 16:49:44 -04004360
Josef Bacik2aaa6652012-08-29 14:27:18 -04004361 if ((offset & (blocksize - 1)) == 0 &&
4362 (!len || ((len & (blocksize - 1)) == 0)))
Chris Masona52d9a82007-08-27 16:49:44 -04004363 goto out;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004364 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
Josef Bacik5d5e1032009-10-13 16:46:49 -04004365 if (ret)
4366 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04004367
Chris Mason211c17f2008-05-15 09:13:45 -04004368again:
Josef Bacik3b16a4e2011-09-21 15:05:58 -04004369 page = find_or_create_page(mapping, index, mask);
Josef Bacik5d5e1032009-10-13 16:46:49 -04004370 if (!page) {
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004371 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
Miao Xieac6a2b32012-12-05 10:56:13 +00004372 ret = -ENOMEM;
Chris Masona52d9a82007-08-27 16:49:44 -04004373 goto out;
Josef Bacik5d5e1032009-10-13 16:46:49 -04004374 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04004375
4376 page_start = page_offset(page);
4377 page_end = page_start + PAGE_CACHE_SIZE - 1;
4378
Chris Masona52d9a82007-08-27 16:49:44 -04004379 if (!PageUptodate(page)) {
4380 ret = btrfs_readpage(NULL, page);
4381 lock_page(page);
Chris Mason211c17f2008-05-15 09:13:45 -04004382 if (page->mapping != mapping) {
4383 unlock_page(page);
4384 page_cache_release(page);
4385 goto again;
4386 }
Chris Masona52d9a82007-08-27 16:49:44 -04004387 if (!PageUptodate(page)) {
4388 ret = -EIO;
Chris Mason89642222008-07-24 09:41:53 -04004389 goto out_unlock;
Chris Masona52d9a82007-08-27 16:49:44 -04004390 }
4391 }
Chris Mason211c17f2008-05-15 09:13:45 -04004392 wait_on_page_writeback(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004393
Jeff Mahoneyd0082372012-03-01 14:57:19 +01004394 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004395 set_page_extent_mapped(page);
4396
4397 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4398 if (ordered) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00004399 unlock_extent_cached(io_tree, page_start, page_end,
4400 &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004401 unlock_page(page);
4402 page_cache_release(page);
Chris Masoneb84ae02008-07-17 13:53:27 -04004403 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004404 btrfs_put_ordered_extent(ordered);
4405 goto again;
4406 }
4407
Josef Bacik2ac55d42010-02-03 19:33:23 +00004408 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
Liu Bo9e8a4a82012-09-05 19:10:51 -06004409 EXTENT_DIRTY | EXTENT_DELALLOC |
4410 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
Josef Bacik2ac55d42010-02-03 19:33:23 +00004411 0, 0, &cached_state, GFP_NOFS);
Josef Bacik5d5e1032009-10-13 16:46:49 -04004412
Josef Bacik2ac55d42010-02-03 19:33:23 +00004413 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4414 &cached_state);
Josef Bacik9ed74f22009-09-11 16:12:44 -04004415 if (ret) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00004416 unlock_extent_cached(io_tree, page_start, page_end,
4417 &cached_state, GFP_NOFS);
Josef Bacik9ed74f22009-09-11 16:12:44 -04004418 goto out_unlock;
4419 }
4420
Chris Masone6dcd2d2008-07-17 12:53:50 -04004421 if (offset != PAGE_CACHE_SIZE) {
Josef Bacik2aaa6652012-08-29 14:27:18 -04004422 if (!len)
4423 len = PAGE_CACHE_SIZE - offset;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004424 kaddr = kmap(page);
Josef Bacik2aaa6652012-08-29 14:27:18 -04004425 if (front)
4426 memset(kaddr, 0, offset);
4427 else
4428 memset(kaddr + offset, 0, len);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004429 flush_dcache_page(page);
4430 kunmap(page);
4431 }
Chris Mason247e7432008-07-17 12:53:51 -04004432 ClearPageChecked(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004433 set_page_dirty(page);
Josef Bacik2ac55d42010-02-03 19:33:23 +00004434 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4435 GFP_NOFS);
Chris Mason39279cc2007-06-12 06:35:45 -04004436
Chris Mason89642222008-07-24 09:41:53 -04004437out_unlock:
Josef Bacik5d5e1032009-10-13 16:46:49 -04004438 if (ret)
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004439 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
Chris Mason39279cc2007-06-12 06:35:45 -04004440 unlock_page(page);
4441 page_cache_release(page);
4442out:
4443 return ret;
4444}
4445
Josef Bacik16e75492013-10-22 12:18:51 -04004446static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4447 u64 offset, u64 len)
4448{
4449 struct btrfs_trans_handle *trans;
4450 int ret;
4451
4452 /*
4453 * Still need to make sure the inode looks like it's been updated so
4454 * that any holes get logged if we fsync.
4455 */
4456 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4457 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4458 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4459 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4460 return 0;
4461 }
4462
4463 /*
4464 * 1 - for the one we're dropping
4465 * 1 - for the one we're adding
4466 * 1 - for updating the inode.
4467 */
4468 trans = btrfs_start_transaction(root, 3);
4469 if (IS_ERR(trans))
4470 return PTR_ERR(trans);
4471
4472 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4473 if (ret) {
4474 btrfs_abort_transaction(trans, root, ret);
4475 btrfs_end_transaction(trans, root);
4476 return ret;
4477 }
4478
4479 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4480 0, 0, len, 0, len, 0, 0, 0);
4481 if (ret)
4482 btrfs_abort_transaction(trans, root, ret);
4483 else
4484 btrfs_update_inode(trans, root, inode);
4485 btrfs_end_transaction(trans, root);
4486 return ret;
4487}
4488
Josef Bacik695a0d02011-03-04 15:46:53 -05004489/*
4490 * This function puts in dummy file extents for the area we're creating a hole
4491 * for. So if we are truncating this file to a larger size we need to insert
4492 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4493 * the range between oldsize and size
4494 */
Josef Bacika41ad392011-01-31 15:30:16 -05004495int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
Yan Zheng9036c102008-10-30 14:19:41 -04004496{
Yan Zheng9036c102008-10-30 14:19:41 -04004497 struct btrfs_root *root = BTRFS_I(inode)->root;
4498 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Yan, Zhenga22285a2010-05-16 10:48:46 -04004499 struct extent_map *em = NULL;
Josef Bacik2ac55d42010-02-03 19:33:23 +00004500 struct extent_state *cached_state = NULL;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004501 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Qu Wenruofda28322013-02-26 08:10:22 +00004502 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4503 u64 block_end = ALIGN(size, root->sectorsize);
Yan Zheng9036c102008-10-30 14:19:41 -04004504 u64 last_byte;
4505 u64 cur_offset;
4506 u64 hole_size;
Josef Bacik9ed74f22009-09-11 16:12:44 -04004507 int err = 0;
Yan Zheng9036c102008-10-30 14:19:41 -04004508
Josef Bacika71754f2013-06-17 17:14:39 -04004509 /*
4510 * If our size started in the middle of a page we need to zero out the
4511 * rest of the page before we expand the i_size, otherwise we could
4512 * expose stale data.
4513 */
4514 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4515 if (err)
4516 return err;
4517
Yan Zheng9036c102008-10-30 14:19:41 -04004518 if (size <= hole_start)
4519 return 0;
4520
Yan Zheng9036c102008-10-30 14:19:41 -04004521 while (1) {
4522 struct btrfs_ordered_extent *ordered;
Miao Xiefa7c1492013-09-26 13:15:27 +08004523
Josef Bacik2ac55d42010-02-03 19:33:23 +00004524 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01004525 &cached_state);
Miao Xiefa7c1492013-09-26 13:15:27 +08004526 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4527 block_end - hole_start);
Yan Zheng9036c102008-10-30 14:19:41 -04004528 if (!ordered)
4529 break;
Josef Bacik2ac55d42010-02-03 19:33:23 +00004530 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4531 &cached_state, GFP_NOFS);
Miao Xiefa7c1492013-09-26 13:15:27 +08004532 btrfs_start_ordered_extent(inode, ordered, 1);
Yan Zheng9036c102008-10-30 14:19:41 -04004533 btrfs_put_ordered_extent(ordered);
4534 }
4535
Yan Zheng9036c102008-10-30 14:19:41 -04004536 cur_offset = hole_start;
4537 while (1) {
4538 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4539 block_end - cur_offset, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004540 if (IS_ERR(em)) {
4541 err = PTR_ERR(em);
Zach Brownf2767952013-01-08 19:37:58 +00004542 em = NULL;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004543 break;
4544 }
Yan Zheng9036c102008-10-30 14:19:41 -04004545 last_byte = min(extent_map_end(em), block_end);
Qu Wenruofda28322013-02-26 08:10:22 +00004546 last_byte = ALIGN(last_byte , root->sectorsize);
Yan, Zheng80825102009-11-12 09:35:36 +00004547 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
Josef Bacik5dc562c2012-08-17 13:14:17 -04004548 struct extent_map *hole_em;
Yan Zheng9036c102008-10-30 14:19:41 -04004549 hole_size = last_byte - cur_offset;
Yan, Zheng80825102009-11-12 09:35:36 +00004550
Josef Bacik16e75492013-10-22 12:18:51 -04004551 err = maybe_insert_hole(root, inode, cur_offset,
4552 hole_size);
4553 if (err)
Chris Mason771ed682008-11-06 22:02:51 -05004554 break;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004555 btrfs_drop_extent_cache(inode, cur_offset,
4556 cur_offset + hole_size - 1, 0);
4557 hole_em = alloc_extent_map();
4558 if (!hole_em) {
4559 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4560 &BTRFS_I(inode)->runtime_flags);
4561 goto next;
4562 }
4563 hole_em->start = cur_offset;
4564 hole_em->len = hole_size;
4565 hole_em->orig_start = cur_offset;
Yan, Zheng80825102009-11-12 09:35:36 +00004566
Josef Bacik5dc562c2012-08-17 13:14:17 -04004567 hole_em->block_start = EXTENT_MAP_HOLE;
4568 hole_em->block_len = 0;
Josef Bacikb4939682012-12-03 10:31:19 -05004569 hole_em->orig_block_len = 0;
Josef Bacikcc95bef2013-04-04 14:31:27 -04004570 hole_em->ram_bytes = hole_size;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004571 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4572 hole_em->compress_type = BTRFS_COMPRESS_NONE;
Josef Bacik16e75492013-10-22 12:18:51 -04004573 hole_em->generation = root->fs_info->generation;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004574
4575 while (1) {
4576 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -04004577 err = add_extent_mapping(em_tree, hole_em, 1);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004578 write_unlock(&em_tree->lock);
4579 if (err != -EEXIST)
4580 break;
4581 btrfs_drop_extent_cache(inode, cur_offset,
4582 cur_offset +
4583 hole_size - 1, 0);
4584 }
4585 free_extent_map(hole_em);
Yan Zheng9036c102008-10-30 14:19:41 -04004586 }
Josef Bacik16e75492013-10-22 12:18:51 -04004587next:
Yan Zheng9036c102008-10-30 14:19:41 -04004588 free_extent_map(em);
Yan, Zhenga22285a2010-05-16 10:48:46 -04004589 em = NULL;
Yan Zheng9036c102008-10-30 14:19:41 -04004590 cur_offset = last_byte;
Yan, Zheng80825102009-11-12 09:35:36 +00004591 if (cur_offset >= block_end)
Yan Zheng9036c102008-10-30 14:19:41 -04004592 break;
4593 }
Yan, Zhenga22285a2010-05-16 10:48:46 -04004594 free_extent_map(em);
Josef Bacik2ac55d42010-02-03 19:33:23 +00004595 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4596 GFP_NOFS);
Yan Zheng9036c102008-10-30 14:19:41 -04004597 return err;
4598}
4599
Filipe Manana9ea24bb2014-10-29 11:57:59 +00004600static int wait_snapshoting_atomic_t(atomic_t *a)
4601{
4602 schedule();
4603 return 0;
4604}
4605
4606static void wait_for_snapshot_creation(struct btrfs_root *root)
4607{
4608 while (true) {
4609 int ret;
4610
4611 ret = btrfs_start_write_no_snapshoting(root);
4612 if (ret)
4613 break;
4614 wait_on_atomic_t(&root->will_be_snapshoted,
4615 wait_snapshoting_atomic_t,
4616 TASK_UNINTERRUPTIBLE);
4617 }
4618}
4619
Eric Sandeen3972f262013-01-12 02:57:22 +00004620static int btrfs_setsize(struct inode *inode, struct iattr *attr)
Yan, Zheng80825102009-11-12 09:35:36 +00004621{
Miao Xief4a2f4c2011-12-14 20:12:01 -05004622 struct btrfs_root *root = BTRFS_I(inode)->root;
4623 struct btrfs_trans_handle *trans;
Josef Bacika41ad392011-01-31 15:30:16 -05004624 loff_t oldsize = i_size_read(inode);
Eric Sandeen3972f262013-01-12 02:57:22 +00004625 loff_t newsize = attr->ia_size;
4626 int mask = attr->ia_valid;
Yan, Zheng80825102009-11-12 09:35:36 +00004627 int ret;
4628
Eric Sandeen3972f262013-01-12 02:57:22 +00004629 /*
4630 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4631 * special case where we need to update the times despite not having
4632 * these flags set. For all other operations the VFS set these flags
4633 * explicitly if it wants a timestamp update.
4634 */
Christoph Hellwigdff6efc2013-11-19 07:17:07 -08004635 if (newsize != oldsize) {
4636 inode_inc_iversion(inode);
4637 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4638 inode->i_ctime = inode->i_mtime =
4639 current_fs_time(inode->i_sb);
4640 }
Eric Sandeen3972f262013-01-12 02:57:22 +00004641
Josef Bacika41ad392011-01-31 15:30:16 -05004642 if (newsize > oldsize) {
Kirill A. Shutemov7caef262013-09-12 15:13:56 -07004643 truncate_pagecache(inode, newsize);
Filipe Manana9ea24bb2014-10-29 11:57:59 +00004644 /*
4645 * Don't do an expanding truncate while snapshoting is ongoing.
4646 * This is to ensure the snapshot captures a fully consistent
4647 * state of this file - if the snapshot captures this expanding
4648 * truncation, it must capture all writes that happened before
4649 * this truncation.
4650 */
4651 wait_for_snapshot_creation(root);
Josef Bacika41ad392011-01-31 15:30:16 -05004652 ret = btrfs_cont_expand(inode, oldsize, newsize);
Filipe Manana9ea24bb2014-10-29 11:57:59 +00004653 if (ret) {
4654 btrfs_end_write_no_snapshoting(root);
Yan, Zheng80825102009-11-12 09:35:36 +00004655 return ret;
Filipe Manana9ea24bb2014-10-29 11:57:59 +00004656 }
Yan, Zheng80825102009-11-12 09:35:36 +00004657
Miao Xief4a2f4c2011-12-14 20:12:01 -05004658 trans = btrfs_start_transaction(root, 1);
Filipe Manana9ea24bb2014-10-29 11:57:59 +00004659 if (IS_ERR(trans)) {
4660 btrfs_end_write_no_snapshoting(root);
Miao Xief4a2f4c2011-12-14 20:12:01 -05004661 return PTR_ERR(trans);
Filipe Manana9ea24bb2014-10-29 11:57:59 +00004662 }
Miao Xief4a2f4c2011-12-14 20:12:01 -05004663
4664 i_size_write(inode, newsize);
4665 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4666 ret = btrfs_update_inode(trans, root, inode);
Filipe Manana9ea24bb2014-10-29 11:57:59 +00004667 btrfs_end_write_no_snapshoting(root);
Josef Bacik7ad85bb2012-01-12 19:10:12 -05004668 btrfs_end_transaction(trans, root);
Josef Bacika41ad392011-01-31 15:30:16 -05004669 } else {
Yan, Zheng80825102009-11-12 09:35:36 +00004670
Josef Bacika41ad392011-01-31 15:30:16 -05004671 /*
4672 * We're truncating a file that used to have good data down to
4673 * zero. Make sure it gets into the ordered flush list so that
4674 * any new writes get down to disk quickly.
4675 */
4676 if (newsize == 0)
Josef Bacik72ac3c02012-05-23 14:13:11 -04004677 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4678 &BTRFS_I(inode)->runtime_flags);
Yan, Zheng80825102009-11-12 09:35:36 +00004679
Josef Bacikf3fe8202013-01-07 17:03:21 -05004680 /*
4681 * 1 for the orphan item we're going to add
4682 * 1 for the orphan item deletion.
4683 */
4684 trans = btrfs_start_transaction(root, 2);
4685 if (IS_ERR(trans))
4686 return PTR_ERR(trans);
4687
4688 /*
4689 * We need to do this in case we fail at _any_ point during the
4690 * actual truncate. Once we do the truncate_setsize we could
4691 * invalidate pages which forces any outstanding ordered io to
4692 * be instantly completed which will give us extents that need
4693 * to be truncated. If we fail to get an orphan inode down we
4694 * could have left over extents that were never meant to live,
4695 * so we need to garuntee from this point on that everything
4696 * will be consistent.
4697 */
4698 ret = btrfs_orphan_add(trans, inode);
4699 btrfs_end_transaction(trans, root);
4700 if (ret)
4701 return ret;
4702
Josef Bacika41ad392011-01-31 15:30:16 -05004703 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4704 truncate_setsize(inode, newsize);
Miao Xie2e60a512013-02-08 07:01:08 +00004705
4706 /* Disable nonlocked read DIO to avoid the end less truncate */
4707 btrfs_inode_block_unlocked_dio(inode);
4708 inode_dio_wait(inode);
4709 btrfs_inode_resume_unlocked_dio(inode);
4710
Josef Bacika41ad392011-01-31 15:30:16 -05004711 ret = btrfs_truncate(inode);
Josef Bacik7f4f6e02013-08-29 16:43:28 -04004712 if (ret && inode->i_nlink) {
4713 int err;
4714
4715 /*
4716 * failed to truncate, disk_i_size is only adjusted down
4717 * as we remove extents, so it should represent the true
4718 * size of the inode, so reset the in memory size and
4719 * delete our orphan entry.
4720 */
4721 trans = btrfs_join_transaction(root);
4722 if (IS_ERR(trans)) {
4723 btrfs_orphan_del(NULL, inode);
4724 return ret;
4725 }
4726 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4727 err = btrfs_orphan_del(trans, inode);
4728 if (err)
4729 btrfs_abort_transaction(trans, root, err);
4730 btrfs_end_transaction(trans, root);
4731 }
Yan, Zheng80825102009-11-12 09:35:36 +00004732 }
4733
Josef Bacika41ad392011-01-31 15:30:16 -05004734 return ret;
Yan, Zheng80825102009-11-12 09:35:36 +00004735}
4736
Chris Mason39279cc2007-06-12 06:35:45 -04004737static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4738{
4739 struct inode *inode = dentry->d_inode;
Li Zefanb83cc962010-12-20 16:04:08 +08004740 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04004741 int err;
4742
Li Zefanb83cc962010-12-20 16:04:08 +08004743 if (btrfs_root_readonly(root))
4744 return -EROFS;
4745
Chris Mason39279cc2007-06-12 06:35:45 -04004746 err = inode_change_ok(inode, attr);
4747 if (err)
4748 return err;
4749
Chris Mason5a3f23d2009-03-31 13:27:11 -04004750 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
Eric Sandeen3972f262013-01-12 02:57:22 +00004751 err = btrfs_setsize(inode, attr);
Yan, Zheng80825102009-11-12 09:35:36 +00004752 if (err)
4753 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04004754 }
Yan Zheng9036c102008-10-30 14:19:41 -04004755
Christoph Hellwig10257742010-06-04 11:30:02 +02004756 if (attr->ia_valid) {
4757 setattr_copy(inode, attr);
Josef Bacik0c4d2d92012-04-05 15:03:02 -04004758 inode_inc_iversion(inode);
Josef Bacik22c44fe2011-11-30 10:45:38 -05004759 err = btrfs_dirty_inode(inode);
Josef Bacik33268ea2008-07-24 12:16:36 -04004760
Josef Bacik22c44fe2011-11-30 10:45:38 -05004761 if (!err && attr->ia_valid & ATTR_MODE)
Christoph Hellwig996a7102013-12-20 05:16:43 -08004762 err = posix_acl_chmod(inode, inode->i_mode);
Christoph Hellwig10257742010-06-04 11:30:02 +02004763 }
4764
Chris Mason39279cc2007-06-12 06:35:45 -04004765 return err;
4766}
Chris Mason61295eb2008-01-14 16:24:38 -05004767
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00004768/*
4769 * While truncating the inode pages during eviction, we get the VFS calling
4770 * btrfs_invalidatepage() against each page of the inode. This is slow because
4771 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4772 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4773 * extent_state structures over and over, wasting lots of time.
4774 *
4775 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4776 * those expensive operations on a per page basis and do only the ordered io
4777 * finishing, while we release here the extent_map and extent_state structures,
4778 * without the excessive merging and splitting.
4779 */
4780static void evict_inode_truncate_pages(struct inode *inode)
4781{
4782 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4783 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4784 struct rb_node *node;
4785
4786 ASSERT(inode->i_state & I_FREEING);
Johannes Weiner91b0abe2014-04-03 14:47:49 -07004787 truncate_inode_pages_final(&inode->i_data);
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00004788
4789 write_lock(&map_tree->lock);
4790 while (!RB_EMPTY_ROOT(&map_tree->map)) {
4791 struct extent_map *em;
4792
4793 node = rb_first(&map_tree->map);
4794 em = rb_entry(node, struct extent_map, rb_node);
Wang Shilong180589e2013-12-14 15:27:31 +08004795 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4796 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00004797 remove_extent_mapping(map_tree, em);
4798 free_extent_map(em);
Filipe Manana7064dd52014-08-08 02:47:05 +01004799 if (need_resched()) {
4800 write_unlock(&map_tree->lock);
4801 cond_resched();
4802 write_lock(&map_tree->lock);
4803 }
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00004804 }
4805 write_unlock(&map_tree->lock);
4806
4807 spin_lock(&io_tree->lock);
4808 while (!RB_EMPTY_ROOT(&io_tree->state)) {
4809 struct extent_state *state;
4810 struct extent_state *cached_state = NULL;
4811
4812 node = rb_first(&io_tree->state);
4813 state = rb_entry(node, struct extent_state, rb_node);
4814 atomic_inc(&state->refs);
4815 spin_unlock(&io_tree->lock);
4816
4817 lock_extent_bits(io_tree, state->start, state->end,
4818 0, &cached_state);
4819 clear_extent_bit(io_tree, state->start, state->end,
4820 EXTENT_LOCKED | EXTENT_DIRTY |
4821 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
4822 EXTENT_DEFRAG, 1, 1,
4823 &cached_state, GFP_NOFS);
4824 free_extent_state(state);
4825
Filipe Manana7064dd52014-08-08 02:47:05 +01004826 cond_resched();
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00004827 spin_lock(&io_tree->lock);
4828 }
4829 spin_unlock(&io_tree->lock);
4830}
4831
Al Virobd555972010-06-07 11:35:40 -04004832void btrfs_evict_inode(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04004833{
4834 struct btrfs_trans_handle *trans;
4835 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik726c35f2011-09-26 15:46:06 -04004836 struct btrfs_block_rsv *rsv, *global_rsv;
Josef Bacik07127182011-08-19 10:29:59 -04004837 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
Chris Mason39279cc2007-06-12 06:35:45 -04004838 int ret;
4839
liubo1abe9b82011-03-24 11:18:59 +00004840 trace_btrfs_inode_evict(inode);
4841
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00004842 evict_inode_truncate_pages(inode);
4843
Stefan Behrens69e9c6c2013-09-05 16:58:43 +02004844 if (inode->i_nlink &&
4845 ((btrfs_root_refs(&root->root_item) != 0 &&
4846 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4847 btrfs_is_free_space_inode(inode)))
Al Virobd555972010-06-07 11:35:40 -04004848 goto no_delete;
4849
Chris Mason39279cc2007-06-12 06:35:45 -04004850 if (is_bad_inode(inode)) {
Josef Bacik7b128762008-07-24 12:17:14 -04004851 btrfs_orphan_del(NULL, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04004852 goto no_delete;
4853 }
Al Virobd555972010-06-07 11:35:40 -04004854 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
Chris Mason4a096752008-07-21 10:29:44 -04004855 btrfs_wait_ordered_range(inode, 0, (u64)-1);
Chris Mason5f39d392007-10-15 16:14:19 -04004856
Miao Xief6124962014-09-12 18:44:04 +08004857 btrfs_free_io_failure_record(inode, 0, (u64)-1);
4858
Yan, Zhengc71bf092009-11-12 09:34:40 +00004859 if (root->fs_info->log_root_recovering) {
Liu Bo6bf02312012-06-25 21:59:09 -06004860 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
Josef Bacik8a35d952012-05-23 14:26:42 -04004861 &BTRFS_I(inode)->runtime_flags));
Yan, Zhengc71bf092009-11-12 09:34:40 +00004862 goto no_delete;
4863 }
4864
Yan, Zheng76dda932009-09-21 16:00:26 -04004865 if (inode->i_nlink > 0) {
Stefan Behrens69e9c6c2013-09-05 16:58:43 +02004866 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4867 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
Yan, Zheng76dda932009-09-21 16:00:26 -04004868 goto no_delete;
4869 }
4870
Miao Xie0e8c36a2012-12-19 06:59:51 +00004871 ret = btrfs_commit_inode_delayed_inode(inode);
4872 if (ret) {
4873 btrfs_orphan_del(NULL, inode);
4874 goto no_delete;
4875 }
4876
Miao Xie66d8f3d2012-09-06 04:02:28 -06004877 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
Josef Bacik4289a662011-08-05 13:22:24 -04004878 if (!rsv) {
4879 btrfs_orphan_del(NULL, inode);
4880 goto no_delete;
4881 }
Josef Bacik4a338542011-08-29 11:01:31 -04004882 rsv->size = min_size;
Josef Bacikca7e70f2012-08-27 17:48:15 -04004883 rsv->failfast = 1;
Josef Bacik726c35f2011-09-26 15:46:06 -04004884 global_rsv = &root->fs_info->global_block_rsv;
Josef Bacik4289a662011-08-05 13:22:24 -04004885
Chris Masondbe674a2008-07-17 12:54:05 -04004886 btrfs_i_size_write(inode, 0);
Chris Mason5f39d392007-10-15 16:14:19 -04004887
Josef Bacik4289a662011-08-05 13:22:24 -04004888 /*
Miao Xie8407aa42012-09-07 01:43:32 -06004889 * This is a bit simpler than btrfs_truncate since we've already
4890 * reserved our space for our orphan item in the unlink, so we just
4891 * need to reserve some slack space in case we add bytes and update
4892 * inode item when doing the truncate.
Josef Bacik4289a662011-08-05 13:22:24 -04004893 */
Yan, Zheng80825102009-11-12 09:35:36 +00004894 while (1) {
Miao Xie08e007d2012-10-16 11:33:38 +00004895 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4896 BTRFS_RESERVE_FLUSH_LIMIT);
Yan, Zheng80825102009-11-12 09:35:36 +00004897
Josef Bacik726c35f2011-09-26 15:46:06 -04004898 /*
4899 * Try and steal from the global reserve since we will
4900 * likely not use this space anyway, we want to try as
4901 * hard as possible to get this to work.
4902 */
4903 if (ret)
4904 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4905
Yan, Zhengd68fc572010-05-16 10:49:58 -04004906 if (ret) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00004907 btrfs_warn(root->fs_info,
4908 "Could not get space for a delete, will truncate on mount %d",
4909 ret);
Josef Bacik4289a662011-08-05 13:22:24 -04004910 btrfs_orphan_del(NULL, inode);
4911 btrfs_free_block_rsv(root, rsv);
4912 goto no_delete;
Yan, Zhengd68fc572010-05-16 10:49:58 -04004913 }
4914
Miao Xie0e8c36a2012-12-19 06:59:51 +00004915 trans = btrfs_join_transaction(root);
Josef Bacik4289a662011-08-05 13:22:24 -04004916 if (IS_ERR(trans)) {
4917 btrfs_orphan_del(NULL, inode);
4918 btrfs_free_block_rsv(root, rsv);
4919 goto no_delete;
4920 }
4921
4922 trans->block_rsv = rsv;
4923
Yan, Zhengd68fc572010-05-16 10:49:58 -04004924 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
Josef Bacikca7e70f2012-08-27 17:48:15 -04004925 if (ret != -ENOSPC)
Yan, Zheng80825102009-11-12 09:35:36 +00004926 break;
4927
Miao Xie8407aa42012-09-07 01:43:32 -06004928 trans->block_rsv = &root->fs_info->trans_block_rsv;
Yan, Zheng80825102009-11-12 09:35:36 +00004929 btrfs_end_transaction(trans, root);
4930 trans = NULL;
Liu Bob53d3f52012-11-14 14:34:34 +00004931 btrfs_btree_balance_dirty(root);
Josef Bacik7b128762008-07-24 12:17:14 -04004932 }
4933
Josef Bacik4289a662011-08-05 13:22:24 -04004934 btrfs_free_block_rsv(root, rsv);
4935
Josef Bacik4ef31a42013-08-13 14:10:08 -04004936 /*
4937 * Errors here aren't a big deal, it just means we leave orphan items
4938 * in the tree. They will be cleaned up on the next mount.
4939 */
Yan, Zheng80825102009-11-12 09:35:36 +00004940 if (ret == 0) {
Josef Bacik4289a662011-08-05 13:22:24 -04004941 trans->block_rsv = root->orphan_block_rsv;
Josef Bacik4ef31a42013-08-13 14:10:08 -04004942 btrfs_orphan_del(trans, inode);
4943 } else {
4944 btrfs_orphan_del(NULL, inode);
Yan, Zheng80825102009-11-12 09:35:36 +00004945 }
Chris Mason85e21ba2008-01-29 15:11:36 -05004946
Josef Bacik4289a662011-08-05 13:22:24 -04004947 trans->block_rsv = &root->fs_info->trans_block_rsv;
Li Zefan581bb052011-04-20 10:06:11 +08004948 if (!(root == root->fs_info->tree_root ||
4949 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
Li Zefan33345d012011-04-20 10:31:50 +08004950 btrfs_return_ino(root, btrfs_ino(inode));
Li Zefan581bb052011-04-20 10:06:11 +08004951
Chris Mason54aa1f42007-06-22 14:16:25 -04004952 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00004953 btrfs_btree_balance_dirty(root);
Chris Mason39279cc2007-06-12 06:35:45 -04004954no_delete:
Miao Xie89042e52013-05-15 07:48:15 +00004955 btrfs_remove_delayed_node(inode);
Jan Karadbd57682012-05-03 14:48:02 +02004956 clear_inode(inode);
Yan, Zheng80825102009-11-12 09:35:36 +00004957 return;
Chris Mason39279cc2007-06-12 06:35:45 -04004958}
4959
4960/*
4961 * this returns the key found in the dir entry in the location pointer.
4962 * If no dir entries were found, location->objectid is 0.
4963 */
4964static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4965 struct btrfs_key *location)
4966{
4967 const char *name = dentry->d_name.name;
4968 int namelen = dentry->d_name.len;
4969 struct btrfs_dir_item *di;
4970 struct btrfs_path *path;
4971 struct btrfs_root *root = BTRFS_I(dir)->root;
Yan0d9f7f32007-10-25 15:48:28 -04004972 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004973
4974 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07004975 if (!path)
4976 return -ENOMEM;
Chris Mason39544012007-12-12 14:38:19 -05004977
Li Zefan33345d012011-04-20 10:31:50 +08004978 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
Chris Mason39279cc2007-06-12 06:35:45 -04004979 namelen, 0);
Yan0d9f7f32007-10-25 15:48:28 -04004980 if (IS_ERR(di))
4981 ret = PTR_ERR(di);
Chris Masond3977122009-01-05 21:25:51 -05004982
David Sterbac7040052011-04-19 18:00:01 +02004983 if (IS_ERR_OR_NULL(di))
Chris Mason39544012007-12-12 14:38:19 -05004984 goto out_err;
Chris Masond3977122009-01-05 21:25:51 -05004985
Chris Mason5f39d392007-10-15 16:14:19 -04004986 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
Chris Mason39279cc2007-06-12 06:35:45 -04004987out:
Chris Mason39279cc2007-06-12 06:35:45 -04004988 btrfs_free_path(path);
4989 return ret;
Chris Mason39544012007-12-12 14:38:19 -05004990out_err:
4991 location->objectid = 0;
4992 goto out;
Chris Mason39279cc2007-06-12 06:35:45 -04004993}
4994
4995/*
4996 * when we hit a tree root in a directory, the btrfs part of the inode
4997 * needs to be changed to reflect the root directory of the tree root. This
4998 * is kind of like crossing a mount point.
4999 */
5000static int fixup_tree_root_location(struct btrfs_root *root,
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005001 struct inode *dir,
5002 struct dentry *dentry,
5003 struct btrfs_key *location,
5004 struct btrfs_root **sub_root)
Chris Mason39279cc2007-06-12 06:35:45 -04005005{
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005006 struct btrfs_path *path;
5007 struct btrfs_root *new_root;
5008 struct btrfs_root_ref *ref;
5009 struct extent_buffer *leaf;
5010 int ret;
5011 int err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04005012
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005013 path = btrfs_alloc_path();
5014 if (!path) {
5015 err = -ENOMEM;
5016 goto out;
5017 }
Chris Mason39279cc2007-06-12 06:35:45 -04005018
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005019 err = -ENOENT;
Kelley Nielsen75ac2dd2013-11-04 19:35:58 -08005020 ret = btrfs_find_item(root->fs_info->tree_root, path,
5021 BTRFS_I(dir)->root->root_key.objectid,
5022 location->objectid, BTRFS_ROOT_REF_KEY, NULL);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005023 if (ret) {
5024 if (ret < 0)
5025 err = ret;
5026 goto out;
5027 }
Chris Mason39279cc2007-06-12 06:35:45 -04005028
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005029 leaf = path->nodes[0];
5030 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
Li Zefan33345d012011-04-20 10:31:50 +08005031 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005032 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5033 goto out;
5034
5035 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5036 (unsigned long)(ref + 1),
5037 dentry->d_name.len);
5038 if (ret)
5039 goto out;
5040
David Sterbab3b4aa72011-04-21 01:20:15 +02005041 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005042
5043 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5044 if (IS_ERR(new_root)) {
5045 err = PTR_ERR(new_root);
5046 goto out;
5047 }
5048
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005049 *sub_root = new_root;
5050 location->objectid = btrfs_root_dirid(&new_root->root_item);
5051 location->type = BTRFS_INODE_ITEM_KEY;
Chris Mason39279cc2007-06-12 06:35:45 -04005052 location->offset = 0;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005053 err = 0;
5054out:
5055 btrfs_free_path(path);
5056 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04005057}
5058
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005059static void inode_tree_add(struct inode *inode)
5060{
5061 struct btrfs_root *root = BTRFS_I(inode)->root;
5062 struct btrfs_inode *entry;
From: Nick Piggin03e860b2009-08-21 10:09:44 +02005063 struct rb_node **p;
5064 struct rb_node *parent;
Filipe David Borba Mananacef21932013-09-02 12:19:13 +01005065 struct rb_node *new = &BTRFS_I(inode)->rb_node;
Li Zefan33345d012011-04-20 10:31:50 +08005066 u64 ino = btrfs_ino(inode);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005067
Al Viro1d3382cb2010-10-23 15:19:20 -04005068 if (inode_unhashed(inode))
Yan, Zheng76dda932009-09-21 16:00:26 -04005069 return;
Miao Xiee1409ce2013-05-15 07:48:16 +00005070 parent = NULL;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005071 spin_lock(&root->inode_lock);
Miao Xiee1409ce2013-05-15 07:48:16 +00005072 p = &root->inode_tree.rb_node;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005073 while (*p) {
5074 parent = *p;
5075 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5076
Li Zefan33345d012011-04-20 10:31:50 +08005077 if (ino < btrfs_ino(&entry->vfs_inode))
From: Nick Piggin03e860b2009-08-21 10:09:44 +02005078 p = &parent->rb_left;
Li Zefan33345d012011-04-20 10:31:50 +08005079 else if (ino > btrfs_ino(&entry->vfs_inode))
From: Nick Piggin03e860b2009-08-21 10:09:44 +02005080 p = &parent->rb_right;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005081 else {
5082 WARN_ON(!(entry->vfs_inode.i_state &
Al Viroa4ffdde2010-06-02 17:38:30 -04005083 (I_WILL_FREE | I_FREEING)));
Filipe David Borba Mananacef21932013-09-02 12:19:13 +01005084 rb_replace_node(parent, new, &root->inode_tree);
From: Nick Piggin03e860b2009-08-21 10:09:44 +02005085 RB_CLEAR_NODE(parent);
5086 spin_unlock(&root->inode_lock);
Filipe David Borba Mananacef21932013-09-02 12:19:13 +01005087 return;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005088 }
5089 }
Filipe David Borba Mananacef21932013-09-02 12:19:13 +01005090 rb_link_node(new, parent, p);
5091 rb_insert_color(new, &root->inode_tree);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005092 spin_unlock(&root->inode_lock);
5093}
5094
5095static void inode_tree_del(struct inode *inode)
5096{
5097 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng76dda932009-09-21 16:00:26 -04005098 int empty = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005099
From: Nick Piggin03e860b2009-08-21 10:09:44 +02005100 spin_lock(&root->inode_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005101 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005102 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005103 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
Yan, Zheng76dda932009-09-21 16:00:26 -04005104 empty = RB_EMPTY_ROOT(&root->inode_tree);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005105 }
From: Nick Piggin03e860b2009-08-21 10:09:44 +02005106 spin_unlock(&root->inode_lock);
Yan, Zheng76dda932009-09-21 16:00:26 -04005107
Stefan Behrens69e9c6c2013-09-05 16:58:43 +02005108 if (empty && btrfs_root_refs(&root->root_item) == 0) {
Yan, Zheng76dda932009-09-21 16:00:26 -04005109 synchronize_srcu(&root->fs_info->subvol_srcu);
5110 spin_lock(&root->inode_lock);
5111 empty = RB_EMPTY_ROOT(&root->inode_tree);
5112 spin_unlock(&root->inode_lock);
5113 if (empty)
5114 btrfs_add_dead_root(root);
5115 }
5116}
5117
Jeff Mahoney143bede2012-03-01 14:56:26 +01005118void btrfs_invalidate_inodes(struct btrfs_root *root)
Yan, Zheng76dda932009-09-21 16:00:26 -04005119{
5120 struct rb_node *node;
5121 struct rb_node *prev;
5122 struct btrfs_inode *entry;
5123 struct inode *inode;
5124 u64 objectid = 0;
5125
Liu Bo7813b3d2014-02-10 17:37:25 +08005126 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5127 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
Yan, Zheng76dda932009-09-21 16:00:26 -04005128
5129 spin_lock(&root->inode_lock);
5130again:
5131 node = root->inode_tree.rb_node;
5132 prev = NULL;
5133 while (node) {
5134 prev = node;
5135 entry = rb_entry(node, struct btrfs_inode, rb_node);
5136
Li Zefan33345d012011-04-20 10:31:50 +08005137 if (objectid < btrfs_ino(&entry->vfs_inode))
Yan, Zheng76dda932009-09-21 16:00:26 -04005138 node = node->rb_left;
Li Zefan33345d012011-04-20 10:31:50 +08005139 else if (objectid > btrfs_ino(&entry->vfs_inode))
Yan, Zheng76dda932009-09-21 16:00:26 -04005140 node = node->rb_right;
5141 else
5142 break;
5143 }
5144 if (!node) {
5145 while (prev) {
5146 entry = rb_entry(prev, struct btrfs_inode, rb_node);
Li Zefan33345d012011-04-20 10:31:50 +08005147 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
Yan, Zheng76dda932009-09-21 16:00:26 -04005148 node = prev;
5149 break;
5150 }
5151 prev = rb_next(prev);
5152 }
5153 }
5154 while (node) {
5155 entry = rb_entry(node, struct btrfs_inode, rb_node);
Li Zefan33345d012011-04-20 10:31:50 +08005156 objectid = btrfs_ino(&entry->vfs_inode) + 1;
Yan, Zheng76dda932009-09-21 16:00:26 -04005157 inode = igrab(&entry->vfs_inode);
5158 if (inode) {
5159 spin_unlock(&root->inode_lock);
5160 if (atomic_read(&inode->i_count) > 1)
5161 d_prune_aliases(inode);
5162 /*
Al Viro45321ac2010-06-07 13:43:19 -04005163 * btrfs_drop_inode will have it removed from
Yan, Zheng76dda932009-09-21 16:00:26 -04005164 * the inode cache when its usage count
5165 * hits zero.
5166 */
5167 iput(inode);
5168 cond_resched();
5169 spin_lock(&root->inode_lock);
5170 goto again;
5171 }
5172
5173 if (cond_resched_lock(&root->inode_lock))
5174 goto again;
5175
5176 node = rb_next(node);
5177 }
5178 spin_unlock(&root->inode_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005179}
5180
Chris Masone02119d2008-09-05 16:13:11 -04005181static int btrfs_init_locked_inode(struct inode *inode, void *p)
5182{
5183 struct btrfs_iget_args *args = p;
Chris Mason90d3e592014-01-09 17:28:00 -08005184 inode->i_ino = args->location->objectid;
5185 memcpy(&BTRFS_I(inode)->location, args->location,
5186 sizeof(*args->location));
Chris Masone02119d2008-09-05 16:13:11 -04005187 BTRFS_I(inode)->root = args->root;
Chris Mason39279cc2007-06-12 06:35:45 -04005188 return 0;
5189}
5190
5191static int btrfs_find_actor(struct inode *inode, void *opaque)
5192{
5193 struct btrfs_iget_args *args = opaque;
Chris Mason90d3e592014-01-09 17:28:00 -08005194 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
Chris Masond3977122009-01-05 21:25:51 -05005195 args->root == BTRFS_I(inode)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04005196}
5197
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005198static struct inode *btrfs_iget_locked(struct super_block *s,
Chris Mason90d3e592014-01-09 17:28:00 -08005199 struct btrfs_key *location,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005200 struct btrfs_root *root)
Chris Mason39279cc2007-06-12 06:35:45 -04005201{
5202 struct inode *inode;
5203 struct btrfs_iget_args args;
Chris Mason90d3e592014-01-09 17:28:00 -08005204 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
Filipe David Borba Manana778ba822013-10-06 22:22:33 +01005205
Chris Mason90d3e592014-01-09 17:28:00 -08005206 args.location = location;
Chris Mason39279cc2007-06-12 06:35:45 -04005207 args.root = root;
5208
Filipe David Borba Manana778ba822013-10-06 22:22:33 +01005209 inode = iget5_locked(s, hashval, btrfs_find_actor,
Chris Mason39279cc2007-06-12 06:35:45 -04005210 btrfs_init_locked_inode,
5211 (void *)&args);
5212 return inode;
5213}
5214
Balaji Rao1a54ef82008-07-21 02:01:04 +05305215/* Get an inode object given its location and corresponding root.
5216 * Returns in *is_new if the inode was read from disk
5217 */
5218struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
Josef Bacik73f73412009-12-04 17:38:27 +00005219 struct btrfs_root *root, int *new)
Balaji Rao1a54ef82008-07-21 02:01:04 +05305220{
5221 struct inode *inode;
5222
Chris Mason90d3e592014-01-09 17:28:00 -08005223 inode = btrfs_iget_locked(s, location, root);
Balaji Rao1a54ef82008-07-21 02:01:04 +05305224 if (!inode)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005225 return ERR_PTR(-ENOMEM);
Balaji Rao1a54ef82008-07-21 02:01:04 +05305226
5227 if (inode->i_state & I_NEW) {
Balaji Rao1a54ef82008-07-21 02:01:04 +05305228 btrfs_read_locked_inode(inode);
Mark Fasheh1748f842011-07-12 11:25:31 -07005229 if (!is_bad_inode(inode)) {
5230 inode_tree_add(inode);
5231 unlock_new_inode(inode);
5232 if (new)
5233 *new = 1;
5234 } else {
Sergei Trofimoviche0b6d652011-09-11 10:52:24 -04005235 unlock_new_inode(inode);
5236 iput(inode);
5237 inode = ERR_PTR(-ESTALE);
Mark Fasheh1748f842011-07-12 11:25:31 -07005238 }
5239 }
5240
Balaji Rao1a54ef82008-07-21 02:01:04 +05305241 return inode;
5242}
5243
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005244static struct inode *new_simple_dir(struct super_block *s,
5245 struct btrfs_key *key,
5246 struct btrfs_root *root)
5247{
5248 struct inode *inode = new_inode(s);
5249
5250 if (!inode)
5251 return ERR_PTR(-ENOMEM);
5252
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005253 BTRFS_I(inode)->root = root;
5254 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
Josef Bacik72ac3c02012-05-23 14:13:11 -04005255 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005256
5257 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
Li Zefan848cce02012-02-21 17:04:28 +08005258 inode->i_op = &btrfs_dir_ro_inode_operations;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005259 inode->i_fop = &simple_dir_operations;
5260 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5261 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5262
5263 return inode;
5264}
5265
Chris Mason3de45862008-11-17 21:02:50 -05005266struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
Chris Mason39279cc2007-06-12 06:35:45 -04005267{
Chris Masond3977122009-01-05 21:25:51 -05005268 struct inode *inode;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005269 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04005270 struct btrfs_root *sub_root = root;
5271 struct btrfs_key location;
Yan, Zheng76dda932009-09-21 16:00:26 -04005272 int index;
Josef Bacikb4aff1f2011-06-28 16:18:59 -04005273 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04005274
5275 if (dentry->d_name.len > BTRFS_NAME_LEN)
5276 return ERR_PTR(-ENAMETOOLONG);
Chris Mason5f39d392007-10-15 16:14:19 -04005277
Jeff Layton39e3c952012-11-28 11:30:53 -05005278 ret = btrfs_inode_by_name(dir, dentry, &location);
Chris Mason39279cc2007-06-12 06:35:45 -04005279 if (ret < 0)
5280 return ERR_PTR(ret);
Chris Mason5f39d392007-10-15 16:14:19 -04005281
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005282 if (location.objectid == 0)
Tsutomu Itoh5662344b32013-12-13 09:51:42 +09005283 return ERR_PTR(-ENOENT);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005284
5285 if (location.type == BTRFS_INODE_ITEM_KEY) {
Josef Bacik73f73412009-12-04 17:38:27 +00005286 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005287 return inode;
Chris Mason39279cc2007-06-12 06:35:45 -04005288 }
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005289
5290 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5291
Yan, Zheng76dda932009-09-21 16:00:26 -04005292 index = srcu_read_lock(&root->fs_info->subvol_srcu);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005293 ret = fixup_tree_root_location(root, dir, dentry,
5294 &location, &sub_root);
5295 if (ret < 0) {
5296 if (ret != -ENOENT)
5297 inode = ERR_PTR(ret);
5298 else
5299 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5300 } else {
Josef Bacik73f73412009-12-04 17:38:27 +00005301 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04005302 }
Yan, Zheng76dda932009-09-21 16:00:26 -04005303 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5304
Julia Lawall34d19ba2011-01-24 19:55:19 +00005305 if (!IS_ERR(inode) && root != sub_root) {
Yan, Zhengc71bf092009-11-12 09:34:40 +00005306 down_read(&root->fs_info->cleanup_work_sem);
5307 if (!(inode->i_sb->s_flags & MS_RDONLY))
Josef Bacik66b4ffd2011-01-31 16:22:42 -05005308 ret = btrfs_orphan_cleanup(sub_root);
Yan, Zhengc71bf092009-11-12 09:34:40 +00005309 up_read(&root->fs_info->cleanup_work_sem);
Josef Bacik01cd3362013-06-03 21:39:49 -04005310 if (ret) {
5311 iput(inode);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05005312 inode = ERR_PTR(ret);
Josef Bacik01cd3362013-06-03 21:39:49 -04005313 }
Yan, Zhengc71bf092009-11-12 09:34:40 +00005314 }
5315
Chris Mason3de45862008-11-17 21:02:50 -05005316 return inode;
5317}
5318
Nick Pigginfe15ce42011-01-07 17:49:23 +11005319static int btrfs_dentry_delete(const struct dentry *dentry)
Yan, Zheng76dda932009-09-21 16:00:26 -04005320{
5321 struct btrfs_root *root;
Li Zefan848cce02012-02-21 17:04:28 +08005322 struct inode *inode = dentry->d_inode;
Yan, Zheng76dda932009-09-21 16:00:26 -04005323
Li Zefan848cce02012-02-21 17:04:28 +08005324 if (!inode && !IS_ROOT(dentry))
5325 inode = dentry->d_parent->d_inode;
Yan, Zheng76dda932009-09-21 16:00:26 -04005326
Li Zefan848cce02012-02-21 17:04:28 +08005327 if (inode) {
5328 root = BTRFS_I(inode)->root;
Yan, Zhengefefb142009-10-09 09:25:16 -04005329 if (btrfs_root_refs(&root->root_item) == 0)
5330 return 1;
Li Zefan848cce02012-02-21 17:04:28 +08005331
5332 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5333 return 1;
Yan, Zhengefefb142009-10-09 09:25:16 -04005334 }
Yan, Zheng76dda932009-09-21 16:00:26 -04005335 return 0;
5336}
5337
Josef Bacikb4aff1f2011-06-28 16:18:59 -04005338static void btrfs_dentry_release(struct dentry *dentry)
5339{
Daeseok Youn944a4512014-04-14 15:37:02 +09005340 kfree(dentry->d_fsdata);
Josef Bacikb4aff1f2011-06-28 16:18:59 -04005341}
5342
Chris Mason3de45862008-11-17 21:02:50 -05005343static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
Al Viro00cd8dd2012-06-10 17:13:09 -04005344 unsigned int flags)
Chris Mason3de45862008-11-17 21:02:50 -05005345{
Tsutomu Itoh5662344b32013-12-13 09:51:42 +09005346 struct inode *inode;
Josef Bacika66e7cc2011-09-18 10:34:03 -04005347
Tsutomu Itoh5662344b32013-12-13 09:51:42 +09005348 inode = btrfs_lookup_dentry(dir, dentry);
5349 if (IS_ERR(inode)) {
5350 if (PTR_ERR(inode) == -ENOENT)
5351 inode = NULL;
5352 else
5353 return ERR_CAST(inode);
5354 }
5355
Al Viro41d28bc2014-10-12 22:24:21 -04005356 return d_splice_alias(inode, dentry);
Chris Mason39279cc2007-06-12 06:35:45 -04005357}
5358
Miao Xie16cdcec2011-04-22 18:12:22 +08005359unsigned char btrfs_filetype_table[] = {
Chris Mason39279cc2007-06-12 06:35:45 -04005360 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5361};
5362
Al Viro9cdda8d2013-05-22 16:48:09 -04005363static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
Chris Mason39279cc2007-06-12 06:35:45 -04005364{
Al Viro9cdda8d2013-05-22 16:48:09 -04005365 struct inode *inode = file_inode(file);
Chris Mason39279cc2007-06-12 06:35:45 -04005366 struct btrfs_root *root = BTRFS_I(inode)->root;
5367 struct btrfs_item *item;
5368 struct btrfs_dir_item *di;
5369 struct btrfs_key key;
Chris Mason5f39d392007-10-15 16:14:19 -04005370 struct btrfs_key found_key;
Chris Mason39279cc2007-06-12 06:35:45 -04005371 struct btrfs_path *path;
Miao Xie16cdcec2011-04-22 18:12:22 +08005372 struct list_head ins_list;
5373 struct list_head del_list;
Chris Mason39279cc2007-06-12 06:35:45 -04005374 int ret;
Chris Mason5f39d392007-10-15 16:14:19 -04005375 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04005376 int slot;
Chris Mason39279cc2007-06-12 06:35:45 -04005377 unsigned char d_type;
5378 int over = 0;
5379 u32 di_cur;
5380 u32 di_total;
5381 u32 di_len;
5382 int key_type = BTRFS_DIR_INDEX_KEY;
Chris Mason5f39d392007-10-15 16:14:19 -04005383 char tmp_name[32];
5384 char *name_ptr;
5385 int name_len;
Al Viro9cdda8d2013-05-22 16:48:09 -04005386 int is_curr = 0; /* ctx->pos points to the current index? */
Chris Mason39279cc2007-06-12 06:35:45 -04005387
5388 /* FIXME, use a real flag for deciding about the key type */
5389 if (root->fs_info->tree_root == root)
5390 key_type = BTRFS_DIR_ITEM_KEY;
Chris Mason5f39d392007-10-15 16:14:19 -04005391
Al Viro9cdda8d2013-05-22 16:48:09 -04005392 if (!dir_emit_dots(file, ctx))
5393 return 0;
5394
David Woodhouse49593bf2008-08-17 17:08:36 +01005395 path = btrfs_alloc_path();
Miao Xie16cdcec2011-04-22 18:12:22 +08005396 if (!path)
5397 return -ENOMEM;
Chris Masonff5714c2011-05-28 07:00:39 -04005398
Josef Bacik026fd312011-05-13 10:32:11 -04005399 path->reada = 1;
David Woodhouse49593bf2008-08-17 17:08:36 +01005400
Miao Xie16cdcec2011-04-22 18:12:22 +08005401 if (key_type == BTRFS_DIR_INDEX_KEY) {
5402 INIT_LIST_HEAD(&ins_list);
5403 INIT_LIST_HEAD(&del_list);
5404 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5405 }
5406
David Sterba962a2982014-06-04 18:41:45 +02005407 key.type = key_type;
Al Viro9cdda8d2013-05-22 16:48:09 -04005408 key.offset = ctx->pos;
Li Zefan33345d012011-04-20 10:31:50 +08005409 key.objectid = btrfs_ino(inode);
Chris Mason5f39d392007-10-15 16:14:19 -04005410
Chris Mason39279cc2007-06-12 06:35:45 -04005411 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5412 if (ret < 0)
5413 goto err;
David Woodhouse49593bf2008-08-17 17:08:36 +01005414
5415 while (1) {
Chris Mason5f39d392007-10-15 16:14:19 -04005416 leaf = path->nodes[0];
Chris Mason39279cc2007-06-12 06:35:45 -04005417 slot = path->slots[0];
Li Zefanb9e03af2011-03-23 10:43:58 +08005418 if (slot >= btrfs_header_nritems(leaf)) {
5419 ret = btrfs_next_leaf(root, path);
5420 if (ret < 0)
5421 goto err;
5422 else if (ret > 0)
5423 break;
5424 continue;
Chris Mason39279cc2007-06-12 06:35:45 -04005425 }
Chris Mason3de45862008-11-17 21:02:50 -05005426
Ross Kirkdd3cc162013-09-16 15:58:09 +01005427 item = btrfs_item_nr(slot);
Chris Mason5f39d392007-10-15 16:14:19 -04005428 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5429
5430 if (found_key.objectid != key.objectid)
Chris Mason39279cc2007-06-12 06:35:45 -04005431 break;
David Sterba962a2982014-06-04 18:41:45 +02005432 if (found_key.type != key_type)
Chris Mason39279cc2007-06-12 06:35:45 -04005433 break;
Al Viro9cdda8d2013-05-22 16:48:09 -04005434 if (found_key.offset < ctx->pos)
Li Zefanb9e03af2011-03-23 10:43:58 +08005435 goto next;
Miao Xie16cdcec2011-04-22 18:12:22 +08005436 if (key_type == BTRFS_DIR_INDEX_KEY &&
5437 btrfs_should_delete_dir_index(&del_list,
5438 found_key.offset))
5439 goto next;
Chris Mason5f39d392007-10-15 16:14:19 -04005440
Al Viro9cdda8d2013-05-22 16:48:09 -04005441 ctx->pos = found_key.offset;
Miao Xie16cdcec2011-04-22 18:12:22 +08005442 is_curr = 1;
David Woodhouse49593bf2008-08-17 17:08:36 +01005443
Chris Mason39279cc2007-06-12 06:35:45 -04005444 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5445 di_cur = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04005446 di_total = btrfs_item_size(leaf, item);
David Woodhouse49593bf2008-08-17 17:08:36 +01005447
5448 while (di_cur < di_total) {
Chris Mason5f39d392007-10-15 16:14:19 -04005449 struct btrfs_key location;
5450
Josef Bacik22a94d42011-03-16 16:47:17 -04005451 if (verify_dir_item(root, leaf, di))
5452 break;
5453
Chris Mason5f39d392007-10-15 16:14:19 -04005454 name_len = btrfs_dir_name_len(leaf, di);
David Woodhouse49593bf2008-08-17 17:08:36 +01005455 if (name_len <= sizeof(tmp_name)) {
Chris Mason5f39d392007-10-15 16:14:19 -04005456 name_ptr = tmp_name;
5457 } else {
5458 name_ptr = kmalloc(name_len, GFP_NOFS);
David Woodhouse49593bf2008-08-17 17:08:36 +01005459 if (!name_ptr) {
5460 ret = -ENOMEM;
5461 goto err;
5462 }
Chris Mason5f39d392007-10-15 16:14:19 -04005463 }
5464 read_extent_buffer(leaf, name_ptr,
5465 (unsigned long)(di + 1), name_len);
5466
5467 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5468 btrfs_dir_item_key_to_cpu(leaf, di, &location);
Chris Mason3de45862008-11-17 21:02:50 -05005469
Josef Bacikb4aff1f2011-06-28 16:18:59 -04005470
Chris Mason3de45862008-11-17 21:02:50 -05005471 /* is this a reference to our own snapshot? If so
Arne Jansen8c9c2bf2012-02-25 09:09:30 +01005472 * skip it.
5473 *
5474 * In contrast to old kernels, we insert the snapshot's
5475 * dir item and dir index after it has been created, so
5476 * we won't find a reference to our own snapshot. We
5477 * still keep the following code for backward
5478 * compatibility.
Chris Mason3de45862008-11-17 21:02:50 -05005479 */
5480 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5481 location.objectid == root->root_key.objectid) {
5482 over = 0;
5483 goto skip;
5484 }
Al Viro9cdda8d2013-05-22 16:48:09 -04005485 over = !dir_emit(ctx, name_ptr, name_len,
5486 location.objectid, d_type);
Chris Mason5f39d392007-10-15 16:14:19 -04005487
Chris Mason3de45862008-11-17 21:02:50 -05005488skip:
Chris Mason5f39d392007-10-15 16:14:19 -04005489 if (name_ptr != tmp_name)
5490 kfree(name_ptr);
5491
Chris Mason39279cc2007-06-12 06:35:45 -04005492 if (over)
5493 goto nopos;
Josef Bacik5103e942007-11-16 11:45:54 -05005494 di_len = btrfs_dir_name_len(leaf, di) +
David Woodhouse49593bf2008-08-17 17:08:36 +01005495 btrfs_dir_data_len(leaf, di) + sizeof(*di);
Chris Mason39279cc2007-06-12 06:35:45 -04005496 di_cur += di_len;
5497 di = (struct btrfs_dir_item *)((char *)di + di_len);
5498 }
Li Zefanb9e03af2011-03-23 10:43:58 +08005499next:
5500 path->slots[0]++;
Chris Mason39279cc2007-06-12 06:35:45 -04005501 }
David Woodhouse49593bf2008-08-17 17:08:36 +01005502
Miao Xie16cdcec2011-04-22 18:12:22 +08005503 if (key_type == BTRFS_DIR_INDEX_KEY) {
5504 if (is_curr)
Al Viro9cdda8d2013-05-22 16:48:09 -04005505 ctx->pos++;
5506 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
Miao Xie16cdcec2011-04-22 18:12:22 +08005507 if (ret)
5508 goto nopos;
5509 }
5510
David Woodhouse49593bf2008-08-17 17:08:36 +01005511 /* Reached end of directory/root. Bump pos past the last item. */
Zach Browndb62efb2013-07-11 16:19:42 -07005512 ctx->pos++;
5513
5514 /*
5515 * Stop new entries from being returned after we return the last
5516 * entry.
5517 *
5518 * New directory entries are assigned a strictly increasing
5519 * offset. This means that new entries created during readdir
5520 * are *guaranteed* to be seen in the future by that readdir.
5521 * This has broken buggy programs which operate on names as
5522 * they're returned by readdir. Until we re-use freed offsets
5523 * we have this hack to stop new entries from being returned
5524 * under the assumption that they'll never reach this huge
5525 * offset.
5526 *
5527 * This is being careful not to overflow 32bit loff_t unless the
5528 * last entry requires it because doing so has broken 32bit apps
5529 * in the past.
5530 */
5531 if (key_type == BTRFS_DIR_INDEX_KEY) {
5532 if (ctx->pos >= INT_MAX)
5533 ctx->pos = LLONG_MAX;
5534 else
5535 ctx->pos = INT_MAX;
5536 }
Chris Mason39279cc2007-06-12 06:35:45 -04005537nopos:
5538 ret = 0;
5539err:
Miao Xie16cdcec2011-04-22 18:12:22 +08005540 if (key_type == BTRFS_DIR_INDEX_KEY)
5541 btrfs_put_delayed_items(&ins_list, &del_list);
Chris Mason39279cc2007-06-12 06:35:45 -04005542 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04005543 return ret;
5544}
5545
Christoph Hellwiga9185b42010-03-05 09:21:37 +01005546int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
Chris Mason39279cc2007-06-12 06:35:45 -04005547{
5548 struct btrfs_root *root = BTRFS_I(inode)->root;
5549 struct btrfs_trans_handle *trans;
5550 int ret = 0;
Josef Bacik0af3d002010-06-21 14:48:16 -04005551 bool nolock = false;
Chris Mason39279cc2007-06-12 06:35:45 -04005552
Josef Bacik72ac3c02012-05-23 14:13:11 -04005553 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
Chris Mason4ca8b412008-08-05 13:30:48 -04005554 return 0;
5555
Liu Bo83eea1f2012-07-10 05:28:39 -06005556 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
Li Zefan82d59022011-04-20 10:33:24 +08005557 nolock = true;
Josef Bacik0af3d002010-06-21 14:48:16 -04005558
Christoph Hellwiga9185b42010-03-05 09:21:37 +01005559 if (wbc->sync_mode == WB_SYNC_ALL) {
Josef Bacik0af3d002010-06-21 14:48:16 -04005560 if (nolock)
Josef Bacik7a7eaa42011-04-13 12:54:33 -04005561 trans = btrfs_join_transaction_nolock(root);
Josef Bacik0af3d002010-06-21 14:48:16 -04005562 else
Josef Bacik7a7eaa42011-04-13 12:54:33 -04005563 trans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00005564 if (IS_ERR(trans))
5565 return PTR_ERR(trans);
Miao Xiea698d0752012-09-20 01:51:59 -06005566 ret = btrfs_commit_transaction(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04005567 }
5568 return ret;
5569}
5570
5571/*
Chris Mason54aa1f42007-06-22 14:16:25 -04005572 * This is somewhat expensive, updating the tree every time the
Chris Mason39279cc2007-06-12 06:35:45 -04005573 * inode changes. But, it is most likely to find the inode in cache.
5574 * FIXME, needs more benchmarking...there are no reasons other than performance
5575 * to keep or drop this code.
5576 */
Eric Sandeen48a3b632013-04-25 20:41:01 +00005577static int btrfs_dirty_inode(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04005578{
5579 struct btrfs_root *root = BTRFS_I(inode)->root;
5580 struct btrfs_trans_handle *trans;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04005581 int ret;
5582
Josef Bacik72ac3c02012-05-23 14:13:11 -04005583 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
Josef Bacik22c44fe2011-11-30 10:45:38 -05005584 return 0;
Chris Mason39279cc2007-06-12 06:35:45 -04005585
Josef Bacik7a7eaa42011-04-13 12:54:33 -04005586 trans = btrfs_join_transaction(root);
Josef Bacik22c44fe2011-11-30 10:45:38 -05005587 if (IS_ERR(trans))
5588 return PTR_ERR(trans);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04005589
5590 ret = btrfs_update_inode(trans, root, inode);
Chris Mason94b60442010-05-26 11:02:00 -04005591 if (ret && ret == -ENOSPC) {
5592 /* whoops, lets try again with the full transaction */
5593 btrfs_end_transaction(trans, root);
5594 trans = btrfs_start_transaction(root, 1);
Josef Bacik22c44fe2011-11-30 10:45:38 -05005595 if (IS_ERR(trans))
5596 return PTR_ERR(trans);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04005597
Chris Mason94b60442010-05-26 11:02:00 -04005598 ret = btrfs_update_inode(trans, root, inode);
Chris Mason94b60442010-05-26 11:02:00 -04005599 }
Chris Mason39279cc2007-06-12 06:35:45 -04005600 btrfs_end_transaction(trans, root);
Miao Xie16cdcec2011-04-22 18:12:22 +08005601 if (BTRFS_I(inode)->delayed_node)
5602 btrfs_balance_delayed_items(root);
Josef Bacik22c44fe2011-11-30 10:45:38 -05005603
5604 return ret;
5605}
5606
5607/*
5608 * This is a copy of file_update_time. We need this so we can return error on
5609 * ENOSPC for updating the inode in the case of file write and mmap writes.
5610 */
Josef Bacike41f9412012-03-26 09:46:47 -04005611static int btrfs_update_time(struct inode *inode, struct timespec *now,
5612 int flags)
Josef Bacik22c44fe2011-11-30 10:45:38 -05005613{
Alexander Block2bc5565282012-06-15 09:49:33 +02005614 struct btrfs_root *root = BTRFS_I(inode)->root;
5615
5616 if (btrfs_root_readonly(root))
5617 return -EROFS;
5618
Josef Bacike41f9412012-03-26 09:46:47 -04005619 if (flags & S_VERSION)
Josef Bacik22c44fe2011-11-30 10:45:38 -05005620 inode_inc_iversion(inode);
Josef Bacike41f9412012-03-26 09:46:47 -04005621 if (flags & S_CTIME)
5622 inode->i_ctime = *now;
5623 if (flags & S_MTIME)
5624 inode->i_mtime = *now;
5625 if (flags & S_ATIME)
5626 inode->i_atime = *now;
5627 return btrfs_dirty_inode(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04005628}
5629
Chris Masond352ac62008-09-29 15:18:18 -04005630/*
5631 * find the highest existing sequence number in a directory
5632 * and then set the in-memory index_cnt variable to reflect
5633 * free sequence numbers
5634 */
Josef Bacikaec74772008-07-24 12:12:38 -04005635static int btrfs_set_inode_index_count(struct inode *inode)
5636{
5637 struct btrfs_root *root = BTRFS_I(inode)->root;
5638 struct btrfs_key key, found_key;
5639 struct btrfs_path *path;
5640 struct extent_buffer *leaf;
5641 int ret;
5642
Li Zefan33345d012011-04-20 10:31:50 +08005643 key.objectid = btrfs_ino(inode);
David Sterba962a2982014-06-04 18:41:45 +02005644 key.type = BTRFS_DIR_INDEX_KEY;
Josef Bacikaec74772008-07-24 12:12:38 -04005645 key.offset = (u64)-1;
5646
5647 path = btrfs_alloc_path();
5648 if (!path)
5649 return -ENOMEM;
5650
5651 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5652 if (ret < 0)
5653 goto out;
5654 /* FIXME: we should be able to handle this */
5655 if (ret == 0)
5656 goto out;
5657 ret = 0;
5658
5659 /*
5660 * MAGIC NUMBER EXPLANATION:
5661 * since we search a directory based on f_pos we have to start at 2
5662 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5663 * else has to start at 2
5664 */
5665 if (path->slots[0] == 0) {
5666 BTRFS_I(inode)->index_cnt = 2;
5667 goto out;
5668 }
5669
5670 path->slots[0]--;
5671
5672 leaf = path->nodes[0];
5673 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5674
Li Zefan33345d012011-04-20 10:31:50 +08005675 if (found_key.objectid != btrfs_ino(inode) ||
David Sterba962a2982014-06-04 18:41:45 +02005676 found_key.type != BTRFS_DIR_INDEX_KEY) {
Josef Bacikaec74772008-07-24 12:12:38 -04005677 BTRFS_I(inode)->index_cnt = 2;
5678 goto out;
5679 }
5680
5681 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5682out:
5683 btrfs_free_path(path);
5684 return ret;
5685}
5686
Chris Masond352ac62008-09-29 15:18:18 -04005687/*
5688 * helper to find a free sequence number in a given directory. This current
5689 * code is very simple, later versions will do smarter things in the btree
5690 */
Chris Mason3de45862008-11-17 21:02:50 -05005691int btrfs_set_inode_index(struct inode *dir, u64 *index)
Josef Bacikaec74772008-07-24 12:12:38 -04005692{
5693 int ret = 0;
5694
5695 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
Miao Xie16cdcec2011-04-22 18:12:22 +08005696 ret = btrfs_inode_delayed_dir_index_count(dir);
5697 if (ret) {
5698 ret = btrfs_set_inode_index_count(dir);
5699 if (ret)
5700 return ret;
5701 }
Josef Bacikaec74772008-07-24 12:12:38 -04005702 }
5703
Chris Mason00e4e6b2008-08-05 11:18:09 -04005704 *index = BTRFS_I(dir)->index_cnt;
Josef Bacikaec74772008-07-24 12:12:38 -04005705 BTRFS_I(dir)->index_cnt++;
5706
5707 return ret;
5708}
5709
Chris Masonb0d5d102014-09-08 13:08:51 -07005710static int btrfs_insert_inode_locked(struct inode *inode)
5711{
5712 struct btrfs_iget_args args;
5713 args.location = &BTRFS_I(inode)->location;
5714 args.root = BTRFS_I(inode)->root;
5715
5716 return insert_inode_locked4(inode,
5717 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
5718 btrfs_find_actor, &args);
5719}
5720
Chris Mason39279cc2007-06-12 06:35:45 -04005721static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5722 struct btrfs_root *root,
Josef Bacikaec74772008-07-24 12:12:38 -04005723 struct inode *dir,
Chris Mason9c583092008-01-29 15:15:18 -05005724 const char *name, int name_len,
Al Viro175a4eb2011-07-26 03:30:54 -04005725 u64 ref_objectid, u64 objectid,
5726 umode_t mode, u64 *index)
Chris Mason39279cc2007-06-12 06:35:45 -04005727{
5728 struct inode *inode;
Chris Mason5f39d392007-10-15 16:14:19 -04005729 struct btrfs_inode_item *inode_item;
Chris Mason39279cc2007-06-12 06:35:45 -04005730 struct btrfs_key *location;
Chris Mason5f39d392007-10-15 16:14:19 -04005731 struct btrfs_path *path;
Chris Mason9c583092008-01-29 15:15:18 -05005732 struct btrfs_inode_ref *ref;
5733 struct btrfs_key key[2];
5734 u32 sizes[2];
Filipe Mananaef3b9af2014-04-27 20:40:45 +01005735 int nitems = name ? 2 : 1;
Chris Mason9c583092008-01-29 15:15:18 -05005736 unsigned long ptr;
Chris Mason39279cc2007-06-12 06:35:45 -04005737 int ret;
Chris Mason39279cc2007-06-12 06:35:45 -04005738
Chris Mason5f39d392007-10-15 16:14:19 -04005739 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07005740 if (!path)
5741 return ERR_PTR(-ENOMEM);
Chris Mason5f39d392007-10-15 16:14:19 -04005742
Chris Mason39279cc2007-06-12 06:35:45 -04005743 inode = new_inode(root->fs_info->sb);
Yoshinori Sano8fb27642011-04-09 02:30:07 +00005744 if (!inode) {
5745 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04005746 return ERR_PTR(-ENOMEM);
Yoshinori Sano8fb27642011-04-09 02:30:07 +00005747 }
Chris Mason39279cc2007-06-12 06:35:45 -04005748
Li Zefan581bb052011-04-20 10:06:11 +08005749 /*
Filipe Manana5762b5c2014-08-01 00:10:32 +01005750 * O_TMPFILE, set link count to 0, so that after this point,
5751 * we fill in an inode item with the correct link count.
5752 */
5753 if (!name)
5754 set_nlink(inode, 0);
5755
5756 /*
Li Zefan581bb052011-04-20 10:06:11 +08005757 * we have to initialize this early, so we can reclaim the inode
5758 * number if we fail afterwards in this function.
5759 */
5760 inode->i_ino = objectid;
5761
Filipe Mananaef3b9af2014-04-27 20:40:45 +01005762 if (dir && name) {
liubo1abe9b82011-03-24 11:18:59 +00005763 trace_btrfs_inode_request(dir);
5764
Chris Mason3de45862008-11-17 21:02:50 -05005765 ret = btrfs_set_inode_index(dir, index);
Shen Feng09771432009-04-02 16:46:06 -04005766 if (ret) {
Yoshinori Sano8fb27642011-04-09 02:30:07 +00005767 btrfs_free_path(path);
Shen Feng09771432009-04-02 16:46:06 -04005768 iput(inode);
Josef Bacikaec74772008-07-24 12:12:38 -04005769 return ERR_PTR(ret);
Shen Feng09771432009-04-02 16:46:06 -04005770 }
Filipe Mananaef3b9af2014-04-27 20:40:45 +01005771 } else if (dir) {
5772 *index = 0;
Josef Bacikaec74772008-07-24 12:12:38 -04005773 }
5774 /*
5775 * index_cnt is ignored for everything but a dir,
5776 * btrfs_get_inode_index_count has an explanation for the magic
5777 * number
5778 */
5779 BTRFS_I(inode)->index_cnt = 2;
Miao Xie67de1172013-12-26 13:07:06 +08005780 BTRFS_I(inode)->dir_index = *index;
Chris Mason39279cc2007-06-12 06:35:45 -04005781 BTRFS_I(inode)->root = root;
Chris Masone02119d2008-09-05 16:13:11 -04005782 BTRFS_I(inode)->generation = trans->transid;
Josef Bacik76195852010-11-19 02:18:02 +00005783 inode->i_generation = BTRFS_I(inode)->generation;
Chris Masonb888db22007-08-27 16:49:44 -04005784
Josef Bacik5dc562c2012-08-17 13:14:17 -04005785 /*
5786 * We could have gotten an inode number from somebody who was fsynced
5787 * and then removed in this same transaction, so let's just set full
5788 * sync since it will be a full sync anyway and this will blow away the
5789 * old info in the log.
5790 */
5791 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5792
Chris Mason9c583092008-01-29 15:15:18 -05005793 key[0].objectid = objectid;
David Sterba962a2982014-06-04 18:41:45 +02005794 key[0].type = BTRFS_INODE_ITEM_KEY;
Chris Mason9c583092008-01-29 15:15:18 -05005795 key[0].offset = 0;
5796
Chris Mason9c583092008-01-29 15:15:18 -05005797 sizes[0] = sizeof(struct btrfs_inode_item);
Filipe Mananaef3b9af2014-04-27 20:40:45 +01005798
5799 if (name) {
5800 /*
5801 * Start new inodes with an inode_ref. This is slightly more
5802 * efficient for small numbers of hard links since they will
5803 * be packed into one item. Extended refs will kick in if we
5804 * add more hard links than can fit in the ref item.
5805 */
5806 key[1].objectid = objectid;
David Sterba962a2982014-06-04 18:41:45 +02005807 key[1].type = BTRFS_INODE_REF_KEY;
Filipe Mananaef3b9af2014-04-27 20:40:45 +01005808 key[1].offset = ref_objectid;
5809
5810 sizes[1] = name_len + sizeof(*ref);
5811 }
Chris Mason9c583092008-01-29 15:15:18 -05005812
Chris Masonb0d5d102014-09-08 13:08:51 -07005813 location = &BTRFS_I(inode)->location;
5814 location->objectid = objectid;
5815 location->offset = 0;
David Sterba962a2982014-06-04 18:41:45 +02005816 location->type = BTRFS_INODE_ITEM_KEY;
Chris Masonb0d5d102014-09-08 13:08:51 -07005817
5818 ret = btrfs_insert_inode_locked(inode);
5819 if (ret < 0)
5820 goto fail;
5821
Chris Masonb9473432009-03-13 11:00:37 -04005822 path->leave_spinning = 1;
Filipe Mananaef3b9af2014-04-27 20:40:45 +01005823 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
Chris Mason9c583092008-01-29 15:15:18 -05005824 if (ret != 0)
Chris Masonb0d5d102014-09-08 13:08:51 -07005825 goto fail_unlock;
Chris Mason5f39d392007-10-15 16:14:19 -04005826
Dmitry Monakhovecc11fab2010-03-04 17:31:47 +03005827 inode_init_owner(inode, dir, mode);
Yan Zhenga76a3cd2008-10-09 11:46:29 -04005828 inode_set_bytes(inode, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04005829 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
Chris Mason5f39d392007-10-15 16:14:19 -04005830 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5831 struct btrfs_inode_item);
Li Zefan293f7e02012-07-10 00:58:58 -06005832 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5833 sizeof(*inode_item));
Chris Masone02119d2008-09-05 16:13:11 -04005834 fill_inode_item(trans, path->nodes[0], inode_item, inode);
Chris Mason9c583092008-01-29 15:15:18 -05005835
Filipe Mananaef3b9af2014-04-27 20:40:45 +01005836 if (name) {
5837 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5838 struct btrfs_inode_ref);
5839 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5840 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5841 ptr = (unsigned long)(ref + 1);
5842 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5843 }
Chris Mason9c583092008-01-29 15:15:18 -05005844
Chris Mason5f39d392007-10-15 16:14:19 -04005845 btrfs_mark_buffer_dirty(path->nodes[0]);
5846 btrfs_free_path(path);
5847
Christoph Hellwig6cbff002009-04-17 10:37:41 +02005848 btrfs_inherit_iflags(inode, dir);
5849
Al Viro569254b2011-07-24 17:08:40 -04005850 if (S_ISREG(mode)) {
Chris Mason94272162009-07-02 12:26:06 -04005851 if (btrfs_test_opt(root, NODATASUM))
5852 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
Liu Bo213490b2012-09-11 08:33:50 -06005853 if (btrfs_test_opt(root, NODATACOW))
Josef Bacikf2bdf9a2013-02-21 15:28:28 -05005854 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5855 BTRFS_INODE_NODATASUM;
Chris Mason94272162009-07-02 12:26:06 -04005856 }
5857
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005858 inode_tree_add(inode);
liubo1abe9b82011-03-24 11:18:59 +00005859
5860 trace_btrfs_inode_new(inode);
Chris Mason1973f0f2011-06-24 13:13:29 -04005861 btrfs_set_inode_last_trans(trans, inode);
liubo1abe9b82011-03-24 11:18:59 +00005862
Alexander Block8ea05e32012-07-25 17:35:53 +02005863 btrfs_update_root_times(trans, root);
5864
Filipe David Borba Manana63541922014-01-07 11:47:46 +00005865 ret = btrfs_inode_inherit_props(trans, inode, dir);
5866 if (ret)
5867 btrfs_err(root->fs_info,
5868 "error inheriting props for ino %llu (root %llu): %d",
5869 btrfs_ino(inode), root->root_key.objectid, ret);
5870
Chris Mason39279cc2007-06-12 06:35:45 -04005871 return inode;
Chris Masonb0d5d102014-09-08 13:08:51 -07005872
5873fail_unlock:
5874 unlock_new_inode(inode);
Chris Mason5f39d392007-10-15 16:14:19 -04005875fail:
Filipe Mananaef3b9af2014-04-27 20:40:45 +01005876 if (dir && name)
Josef Bacikaec74772008-07-24 12:12:38 -04005877 BTRFS_I(dir)->index_cnt--;
Chris Mason5f39d392007-10-15 16:14:19 -04005878 btrfs_free_path(path);
Shen Feng09771432009-04-02 16:46:06 -04005879 iput(inode);
Chris Mason5f39d392007-10-15 16:14:19 -04005880 return ERR_PTR(ret);
Chris Mason39279cc2007-06-12 06:35:45 -04005881}
5882
5883static inline u8 btrfs_inode_type(struct inode *inode)
5884{
5885 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5886}
5887
Chris Masond352ac62008-09-29 15:18:18 -04005888/*
5889 * utility function to add 'inode' into 'parent_inode' with
5890 * a give name and a given sequence number.
5891 * if 'add_backref' is true, also insert a backref from the
5892 * inode to the parent directory.
5893 */
Chris Masone02119d2008-09-05 16:13:11 -04005894int btrfs_add_link(struct btrfs_trans_handle *trans,
5895 struct inode *parent_inode, struct inode *inode,
5896 const char *name, int name_len, int add_backref, u64 index)
Chris Mason39279cc2007-06-12 06:35:45 -04005897{
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005898 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04005899 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -04005900 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
Li Zefan33345d012011-04-20 10:31:50 +08005901 u64 ino = btrfs_ino(inode);
5902 u64 parent_ino = btrfs_ino(parent_inode);
Chris Mason5f39d392007-10-15 16:14:19 -04005903
Li Zefan33345d012011-04-20 10:31:50 +08005904 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005905 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5906 } else {
Li Zefan33345d012011-04-20 10:31:50 +08005907 key.objectid = ino;
David Sterba962a2982014-06-04 18:41:45 +02005908 key.type = BTRFS_INODE_ITEM_KEY;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005909 key.offset = 0;
5910 }
Chris Mason39279cc2007-06-12 06:35:45 -04005911
Li Zefan33345d012011-04-20 10:31:50 +08005912 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005913 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5914 key.objectid, root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +08005915 parent_ino, index, name, name_len);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005916 } else if (add_backref) {
Li Zefan33345d012011-04-20 10:31:50 +08005917 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5918 parent_ino, index);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005919 }
5920
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005921 /* Nothing to clean up yet */
5922 if (ret)
5923 return ret;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005924
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005925 ret = btrfs_insert_dir_item(trans, root, name, name_len,
5926 parent_inode, &key,
5927 btrfs_inode_type(inode), index);
Chris Mason9c520572012-12-17 14:26:57 -05005928 if (ret == -EEXIST || ret == -EOVERFLOW)
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005929 goto fail_dir_item;
5930 else if (ret) {
5931 btrfs_abort_transaction(trans, root, ret);
5932 return ret;
Chris Mason39279cc2007-06-12 06:35:45 -04005933 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005934
5935 btrfs_i_size_write(parent_inode, parent_inode->i_size +
5936 name_len * 2);
Josef Bacik0c4d2d92012-04-05 15:03:02 -04005937 inode_inc_iversion(parent_inode);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005938 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5939 ret = btrfs_update_inode(trans, root, parent_inode);
5940 if (ret)
5941 btrfs_abort_transaction(trans, root, ret);
Chris Mason39279cc2007-06-12 06:35:45 -04005942 return ret;
Chris Masonfe66a052012-02-20 08:40:56 -05005943
5944fail_dir_item:
5945 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5946 u64 local_index;
5947 int err;
5948 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5949 key.objectid, root->root_key.objectid,
5950 parent_ino, &local_index, name, name_len);
5951
5952 } else if (add_backref) {
5953 u64 local_index;
5954 int err;
5955
5956 err = btrfs_del_inode_ref(trans, root, name, name_len,
5957 ino, parent_ino, &local_index);
5958 }
5959 return ret;
Chris Mason39279cc2007-06-12 06:35:45 -04005960}
5961
5962static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
Josef Bacika1b075d2010-11-19 20:36:11 +00005963 struct inode *dir, struct dentry *dentry,
5964 struct inode *inode, int backref, u64 index)
Chris Mason39279cc2007-06-12 06:35:45 -04005965{
Josef Bacika1b075d2010-11-19 20:36:11 +00005966 int err = btrfs_add_link(trans, dir, inode,
5967 dentry->d_name.name, dentry->d_name.len,
5968 backref, index);
Chris Mason39279cc2007-06-12 06:35:45 -04005969 if (err > 0)
5970 err = -EEXIST;
5971 return err;
5972}
5973
Josef Bacik618e21d2007-07-11 10:18:17 -04005974static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
Al Viro1a67aaf2011-07-26 01:52:52 -04005975 umode_t mode, dev_t rdev)
Josef Bacik618e21d2007-07-11 10:18:17 -04005976{
5977 struct btrfs_trans_handle *trans;
5978 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason1832a6d2007-12-21 16:27:21 -05005979 struct inode *inode = NULL;
Josef Bacik618e21d2007-07-11 10:18:17 -04005980 int err;
5981 int drop_inode = 0;
5982 u64 objectid;
Chris Mason00e4e6b2008-08-05 11:18:09 -04005983 u64 index = 0;
Josef Bacik618e21d2007-07-11 10:18:17 -04005984
5985 if (!new_valid_dev(rdev))
5986 return -EINVAL;
5987
Josef Bacik9ed74f22009-09-11 16:12:44 -04005988 /*
5989 * 2 for inode item and ref
5990 * 2 for dir items
5991 * 1 for xattr if selinux is on
5992 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04005993 trans = btrfs_start_transaction(root, 5);
5994 if (IS_ERR(trans))
5995 return PTR_ERR(trans);
Chris Mason1832a6d2007-12-21 16:27:21 -05005996
Li Zefan581bb052011-04-20 10:06:11 +08005997 err = btrfs_find_free_ino(root, &objectid);
5998 if (err)
5999 goto out_unlock;
6000
Josef Bacikaec74772008-07-24 12:12:38 -04006001 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Li Zefan33345d012011-04-20 10:31:50 +08006002 dentry->d_name.len, btrfs_ino(dir), objectid,
Josef Bacikd82a6f12011-05-11 15:26:06 -04006003 mode, &index);
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04006004 if (IS_ERR(inode)) {
6005 err = PTR_ERR(inode);
Josef Bacik618e21d2007-07-11 10:18:17 -04006006 goto out_unlock;
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04006007 }
Josef Bacik618e21d2007-07-11 10:18:17 -04006008
Casey Schauflerad19db72011-12-15 10:09:07 -05006009 /*
6010 * If the active LSM wants to access the inode during
6011 * d_instantiate it needs these. Smack checks to see
6012 * if the filesystem supports xattrs by looking at the
6013 * ops vector.
6014 */
Casey Schauflerad19db72011-12-15 10:09:07 -05006015 inode->i_op = &btrfs_special_inode_operations;
Chris Masonb0d5d102014-09-08 13:08:51 -07006016 init_special_inode(inode, inode->i_mode, rdev);
6017
6018 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
Josef Bacik618e21d2007-07-11 10:18:17 -04006019 if (err)
Chris Masonb0d5d102014-09-08 13:08:51 -07006020 goto out_unlock_inode;
6021
6022 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6023 if (err) {
6024 goto out_unlock_inode;
6025 } else {
Yan1b4ab1b2007-08-29 09:11:44 -04006026 btrfs_update_inode(trans, root, inode);
Chris Masonb0d5d102014-09-08 13:08:51 -07006027 unlock_new_inode(inode);
Al Viro08c422c2011-12-23 07:58:13 -05006028 d_instantiate(dentry, inode);
Josef Bacik618e21d2007-07-11 10:18:17 -04006029 }
Chris Masonb0d5d102014-09-08 13:08:51 -07006030
Josef Bacik618e21d2007-07-11 10:18:17 -04006031out_unlock:
Josef Bacik7ad85bb2012-01-12 19:10:12 -05006032 btrfs_end_transaction(trans, root);
Josef Bacikc581afc2014-02-06 16:06:06 -05006033 btrfs_balance_delayed_items(root);
Liu Bob53d3f52012-11-14 14:34:34 +00006034 btrfs_btree_balance_dirty(root);
Josef Bacik618e21d2007-07-11 10:18:17 -04006035 if (drop_inode) {
6036 inode_dec_link_count(inode);
6037 iput(inode);
6038 }
Josef Bacik618e21d2007-07-11 10:18:17 -04006039 return err;
Chris Masonb0d5d102014-09-08 13:08:51 -07006040
6041out_unlock_inode:
6042 drop_inode = 1;
6043 unlock_new_inode(inode);
6044 goto out_unlock;
6045
Josef Bacik618e21d2007-07-11 10:18:17 -04006046}
6047
Chris Mason39279cc2007-06-12 06:35:45 -04006048static int btrfs_create(struct inode *dir, struct dentry *dentry,
Al Viroebfc3b42012-06-10 18:05:36 -04006049 umode_t mode, bool excl)
Chris Mason39279cc2007-06-12 06:35:45 -04006050{
6051 struct btrfs_trans_handle *trans;
6052 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason1832a6d2007-12-21 16:27:21 -05006053 struct inode *inode = NULL;
Filipe Brandenburger43baa572012-11-30 03:40:09 +00006054 int drop_inode_on_err = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -04006055 int err;
Chris Mason39279cc2007-06-12 06:35:45 -04006056 u64 objectid;
Chris Mason00e4e6b2008-08-05 11:18:09 -04006057 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04006058
Josef Bacik9ed74f22009-09-11 16:12:44 -04006059 /*
6060 * 2 for inode item and ref
6061 * 2 for dir items
6062 * 1 for xattr if selinux is on
6063 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04006064 trans = btrfs_start_transaction(root, 5);
6065 if (IS_ERR(trans))
6066 return PTR_ERR(trans);
Josef Bacik9ed74f22009-09-11 16:12:44 -04006067
Li Zefan581bb052011-04-20 10:06:11 +08006068 err = btrfs_find_free_ino(root, &objectid);
6069 if (err)
6070 goto out_unlock;
6071
Josef Bacikaec74772008-07-24 12:12:38 -04006072 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Li Zefan33345d012011-04-20 10:31:50 +08006073 dentry->d_name.len, btrfs_ino(dir), objectid,
Josef Bacikd82a6f12011-05-11 15:26:06 -04006074 mode, &index);
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04006075 if (IS_ERR(inode)) {
6076 err = PTR_ERR(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04006077 goto out_unlock;
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04006078 }
Filipe Brandenburger43baa572012-11-30 03:40:09 +00006079 drop_inode_on_err = 1;
Casey Schauflerad19db72011-12-15 10:09:07 -05006080 /*
6081 * If the active LSM wants to access the inode during
6082 * d_instantiate it needs these. Smack checks to see
6083 * if the filesystem supports xattrs by looking at the
6084 * ops vector.
6085 */
6086 inode->i_fop = &btrfs_file_operations;
6087 inode->i_op = &btrfs_file_inode_operations;
Chris Masonb0d5d102014-09-08 13:08:51 -07006088 inode->i_mapping->a_ops = &btrfs_aops;
Chris Masonb0d5d102014-09-08 13:08:51 -07006089
6090 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6091 if (err)
6092 goto out_unlock_inode;
6093
6094 err = btrfs_update_inode(trans, root, inode);
6095 if (err)
6096 goto out_unlock_inode;
Casey Schauflerad19db72011-12-15 10:09:07 -05006097
Josef Bacika1b075d2010-11-19 20:36:11 +00006098 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04006099 if (err)
Chris Masonb0d5d102014-09-08 13:08:51 -07006100 goto out_unlock_inode;
Filipe Brandenburger43baa572012-11-30 03:40:09 +00006101
Filipe Brandenburger43baa572012-11-30 03:40:09 +00006102 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
Chris Masonb0d5d102014-09-08 13:08:51 -07006103 unlock_new_inode(inode);
Filipe Brandenburger43baa572012-11-30 03:40:09 +00006104 d_instantiate(dentry, inode);
6105
Chris Mason39279cc2007-06-12 06:35:45 -04006106out_unlock:
Josef Bacik7ad85bb2012-01-12 19:10:12 -05006107 btrfs_end_transaction(trans, root);
Filipe Brandenburger43baa572012-11-30 03:40:09 +00006108 if (err && drop_inode_on_err) {
Chris Mason39279cc2007-06-12 06:35:45 -04006109 inode_dec_link_count(inode);
6110 iput(inode);
6111 }
Josef Bacikc581afc2014-02-06 16:06:06 -05006112 btrfs_balance_delayed_items(root);
Liu Bob53d3f52012-11-14 14:34:34 +00006113 btrfs_btree_balance_dirty(root);
Chris Mason39279cc2007-06-12 06:35:45 -04006114 return err;
Chris Masonb0d5d102014-09-08 13:08:51 -07006115
6116out_unlock_inode:
6117 unlock_new_inode(inode);
6118 goto out_unlock;
6119
Chris Mason39279cc2007-06-12 06:35:45 -04006120}
6121
6122static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6123 struct dentry *dentry)
6124{
6125 struct btrfs_trans_handle *trans;
6126 struct btrfs_root *root = BTRFS_I(dir)->root;
6127 struct inode *inode = old_dentry->d_inode;
Chris Mason00e4e6b2008-08-05 11:18:09 -04006128 u64 index;
Chris Mason39279cc2007-06-12 06:35:45 -04006129 int err;
6130 int drop_inode = 0;
6131
TARUISI Hiroaki4a8be422009-11-12 07:14:26 +00006132 /* do not allow sys_link's with other subvols of the same device */
6133 if (root->objectid != BTRFS_I(inode)->root->objectid)
Mark Fasheh3ab35642011-03-22 17:20:26 +00006134 return -EXDEV;
TARUISI Hiroaki4a8be422009-11-12 07:14:26 +00006135
Mark Fashehf1863732012-08-08 11:32:27 -07006136 if (inode->i_nlink >= BTRFS_LINK_MAX)
Al Viroc055e992011-03-04 17:15:18 +00006137 return -EMLINK;
Josef Bacik9ed74f22009-09-11 16:12:44 -04006138
Chris Mason3de45862008-11-17 21:02:50 -05006139 err = btrfs_set_inode_index(dir, &index);
Josef Bacikaec74772008-07-24 12:12:38 -04006140 if (err)
6141 goto fail;
6142
Yan, Zhenga22285a2010-05-16 10:48:46 -04006143 /*
Miao Xie7e6b6462011-02-18 09:21:17 +00006144 * 2 items for inode and inode ref
Yan, Zhenga22285a2010-05-16 10:48:46 -04006145 * 2 items for dir items
Miao Xie7e6b6462011-02-18 09:21:17 +00006146 * 1 item for parent inode
Yan, Zhenga22285a2010-05-16 10:48:46 -04006147 */
Miao Xie7e6b6462011-02-18 09:21:17 +00006148 trans = btrfs_start_transaction(root, 5);
Yan, Zhenga22285a2010-05-16 10:48:46 -04006149 if (IS_ERR(trans)) {
6150 err = PTR_ERR(trans);
6151 goto fail;
6152 }
Chris Mason5f39d392007-10-15 16:14:19 -04006153
Miao Xie67de1172013-12-26 13:07:06 +08006154 /* There are several dir indexes for this inode, clear the cache. */
6155 BTRFS_I(inode)->dir_index = 0ULL;
Zach Brown8b558c52013-10-16 12:10:34 -07006156 inc_nlink(inode);
Josef Bacik0c4d2d92012-04-05 15:03:02 -04006157 inode_inc_iversion(inode);
Miao Xie31534952011-04-13 13:19:21 +08006158 inode->i_ctime = CURRENT_TIME;
Al Viro7de9c6ee2010-10-23 11:11:40 -04006159 ihold(inode);
Josef Bacike9976152012-10-11 15:53:56 -04006160 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
Josef Bacikaec74772008-07-24 12:12:38 -04006161
Josef Bacika1b075d2010-11-19 20:36:11 +00006162 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
Chris Mason5f39d392007-10-15 16:14:19 -04006163
Yan, Zhenga5719522009-09-24 09:17:31 -04006164 if (err) {
Chris Mason39279cc2007-06-12 06:35:45 -04006165 drop_inode = 1;
Yan, Zhenga5719522009-09-24 09:17:31 -04006166 } else {
Al Viro10d9f302011-07-16 23:09:10 -04006167 struct dentry *parent = dentry->d_parent;
Yan, Zhenga5719522009-09-24 09:17:31 -04006168 err = btrfs_update_inode(trans, root, inode);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006169 if (err)
6170 goto fail;
Filipe Mananaef3b9af2014-04-27 20:40:45 +01006171 if (inode->i_nlink == 1) {
6172 /*
6173 * If new hard link count is 1, it's a file created
6174 * with open(2) O_TMPFILE flag.
6175 */
6176 err = btrfs_orphan_del(trans, inode);
6177 if (err)
6178 goto fail;
6179 }
Al Viro08c422c2011-12-23 07:58:13 -05006180 d_instantiate(dentry, inode);
Josef Bacik6a912212010-11-20 09:48:00 +00006181 btrfs_log_new_name(trans, inode, NULL, parent);
Yan, Zhenga5719522009-09-24 09:17:31 -04006182 }
Chris Mason39279cc2007-06-12 06:35:45 -04006183
Josef Bacik7ad85bb2012-01-12 19:10:12 -05006184 btrfs_end_transaction(trans, root);
Josef Bacikc581afc2014-02-06 16:06:06 -05006185 btrfs_balance_delayed_items(root);
Chris Mason1832a6d2007-12-21 16:27:21 -05006186fail:
Chris Mason39279cc2007-06-12 06:35:45 -04006187 if (drop_inode) {
6188 inode_dec_link_count(inode);
6189 iput(inode);
6190 }
Liu Bob53d3f52012-11-14 14:34:34 +00006191 btrfs_btree_balance_dirty(root);
Chris Mason39279cc2007-06-12 06:35:45 -04006192 return err;
6193}
6194
Al Viro18bb1db2011-07-26 01:41:39 -04006195static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Chris Mason39279cc2007-06-12 06:35:45 -04006196{
Chris Masonb9d86662008-05-02 16:13:49 -04006197 struct inode *inode = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04006198 struct btrfs_trans_handle *trans;
6199 struct btrfs_root *root = BTRFS_I(dir)->root;
6200 int err = 0;
6201 int drop_on_err = 0;
Chris Masonb9d86662008-05-02 16:13:49 -04006202 u64 objectid = 0;
Chris Mason00e4e6b2008-08-05 11:18:09 -04006203 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04006204
Josef Bacik9ed74f22009-09-11 16:12:44 -04006205 /*
6206 * 2 items for inode and ref
6207 * 2 items for dir items
6208 * 1 for xattr if selinux is on
6209 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04006210 trans = btrfs_start_transaction(root, 5);
6211 if (IS_ERR(trans))
6212 return PTR_ERR(trans);
Chris Mason39279cc2007-06-12 06:35:45 -04006213
Li Zefan581bb052011-04-20 10:06:11 +08006214 err = btrfs_find_free_ino(root, &objectid);
6215 if (err)
6216 goto out_fail;
6217
Josef Bacikaec74772008-07-24 12:12:38 -04006218 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Li Zefan33345d012011-04-20 10:31:50 +08006219 dentry->d_name.len, btrfs_ino(dir), objectid,
Josef Bacikd82a6f12011-05-11 15:26:06 -04006220 S_IFDIR | mode, &index);
Chris Mason39279cc2007-06-12 06:35:45 -04006221 if (IS_ERR(inode)) {
6222 err = PTR_ERR(inode);
6223 goto out_fail;
6224 }
Chris Mason5f39d392007-10-15 16:14:19 -04006225
Chris Mason39279cc2007-06-12 06:35:45 -04006226 drop_on_err = 1;
Chris Masonb0d5d102014-09-08 13:08:51 -07006227 /* these must be set before we unlock the inode */
6228 inode->i_op = &btrfs_dir_inode_operations;
6229 inode->i_fop = &btrfs_dir_file_operations;
Josef Bacik33268ea2008-07-24 12:16:36 -04006230
Eric Paris2a7dba32011-02-01 11:05:39 -05006231 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
Josef Bacik33268ea2008-07-24 12:16:36 -04006232 if (err)
Chris Masonb0d5d102014-09-08 13:08:51 -07006233 goto out_fail_inode;
Chris Mason39279cc2007-06-12 06:35:45 -04006234
Chris Masondbe674a2008-07-17 12:54:05 -04006235 btrfs_i_size_write(inode, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04006236 err = btrfs_update_inode(trans, root, inode);
6237 if (err)
Chris Masonb0d5d102014-09-08 13:08:51 -07006238 goto out_fail_inode;
Chris Mason5f39d392007-10-15 16:14:19 -04006239
Josef Bacika1b075d2010-11-19 20:36:11 +00006240 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6241 dentry->d_name.len, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04006242 if (err)
Chris Masonb0d5d102014-09-08 13:08:51 -07006243 goto out_fail_inode;
Chris Mason5f39d392007-10-15 16:14:19 -04006244
Chris Mason39279cc2007-06-12 06:35:45 -04006245 d_instantiate(dentry, inode);
Chris Masonb0d5d102014-09-08 13:08:51 -07006246 /*
6247 * mkdir is special. We're unlocking after we call d_instantiate
6248 * to avoid a race with nfsd calling d_instantiate.
6249 */
6250 unlock_new_inode(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04006251 drop_on_err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04006252
6253out_fail:
Josef Bacik7ad85bb2012-01-12 19:10:12 -05006254 btrfs_end_transaction(trans, root);
Wang Shilongc7cfb8a2014-12-24 14:45:30 +08006255 if (drop_on_err) {
6256 inode_dec_link_count(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04006257 iput(inode);
Wang Shilongc7cfb8a2014-12-24 14:45:30 +08006258 }
Josef Bacikc581afc2014-02-06 16:06:06 -05006259 btrfs_balance_delayed_items(root);
Liu Bob53d3f52012-11-14 14:34:34 +00006260 btrfs_btree_balance_dirty(root);
Chris Mason39279cc2007-06-12 06:35:45 -04006261 return err;
Chris Masonb0d5d102014-09-08 13:08:51 -07006262
6263out_fail_inode:
6264 unlock_new_inode(inode);
6265 goto out_fail;
Chris Mason39279cc2007-06-12 06:35:45 -04006266}
6267
Qu Wenruoe6c4efd2014-09-17 11:53:35 +08006268/* Find next extent map of a given extent map, caller needs to ensure locks */
6269static struct extent_map *next_extent_map(struct extent_map *em)
6270{
6271 struct rb_node *next;
6272
6273 next = rb_next(&em->rb_node);
6274 if (!next)
6275 return NULL;
6276 return container_of(next, struct extent_map, rb_node);
6277}
6278
6279static struct extent_map *prev_extent_map(struct extent_map *em)
6280{
6281 struct rb_node *prev;
6282
6283 prev = rb_prev(&em->rb_node);
6284 if (!prev)
6285 return NULL;
6286 return container_of(prev, struct extent_map, rb_node);
6287}
6288
Chris Masond352ac62008-09-29 15:18:18 -04006289/* helper for btfs_get_extent. Given an existing extent in the tree,
Qu Wenruoe6c4efd2014-09-17 11:53:35 +08006290 * the existing extent is the nearest extent to map_start,
Chris Masond352ac62008-09-29 15:18:18 -04006291 * and an extent that you want to insert, deal with overlap and insert
Qu Wenruoe6c4efd2014-09-17 11:53:35 +08006292 * the best fitted new extent into the tree.
Chris Masond352ac62008-09-29 15:18:18 -04006293 */
Chris Mason3b951512008-04-17 11:29:12 -04006294static int merge_extent_mapping(struct extent_map_tree *em_tree,
6295 struct extent_map *existing,
Chris Masone6dcd2d2008-07-17 12:53:50 -04006296 struct extent_map *em,
Qu Wenruo51f395a2014-08-08 13:06:20 +08006297 u64 map_start)
Chris Mason3b951512008-04-17 11:29:12 -04006298{
Qu Wenruoe6c4efd2014-09-17 11:53:35 +08006299 struct extent_map *prev;
6300 struct extent_map *next;
6301 u64 start;
6302 u64 end;
Chris Mason3b951512008-04-17 11:29:12 -04006303 u64 start_diff;
Chris Mason3b951512008-04-17 11:29:12 -04006304
Chris Masone6dcd2d2008-07-17 12:53:50 -04006305 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
Qu Wenruoe6c4efd2014-09-17 11:53:35 +08006306
6307 if (existing->start > map_start) {
6308 next = existing;
6309 prev = prev_extent_map(next);
6310 } else {
6311 prev = existing;
6312 next = next_extent_map(prev);
6313 }
6314
6315 start = prev ? extent_map_end(prev) : em->start;
6316 start = max_t(u64, start, em->start);
6317 end = next ? next->start : extent_map_end(em);
6318 end = min_t(u64, end, extent_map_end(em));
6319 start_diff = start - em->start;
6320 em->start = start;
6321 em->len = end - start;
Chris Masonc8b97812008-10-29 14:49:59 -04006322 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6323 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
Chris Masone6dcd2d2008-07-17 12:53:50 -04006324 em->block_start += start_diff;
Chris Masonc8b97812008-10-29 14:49:59 -04006325 em->block_len -= start_diff;
6326 }
Josef Bacik09a2a8f92013-04-05 16:51:15 -04006327 return add_extent_mapping(em_tree, em, 0);
Chris Mason3b951512008-04-17 11:29:12 -04006328}
6329
Chris Masonc8b97812008-10-29 14:49:59 -04006330static noinline int uncompress_inline(struct btrfs_path *path,
6331 struct inode *inode, struct page *page,
6332 size_t pg_offset, u64 extent_offset,
6333 struct btrfs_file_extent_item *item)
6334{
6335 int ret;
6336 struct extent_buffer *leaf = path->nodes[0];
6337 char *tmp;
6338 size_t max_size;
6339 unsigned long inline_size;
6340 unsigned long ptr;
Li Zefan261507a02010-12-17 14:21:50 +08006341 int compress_type;
Chris Masonc8b97812008-10-29 14:49:59 -04006342
6343 WARN_ON(pg_offset != 0);
Li Zefan261507a02010-12-17 14:21:50 +08006344 compress_type = btrfs_file_extent_compression(leaf, item);
Chris Masonc8b97812008-10-29 14:49:59 -04006345 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6346 inline_size = btrfs_file_extent_inline_item_len(leaf,
Ross Kirkdd3cc162013-09-16 15:58:09 +01006347 btrfs_item_nr(path->slots[0]));
Chris Masonc8b97812008-10-29 14:49:59 -04006348 tmp = kmalloc(inline_size, GFP_NOFS);
Tsutomu Itoh8d413712011-04-25 19:43:52 -04006349 if (!tmp)
6350 return -ENOMEM;
Chris Masonc8b97812008-10-29 14:49:59 -04006351 ptr = btrfs_file_extent_inline_start(item);
6352
6353 read_extent_buffer(leaf, tmp, ptr, inline_size);
6354
Chris Mason5b050f02008-11-11 09:34:41 -05006355 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
Li Zefan261507a02010-12-17 14:21:50 +08006356 ret = btrfs_decompress(compress_type, tmp, page,
6357 extent_offset, inline_size, max_size);
Chris Masonc8b97812008-10-29 14:49:59 -04006358 kfree(tmp);
Zach Brown166ae5a2014-05-09 17:15:10 -04006359 return ret;
Chris Masonc8b97812008-10-29 14:49:59 -04006360}
6361
Chris Masond352ac62008-09-29 15:18:18 -04006362/*
6363 * a bit scary, this does extent mapping from logical file offset to the disk.
Chris Masond3977122009-01-05 21:25:51 -05006364 * the ugly parts come from merging extents from the disk with the in-ram
6365 * representation. This gets more complex because of the data=ordered code,
Chris Masond352ac62008-09-29 15:18:18 -04006366 * where the in-ram extents might be locked pending data=ordered completion.
6367 *
6368 * This also copies inline extents directly into the page.
6369 */
Chris Masond3977122009-01-05 21:25:51 -05006370
Chris Masona52d9a82007-08-27 16:49:44 -04006371struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
Chris Mason70dec802008-01-29 09:59:12 -05006372 size_t pg_offset, u64 start, u64 len,
Chris Masona52d9a82007-08-27 16:49:44 -04006373 int create)
6374{
6375 int ret;
6376 int err = 0;
Chris Masona52d9a82007-08-27 16:49:44 -04006377 u64 extent_start = 0;
6378 u64 extent_end = 0;
Li Zefan33345d012011-04-20 10:31:50 +08006379 u64 objectid = btrfs_ino(inode);
Chris Masona52d9a82007-08-27 16:49:44 -04006380 u32 found_type;
Chris Masonf4219502008-07-22 11:18:09 -04006381 struct btrfs_path *path = NULL;
Chris Masona52d9a82007-08-27 16:49:44 -04006382 struct btrfs_root *root = BTRFS_I(inode)->root;
6383 struct btrfs_file_extent_item *item;
Chris Mason5f39d392007-10-15 16:14:19 -04006384 struct extent_buffer *leaf;
6385 struct btrfs_key found_key;
Chris Masona52d9a82007-08-27 16:49:44 -04006386 struct extent_map *em = NULL;
6387 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Chris Masond1310b22008-01-24 16:13:08 -05006388 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Chris Masona52d9a82007-08-27 16:49:44 -04006389 struct btrfs_trans_handle *trans = NULL;
Filipe Manana7ffbb592014-06-09 03:48:05 +01006390 const bool new_inline = !page || create;
Chris Masona52d9a82007-08-27 16:49:44 -04006391
Chris Masona52d9a82007-08-27 16:49:44 -04006392again:
Chris Mason890871b2009-09-02 16:24:52 -04006393 read_lock(&em_tree->lock);
Chris Masond1310b22008-01-24 16:13:08 -05006394 em = lookup_extent_mapping(em_tree, start, len);
Chris Masona061fc82008-05-07 11:43:44 -04006395 if (em)
6396 em->bdev = root->fs_info->fs_devices->latest_bdev;
Chris Mason890871b2009-09-02 16:24:52 -04006397 read_unlock(&em_tree->lock);
Chris Masond1310b22008-01-24 16:13:08 -05006398
Chris Masona52d9a82007-08-27 16:49:44 -04006399 if (em) {
Chris Masone1c4b742008-04-22 13:26:46 -04006400 if (em->start > start || em->start + em->len <= start)
6401 free_extent_map(em);
6402 else if (em->block_start == EXTENT_MAP_INLINE && page)
Chris Mason70dec802008-01-29 09:59:12 -05006403 free_extent_map(em);
6404 else
6405 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04006406 }
David Sterba172ddd62011-04-21 00:48:27 +02006407 em = alloc_extent_map();
Chris Masona52d9a82007-08-27 16:49:44 -04006408 if (!em) {
Chris Masond1310b22008-01-24 16:13:08 -05006409 err = -ENOMEM;
6410 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04006411 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04006412 em->bdev = root->fs_info->fs_devices->latest_bdev;
Chris Masond1310b22008-01-24 16:13:08 -05006413 em->start = EXTENT_MAP_HOLE;
Chris Mason445a6942008-11-10 11:53:33 -05006414 em->orig_start = EXTENT_MAP_HOLE;
Chris Masond1310b22008-01-24 16:13:08 -05006415 em->len = (u64)-1;
Chris Masonc8b97812008-10-29 14:49:59 -04006416 em->block_len = (u64)-1;
Chris Masonf4219502008-07-22 11:18:09 -04006417
6418 if (!path) {
6419 path = btrfs_alloc_path();
Josef Bacik026fd312011-05-13 10:32:11 -04006420 if (!path) {
6421 err = -ENOMEM;
6422 goto out;
6423 }
6424 /*
6425 * Chances are we'll be called again, so go ahead and do
6426 * readahead
6427 */
6428 path->reada = 1;
Chris Masonf4219502008-07-22 11:18:09 -04006429 }
6430
Chris Mason179e29e2007-11-01 11:28:41 -04006431 ret = btrfs_lookup_file_extent(trans, root, path,
6432 objectid, start, trans != NULL);
Chris Masona52d9a82007-08-27 16:49:44 -04006433 if (ret < 0) {
6434 err = ret;
6435 goto out;
6436 }
6437
6438 if (ret != 0) {
6439 if (path->slots[0] == 0)
6440 goto not_found;
6441 path->slots[0]--;
6442 }
6443
Chris Mason5f39d392007-10-15 16:14:19 -04006444 leaf = path->nodes[0];
6445 item = btrfs_item_ptr(leaf, path->slots[0],
Chris Masona52d9a82007-08-27 16:49:44 -04006446 struct btrfs_file_extent_item);
Chris Masona52d9a82007-08-27 16:49:44 -04006447 /* are we inside the extent that was found? */
Chris Mason5f39d392007-10-15 16:14:19 -04006448 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
David Sterba962a2982014-06-04 18:41:45 +02006449 found_type = found_key.type;
Chris Mason5f39d392007-10-15 16:14:19 -04006450 if (found_key.objectid != objectid ||
Chris Masona52d9a82007-08-27 16:49:44 -04006451 found_type != BTRFS_EXTENT_DATA_KEY) {
Josef Bacik25a50342013-10-14 12:08:38 -04006452 /*
6453 * If we backup past the first extent we want to move forward
6454 * and see if there is an extent in front of us, otherwise we'll
6455 * say there is a hole for our whole search range which can
6456 * cause problems.
6457 */
6458 extent_end = start;
6459 goto next;
Chris Masona52d9a82007-08-27 16:49:44 -04006460 }
6461
Chris Mason5f39d392007-10-15 16:14:19 -04006462 found_type = btrfs_file_extent_type(leaf, item);
6463 extent_start = found_key.offset;
Yan Zhengd899e052008-10-30 14:25:28 -04006464 if (found_type == BTRFS_FILE_EXTENT_REG ||
6465 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Chris Masona52d9a82007-08-27 16:49:44 -04006466 extent_end = extent_start +
Chris Masondb945352007-10-15 16:15:53 -04006467 btrfs_file_extent_num_bytes(leaf, item);
Yan Zheng9036c102008-10-30 14:19:41 -04006468 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6469 size_t size;
Chris Mason514ac8a2014-01-03 21:07:00 -08006470 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
Qu Wenruofda28322013-02-26 08:10:22 +00006471 extent_end = ALIGN(extent_start + size, root->sectorsize);
Yan Zheng9036c102008-10-30 14:19:41 -04006472 }
Josef Bacik25a50342013-10-14 12:08:38 -04006473next:
Yan Zheng9036c102008-10-30 14:19:41 -04006474 if (start >= extent_end) {
6475 path->slots[0]++;
6476 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6477 ret = btrfs_next_leaf(root, path);
6478 if (ret < 0) {
6479 err = ret;
6480 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04006481 }
Yan Zheng9036c102008-10-30 14:19:41 -04006482 if (ret > 0)
6483 goto not_found;
6484 leaf = path->nodes[0];
Chris Masona52d9a82007-08-27 16:49:44 -04006485 }
Yan Zheng9036c102008-10-30 14:19:41 -04006486 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6487 if (found_key.objectid != objectid ||
6488 found_key.type != BTRFS_EXTENT_DATA_KEY)
6489 goto not_found;
6490 if (start + len <= found_key.offset)
6491 goto not_found;
Wang Shilonge2eca692014-07-17 11:44:14 +08006492 if (start > found_key.offset)
6493 goto next;
Yan Zheng9036c102008-10-30 14:19:41 -04006494 em->start = start;
Josef Bacik70c8a912012-10-11 16:54:30 -04006495 em->orig_start = start;
Yan Zheng9036c102008-10-30 14:19:41 -04006496 em->len = found_key.offset - start;
6497 goto not_found_em;
6498 }
6499
Filipe Manana7ffbb592014-06-09 03:48:05 +01006500 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6501
Yan Zhengd899e052008-10-30 14:25:28 -04006502 if (found_type == BTRFS_FILE_EXTENT_REG ||
6503 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Chris Masona52d9a82007-08-27 16:49:44 -04006504 goto insert;
6505 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason5f39d392007-10-15 16:14:19 -04006506 unsigned long ptr;
Chris Masona52d9a82007-08-27 16:49:44 -04006507 char *map;
Chris Mason3326d1b2007-10-15 16:18:25 -04006508 size_t size;
6509 size_t extent_offset;
6510 size_t copy_size;
Chris Masona52d9a82007-08-27 16:49:44 -04006511
Filipe Manana7ffbb592014-06-09 03:48:05 +01006512 if (new_inline)
Yan689f9342007-10-29 11:41:07 -04006513 goto out;
Yan689f9342007-10-29 11:41:07 -04006514
Chris Mason514ac8a2014-01-03 21:07:00 -08006515 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
Yan Zheng9036c102008-10-30 14:19:41 -04006516 extent_offset = page_offset(page) + pg_offset - extent_start;
Chris Mason70dec802008-01-29 09:59:12 -05006517 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
Yan689f9342007-10-29 11:41:07 -04006518 size - extent_offset);
Chris Mason3326d1b2007-10-15 16:18:25 -04006519 em->start = extent_start + extent_offset;
Qu Wenruofda28322013-02-26 08:10:22 +00006520 em->len = ALIGN(copy_size, root->sectorsize);
Josef Bacikb4939682012-12-03 10:31:19 -05006521 em->orig_block_len = em->len;
Josef Bacik70c8a912012-10-11 16:54:30 -04006522 em->orig_start = em->start;
Yan689f9342007-10-29 11:41:07 -04006523 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
Chris Mason179e29e2007-11-01 11:28:41 -04006524 if (create == 0 && !PageUptodate(page)) {
Li Zefan261507a02010-12-17 14:21:50 +08006525 if (btrfs_file_extent_compression(leaf, item) !=
6526 BTRFS_COMPRESS_NONE) {
Chris Masonc8b97812008-10-29 14:49:59 -04006527 ret = uncompress_inline(path, inode, page,
6528 pg_offset,
6529 extent_offset, item);
Zach Brown166ae5a2014-05-09 17:15:10 -04006530 if (ret) {
6531 err = ret;
6532 goto out;
6533 }
Chris Masonc8b97812008-10-29 14:49:59 -04006534 } else {
6535 map = kmap(page);
6536 read_extent_buffer(leaf, map + pg_offset, ptr,
6537 copy_size);
Chris Mason93c82d52009-09-11 12:36:29 -04006538 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6539 memset(map + pg_offset + copy_size, 0,
6540 PAGE_CACHE_SIZE - pg_offset -
6541 copy_size);
6542 }
Chris Masonc8b97812008-10-29 14:49:59 -04006543 kunmap(page);
6544 }
Chris Mason179e29e2007-11-01 11:28:41 -04006545 flush_dcache_page(page);
6546 } else if (create && PageUptodate(page)) {
Jan Schmidt6bf7e082011-12-01 14:35:19 +01006547 BUG();
Chris Mason179e29e2007-11-01 11:28:41 -04006548 if (!trans) {
6549 kunmap(page);
6550 free_extent_map(em);
6551 em = NULL;
Chris Masonff5714c2011-05-28 07:00:39 -04006552
David Sterbab3b4aa72011-04-21 01:20:15 +02006553 btrfs_release_path(path);
Josef Bacik7a7eaa42011-04-13 12:54:33 -04006554 trans = btrfs_join_transaction(root);
Chris Masonff5714c2011-05-28 07:00:39 -04006555
Tsutomu Itoh3612b492011-01-25 02:51:38 +00006556 if (IS_ERR(trans))
6557 return ERR_CAST(trans);
Chris Mason179e29e2007-11-01 11:28:41 -04006558 goto again;
6559 }
Chris Masonc8b97812008-10-29 14:49:59 -04006560 map = kmap(page);
Chris Mason70dec802008-01-29 09:59:12 -05006561 write_extent_buffer(leaf, map + pg_offset, ptr,
Chris Mason179e29e2007-11-01 11:28:41 -04006562 copy_size);
Chris Masonc8b97812008-10-29 14:49:59 -04006563 kunmap(page);
Chris Mason179e29e2007-11-01 11:28:41 -04006564 btrfs_mark_buffer_dirty(leaf);
Chris Masona52d9a82007-08-27 16:49:44 -04006565 }
Chris Masond1310b22008-01-24 16:13:08 -05006566 set_extent_uptodate(io_tree, em->start,
Arne Jansen507903b2011-04-06 10:02:20 +00006567 extent_map_end(em) - 1, NULL, GFP_NOFS);
Chris Masona52d9a82007-08-27 16:49:44 -04006568 goto insert;
Chris Masona52d9a82007-08-27 16:49:44 -04006569 }
6570not_found:
6571 em->start = start;
Josef Bacik70c8a912012-10-11 16:54:30 -04006572 em->orig_start = start;
Chris Masond1310b22008-01-24 16:13:08 -05006573 em->len = len;
Chris Masona52d9a82007-08-27 16:49:44 -04006574not_found_em:
Chris Mason5f39d392007-10-15 16:14:19 -04006575 em->block_start = EXTENT_MAP_HOLE;
Yan Zheng9036c102008-10-30 14:19:41 -04006576 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
Chris Masona52d9a82007-08-27 16:49:44 -04006577insert:
David Sterbab3b4aa72011-04-21 01:20:15 +02006578 btrfs_release_path(path);
Chris Masond1310b22008-01-24 16:13:08 -05006579 if (em->start > start || extent_map_end(em) <= start) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00006580 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02006581 em->start, em->len, start, len);
Chris Masona52d9a82007-08-27 16:49:44 -04006582 err = -EIO;
6583 goto out;
6584 }
Chris Masond1310b22008-01-24 16:13:08 -05006585
6586 err = 0;
Chris Mason890871b2009-09-02 16:24:52 -04006587 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -04006588 ret = add_extent_mapping(em_tree, em, 0);
Chris Mason3b951512008-04-17 11:29:12 -04006589 /* it is possible that someone inserted the extent into the tree
6590 * while we had the lock dropped. It is also possible that
6591 * an overlapping map exists in the tree
6592 */
Chris Masona52d9a82007-08-27 16:49:44 -04006593 if (ret == -EEXIST) {
Chris Mason3b951512008-04-17 11:29:12 -04006594 struct extent_map *existing;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006595
6596 ret = 0;
6597
Qu Wenruoe6c4efd2014-09-17 11:53:35 +08006598 existing = search_extent_mapping(em_tree, start, len);
6599 /*
6600 * existing will always be non-NULL, since there must be
6601 * extent causing the -EEXIST.
6602 */
6603 if (start >= extent_map_end(existing) ||
Qu Wenruo32be3a12014-09-22 09:13:03 +08006604 start <= existing->start) {
Qu Wenruoe6c4efd2014-09-17 11:53:35 +08006605 /*
6606 * The existing extent map is the one nearest to
6607 * the [start, start + len) range which overlaps
6608 */
6609 err = merge_extent_mapping(em_tree, existing,
6610 em, start);
Chris Masone1c4b742008-04-22 13:26:46 -04006611 free_extent_map(existing);
Qu Wenruoe6c4efd2014-09-17 11:53:35 +08006612 if (err) {
Chris Mason3b951512008-04-17 11:29:12 -04006613 free_extent_map(em);
6614 em = NULL;
6615 }
6616 } else {
6617 free_extent_map(em);
6618 em = existing;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006619 err = 0;
Chris Masona52d9a82007-08-27 16:49:44 -04006620 }
Chris Masona52d9a82007-08-27 16:49:44 -04006621 }
Chris Mason890871b2009-09-02 16:24:52 -04006622 write_unlock(&em_tree->lock);
Chris Masona52d9a82007-08-27 16:49:44 -04006623out:
liubo1abe9b82011-03-24 11:18:59 +00006624
Steven Rostedt4cd85872013-11-14 22:57:29 -05006625 trace_btrfs_get_extent(root, em);
liubo1abe9b82011-03-24 11:18:59 +00006626
Chris Masonf4219502008-07-22 11:18:09 -04006627 if (path)
6628 btrfs_free_path(path);
Chris Masona52d9a82007-08-27 16:49:44 -04006629 if (trans) {
6630 ret = btrfs_end_transaction(trans, root);
Chris Masond3977122009-01-05 21:25:51 -05006631 if (!err)
Chris Masona52d9a82007-08-27 16:49:44 -04006632 err = ret;
6633 }
Chris Masona52d9a82007-08-27 16:49:44 -04006634 if (err) {
6635 free_extent_map(em);
Chris Masona52d9a82007-08-27 16:49:44 -04006636 return ERR_PTR(err);
6637 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006638 BUG_ON(!em); /* Error is always set */
Chris Masona52d9a82007-08-27 16:49:44 -04006639 return em;
6640}
6641
Chris Masonec29ed52011-02-23 16:23:20 -05006642struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6643 size_t pg_offset, u64 start, u64 len,
6644 int create)
6645{
6646 struct extent_map *em;
6647 struct extent_map *hole_em = NULL;
6648 u64 range_start = start;
6649 u64 end;
6650 u64 found;
6651 u64 found_end;
6652 int err = 0;
6653
6654 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6655 if (IS_ERR(em))
6656 return em;
6657 if (em) {
6658 /*
Liu Bof9e4fb52013-01-07 10:10:12 +00006659 * if our em maps to
6660 * - a hole or
6661 * - a pre-alloc extent,
6662 * there might actually be delalloc bytes behind it.
Chris Masonec29ed52011-02-23 16:23:20 -05006663 */
Liu Bof9e4fb52013-01-07 10:10:12 +00006664 if (em->block_start != EXTENT_MAP_HOLE &&
6665 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
Chris Masonec29ed52011-02-23 16:23:20 -05006666 return em;
6667 else
6668 hole_em = em;
6669 }
6670
6671 /* check to see if we've wrapped (len == -1 or similar) */
6672 end = start + len;
6673 if (end < start)
6674 end = (u64)-1;
6675 else
6676 end -= 1;
6677
6678 em = NULL;
6679
6680 /* ok, we didn't find anything, lets look for delalloc */
6681 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6682 end, len, EXTENT_DELALLOC, 1);
6683 found_end = range_start + found;
6684 if (found_end < range_start)
6685 found_end = (u64)-1;
6686
6687 /*
6688 * we didn't find anything useful, return
6689 * the original results from get_extent()
6690 */
6691 if (range_start > end || found_end <= start) {
6692 em = hole_em;
6693 hole_em = NULL;
6694 goto out;
6695 }
6696
6697 /* adjust the range_start to make sure it doesn't
6698 * go backwards from the start they passed in
6699 */
Dulshani Gunawardhana67871252013-10-31 10:33:04 +05306700 range_start = max(start, range_start);
Chris Masonec29ed52011-02-23 16:23:20 -05006701 found = found_end - range_start;
6702
6703 if (found > 0) {
6704 u64 hole_start = start;
6705 u64 hole_len = len;
6706
David Sterba172ddd62011-04-21 00:48:27 +02006707 em = alloc_extent_map();
Chris Masonec29ed52011-02-23 16:23:20 -05006708 if (!em) {
6709 err = -ENOMEM;
6710 goto out;
6711 }
6712 /*
6713 * when btrfs_get_extent can't find anything it
6714 * returns one huge hole
6715 *
6716 * make sure what it found really fits our range, and
6717 * adjust to make sure it is based on the start from
6718 * the caller
6719 */
6720 if (hole_em) {
6721 u64 calc_end = extent_map_end(hole_em);
6722
6723 if (calc_end <= start || (hole_em->start > end)) {
6724 free_extent_map(hole_em);
6725 hole_em = NULL;
6726 } else {
6727 hole_start = max(hole_em->start, start);
6728 hole_len = calc_end - hole_start;
6729 }
6730 }
6731 em->bdev = NULL;
6732 if (hole_em && range_start > hole_start) {
6733 /* our hole starts before our delalloc, so we
6734 * have to return just the parts of the hole
6735 * that go until the delalloc starts
6736 */
6737 em->len = min(hole_len,
6738 range_start - hole_start);
6739 em->start = hole_start;
6740 em->orig_start = hole_start;
6741 /*
6742 * don't adjust block start at all,
6743 * it is fixed at EXTENT_MAP_HOLE
6744 */
6745 em->block_start = hole_em->block_start;
6746 em->block_len = hole_len;
Liu Bof9e4fb52013-01-07 10:10:12 +00006747 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6748 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
Chris Masonec29ed52011-02-23 16:23:20 -05006749 } else {
6750 em->start = range_start;
6751 em->len = found;
6752 em->orig_start = range_start;
6753 em->block_start = EXTENT_MAP_DELALLOC;
6754 em->block_len = found;
6755 }
6756 } else if (hole_em) {
6757 return hole_em;
6758 }
6759out:
6760
6761 free_extent_map(hole_em);
6762 if (err) {
6763 free_extent_map(em);
6764 return ERR_PTR(err);
6765 }
6766 return em;
6767}
6768
Josef Bacik4b46fce2010-05-23 11:00:55 -04006769static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6770 u64 start, u64 len)
6771{
6772 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik70c8a912012-10-11 16:54:30 -04006773 struct extent_map *em;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006774 struct btrfs_key ins;
6775 u64 alloc_hint;
6776 int ret;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006777
Josef Bacik4b46fce2010-05-23 11:00:55 -04006778 alloc_hint = get_extent_allocation_hint(inode, start, len);
Josef Bacik00361582013-08-14 14:02:47 -04006779 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
Miao Xiee570fd22014-06-19 10:42:50 +08006780 alloc_hint, &ins, 1, 1);
Josef Bacik00361582013-08-14 14:02:47 -04006781 if (ret)
6782 return ERR_PTR(ret);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006783
Josef Bacik70c8a912012-10-11 16:54:30 -04006784 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
Josef Bacikcc95bef2013-04-04 14:31:27 -04006785 ins.offset, ins.offset, ins.offset, 0);
Josef Bacik00361582013-08-14 14:02:47 -04006786 if (IS_ERR(em)) {
Miao Xiee570fd22014-06-19 10:42:50 +08006787 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
Josef Bacik00361582013-08-14 14:02:47 -04006788 return em;
6789 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04006790
6791 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6792 ins.offset, ins.offset, 0);
6793 if (ret) {
Miao Xiee570fd22014-06-19 10:42:50 +08006794 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
Josef Bacik00361582013-08-14 14:02:47 -04006795 free_extent_map(em);
6796 return ERR_PTR(ret);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006797 }
Josef Bacik00361582013-08-14 14:02:47 -04006798
Josef Bacik4b46fce2010-05-23 11:00:55 -04006799 return em;
6800}
6801
Chris Mason46bfbb52010-05-26 11:04:10 -04006802/*
6803 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6804 * block must be cow'd
6805 */
Josef Bacik00361582013-08-14 14:02:47 -04006806noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
Josef Bacik7ee9e442013-06-21 16:37:03 -04006807 u64 *orig_start, u64 *orig_block_len,
6808 u64 *ram_bytes)
Chris Mason46bfbb52010-05-26 11:04:10 -04006809{
Josef Bacik00361582013-08-14 14:02:47 -04006810 struct btrfs_trans_handle *trans;
Chris Mason46bfbb52010-05-26 11:04:10 -04006811 struct btrfs_path *path;
6812 int ret;
6813 struct extent_buffer *leaf;
6814 struct btrfs_root *root = BTRFS_I(inode)->root;
Miao Xie7b2b7082014-02-27 13:58:05 +08006815 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Chris Mason46bfbb52010-05-26 11:04:10 -04006816 struct btrfs_file_extent_item *fi;
6817 struct btrfs_key key;
6818 u64 disk_bytenr;
6819 u64 backref_offset;
6820 u64 extent_end;
6821 u64 num_bytes;
6822 int slot;
6823 int found_type;
Josef Bacik7ee9e442013-06-21 16:37:03 -04006824 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
Miao Xiee77751a2013-12-27 21:11:50 +08006825
Chris Mason46bfbb52010-05-26 11:04:10 -04006826 path = btrfs_alloc_path();
6827 if (!path)
6828 return -ENOMEM;
6829
Josef Bacik00361582013-08-14 14:02:47 -04006830 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
Chris Mason46bfbb52010-05-26 11:04:10 -04006831 offset, 0);
6832 if (ret < 0)
6833 goto out;
6834
6835 slot = path->slots[0];
6836 if (ret == 1) {
6837 if (slot == 0) {
6838 /* can't find the item, must cow */
6839 ret = 0;
6840 goto out;
6841 }
6842 slot--;
6843 }
6844 ret = 0;
6845 leaf = path->nodes[0];
6846 btrfs_item_key_to_cpu(leaf, &key, slot);
Li Zefan33345d012011-04-20 10:31:50 +08006847 if (key.objectid != btrfs_ino(inode) ||
Chris Mason46bfbb52010-05-26 11:04:10 -04006848 key.type != BTRFS_EXTENT_DATA_KEY) {
6849 /* not our file or wrong item type, must cow */
6850 goto out;
6851 }
6852
6853 if (key.offset > offset) {
6854 /* Wrong offset, must cow */
6855 goto out;
6856 }
6857
6858 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6859 found_type = btrfs_file_extent_type(leaf, fi);
6860 if (found_type != BTRFS_FILE_EXTENT_REG &&
6861 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6862 /* not a regular extent, must cow */
6863 goto out;
6864 }
Josef Bacik7ee9e442013-06-21 16:37:03 -04006865
6866 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6867 goto out;
6868
Miao Xiee77751a2013-12-27 21:11:50 +08006869 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6870 if (extent_end <= offset)
6871 goto out;
6872
Chris Mason46bfbb52010-05-26 11:04:10 -04006873 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
Josef Bacik7ee9e442013-06-21 16:37:03 -04006874 if (disk_bytenr == 0)
6875 goto out;
6876
6877 if (btrfs_file_extent_compression(leaf, fi) ||
6878 btrfs_file_extent_encryption(leaf, fi) ||
6879 btrfs_file_extent_other_encoding(leaf, fi))
6880 goto out;
6881
Chris Mason46bfbb52010-05-26 11:04:10 -04006882 backref_offset = btrfs_file_extent_offset(leaf, fi);
6883
Josef Bacik7ee9e442013-06-21 16:37:03 -04006884 if (orig_start) {
6885 *orig_start = key.offset - backref_offset;
6886 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6887 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6888 }
Josef Bacikeb384b52013-04-24 16:32:55 -04006889
Chris Mason46bfbb52010-05-26 11:04:10 -04006890 if (btrfs_extent_readonly(root, disk_bytenr))
6891 goto out;
Miao Xie7b2b7082014-02-27 13:58:05 +08006892
6893 num_bytes = min(offset + *len, extent_end) - offset;
6894 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6895 u64 range_end;
6896
6897 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
6898 ret = test_range_bit(io_tree, offset, range_end,
6899 EXTENT_DELALLOC, 0, NULL);
6900 if (ret) {
6901 ret = -EAGAIN;
6902 goto out;
6903 }
6904 }
6905
Josef Bacik1bda19e2013-10-18 12:10:36 -04006906 btrfs_release_path(path);
Chris Mason46bfbb52010-05-26 11:04:10 -04006907
6908 /*
6909 * look for other files referencing this extent, if we
6910 * find any we must cow
6911 */
Josef Bacik00361582013-08-14 14:02:47 -04006912 trans = btrfs_join_transaction(root);
6913 if (IS_ERR(trans)) {
6914 ret = 0;
Chris Mason46bfbb52010-05-26 11:04:10 -04006915 goto out;
Josef Bacik00361582013-08-14 14:02:47 -04006916 }
6917
6918 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6919 key.offset - backref_offset, disk_bytenr);
6920 btrfs_end_transaction(trans, root);
6921 if (ret) {
6922 ret = 0;
6923 goto out;
6924 }
Chris Mason46bfbb52010-05-26 11:04:10 -04006925
6926 /*
6927 * adjust disk_bytenr and num_bytes to cover just the bytes
6928 * in this extent we are about to write. If there
6929 * are any csums in that range we have to cow in order
6930 * to keep the csums correct
6931 */
6932 disk_bytenr += backref_offset;
6933 disk_bytenr += offset - key.offset;
Chris Mason46bfbb52010-05-26 11:04:10 -04006934 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6935 goto out;
6936 /*
6937 * all of the above have passed, it is safe to overwrite this extent
6938 * without cow
6939 */
Josef Bacikeb384b52013-04-24 16:32:55 -04006940 *len = num_bytes;
Chris Mason46bfbb52010-05-26 11:04:10 -04006941 ret = 1;
6942out:
6943 btrfs_free_path(path);
6944 return ret;
6945}
6946
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07006947bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
6948{
6949 struct radix_tree_root *root = &inode->i_mapping->page_tree;
6950 int found = false;
6951 void **pagep = NULL;
6952 struct page *page = NULL;
6953 int start_idx;
6954 int end_idx;
6955
6956 start_idx = start >> PAGE_CACHE_SHIFT;
6957
6958 /*
6959 * end is the last byte in the last page. end == start is legal
6960 */
6961 end_idx = end >> PAGE_CACHE_SHIFT;
6962
6963 rcu_read_lock();
6964
6965 /* Most of the code in this while loop is lifted from
6966 * find_get_page. It's been modified to begin searching from a
6967 * page and return just the first page found in that range. If the
6968 * found idx is less than or equal to the end idx then we know that
6969 * a page exists. If no pages are found or if those pages are
6970 * outside of the range then we're fine (yay!) */
6971 while (page == NULL &&
6972 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
6973 page = radix_tree_deref_slot(pagep);
6974 if (unlikely(!page))
6975 break;
6976
6977 if (radix_tree_exception(page)) {
Filipe Manana809f9012014-06-05 13:22:25 +01006978 if (radix_tree_deref_retry(page)) {
6979 page = NULL;
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07006980 continue;
Filipe Manana809f9012014-06-05 13:22:25 +01006981 }
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07006982 /*
6983 * Otherwise, shmem/tmpfs must be storing a swap entry
6984 * here as an exceptional entry: so return it without
6985 * attempting to raise page count.
6986 */
Filipe Manana6fdef6d2014-06-05 13:22:26 +01006987 page = NULL;
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07006988 break; /* TODO: Is this relevant for this use case? */
6989 }
6990
Filipe Manana91405152014-06-05 13:22:24 +01006991 if (!page_cache_get_speculative(page)) {
6992 page = NULL;
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07006993 continue;
Filipe Manana91405152014-06-05 13:22:24 +01006994 }
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07006995
6996 /*
6997 * Has the page moved?
6998 * This is part of the lockless pagecache protocol. See
6999 * include/linux/pagemap.h for details.
7000 */
7001 if (unlikely(page != *pagep)) {
7002 page_cache_release(page);
7003 page = NULL;
7004 }
7005 }
7006
7007 if (page) {
7008 if (page->index <= end_idx)
7009 found = true;
7010 page_cache_release(page);
7011 }
7012
7013 rcu_read_unlock();
7014 return found;
7015}
7016
Josef Bacikeb838e72012-07-31 16:28:48 -04007017static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7018 struct extent_state **cached_state, int writing)
7019{
7020 struct btrfs_ordered_extent *ordered;
7021 int ret = 0;
7022
7023 while (1) {
7024 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7025 0, cached_state);
7026 /*
7027 * We're concerned with the entire range that we're going to be
7028 * doing DIO to, so we need to make sure theres no ordered
7029 * extents in this range.
7030 */
7031 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7032 lockend - lockstart + 1);
7033
7034 /*
7035 * We need to make sure there are no buffered pages in this
7036 * range either, we could have raced between the invalidate in
7037 * generic_file_direct_write and locking the extent. The
7038 * invalidate needs to happen so that reads after a write do not
7039 * get stale data.
7040 */
Alex Gartrellfc4adbf2014-05-20 13:07:56 -07007041 if (!ordered &&
7042 (!writing ||
7043 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
Josef Bacikeb838e72012-07-31 16:28:48 -04007044 break;
7045
7046 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7047 cached_state, GFP_NOFS);
7048
7049 if (ordered) {
7050 btrfs_start_ordered_extent(inode, ordered, 1);
7051 btrfs_put_ordered_extent(ordered);
7052 } else {
7053 /* Screw you mmap */
Filipe Manana728404d2014-10-10 09:43:11 +01007054 ret = btrfs_fdatawrite_range(inode, lockstart, lockend);
Filipe Manana075bdbd2014-10-09 21:18:55 +01007055 if (ret)
7056 break;
7057 ret = filemap_fdatawait_range(inode->i_mapping,
7058 lockstart,
7059 lockend);
Josef Bacikeb838e72012-07-31 16:28:48 -04007060 if (ret)
7061 break;
7062
7063 /*
7064 * If we found a page that couldn't be invalidated just
7065 * fall back to buffered.
7066 */
7067 ret = invalidate_inode_pages2_range(inode->i_mapping,
7068 lockstart >> PAGE_CACHE_SHIFT,
7069 lockend >> PAGE_CACHE_SHIFT);
7070 if (ret)
7071 break;
7072 }
7073
7074 cond_resched();
7075 }
7076
7077 return ret;
7078}
7079
Josef Bacik69ffb542012-09-11 15:40:07 -04007080static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7081 u64 len, u64 orig_start,
7082 u64 block_start, u64 block_len,
Josef Bacikcc95bef2013-04-04 14:31:27 -04007083 u64 orig_block_len, u64 ram_bytes,
7084 int type)
Josef Bacik69ffb542012-09-11 15:40:07 -04007085{
7086 struct extent_map_tree *em_tree;
7087 struct extent_map *em;
7088 struct btrfs_root *root = BTRFS_I(inode)->root;
7089 int ret;
7090
7091 em_tree = &BTRFS_I(inode)->extent_tree;
7092 em = alloc_extent_map();
7093 if (!em)
7094 return ERR_PTR(-ENOMEM);
7095
7096 em->start = start;
7097 em->orig_start = orig_start;
Josef Bacik2ab28f32012-10-12 15:27:49 -04007098 em->mod_start = start;
7099 em->mod_len = len;
Josef Bacik69ffb542012-09-11 15:40:07 -04007100 em->len = len;
7101 em->block_len = block_len;
7102 em->block_start = block_start;
7103 em->bdev = root->fs_info->fs_devices->latest_bdev;
Josef Bacikb4939682012-12-03 10:31:19 -05007104 em->orig_block_len = orig_block_len;
Josef Bacikcc95bef2013-04-04 14:31:27 -04007105 em->ram_bytes = ram_bytes;
Josef Bacik70c8a912012-10-11 16:54:30 -04007106 em->generation = -1;
Josef Bacik69ffb542012-09-11 15:40:07 -04007107 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7108 if (type == BTRFS_ORDERED_PREALLOC)
Josef Bacikb11e2342012-12-03 10:58:15 -05007109 set_bit(EXTENT_FLAG_FILLING, &em->flags);
Josef Bacik69ffb542012-09-11 15:40:07 -04007110
7111 do {
7112 btrfs_drop_extent_cache(inode, em->start,
7113 em->start + em->len - 1, 0);
7114 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -04007115 ret = add_extent_mapping(em_tree, em, 1);
Josef Bacik69ffb542012-09-11 15:40:07 -04007116 write_unlock(&em_tree->lock);
7117 } while (ret == -EEXIST);
7118
7119 if (ret) {
7120 free_extent_map(em);
7121 return ERR_PTR(ret);
7122 }
7123
7124 return em;
7125}
7126
7127
Josef Bacik4b46fce2010-05-23 11:00:55 -04007128static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7129 struct buffer_head *bh_result, int create)
7130{
7131 struct extent_map *em;
7132 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacikeb838e72012-07-31 16:28:48 -04007133 struct extent_state *cached_state = NULL;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007134 u64 start = iblock << inode->i_blkbits;
Josef Bacikeb838e72012-07-31 16:28:48 -04007135 u64 lockstart, lockend;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007136 u64 len = bh_result->b_size;
Josef Bacikeb838e72012-07-31 16:28:48 -04007137 int unlock_bits = EXTENT_LOCKED;
Miao Xie09348562013-02-07 10:12:07 +00007138 int ret = 0;
Josef Bacikeb838e72012-07-31 16:28:48 -04007139
Miao Xie172a5042013-02-21 02:48:22 -07007140 if (create)
Josef Bacikeb838e72012-07-31 16:28:48 -04007141 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
Miao Xie172a5042013-02-21 02:48:22 -07007142 else
Josef Bacikc3298612012-08-03 16:49:19 -04007143 len = min_t(u64, len, root->sectorsize);
Josef Bacikeb838e72012-07-31 16:28:48 -04007144
Josef Bacikc3298612012-08-03 16:49:19 -04007145 lockstart = start;
7146 lockend = start + len - 1;
7147
Josef Bacikeb838e72012-07-31 16:28:48 -04007148 /*
7149 * If this errors out it's because we couldn't invalidate pagecache for
7150 * this range and we need to fallback to buffered.
7151 */
7152 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
7153 return -ENOTBLK;
7154
Josef Bacik4b46fce2010-05-23 11:00:55 -04007155 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
Josef Bacikeb838e72012-07-31 16:28:48 -04007156 if (IS_ERR(em)) {
7157 ret = PTR_ERR(em);
7158 goto unlock_err;
7159 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04007160
7161 /*
7162 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7163 * io. INLINE is special, and we could probably kludge it in here, but
7164 * it's still buffered so for safety lets just fall back to the generic
7165 * buffered path.
7166 *
7167 * For COMPRESSED we _have_ to read the entire extent in so we can
7168 * decompress it, so there will be buffering required no matter what we
7169 * do, so go ahead and fallback to buffered.
7170 *
7171 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7172 * to buffered IO. Don't blame me, this is the price we pay for using
7173 * the generic code.
7174 */
7175 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7176 em->block_start == EXTENT_MAP_INLINE) {
7177 free_extent_map(em);
Josef Bacikeb838e72012-07-31 16:28:48 -04007178 ret = -ENOTBLK;
7179 goto unlock_err;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007180 }
7181
7182 /* Just a good old fashioned hole, return */
7183 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7184 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7185 free_extent_map(em);
Josef Bacikeb838e72012-07-31 16:28:48 -04007186 goto unlock_err;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007187 }
7188
7189 /*
7190 * We don't allocate a new extent in the following cases
7191 *
7192 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7193 * existing extent.
7194 * 2) The extent is marked as PREALLOC. We're good to go here and can
7195 * just use the extent.
7196 *
7197 */
Chris Mason46bfbb52010-05-26 11:04:10 -04007198 if (!create) {
Josef Bacikeb838e72012-07-31 16:28:48 -04007199 len = min(len, em->len - (start - em->start));
7200 lockstart = start + len;
7201 goto unlock;
Chris Mason46bfbb52010-05-26 11:04:10 -04007202 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04007203
7204 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7205 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7206 em->block_start != EXTENT_MAP_HOLE)) {
Josef Bacik4b46fce2010-05-23 11:00:55 -04007207 int type;
7208 int ret;
Josef Bacikeb384b52013-04-24 16:32:55 -04007209 u64 block_start, orig_start, orig_block_len, ram_bytes;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007210
7211 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7212 type = BTRFS_ORDERED_PREALLOC;
7213 else
7214 type = BTRFS_ORDERED_NOCOW;
Chris Mason46bfbb52010-05-26 11:04:10 -04007215 len = min(len, em->len - (start - em->start));
Josef Bacik4b46fce2010-05-23 11:00:55 -04007216 block_start = em->block_start + (start - em->start);
Chris Mason46bfbb52010-05-26 11:04:10 -04007217
Josef Bacik00361582013-08-14 14:02:47 -04007218 if (can_nocow_extent(inode, start, &len, &orig_start,
Josef Bacik7ee9e442013-06-21 16:37:03 -04007219 &orig_block_len, &ram_bytes) == 1) {
Josef Bacik69ffb542012-09-11 15:40:07 -04007220 if (type == BTRFS_ORDERED_PREALLOC) {
7221 free_extent_map(em);
7222 em = create_pinned_em(inode, start, len,
7223 orig_start,
Josef Bacikb4939682012-12-03 10:31:19 -05007224 block_start, len,
Josef Bacikcc95bef2013-04-04 14:31:27 -04007225 orig_block_len,
7226 ram_bytes, type);
Filipe Manana555e1282014-07-07 12:35:21 +01007227 if (IS_ERR(em)) {
7228 ret = PTR_ERR(em);
Josef Bacik69ffb542012-09-11 15:40:07 -04007229 goto unlock_err;
Filipe Manana555e1282014-07-07 12:35:21 +01007230 }
Josef Bacik69ffb542012-09-11 15:40:07 -04007231 }
7232
Chris Mason46bfbb52010-05-26 11:04:10 -04007233 ret = btrfs_add_ordered_extent_dio(inode, start,
7234 block_start, len, len, type);
Chris Mason46bfbb52010-05-26 11:04:10 -04007235 if (ret) {
7236 free_extent_map(em);
Josef Bacikeb838e72012-07-31 16:28:48 -04007237 goto unlock_err;
Chris Mason46bfbb52010-05-26 11:04:10 -04007238 }
7239 goto unlock;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007240 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04007241 }
Josef Bacik00361582013-08-14 14:02:47 -04007242
Chris Mason46bfbb52010-05-26 11:04:10 -04007243 /*
7244 * this will cow the extent, reset the len in case we changed
7245 * it above
7246 */
7247 len = bh_result->b_size;
Josef Bacik70c8a912012-10-11 16:54:30 -04007248 free_extent_map(em);
7249 em = btrfs_new_extent_direct(inode, start, len);
Josef Bacikeb838e72012-07-31 16:28:48 -04007250 if (IS_ERR(em)) {
7251 ret = PTR_ERR(em);
7252 goto unlock_err;
7253 }
Chris Mason46bfbb52010-05-26 11:04:10 -04007254 len = min(len, em->len - (start - em->start));
7255unlock:
Josef Bacik4b46fce2010-05-23 11:00:55 -04007256 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7257 inode->i_blkbits;
Chris Mason46bfbb52010-05-26 11:04:10 -04007258 bh_result->b_size = len;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007259 bh_result->b_bdev = em->bdev;
7260 set_buffer_mapped(bh_result);
Josef Bacikc3473e82012-06-19 10:59:00 -04007261 if (create) {
7262 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7263 set_buffer_new(bh_result);
7264
7265 /*
7266 * Need to update the i_size under the extent lock so buffered
7267 * readers will get the updated i_size when we unlock.
7268 */
7269 if (start + len > i_size_read(inode))
7270 i_size_write(inode, start + len);
Miao Xie09348562013-02-07 10:12:07 +00007271
Miao Xie172a5042013-02-21 02:48:22 -07007272 spin_lock(&BTRFS_I(inode)->lock);
7273 BTRFS_I(inode)->outstanding_extents++;
7274 spin_unlock(&BTRFS_I(inode)->lock);
7275
Miao Xie09348562013-02-07 10:12:07 +00007276 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7277 lockstart + len - 1, EXTENT_DELALLOC, NULL,
7278 &cached_state, GFP_NOFS);
7279 BUG_ON(ret);
Josef Bacikc3473e82012-06-19 10:59:00 -04007280 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04007281
Josef Bacikeb838e72012-07-31 16:28:48 -04007282 /*
7283 * In the case of write we need to clear and unlock the entire range,
7284 * in the case of read we need to unlock only the end area that we
7285 * aren't using if there is any left over space.
7286 */
Liu Bo24c03fa2012-08-22 20:10:38 -06007287 if (lockstart < lockend) {
Miao Xie09348562013-02-07 10:12:07 +00007288 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7289 lockend, unlock_bits, 1, 0,
7290 &cached_state, GFP_NOFS);
Liu Bo24c03fa2012-08-22 20:10:38 -06007291 } else {
Josef Bacikeb838e72012-07-31 16:28:48 -04007292 free_extent_state(cached_state);
Liu Bo24c03fa2012-08-22 20:10:38 -06007293 }
Josef Bacikeb838e72012-07-31 16:28:48 -04007294
Josef Bacik4b46fce2010-05-23 11:00:55 -04007295 free_extent_map(em);
7296
7297 return 0;
Josef Bacikeb838e72012-07-31 16:28:48 -04007298
7299unlock_err:
Josef Bacikeb838e72012-07-31 16:28:48 -04007300 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7301 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7302 return ret;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007303}
7304
Miao Xie8b110e32014-09-12 18:44:03 +08007305static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7306 int rw, int mirror_num)
7307{
7308 struct btrfs_root *root = BTRFS_I(inode)->root;
7309 int ret;
7310
7311 BUG_ON(rw & REQ_WRITE);
7312
7313 bio_get(bio);
7314
7315 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7316 BTRFS_WQ_ENDIO_DIO_REPAIR);
7317 if (ret)
7318 goto err;
7319
7320 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7321err:
7322 bio_put(bio);
7323 return ret;
7324}
7325
7326static int btrfs_check_dio_repairable(struct inode *inode,
7327 struct bio *failed_bio,
7328 struct io_failure_record *failrec,
7329 int failed_mirror)
7330{
7331 int num_copies;
7332
7333 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7334 failrec->logical, failrec->len);
7335 if (num_copies == 1) {
7336 /*
7337 * we only have a single copy of the data, so don't bother with
7338 * all the retry and error correction code that follows. no
7339 * matter what the error is, it is very likely to persist.
7340 */
7341 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7342 num_copies, failrec->this_mirror, failed_mirror);
7343 return 0;
7344 }
7345
7346 failrec->failed_mirror = failed_mirror;
7347 failrec->this_mirror++;
7348 if (failrec->this_mirror == failed_mirror)
7349 failrec->this_mirror++;
7350
7351 if (failrec->this_mirror > num_copies) {
7352 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7353 num_copies, failrec->this_mirror, failed_mirror);
7354 return 0;
7355 }
7356
7357 return 1;
7358}
7359
7360static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7361 struct page *page, u64 start, u64 end,
7362 int failed_mirror, bio_end_io_t *repair_endio,
7363 void *repair_arg)
7364{
7365 struct io_failure_record *failrec;
7366 struct bio *bio;
7367 int isector;
7368 int read_mode;
7369 int ret;
7370
7371 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7372
7373 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7374 if (ret)
7375 return ret;
7376
7377 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7378 failed_mirror);
7379 if (!ret) {
7380 free_io_failure(inode, failrec);
7381 return -EIO;
7382 }
7383
7384 if (failed_bio->bi_vcnt > 1)
7385 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7386 else
7387 read_mode = READ_SYNC;
7388
7389 isector = start - btrfs_io_bio(failed_bio)->logical;
7390 isector >>= inode->i_sb->s_blocksize_bits;
7391 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7392 0, isector, repair_endio, repair_arg);
7393 if (!bio) {
7394 free_io_failure(inode, failrec);
7395 return -EIO;
7396 }
7397
7398 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7399 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7400 read_mode, failrec->this_mirror, failrec->in_validation);
7401
7402 ret = submit_dio_repair_bio(inode, bio, read_mode,
7403 failrec->this_mirror);
7404 if (ret) {
7405 free_io_failure(inode, failrec);
7406 bio_put(bio);
7407 }
7408
7409 return ret;
7410}
7411
7412struct btrfs_retry_complete {
7413 struct completion done;
7414 struct inode *inode;
7415 u64 start;
7416 int uptodate;
7417};
7418
7419static void btrfs_retry_endio_nocsum(struct bio *bio, int err)
7420{
7421 struct btrfs_retry_complete *done = bio->bi_private;
7422 struct bio_vec *bvec;
7423 int i;
7424
7425 if (err)
7426 goto end;
7427
7428 done->uptodate = 1;
7429 bio_for_each_segment_all(bvec, bio, i)
7430 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7431end:
7432 complete(&done->done);
7433 bio_put(bio);
7434}
7435
7436static int __btrfs_correct_data_nocsum(struct inode *inode,
7437 struct btrfs_io_bio *io_bio)
Josef Bacik4b46fce2010-05-23 11:00:55 -04007438{
Kent Overstreet2c30c712013-11-07 12:20:26 -08007439 struct bio_vec *bvec;
Miao Xie8b110e32014-09-12 18:44:03 +08007440 struct btrfs_retry_complete done;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007441 u64 start;
Kent Overstreet2c30c712013-11-07 12:20:26 -08007442 int i;
Miao Xiec1dc0892014-09-12 18:43:56 +08007443 int ret;
Miao Xiedc380ae2014-09-12 18:43:55 +08007444
Miao Xiec1dc0892014-09-12 18:43:56 +08007445 start = io_bio->logical;
Miao Xie8b110e32014-09-12 18:44:03 +08007446 done.inode = inode;
7447
7448 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7449try_again:
7450 done.uptodate = 0;
7451 done.start = start;
7452 init_completion(&done.done);
7453
7454 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7455 start + bvec->bv_len - 1,
7456 io_bio->mirror_num,
7457 btrfs_retry_endio_nocsum, &done);
7458 if (ret)
7459 return ret;
7460
7461 wait_for_completion(&done.done);
7462
7463 if (!done.uptodate) {
7464 /* We might have another mirror, so try again */
7465 goto try_again;
7466 }
7467
7468 start += bvec->bv_len;
7469 }
7470
7471 return 0;
7472}
7473
7474static void btrfs_retry_endio(struct bio *bio, int err)
7475{
7476 struct btrfs_retry_complete *done = bio->bi_private;
7477 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7478 struct bio_vec *bvec;
7479 int uptodate;
7480 int ret;
7481 int i;
7482
7483 if (err)
7484 goto end;
7485
7486 uptodate = 1;
7487 bio_for_each_segment_all(bvec, bio, i) {
7488 ret = __readpage_endio_check(done->inode, io_bio, i,
7489 bvec->bv_page, 0,
7490 done->start, bvec->bv_len);
7491 if (!ret)
7492 clean_io_failure(done->inode, done->start,
7493 bvec->bv_page, 0);
7494 else
7495 uptodate = 0;
7496 }
7497
7498 done->uptodate = uptodate;
7499end:
7500 complete(&done->done);
7501 bio_put(bio);
7502}
7503
7504static int __btrfs_subio_endio_read(struct inode *inode,
7505 struct btrfs_io_bio *io_bio, int err)
7506{
7507 struct bio_vec *bvec;
7508 struct btrfs_retry_complete done;
7509 u64 start;
7510 u64 offset = 0;
7511 int i;
7512 int ret;
7513
7514 err = 0;
7515 start = io_bio->logical;
7516 done.inode = inode;
7517
Miao Xiec1dc0892014-09-12 18:43:56 +08007518 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
Miao Xiedc380ae2014-09-12 18:43:55 +08007519 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7520 0, start, bvec->bv_len);
Miao Xie8b110e32014-09-12 18:44:03 +08007521 if (likely(!ret))
7522 goto next;
7523try_again:
7524 done.uptodate = 0;
7525 done.start = start;
7526 init_completion(&done.done);
7527
7528 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7529 start + bvec->bv_len - 1,
7530 io_bio->mirror_num,
7531 btrfs_retry_endio, &done);
7532 if (ret) {
7533 err = ret;
7534 goto next;
7535 }
7536
7537 wait_for_completion(&done.done);
7538
7539 if (!done.uptodate) {
7540 /* We might have another mirror, so try again */
7541 goto try_again;
7542 }
7543next:
7544 offset += bvec->bv_len;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007545 start += bvec->bv_len;
Kent Overstreet2c30c712013-11-07 12:20:26 -08007546 }
Miao Xiec1dc0892014-09-12 18:43:56 +08007547
7548 return err;
7549}
7550
Miao Xie8b110e32014-09-12 18:44:03 +08007551static int btrfs_subio_endio_read(struct inode *inode,
7552 struct btrfs_io_bio *io_bio, int err)
7553{
7554 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7555
7556 if (skip_csum) {
7557 if (unlikely(err))
7558 return __btrfs_correct_data_nocsum(inode, io_bio);
7559 else
7560 return 0;
7561 } else {
7562 return __btrfs_subio_endio_read(inode, io_bio, err);
7563 }
7564}
7565
Miao Xiec1dc0892014-09-12 18:43:56 +08007566static void btrfs_endio_direct_read(struct bio *bio, int err)
7567{
7568 struct btrfs_dio_private *dip = bio->bi_private;
7569 struct inode *inode = dip->inode;
7570 struct bio *dio_bio;
7571 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7572
Miao Xie8b110e32014-09-12 18:44:03 +08007573 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
7574 err = btrfs_subio_endio_read(inode, io_bio, err);
Miao Xiec1dc0892014-09-12 18:43:56 +08007575
Josef Bacik4b46fce2010-05-23 11:00:55 -04007576 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01007577 dip->logical_offset + dip->bytes - 1);
Chris Mason9be33952013-05-17 18:30:14 -04007578 dio_bio = dip->dio_bio;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007579
Josef Bacik4b46fce2010-05-23 11:00:55 -04007580 kfree(dip);
Josef Bacikc0da7aa2011-03-22 11:05:07 -04007581
7582 /* If we had a csum failure make sure to clear the uptodate flag */
7583 if (err)
Chris Mason9be33952013-05-17 18:30:14 -04007584 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7585 dio_end_io(dio_bio, err);
Miao Xie23ea8e52014-09-12 18:43:54 +08007586
7587 if (io_bio->end_io)
7588 io_bio->end_io(io_bio, err);
Chris Mason9be33952013-05-17 18:30:14 -04007589 bio_put(bio);
Josef Bacik4b46fce2010-05-23 11:00:55 -04007590}
7591
7592static void btrfs_endio_direct_write(struct bio *bio, int err)
7593{
7594 struct btrfs_dio_private *dip = bio->bi_private;
7595 struct inode *inode = dip->inode;
7596 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007597 struct btrfs_ordered_extent *ordered = NULL;
Chris Mason163cf092010-11-28 19:56:33 -05007598 u64 ordered_offset = dip->logical_offset;
7599 u64 ordered_bytes = dip->bytes;
Chris Mason9be33952013-05-17 18:30:14 -04007600 struct bio *dio_bio;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007601 int ret;
7602
7603 if (err)
7604 goto out_done;
Chris Mason163cf092010-11-28 19:56:33 -05007605again:
7606 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7607 &ordered_offset,
Josef Bacik5fd02042012-05-02 14:00:54 -04007608 ordered_bytes, !err);
Josef Bacik4b46fce2010-05-23 11:00:55 -04007609 if (!ret)
Chris Mason163cf092010-11-28 19:56:33 -05007610 goto out_test;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007611
Liu Bo9e0af232014-08-15 23:36:53 +08007612 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
7613 finish_ordered_fn, NULL, NULL);
Qu Wenruofccb5d82014-02-28 10:46:10 +08007614 btrfs_queue_work(root->fs_info->endio_write_workers,
7615 &ordered->work);
Chris Mason163cf092010-11-28 19:56:33 -05007616out_test:
7617 /*
7618 * our bio might span multiple ordered extents. If we haven't
7619 * completed the accounting for the whole dio, go back and try again
7620 */
7621 if (ordered_offset < dip->logical_offset + dip->bytes) {
7622 ordered_bytes = dip->logical_offset + dip->bytes -
7623 ordered_offset;
Josef Bacik5fd02042012-05-02 14:00:54 -04007624 ordered = NULL;
Chris Mason163cf092010-11-28 19:56:33 -05007625 goto again;
7626 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04007627out_done:
Chris Mason9be33952013-05-17 18:30:14 -04007628 dio_bio = dip->dio_bio;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007629
Josef Bacik4b46fce2010-05-23 11:00:55 -04007630 kfree(dip);
Josef Bacikc0da7aa2011-03-22 11:05:07 -04007631
7632 /* If we had an error make sure to clear the uptodate flag */
7633 if (err)
Chris Mason9be33952013-05-17 18:30:14 -04007634 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7635 dio_end_io(dio_bio, err);
7636 bio_put(bio);
Josef Bacik4b46fce2010-05-23 11:00:55 -04007637}
7638
Chris Masoneaf25d92010-05-25 09:48:28 -04007639static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7640 struct bio *bio, int mirror_num,
7641 unsigned long bio_flags, u64 offset)
7642{
7643 int ret;
7644 struct btrfs_root *root = BTRFS_I(inode)->root;
7645 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007646 BUG_ON(ret); /* -ENOMEM */
Chris Masoneaf25d92010-05-25 09:48:28 -04007647 return 0;
7648}
7649
Miao Xiee65e1532010-11-22 03:04:43 +00007650static void btrfs_end_dio_bio(struct bio *bio, int err)
7651{
7652 struct btrfs_dio_private *dip = bio->bi_private;
7653
Miao Xie8b110e32014-09-12 18:44:03 +08007654 if (err)
7655 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7656 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7657 btrfs_ino(dip->inode), bio->bi_rw,
7658 (unsigned long long)bio->bi_iter.bi_sector,
7659 bio->bi_iter.bi_size, err);
7660
7661 if (dip->subio_endio)
7662 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
Miao Xiec1dc0892014-09-12 18:43:56 +08007663
7664 if (err) {
Miao Xiee65e1532010-11-22 03:04:43 +00007665 dip->errors = 1;
7666
7667 /*
7668 * before atomic variable goto zero, we must make sure
7669 * dip->errors is perceived to be set.
7670 */
Peter Zijlstra4e857c52014-03-17 18:06:10 +01007671 smp_mb__before_atomic();
Miao Xiee65e1532010-11-22 03:04:43 +00007672 }
7673
7674 /* if there are more bios still pending for this dio, just exit */
7675 if (!atomic_dec_and_test(&dip->pending_bios))
7676 goto out;
7677
Chris Mason9be33952013-05-17 18:30:14 -04007678 if (dip->errors) {
Miao Xiee65e1532010-11-22 03:04:43 +00007679 bio_io_error(dip->orig_bio);
Chris Mason9be33952013-05-17 18:30:14 -04007680 } else {
7681 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
Miao Xiee65e1532010-11-22 03:04:43 +00007682 bio_endio(dip->orig_bio, 0);
7683 }
7684out:
7685 bio_put(bio);
7686}
7687
7688static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7689 u64 first_sector, gfp_t gfp_flags)
7690{
7691 int nr_vecs = bio_get_nr_vecs(bdev);
7692 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7693}
7694
Miao Xiec1dc0892014-09-12 18:43:56 +08007695static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
7696 struct inode *inode,
7697 struct btrfs_dio_private *dip,
7698 struct bio *bio,
7699 u64 file_offset)
7700{
7701 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7702 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
7703 int ret;
7704
7705 /*
7706 * We load all the csum data we need when we submit
7707 * the first bio to reduce the csum tree search and
7708 * contention.
7709 */
7710 if (dip->logical_offset == file_offset) {
7711 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
7712 file_offset);
7713 if (ret)
7714 return ret;
7715 }
7716
7717 if (bio == dip->orig_bio)
7718 return 0;
7719
7720 file_offset -= dip->logical_offset;
7721 file_offset >>= inode->i_sb->s_blocksize_bits;
7722 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
7723
7724 return 0;
7725}
7726
Miao Xiee65e1532010-11-22 03:04:43 +00007727static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7728 int rw, u64 file_offset, int skip_sum,
Josef Bacikc3298612012-08-03 16:49:19 -04007729 int async_submit)
Miao Xiee65e1532010-11-22 03:04:43 +00007730{
Miao Xiefacc8a222013-07-25 19:22:34 +08007731 struct btrfs_dio_private *dip = bio->bi_private;
Miao Xiee65e1532010-11-22 03:04:43 +00007732 int write = rw & REQ_WRITE;
7733 struct btrfs_root *root = BTRFS_I(inode)->root;
7734 int ret;
7735
Josef Bacikb812ce22012-11-16 13:56:32 -05007736 if (async_submit)
7737 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7738
Miao Xiee65e1532010-11-22 03:04:43 +00007739 bio_get(bio);
Josef Bacik5fd02042012-05-02 14:00:54 -04007740
7741 if (!write) {
David Sterbabfebd8b2014-07-30 00:25:45 +02007742 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7743 BTRFS_WQ_ENDIO_DATA);
Josef Bacik5fd02042012-05-02 14:00:54 -04007744 if (ret)
7745 goto err;
7746 }
Miao Xiee65e1532010-11-22 03:04:43 +00007747
Josef Bacik1ae39932011-04-06 14:41:34 -04007748 if (skip_sum)
7749 goto map;
7750
7751 if (write && async_submit) {
Miao Xiee65e1532010-11-22 03:04:43 +00007752 ret = btrfs_wq_submit_bio(root->fs_info,
7753 inode, rw, bio, 0, 0,
7754 file_offset,
7755 __btrfs_submit_bio_start_direct_io,
7756 __btrfs_submit_bio_done);
7757 goto err;
Josef Bacik1ae39932011-04-06 14:41:34 -04007758 } else if (write) {
7759 /*
7760 * If we aren't doing async submit, calculate the csum of the
7761 * bio now.
7762 */
7763 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7764 if (ret)
7765 goto err;
Miao Xie23ea8e52014-09-12 18:43:54 +08007766 } else {
Miao Xiec1dc0892014-09-12 18:43:56 +08007767 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
7768 file_offset);
Tsutomu Itohc2db1072011-03-01 06:48:31 +00007769 if (ret)
7770 goto err;
7771 }
Josef Bacik1ae39932011-04-06 14:41:34 -04007772map:
7773 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
Miao Xiee65e1532010-11-22 03:04:43 +00007774err:
7775 bio_put(bio);
7776 return ret;
7777}
7778
7779static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7780 int skip_sum)
7781{
7782 struct inode *inode = dip->inode;
7783 struct btrfs_root *root = BTRFS_I(inode)->root;
Miao Xiee65e1532010-11-22 03:04:43 +00007784 struct bio *bio;
7785 struct bio *orig_bio = dip->orig_bio;
7786 struct bio_vec *bvec = orig_bio->bi_io_vec;
Kent Overstreet4f024f32013-10-11 15:44:27 -07007787 u64 start_sector = orig_bio->bi_iter.bi_sector;
Miao Xiee65e1532010-11-22 03:04:43 +00007788 u64 file_offset = dip->logical_offset;
7789 u64 submit_len = 0;
7790 u64 map_length;
7791 int nr_pages = 0;
Miao Xie23ea8e52014-09-12 18:43:54 +08007792 int ret;
Josef Bacik1ae39932011-04-06 14:41:34 -04007793 int async_submit = 0;
Miao Xiee65e1532010-11-22 03:04:43 +00007794
Kent Overstreet4f024f32013-10-11 15:44:27 -07007795 map_length = orig_bio->bi_iter.bi_size;
David Woodhouse53b381b2013-01-29 18:40:14 -05007796 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
Miao Xiee65e1532010-11-22 03:04:43 +00007797 &map_length, NULL, 0);
Miao Xie7a5c3c92014-06-17 18:58:59 +08007798 if (ret)
Miao Xiee65e1532010-11-22 03:04:43 +00007799 return -EIO;
Miao Xiefacc8a222013-07-25 19:22:34 +08007800
Kent Overstreet4f024f32013-10-11 15:44:27 -07007801 if (map_length >= orig_bio->bi_iter.bi_size) {
Josef Bacik02f57c72011-04-06 14:25:44 -04007802 bio = orig_bio;
Miao Xiec1dc0892014-09-12 18:43:56 +08007803 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
Josef Bacik02f57c72011-04-06 14:25:44 -04007804 goto submit;
7805 }
7806
David Woodhouse53b381b2013-01-29 18:40:14 -05007807 /* async crcs make it difficult to collect full stripe writes. */
7808 if (btrfs_get_alloc_profile(root, 1) &
7809 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7810 async_submit = 0;
7811 else
7812 async_submit = 1;
7813
Josef Bacik02f57c72011-04-06 14:25:44 -04007814 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7815 if (!bio)
7816 return -ENOMEM;
Miao Xie7a5c3c92014-06-17 18:58:59 +08007817
Josef Bacik02f57c72011-04-06 14:25:44 -04007818 bio->bi_private = dip;
7819 bio->bi_end_io = btrfs_end_dio_bio;
Miao Xiec1dc0892014-09-12 18:43:56 +08007820 btrfs_io_bio(bio)->logical = file_offset;
Josef Bacik02f57c72011-04-06 14:25:44 -04007821 atomic_inc(&dip->pending_bios);
7822
Miao Xiee65e1532010-11-22 03:04:43 +00007823 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
David Sterbaee39b432014-09-30 01:33:33 +02007824 if (map_length < submit_len + bvec->bv_len ||
Miao Xiee65e1532010-11-22 03:04:43 +00007825 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
David Sterbaee39b432014-09-30 01:33:33 +02007826 bvec->bv_offset) < bvec->bv_len) {
Miao Xiee65e1532010-11-22 03:04:43 +00007827 /*
7828 * inc the count before we submit the bio so
7829 * we know the end IO handler won't happen before
7830 * we inc the count. Otherwise, the dip might get freed
7831 * before we're done setting it up
7832 */
7833 atomic_inc(&dip->pending_bios);
7834 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7835 file_offset, skip_sum,
Josef Bacikc3298612012-08-03 16:49:19 -04007836 async_submit);
Miao Xiee65e1532010-11-22 03:04:43 +00007837 if (ret) {
7838 bio_put(bio);
7839 atomic_dec(&dip->pending_bios);
7840 goto out_err;
7841 }
7842
Miao Xiee65e1532010-11-22 03:04:43 +00007843 start_sector += submit_len >> 9;
7844 file_offset += submit_len;
7845
7846 submit_len = 0;
7847 nr_pages = 0;
7848
7849 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7850 start_sector, GFP_NOFS);
7851 if (!bio)
7852 goto out_err;
7853 bio->bi_private = dip;
7854 bio->bi_end_io = btrfs_end_dio_bio;
Miao Xiec1dc0892014-09-12 18:43:56 +08007855 btrfs_io_bio(bio)->logical = file_offset;
Miao Xiee65e1532010-11-22 03:04:43 +00007856
Kent Overstreet4f024f32013-10-11 15:44:27 -07007857 map_length = orig_bio->bi_iter.bi_size;
David Woodhouse53b381b2013-01-29 18:40:14 -05007858 ret = btrfs_map_block(root->fs_info, rw,
Stefan Behrens3ec706c2012-11-05 15:46:42 +01007859 start_sector << 9,
Miao Xiee65e1532010-11-22 03:04:43 +00007860 &map_length, NULL, 0);
7861 if (ret) {
7862 bio_put(bio);
7863 goto out_err;
7864 }
7865 } else {
7866 submit_len += bvec->bv_len;
Dulshani Gunawardhana67871252013-10-31 10:33:04 +05307867 nr_pages++;
Miao Xiee65e1532010-11-22 03:04:43 +00007868 bvec++;
7869 }
7870 }
7871
Josef Bacik02f57c72011-04-06 14:25:44 -04007872submit:
Miao Xiee65e1532010-11-22 03:04:43 +00007873 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
Josef Bacikc3298612012-08-03 16:49:19 -04007874 async_submit);
Miao Xiee65e1532010-11-22 03:04:43 +00007875 if (!ret)
7876 return 0;
7877
7878 bio_put(bio);
7879out_err:
7880 dip->errors = 1;
7881 /*
7882 * before atomic variable goto zero, we must
7883 * make sure dip->errors is perceived to be set.
7884 */
Peter Zijlstra4e857c52014-03-17 18:06:10 +01007885 smp_mb__before_atomic();
Miao Xiee65e1532010-11-22 03:04:43 +00007886 if (atomic_dec_and_test(&dip->pending_bios))
7887 bio_io_error(dip->orig_bio);
7888
7889 /* bio_end_io() will handle error, so we needn't return it */
7890 return 0;
7891}
7892
Chris Mason9be33952013-05-17 18:30:14 -04007893static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7894 struct inode *inode, loff_t file_offset)
Josef Bacik4b46fce2010-05-23 11:00:55 -04007895{
7896 struct btrfs_root *root = BTRFS_I(inode)->root;
7897 struct btrfs_dio_private *dip;
Chris Mason9be33952013-05-17 18:30:14 -04007898 struct bio *io_bio;
Miao Xie23ea8e52014-09-12 18:43:54 +08007899 struct btrfs_io_bio *btrfs_bio;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007900 int skip_sum;
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02007901 int write = rw & REQ_WRITE;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007902 int ret = 0;
7903
7904 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7905
Chris Mason9be33952013-05-17 18:30:14 -04007906 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
Chris Mason9be33952013-05-17 18:30:14 -04007907 if (!io_bio) {
Josef Bacik4b46fce2010-05-23 11:00:55 -04007908 ret = -ENOMEM;
7909 goto free_ordered;
7910 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04007911
Miao Xiec1dc0892014-09-12 18:43:56 +08007912 dip = kzalloc(sizeof(*dip), GFP_NOFS);
Chris Mason9be33952013-05-17 18:30:14 -04007913 if (!dip) {
7914 ret = -ENOMEM;
7915 goto free_io_bio;
7916 }
7917
7918 dip->private = dio_bio->bi_private;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007919 dip->inode = inode;
7920 dip->logical_offset = file_offset;
Kent Overstreet4f024f32013-10-11 15:44:27 -07007921 dip->bytes = dio_bio->bi_iter.bi_size;
7922 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
Chris Mason9be33952013-05-17 18:30:14 -04007923 io_bio->bi_private = dip;
Chris Mason9be33952013-05-17 18:30:14 -04007924 dip->orig_bio = io_bio;
7925 dip->dio_bio = dio_bio;
Miao Xiee65e1532010-11-22 03:04:43 +00007926 atomic_set(&dip->pending_bios, 0);
Miao Xiec1dc0892014-09-12 18:43:56 +08007927 btrfs_bio = btrfs_io_bio(io_bio);
7928 btrfs_bio->logical = file_offset;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007929
Miao Xiec1dc0892014-09-12 18:43:56 +08007930 if (write) {
Chris Mason9be33952013-05-17 18:30:14 -04007931 io_bio->bi_end_io = btrfs_endio_direct_write;
Miao Xiec1dc0892014-09-12 18:43:56 +08007932 } else {
Chris Mason9be33952013-05-17 18:30:14 -04007933 io_bio->bi_end_io = btrfs_endio_direct_read;
Miao Xiec1dc0892014-09-12 18:43:56 +08007934 dip->subio_endio = btrfs_subio_endio_read;
7935 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04007936
Miao Xiee65e1532010-11-22 03:04:43 +00007937 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7938 if (!ret)
Chris Masoneaf25d92010-05-25 09:48:28 -04007939 return;
Chris Mason9be33952013-05-17 18:30:14 -04007940
Miao Xie23ea8e52014-09-12 18:43:54 +08007941 if (btrfs_bio->end_io)
7942 btrfs_bio->end_io(btrfs_bio, ret);
Chris Mason9be33952013-05-17 18:30:14 -04007943free_io_bio:
7944 bio_put(io_bio);
7945
Josef Bacik4b46fce2010-05-23 11:00:55 -04007946free_ordered:
7947 /*
7948 * If this is a write, we need to clean up the reserved space and kill
7949 * the ordered extent.
7950 */
7951 if (write) {
7952 struct btrfs_ordered_extent *ordered;
Josef Bacik955256f2010-11-19 09:41:10 -05007953 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
Josef Bacik4b46fce2010-05-23 11:00:55 -04007954 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7955 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7956 btrfs_free_reserved_extent(root, ordered->start,
Miao Xiee570fd22014-06-19 10:42:50 +08007957 ordered->disk_len, 1);
Josef Bacik4b46fce2010-05-23 11:00:55 -04007958 btrfs_put_ordered_extent(ordered);
7959 btrfs_put_ordered_extent(ordered);
7960 }
Chris Mason9be33952013-05-17 18:30:14 -04007961 bio_endio(dio_bio, ret);
Josef Bacik4b46fce2010-05-23 11:00:55 -04007962}
7963
Chris Mason5a5f79b2010-05-26 21:33:37 -04007964static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
Al Viro28060d52014-03-22 05:15:17 -04007965 const struct iov_iter *iter, loff_t offset)
Chris Mason5a5f79b2010-05-26 21:33:37 -04007966{
7967 int seg;
Josef Bacika1b75f72011-04-08 15:51:18 +00007968 int i;
Chris Mason5a5f79b2010-05-26 21:33:37 -04007969 unsigned blocksize_mask = root->sectorsize - 1;
7970 ssize_t retval = -EINVAL;
Chris Mason5a5f79b2010-05-26 21:33:37 -04007971
7972 if (offset & blocksize_mask)
7973 goto out;
7974
Al Viro28060d52014-03-22 05:15:17 -04007975 if (iov_iter_alignment(iter) & blocksize_mask)
7976 goto out;
Josef Bacika1b75f72011-04-08 15:51:18 +00007977
Al Viro28060d52014-03-22 05:15:17 -04007978 /* If this is a write we don't need to check anymore */
7979 if (rw & WRITE)
7980 return 0;
7981 /*
7982 * Check to make sure we don't have duplicate iov_base's in this
7983 * iovec, if so return EINVAL, otherwise we'll get csum errors
7984 * when reading back.
7985 */
7986 for (seg = 0; seg < iter->nr_segs; seg++) {
7987 for (i = seg + 1; i < iter->nr_segs; i++) {
7988 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
Josef Bacika1b75f72011-04-08 15:51:18 +00007989 goto out;
7990 }
Chris Mason5a5f79b2010-05-26 21:33:37 -04007991 }
7992 retval = 0;
7993out:
7994 return retval;
7995}
Josef Bacikeb838e72012-07-31 16:28:48 -04007996
Chris Mason16432982008-04-10 10:23:21 -04007997static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
Al Virod8d3d942014-03-04 21:27:34 -05007998 struct iov_iter *iter, loff_t offset)
Chris Mason16432982008-04-10 10:23:21 -04007999{
Josef Bacik4b46fce2010-05-23 11:00:55 -04008000 struct file *file = iocb->ki_filp;
8001 struct inode *inode = file->f_mapping->host;
Miao Xie09348562013-02-07 10:12:07 +00008002 size_t count = 0;
Miao Xie2e60a512013-02-08 07:01:08 +00008003 int flags = 0;
Miao Xie38851cc2013-02-08 07:04:11 +00008004 bool wakeup = true;
8005 bool relock = false;
Miao Xie09348562013-02-07 10:12:07 +00008006 ssize_t ret;
Josef Bacik4b46fce2010-05-23 11:00:55 -04008007
Al Viro28060d52014-03-22 05:15:17 -04008008 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iter, offset))
Chris Mason5a5f79b2010-05-26 21:33:37 -04008009 return 0;
Chris Mason5a5f79b2010-05-26 21:33:37 -04008010
Miao Xie38851cc2013-02-08 07:04:11 +00008011 atomic_inc(&inode->i_dio_count);
Peter Zijlstra4e857c52014-03-17 18:06:10 +01008012 smp_mb__after_atomic();
Miao Xie38851cc2013-02-08 07:04:11 +00008013
Josef Bacik0e267c42013-07-02 10:38:02 -04008014 /*
Miao Xie41bd9ca2014-03-06 13:54:57 +08008015 * The generic stuff only does filemap_write_and_wait_range, which
8016 * isn't enough if we've written compressed pages to this area, so
8017 * we need to flush the dirty pages again to make absolutely sure
8018 * that any outstanding dirty pages are on disk.
Josef Bacik0e267c42013-07-02 10:38:02 -04008019 */
Al Viroa6cbcd42014-03-04 22:38:00 -05008020 count = iov_iter_count(iter);
Miao Xie41bd9ca2014-03-06 13:54:57 +08008021 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8022 &BTRFS_I(inode)->runtime_flags))
Wang Shilong9a025a02014-07-17 11:44:13 +08008023 filemap_fdatawrite_range(inode->i_mapping, offset,
8024 offset + count - 1);
Josef Bacik0e267c42013-07-02 10:38:02 -04008025
Miao Xie09348562013-02-07 10:12:07 +00008026 if (rw & WRITE) {
Miao Xie38851cc2013-02-08 07:04:11 +00008027 /*
8028 * If the write DIO is beyond the EOF, we need update
8029 * the isize, but it is protected by i_mutex. So we can
8030 * not unlock the i_mutex at this case.
8031 */
8032 if (offset + count <= inode->i_size) {
8033 mutex_unlock(&inode->i_mutex);
8034 relock = true;
8035 }
Miao Xie09348562013-02-07 10:12:07 +00008036 ret = btrfs_delalloc_reserve_space(inode, count);
8037 if (ret)
Miao Xie38851cc2013-02-08 07:04:11 +00008038 goto out;
David Sterbaee39b432014-09-30 01:33:33 +02008039 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8040 &BTRFS_I(inode)->runtime_flags)) {
Miao Xie38851cc2013-02-08 07:04:11 +00008041 inode_dio_done(inode);
8042 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8043 wakeup = false;
Miao Xie09348562013-02-07 10:12:07 +00008044 }
8045
8046 ret = __blockdev_direct_IO(rw, iocb, inode,
8047 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
Al Viro31b14032014-03-05 01:33:16 -05008048 iter, offset, btrfs_get_blocks_direct, NULL,
Miao Xie2e60a512013-02-08 07:01:08 +00008049 btrfs_submit_direct, flags);
Miao Xie09348562013-02-07 10:12:07 +00008050 if (rw & WRITE) {
8051 if (ret < 0 && ret != -EIOCBQUEUED)
8052 btrfs_delalloc_release_space(inode, count);
Miao Xie172a5042013-02-21 02:48:22 -07008053 else if (ret >= 0 && (size_t)ret < count)
Miao Xie09348562013-02-07 10:12:07 +00008054 btrfs_delalloc_release_space(inode,
8055 count - (size_t)ret);
Miao Xie172a5042013-02-21 02:48:22 -07008056 else
8057 btrfs_delalloc_release_metadata(inode, 0);
Miao Xie09348562013-02-07 10:12:07 +00008058 }
Miao Xie38851cc2013-02-08 07:04:11 +00008059out:
Miao Xie2e60a512013-02-08 07:01:08 +00008060 if (wakeup)
8061 inode_dio_done(inode);
Miao Xie38851cc2013-02-08 07:04:11 +00008062 if (relock)
8063 mutex_lock(&inode->i_mutex);
Miao Xie09348562013-02-07 10:12:07 +00008064
8065 return ret;
Chris Mason16432982008-04-10 10:23:21 -04008066}
8067
Tsutomu Itoh05dadc02012-11-29 05:08:26 +00008068#define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8069
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -05008070static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8071 __u64 start, __u64 len)
8072{
Tsutomu Itoh05dadc02012-11-29 05:08:26 +00008073 int ret;
8074
8075 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8076 if (ret)
8077 return ret;
8078
Chris Masonec29ed52011-02-23 16:23:20 -05008079 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -05008080}
8081
Chris Mason9ebefb182007-06-15 13:50:00 -04008082int btrfs_readpage(struct file *file, struct page *page)
8083{
Chris Masond1310b22008-01-24 16:13:08 -05008084 struct extent_io_tree *tree;
8085 tree = &BTRFS_I(page->mapping->host)->io_tree;
Jan Schmidt8ddc7d92011-06-13 20:02:58 +02008086 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04008087}
Chris Mason1832a6d2007-12-21 16:27:21 -05008088
Chris Mason39279cc2007-06-12 06:35:45 -04008089static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8090{
Chris Masond1310b22008-01-24 16:13:08 -05008091 struct extent_io_tree *tree;
Chris Masonb888db22007-08-27 16:49:44 -04008092
8093
8094 if (current->flags & PF_MEMALLOC) {
8095 redirty_page_for_writepage(wbc, page);
8096 unlock_page(page);
8097 return 0;
8098 }
Chris Masond1310b22008-01-24 16:13:08 -05008099 tree = &BTRFS_I(page->mapping->host)->io_tree;
Chris Masona52d9a82007-08-27 16:49:44 -04008100 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8101}
Chris Mason39279cc2007-06-12 06:35:45 -04008102
Eric Sandeen48a3b632013-04-25 20:41:01 +00008103static int btrfs_writepages(struct address_space *mapping,
8104 struct writeback_control *wbc)
Chris Masonb293f022007-11-01 19:45:34 -04008105{
Chris Masond1310b22008-01-24 16:13:08 -05008106 struct extent_io_tree *tree;
Chris Mason771ed682008-11-06 22:02:51 -05008107
Chris Masond1310b22008-01-24 16:13:08 -05008108 tree = &BTRFS_I(mapping->host)->io_tree;
Chris Masonb293f022007-11-01 19:45:34 -04008109 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8110}
8111
Chris Mason3ab2fb52007-11-08 10:59:22 -05008112static int
8113btrfs_readpages(struct file *file, struct address_space *mapping,
8114 struct list_head *pages, unsigned nr_pages)
8115{
Chris Masond1310b22008-01-24 16:13:08 -05008116 struct extent_io_tree *tree;
8117 tree = &BTRFS_I(mapping->host)->io_tree;
Chris Mason3ab2fb52007-11-08 10:59:22 -05008118 return extent_readpages(tree, mapping, pages, nr_pages,
8119 btrfs_get_extent);
8120}
Chris Masone6dcd2d2008-07-17 12:53:50 -04008121static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
Chris Masona52d9a82007-08-27 16:49:44 -04008122{
Chris Masond1310b22008-01-24 16:13:08 -05008123 struct extent_io_tree *tree;
8124 struct extent_map_tree *map;
Chris Masona52d9a82007-08-27 16:49:44 -04008125 int ret;
Chris Mason39279cc2007-06-12 06:35:45 -04008126
Chris Masond1310b22008-01-24 16:13:08 -05008127 tree = &BTRFS_I(page->mapping->host)->io_tree;
8128 map = &BTRFS_I(page->mapping->host)->extent_tree;
Chris Mason70dec802008-01-29 09:59:12 -05008129 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
Chris Masona52d9a82007-08-27 16:49:44 -04008130 if (ret == 1) {
8131 ClearPagePrivate(page);
8132 set_page_private(page, 0);
8133 page_cache_release(page);
8134 }
8135 return ret;
8136}
Chris Mason39279cc2007-06-12 06:35:45 -04008137
Chris Masone6dcd2d2008-07-17 12:53:50 -04008138static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8139{
Chris Mason98509cf2008-09-11 15:51:43 -04008140 if (PageWriteback(page) || PageDirty(page))
8141 return 0;
Yan Zhengb335b002009-02-12 10:06:04 -05008142 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04008143}
8144
Lukas Czernerd47992f2013-05-21 23:17:23 -04008145static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8146 unsigned int length)
Chris Masona52d9a82007-08-27 16:49:44 -04008147{
Josef Bacik5fd02042012-05-02 14:00:54 -04008148 struct inode *inode = page->mapping->host;
Chris Masond1310b22008-01-24 16:13:08 -05008149 struct extent_io_tree *tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -04008150 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00008151 struct extent_state *cached_state = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04008152 u64 page_start = page_offset(page);
8153 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00008154 int inode_evicting = inode->i_state & I_FREEING;
Chris Masona52d9a82007-08-27 16:49:44 -04008155
Chris Mason8b62b722009-09-02 16:53:46 -04008156 /*
8157 * we have the page locked, so new writeback can't start,
8158 * and the dirty bit won't be cleared while we are here.
8159 *
8160 * Wait for IO on this page so that we can safely clear
8161 * the PagePrivate2 bit and do ordered accounting
8162 */
Chris Masone6dcd2d2008-07-17 12:53:50 -04008163 wait_on_page_writeback(page);
Chris Mason8b62b722009-09-02 16:53:46 -04008164
Josef Bacik5fd02042012-05-02 14:00:54 -04008165 tree = &BTRFS_I(inode)->io_tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -04008166 if (offset) {
8167 btrfs_releasepage(page, GFP_NOFS);
8168 return;
8169 }
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00008170
8171 if (!inode_evicting)
8172 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
8173 ordered = btrfs_lookup_ordered_extent(inode, page_start);
Chris Masone6dcd2d2008-07-17 12:53:50 -04008174 if (ordered) {
Chris Masoneb84ae02008-07-17 13:53:27 -04008175 /*
8176 * IO on this page will never be started, so we need
8177 * to account for any ordered extents now
8178 */
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00008179 if (!inode_evicting)
8180 clear_extent_bit(tree, page_start, page_end,
8181 EXTENT_DIRTY | EXTENT_DELALLOC |
8182 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8183 EXTENT_DEFRAG, 1, 0, &cached_state,
8184 GFP_NOFS);
Chris Mason8b62b722009-09-02 16:53:46 -04008185 /*
8186 * whoever cleared the private bit is responsible
8187 * for the finish_ordered_io
8188 */
Josef Bacik77cef2e2013-08-29 13:57:21 -04008189 if (TestClearPagePrivate2(page)) {
8190 struct btrfs_ordered_inode_tree *tree;
8191 u64 new_len;
8192
8193 tree = &BTRFS_I(inode)->ordered_tree;
8194
8195 spin_lock_irq(&tree->lock);
8196 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8197 new_len = page_start - ordered->file_offset;
8198 if (new_len < ordered->truncated_len)
8199 ordered->truncated_len = new_len;
8200 spin_unlock_irq(&tree->lock);
8201
8202 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8203 page_start,
8204 PAGE_CACHE_SIZE, 1))
8205 btrfs_finish_ordered_io(ordered);
Chris Mason8b62b722009-09-02 16:53:46 -04008206 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04008207 btrfs_put_ordered_extent(ordered);
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00008208 if (!inode_evicting) {
8209 cached_state = NULL;
8210 lock_extent_bits(tree, page_start, page_end, 0,
8211 &cached_state);
8212 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04008213 }
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00008214
8215 if (!inode_evicting) {
8216 clear_extent_bit(tree, page_start, page_end,
8217 EXTENT_LOCKED | EXTENT_DIRTY |
8218 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8219 EXTENT_DEFRAG, 1, 1,
8220 &cached_state, GFP_NOFS);
8221
8222 __btrfs_releasepage(page, GFP_NOFS);
8223 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04008224
Chris Mason4a096752008-07-21 10:29:44 -04008225 ClearPageChecked(page);
Chris Mason9ad6b7b2008-04-18 16:11:30 -04008226 if (PagePrivate(page)) {
Chris Mason9ad6b7b2008-04-18 16:11:30 -04008227 ClearPagePrivate(page);
8228 set_page_private(page, 0);
8229 page_cache_release(page);
8230 }
Chris Mason39279cc2007-06-12 06:35:45 -04008231}
8232
Chris Mason9ebefb182007-06-15 13:50:00 -04008233/*
8234 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8235 * called from a page fault handler when a page is first dirtied. Hence we must
8236 * be careful to check for EOF conditions here. We set the page up correctly
8237 * for a written page which means we get ENOSPC checking when writing into
8238 * holes and correct delalloc and unwritten extent mapping on filesystems that
8239 * support these features.
8240 *
8241 * We are not allowed to take the i_mutex here so we have to play games to
8242 * protect against truncate races as the page could now be beyond EOF. Because
8243 * vmtruncate() writes the inode size before removing pages, once we have the
8244 * page lock we can determine safely if the page is beyond EOF. If it is not
8245 * beyond EOF, then the page is guaranteed safe against truncation until we
8246 * unlock the page.
8247 */
Nick Pigginc2ec1752009-03-31 15:23:21 -07008248int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Chris Mason9ebefb182007-06-15 13:50:00 -04008249{
Nick Pigginc2ec1752009-03-31 15:23:21 -07008250 struct page *page = vmf->page;
Al Viro496ad9a2013-01-23 17:07:38 -05008251 struct inode *inode = file_inode(vma->vm_file);
Chris Mason1832a6d2007-12-21 16:27:21 -05008252 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masone6dcd2d2008-07-17 12:53:50 -04008253 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8254 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00008255 struct extent_state *cached_state = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04008256 char *kaddr;
8257 unsigned long zero_start;
Chris Mason9ebefb182007-06-15 13:50:00 -04008258 loff_t size;
Chris Mason1832a6d2007-12-21 16:27:21 -05008259 int ret;
Chris Mason9998eb72012-01-25 13:47:40 -05008260 int reserved = 0;
Chris Masona52d9a82007-08-27 16:49:44 -04008261 u64 page_start;
Chris Masone6dcd2d2008-07-17 12:53:50 -04008262 u64 page_end;
Chris Mason9ebefb182007-06-15 13:50:00 -04008263
Jan Karab2b5ef52012-06-12 16:20:45 +02008264 sb_start_pagefault(inode->i_sb);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04008265 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
Chris Mason9998eb72012-01-25 13:47:40 -05008266 if (!ret) {
Josef Bacike41f9412012-03-26 09:46:47 -04008267 ret = file_update_time(vma->vm_file);
Chris Mason9998eb72012-01-25 13:47:40 -05008268 reserved = 1;
8269 }
Nick Piggin56a76f82009-03-31 15:23:23 -07008270 if (ret) {
8271 if (ret == -ENOMEM)
8272 ret = VM_FAULT_OOM;
8273 else /* -ENOSPC, -EIO, etc */
8274 ret = VM_FAULT_SIGBUS;
Chris Mason9998eb72012-01-25 13:47:40 -05008275 if (reserved)
8276 goto out;
8277 goto out_noreserve;
Nick Piggin56a76f82009-03-31 15:23:23 -07008278 }
Chris Mason1832a6d2007-12-21 16:27:21 -05008279
Nick Piggin56a76f82009-03-31 15:23:23 -07008280 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
Chris Masone6dcd2d2008-07-17 12:53:50 -04008281again:
Chris Mason9ebefb182007-06-15 13:50:00 -04008282 lock_page(page);
Chris Mason9ebefb182007-06-15 13:50:00 -04008283 size = i_size_read(inode);
Chris Masone6dcd2d2008-07-17 12:53:50 -04008284 page_start = page_offset(page);
8285 page_end = page_start + PAGE_CACHE_SIZE - 1;
Chris Masona52d9a82007-08-27 16:49:44 -04008286
Chris Mason9ebefb182007-06-15 13:50:00 -04008287 if ((page->mapping != inode->i_mapping) ||
Chris Masone6dcd2d2008-07-17 12:53:50 -04008288 (page_start >= size)) {
Chris Mason9ebefb182007-06-15 13:50:00 -04008289 /* page got truncated out from underneath us */
8290 goto out_unlock;
8291 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04008292 wait_on_page_writeback(page);
8293
Jeff Mahoneyd0082372012-03-01 14:57:19 +01008294 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
Chris Masone6dcd2d2008-07-17 12:53:50 -04008295 set_page_extent_mapped(page);
8296
Chris Masoneb84ae02008-07-17 13:53:27 -04008297 /*
8298 * we can't set the delalloc bits if there are pending ordered
8299 * extents. Drop our locks and wait for them to finish
8300 */
Chris Masone6dcd2d2008-07-17 12:53:50 -04008301 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8302 if (ordered) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00008303 unlock_extent_cached(io_tree, page_start, page_end,
8304 &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04008305 unlock_page(page);
Chris Masoneb84ae02008-07-17 13:53:27 -04008306 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04008307 btrfs_put_ordered_extent(ordered);
8308 goto again;
8309 }
8310
Josef Bacikfbf19082009-10-01 17:10:23 -04008311 /*
8312 * XXX - page_mkwrite gets called every time the page is dirtied, even
8313 * if it was already dirty, so for space accounting reasons we need to
8314 * clear any delalloc bits for the range we are fixing to save. There
8315 * is probably a better way to do this, but for now keep consistent with
8316 * prepare_pages in the normal write path.
8317 */
Josef Bacik2ac55d42010-02-03 19:33:23 +00008318 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
Liu Bo9e8a4a82012-09-05 19:10:51 -06008319 EXTENT_DIRTY | EXTENT_DELALLOC |
8320 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
Josef Bacik2ac55d42010-02-03 19:33:23 +00008321 0, 0, &cached_state, GFP_NOFS);
Josef Bacikfbf19082009-10-01 17:10:23 -04008322
Josef Bacik2ac55d42010-02-03 19:33:23 +00008323 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
8324 &cached_state);
Josef Bacik9ed74f22009-09-11 16:12:44 -04008325 if (ret) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00008326 unlock_extent_cached(io_tree, page_start, page_end,
8327 &cached_state, GFP_NOFS);
Josef Bacik9ed74f22009-09-11 16:12:44 -04008328 ret = VM_FAULT_SIGBUS;
8329 goto out_unlock;
8330 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04008331 ret = 0;
Chris Mason9ebefb182007-06-15 13:50:00 -04008332
8333 /* page is wholly or partially inside EOF */
Chris Masona52d9a82007-08-27 16:49:44 -04008334 if (page_start + PAGE_CACHE_SIZE > size)
Chris Masone6dcd2d2008-07-17 12:53:50 -04008335 zero_start = size & ~PAGE_CACHE_MASK;
Chris Mason9ebefb182007-06-15 13:50:00 -04008336 else
Chris Masone6dcd2d2008-07-17 12:53:50 -04008337 zero_start = PAGE_CACHE_SIZE;
Chris Mason9ebefb182007-06-15 13:50:00 -04008338
Chris Masone6dcd2d2008-07-17 12:53:50 -04008339 if (zero_start != PAGE_CACHE_SIZE) {
8340 kaddr = kmap(page);
8341 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
8342 flush_dcache_page(page);
8343 kunmap(page);
8344 }
Chris Mason247e7432008-07-17 12:53:51 -04008345 ClearPageChecked(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04008346 set_page_dirty(page);
Chris Mason50a9b212009-09-11 12:33:12 -04008347 SetPageUptodate(page);
Chris Mason5a3f23d2009-03-31 13:27:11 -04008348
Chris Mason257c62e2009-10-13 13:21:08 -04008349 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8350 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
Liu Bo46d8bc32012-08-29 01:07:55 -06008351 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
Chris Mason257c62e2009-10-13 13:21:08 -04008352
Josef Bacik2ac55d42010-02-03 19:33:23 +00008353 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
Chris Mason9ebefb182007-06-15 13:50:00 -04008354
8355out_unlock:
Jan Karab2b5ef52012-06-12 16:20:45 +02008356 if (!ret) {
8357 sb_end_pagefault(inode->i_sb);
Chris Mason50a9b212009-09-11 12:33:12 -04008358 return VM_FAULT_LOCKED;
Jan Karab2b5ef52012-06-12 16:20:45 +02008359 }
Chris Mason9ebefb182007-06-15 13:50:00 -04008360 unlock_page(page);
Chris Mason1832a6d2007-12-21 16:27:21 -05008361out:
Josef Bacikec39e182012-01-12 19:10:12 -05008362 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
Chris Mason9998eb72012-01-25 13:47:40 -05008363out_noreserve:
Jan Karab2b5ef52012-06-12 16:20:45 +02008364 sb_end_pagefault(inode->i_sb);
Chris Mason9ebefb182007-06-15 13:50:00 -04008365 return ret;
8366}
8367
Josef Bacika41ad392011-01-31 15:30:16 -05008368static int btrfs_truncate(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04008369{
8370 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacikfcb80c22011-05-03 10:40:22 -04008371 struct btrfs_block_rsv *rsv;
Josef Bacika71754f2013-06-17 17:14:39 -04008372 int ret = 0;
Josef Bacik3893e332011-01-31 16:03:11 -05008373 int err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04008374 struct btrfs_trans_handle *trans;
Chris Masondbe674a2008-07-17 12:54:05 -04008375 u64 mask = root->sectorsize - 1;
Josef Bacik07127182011-08-19 10:29:59 -04008376 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
Chris Mason39279cc2007-06-12 06:35:45 -04008377
Josef Bacik0ef8b722013-10-25 16:13:35 -04008378 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8379 (u64)-1);
8380 if (ret)
8381 return ret;
Chris Mason39279cc2007-06-12 06:35:45 -04008382
Josef Bacikfcb80c22011-05-03 10:40:22 -04008383 /*
8384 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8385 * 3 things going on here
8386 *
8387 * 1) We need to reserve space for our orphan item and the space to
8388 * delete our orphan item. Lord knows we don't want to have a dangling
8389 * orphan item because we didn't reserve space to remove it.
8390 *
8391 * 2) We need to reserve space to update our inode.
8392 *
8393 * 3) We need to have something to cache all the space that is going to
8394 * be free'd up by the truncate operation, but also have some slack
8395 * space reserved in case it uses space during the truncate (thank you
8396 * very much snapshotting).
8397 *
8398 * And we need these to all be seperate. The fact is we can use alot of
8399 * space doing the truncate, and we have no earthly idea how much space
8400 * we will use, so we need the truncate reservation to be seperate so it
8401 * doesn't end up using space reserved for updating the inode or
8402 * removing the orphan item. We also need to be able to stop the
8403 * transaction and start a new one, which means we need to be able to
8404 * update the inode several times, and we have no idea of knowing how
8405 * many times that will be, so we can't just reserve 1 item for the
8406 * entirety of the opration, so that has to be done seperately as well.
8407 * Then there is the orphan item, which does indeed need to be held on
8408 * to for the whole operation, and we need nobody to touch this reserved
8409 * space except the orphan code.
8410 *
8411 * So that leaves us with
8412 *
8413 * 1) root->orphan_block_rsv - for the orphan deletion.
8414 * 2) rsv - for the truncate reservation, which we will steal from the
8415 * transaction reservation.
8416 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8417 * updating the inode.
8418 */
Miao Xie66d8f3d2012-09-06 04:02:28 -06008419 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
Josef Bacikfcb80c22011-05-03 10:40:22 -04008420 if (!rsv)
8421 return -ENOMEM;
Josef Bacik4a338542011-08-29 11:01:31 -04008422 rsv->size = min_size;
Josef Bacikca7e70f2012-08-27 17:48:15 -04008423 rsv->failfast = 1;
Josef Bacikf0cd8462011-03-04 14:37:08 -05008424
Josef Bacik907cbce2011-08-08 13:46:15 -04008425 /*
Josef Bacik07127182011-08-19 10:29:59 -04008426 * 1 for the truncate slack space
Josef Bacik907cbce2011-08-08 13:46:15 -04008427 * 1 for updating the inode.
8428 */
Josef Bacikf3fe8202013-01-07 17:03:21 -05008429 trans = btrfs_start_transaction(root, 2);
Josef Bacikfcb80c22011-05-03 10:40:22 -04008430 if (IS_ERR(trans)) {
8431 err = PTR_ERR(trans);
8432 goto out;
8433 }
Josef Bacikf0cd8462011-03-04 14:37:08 -05008434
Josef Bacik907cbce2011-08-08 13:46:15 -04008435 /* Migrate the slack space for the truncate to our reserve */
8436 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8437 min_size);
Josef Bacikfcb80c22011-05-03 10:40:22 -04008438 BUG_ON(ret);
Josef Bacikf0cd8462011-03-04 14:37:08 -05008439
Chris Mason5a3f23d2009-03-31 13:27:11 -04008440 /*
Josef Bacik5dc562c2012-08-17 13:14:17 -04008441 * So if we truncate and then write and fsync we normally would just
8442 * write the extents that changed, which is a problem if we need to
8443 * first truncate that entire inode. So set this flag so we write out
8444 * all of the extents in the inode to the sync log so we're completely
8445 * safe.
8446 */
8447 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
Josef Bacikca7e70f2012-08-27 17:48:15 -04008448 trans->block_rsv = rsv;
Josef Bacik5dc562c2012-08-17 13:14:17 -04008449
Yan, Zheng80825102009-11-12 09:35:36 +00008450 while (1) {
8451 ret = btrfs_truncate_inode_items(trans, root, inode,
8452 inode->i_size,
8453 BTRFS_EXTENT_DATA_KEY);
Josef Bacikca7e70f2012-08-27 17:48:15 -04008454 if (ret != -ENOSPC) {
Josef Bacik3893e332011-01-31 16:03:11 -05008455 err = ret;
Yan, Zheng80825102009-11-12 09:35:36 +00008456 break;
Josef Bacik3893e332011-01-31 16:03:11 -05008457 }
Chris Mason39279cc2007-06-12 06:35:45 -04008458
Josef Bacikfcb80c22011-05-03 10:40:22 -04008459 trans->block_rsv = &root->fs_info->trans_block_rsv;
Yan, Zheng80825102009-11-12 09:35:36 +00008460 ret = btrfs_update_inode(trans, root, inode);
Josef Bacik3893e332011-01-31 16:03:11 -05008461 if (ret) {
8462 err = ret;
8463 break;
8464 }
Josef Bacikca7e70f2012-08-27 17:48:15 -04008465
Yan, Zheng80825102009-11-12 09:35:36 +00008466 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00008467 btrfs_btree_balance_dirty(root);
Josef Bacikca7e70f2012-08-27 17:48:15 -04008468
8469 trans = btrfs_start_transaction(root, 2);
8470 if (IS_ERR(trans)) {
8471 ret = err = PTR_ERR(trans);
8472 trans = NULL;
8473 break;
8474 }
8475
8476 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
8477 rsv, min_size);
8478 BUG_ON(ret); /* shouldn't happen */
8479 trans->block_rsv = rsv;
Yan, Zheng80825102009-11-12 09:35:36 +00008480 }
8481
8482 if (ret == 0 && inode->i_nlink > 0) {
Josef Bacikfcb80c22011-05-03 10:40:22 -04008483 trans->block_rsv = root->orphan_block_rsv;
Yan, Zheng80825102009-11-12 09:35:36 +00008484 ret = btrfs_orphan_del(trans, inode);
Josef Bacik3893e332011-01-31 16:03:11 -05008485 if (ret)
8486 err = ret;
Yan, Zheng80825102009-11-12 09:35:36 +00008487 }
8488
Chris Mason917c16b2011-11-08 14:49:59 -05008489 if (trans) {
8490 trans->block_rsv = &root->fs_info->trans_block_rsv;
8491 ret = btrfs_update_inode(trans, root, inode);
8492 if (ret && !err)
8493 err = ret;
Josef Bacik7b128762008-07-24 12:17:14 -04008494
Josef Bacik7ad85bb2012-01-12 19:10:12 -05008495 ret = btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00008496 btrfs_btree_balance_dirty(root);
Chris Mason917c16b2011-11-08 14:49:59 -05008497 }
Josef Bacikfcb80c22011-05-03 10:40:22 -04008498
8499out:
8500 btrfs_free_block_rsv(root, rsv);
8501
Josef Bacik3893e332011-01-31 16:03:11 -05008502 if (ret && !err)
8503 err = ret;
Josef Bacika41ad392011-01-31 15:30:16 -05008504
Josef Bacik3893e332011-01-31 16:03:11 -05008505 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04008506}
8507
Sven Wegener3b963622008-06-09 21:57:42 -04008508/*
Chris Masond352ac62008-09-29 15:18:18 -04008509 * create a new subvolume directory/inode (helper for the ioctl).
8510 */
Yan Zhengd2fb3432008-12-11 16:30:39 -05008511int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
Filipe David Borba Manana63541922014-01-07 11:47:46 +00008512 struct btrfs_root *new_root,
8513 struct btrfs_root *parent_root,
8514 u64 new_dirid)
Chris Mason39279cc2007-06-12 06:35:45 -04008515{
Chris Mason39279cc2007-06-12 06:35:45 -04008516 struct inode *inode;
Yan, Zheng76dda932009-09-21 16:00:26 -04008517 int err;
Chris Mason00e4e6b2008-08-05 11:18:09 -04008518 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04008519
Florian Albrechtskirchinger12fc9d02012-02-10 22:15:54 +01008520 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8521 new_dirid, new_dirid,
8522 S_IFDIR | (~current_umask() & S_IRWXUGO),
8523 &index);
Chris Mason54aa1f42007-06-22 14:16:25 -04008524 if (IS_ERR(inode))
Christoph Hellwigf46b5a62008-06-11 21:53:53 -04008525 return PTR_ERR(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04008526 inode->i_op = &btrfs_dir_inode_operations;
8527 inode->i_fop = &btrfs_dir_file_operations;
8528
Miklos Szeredibfe86842011-10-28 14:13:29 +02008529 set_nlink(inode, 1);
Chris Masondbe674a2008-07-17 12:54:05 -04008530 btrfs_i_size_write(inode, 0);
Chris Masonb0d5d102014-09-08 13:08:51 -07008531 unlock_new_inode(inode);
Sven Wegener3b963622008-06-09 21:57:42 -04008532
Filipe David Borba Manana63541922014-01-07 11:47:46 +00008533 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8534 if (err)
8535 btrfs_err(new_root->fs_info,
David Sterba351fd352014-05-15 16:48:20 +02008536 "error inheriting subvolume %llu properties: %d",
Filipe David Borba Manana63541922014-01-07 11:47:46 +00008537 new_root->root_key.objectid, err);
8538
Yan, Zheng76dda932009-09-21 16:00:26 -04008539 err = btrfs_update_inode(trans, new_root, inode);
Christoph Hellwigcb8e7092008-10-09 13:39:39 -04008540
Yan, Zheng76dda932009-09-21 16:00:26 -04008541 iput(inode);
Mark Fashehce598972011-07-26 11:32:23 -07008542 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04008543}
8544
Chris Mason39279cc2007-06-12 06:35:45 -04008545struct inode *btrfs_alloc_inode(struct super_block *sb)
8546{
8547 struct btrfs_inode *ei;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04008548 struct inode *inode;
Chris Mason39279cc2007-06-12 06:35:45 -04008549
8550 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
8551 if (!ei)
8552 return NULL;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04008553
8554 ei->root = NULL;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04008555 ei->generation = 0;
Josef Bacik15ee9bc2007-08-10 16:22:09 -04008556 ei->last_trans = 0;
Chris Mason257c62e2009-10-13 13:21:08 -04008557 ei->last_sub_trans = 0;
Chris Masone02119d2008-09-05 16:13:11 -04008558 ei->logged_trans = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04008559 ei->delalloc_bytes = 0;
Wang Shilong47059d92014-07-03 18:22:07 +08008560 ei->defrag_bytes = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04008561 ei->disk_i_size = 0;
8562 ei->flags = 0;
Josef Bacik7709cde2011-08-04 10:25:02 -04008563 ei->csum_bytes = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04008564 ei->index_cnt = (u64)-1;
Miao Xie67de1172013-12-26 13:07:06 +08008565 ei->dir_index = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04008566 ei->last_unlink_trans = 0;
Liu Bo46d8bc32012-08-29 01:07:55 -06008567 ei->last_log_commit = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04008568
Josef Bacik9e0baf62011-07-15 15:16:44 +00008569 spin_lock_init(&ei->lock);
8570 ei->outstanding_extents = 0;
8571 ei->reserved_extents = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04008572
Josef Bacik72ac3c02012-05-23 14:13:11 -04008573 ei->runtime_flags = 0;
Li Zefan261507a02010-12-17 14:21:50 +08008574 ei->force_compress = BTRFS_COMPRESS_NONE;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04008575
Miao Xie16cdcec2011-04-22 18:12:22 +08008576 ei->delayed_node = NULL;
8577
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04008578 inode = &ei->vfs_inode;
David Sterbaa8067e02011-04-21 00:34:43 +02008579 extent_map_tree_init(&ei->extent_tree);
David Sterbaf993c882011-04-20 23:35:57 +02008580 extent_io_tree_init(&ei->io_tree, &inode->i_data);
8581 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
Josef Bacik0b32f4b2012-03-13 09:38:00 -04008582 ei->io_tree.track_uptodate = 1;
8583 ei->io_failure_tree.track_uptodate = 1;
Josef Bacikb812ce22012-11-16 13:56:32 -05008584 atomic_set(&ei->sync_writers, 0);
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04008585 mutex_init(&ei->log_mutex);
Josef Bacikf2486792012-01-13 12:09:22 -05008586 mutex_init(&ei->delalloc_mutex);
Chris Masone6dcd2d2008-07-17 12:53:50 -04008587 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04008588 INIT_LIST_HEAD(&ei->delalloc_inodes);
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04008589 RB_CLEAR_NODE(&ei->rb_node);
8590
8591 return inode;
Chris Mason39279cc2007-06-12 06:35:45 -04008592}
8593
Josef Bacikaaedb552013-10-11 14:44:09 -04008594#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8595void btrfs_test_destroy_inode(struct inode *inode)
8596{
8597 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8598 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8599}
8600#endif
8601
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11008602static void btrfs_i_callback(struct rcu_head *head)
8603{
8604 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11008605 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8606}
8607
Chris Mason39279cc2007-06-12 06:35:45 -04008608void btrfs_destroy_inode(struct inode *inode)
8609{
Chris Masone6dcd2d2008-07-17 12:53:50 -04008610 struct btrfs_ordered_extent *ordered;
Chris Mason5a3f23d2009-03-31 13:27:11 -04008611 struct btrfs_root *root = BTRFS_I(inode)->root;
8612
Al Virob3d9b7a2012-06-09 13:51:19 -04008613 WARN_ON(!hlist_empty(&inode->i_dentry));
Chris Mason39279cc2007-06-12 06:35:45 -04008614 WARN_ON(inode->i_data.nrpages);
Josef Bacik9e0baf62011-07-15 15:16:44 +00008615 WARN_ON(BTRFS_I(inode)->outstanding_extents);
8616 WARN_ON(BTRFS_I(inode)->reserved_extents);
Josef Bacik7709cde2011-08-04 10:25:02 -04008617 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8618 WARN_ON(BTRFS_I(inode)->csum_bytes);
Wang Shilong47059d92014-07-03 18:22:07 +08008619 WARN_ON(BTRFS_I(inode)->defrag_bytes);
Chris Mason39279cc2007-06-12 06:35:45 -04008620
Chris Mason5a3f23d2009-03-31 13:27:11 -04008621 /*
Josef Bacika6dbd422009-11-11 15:53:34 -05008622 * This can happen where we create an inode, but somebody else also
8623 * created the same inode and we need to destroy the one we already
8624 * created.
8625 */
8626 if (!root)
8627 goto free;
8628
Josef Bacik8a35d952012-05-23 14:26:42 -04008629 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8630 &BTRFS_I(inode)->runtime_flags)) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00008631 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02008632 btrfs_ino(inode));
Josef Bacik8a35d952012-05-23 14:26:42 -04008633 atomic_dec(&root->orphan_inodes);
Josef Bacik7b128762008-07-24 12:17:14 -04008634 }
Josef Bacik7b128762008-07-24 12:17:14 -04008635
Chris Masond3977122009-01-05 21:25:51 -05008636 while (1) {
Chris Masone6dcd2d2008-07-17 12:53:50 -04008637 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8638 if (!ordered)
8639 break;
8640 else {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00008641 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02008642 ordered->file_offset, ordered->len);
Chris Masone6dcd2d2008-07-17 12:53:50 -04008643 btrfs_remove_ordered_extent(inode, ordered);
8644 btrfs_put_ordered_extent(ordered);
8645 btrfs_put_ordered_extent(ordered);
8646 }
8647 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04008648 inode_tree_del(inode);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04008649 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
Josef Bacika6dbd422009-11-11 15:53:34 -05008650free:
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11008651 call_rcu(&inode->i_rcu, btrfs_i_callback);
Chris Mason39279cc2007-06-12 06:35:45 -04008652}
8653
Al Viro45321ac2010-06-07 13:43:19 -04008654int btrfs_drop_inode(struct inode *inode)
Yan, Zheng76dda932009-09-21 16:00:26 -04008655{
8656 struct btrfs_root *root = BTRFS_I(inode)->root;
Al Viro45321ac2010-06-07 13:43:19 -04008657
Naohiro Aota6379ef92013-06-06 09:56:34 +00008658 if (root == NULL)
8659 return 1;
8660
Liu Bofa6ac872013-02-20 14:10:23 +00008661 /* the snap/subvol tree is on deleting */
Stefan Behrens69e9c6c2013-09-05 16:58:43 +02008662 if (btrfs_root_refs(&root->root_item) == 0)
Al Viro45321ac2010-06-07 13:43:19 -04008663 return 1;
Yan, Zheng76dda932009-09-21 16:00:26 -04008664 else
Al Viro45321ac2010-06-07 13:43:19 -04008665 return generic_drop_inode(inode);
Yan, Zheng76dda932009-09-21 16:00:26 -04008666}
8667
Sven Wegener0ee0fda2008-07-30 16:54:26 -04008668static void init_once(void *foo)
Chris Mason39279cc2007-06-12 06:35:45 -04008669{
8670 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8671
8672 inode_init_once(&ei->vfs_inode);
8673}
8674
8675void btrfs_destroy_cachep(void)
8676{
Kirill A. Shutemov8c0a8532012-09-26 11:33:07 +10008677 /*
8678 * Make sure all delayed rcu free inodes are flushed before we
8679 * destroy cache.
8680 */
8681 rcu_barrier();
Chris Mason39279cc2007-06-12 06:35:45 -04008682 if (btrfs_inode_cachep)
8683 kmem_cache_destroy(btrfs_inode_cachep);
8684 if (btrfs_trans_handle_cachep)
8685 kmem_cache_destroy(btrfs_trans_handle_cachep);
8686 if (btrfs_transaction_cachep)
8687 kmem_cache_destroy(btrfs_transaction_cachep);
Chris Mason39279cc2007-06-12 06:35:45 -04008688 if (btrfs_path_cachep)
8689 kmem_cache_destroy(btrfs_path_cachep);
Josef Bacikdc89e982011-01-28 17:05:48 -05008690 if (btrfs_free_space_cachep)
8691 kmem_cache_destroy(btrfs_free_space_cachep);
Miao Xie8ccf6f192012-10-25 09:28:04 +00008692 if (btrfs_delalloc_work_cachep)
8693 kmem_cache_destroy(btrfs_delalloc_work_cachep);
Chris Mason39279cc2007-06-12 06:35:45 -04008694}
8695
8696int btrfs_init_cachep(void)
8697{
David Sterba837e1972012-09-07 03:00:48 -06008698 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02008699 sizeof(struct btrfs_inode), 0,
8700 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
Chris Mason39279cc2007-06-12 06:35:45 -04008701 if (!btrfs_inode_cachep)
8702 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02008703
David Sterba837e1972012-09-07 03:00:48 -06008704 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02008705 sizeof(struct btrfs_trans_handle), 0,
8706 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04008707 if (!btrfs_trans_handle_cachep)
8708 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02008709
David Sterba837e1972012-09-07 03:00:48 -06008710 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02008711 sizeof(struct btrfs_transaction), 0,
8712 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04008713 if (!btrfs_transaction_cachep)
8714 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02008715
David Sterba837e1972012-09-07 03:00:48 -06008716 btrfs_path_cachep = kmem_cache_create("btrfs_path",
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02008717 sizeof(struct btrfs_path), 0,
8718 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04008719 if (!btrfs_path_cachep)
8720 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02008721
David Sterba837e1972012-09-07 03:00:48 -06008722 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
Josef Bacikdc89e982011-01-28 17:05:48 -05008723 sizeof(struct btrfs_free_space), 0,
8724 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8725 if (!btrfs_free_space_cachep)
8726 goto fail;
8727
Miao Xie8ccf6f192012-10-25 09:28:04 +00008728 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8729 sizeof(struct btrfs_delalloc_work), 0,
8730 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8731 NULL);
8732 if (!btrfs_delalloc_work_cachep)
8733 goto fail;
8734
Chris Mason39279cc2007-06-12 06:35:45 -04008735 return 0;
8736fail:
8737 btrfs_destroy_cachep();
8738 return -ENOMEM;
8739}
8740
8741static int btrfs_getattr(struct vfsmount *mnt,
8742 struct dentry *dentry, struct kstat *stat)
8743{
Miao Xiedf0af1a2013-01-29 10:11:59 +00008744 u64 delalloc_bytes;
Chris Mason39279cc2007-06-12 06:35:45 -04008745 struct inode *inode = dentry->d_inode;
David Sterbafadc0d82011-11-20 07:33:38 -05008746 u32 blocksize = inode->i_sb->s_blocksize;
8747
Chris Mason39279cc2007-06-12 06:35:45 -04008748 generic_fillattr(inode, stat);
Al Viro0ee5dc62011-07-07 15:44:25 -04008749 stat->dev = BTRFS_I(inode)->root->anon_dev;
Chris Masond6667462008-01-03 14:51:00 -05008750 stat->blksize = PAGE_CACHE_SIZE;
Miao Xiedf0af1a2013-01-29 10:11:59 +00008751
8752 spin_lock(&BTRFS_I(inode)->lock);
8753 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8754 spin_unlock(&BTRFS_I(inode)->lock);
David Sterbafadc0d82011-11-20 07:33:38 -05008755 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
Miao Xiedf0af1a2013-01-29 10:11:59 +00008756 ALIGN(delalloc_bytes, blocksize)) >> 9;
Chris Mason39279cc2007-06-12 06:35:45 -04008757 return 0;
8758}
8759
Chris Masond3977122009-01-05 21:25:51 -05008760static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8761 struct inode *new_dir, struct dentry *new_dentry)
Chris Mason39279cc2007-06-12 06:35:45 -04008762{
8763 struct btrfs_trans_handle *trans;
8764 struct btrfs_root *root = BTRFS_I(old_dir)->root;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008765 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04008766 struct inode *new_inode = new_dentry->d_inode;
8767 struct inode *old_inode = old_dentry->d_inode;
8768 struct timespec ctime = CURRENT_TIME;
Chris Mason00e4e6b2008-08-05 11:18:09 -04008769 u64 index = 0;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008770 u64 root_objectid;
Chris Mason39279cc2007-06-12 06:35:45 -04008771 int ret;
Li Zefan33345d012011-04-20 10:31:50 +08008772 u64 old_ino = btrfs_ino(old_inode);
Chris Mason39279cc2007-06-12 06:35:45 -04008773
Li Zefan33345d012011-04-20 10:31:50 +08008774 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
Yan, Zhengf679a842009-09-24 09:17:31 -04008775 return -EPERM;
8776
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008777 /* we only allow rename subvolume link between subvolumes */
Li Zefan33345d012011-04-20 10:31:50 +08008778 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
Chris Mason3394e162008-11-17 20:42:26 -05008779 return -EXDEV;
8780
Li Zefan33345d012011-04-20 10:31:50 +08008781 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8782 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008783 return -ENOTEMPTY;
8784
Chris Mason39279cc2007-06-12 06:35:45 -04008785 if (S_ISDIR(old_inode->i_mode) && new_inode &&
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008786 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
Chris Mason39279cc2007-06-12 06:35:45 -04008787 return -ENOTEMPTY;
Chris Mason9c520572012-12-17 14:26:57 -05008788
8789
8790 /* check for collisions, even if the name isn't there */
Josef Bacik4871c152013-10-09 12:24:04 -04008791 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
Chris Mason9c520572012-12-17 14:26:57 -05008792 new_dentry->d_name.name,
8793 new_dentry->d_name.len);
8794
8795 if (ret) {
8796 if (ret == -EEXIST) {
8797 /* we shouldn't get
8798 * eexist without a new_inode */
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05308799 if (WARN_ON(!new_inode)) {
Chris Mason9c520572012-12-17 14:26:57 -05008800 return ret;
8801 }
8802 } else {
8803 /* maybe -EOVERFLOW */
8804 return ret;
8805 }
8806 }
8807 ret = 0;
8808
Chris Mason5a3f23d2009-03-31 13:27:11 -04008809 /*
Chris Mason8d875f92014-08-12 10:47:42 -07008810 * we're using rename to replace one file with another. Start IO on it
8811 * now so we don't add too much work to the end of the transaction
Chris Mason5a3f23d2009-03-31 13:27:11 -04008812 */
Chris Mason8d875f92014-08-12 10:47:42 -07008813 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
Chris Mason5a3f23d2009-03-31 13:27:11 -04008814 filemap_flush(old_inode->i_mapping);
8815
Yan, Zheng76dda932009-09-21 16:00:26 -04008816 /* close the racy window with snapshot create/destroy ioctl */
Li Zefan33345d012011-04-20 10:31:50 +08008817 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
Yan, Zheng76dda932009-09-21 16:00:26 -04008818 down_read(&root->fs_info->subvol_sem);
Yan, Zhenga22285a2010-05-16 10:48:46 -04008819 /*
8820 * We want to reserve the absolute worst case amount of items. So if
8821 * both inodes are subvols and we need to unlink them then that would
8822 * require 4 item modifications, but if they are both normal inodes it
8823 * would require 5 item modifications, so we'll assume their normal
8824 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8825 * should cover the worst case number of items we'll modify.
8826 */
Josef Bacik6e137ed2013-03-26 15:26:55 -04008827 trans = btrfs_start_transaction(root, 11);
Johann Lombardib44c59a2011-03-31 13:23:47 +00008828 if (IS_ERR(trans)) {
8829 ret = PTR_ERR(trans);
8830 goto out_notrans;
8831 }
Chris Mason5f39d392007-10-15 16:14:19 -04008832
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008833 if (dest != root)
8834 btrfs_record_root_in_trans(trans, dest);
8835
Yan, Zhenga5719522009-09-24 09:17:31 -04008836 ret = btrfs_set_inode_index(new_dir, &index);
8837 if (ret)
8838 goto out_fail;
Chris Mason5a3f23d2009-03-31 13:27:11 -04008839
Miao Xie67de1172013-12-26 13:07:06 +08008840 BTRFS_I(old_inode)->dir_index = 0ULL;
Li Zefan33345d012011-04-20 10:31:50 +08008841 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008842 /* force full log commit if subvolume involved. */
Miao Xie995946d2014-04-02 19:51:06 +08008843 btrfs_set_log_full_commit(root->fs_info, trans);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008844 } else {
Yan, Zhenga5719522009-09-24 09:17:31 -04008845 ret = btrfs_insert_inode_ref(trans, dest,
8846 new_dentry->d_name.name,
8847 new_dentry->d_name.len,
Li Zefan33345d012011-04-20 10:31:50 +08008848 old_ino,
8849 btrfs_ino(new_dir), index);
Yan, Zhenga5719522009-09-24 09:17:31 -04008850 if (ret)
8851 goto out_fail;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008852 /*
8853 * this is an ugly little race, but the rename is required
8854 * to make sure that if we crash, the inode is either at the
8855 * old name or the new one. pinning the log transaction lets
8856 * us make sure we don't allow a log commit to come in after
8857 * we unlink the name but before we add the new name back in.
8858 */
8859 btrfs_pin_log_trans(root);
8860 }
Chris Mason5a3f23d2009-03-31 13:27:11 -04008861
Josef Bacik0c4d2d92012-04-05 15:03:02 -04008862 inode_inc_iversion(old_dir);
8863 inode_inc_iversion(new_dir);
8864 inode_inc_iversion(old_inode);
Chris Mason39279cc2007-06-12 06:35:45 -04008865 old_dir->i_ctime = old_dir->i_mtime = ctime;
8866 new_dir->i_ctime = new_dir->i_mtime = ctime;
8867 old_inode->i_ctime = ctime;
Chris Mason5f39d392007-10-15 16:14:19 -04008868
Chris Mason12fcfd22009-03-24 10:24:20 -04008869 if (old_dentry->d_parent != new_dentry->d_parent)
8870 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8871
Li Zefan33345d012011-04-20 10:31:50 +08008872 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008873 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8874 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8875 old_dentry->d_name.name,
8876 old_dentry->d_name.len);
8877 } else {
Al Viro92986792011-03-04 17:14:37 +00008878 ret = __btrfs_unlink_inode(trans, root, old_dir,
8879 old_dentry->d_inode,
8880 old_dentry->d_name.name,
8881 old_dentry->d_name.len);
8882 if (!ret)
8883 ret = btrfs_update_inode(trans, root, old_inode);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008884 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01008885 if (ret) {
8886 btrfs_abort_transaction(trans, root, ret);
8887 goto out_fail;
8888 }
Chris Mason39279cc2007-06-12 06:35:45 -04008889
8890 if (new_inode) {
Josef Bacik0c4d2d92012-04-05 15:03:02 -04008891 inode_inc_iversion(new_inode);
Chris Mason39279cc2007-06-12 06:35:45 -04008892 new_inode->i_ctime = CURRENT_TIME;
Li Zefan33345d012011-04-20 10:31:50 +08008893 if (unlikely(btrfs_ino(new_inode) ==
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008894 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8895 root_objectid = BTRFS_I(new_inode)->location.objectid;
8896 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8897 root_objectid,
8898 new_dentry->d_name.name,
8899 new_dentry->d_name.len);
8900 BUG_ON(new_inode->i_nlink == 0);
8901 } else {
8902 ret = btrfs_unlink_inode(trans, dest, new_dir,
8903 new_dentry->d_inode,
8904 new_dentry->d_name.name,
8905 new_dentry->d_name.len);
8906 }
Josef Bacik4ef31a42013-08-13 14:10:08 -04008907 if (!ret && new_inode->i_nlink == 0)
Chris Masone02119d2008-09-05 16:13:11 -04008908 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01008909 if (ret) {
8910 btrfs_abort_transaction(trans, root, ret);
8911 goto out_fail;
8912 }
Chris Mason39279cc2007-06-12 06:35:45 -04008913 }
Josef Bacikaec74772008-07-24 12:12:38 -04008914
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008915 ret = btrfs_add_link(trans, new_dir, old_inode,
8916 new_dentry->d_name.name,
Yan, Zhenga5719522009-09-24 09:17:31 -04008917 new_dentry->d_name.len, 0, index);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01008918 if (ret) {
8919 btrfs_abort_transaction(trans, root, ret);
8920 goto out_fail;
8921 }
Chris Mason39279cc2007-06-12 06:35:45 -04008922
Miao Xie67de1172013-12-26 13:07:06 +08008923 if (old_inode->i_nlink == 1)
8924 BTRFS_I(old_inode)->dir_index = index;
8925
Li Zefan33345d012011-04-20 10:31:50 +08008926 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
Al Viro10d9f302011-07-16 23:09:10 -04008927 struct dentry *parent = new_dentry->d_parent;
Josef Bacik6a912212010-11-20 09:48:00 +00008928 btrfs_log_new_name(trans, old_inode, old_dir, parent);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008929 btrfs_end_log_trans(root);
8930 }
Chris Mason39279cc2007-06-12 06:35:45 -04008931out_fail:
Josef Bacik7ad85bb2012-01-12 19:10:12 -05008932 btrfs_end_transaction(trans, root);
Johann Lombardib44c59a2011-03-31 13:23:47 +00008933out_notrans:
Li Zefan33345d012011-04-20 10:31:50 +08008934 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
Yan, Zheng76dda932009-09-21 16:00:26 -04008935 up_read(&root->fs_info->subvol_sem);
Josef Bacik9ed74f22009-09-11 16:12:44 -04008936
Chris Mason39279cc2007-06-12 06:35:45 -04008937 return ret;
8938}
8939
Miklos Szeredi80ace852014-07-23 15:15:32 +02008940static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
8941 struct inode *new_dir, struct dentry *new_dentry,
8942 unsigned int flags)
8943{
8944 if (flags & ~RENAME_NOREPLACE)
8945 return -EINVAL;
8946
8947 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
8948}
8949
Miao Xie8ccf6f192012-10-25 09:28:04 +00008950static void btrfs_run_delalloc_work(struct btrfs_work *work)
8951{
8952 struct btrfs_delalloc_work *delalloc_work;
Josef Bacik9f23e282013-10-28 15:03:41 -04008953 struct inode *inode;
Miao Xie8ccf6f192012-10-25 09:28:04 +00008954
8955 delalloc_work = container_of(work, struct btrfs_delalloc_work,
8956 work);
Josef Bacik9f23e282013-10-28 15:03:41 -04008957 inode = delalloc_work->inode;
8958 if (delalloc_work->wait) {
8959 btrfs_wait_ordered_range(inode, 0, (u64)-1);
8960 } else {
8961 filemap_flush(inode->i_mapping);
8962 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8963 &BTRFS_I(inode)->runtime_flags))
8964 filemap_flush(inode->i_mapping);
8965 }
Miao Xie8ccf6f192012-10-25 09:28:04 +00008966
8967 if (delalloc_work->delay_iput)
Josef Bacik9f23e282013-10-28 15:03:41 -04008968 btrfs_add_delayed_iput(inode);
Miao Xie8ccf6f192012-10-25 09:28:04 +00008969 else
Josef Bacik9f23e282013-10-28 15:03:41 -04008970 iput(inode);
Miao Xie8ccf6f192012-10-25 09:28:04 +00008971 complete(&delalloc_work->completion);
8972}
8973
8974struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8975 int wait, int delay_iput)
8976{
8977 struct btrfs_delalloc_work *work;
8978
8979 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8980 if (!work)
8981 return NULL;
8982
8983 init_completion(&work->completion);
8984 INIT_LIST_HEAD(&work->list);
8985 work->inode = inode;
8986 work->wait = wait;
8987 work->delay_iput = delay_iput;
Liu Bo9e0af232014-08-15 23:36:53 +08008988 WARN_ON_ONCE(!inode);
8989 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
8990 btrfs_run_delalloc_work, NULL, NULL);
Miao Xie8ccf6f192012-10-25 09:28:04 +00008991
8992 return work;
8993}
8994
8995void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8996{
8997 wait_for_completion(&work->completion);
8998 kmem_cache_free(btrfs_delalloc_work_cachep, work);
8999}
9000
Chris Masond352ac62008-09-29 15:18:18 -04009001/*
9002 * some fairly slow code that needs optimization. This walks the list
9003 * of all the inodes with pending delalloc and forces them to disk.
9004 */
Miao Xie6c255e62014-03-06 13:55:01 +08009005static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9006 int nr)
Chris Masonea8c2812008-08-04 23:17:27 -04009007{
Chris Masonea8c2812008-08-04 23:17:27 -04009008 struct btrfs_inode *binode;
Zheng Yan5b21f2e2008-09-26 10:05:38 -04009009 struct inode *inode;
Miao Xie8ccf6f192012-10-25 09:28:04 +00009010 struct btrfs_delalloc_work *work, *next;
9011 struct list_head works;
Miao Xie1eafa6c2013-01-22 10:49:00 +00009012 struct list_head splice;
Miao Xie8ccf6f192012-10-25 09:28:04 +00009013 int ret = 0;
Chris Masonea8c2812008-08-04 23:17:27 -04009014
Miao Xie8ccf6f192012-10-25 09:28:04 +00009015 INIT_LIST_HEAD(&works);
Miao Xie1eafa6c2013-01-22 10:49:00 +00009016 INIT_LIST_HEAD(&splice);
Miao Xie63607cc2013-01-22 10:50:35 +00009017
Miao Xie573bfb72014-03-06 13:55:03 +08009018 mutex_lock(&root->delalloc_mutex);
Miao Xieeb73c1b2013-05-15 07:48:22 +00009019 spin_lock(&root->delalloc_lock);
9020 list_splice_init(&root->delalloc_inodes, &splice);
Miao Xie1eafa6c2013-01-22 10:49:00 +00009021 while (!list_empty(&splice)) {
9022 binode = list_entry(splice.next, struct btrfs_inode,
Chris Masonea8c2812008-08-04 23:17:27 -04009023 delalloc_inodes);
Miao Xie1eafa6c2013-01-22 10:49:00 +00009024
Miao Xieeb73c1b2013-05-15 07:48:22 +00009025 list_move_tail(&binode->delalloc_inodes,
9026 &root->delalloc_inodes);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04009027 inode = igrab(&binode->vfs_inode);
Miao Xiedf0af1a2013-01-29 10:11:59 +00009028 if (!inode) {
Miao Xieeb73c1b2013-05-15 07:48:22 +00009029 cond_resched_lock(&root->delalloc_lock);
Miao Xie1eafa6c2013-01-22 10:49:00 +00009030 continue;
Miao Xiedf0af1a2013-01-29 10:11:59 +00009031 }
Miao Xieeb73c1b2013-05-15 07:48:22 +00009032 spin_unlock(&root->delalloc_lock);
Miao Xie1eafa6c2013-01-22 10:49:00 +00009033
9034 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
David Sterba5d99a9982014-09-29 19:20:37 +02009035 if (!work) {
Josef Bacikf4ab9ea2013-09-17 11:25:44 -04009036 if (delay_iput)
9037 btrfs_add_delayed_iput(inode);
9038 else
9039 iput(inode);
Miao Xie1eafa6c2013-01-22 10:49:00 +00009040 ret = -ENOMEM;
Wang Shilonga1ecaab2014-04-02 19:53:32 +08009041 goto out;
Zheng Yan5b21f2e2008-09-26 10:05:38 -04009042 }
Miao Xie1eafa6c2013-01-22 10:49:00 +00009043 list_add_tail(&work->list, &works);
Qu Wenruoa44903a2014-02-28 10:46:09 +08009044 btrfs_queue_work(root->fs_info->flush_workers,
9045 &work->work);
Miao Xie6c255e62014-03-06 13:55:01 +08009046 ret++;
9047 if (nr != -1 && ret >= nr)
Wang Shilonga1ecaab2014-04-02 19:53:32 +08009048 goto out;
Zheng Yan5b21f2e2008-09-26 10:05:38 -04009049 cond_resched();
Miao Xieeb73c1b2013-05-15 07:48:22 +00009050 spin_lock(&root->delalloc_lock);
Chris Masonea8c2812008-08-04 23:17:27 -04009051 }
Miao Xieeb73c1b2013-05-15 07:48:22 +00009052 spin_unlock(&root->delalloc_lock);
Chris Mason8c8bee12008-09-29 11:19:10 -04009053
Wang Shilonga1ecaab2014-04-02 19:53:32 +08009054out:
Miao Xie1eafa6c2013-01-22 10:49:00 +00009055 list_for_each_entry_safe(work, next, &works, list) {
9056 list_del_init(&work->list);
9057 btrfs_wait_and_free_delalloc_work(work);
9058 }
9059
Miao Xieeb73c1b2013-05-15 07:48:22 +00009060 if (!list_empty_careful(&splice)) {
9061 spin_lock(&root->delalloc_lock);
9062 list_splice_tail(&splice, &root->delalloc_inodes);
9063 spin_unlock(&root->delalloc_lock);
9064 }
Miao Xie573bfb72014-03-06 13:55:03 +08009065 mutex_unlock(&root->delalloc_mutex);
Miao Xieeb73c1b2013-05-15 07:48:22 +00009066 return ret;
9067}
9068
9069int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9070{
9071 int ret;
9072
Wang Shilong2c21b4d2014-01-14 19:42:20 +08009073 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
Miao Xieeb73c1b2013-05-15 07:48:22 +00009074 return -EROFS;
9075
Miao Xie6c255e62014-03-06 13:55:01 +08009076 ret = __start_delalloc_inodes(root, delay_iput, -1);
9077 if (ret > 0)
9078 ret = 0;
Miao Xieeb73c1b2013-05-15 07:48:22 +00009079 /*
9080 * the filemap_flush will queue IO into the worker threads, but
Chris Mason8c8bee12008-09-29 11:19:10 -04009081 * we have to make sure the IO is actually started and that
9082 * ordered extents get created before we return
9083 */
9084 atomic_inc(&root->fs_info->async_submit_draining);
Chris Masond3977122009-01-05 21:25:51 -05009085 while (atomic_read(&root->fs_info->nr_async_submits) ||
Chris Mason771ed682008-11-06 22:02:51 -05009086 atomic_read(&root->fs_info->async_delalloc_pages)) {
Chris Mason8c8bee12008-09-29 11:19:10 -04009087 wait_event(root->fs_info->async_submit_wait,
Chris Mason771ed682008-11-06 22:02:51 -05009088 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9089 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
Chris Mason8c8bee12008-09-29 11:19:10 -04009090 }
9091 atomic_dec(&root->fs_info->async_submit_draining);
Miao Xieeb73c1b2013-05-15 07:48:22 +00009092 return ret;
9093}
9094
Miao Xie6c255e62014-03-06 13:55:01 +08009095int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9096 int nr)
Miao Xieeb73c1b2013-05-15 07:48:22 +00009097{
9098 struct btrfs_root *root;
9099 struct list_head splice;
9100 int ret;
9101
Wang Shilong2c21b4d2014-01-14 19:42:20 +08009102 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
Miao Xieeb73c1b2013-05-15 07:48:22 +00009103 return -EROFS;
9104
9105 INIT_LIST_HEAD(&splice);
9106
Miao Xie573bfb72014-03-06 13:55:03 +08009107 mutex_lock(&fs_info->delalloc_root_mutex);
Miao Xieeb73c1b2013-05-15 07:48:22 +00009108 spin_lock(&fs_info->delalloc_root_lock);
9109 list_splice_init(&fs_info->delalloc_roots, &splice);
Miao Xie6c255e62014-03-06 13:55:01 +08009110 while (!list_empty(&splice) && nr) {
Miao Xieeb73c1b2013-05-15 07:48:22 +00009111 root = list_first_entry(&splice, struct btrfs_root,
9112 delalloc_root);
9113 root = btrfs_grab_fs_root(root);
9114 BUG_ON(!root);
9115 list_move_tail(&root->delalloc_root,
9116 &fs_info->delalloc_roots);
9117 spin_unlock(&fs_info->delalloc_root_lock);
9118
Miao Xie6c255e62014-03-06 13:55:01 +08009119 ret = __start_delalloc_inodes(root, delay_iput, nr);
Miao Xieeb73c1b2013-05-15 07:48:22 +00009120 btrfs_put_fs_root(root);
Miao Xie6c255e62014-03-06 13:55:01 +08009121 if (ret < 0)
Miao Xieeb73c1b2013-05-15 07:48:22 +00009122 goto out;
9123
Miao Xie6c255e62014-03-06 13:55:01 +08009124 if (nr != -1) {
9125 nr -= ret;
9126 WARN_ON(nr < 0);
9127 }
Miao Xieeb73c1b2013-05-15 07:48:22 +00009128 spin_lock(&fs_info->delalloc_root_lock);
9129 }
9130 spin_unlock(&fs_info->delalloc_root_lock);
9131
Miao Xie6c255e62014-03-06 13:55:01 +08009132 ret = 0;
Miao Xieeb73c1b2013-05-15 07:48:22 +00009133 atomic_inc(&fs_info->async_submit_draining);
9134 while (atomic_read(&fs_info->nr_async_submits) ||
9135 atomic_read(&fs_info->async_delalloc_pages)) {
9136 wait_event(fs_info->async_submit_wait,
9137 (atomic_read(&fs_info->nr_async_submits) == 0 &&
9138 atomic_read(&fs_info->async_delalloc_pages) == 0));
9139 }
9140 atomic_dec(&fs_info->async_submit_draining);
Miao Xie8ccf6f192012-10-25 09:28:04 +00009141out:
Miao Xie1eafa6c2013-01-22 10:49:00 +00009142 if (!list_empty_careful(&splice)) {
Miao Xieeb73c1b2013-05-15 07:48:22 +00009143 spin_lock(&fs_info->delalloc_root_lock);
9144 list_splice_tail(&splice, &fs_info->delalloc_roots);
9145 spin_unlock(&fs_info->delalloc_root_lock);
Miao Xie1eafa6c2013-01-22 10:49:00 +00009146 }
Miao Xie573bfb72014-03-06 13:55:03 +08009147 mutex_unlock(&fs_info->delalloc_root_mutex);
Miao Xie8ccf6f192012-10-25 09:28:04 +00009148 return ret;
Chris Masonea8c2812008-08-04 23:17:27 -04009149}
9150
Chris Mason39279cc2007-06-12 06:35:45 -04009151static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9152 const char *symname)
9153{
9154 struct btrfs_trans_handle *trans;
9155 struct btrfs_root *root = BTRFS_I(dir)->root;
9156 struct btrfs_path *path;
9157 struct btrfs_key key;
Chris Mason1832a6d2007-12-21 16:27:21 -05009158 struct inode *inode = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04009159 int err;
9160 int drop_inode = 0;
9161 u64 objectid;
Dulshani Gunawardhana67871252013-10-31 10:33:04 +05309162 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04009163 int name_len;
9164 int datasize;
Chris Mason5f39d392007-10-15 16:14:19 -04009165 unsigned long ptr;
Chris Mason39279cc2007-06-12 06:35:45 -04009166 struct btrfs_file_extent_item *ei;
Chris Mason5f39d392007-10-15 16:14:19 -04009167 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04009168
Filipe David Borba Mananaf06becc2013-09-16 09:53:28 +01009169 name_len = strlen(symname);
Chris Mason39279cc2007-06-12 06:35:45 -04009170 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9171 return -ENAMETOOLONG;
Chris Mason1832a6d2007-12-21 16:27:21 -05009172
Josef Bacik9ed74f22009-09-11 16:12:44 -04009173 /*
9174 * 2 items for inode item and ref
9175 * 2 items for dir items
9176 * 1 item for xattr if selinux is on
9177 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04009178 trans = btrfs_start_transaction(root, 5);
9179 if (IS_ERR(trans))
9180 return PTR_ERR(trans);
Chris Mason1832a6d2007-12-21 16:27:21 -05009181
Li Zefan581bb052011-04-20 10:06:11 +08009182 err = btrfs_find_free_ino(root, &objectid);
9183 if (err)
9184 goto out_unlock;
9185
Josef Bacikaec74772008-07-24 12:12:38 -04009186 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Li Zefan33345d012011-04-20 10:31:50 +08009187 dentry->d_name.len, btrfs_ino(dir), objectid,
Josef Bacikd82a6f12011-05-11 15:26:06 -04009188 S_IFLNK|S_IRWXUGO, &index);
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04009189 if (IS_ERR(inode)) {
9190 err = PTR_ERR(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04009191 goto out_unlock;
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04009192 }
Chris Mason39279cc2007-06-12 06:35:45 -04009193
Casey Schauflerad19db72011-12-15 10:09:07 -05009194 /*
9195 * If the active LSM wants to access the inode during
9196 * d_instantiate it needs these. Smack checks to see
9197 * if the filesystem supports xattrs by looking at the
9198 * ops vector.
9199 */
9200 inode->i_fop = &btrfs_file_operations;
9201 inode->i_op = &btrfs_file_inode_operations;
Chris Masonb0d5d102014-09-08 13:08:51 -07009202 inode->i_mapping->a_ops = &btrfs_aops;
Chris Masonb0d5d102014-09-08 13:08:51 -07009203 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9204
9205 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9206 if (err)
9207 goto out_unlock_inode;
Casey Schauflerad19db72011-12-15 10:09:07 -05009208
Josef Bacika1b075d2010-11-19 20:36:11 +00009209 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04009210 if (err)
Chris Masonb0d5d102014-09-08 13:08:51 -07009211 goto out_unlock_inode;
Chris Mason39279cc2007-06-12 06:35:45 -04009212
9213 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07009214 if (!path) {
9215 err = -ENOMEM;
Chris Masonb0d5d102014-09-08 13:08:51 -07009216 goto out_unlock_inode;
Mark Fashehd8926bb2011-07-13 10:38:47 -07009217 }
Li Zefan33345d012011-04-20 10:31:50 +08009218 key.objectid = btrfs_ino(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04009219 key.offset = 0;
David Sterba962a2982014-06-04 18:41:45 +02009220 key.type = BTRFS_EXTENT_DATA_KEY;
Chris Mason39279cc2007-06-12 06:35:45 -04009221 datasize = btrfs_file_extent_calc_inline_size(name_len);
9222 err = btrfs_insert_empty_item(trans, root, path, &key,
9223 datasize);
Chris Mason54aa1f42007-06-22 14:16:25 -04009224 if (err) {
Julia Lawallb0839162011-05-14 07:10:51 +00009225 btrfs_free_path(path);
Chris Masonb0d5d102014-09-08 13:08:51 -07009226 goto out_unlock_inode;
Chris Mason54aa1f42007-06-22 14:16:25 -04009227 }
Chris Mason5f39d392007-10-15 16:14:19 -04009228 leaf = path->nodes[0];
9229 ei = btrfs_item_ptr(leaf, path->slots[0],
9230 struct btrfs_file_extent_item);
9231 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9232 btrfs_set_file_extent_type(leaf, ei,
Chris Mason39279cc2007-06-12 06:35:45 -04009233 BTRFS_FILE_EXTENT_INLINE);
Chris Masonc8b97812008-10-29 14:49:59 -04009234 btrfs_set_file_extent_encryption(leaf, ei, 0);
9235 btrfs_set_file_extent_compression(leaf, ei, 0);
9236 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9237 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9238
Chris Mason39279cc2007-06-12 06:35:45 -04009239 ptr = btrfs_file_extent_inline_start(ei);
Chris Mason5f39d392007-10-15 16:14:19 -04009240 write_extent_buffer(leaf, symname, ptr, name_len);
9241 btrfs_mark_buffer_dirty(leaf);
Chris Mason39279cc2007-06-12 06:35:45 -04009242 btrfs_free_path(path);
Chris Mason5f39d392007-10-15 16:14:19 -04009243
Chris Mason39279cc2007-06-12 06:35:45 -04009244 inode->i_op = &btrfs_symlink_inode_operations;
9245 inode->i_mapping->a_ops = &btrfs_symlink_aops;
Yan Zhengd899e052008-10-30 14:25:28 -04009246 inode_set_bytes(inode, name_len);
Filipe David Borba Mananaf06becc2013-09-16 09:53:28 +01009247 btrfs_i_size_write(inode, name_len);
Chris Mason54aa1f42007-06-22 14:16:25 -04009248 err = btrfs_update_inode(trans, root, inode);
Chris Masonb0d5d102014-09-08 13:08:51 -07009249 if (err) {
Chris Mason54aa1f42007-06-22 14:16:25 -04009250 drop_inode = 1;
Chris Masonb0d5d102014-09-08 13:08:51 -07009251 goto out_unlock_inode;
9252 }
9253
9254 unlock_new_inode(inode);
9255 d_instantiate(dentry, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04009256
9257out_unlock:
Josef Bacik7ad85bb2012-01-12 19:10:12 -05009258 btrfs_end_transaction(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04009259 if (drop_inode) {
9260 inode_dec_link_count(inode);
9261 iput(inode);
9262 }
Liu Bob53d3f52012-11-14 14:34:34 +00009263 btrfs_btree_balance_dirty(root);
Chris Mason39279cc2007-06-12 06:35:45 -04009264 return err;
Chris Masonb0d5d102014-09-08 13:08:51 -07009265
9266out_unlock_inode:
9267 drop_inode = 1;
9268 unlock_new_inode(inode);
9269 goto out_unlock;
Chris Mason39279cc2007-06-12 06:35:45 -04009270}
Chris Mason16432982008-04-10 10:23:21 -04009271
Josef Bacik0af3d002010-06-21 14:48:16 -04009272static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9273 u64 start, u64 num_bytes, u64 min_size,
9274 loff_t actual_len, u64 *alloc_hint,
9275 struct btrfs_trans_handle *trans)
Yan Zhengd899e052008-10-30 14:25:28 -04009276{
Josef Bacik5dc562c2012-08-17 13:14:17 -04009277 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9278 struct extent_map *em;
Yan Zhengd899e052008-10-30 14:25:28 -04009279 struct btrfs_root *root = BTRFS_I(inode)->root;
9280 struct btrfs_key ins;
Yan Zhengd899e052008-10-30 14:25:28 -04009281 u64 cur_offset = start;
Josef Bacik55a61d12010-11-22 18:50:32 +00009282 u64 i_size;
Chris Mason154ea282013-03-05 11:11:26 -05009283 u64 cur_bytes;
Yan Zhengd899e052008-10-30 14:25:28 -04009284 int ret = 0;
Josef Bacik0af3d002010-06-21 14:48:16 -04009285 bool own_trans = true;
Yan Zhengd899e052008-10-30 14:25:28 -04009286
Josef Bacik0af3d002010-06-21 14:48:16 -04009287 if (trans)
9288 own_trans = false;
Yan Zhengd899e052008-10-30 14:25:28 -04009289 while (num_bytes > 0) {
Josef Bacik0af3d002010-06-21 14:48:16 -04009290 if (own_trans) {
9291 trans = btrfs_start_transaction(root, 3);
9292 if (IS_ERR(trans)) {
9293 ret = PTR_ERR(trans);
9294 break;
9295 }
Yan Zhengd899e052008-10-30 14:25:28 -04009296 }
Yan, Zheng5a303d52009-11-12 09:34:52 +00009297
Chris Mason154ea282013-03-05 11:11:26 -05009298 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
9299 cur_bytes = max(cur_bytes, min_size);
Josef Bacik00361582013-08-14 14:02:47 -04009300 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
Miao Xiee570fd22014-06-19 10:42:50 +08009301 *alloc_hint, &ins, 1, 0);
Yan, Zheng5a303d52009-11-12 09:34:52 +00009302 if (ret) {
Josef Bacik0af3d002010-06-21 14:48:16 -04009303 if (own_trans)
9304 btrfs_end_transaction(trans, root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04009305 break;
Yan, Zheng5a303d52009-11-12 09:34:52 +00009306 }
9307
Yan Zhengd899e052008-10-30 14:25:28 -04009308 ret = insert_reserved_file_extent(trans, inode,
9309 cur_offset, ins.objectid,
9310 ins.offset, ins.offset,
Yan, Zheng920bbbf2009-11-12 09:34:08 +00009311 ins.offset, 0, 0, 0,
Yan Zhengd899e052008-10-30 14:25:28 -04009312 BTRFS_FILE_EXTENT_PREALLOC);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01009313 if (ret) {
Josef Bacik857cc2f2013-10-07 15:21:08 -04009314 btrfs_free_reserved_extent(root, ins.objectid,
Miao Xiee570fd22014-06-19 10:42:50 +08009315 ins.offset, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01009316 btrfs_abort_transaction(trans, root, ret);
9317 if (own_trans)
9318 btrfs_end_transaction(trans, root);
9319 break;
9320 }
Chris Masona1ed8352009-09-11 12:27:37 -04009321 btrfs_drop_extent_cache(inode, cur_offset,
9322 cur_offset + ins.offset -1, 0);
Yan, Zheng5a303d52009-11-12 09:34:52 +00009323
Josef Bacik5dc562c2012-08-17 13:14:17 -04009324 em = alloc_extent_map();
9325 if (!em) {
9326 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9327 &BTRFS_I(inode)->runtime_flags);
9328 goto next;
9329 }
9330
9331 em->start = cur_offset;
9332 em->orig_start = cur_offset;
9333 em->len = ins.offset;
9334 em->block_start = ins.objectid;
9335 em->block_len = ins.offset;
Josef Bacikb4939682012-12-03 10:31:19 -05009336 em->orig_block_len = ins.offset;
Josef Bacikcc95bef2013-04-04 14:31:27 -04009337 em->ram_bytes = ins.offset;
Josef Bacik5dc562c2012-08-17 13:14:17 -04009338 em->bdev = root->fs_info->fs_devices->latest_bdev;
9339 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9340 em->generation = trans->transid;
9341
9342 while (1) {
9343 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -04009344 ret = add_extent_mapping(em_tree, em, 1);
Josef Bacik5dc562c2012-08-17 13:14:17 -04009345 write_unlock(&em_tree->lock);
9346 if (ret != -EEXIST)
9347 break;
9348 btrfs_drop_extent_cache(inode, cur_offset,
9349 cur_offset + ins.offset - 1,
9350 0);
9351 }
9352 free_extent_map(em);
9353next:
Yan Zhengd899e052008-10-30 14:25:28 -04009354 num_bytes -= ins.offset;
9355 cur_offset += ins.offset;
Yan, Zhengefa56462010-05-16 10:49:59 -04009356 *alloc_hint = ins.objectid + ins.offset;
Yan, Zheng5a303d52009-11-12 09:34:52 +00009357
Josef Bacik0c4d2d92012-04-05 15:03:02 -04009358 inode_inc_iversion(inode);
Yan Zhengd899e052008-10-30 14:25:28 -04009359 inode->i_ctime = CURRENT_TIME;
Christoph Hellwig6cbff002009-04-17 10:37:41 +02009360 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
Yan Zhengd899e052008-10-30 14:25:28 -04009361 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
Yan, Zhengefa56462010-05-16 10:49:59 -04009362 (actual_len > inode->i_size) &&
9363 (cur_offset > inode->i_size)) {
Aneesh Kumar K.Vd1ea6a62010-01-20 07:28:54 +00009364 if (cur_offset > actual_len)
Josef Bacik55a61d12010-11-22 18:50:32 +00009365 i_size = actual_len;
Aneesh Kumar K.Vd1ea6a62010-01-20 07:28:54 +00009366 else
Josef Bacik55a61d12010-11-22 18:50:32 +00009367 i_size = cur_offset;
9368 i_size_write(inode, i_size);
9369 btrfs_ordered_update_i_size(inode, i_size, NULL);
Yan, Zheng5a303d52009-11-12 09:34:52 +00009370 }
9371
Yan Zhengd899e052008-10-30 14:25:28 -04009372 ret = btrfs_update_inode(trans, root, inode);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01009373
9374 if (ret) {
9375 btrfs_abort_transaction(trans, root, ret);
9376 if (own_trans)
9377 btrfs_end_transaction(trans, root);
9378 break;
9379 }
Yan Zhengd899e052008-10-30 14:25:28 -04009380
Josef Bacik0af3d002010-06-21 14:48:16 -04009381 if (own_trans)
9382 btrfs_end_transaction(trans, root);
Yan, Zheng5a303d52009-11-12 09:34:52 +00009383 }
Yan Zhengd899e052008-10-30 14:25:28 -04009384 return ret;
9385}
9386
Josef Bacik0af3d002010-06-21 14:48:16 -04009387int btrfs_prealloc_file_range(struct inode *inode, int mode,
9388 u64 start, u64 num_bytes, u64 min_size,
9389 loff_t actual_len, u64 *alloc_hint)
9390{
9391 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9392 min_size, actual_len, alloc_hint,
9393 NULL);
9394}
9395
9396int btrfs_prealloc_file_range_trans(struct inode *inode,
9397 struct btrfs_trans_handle *trans, int mode,
9398 u64 start, u64 num_bytes, u64 min_size,
9399 loff_t actual_len, u64 *alloc_hint)
9400{
9401 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9402 min_size, actual_len, alloc_hint, trans);
9403}
9404
Chris Masone6dcd2d2008-07-17 12:53:50 -04009405static int btrfs_set_page_dirty(struct page *page)
9406{
Chris Masone6dcd2d2008-07-17 12:53:50 -04009407 return __set_page_dirty_nobuffers(page);
9408}
9409
Al Viro10556cb2011-06-20 19:28:19 -04009410static int btrfs_permission(struct inode *inode, int mask)
Yanfdebe2b2008-01-14 13:26:08 -05009411{
Li Zefanb83cc962010-12-20 16:04:08 +08009412 struct btrfs_root *root = BTRFS_I(inode)->root;
Jeff Mahoneycb6db4e2011-08-15 17:27:21 +00009413 umode_t mode = inode->i_mode;
Li Zefanb83cc962010-12-20 16:04:08 +08009414
Jeff Mahoneycb6db4e2011-08-15 17:27:21 +00009415 if (mask & MAY_WRITE &&
9416 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9417 if (btrfs_root_readonly(root))
9418 return -EROFS;
9419 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9420 return -EACCES;
9421 }
Al Viro2830ba72011-06-20 19:16:29 -04009422 return generic_permission(inode, mask);
Yanfdebe2b2008-01-14 13:26:08 -05009423}
Chris Mason39279cc2007-06-12 06:35:45 -04009424
Filipe Mananaef3b9af2014-04-27 20:40:45 +01009425static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9426{
9427 struct btrfs_trans_handle *trans;
9428 struct btrfs_root *root = BTRFS_I(dir)->root;
9429 struct inode *inode = NULL;
9430 u64 objectid;
9431 u64 index;
9432 int ret = 0;
9433
9434 /*
9435 * 5 units required for adding orphan entry
9436 */
9437 trans = btrfs_start_transaction(root, 5);
9438 if (IS_ERR(trans))
9439 return PTR_ERR(trans);
9440
9441 ret = btrfs_find_free_ino(root, &objectid);
9442 if (ret)
9443 goto out;
9444
9445 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9446 btrfs_ino(dir), objectid, mode, &index);
9447 if (IS_ERR(inode)) {
9448 ret = PTR_ERR(inode);
9449 inode = NULL;
9450 goto out;
9451 }
9452
Filipe Mananaef3b9af2014-04-27 20:40:45 +01009453 inode->i_fop = &btrfs_file_operations;
9454 inode->i_op = &btrfs_file_inode_operations;
9455
9456 inode->i_mapping->a_ops = &btrfs_aops;
Filipe Mananaef3b9af2014-04-27 20:40:45 +01009457 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9458
Chris Masonb0d5d102014-09-08 13:08:51 -07009459 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9460 if (ret)
9461 goto out_inode;
9462
9463 ret = btrfs_update_inode(trans, root, inode);
9464 if (ret)
9465 goto out_inode;
Filipe Mananaef3b9af2014-04-27 20:40:45 +01009466 ret = btrfs_orphan_add(trans, inode);
9467 if (ret)
Chris Masonb0d5d102014-09-08 13:08:51 -07009468 goto out_inode;
Filipe Mananaef3b9af2014-04-27 20:40:45 +01009469
Filipe Manana5762b5c2014-08-01 00:10:32 +01009470 /*
9471 * We set number of links to 0 in btrfs_new_inode(), and here we set
9472 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9473 * through:
9474 *
9475 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9476 */
9477 set_nlink(inode, 1);
Chris Masonb0d5d102014-09-08 13:08:51 -07009478 unlock_new_inode(inode);
Filipe Mananaef3b9af2014-04-27 20:40:45 +01009479 d_tmpfile(dentry, inode);
9480 mark_inode_dirty(inode);
9481
9482out:
9483 btrfs_end_transaction(trans, root);
9484 if (ret)
9485 iput(inode);
9486 btrfs_balance_delayed_items(root);
9487 btrfs_btree_balance_dirty(root);
Filipe Mananaef3b9af2014-04-27 20:40:45 +01009488 return ret;
Chris Masonb0d5d102014-09-08 13:08:51 -07009489
9490out_inode:
9491 unlock_new_inode(inode);
9492 goto out;
9493
Filipe Mananaef3b9af2014-04-27 20:40:45 +01009494}
9495
Filipe Mananab38ef712014-11-13 17:01:45 +00009496/* Inspired by filemap_check_errors() */
9497int btrfs_inode_check_errors(struct inode *inode)
9498{
9499 int ret = 0;
9500
9501 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
9502 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
9503 ret = -ENOSPC;
9504 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
9505 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
9506 ret = -EIO;
9507
9508 return ret;
9509}
9510
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07009511static const struct inode_operations btrfs_dir_inode_operations = {
Chris Mason3394e162008-11-17 20:42:26 -05009512 .getattr = btrfs_getattr,
Chris Mason39279cc2007-06-12 06:35:45 -04009513 .lookup = btrfs_lookup,
9514 .create = btrfs_create,
9515 .unlink = btrfs_unlink,
9516 .link = btrfs_link,
9517 .mkdir = btrfs_mkdir,
9518 .rmdir = btrfs_rmdir,
Miklos Szeredi80ace852014-07-23 15:15:32 +02009519 .rename2 = btrfs_rename2,
Chris Mason39279cc2007-06-12 06:35:45 -04009520 .symlink = btrfs_symlink,
9521 .setattr = btrfs_setattr,
Josef Bacik618e21d2007-07-11 10:18:17 -04009522 .mknod = btrfs_mknod,
Christoph Hellwig95819c02008-08-28 06:21:17 -04009523 .setxattr = btrfs_setxattr,
9524 .getxattr = btrfs_getxattr,
Josef Bacik5103e942007-11-16 11:45:54 -05009525 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04009526 .removexattr = btrfs_removexattr,
Yanfdebe2b2008-01-14 13:26:08 -05009527 .permission = btrfs_permission,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02009528 .get_acl = btrfs_get_acl,
Christoph Hellwig996a7102013-12-20 05:16:43 -08009529 .set_acl = btrfs_set_acl,
Guangyu Sun93fd63c2013-09-16 10:42:03 -07009530 .update_time = btrfs_update_time,
Filipe Mananaef3b9af2014-04-27 20:40:45 +01009531 .tmpfile = btrfs_tmpfile,
Chris Mason39279cc2007-06-12 06:35:45 -04009532};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07009533static const struct inode_operations btrfs_dir_ro_inode_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04009534 .lookup = btrfs_lookup,
Yanfdebe2b2008-01-14 13:26:08 -05009535 .permission = btrfs_permission,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02009536 .get_acl = btrfs_get_acl,
Christoph Hellwig996a7102013-12-20 05:16:43 -08009537 .set_acl = btrfs_set_acl,
Guangyu Sun93fd63c2013-09-16 10:42:03 -07009538 .update_time = btrfs_update_time,
Chris Mason39279cc2007-06-12 06:35:45 -04009539};
Yan, Zheng76dda932009-09-21 16:00:26 -04009540
Alexey Dobriyan828c0952009-10-01 15:43:56 -07009541static const struct file_operations btrfs_dir_file_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04009542 .llseek = generic_file_llseek,
9543 .read = generic_read_dir,
Al Viro9cdda8d2013-05-22 16:48:09 -04009544 .iterate = btrfs_real_readdir,
Christoph Hellwig34287aa2007-09-14 10:22:47 -04009545 .unlocked_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -04009546#ifdef CONFIG_COMPAT
Christoph Hellwig34287aa2007-09-14 10:22:47 -04009547 .compat_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -04009548#endif
Sage Weil6bf13c02008-06-10 10:07:39 -04009549 .release = btrfs_release_file,
Chris Masone02119d2008-09-05 16:13:11 -04009550 .fsync = btrfs_sync_file,
Chris Mason39279cc2007-06-12 06:35:45 -04009551};
9552
Chris Masond1310b22008-01-24 16:13:08 -05009553static struct extent_io_ops btrfs_extent_io_ops = {
Chris Mason07157aa2007-08-30 08:50:51 -04009554 .fill_delalloc = run_delalloc_range,
Chris Mason065631f2008-02-20 12:07:25 -05009555 .submit_bio_hook = btrfs_submit_bio_hook,
Chris Mason239b14b2008-03-24 15:02:07 -04009556 .merge_bio_hook = btrfs_merge_bio_hook,
Chris Mason07157aa2007-08-30 08:50:51 -04009557 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
Chris Masone6dcd2d2008-07-17 12:53:50 -04009558 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
Chris Mason247e7432008-07-17 12:53:51 -04009559 .writepage_start_hook = btrfs_writepage_start_hook,
Chris Masonb0c68f82008-01-31 11:05:37 -05009560 .set_bit_hook = btrfs_set_bit_hook,
9561 .clear_bit_hook = btrfs_clear_bit_hook,
Josef Bacik9ed74f22009-09-11 16:12:44 -04009562 .merge_extent_hook = btrfs_merge_extent_hook,
9563 .split_extent_hook = btrfs_split_extent_hook,
Chris Mason07157aa2007-08-30 08:50:51 -04009564};
9565
Chris Mason35054392009-01-21 13:11:13 -05009566/*
9567 * btrfs doesn't support the bmap operation because swapfiles
9568 * use bmap to make a mapping of extents in the file. They assume
9569 * these extents won't change over the life of the file and they
9570 * use the bmap result to do IO directly to the drive.
9571 *
9572 * the btrfs bmap call would return logical addresses that aren't
9573 * suitable for IO and they also will change frequently as COW
9574 * operations happen. So, swapfile + btrfs == corruption.
9575 *
9576 * For now we're avoiding this by dropping bmap.
9577 */
Alexey Dobriyan7f094102009-09-21 17:01:10 -07009578static const struct address_space_operations btrfs_aops = {
Chris Mason39279cc2007-06-12 06:35:45 -04009579 .readpage = btrfs_readpage,
9580 .writepage = btrfs_writepage,
Chris Masonb293f022007-11-01 19:45:34 -04009581 .writepages = btrfs_writepages,
Chris Mason3ab2fb52007-11-08 10:59:22 -05009582 .readpages = btrfs_readpages,
Chris Mason16432982008-04-10 10:23:21 -04009583 .direct_IO = btrfs_direct_IO,
Chris Masona52d9a82007-08-27 16:49:44 -04009584 .invalidatepage = btrfs_invalidatepage,
9585 .releasepage = btrfs_releasepage,
Chris Masone6dcd2d2008-07-17 12:53:50 -04009586 .set_page_dirty = btrfs_set_page_dirty,
Andi Kleen465fdd92009-09-16 11:50:18 +02009587 .error_remove_page = generic_error_remove_page,
Chris Mason39279cc2007-06-12 06:35:45 -04009588};
9589
Alexey Dobriyan7f094102009-09-21 17:01:10 -07009590static const struct address_space_operations btrfs_symlink_aops = {
Chris Mason39279cc2007-06-12 06:35:45 -04009591 .readpage = btrfs_readpage,
9592 .writepage = btrfs_writepage,
Chris Mason2bf5a722007-08-30 11:54:02 -04009593 .invalidatepage = btrfs_invalidatepage,
9594 .releasepage = btrfs_releasepage,
Chris Mason39279cc2007-06-12 06:35:45 -04009595};
9596
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07009597static const struct inode_operations btrfs_file_inode_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04009598 .getattr = btrfs_getattr,
9599 .setattr = btrfs_setattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04009600 .setxattr = btrfs_setxattr,
9601 .getxattr = btrfs_getxattr,
Josef Bacik5103e942007-11-16 11:45:54 -05009602 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04009603 .removexattr = btrfs_removexattr,
Yanfdebe2b2008-01-14 13:26:08 -05009604 .permission = btrfs_permission,
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -05009605 .fiemap = btrfs_fiemap,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02009606 .get_acl = btrfs_get_acl,
Christoph Hellwig996a7102013-12-20 05:16:43 -08009607 .set_acl = btrfs_set_acl,
Josef Bacike41f9412012-03-26 09:46:47 -04009608 .update_time = btrfs_update_time,
Chris Mason39279cc2007-06-12 06:35:45 -04009609};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07009610static const struct inode_operations btrfs_special_inode_operations = {
Josef Bacik618e21d2007-07-11 10:18:17 -04009611 .getattr = btrfs_getattr,
9612 .setattr = btrfs_setattr,
Yanfdebe2b2008-01-14 13:26:08 -05009613 .permission = btrfs_permission,
Christoph Hellwig95819c02008-08-28 06:21:17 -04009614 .setxattr = btrfs_setxattr,
9615 .getxattr = btrfs_getxattr,
Josef Bacik33268ea2008-07-24 12:16:36 -04009616 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04009617 .removexattr = btrfs_removexattr,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02009618 .get_acl = btrfs_get_acl,
Christoph Hellwig996a7102013-12-20 05:16:43 -08009619 .set_acl = btrfs_set_acl,
Josef Bacike41f9412012-03-26 09:46:47 -04009620 .update_time = btrfs_update_time,
Josef Bacik618e21d2007-07-11 10:18:17 -04009621};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07009622static const struct inode_operations btrfs_symlink_inode_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04009623 .readlink = generic_readlink,
9624 .follow_link = page_follow_link_light,
9625 .put_link = page_put_link,
Li Zefanf2095612010-11-19 02:05:24 +00009626 .getattr = btrfs_getattr,
Josef Bacik22c44fe2011-11-30 10:45:38 -05009627 .setattr = btrfs_setattr,
Yanfdebe2b2008-01-14 13:26:08 -05009628 .permission = btrfs_permission,
Jim Owens0279b4c2009-02-04 09:29:13 -05009629 .setxattr = btrfs_setxattr,
9630 .getxattr = btrfs_getxattr,
9631 .listxattr = btrfs_listxattr,
9632 .removexattr = btrfs_removexattr,
Josef Bacike41f9412012-03-26 09:46:47 -04009633 .update_time = btrfs_update_time,
Chris Mason39279cc2007-06-12 06:35:45 -04009634};
Yan, Zheng76dda932009-09-21 16:00:26 -04009635
Alexey Dobriyan82d339d2009-10-09 09:54:36 -04009636const struct dentry_operations btrfs_dentry_operations = {
Yan, Zheng76dda932009-09-21 16:00:26 -04009637 .d_delete = btrfs_dentry_delete,
Josef Bacikb4aff1f2011-06-28 16:18:59 -04009638 .d_release = btrfs_dentry_release,
Yan, Zheng76dda932009-09-21 16:00:26 -04009639};