projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branch 'akpm' (patches from Andrew)
[linux-2.6-block.git] / fs / btrfs / inode.c
CommitLineData
6cbd5570
CM		 1/*
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
8f18cf13 19#include <linux/kernel.h>
065631f6 20#include <linux/bio.h>
39279cc3 21#include <linux/buffer_head.h>
f2eb0a24 22#include <linux/file.h>
39279cc3
CM		 23#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>
39279cc3
CM		 29#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>
9ebefb18 35#include <linux/bit_spinlock.h>
5103e947 36#include <linux/xattr.h>
33268eaf 37#include <linux/posix_acl.h>
d899e052 38#include <linux/falloc.h>
5a0e3ad6 39#include <linux/slab.h>
7a36ddec 40#include <linux/ratelimit.h>
22c44fe6 41#include <linux/mount.h>
55e301fd 42#include <linux/btrfs.h>
53b381b3 43#include <linux/blkdev.h>
f23b5a59 44#include <linux/posix_acl_xattr.h>
e2e40f2c 45#include <linux/uio.h>
39279cc3
CM		 46#include "ctree.h"
47#include "disk-io.h"
48#include "transaction.h"
49#include "btrfs_inode.h"
39279cc3 50#include "print-tree.h"
e6dcd2dc 51#include "ordered-data.h"
95819c05 52#include "xattr.h"
e02119d5 53#include "tree-log.h"
4a54c8c1 54#include "volumes.h"
c8b97818 55#include "compression.h"
b4ce94de 56#include "locking.h"
dc89e982 57#include "free-space-cache.h"
581bb050 58#include "inode-map.h"
38c227d8 59#include "backref.h"
f23b5a59 60#include "hash.h"
63541927 61#include "props.h"
31193213 62#include "qgroup.h"
39279cc3
CM		 63
64struct btrfs_iget_args {
90d3e592 65 struct btrfs_key *location;
39279cc3
CM		 66 struct btrfs_root *root;
67};
68
6e1d5dcc
AD		 69static const struct inode_operations btrfs_dir_inode_operations;
70static const struct inode_operations btrfs_symlink_inode_operations;
71static const struct inode_operations btrfs_dir_ro_inode_operations;
72static const struct inode_operations btrfs_special_inode_operations;
73static const struct inode_operations btrfs_file_inode_operations;
7f09410b
AD		 74static const struct address_space_operations btrfs_aops;
75static const struct address_space_operations btrfs_symlink_aops;
828c0950 76static const struct file_operations btrfs_dir_file_operations;
d1310b2e 77static struct extent_io_ops btrfs_extent_io_ops;
39279cc3
CM		 78
79static struct kmem_cache *btrfs_inode_cachep;
8ccf6f19 80static struct kmem_cache *btrfs_delalloc_work_cachep;
39279cc3
CM		 81struct kmem_cache *btrfs_trans_handle_cachep;
82struct kmem_cache *btrfs_transaction_cachep;
39279cc3 83struct kmem_cache *btrfs_path_cachep;
dc89e982 84struct kmem_cache *btrfs_free_space_cachep;
39279cc3
CM		 85
86#define S_SHIFT 12
87static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
88 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
89 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
90 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
91 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
92 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
93 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
94 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
95};
96
3972f260 97static int btrfs_setsize(struct inode *inode, struct iattr *attr);
a41ad394 98static int btrfs_truncate(struct inode *inode);
5fd02043 99static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
771ed689
CM		 100static noinline int cow_file_range(struct inode *inode,
101 struct page *locked_page,
102 u64 start, u64 end, int *page_started,
103 unsigned long *nr_written, int unlock);
70c8a91c
JB		 104static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
105 u64 len, u64 orig_start,
106 u64 block_start, u64 block_len,
cc95bef6
JB		 107 u64 orig_block_len, u64 ram_bytes,
108 int type);
7b128766 109
48a3b636 110static int btrfs_dirty_inode(struct inode *inode);
7b128766 111
6a3891c5
JB		 112#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
113void btrfs_test_inode_set_ops(struct inode *inode)
114{
115 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
116}
117#endif
118
f34f57a3 119static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
2a7dba39
EP		 120 struct inode *inode, struct inode *dir,
121 const struct qstr *qstr)
0279b4cd
JO		 122{
123 int err;
124
f34f57a3 125 err = btrfs_init_acl(trans, inode, dir);
0279b4cd 126 if (!err)
2a7dba39 127 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
0279b4cd
JO		 128 return err;
129}
130
c8b97818
CM		 131/*
132 * this does all the hard work for inserting an inline extent into
133 * the btree. The caller should have done a btrfs_drop_extents so that
134 * no overlapping inline items exist in the btree
135 */
40f76580 136static int insert_inline_extent(struct btrfs_trans_handle *trans,
1acae57b 137 struct btrfs_path *path, int extent_inserted,
c8b97818
CM		 138 struct btrfs_root *root, struct inode *inode,
139 u64 start, size_t size, size_t compressed_size,
fe3f566c 140 int compress_type,
c8b97818
CM		 141 struct page **compressed_pages)
142{
c8b97818
CM		 143 struct extent_buffer *leaf;
144 struct page *page = NULL;
145 char *kaddr;
146 unsigned long ptr;
147 struct btrfs_file_extent_item *ei;
148 int err = 0;
149 int ret;
150 size_t cur_size = size;
c8b97818 151 unsigned long offset;
c8b97818 152
fe3f566c 153 if (compressed_size && compressed_pages)
c8b97818 154 cur_size = compressed_size;
c8b97818 155
1acae57b 156 inode_add_bytes(inode, size);
c8b97818 157
1acae57b
FDBM		 158 if (!extent_inserted) {
159 struct btrfs_key key;
160 size_t datasize;
c8b97818 161
1acae57b
FDBM		 162 key.objectid = btrfs_ino(inode);
163 key.offset = start;
962a298f 164 key.type = BTRFS_EXTENT_DATA_KEY;
c8b97818 165
1acae57b
FDBM		 166 datasize = btrfs_file_extent_calc_inline_size(cur_size);
167 path->leave_spinning = 1;
168 ret = btrfs_insert_empty_item(trans, root, path, &key,
169 datasize);
170 if (ret) {
171 err = ret;
172 goto fail;
173 }
c8b97818
CM		 174 }
175 leaf = path->nodes[0];
176 ei = btrfs_item_ptr(leaf, path->slots[0],
177 struct btrfs_file_extent_item);
178 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
179 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
180 btrfs_set_file_extent_encryption(leaf, ei, 0);
181 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
182 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
183 ptr = btrfs_file_extent_inline_start(ei);
184
261507a0 185 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818
CM		 186 struct page *cpage;
187 int i = 0;
d397712b 188 while (compressed_size > 0) {
c8b97818 189 cpage = compressed_pages[i];
5b050f04 190 cur_size = min_t(unsigned long, compressed_size,
c8b97818
CM		 191 PAGE_CACHE_SIZE);
192
7ac687d9 193 kaddr = kmap_atomic(cpage);
c8b97818 194 write_extent_buffer(leaf, kaddr, ptr, cur_size);
7ac687d9 195 kunmap_atomic(kaddr);
c8b97818
CM		 196
197 i++;
198 ptr += cur_size;
199 compressed_size -= cur_size;
200 }
201 btrfs_set_file_extent_compression(leaf, ei,
261507a0 202 compress_type);
c8b97818
CM		 203 } else {
204 page = find_get_page(inode->i_mapping,
205 start >> PAGE_CACHE_SHIFT);
206 btrfs_set_file_extent_compression(leaf, ei, 0);
7ac687d9 207 kaddr = kmap_atomic(page);
c8b97818
CM		 208 offset = start & (PAGE_CACHE_SIZE - 1);
209 write_extent_buffer(leaf, kaddr + offset, ptr, size);
7ac687d9 210 kunmap_atomic(kaddr);
c8b97818
CM		 211 page_cache_release(page);
212 }
213 btrfs_mark_buffer_dirty(leaf);
1acae57b 214 btrfs_release_path(path);
c8b97818 215
c2167754
YZ		 216 /*
217 * we're an inline extent, so nobody can
218 * extend the file past i_size without locking
219 * a page we already have locked.
220 *
221 * We must do any isize and inode updates
222 * before we unlock the pages. Otherwise we
223 * could end up racing with unlink.
224 */
c8b97818 225 BTRFS_I(inode)->disk_i_size = inode->i_size;
79787eaa 226 ret = btrfs_update_inode(trans, root, inode);
c2167754 227
79787eaa 228 return ret;
c8b97818 229fail:
c8b97818
CM		 230 return err;
231}
232
233
234/*
235 * conditionally insert an inline extent into the file. This
236 * does the checks required to make sure the data is small enough
237 * to fit as an inline extent.
238 */
00361589
JB		 239static noinline int cow_file_range_inline(struct btrfs_root *root,
240 struct inode *inode, u64 start,
241 u64 end, size_t compressed_size,
242 int compress_type,
243 struct page **compressed_pages)
c8b97818 244{
00361589 245 struct btrfs_trans_handle *trans;
c8b97818
CM		 246 u64 isize = i_size_read(inode);
247 u64 actual_end = min(end + 1, isize);
248 u64 inline_len = actual_end - start;
fda2832f 249 u64 aligned_end = ALIGN(end, root->sectorsize);
c8b97818
CM		 250 u64 data_len = inline_len;
251 int ret;
1acae57b
FDBM		 252 struct btrfs_path *path;
253 int extent_inserted = 0;
254 u32 extent_item_size;
c8b97818
CM		 255
256 if (compressed_size)
257 data_len = compressed_size;
258
259 if (start > 0 ||
354877be
WS		 260 actual_end > PAGE_CACHE_SIZE ||
261 data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
c8b97818
CM		 262 (!compressed_size &&
263 (actual_end & (root->sectorsize - 1)) == 0) ||
264 end + 1 < isize ||
265 data_len > root->fs_info->max_inline) {
266 return 1;
267 }
268
1acae57b
FDBM		 269 path = btrfs_alloc_path();
270 if (!path)
271 return -ENOMEM;
272
00361589 273 trans = btrfs_join_transaction(root);
1acae57b
FDBM		 274 if (IS_ERR(trans)) {
275 btrfs_free_path(path);
00361589 276 return PTR_ERR(trans);
1acae57b 277 }
00361589
JB		 278 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
279
1acae57b
FDBM		 280 if (compressed_size && compressed_pages)
281 extent_item_size = btrfs_file_extent_calc_inline_size(
282 compressed_size);
283 else
284 extent_item_size = btrfs_file_extent_calc_inline_size(
285 inline_len);
286
287 ret = __btrfs_drop_extents(trans, root, inode, path,
288 start, aligned_end, NULL,
289 1, 1, extent_item_size, &extent_inserted);
00361589
JB		 290 if (ret) {
291 btrfs_abort_transaction(trans, root, ret);
292 goto out;
293 }
c8b97818
CM		 294
295 if (isize > actual_end)
296 inline_len = min_t(u64, isize, actual_end);
1acae57b
FDBM		 297 ret = insert_inline_extent(trans, path, extent_inserted,
298 root, inode, start,
c8b97818 299 inline_len, compressed_size,
fe3f566c 300 compress_type, compressed_pages);
2adcac1a 301 if (ret && ret != -ENOSPC) {
79787eaa 302 btrfs_abort_transaction(trans, root, ret);
00361589 303 goto out;
2adcac1a 304 } else if (ret == -ENOSPC) {
00361589
JB		 305 ret = 1;
306 goto out;
79787eaa 307 }
2adcac1a 308
bdc20e67 309 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
0ca1f7ce 310 btrfs_delalloc_release_metadata(inode, end + 1 - start);
a1ed835e 311 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
00361589 312out:
1acae57b 313 btrfs_free_path(path);
00361589
JB		 314 btrfs_end_transaction(trans, root);
315 return ret;
c8b97818
CM		 316}
317
771ed689
CM		 318struct async_extent {
319 u64 start;
320 u64 ram_size;
321 u64 compressed_size;
322 struct page **pages;
323 unsigned long nr_pages;
261507a0 324 int compress_type;
771ed689
CM		 325 struct list_head list;
326};
327
328struct async_cow {
329 struct inode *inode;
330 struct btrfs_root *root;
331 struct page *locked_page;
332 u64 start;
333 u64 end;
334 struct list_head extents;
335 struct btrfs_work work;
336};
337
338static noinline int add_async_extent(struct async_cow *cow,
339 u64 start, u64 ram_size,
340 u64 compressed_size,
341 struct page **pages,
261507a0
LZ		 342 unsigned long nr_pages,
343 int compress_type)
771ed689
CM		 344{
345 struct async_extent *async_extent;
346
347 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
79787eaa 348 BUG_ON(!async_extent); /* -ENOMEM */
771ed689
CM		 349 async_extent->start = start;
350 async_extent->ram_size = ram_size;
351 async_extent->compressed_size = compressed_size;
352 async_extent->pages = pages;
353 async_extent->nr_pages = nr_pages;
261507a0 354 async_extent->compress_type = compress_type;
771ed689
CM		 355 list_add_tail(&async_extent->list, &cow->extents);
356 return 0;
357}
358
f79707b0
WS		 359static inline int inode_need_compress(struct inode *inode)
360{
361 struct btrfs_root *root = BTRFS_I(inode)->root;
362
363 /* force compress */
364 if (btrfs_test_opt(root, FORCE_COMPRESS))
365 return 1;
366 /* bad compression ratios */
367 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
368 return 0;
369 if (btrfs_test_opt(root, COMPRESS) ||
370 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
371 BTRFS_I(inode)->force_compress)
372 return 1;
373 return 0;
374}
375
d352ac68 376/*
771ed689
CM		 377 * we create compressed extents in two phases. The first
378 * phase compresses a range of pages that have already been
379 * locked (both pages and state bits are locked).
c8b97818 380 *
771ed689
CM		 381 * This is done inside an ordered work queue, and the compression
382 * is spread across many cpus. The actual IO submission is step
383 * two, and the ordered work queue takes care of making sure that
384 * happens in the same order things were put onto the queue by
385 * writepages and friends.
c8b97818 386 *
771ed689
CM		 387 * If this code finds it can't get good compression, it puts an
388 * entry onto the work queue to write the uncompressed bytes. This
389 * makes sure that both compressed inodes and uncompressed inodes
b2570314
AB		 390 * are written in the same order that the flusher thread sent them
391 * down.
d352ac68 392 */
c44f649e 393static noinline void compress_file_range(struct inode *inode,
771ed689
CM		 394 struct page *locked_page,
395 u64 start, u64 end,
396 struct async_cow *async_cow,
397 int *num_added)
b888db2b
CM		 398{
399 struct btrfs_root *root = BTRFS_I(inode)->root;
db94535d 400 u64 num_bytes;
db94535d 401 u64 blocksize = root->sectorsize;
c8b97818 402 u64 actual_end;
42dc7bab 403 u64 isize = i_size_read(inode);
e6dcd2dc 404 int ret = 0;
c8b97818
CM		 405 struct page **pages = NULL;
406 unsigned long nr_pages;
407 unsigned long nr_pages_ret = 0;
408 unsigned long total_compressed = 0;
409 unsigned long total_in = 0;
410 unsigned long max_compressed = 128 * 1024;
771ed689 411 unsigned long max_uncompressed = 128 * 1024;
c8b97818
CM		 412 int i;
413 int will_compress;
261507a0 414 int compress_type = root->fs_info->compress_type;
4adaa611 415 int redirty = 0;
b888db2b 416
4cb13e5d
LB		 417 /* if this is a small write inside eof, kick off a defrag */
418 if ((end - start + 1) < 16 * 1024 &&
419 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
4cb5300b
CM		 420 btrfs_add_inode_defrag(NULL, inode);
421
42dc7bab 422 actual_end = min_t(u64, isize, end + 1);
c8b97818
CM		 423again:
424 will_compress = 0;
425 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
426 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
be20aa9d 427
f03d9301
CM		 428 /*
429 * we don't want to send crud past the end of i_size through
430 * compression, that's just a waste of CPU time. So, if the
431 * end of the file is before the start of our current
432 * requested range of bytes, we bail out to the uncompressed
433 * cleanup code that can deal with all of this.
434 *
435 * It isn't really the fastest way to fix things, but this is a
436 * very uncommon corner.
437 */
438 if (actual_end <= start)
439 goto cleanup_and_bail_uncompressed;
440
c8b97818
CM		 441 total_compressed = actual_end - start;
442
4bcbb332
SW		 443 /*
444 * skip compression for a small file range(<=blocksize) that
445 * isn't an inline extent, since it dosen't save disk space at all.
446 */
447 if (total_compressed <= blocksize &&
448 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
449 goto cleanup_and_bail_uncompressed;
450
c8b97818
CM		 451 /* we want to make sure that amount of ram required to uncompress
452 * an extent is reasonable, so we limit the total size in ram
771ed689
CM		 453 * of a compressed extent to 128k. This is a crucial number
454 * because it also controls how easily we can spread reads across
455 * cpus for decompression.
456 *
457 * We also want to make sure the amount of IO required to do
458 * a random read is reasonably small, so we limit the size of
459 * a compressed extent to 128k.
c8b97818
CM		 460 */
461 total_compressed = min(total_compressed, max_uncompressed);
fda2832f 462 num_bytes = ALIGN(end - start + 1, blocksize);
be20aa9d 463 num_bytes = max(blocksize, num_bytes);
c8b97818
CM		 464 total_in = 0;
465 ret = 0;
db94535d 466
771ed689
CM		 467 /*
468 * we do compression for mount -o compress and when the
469 * inode has not been flagged as nocompress. This flag can
470 * change at any time if we discover bad compression ratios.
c8b97818 471 */
f79707b0 472 if (inode_need_compress(inode)) {
c8b97818 473 WARN_ON(pages);
31e818fe 474 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
560f7d75
LZ		 475 if (!pages) {
476 /* just bail out to the uncompressed code */
477 goto cont;
478 }
c8b97818 479
261507a0
LZ		 480 if (BTRFS_I(inode)->force_compress)
481 compress_type = BTRFS_I(inode)->force_compress;
482
4adaa611
CM		 483 /*
484 * we need to call clear_page_dirty_for_io on each
485 * page in the range. Otherwise applications with the file
486 * mmap'd can wander in and change the page contents while
487 * we are compressing them.
488 *
489 * If the compression fails for any reason, we set the pages
490 * dirty again later on.
491 */
492 extent_range_clear_dirty_for_io(inode, start, end);
493 redirty = 1;
261507a0
LZ		 494 ret = btrfs_compress_pages(compress_type,
495 inode->i_mapping, start,
496 total_compressed, pages,
497 nr_pages, &nr_pages_ret,
498 &total_in,
499 &total_compressed,
500 max_compressed);
c8b97818
CM		 501
502 if (!ret) {
503 unsigned long offset = total_compressed &
504 (PAGE_CACHE_SIZE - 1);
505 struct page *page = pages[nr_pages_ret - 1];
506 char *kaddr;
507
508 /* zero the tail end of the last page, we might be
509 * sending it down to disk
510 */
511 if (offset) {
7ac687d9 512 kaddr = kmap_atomic(page);
c8b97818
CM		 513 memset(kaddr + offset, 0,
514 PAGE_CACHE_SIZE - offset);
7ac687d9 515 kunmap_atomic(kaddr);
c8b97818
CM		 516 }
517 will_compress = 1;
518 }
519 }
560f7d75 520cont:
c8b97818
CM		 521 if (start == 0) {
522 /* lets try to make an inline extent */
771ed689 523 if (ret || total_in < (actual_end - start)) {
c8b97818 524 /* we didn't compress the entire range, try
771ed689 525 * to make an uncompressed inline extent.
c8b97818 526 */
00361589
JB		 527 ret = cow_file_range_inline(root, inode, start, end,
528 0, 0, NULL);
c8b97818 529 } else {
771ed689 530 /* try making a compressed inline extent */
00361589 531 ret = cow_file_range_inline(root, inode, start, end,
fe3f566c
LZ		 532 total_compressed,
533 compress_type, pages);
c8b97818 534 }
79787eaa 535 if (ret <= 0) {
151a41bc
JB		 536 unsigned long clear_flags = EXTENT_DELALLOC |
537 EXTENT_DEFRAG;
e6eb4314
FM		 538 unsigned long page_error_op;
539
151a41bc 540 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
e6eb4314 541 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
151a41bc 542
771ed689 543 /*
79787eaa
JM		 544 * inline extent creation worked or returned error,
545 * we don't need to create any more async work items.
546 * Unlock and free up our temp pages.
771ed689 547 */
c2790a2e 548 extent_clear_unlock_delalloc(inode, start, end, NULL,
151a41bc 549 clear_flags, PAGE_UNLOCK |
c2790a2e
JB		 550 PAGE_CLEAR_DIRTY |
551 PAGE_SET_WRITEBACK |
e6eb4314 552 page_error_op |
c2790a2e 553 PAGE_END_WRITEBACK);
c8b97818
CM		 554 goto free_pages_out;
555 }
556 }
557
558 if (will_compress) {
559 /*
560 * we aren't doing an inline extent round the compressed size
561 * up to a block size boundary so the allocator does sane
562 * things
563 */
fda2832f 564 total_compressed = ALIGN(total_compressed, blocksize);
c8b97818
CM		 565
566 /*
567 * one last check to make sure the compression is really a
568 * win, compare the page count read with the blocks on disk
569 */
fda2832f 570 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
c8b97818
CM		 571 if (total_compressed >= total_in) {
572 will_compress = 0;
573 } else {
c8b97818
CM		 574 num_bytes = total_in;
575 }
576 }
577 if (!will_compress && pages) {
578 /*
579 * the compression code ran but failed to make things smaller,
580 * free any pages it allocated and our page pointer array
581 */
582 for (i = 0; i < nr_pages_ret; i++) {
70b99e69 583 WARN_ON(pages[i]->mapping);
c8b97818
CM		 584 page_cache_release(pages[i]);
585 }
586 kfree(pages);
587 pages = NULL;
588 total_compressed = 0;
589 nr_pages_ret = 0;
590
591 /* flag the file so we don't compress in the future */
1e701a32
CM		 592 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
593 !(BTRFS_I(inode)->force_compress)) {
a555f810 594 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
1e701a32 595 }
c8b97818 596 }
771ed689
CM		 597 if (will_compress) {
598 *num_added += 1;
c8b97818 599
771ed689
CM		 600 /* the async work queues will take care of doing actual
601 * allocation on disk for these compressed pages,
602 * and will submit them to the elevator.
603 */
604 add_async_extent(async_cow, start, num_bytes,
261507a0
LZ		 605 total_compressed, pages, nr_pages_ret,
606 compress_type);
179e29e4 607
24ae6365 608 if (start + num_bytes < end) {
771ed689
CM		 609 start += num_bytes;
610 pages = NULL;
611 cond_resched();
612 goto again;
613 }
614 } else {
f03d9301 615cleanup_and_bail_uncompressed:
771ed689
CM		 616 /*
617 * No compression, but we still need to write the pages in
618 * the file we've been given so far. redirty the locked
619 * page if it corresponds to our extent and set things up
620 * for the async work queue to run cow_file_range to do
621 * the normal delalloc dance
622 */
623 if (page_offset(locked_page) >= start &&
624 page_offset(locked_page) <= end) {
625 __set_page_dirty_nobuffers(locked_page);
626 /* unlocked later on in the async handlers */
627 }
4adaa611
CM		 628 if (redirty)
629 extent_range_redirty_for_io(inode, start, end);
261507a0
LZ		 630 add_async_extent(async_cow, start, end - start + 1,
631 0, NULL, 0, BTRFS_COMPRESS_NONE);
771ed689
CM		 632 *num_added += 1;
633 }
3b951516 634
c44f649e 635 return;
771ed689
CM		 636
637free_pages_out:
638 for (i = 0; i < nr_pages_ret; i++) {
639 WARN_ON(pages[i]->mapping);
640 page_cache_release(pages[i]);
641 }
d397712b 642 kfree(pages);
771ed689 643}
771ed689 644
40ae837b
FM		 645static void free_async_extent_pages(struct async_extent *async_extent)
646{
647 int i;
648
649 if (!async_extent->pages)
650 return;
651
652 for (i = 0; i < async_extent->nr_pages; i++) {
653 WARN_ON(async_extent->pages[i]->mapping);
654 page_cache_release(async_extent->pages[i]);
655 }
656 kfree(async_extent->pages);
657 async_extent->nr_pages = 0;
658 async_extent->pages = NULL;
771ed689
CM		 659}
660
661/*
662 * phase two of compressed writeback. This is the ordered portion
663 * of the code, which only gets called in the order the work was
664 * queued. We walk all the async extents created by compress_file_range
665 * and send them down to the disk.
666 */
dec8f175 667static noinline void submit_compressed_extents(struct inode *inode,
771ed689
CM		 668 struct async_cow *async_cow)
669{
670 struct async_extent *async_extent;
671 u64 alloc_hint = 0;
771ed689
CM		 672 struct btrfs_key ins;
673 struct extent_map *em;
674 struct btrfs_root *root = BTRFS_I(inode)->root;
675 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
676 struct extent_io_tree *io_tree;
f5a84ee3 677 int ret = 0;
771ed689 678
3e04e7f1 679again:
d397712b 680 while (!list_empty(&async_cow->extents)) {
771ed689
CM		 681 async_extent = list_entry(async_cow->extents.next,
682 struct async_extent, list);
683 list_del(&async_extent->list);
c8b97818 684
771ed689
CM		 685 io_tree = &BTRFS_I(inode)->io_tree;
686
f5a84ee3 687retry:
771ed689
CM		 688 /* did the compression code fall back to uncompressed IO? */
689 if (!async_extent->pages) {
690 int page_started = 0;
691 unsigned long nr_written = 0;
692
693 lock_extent(io_tree, async_extent->start,
2ac55d41 694 async_extent->start +
d0082371 695 async_extent->ram_size - 1);
771ed689
CM		 696
697 /* allocate blocks */
f5a84ee3
JB		 698 ret = cow_file_range(inode, async_cow->locked_page,
699 async_extent->start,
700 async_extent->start +
701 async_extent->ram_size - 1,
702 &page_started, &nr_written, 0);
771ed689 703
79787eaa
JM		 704 /* JDM XXX */
705
771ed689
CM		 706 /*
707 * if page_started, cow_file_range inserted an
708 * inline extent and took care of all the unlocking
709 * and IO for us. Otherwise, we need to submit
710 * all those pages down to the drive.
711 */
f5a84ee3 712 if (!page_started && !ret)
771ed689
CM		 713 extent_write_locked_range(io_tree,
714 inode, async_extent->start,
d397712b 715 async_extent->start +
771ed689
CM		 716 async_extent->ram_size - 1,
717 btrfs_get_extent,
718 WB_SYNC_ALL);
3e04e7f1
JB		 719 else if (ret)
720 unlock_page(async_cow->locked_page);
771ed689
CM		 721 kfree(async_extent);
722 cond_resched();
723 continue;
724 }
725
726 lock_extent(io_tree, async_extent->start,
d0082371 727 async_extent->start + async_extent->ram_size - 1);
771ed689 728
00361589 729 ret = btrfs_reserve_extent(root,
771ed689
CM		 730 async_extent->compressed_size,
731 async_extent->compressed_size,
e570fd27 732 0, alloc_hint, &ins, 1, 1);
f5a84ee3 733 if (ret) {
40ae837b 734 free_async_extent_pages(async_extent);
3e04e7f1 735
fdf8e2ea
JB		 736 if (ret == -ENOSPC) {
737 unlock_extent(io_tree, async_extent->start,
738 async_extent->start +
739 async_extent->ram_size - 1);
ce62003f
LB		 740
741 /*
742 * we need to redirty the pages if we decide to
743 * fallback to uncompressed IO, otherwise we
744 * will not submit these pages down to lower
745 * layers.
746 */
747 extent_range_redirty_for_io(inode,
748 async_extent->start,
749 async_extent->start +
750 async_extent->ram_size - 1);
751
79787eaa 752 goto retry;
fdf8e2ea 753 }
3e04e7f1 754 goto out_free;
f5a84ee3 755 }
c2167754
YZ		 756 /*
757 * here we're doing allocation and writeback of the
758 * compressed pages
759 */
760 btrfs_drop_extent_cache(inode, async_extent->start,
761 async_extent->start +
762 async_extent->ram_size - 1, 0);
763
172ddd60 764 em = alloc_extent_map();
b9aa55be
LB		 765 if (!em) {
766 ret = -ENOMEM;
3e04e7f1 767 goto out_free_reserve;
b9aa55be 768 }
771ed689
CM		 769 em->start = async_extent->start;
770 em->len = async_extent->ram_size;
445a6944 771 em->orig_start = em->start;
2ab28f32
JB		 772 em->mod_start = em->start;
773 em->mod_len = em->len;
c8b97818 774
771ed689
CM		 775 em->block_start = ins.objectid;
776 em->block_len = ins.offset;
b4939680 777 em->orig_block_len = ins.offset;
cc95bef6 778 em->ram_bytes = async_extent->ram_size;
771ed689 779 em->bdev = root->fs_info->fs_devices->latest_bdev;
261507a0 780 em->compress_type = async_extent->compress_type;
771ed689
CM		 781 set_bit(EXTENT_FLAG_PINNED, &em->flags);
782 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
70c8a91c 783 em->generation = -1;
771ed689 784
d397712b 785 while (1) {
890871be 786 write_lock(&em_tree->lock);
09a2a8f9 787 ret = add_extent_mapping(em_tree, em, 1);
890871be 788 write_unlock(&em_tree->lock);
771ed689
CM		 789 if (ret != -EEXIST) {
790 free_extent_map(em);
791 break;
792 }
793 btrfs_drop_extent_cache(inode, async_extent->start,
794 async_extent->start +
795 async_extent->ram_size - 1, 0);
796 }
797
3e04e7f1
JB		 798 if (ret)
799 goto out_free_reserve;
800
261507a0
LZ		 801 ret = btrfs_add_ordered_extent_compress(inode,
802 async_extent->start,
803 ins.objectid,
804 async_extent->ram_size,
805 ins.offset,
806 BTRFS_ORDERED_COMPRESSED,
807 async_extent->compress_type);
d9f85963
FM		 808 if (ret) {
809 btrfs_drop_extent_cache(inode, async_extent->start,
810 async_extent->start +
811 async_extent->ram_size - 1, 0);
3e04e7f1 812 goto out_free_reserve;
d9f85963 813 }
771ed689 814
771ed689
CM		 815 /*
816 * clear dirty, set writeback and unlock the pages.
817 */
c2790a2e 818 extent_clear_unlock_delalloc(inode, async_extent->start,
a791e35e
CM		 819 async_extent->start +
820 async_extent->ram_size - 1,
151a41bc
JB		 821 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
822 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
c2790a2e 823 PAGE_SET_WRITEBACK);
771ed689 824 ret = btrfs_submit_compressed_write(inode,
d397712b
CM		 825 async_extent->start,
826 async_extent->ram_size,
827 ins.objectid,
828 ins.offset, async_extent->pages,
829 async_extent->nr_pages);
fce2a4e6
FM		 830 if (ret) {
831 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
832 struct page *p = async_extent->pages[0];
833 const u64 start = async_extent->start;
834 const u64 end = start + async_extent->ram_size - 1;
835
836 p->mapping = inode->i_mapping;
837 tree->ops->writepage_end_io_hook(p, start, end,
838 NULL, 0);
839 p->mapping = NULL;
840 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
841 PAGE_END_WRITEBACK |
842 PAGE_SET_ERROR);
40ae837b 843 free_async_extent_pages(async_extent);
fce2a4e6 844 }
771ed689
CM		 845 alloc_hint = ins.objectid + ins.offset;
846 kfree(async_extent);
847 cond_resched();
848 }
dec8f175 849 return;
3e04e7f1 850out_free_reserve:
e570fd27 851 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
79787eaa 852out_free:
c2790a2e 853 extent_clear_unlock_delalloc(inode, async_extent->start,
3e04e7f1
JB		 854 async_extent->start +
855 async_extent->ram_size - 1,
c2790a2e 856 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc
JB		 857 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
858 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
704de49d
FM		 859 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
860 PAGE_SET_ERROR);
40ae837b 861 free_async_extent_pages(async_extent);
79787eaa 862 kfree(async_extent);
3e04e7f1 863 goto again;
771ed689
CM		 864}
865
4b46fce2
JB		 866static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
867 u64 num_bytes)
868{
869 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
870 struct extent_map *em;
871 u64 alloc_hint = 0;
872
873 read_lock(&em_tree->lock);
874 em = search_extent_mapping(em_tree, start, num_bytes);
875 if (em) {
876 /*
877 * if block start isn't an actual block number then find the
878 * first block in this inode and use that as a hint. If that
879 * block is also bogus then just don't worry about it.
880 */
881 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
882 free_extent_map(em);
883 em = search_extent_mapping(em_tree, 0, 0);
884 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
885 alloc_hint = em->block_start;
886 if (em)
887 free_extent_map(em);
888 } else {
889 alloc_hint = em->block_start;
890 free_extent_map(em);
891 }
892 }
893 read_unlock(&em_tree->lock);
894
895 return alloc_hint;
896}
897
771ed689
CM		 898/*
899 * when extent_io.c finds a delayed allocation range in the file,
900 * the call backs end up in this code. The basic idea is to
901 * allocate extents on disk for the range, and create ordered data structs
902 * in ram to track those extents.
903 *
904 * locked_page is the page that writepage had locked already. We use
905 * it to make sure we don't do extra locks or unlocks.
906 *
907 * *page_started is set to one if we unlock locked_page and do everything
908 * required to start IO on it. It may be clean and already done with
909 * IO when we return.
910 */
00361589
JB		 911static noinline int cow_file_range(struct inode *inode,
912 struct page *locked_page,
913 u64 start, u64 end, int *page_started,
914 unsigned long *nr_written,
915 int unlock)
771ed689 916{
00361589 917 struct btrfs_root *root = BTRFS_I(inode)->root;
771ed689
CM		 918 u64 alloc_hint = 0;
919 u64 num_bytes;
920 unsigned long ram_size;
921 u64 disk_num_bytes;
922 u64 cur_alloc_size;
923 u64 blocksize = root->sectorsize;
771ed689
CM		 924 struct btrfs_key ins;
925 struct extent_map *em;
926 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
927 int ret = 0;
928
02ecd2c2
JB		 929 if (btrfs_is_free_space_inode(inode)) {
930 WARN_ON_ONCE(1);
29bce2f3
JB		 931 ret = -EINVAL;
932 goto out_unlock;
02ecd2c2 933 }
771ed689 934
fda2832f 935 num_bytes = ALIGN(end - start + 1, blocksize);
771ed689
CM		 936 num_bytes = max(blocksize, num_bytes);
937 disk_num_bytes = num_bytes;
771ed689 938
4cb5300b 939 /* if this is a small write inside eof, kick off defrag */
4cb13e5d
LB		 940 if (num_bytes < 64 * 1024 &&
941 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
00361589 942 btrfs_add_inode_defrag(NULL, inode);
4cb5300b 943
771ed689
CM		 944 if (start == 0) {
945 /* lets try to make an inline extent */
00361589
JB		 946 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
947 NULL);
771ed689 948 if (ret == 0) {
c2790a2e
JB		 949 extent_clear_unlock_delalloc(inode, start, end, NULL,
950 EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc 951 EXTENT_DEFRAG, PAGE_UNLOCK |
c2790a2e
JB		 952 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
953 PAGE_END_WRITEBACK);
c2167754 954
771ed689
CM		 955 *nr_written = *nr_written +
956 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
957 *page_started = 1;
771ed689 958 goto out;
79787eaa 959 } else if (ret < 0) {
79787eaa 960 goto out_unlock;
771ed689
CM		 961 }
962 }
963
964 BUG_ON(disk_num_bytes >
6c41761f 965 btrfs_super_total_bytes(root->fs_info->super_copy));
771ed689 966
4b46fce2 967 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
771ed689
CM		 968 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
969
d397712b 970 while (disk_num_bytes > 0) {
a791e35e
CM		 971 unsigned long op;
972
287a0ab9 973 cur_alloc_size = disk_num_bytes;
00361589 974 ret = btrfs_reserve_extent(root, cur_alloc_size,
771ed689 975 root->sectorsize, 0, alloc_hint,
e570fd27 976 &ins, 1, 1);
00361589 977 if (ret < 0)
79787eaa 978 goto out_unlock;
d397712b 979
172ddd60 980 em = alloc_extent_map();
b9aa55be
LB		 981 if (!em) {
982 ret = -ENOMEM;
ace68bac 983 goto out_reserve;
b9aa55be 984 }
e6dcd2dc 985 em->start = start;
445a6944 986 em->orig_start = em->start;
771ed689
CM		 987 ram_size = ins.offset;
988 em->len = ins.offset;
2ab28f32
JB		 989 em->mod_start = em->start;
990 em->mod_len = em->len;
c8b97818 991
e6dcd2dc 992 em->block_start = ins.objectid;
c8b97818 993 em->block_len = ins.offset;
b4939680 994 em->orig_block_len = ins.offset;
cc95bef6 995 em->ram_bytes = ram_size;
e6dcd2dc 996 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 997 set_bit(EXTENT_FLAG_PINNED, &em->flags);
70c8a91c 998 em->generation = -1;
c8b97818 999
d397712b 1000 while (1) {
890871be 1001 write_lock(&em_tree->lock);
09a2a8f9 1002 ret = add_extent_mapping(em_tree, em, 1);
890871be 1003 write_unlock(&em_tree->lock);
e6dcd2dc
CM		 1004 if (ret != -EEXIST) {
1005 free_extent_map(em);
1006 break;
1007 }
1008 btrfs_drop_extent_cache(inode, start,
c8b97818 1009 start + ram_size - 1, 0);
e6dcd2dc 1010 }
ace68bac
LB		 1011 if (ret)
1012 goto out_reserve;
e6dcd2dc 1013
98d20f67 1014 cur_alloc_size = ins.offset;
e6dcd2dc 1015 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771ed689 1016 ram_size, cur_alloc_size, 0);
ace68bac 1017 if (ret)
d9f85963 1018 goto out_drop_extent_cache;
c8b97818 1019
17d217fe
YZ		 1020 if (root->root_key.objectid ==
1021 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1022 ret = btrfs_reloc_clone_csums(inode, start,
1023 cur_alloc_size);
00361589 1024 if (ret)
d9f85963 1025 goto out_drop_extent_cache;
17d217fe
YZ		 1026 }
1027
d397712b 1028 if (disk_num_bytes < cur_alloc_size)
3b951516 1029 break;
d397712b 1030
c8b97818
CM		 1031 /* we're not doing compressed IO, don't unlock the first
1032 * page (which the caller expects to stay locked), don't
1033 * clear any dirty bits and don't set any writeback bits
8b62b72b
CM		 1034 *
1035 * Do set the Private2 bit so we know this page was properly
1036 * setup for writepage
c8b97818 1037 */
c2790a2e
JB		 1038 op = unlock ? PAGE_UNLOCK : 0;
1039 op |= PAGE_SET_PRIVATE2;
a791e35e 1040
c2790a2e
JB		 1041 extent_clear_unlock_delalloc(inode, start,
1042 start + ram_size - 1, locked_page,
1043 EXTENT_LOCKED | EXTENT_DELALLOC,
1044 op);
c8b97818 1045 disk_num_bytes -= cur_alloc_size;
c59f8951
CM		 1046 num_bytes -= cur_alloc_size;
1047 alloc_hint = ins.objectid + ins.offset;
1048 start += cur_alloc_size;
b888db2b 1049 }
79787eaa 1050out:
be20aa9d 1051 return ret;
b7d5b0a8 1052
d9f85963
FM		 1053out_drop_extent_cache:
1054 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
ace68bac 1055out_reserve:
e570fd27 1056 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
79787eaa 1057out_unlock:
c2790a2e 1058 extent_clear_unlock_delalloc(inode, start, end, locked_page,
151a41bc
JB		 1059 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1060 EXTENT_DELALLOC | EXTENT_DEFRAG,
1061 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1062 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
79787eaa 1063 goto out;
771ed689 1064}
c8b97818 1065
771ed689
CM		 1066/*
1067 * work queue call back to started compression on a file and pages
1068 */
1069static noinline void async_cow_start(struct btrfs_work *work)
1070{
1071 struct async_cow *async_cow;
1072 int num_added = 0;
1073 async_cow = container_of(work, struct async_cow, work);
1074
1075 compress_file_range(async_cow->inode, async_cow->locked_page,
1076 async_cow->start, async_cow->end, async_cow,
1077 &num_added);
8180ef88 1078 if (num_added == 0) {
cb77fcd8 1079 btrfs_add_delayed_iput(async_cow->inode);
771ed689 1080 async_cow->inode = NULL;
8180ef88 1081 }
771ed689
CM		 1082}
1083
1084/*
1085 * work queue call back to submit previously compressed pages
1086 */
1087static noinline void async_cow_submit(struct btrfs_work *work)
1088{
1089 struct async_cow *async_cow;
1090 struct btrfs_root *root;
1091 unsigned long nr_pages;
1092
1093 async_cow = container_of(work, struct async_cow, work);
1094
1095 root = async_cow->root;
1096 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1097 PAGE_CACHE_SHIFT;
1098
66657b31 1099 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
287082b0 1100 5 * 1024 * 1024 &&
771ed689
CM		 1101 waitqueue_active(&root->fs_info->async_submit_wait))
1102 wake_up(&root->fs_info->async_submit_wait);
1103
d397712b 1104 if (async_cow->inode)
771ed689 1105 submit_compressed_extents(async_cow->inode, async_cow);
771ed689 1106}
c8b97818 1107
771ed689
CM		 1108static noinline void async_cow_free(struct btrfs_work *work)
1109{
1110 struct async_cow *async_cow;
1111 async_cow = container_of(work, struct async_cow, work);
8180ef88 1112 if (async_cow->inode)
cb77fcd8 1113 btrfs_add_delayed_iput(async_cow->inode);
771ed689
CM		 1114 kfree(async_cow);
1115}
1116
1117static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1118 u64 start, u64 end, int *page_started,
1119 unsigned long *nr_written)
1120{
1121 struct async_cow *async_cow;
1122 struct btrfs_root *root = BTRFS_I(inode)->root;
1123 unsigned long nr_pages;
1124 u64 cur_end;
287082b0 1125 int limit = 10 * 1024 * 1024;
771ed689 1126
a3429ab7
CM		 1127 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1128 1, 0, NULL, GFP_NOFS);
d397712b 1129 while (start < end) {
771ed689 1130 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
79787eaa 1131 BUG_ON(!async_cow); /* -ENOMEM */
8180ef88 1132 async_cow->inode = igrab(inode);
771ed689
CM		 1133 async_cow->root = root;
1134 async_cow->locked_page = locked_page;
1135 async_cow->start = start;
1136
f79707b0
WS		 1137 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1138 !btrfs_test_opt(root, FORCE_COMPRESS))
771ed689
CM		 1139 cur_end = end;
1140 else
1141 cur_end = min(end, start + 512 * 1024 - 1);
1142
1143 async_cow->end = cur_end;
1144 INIT_LIST_HEAD(&async_cow->extents);
1145
9e0af237
LB		 1146 btrfs_init_work(&async_cow->work,
1147 btrfs_delalloc_helper,
1148 async_cow_start, async_cow_submit,
1149 async_cow_free);
771ed689 1150
771ed689
CM		 1151 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1152 PAGE_CACHE_SHIFT;
1153 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1154
afe3d242
QW		 1155 btrfs_queue_work(root->fs_info->delalloc_workers,
1156 &async_cow->work);
771ed689
CM		 1157
1158 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1159 wait_event(root->fs_info->async_submit_wait,
1160 (atomic_read(&root->fs_info->async_delalloc_pages) <
1161 limit));
1162 }
1163
d397712b 1164 while (atomic_read(&root->fs_info->async_submit_draining) &&
771ed689
CM		 1165 atomic_read(&root->fs_info->async_delalloc_pages)) {
1166 wait_event(root->fs_info->async_submit_wait,
1167 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1168 0));
1169 }
1170
1171 *nr_written += nr_pages;
1172 start = cur_end + 1;
1173 }
1174 *page_started = 1;
1175 return 0;
be20aa9d
CM		 1176}
1177
d397712b 1178static noinline int csum_exist_in_range(struct btrfs_root *root,
17d217fe
YZ		 1179 u64 bytenr, u64 num_bytes)
1180{
1181 int ret;
1182 struct btrfs_ordered_sum *sums;
1183 LIST_HEAD(list);
1184
07d400a6 1185 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
a2de733c 1186 bytenr + num_bytes - 1, &list, 0);
17d217fe
YZ		 1187 if (ret == 0 && list_empty(&list))
1188 return 0;
1189
1190 while (!list_empty(&list)) {
1191 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1192 list_del(&sums->list);
1193 kfree(sums);
1194 }
1195 return 1;
1196}
1197
d352ac68
CM		 1198/*
1199 * when nowcow writeback call back. This checks for snapshots or COW copies
1200 * of the extents that exist in the file, and COWs the file as required.
1201 *
1202 * If no cow copies or snapshots exist, we write directly to the existing
1203 * blocks on disk
1204 */
7f366cfe
CM		 1205static noinline int run_delalloc_nocow(struct inode *inode,
1206 struct page *locked_page,
771ed689
CM		 1207 u64 start, u64 end, int *page_started, int force,
1208 unsigned long *nr_written)
be20aa9d 1209{
be20aa9d 1210 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 1211 struct btrfs_trans_handle *trans;
be20aa9d 1212 struct extent_buffer *leaf;
be20aa9d 1213 struct btrfs_path *path;
80ff3856 1214 struct btrfs_file_extent_item *fi;
be20aa9d 1215 struct btrfs_key found_key;
80ff3856
YZ		 1216 u64 cow_start;
1217 u64 cur_offset;
1218 u64 extent_end;
5d4f98a2 1219 u64 extent_offset;
80ff3856
YZ		 1220 u64 disk_bytenr;
1221 u64 num_bytes;
b4939680 1222 u64 disk_num_bytes;
cc95bef6 1223 u64 ram_bytes;
80ff3856 1224 int extent_type;
79787eaa 1225 int ret, err;
d899e052 1226 int type;
80ff3856
YZ		 1227 int nocow;
1228 int check_prev = 1;
82d5902d 1229 bool nolock;
33345d01 1230 u64 ino = btrfs_ino(inode);
be20aa9d
CM		 1231
1232 path = btrfs_alloc_path();
17ca04af 1233 if (!path) {
c2790a2e
JB		 1234 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1235 EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc
JB		 1236 EXTENT_DO_ACCOUNTING |
1237 EXTENT_DEFRAG, PAGE_UNLOCK |
c2790a2e
JB		 1238 PAGE_CLEAR_DIRTY |
1239 PAGE_SET_WRITEBACK |
1240 PAGE_END_WRITEBACK);
d8926bb3 1241 return -ENOMEM;
17ca04af 1242 }
82d5902d 1243
83eea1f1 1244 nolock = btrfs_is_free_space_inode(inode);
82d5902d
LZ		 1245
1246 if (nolock)
7a7eaa40 1247 trans = btrfs_join_transaction_nolock(root);
82d5902d 1248 else
7a7eaa40 1249 trans = btrfs_join_transaction(root);
ff5714cc 1250
79787eaa 1251 if (IS_ERR(trans)) {
c2790a2e
JB		 1252 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1253 EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc
JB		 1254 EXTENT_DO_ACCOUNTING |
1255 EXTENT_DEFRAG, PAGE_UNLOCK |
c2790a2e
JB		 1256 PAGE_CLEAR_DIRTY |
1257 PAGE_SET_WRITEBACK |
1258 PAGE_END_WRITEBACK);
79787eaa
JM		 1259 btrfs_free_path(path);
1260 return PTR_ERR(trans);
1261 }
1262
74b21075 1263 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
be20aa9d 1264
80ff3856
YZ		 1265 cow_start = (u64)-1;
1266 cur_offset = start;
1267 while (1) {
33345d01 1268 ret = btrfs_lookup_file_extent(trans, root, path, ino,
80ff3856 1269 cur_offset, 0);
d788a349 1270 if (ret < 0)
79787eaa 1271 goto error;
80ff3856
YZ		 1272 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1273 leaf = path->nodes[0];
1274 btrfs_item_key_to_cpu(leaf, &found_key,
1275 path->slots[0] - 1);
33345d01 1276 if (found_key.objectid == ino &&
80ff3856
YZ		 1277 found_key.type == BTRFS_EXTENT_DATA_KEY)
1278 path->slots[0]--;
1279 }
1280 check_prev = 0;
1281next_slot:
1282 leaf = path->nodes[0];
1283 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1284 ret = btrfs_next_leaf(root, path);
d788a349 1285 if (ret < 0)
79787eaa 1286 goto error;
80ff3856
YZ		 1287 if (ret > 0)
1288 break;
1289 leaf = path->nodes[0];
1290 }
be20aa9d 1291
80ff3856
YZ		 1292 nocow = 0;
1293 disk_bytenr = 0;
17d217fe 1294 num_bytes = 0;
80ff3856
YZ		 1295 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1296
33345d01 1297 if (found_key.objectid > ino ||
80ff3856
YZ		 1298 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1299 found_key.offset > end)
1300 break;
1301
1302 if (found_key.offset > cur_offset) {
1303 extent_end = found_key.offset;
e9061e21 1304 extent_type = 0;
80ff3856
YZ		 1305 goto out_check;
1306 }
1307
1308 fi = btrfs_item_ptr(leaf, path->slots[0],
1309 struct btrfs_file_extent_item);
1310 extent_type = btrfs_file_extent_type(leaf, fi);
1311
cc95bef6 1312 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
d899e052
YZ		 1313 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1314 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856 1315 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5d4f98a2 1316 extent_offset = btrfs_file_extent_offset(leaf, fi);
80ff3856
YZ		 1317 extent_end = found_key.offset +
1318 btrfs_file_extent_num_bytes(leaf, fi);
b4939680
JB		 1319 disk_num_bytes =
1320 btrfs_file_extent_disk_num_bytes(leaf, fi);
80ff3856
YZ		 1321 if (extent_end <= start) {
1322 path->slots[0]++;
1323 goto next_slot;
1324 }
17d217fe
YZ		 1325 if (disk_bytenr == 0)
1326 goto out_check;
80ff3856
YZ		 1327 if (btrfs_file_extent_compression(leaf, fi) ||
1328 btrfs_file_extent_encryption(leaf, fi) ||
1329 btrfs_file_extent_other_encoding(leaf, fi))
1330 goto out_check;
d899e052
YZ		 1331 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1332 goto out_check;
d2fb3437 1333 if (btrfs_extent_readonly(root, disk_bytenr))
80ff3856 1334 goto out_check;
33345d01 1335 if (btrfs_cross_ref_exist(trans, root, ino,
5d4f98a2
YZ		 1336 found_key.offset -
1337 extent_offset, disk_bytenr))
17d217fe 1338 goto out_check;
5d4f98a2 1339 disk_bytenr += extent_offset;
17d217fe
YZ		 1340 disk_bytenr += cur_offset - found_key.offset;
1341 num_bytes = min(end + 1, extent_end) - cur_offset;
e9894fd3
WS		 1342 /*
1343 * if there are pending snapshots for this root,
1344 * we fall into common COW way.
1345 */
1346 if (!nolock) {
9ea24bbe 1347 err = btrfs_start_write_no_snapshoting(root);
e9894fd3
WS		 1348 if (!err)
1349 goto out_check;
1350 }
17d217fe
YZ		 1351 /*
1352 * force cow if csum exists in the range.
1353 * this ensure that csum for a given extent are
1354 * either valid or do not exist.
1355 */
1356 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1357 goto out_check;
80ff3856
YZ		 1358 nocow = 1;
1359 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1360 extent_end = found_key.offset +
514ac8ad
CM		 1361 btrfs_file_extent_inline_len(leaf,
1362 path->slots[0], fi);
80ff3856
YZ		 1363 extent_end = ALIGN(extent_end, root->sectorsize);
1364 } else {
1365 BUG_ON(1);
1366 }
1367out_check:
1368 if (extent_end <= start) {
1369 path->slots[0]++;
e9894fd3 1370 if (!nolock && nocow)
9ea24bbe 1371 btrfs_end_write_no_snapshoting(root);
80ff3856
YZ		 1372 goto next_slot;
1373 }
1374 if (!nocow) {
1375 if (cow_start == (u64)-1)
1376 cow_start = cur_offset;
1377 cur_offset = extent_end;
1378 if (cur_offset > end)
1379 break;
1380 path->slots[0]++;
1381 goto next_slot;
7ea394f1
YZ		 1382 }
1383
b3b4aa74 1384 btrfs_release_path(path);
80ff3856 1385 if (cow_start != (u64)-1) {
00361589
JB		 1386 ret = cow_file_range(inode, locked_page,
1387 cow_start, found_key.offset - 1,
1388 page_started, nr_written, 1);
e9894fd3
WS		 1389 if (ret) {
1390 if (!nolock && nocow)
9ea24bbe 1391 btrfs_end_write_no_snapshoting(root);
79787eaa 1392 goto error;
e9894fd3 1393 }
80ff3856 1394 cow_start = (u64)-1;
7ea394f1 1395 }
80ff3856 1396
d899e052
YZ		 1397 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1398 struct extent_map *em;
1399 struct extent_map_tree *em_tree;
1400 em_tree = &BTRFS_I(inode)->extent_tree;
172ddd60 1401 em = alloc_extent_map();
79787eaa 1402 BUG_ON(!em); /* -ENOMEM */
d899e052 1403 em->start = cur_offset;
70c8a91c 1404 em->orig_start = found_key.offset - extent_offset;
d899e052
YZ		 1405 em->len = num_bytes;
1406 em->block_len = num_bytes;
1407 em->block_start = disk_bytenr;
b4939680 1408 em->orig_block_len = disk_num_bytes;
cc95bef6 1409 em->ram_bytes = ram_bytes;
d899e052 1410 em->bdev = root->fs_info->fs_devices->latest_bdev;
2ab28f32
JB		 1411 em->mod_start = em->start;
1412 em->mod_len = em->len;
d899e052 1413 set_bit(EXTENT_FLAG_PINNED, &em->flags);
b11e234d 1414 set_bit(EXTENT_FLAG_FILLING, &em->flags);
70c8a91c 1415 em->generation = -1;
d899e052 1416 while (1) {
890871be 1417 write_lock(&em_tree->lock);
09a2a8f9 1418 ret = add_extent_mapping(em_tree, em, 1);
890871be 1419 write_unlock(&em_tree->lock);
d899e052
YZ		 1420 if (ret != -EEXIST) {
1421 free_extent_map(em);
1422 break;
1423 }
1424 btrfs_drop_extent_cache(inode, em->start,
1425 em->start + em->len - 1, 0);
1426 }
1427 type = BTRFS_ORDERED_PREALLOC;
1428 } else {
1429 type = BTRFS_ORDERED_NOCOW;
1430 }
80ff3856
YZ		 1431
1432 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052 1433 num_bytes, num_bytes, type);
79787eaa 1434 BUG_ON(ret); /* -ENOMEM */
771ed689 1435
efa56464
YZ		 1436 if (root->root_key.objectid ==
1437 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1438 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1439 num_bytes);
e9894fd3
WS		 1440 if (ret) {
1441 if (!nolock && nocow)
9ea24bbe 1442 btrfs_end_write_no_snapshoting(root);
79787eaa 1443 goto error;
e9894fd3 1444 }
efa56464
YZ		 1445 }
1446
c2790a2e
JB		 1447 extent_clear_unlock_delalloc(inode, cur_offset,
1448 cur_offset + num_bytes - 1,
1449 locked_page, EXTENT_LOCKED |
1450 EXTENT_DELALLOC, PAGE_UNLOCK |
1451 PAGE_SET_PRIVATE2);
e9894fd3 1452 if (!nolock && nocow)
9ea24bbe 1453 btrfs_end_write_no_snapshoting(root);
80ff3856
YZ		 1454 cur_offset = extent_end;
1455 if (cur_offset > end)
1456 break;
be20aa9d 1457 }
b3b4aa74 1458 btrfs_release_path(path);
80ff3856 1459
17ca04af 1460 if (cur_offset <= end && cow_start == (u64)-1) {
80ff3856 1461 cow_start = cur_offset;
17ca04af
JB		 1462 cur_offset = end;
1463 }
1464
80ff3856 1465 if (cow_start != (u64)-1) {
00361589
JB		 1466 ret = cow_file_range(inode, locked_page, cow_start, end,
1467 page_started, nr_written, 1);
d788a349 1468 if (ret)
79787eaa 1469 goto error;
80ff3856
YZ		 1470 }
1471
79787eaa 1472error:
a698d075 1473 err = btrfs_end_transaction(trans, root);
79787eaa
JM		 1474 if (!ret)
1475 ret = err;
1476
17ca04af 1477 if (ret && cur_offset < end)
c2790a2e
JB		 1478 extent_clear_unlock_delalloc(inode, cur_offset, end,
1479 locked_page, EXTENT_LOCKED |
151a41bc
JB		 1480 EXTENT_DELALLOC | EXTENT_DEFRAG |
1481 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1482 PAGE_CLEAR_DIRTY |
c2790a2e
JB		 1483 PAGE_SET_WRITEBACK |
1484 PAGE_END_WRITEBACK);
7ea394f1 1485 btrfs_free_path(path);
79787eaa 1486 return ret;
be20aa9d
CM		 1487}
1488
47059d93
WS		 1489static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1490{
1491
1492 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1493 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1494 return 0;
1495
1496 /*
1497 * @defrag_bytes is a hint value, no spinlock held here,
1498 * if is not zero, it means the file is defragging.
1499 * Force cow if given extent needs to be defragged.
1500 */
1501 if (BTRFS_I(inode)->defrag_bytes &&
1502 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1503 EXTENT_DEFRAG, 0, NULL))
1504 return 1;
1505
1506 return 0;
1507}
1508
d352ac68
CM		 1509/*
1510 * extent_io.c call back to do delayed allocation processing
1511 */
c8b97818 1512static int run_delalloc_range(struct inode *inode, struct page *locked_page,
771ed689
CM		 1513 u64 start, u64 end, int *page_started,
1514 unsigned long *nr_written)
be20aa9d 1515{
be20aa9d 1516 int ret;
47059d93 1517 int force_cow = need_force_cow(inode, start, end);
a2135011 1518
47059d93 1519 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
c8b97818 1520 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1521 page_started, 1, nr_written);
47059d93 1522 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
d899e052 1523 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1524 page_started, 0, nr_written);
7816030e 1525 } else if (!inode_need_compress(inode)) {
7f366cfe
CM		 1526 ret = cow_file_range(inode, locked_page, start, end,
1527 page_started, nr_written, 1);
7ddf5a42
JB		 1528 } else {
1529 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1530 &BTRFS_I(inode)->runtime_flags);
771ed689 1531 ret = cow_file_range_async(inode, locked_page, start, end,
d397712b 1532 page_started, nr_written);
7ddf5a42 1533 }
b888db2b
CM		 1534 return ret;
1535}
1536
1bf85046
JM		 1537static void btrfs_split_extent_hook(struct inode *inode,
1538 struct extent_state *orig, u64 split)
9ed74f2d 1539{
dcab6a3b
JB		 1540 u64 size;
1541
0ca1f7ce 1542 /* not delalloc, ignore it */
9ed74f2d 1543 if (!(orig->state & EXTENT_DELALLOC))
1bf85046 1544 return;
9ed74f2d 1545
dcab6a3b
JB		 1546 size = orig->end - orig->start + 1;
1547 if (size > BTRFS_MAX_EXTENT_SIZE) {
1548 u64 num_extents;
1549 u64 new_size;
1550
1551 /*
ba117213
JB		 1552 * See the explanation in btrfs_merge_extent_hook, the same
1553 * applies here, just in reverse.
dcab6a3b
JB		 1554 */
1555 new_size = orig->end - split + 1;
ba117213 1556 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
dcab6a3b 1557 BTRFS_MAX_EXTENT_SIZE);
ba117213
JB		 1558 new_size = split - orig->start;
1559 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1560 BTRFS_MAX_EXTENT_SIZE);
1561 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1562 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
dcab6a3b
JB		 1563 return;
1564 }
1565
9e0baf60
JB		 1566 spin_lock(&BTRFS_I(inode)->lock);
1567 BTRFS_I(inode)->outstanding_extents++;
1568 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1569}
1570
1571/*
1572 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1573 * extents so we can keep track of new extents that are just merged onto old
1574 * extents, such as when we are doing sequential writes, so we can properly
1575 * account for the metadata space we'll need.
1576 */
1bf85046
JM		 1577static void btrfs_merge_extent_hook(struct inode *inode,
1578 struct extent_state *new,
1579 struct extent_state *other)
9ed74f2d 1580{
dcab6a3b
JB		 1581 u64 new_size, old_size;
1582 u64 num_extents;
1583
9ed74f2d
JB		 1584 /* not delalloc, ignore it */
1585 if (!(other->state & EXTENT_DELALLOC))
1bf85046 1586 return;
9ed74f2d 1587
8461a3de
JB		 1588 if (new->start > other->start)
1589 new_size = new->end - other->start + 1;
1590 else
1591 new_size = other->end - new->start + 1;
dcab6a3b
JB		 1592
1593 /* we're not bigger than the max, unreserve the space and go */
1594 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1595 spin_lock(&BTRFS_I(inode)->lock);
1596 BTRFS_I(inode)->outstanding_extents--;
1597 spin_unlock(&BTRFS_I(inode)->lock);
1598 return;
1599 }
1600
1601 /*
ba117213
JB		 1602 * We have to add up either side to figure out how many extents were
1603 * accounted for before we merged into one big extent. If the number of
1604 * extents we accounted for is <= the amount we need for the new range
1605 * then we can return, otherwise drop. Think of it like this
1606 *
1607 * [ 4k][MAX_SIZE]
1608 *
1609 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1610 * need 2 outstanding extents, on one side we have 1 and the other side
1611 * we have 1 so they are == and we can return. But in this case
1612 *
1613 * [MAX_SIZE+4k][MAX_SIZE+4k]
1614 *
1615 * Each range on their own accounts for 2 extents, but merged together
1616 * they are only 3 extents worth of accounting, so we need to drop in
1617 * this case.
dcab6a3b 1618 */
ba117213 1619 old_size = other->end - other->start + 1;
dcab6a3b
JB		 1620 num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1621 BTRFS_MAX_EXTENT_SIZE);
ba117213
JB		 1622 old_size = new->end - new->start + 1;
1623 num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1624 BTRFS_MAX_EXTENT_SIZE);
1625
dcab6a3b 1626 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
ba117213 1627 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
dcab6a3b
JB		 1628 return;
1629
9e0baf60
JB		 1630 spin_lock(&BTRFS_I(inode)->lock);
1631 BTRFS_I(inode)->outstanding_extents--;
1632 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1633}
1634
eb73c1b7
MX		 1635static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1636 struct inode *inode)
1637{
1638 spin_lock(&root->delalloc_lock);
1639 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1640 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1641 &root->delalloc_inodes);
1642 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1643 &BTRFS_I(inode)->runtime_flags);
1644 root->nr_delalloc_inodes++;
1645 if (root->nr_delalloc_inodes == 1) {
1646 spin_lock(&root->fs_info->delalloc_root_lock);
1647 BUG_ON(!list_empty(&root->delalloc_root));
1648 list_add_tail(&root->delalloc_root,
1649 &root->fs_info->delalloc_roots);
1650 spin_unlock(&root->fs_info->delalloc_root_lock);
1651 }
1652 }
1653 spin_unlock(&root->delalloc_lock);
1654}
1655
1656static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1657 struct inode *inode)
1658{
1659 spin_lock(&root->delalloc_lock);
1660 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1661 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1662 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1663 &BTRFS_I(inode)->runtime_flags);
1664 root->nr_delalloc_inodes--;
1665 if (!root->nr_delalloc_inodes) {
1666 spin_lock(&root->fs_info->delalloc_root_lock);
1667 BUG_ON(list_empty(&root->delalloc_root));
1668 list_del_init(&root->delalloc_root);
1669 spin_unlock(&root->fs_info->delalloc_root_lock);
1670 }
1671 }
1672 spin_unlock(&root->delalloc_lock);
1673}
1674
d352ac68
CM		 1675/*
1676 * extent_io.c set_bit_hook, used to track delayed allocation
1677 * bytes in this file, and to maintain the list of inodes that
1678 * have pending delalloc work to be done.
1679 */
1bf85046 1680static void btrfs_set_bit_hook(struct inode *inode,
9ee49a04 1681 struct extent_state *state, unsigned *bits)
291d673e 1682{
9ed74f2d 1683
47059d93
WS		 1684 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1685 WARN_ON(1);
75eff68e
CM		 1686 /*
1687 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1688 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1689 * bit, which is only set or cleared with irqs on
1690 */
0ca1f7ce 1691 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1692 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1693 u64 len = state->end + 1 - state->start;
83eea1f1 1694 bool do_list = !btrfs_is_free_space_inode(inode);
9ed74f2d 1695
9e0baf60 1696 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1697 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1698 } else {
1699 spin_lock(&BTRFS_I(inode)->lock);
1700 BTRFS_I(inode)->outstanding_extents++;
1701 spin_unlock(&BTRFS_I(inode)->lock);
1702 }
287a0ab9 1703
6a3891c5
JB		 1704 /* For sanity tests */
1705 if (btrfs_test_is_dummy_root(root))
1706 return;
1707
963d678b
MX		 1708 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1709 root->fs_info->delalloc_batch);
df0af1a5 1710 spin_lock(&BTRFS_I(inode)->lock);
0ca1f7ce 1711 BTRFS_I(inode)->delalloc_bytes += len;
47059d93
WS		 1712 if (*bits & EXTENT_DEFRAG)
1713 BTRFS_I(inode)->defrag_bytes += len;
df0af1a5 1714 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
eb73c1b7
MX		 1715 &BTRFS_I(inode)->runtime_flags))
1716 btrfs_add_delalloc_inodes(root, inode);
df0af1a5 1717 spin_unlock(&BTRFS_I(inode)->lock);
291d673e 1718 }
291d673e
CM		 1719}
1720
d352ac68
CM		 1721/*
1722 * extent_io.c clear_bit_hook, see set_bit_hook for why
1723 */
1bf85046 1724static void btrfs_clear_bit_hook(struct inode *inode,
41074888 1725 struct extent_state *state,
9ee49a04 1726 unsigned *bits)
291d673e 1727{
47059d93 1728 u64 len = state->end + 1 - state->start;
dcab6a3b
JB		 1729 u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1730 BTRFS_MAX_EXTENT_SIZE);
47059d93
WS		 1731
1732 spin_lock(&BTRFS_I(inode)->lock);
1733 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1734 BTRFS_I(inode)->defrag_bytes -= len;
1735 spin_unlock(&BTRFS_I(inode)->lock);
1736
75eff68e
CM		 1737 /*
1738 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1739 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1740 * bit, which is only set or cleared with irqs on
1741 */
0ca1f7ce 1742 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1743 struct btrfs_root *root = BTRFS_I(inode)->root;
83eea1f1 1744 bool do_list = !btrfs_is_free_space_inode(inode);
bcbfce8a 1745
9e0baf60 1746 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1747 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1748 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1749 spin_lock(&BTRFS_I(inode)->lock);
dcab6a3b 1750 BTRFS_I(inode)->outstanding_extents -= num_extents;
9e0baf60
JB		 1751 spin_unlock(&BTRFS_I(inode)->lock);
1752 }
0ca1f7ce 1753
b6d08f06
JB		 1754 /*
1755 * We don't reserve metadata space for space cache inodes so we
1756 * don't need to call dellalloc_release_metadata if there is an
1757 * error.
1758 */
1759 if (*bits & EXTENT_DO_ACCOUNTING &&
1760 root != root->fs_info->tree_root)
0ca1f7ce
YZ		 1761 btrfs_delalloc_release_metadata(inode, len);
1762
6a3891c5
JB		 1763 /* For sanity tests. */
1764 if (btrfs_test_is_dummy_root(root))
1765 return;
1766
0cb59c99 1767 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
7ee9e440 1768 && do_list && !(state->state & EXTENT_NORESERVE))
0ca1f7ce 1769 btrfs_free_reserved_data_space(inode, len);
9ed74f2d 1770
963d678b
MX		 1771 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1772 root->fs_info->delalloc_batch);
df0af1a5 1773 spin_lock(&BTRFS_I(inode)->lock);
0ca1f7ce 1774 BTRFS_I(inode)->delalloc_bytes -= len;
0cb59c99 1775 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
df0af1a5 1776 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
eb73c1b7
MX		 1777 &BTRFS_I(inode)->runtime_flags))
1778 btrfs_del_delalloc_inode(root, inode);
df0af1a5 1779 spin_unlock(&BTRFS_I(inode)->lock);
291d673e 1780 }
291d673e
CM		 1781}
1782
d352ac68
CM		 1783/*
1784 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1785 * we don't create bios that span stripes or chunks
1786 */
64a16701 1787int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
c8b97818
CM		 1788 size_t size, struct bio *bio,
1789 unsigned long bio_flags)
239b14b3
CM		 1790{
1791 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
4f024f37 1792 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
239b14b3
CM		 1793 u64 length = 0;
1794 u64 map_length;
239b14b3
CM		 1795 int ret;
1796
771ed689
CM		 1797 if (bio_flags & EXTENT_BIO_COMPRESSED)
1798 return 0;
1799
4f024f37 1800 length = bio->bi_iter.bi_size;
239b14b3 1801 map_length = length;
64a16701 1802 ret = btrfs_map_block(root->fs_info, rw, logical,
f188591e 1803 &map_length, NULL, 0);
3ec706c8 1804 /* Will always return 0 with map_multi == NULL */
3444a972 1805 BUG_ON(ret < 0);
d397712b 1806 if (map_length < length + size)
239b14b3 1807 return 1;
3444a972 1808 return 0;
239b14b3
CM		 1809}
1810
d352ac68
CM		 1811/*
1812 * in order to insert checksums into the metadata in large chunks,
1813 * we wait until bio submission time. All the pages in the bio are
1814 * checksummed and sums are attached onto the ordered extent record.
1815 *
1816 * At IO completion time the cums attached on the ordered extent record
1817 * are inserted into the btree
1818 */
d397712b
CM		 1819static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1820 struct bio *bio, int mirror_num,
eaf25d93
CM		 1821 unsigned long bio_flags,
1822 u64 bio_offset)
065631f6 1823{
065631f6 1824 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 1825 int ret = 0;
e015640f 1826
d20f7043 1827 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
79787eaa 1828 BUG_ON(ret); /* -ENOMEM */
4a69a410
CM		 1829 return 0;
1830}
e015640f 1831
4a69a410
CM		 1832/*
1833 * in order to insert checksums into the metadata in large chunks,
1834 * we wait until bio submission time. All the pages in the bio are
1835 * checksummed and sums are attached onto the ordered extent record.
1836 *
1837 * At IO completion time the cums attached on the ordered extent record
1838 * are inserted into the btree
1839 */
b2950863 1840static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1841 int mirror_num, unsigned long bio_flags,
1842 u64 bio_offset)
4a69a410
CM		 1843{
1844 struct btrfs_root *root = BTRFS_I(inode)->root;
61891923
SB		 1845 int ret;
1846
1847 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
4246a0b6
CH		 1848 if (ret) {
1849 bio->bi_error = ret;
1850 bio_endio(bio);
1851 }
61891923 1852 return ret;
44b8bd7e
CM		 1853}
1854
d352ac68 1855/*
cad321ad
CM		 1856 * extent_io.c submission hook. This does the right thing for csum calculation
1857 * on write, or reading the csums from the tree before a read
d352ac68 1858 */
b2950863 1859static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1860 int mirror_num, unsigned long bio_flags,
1861 u64 bio_offset)
44b8bd7e
CM		 1862{
1863 struct btrfs_root *root = BTRFS_I(inode)->root;
1864 int ret = 0;
19b9bdb0 1865 int skip_sum;
0417341e 1866 int metadata = 0;
b812ce28 1867 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
44b8bd7e 1868
6cbff00f 1869 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
cad321ad 1870
83eea1f1 1871 if (btrfs_is_free_space_inode(inode))
0417341e
JM		 1872 metadata = 2;
1873
7b6d91da 1874 if (!(rw & REQ_WRITE)) {
5fd02043
JB		 1875 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1876 if (ret)
61891923 1877 goto out;
5fd02043 1878
d20f7043 1879 if (bio_flags & EXTENT_BIO_COMPRESSED) {
61891923
SB		 1880 ret = btrfs_submit_compressed_read(inode, bio,
1881 mirror_num,
1882 bio_flags);
1883 goto out;
c2db1073
TI		 1884 } else if (!skip_sum) {
1885 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1886 if (ret)
61891923 1887 goto out;
c2db1073 1888 }
4d1b5fb4 1889 goto mapit;
b812ce28 1890 } else if (async && !skip_sum) {
17d217fe
YZ		 1891 /* csum items have already been cloned */
1892 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1893 goto mapit;
19b9bdb0 1894 /* we're doing a write, do the async checksumming */
61891923 1895 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 1896 inode, rw, bio, mirror_num,
eaf25d93
CM		 1897 bio_flags, bio_offset,
1898 __btrfs_submit_bio_start,
4a69a410 1899 __btrfs_submit_bio_done);
61891923 1900 goto out;
b812ce28
JB		 1901 } else if (!skip_sum) {
1902 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1903 if (ret)
1904 goto out;
19b9bdb0
CM		 1905 }
1906
0b86a832 1907mapit:
61891923
SB		 1908 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1909
1910out:
4246a0b6
CH		 1911 if (ret < 0) {
1912 bio->bi_error = ret;
1913 bio_endio(bio);
1914 }
61891923 1915 return ret;
065631f6 1916}
6885f308 1917
d352ac68
CM		 1918/*
1919 * given a list of ordered sums record them in the inode. This happens
1920 * at IO completion time based on sums calculated at bio submission time.
1921 */
ba1da2f4 1922static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM		 1923 struct inode *inode, u64 file_offset,
1924 struct list_head *list)
1925{
e6dcd2dc
CM		 1926 struct btrfs_ordered_sum *sum;
1927
c6e30871 1928 list_for_each_entry(sum, list, list) {
39847c4d 1929 trans->adding_csums = 1;
d20f7043
CM		 1930 btrfs_csum_file_blocks(trans,
1931 BTRFS_I(inode)->root->fs_info->csum_root, sum);
39847c4d 1932 trans->adding_csums = 0;
e6dcd2dc
CM		 1933 }
1934 return 0;
1935}
1936
2ac55d41
JB		 1937int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1938 struct extent_state **cached_state)
ea8c2819 1939{
6c1500f2 1940 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
ea8c2819 1941 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2ac55d41 1942 cached_state, GFP_NOFS);
ea8c2819
CM		 1943}
1944
d352ac68 1945/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM		 1946struct btrfs_writepage_fixup {
1947 struct page *page;
1948 struct btrfs_work work;
1949};
1950
b2950863 1951static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
247e743c
CM		 1952{
1953 struct btrfs_writepage_fixup *fixup;
1954 struct btrfs_ordered_extent *ordered;
2ac55d41 1955 struct extent_state *cached_state = NULL;
247e743c
CM		 1956 struct page *page;
1957 struct inode *inode;
1958 u64 page_start;
1959 u64 page_end;
87826df0 1960 int ret;
247e743c
CM		 1961
1962 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1963 page = fixup->page;
4a096752 1964again:
247e743c
CM		 1965 lock_page(page);
1966 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1967 ClearPageChecked(page);
1968 goto out_page;
1969 }
1970
1971 inode = page->mapping->host;
1972 page_start = page_offset(page);
1973 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1974
2ac55d41 1975 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
d0082371 1976 &cached_state);
4a096752
CM		 1977
1978 /* already ordered? We're done */
8b62b72b 1979 if (PagePrivate2(page))
247e743c 1980 goto out;
4a096752
CM		 1981
1982 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1983 if (ordered) {
2ac55d41
JB		 1984 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1985 page_end, &cached_state, GFP_NOFS);
4a096752
CM		 1986 unlock_page(page);
1987 btrfs_start_ordered_extent(inode, ordered, 1);
87826df0 1988 btrfs_put_ordered_extent(ordered);
4a096752
CM		 1989 goto again;
1990 }
247e743c 1991
87826df0
JM		 1992 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1993 if (ret) {
1994 mapping_set_error(page->mapping, ret);
1995 end_extent_writepage(page, ret, page_start, page_end);
1996 ClearPageChecked(page);
1997 goto out;
1998 }
1999
2ac55d41 2000 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
247e743c 2001 ClearPageChecked(page);
87826df0 2002 set_page_dirty(page);
247e743c 2003out:
2ac55d41
JB		 2004 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2005 &cached_state, GFP_NOFS);
247e743c
CM		 2006out_page:
2007 unlock_page(page);
2008 page_cache_release(page);
b897abec 2009 kfree(fixup);
247e743c
CM		 2010}
2011
2012/*
2013 * There are a few paths in the higher layers of the kernel that directly
2014 * set the page dirty bit without asking the filesystem if it is a
2015 * good idea. This causes problems because we want to make sure COW
2016 * properly happens and the data=ordered rules are followed.
2017 *
c8b97818 2018 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 2019 * hasn't been properly setup for IO. We kick off an async process
2020 * to fix it up. The async helper will wait for ordered extents, set
2021 * the delalloc bit and make it safe to write the page.
2022 */
b2950863 2023static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
247e743c
CM		 2024{
2025 struct inode *inode = page->mapping->host;
2026 struct btrfs_writepage_fixup *fixup;
2027 struct btrfs_root *root = BTRFS_I(inode)->root;
247e743c 2028
8b62b72b
CM		 2029 /* this page is properly in the ordered list */
2030 if (TestClearPagePrivate2(page))
247e743c
CM		 2031 return 0;
2032
2033 if (PageChecked(page))
2034 return -EAGAIN;
2035
2036 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2037 if (!fixup)
2038 return -EAGAIN;
f421950f 2039
247e743c
CM		 2040 SetPageChecked(page);
2041 page_cache_get(page);
9e0af237
LB		 2042 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2043 btrfs_writepage_fixup_worker, NULL, NULL);
247e743c 2044 fixup->page = page;
dc6e3209 2045 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
87826df0 2046 return -EBUSY;
247e743c
CM		 2047}
2048
d899e052
YZ		 2049static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2050 struct inode *inode, u64 file_pos,
2051 u64 disk_bytenr, u64 disk_num_bytes,
2052 u64 num_bytes, u64 ram_bytes,
2053 u8 compression, u8 encryption,
2054 u16 other_encoding, int extent_type)
2055{
2056 struct btrfs_root *root = BTRFS_I(inode)->root;
2057 struct btrfs_file_extent_item *fi;
2058 struct btrfs_path *path;
2059 struct extent_buffer *leaf;
2060 struct btrfs_key ins;
1acae57b 2061 int extent_inserted = 0;
d899e052
YZ		 2062 int ret;
2063
2064 path = btrfs_alloc_path();
d8926bb3
MF		 2065 if (!path)
2066 return -ENOMEM;
d899e052 2067
a1ed835e
CM		 2068 /*
2069 * we may be replacing one extent in the tree with another.
2070 * The new extent is pinned in the extent map, and we don't want
2071 * to drop it from the cache until it is completely in the btree.
2072 *
2073 * So, tell btrfs_drop_extents to leave this extent in the cache.
2074 * the caller is expected to unpin it and allow it to be merged
2075 * with the others.
2076 */
1acae57b
FDBM		 2077 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2078 file_pos + num_bytes, NULL, 0,
2079 1, sizeof(*fi), &extent_inserted);
79787eaa
JM		 2080 if (ret)
2081 goto out;
d899e052 2082
1acae57b
FDBM		 2083 if (!extent_inserted) {
2084 ins.objectid = btrfs_ino(inode);
2085 ins.offset = file_pos;
2086 ins.type = BTRFS_EXTENT_DATA_KEY;
2087
2088 path->leave_spinning = 1;
2089 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2090 sizeof(*fi));
2091 if (ret)
2092 goto out;
2093 }
d899e052
YZ		 2094 leaf = path->nodes[0];
2095 fi = btrfs_item_ptr(leaf, path->slots[0],
2096 struct btrfs_file_extent_item);
2097 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2098 btrfs_set_file_extent_type(leaf, fi, extent_type);
2099 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2100 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2101 btrfs_set_file_extent_offset(leaf, fi, 0);
2102 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2103 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2104 btrfs_set_file_extent_compression(leaf, fi, compression);
2105 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2106 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
b9473439 2107
d899e052 2108 btrfs_mark_buffer_dirty(leaf);
ce195332 2109 btrfs_release_path(path);
d899e052
YZ		 2110
2111 inode_add_bytes(inode, num_bytes);
d899e052
YZ		 2112
2113 ins.objectid = disk_bytenr;
2114 ins.offset = disk_num_bytes;
2115 ins.type = BTRFS_EXTENT_ITEM_KEY;
5d4f98a2
YZ		 2116 ret = btrfs_alloc_reserved_file_extent(trans, root,
2117 root->root_key.objectid,
33345d01 2118 btrfs_ino(inode), file_pos, &ins);
79787eaa 2119out:
d899e052 2120 btrfs_free_path(path);
b9473439 2121
79787eaa 2122 return ret;
d899e052
YZ		 2123}
2124
38c227d8
LB		 2125/* snapshot-aware defrag */
2126struct sa_defrag_extent_backref {
2127 struct rb_node node;
2128 struct old_sa_defrag_extent *old;
2129 u64 root_id;
2130 u64 inum;
2131 u64 file_pos;
2132 u64 extent_offset;
2133 u64 num_bytes;
2134 u64 generation;
2135};
2136
2137struct old_sa_defrag_extent {
2138 struct list_head list;
2139 struct new_sa_defrag_extent *new;
2140
2141 u64 extent_offset;
2142 u64 bytenr;
2143 u64 offset;
2144 u64 len;
2145 int count;
2146};
2147
2148struct new_sa_defrag_extent {
2149 struct rb_root root;
2150 struct list_head head;
2151 struct btrfs_path *path;
2152 struct inode *inode;
2153 u64 file_pos;
2154 u64 len;
2155 u64 bytenr;
2156 u64 disk_len;
2157 u8 compress_type;
2158};
2159
2160static int backref_comp(struct sa_defrag_extent_backref *b1,
2161 struct sa_defrag_extent_backref *b2)
2162{
2163 if (b1->root_id < b2->root_id)
2164 return -1;
2165 else if (b1->root_id > b2->root_id)
2166 return 1;
2167
2168 if (b1->inum < b2->inum)
2169 return -1;
2170 else if (b1->inum > b2->inum)
2171 return 1;
2172
2173 if (b1->file_pos < b2->file_pos)
2174 return -1;
2175 else if (b1->file_pos > b2->file_pos)
2176 return 1;
2177
2178 /*
2179 * [------------------------------] ===> (a range of space)
2180 * |<--->| |<---->| =============> (fs/file tree A)
2181 * |<---------------------------->| ===> (fs/file tree B)
2182 *
2183 * A range of space can refer to two file extents in one tree while
2184 * refer to only one file extent in another tree.
2185 *
2186 * So we may process a disk offset more than one time(two extents in A)
2187 * and locate at the same extent(one extent in B), then insert two same
2188 * backrefs(both refer to the extent in B).
2189 */
2190 return 0;
2191}
2192
2193static void backref_insert(struct rb_root *root,
2194 struct sa_defrag_extent_backref *backref)
2195{
2196 struct rb_node **p = &root->rb_node;
2197 struct rb_node *parent = NULL;
2198 struct sa_defrag_extent_backref *entry;
2199 int ret;
2200
2201 while (*p) {
2202 parent = *p;
2203 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2204
2205 ret = backref_comp(backref, entry);
2206 if (ret < 0)
2207 p = &(*p)->rb_left;
2208 else
2209 p = &(*p)->rb_right;
2210 }
2211
2212 rb_link_node(&backref->node, parent, p);
2213 rb_insert_color(&backref->node, root);
2214}
2215
2216/*
2217 * Note the backref might has changed, and in this case we just return 0.
2218 */
2219static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2220 void *ctx)
2221{
2222 struct btrfs_file_extent_item *extent;
2223 struct btrfs_fs_info *fs_info;
2224 struct old_sa_defrag_extent *old = ctx;
2225 struct new_sa_defrag_extent *new = old->new;
2226 struct btrfs_path *path = new->path;
2227 struct btrfs_key key;
2228 struct btrfs_root *root;
2229 struct sa_defrag_extent_backref *backref;
2230 struct extent_buffer *leaf;
2231 struct inode *inode = new->inode;
2232 int slot;
2233 int ret;
2234 u64 extent_offset;
2235 u64 num_bytes;
2236
2237 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2238 inum == btrfs_ino(inode))
2239 return 0;
2240
2241 key.objectid = root_id;
2242 key.type = BTRFS_ROOT_ITEM_KEY;
2243 key.offset = (u64)-1;
2244
2245 fs_info = BTRFS_I(inode)->root->fs_info;
2246 root = btrfs_read_fs_root_no_name(fs_info, &key);
2247 if (IS_ERR(root)) {
2248 if (PTR_ERR(root) == -ENOENT)
2249 return 0;
2250 WARN_ON(1);
2251 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2252 inum, offset, root_id);
2253 return PTR_ERR(root);
2254 }
2255
2256 key.objectid = inum;
2257 key.type = BTRFS_EXTENT_DATA_KEY;
2258 if (offset > (u64)-1 << 32)
2259 key.offset = 0;
2260 else
2261 key.offset = offset;
2262
2263 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
fae7f21c 2264 if (WARN_ON(ret < 0))
38c227d8 2265 return ret;
50f1319c 2266 ret = 0;
38c227d8
LB		 2267
2268 while (1) {
2269 cond_resched();
2270
2271 leaf = path->nodes[0];
2272 slot = path->slots[0];
2273
2274 if (slot >= btrfs_header_nritems(leaf)) {
2275 ret = btrfs_next_leaf(root, path);
2276 if (ret < 0) {
2277 goto out;
2278 } else if (ret > 0) {
2279 ret = 0;
2280 goto out;
2281 }
2282 continue;
2283 }
2284
2285 path->slots[0]++;
2286
2287 btrfs_item_key_to_cpu(leaf, &key, slot);
2288
2289 if (key.objectid > inum)
2290 goto out;
2291
2292 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2293 continue;
2294
2295 extent = btrfs_item_ptr(leaf, slot,
2296 struct btrfs_file_extent_item);
2297
2298 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2299 continue;
2300
e68afa49
LB		 2301 /*
2302 * 'offset' refers to the exact key.offset,
2303 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2304 * (key.offset - extent_offset).
2305 */
2306 if (key.offset != offset)
38c227d8
LB		 2307 continue;
2308
e68afa49 2309 extent_offset = btrfs_file_extent_offset(leaf, extent);
38c227d8 2310 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
e68afa49 2311
38c227d8
LB		 2312 if (extent_offset >= old->extent_offset + old->offset +
2313 old->len || extent_offset + num_bytes <=
2314 old->extent_offset + old->offset)
2315 continue;
38c227d8
LB		 2316 break;
2317 }
2318
2319 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2320 if (!backref) {
2321 ret = -ENOENT;
2322 goto out;
2323 }
2324
2325 backref->root_id = root_id;
2326 backref->inum = inum;
e68afa49 2327 backref->file_pos = offset;
38c227d8
LB		 2328 backref->num_bytes = num_bytes;
2329 backref->extent_offset = extent_offset;
2330 backref->generation = btrfs_file_extent_generation(leaf, extent);
2331 backref->old = old;
2332 backref_insert(&new->root, backref);
2333 old->count++;
2334out:
2335 btrfs_release_path(path);
2336 WARN_ON(ret);
2337 return ret;
2338}
2339
2340static noinline bool record_extent_backrefs(struct btrfs_path *path,
2341 struct new_sa_defrag_extent *new)
2342{
2343 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2344 struct old_sa_defrag_extent *old, *tmp;
2345 int ret;
2346
2347 new->path = path;
2348
2349 list_for_each_entry_safe(old, tmp, &new->head, list) {
e68afa49
LB		 2350 ret = iterate_inodes_from_logical(old->bytenr +
2351 old->extent_offset, fs_info,
38c227d8
LB		 2352 path, record_one_backref,
2353 old);
4724b106
JB		 2354 if (ret < 0 && ret != -ENOENT)
2355 return false;
38c227d8
LB		 2356
2357 /* no backref to be processed for this extent */
2358 if (!old->count) {
2359 list_del(&old->list);
2360 kfree(old);
2361 }
2362 }
2363
2364 if (list_empty(&new->head))
2365 return false;
2366
2367 return true;
2368}
2369
2370static int relink_is_mergable(struct extent_buffer *leaf,
2371 struct btrfs_file_extent_item *fi,
116e0024 2372 struct new_sa_defrag_extent *new)
38c227d8 2373{
116e0024 2374 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
38c227d8
LB		 2375 return 0;
2376
2377 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2378 return 0;
2379
116e0024
LB		 2380 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2381 return 0;
2382
2383 if (btrfs_file_extent_encryption(leaf, fi) ||
38c227d8
LB		 2384 btrfs_file_extent_other_encoding(leaf, fi))
2385 return 0;
2386
2387 return 1;
2388}
2389
2390/*
2391 * Note the backref might has changed, and in this case we just return 0.
2392 */
2393static noinline int relink_extent_backref(struct btrfs_path *path,
2394 struct sa_defrag_extent_backref *prev,
2395 struct sa_defrag_extent_backref *backref)
2396{
2397 struct btrfs_file_extent_item *extent;
2398 struct btrfs_file_extent_item *item;
2399 struct btrfs_ordered_extent *ordered;
2400 struct btrfs_trans_handle *trans;
2401 struct btrfs_fs_info *fs_info;
2402 struct btrfs_root *root;
2403 struct btrfs_key key;
2404 struct extent_buffer *leaf;
2405 struct old_sa_defrag_extent *old = backref->old;
2406 struct new_sa_defrag_extent *new = old->new;
2407 struct inode *src_inode = new->inode;
2408 struct inode *inode;
2409 struct extent_state *cached = NULL;
2410 int ret = 0;
2411 u64 start;
2412 u64 len;
2413 u64 lock_start;
2414 u64 lock_end;
2415 bool merge = false;
2416 int index;
2417
2418 if (prev && prev->root_id == backref->root_id &&
2419 prev->inum == backref->inum &&
2420 prev->file_pos + prev->num_bytes == backref->file_pos)
2421 merge = true;
2422
2423 /* step 1: get root */
2424 key.objectid = backref->root_id;
2425 key.type = BTRFS_ROOT_ITEM_KEY;
2426 key.offset = (u64)-1;
2427
2428 fs_info = BTRFS_I(src_inode)->root->fs_info;
2429 index = srcu_read_lock(&fs_info->subvol_srcu);
2430
2431 root = btrfs_read_fs_root_no_name(fs_info, &key);
2432 if (IS_ERR(root)) {
2433 srcu_read_unlock(&fs_info->subvol_srcu, index);
2434 if (PTR_ERR(root) == -ENOENT)
2435 return 0;
2436 return PTR_ERR(root);
2437 }
38c227d8 2438
bcbba5e6
WS		 2439 if (btrfs_root_readonly(root)) {
2440 srcu_read_unlock(&fs_info->subvol_srcu, index);
2441 return 0;
2442 }
2443
38c227d8
LB		 2444 /* step 2: get inode */
2445 key.objectid = backref->inum;
2446 key.type = BTRFS_INODE_ITEM_KEY;
2447 key.offset = 0;
2448
2449 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2450 if (IS_ERR(inode)) {
2451 srcu_read_unlock(&fs_info->subvol_srcu, index);
2452 return 0;
2453 }
2454
2455 srcu_read_unlock(&fs_info->subvol_srcu, index);
2456
2457 /* step 3: relink backref */
2458 lock_start = backref->file_pos;
2459 lock_end = backref->file_pos + backref->num_bytes - 1;
2460 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2461 0, &cached);
2462
2463 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2464 if (ordered) {
2465 btrfs_put_ordered_extent(ordered);
2466 goto out_unlock;
2467 }
2468
2469 trans = btrfs_join_transaction(root);
2470 if (IS_ERR(trans)) {
2471 ret = PTR_ERR(trans);
2472 goto out_unlock;
2473 }
2474
2475 key.objectid = backref->inum;
2476 key.type = BTRFS_EXTENT_DATA_KEY;
2477 key.offset = backref->file_pos;
2478
2479 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2480 if (ret < 0) {
2481 goto out_free_path;
2482 } else if (ret > 0) {
2483 ret = 0;
2484 goto out_free_path;
2485 }
2486
2487 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2488 struct btrfs_file_extent_item);
2489
2490 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2491 backref->generation)
2492 goto out_free_path;
2493
2494 btrfs_release_path(path);
2495
2496 start = backref->file_pos;
2497 if (backref->extent_offset < old->extent_offset + old->offset)
2498 start += old->extent_offset + old->offset -
2499 backref->extent_offset;
2500
2501 len = min(backref->extent_offset + backref->num_bytes,
2502 old->extent_offset + old->offset + old->len);
2503 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2504
2505 ret = btrfs_drop_extents(trans, root, inode, start,
2506 start + len, 1);
2507 if (ret)
2508 goto out_free_path;
2509again:
2510 key.objectid = btrfs_ino(inode);
2511 key.type = BTRFS_EXTENT_DATA_KEY;
2512 key.offset = start;
2513
a09a0a70 2514 path->leave_spinning = 1;
38c227d8
LB		 2515 if (merge) {
2516 struct btrfs_file_extent_item *fi;
2517 u64 extent_len;
2518 struct btrfs_key found_key;
2519
3c9665df 2520 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
38c227d8
LB		 2521 if (ret < 0)
2522 goto out_free_path;
2523
2524 path->slots[0]--;
2525 leaf = path->nodes[0];
2526 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2527
2528 fi = btrfs_item_ptr(leaf, path->slots[0],
2529 struct btrfs_file_extent_item);
2530 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2531
116e0024
LB		 2532 if (extent_len + found_key.offset == start &&
2533 relink_is_mergable(leaf, fi, new)) {
38c227d8
LB		 2534 btrfs_set_file_extent_num_bytes(leaf, fi,
2535 extent_len + len);
2536 btrfs_mark_buffer_dirty(leaf);
2537 inode_add_bytes(inode, len);
2538
2539 ret = 1;
2540 goto out_free_path;
2541 } else {
2542 merge = false;
2543 btrfs_release_path(path);
2544 goto again;
2545 }
2546 }
2547
2548 ret = btrfs_insert_empty_item(trans, root, path, &key,
2549 sizeof(*extent));
2550 if (ret) {
2551 btrfs_abort_transaction(trans, root, ret);
2552 goto out_free_path;
2553 }
2554
2555 leaf = path->nodes[0];
2556 item = btrfs_item_ptr(leaf, path->slots[0],
2557 struct btrfs_file_extent_item);
2558 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2559 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2560 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2561 btrfs_set_file_extent_num_bytes(leaf, item, len);
2562 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2563 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2564 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2565 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2566 btrfs_set_file_extent_encryption(leaf, item, 0);
2567 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2568
2569 btrfs_mark_buffer_dirty(leaf);
2570 inode_add_bytes(inode, len);
a09a0a70 2571 btrfs_release_path(path);
38c227d8
LB		 2572
2573 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2574 new->disk_len, 0,
2575 backref->root_id, backref->inum,
2576 new->file_pos, 0); /* start - extent_offset */
2577 if (ret) {
2578 btrfs_abort_transaction(trans, root, ret);
2579 goto out_free_path;
2580 }
2581
2582 ret = 1;
2583out_free_path:
2584 btrfs_release_path(path);
a09a0a70 2585 path->leave_spinning = 0;
38c227d8
LB		 2586 btrfs_end_transaction(trans, root);
2587out_unlock:
2588 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2589 &cached, GFP_NOFS);
2590 iput(inode);
2591 return ret;
2592}
2593
6f519564
LB		 2594static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2595{
2596 struct old_sa_defrag_extent *old, *tmp;
2597
2598 if (!new)
2599 return;
2600
2601 list_for_each_entry_safe(old, tmp, &new->head, list) {
2602 list_del(&old->list);
2603 kfree(old);
2604 }
2605 kfree(new);
2606}
2607
38c227d8
LB		 2608static void relink_file_extents(struct new_sa_defrag_extent *new)
2609{
2610 struct btrfs_path *path;
38c227d8
LB		 2611 struct sa_defrag_extent_backref *backref;
2612 struct sa_defrag_extent_backref *prev = NULL;
2613 struct inode *inode;
2614 struct btrfs_root *root;
2615 struct rb_node *node;
2616 int ret;
2617
2618 inode = new->inode;
2619 root = BTRFS_I(inode)->root;
2620
2621 path = btrfs_alloc_path();
2622 if (!path)
2623 return;
2624
2625 if (!record_extent_backrefs(path, new)) {
2626 btrfs_free_path(path);
2627 goto out;
2628 }
2629 btrfs_release_path(path);
2630
2631 while (1) {
2632 node = rb_first(&new->root);
2633 if (!node)
2634 break;
2635 rb_erase(node, &new->root);
2636
2637 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2638
2639 ret = relink_extent_backref(path, prev, backref);
2640 WARN_ON(ret < 0);
2641
2642 kfree(prev);
2643
2644 if (ret == 1)
2645 prev = backref;
2646 else
2647 prev = NULL;
2648 cond_resched();
2649 }
2650 kfree(prev);
2651
2652 btrfs_free_path(path);
38c227d8 2653out:
6f519564
LB		 2654 free_sa_defrag_extent(new);
2655
38c227d8
LB		 2656 atomic_dec(&root->fs_info->defrag_running);
2657 wake_up(&root->fs_info->transaction_wait);
38c227d8
LB		 2658}
2659
2660static struct new_sa_defrag_extent *
2661record_old_file_extents(struct inode *inode,
2662 struct btrfs_ordered_extent *ordered)
2663{
2664 struct btrfs_root *root = BTRFS_I(inode)->root;
2665 struct btrfs_path *path;
2666 struct btrfs_key key;
6f519564 2667 struct old_sa_defrag_extent *old;
38c227d8
LB		 2668 struct new_sa_defrag_extent *new;
2669 int ret;
2670
2671 new = kmalloc(sizeof(*new), GFP_NOFS);
2672 if (!new)
2673 return NULL;
2674
2675 new->inode = inode;
2676 new->file_pos = ordered->file_offset;
2677 new->len = ordered->len;
2678 new->bytenr = ordered->start;
2679 new->disk_len = ordered->disk_len;
2680 new->compress_type = ordered->compress_type;
2681 new->root = RB_ROOT;
2682 INIT_LIST_HEAD(&new->head);
2683
2684 path = btrfs_alloc_path();
2685 if (!path)
2686 goto out_kfree;
2687
2688 key.objectid = btrfs_ino(inode);
2689 key.type = BTRFS_EXTENT_DATA_KEY;
2690 key.offset = new->file_pos;
2691
2692 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2693 if (ret < 0)
2694 goto out_free_path;
2695 if (ret > 0 && path->slots[0] > 0)
2696 path->slots[0]--;
2697
2698 /* find out all the old extents for the file range */
2699 while (1) {
2700 struct btrfs_file_extent_item *extent;
2701 struct extent_buffer *l;
2702 int slot;
2703 u64 num_bytes;
2704 u64 offset;
2705 u64 end;
2706 u64 disk_bytenr;
2707 u64 extent_offset;
2708
2709 l = path->nodes[0];
2710 slot = path->slots[0];
2711
2712 if (slot >= btrfs_header_nritems(l)) {
2713 ret = btrfs_next_leaf(root, path);
2714 if (ret < 0)
6f519564 2715 goto out_free_path;
38c227d8
LB		 2716 else if (ret > 0)
2717 break;
2718 continue;
2719 }
2720
2721 btrfs_item_key_to_cpu(l, &key, slot);
2722
2723 if (key.objectid != btrfs_ino(inode))
2724 break;
2725 if (key.type != BTRFS_EXTENT_DATA_KEY)
2726 break;
2727 if (key.offset >= new->file_pos + new->len)
2728 break;
2729
2730 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2731
2732 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2733 if (key.offset + num_bytes < new->file_pos)
2734 goto next;
2735
2736 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2737 if (!disk_bytenr)
2738 goto next;
2739
2740 extent_offset = btrfs_file_extent_offset(l, extent);
2741
2742 old = kmalloc(sizeof(*old), GFP_NOFS);
2743 if (!old)
6f519564 2744 goto out_free_path;
38c227d8
LB		 2745
2746 offset = max(new->file_pos, key.offset);
2747 end = min(new->file_pos + new->len, key.offset + num_bytes);
2748
2749 old->bytenr = disk_bytenr;
2750 old->extent_offset = extent_offset;
2751 old->offset = offset - key.offset;
2752 old->len = end - offset;
2753 old->new = new;
2754 old->count = 0;
2755 list_add_tail(&old->list, &new->head);
2756next:
2757 path->slots[0]++;
2758 cond_resched();
2759 }
2760
2761 btrfs_free_path(path);
2762 atomic_inc(&root->fs_info->defrag_running);
2763
2764 return new;
2765
38c227d8
LB		 2766out_free_path:
2767 btrfs_free_path(path);
2768out_kfree:
6f519564 2769 free_sa_defrag_extent(new);
38c227d8
LB		 2770 return NULL;
2771}
2772
e570fd27
MX		 2773static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2774 u64 start, u64 len)
2775{
2776 struct btrfs_block_group_cache *cache;
2777
2778 cache = btrfs_lookup_block_group(root->fs_info, start);
2779 ASSERT(cache);
2780
2781 spin_lock(&cache->lock);
2782 cache->delalloc_bytes -= len;
2783 spin_unlock(&cache->lock);
2784
2785 btrfs_put_block_group(cache);
2786}
2787
d352ac68
CM		 2788/* as ordered data IO finishes, this gets called so we can finish
2789 * an ordered extent if the range of bytes in the file it covers are
2790 * fully written.
2791 */
5fd02043 2792static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
e6dcd2dc 2793{
5fd02043 2794 struct inode *inode = ordered_extent->inode;
e6dcd2dc 2795 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 2796 struct btrfs_trans_handle *trans = NULL;
e6dcd2dc 2797 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2ac55d41 2798 struct extent_state *cached_state = NULL;
38c227d8 2799 struct new_sa_defrag_extent *new = NULL;
261507a0 2800 int compress_type = 0;
77cef2ec
JB		 2801 int ret = 0;
2802 u64 logical_len = ordered_extent->len;
82d5902d 2803 bool nolock;
77cef2ec 2804 bool truncated = false;
e6dcd2dc 2805
83eea1f1 2806 nolock = btrfs_is_free_space_inode(inode);
0cb59c99 2807
5fd02043
JB		 2808 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2809 ret = -EIO;
2810 goto out;
2811 }
2812
f612496b
MX		 2813 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2814 ordered_extent->file_offset +
2815 ordered_extent->len - 1);
2816
77cef2ec
JB		 2817 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2818 truncated = true;
2819 logical_len = ordered_extent->truncated_len;
2820 /* Truncated the entire extent, don't bother adding */
2821 if (!logical_len)
2822 goto out;
2823 }
2824
c2167754 2825 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
79787eaa 2826 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
6c760c07
JB		 2827 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2828 if (nolock)
2829 trans = btrfs_join_transaction_nolock(root);
2830 else
2831 trans = btrfs_join_transaction(root);
2832 if (IS_ERR(trans)) {
2833 ret = PTR_ERR(trans);
2834 trans = NULL;
2835 goto out;
c2167754 2836 }
6c760c07
JB		 2837 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2838 ret = btrfs_update_inode_fallback(trans, root, inode);
2839 if (ret) /* -ENOMEM or corruption */
2840 btrfs_abort_transaction(trans, root, ret);
c2167754
YZ		 2841 goto out;
2842 }
e6dcd2dc 2843
2ac55d41
JB		 2844 lock_extent_bits(io_tree, ordered_extent->file_offset,
2845 ordered_extent->file_offset + ordered_extent->len - 1,
d0082371 2846 0, &cached_state);
e6dcd2dc 2847
38c227d8
LB		 2848 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2849 ordered_extent->file_offset + ordered_extent->len - 1,
2850 EXTENT_DEFRAG, 1, cached_state);
2851 if (ret) {
2852 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
8101c8db 2853 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
38c227d8
LB		 2854 /* the inode is shared */
2855 new = record_old_file_extents(inode, ordered_extent);
2856
2857 clear_extent_bit(io_tree, ordered_extent->file_offset,
2858 ordered_extent->file_offset + ordered_extent->len - 1,
2859 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2860 }
2861
0cb59c99 2862 if (nolock)
7a7eaa40 2863 trans = btrfs_join_transaction_nolock(root);
0cb59c99 2864 else
7a7eaa40 2865 trans = btrfs_join_transaction(root);
79787eaa
JM		 2866 if (IS_ERR(trans)) {
2867 ret = PTR_ERR(trans);
2868 trans = NULL;
2869 goto out_unlock;
2870 }
a79b7d4b 2871
0ca1f7ce 2872 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
c2167754 2873
c8b97818 2874 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
261507a0 2875 compress_type = ordered_extent->compress_type;
d899e052 2876 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
261507a0 2877 BUG_ON(compress_type);
920bbbfb 2878 ret = btrfs_mark_extent_written(trans, inode,
d899e052
YZ		 2879 ordered_extent->file_offset,
2880 ordered_extent->file_offset +
77cef2ec 2881 logical_len);
d899e052 2882 } else {
0af3d00b 2883 BUG_ON(root == root->fs_info->tree_root);
d899e052
YZ		 2884 ret = insert_reserved_file_extent(trans, inode,
2885 ordered_extent->file_offset,
2886 ordered_extent->start,
2887 ordered_extent->disk_len,
77cef2ec 2888 logical_len, logical_len,
261507a0 2889 compress_type, 0, 0,
d899e052 2890 BTRFS_FILE_EXTENT_REG);
e570fd27
MX		 2891 if (!ret)
2892 btrfs_release_delalloc_bytes(root,
2893 ordered_extent->start,
2894 ordered_extent->disk_len);
d899e052 2895 }
5dc562c5
JB		 2896 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2897 ordered_extent->file_offset, ordered_extent->len,
2898 trans->transid);
79787eaa
JM		 2899 if (ret < 0) {
2900 btrfs_abort_transaction(trans, root, ret);
5fd02043 2901 goto out_unlock;
79787eaa 2902 }
2ac55d41 2903
e6dcd2dc
CM		 2904 add_pending_csums(trans, inode, ordered_extent->file_offset,
2905 &ordered_extent->list);
2906
6c760c07
JB		 2907 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2908 ret = btrfs_update_inode_fallback(trans, root, inode);
2909 if (ret) { /* -ENOMEM or corruption */
2910 btrfs_abort_transaction(trans, root, ret);
2911 goto out_unlock;
1ef30be1
JB		 2912 }
2913 ret = 0;
5fd02043
JB		 2914out_unlock:
2915 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2916 ordered_extent->file_offset +
2917 ordered_extent->len - 1, &cached_state, GFP_NOFS);
c2167754 2918out:
5b0e95bf 2919 if (root != root->fs_info->tree_root)
0cb59c99 2920 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
a698d075
MX		 2921 if (trans)
2922 btrfs_end_transaction(trans, root);
0cb59c99 2923
77cef2ec
JB		 2924 if (ret || truncated) {
2925 u64 start, end;
2926
2927 if (truncated)
2928 start = ordered_extent->file_offset + logical_len;
2929 else
2930 start = ordered_extent->file_offset;
2931 end = ordered_extent->file_offset + ordered_extent->len - 1;
2932 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2933
2934 /* Drop the cache for the part of the extent we didn't write. */
2935 btrfs_drop_extent_cache(inode, start, end, 0);
5fd02043 2936
0bec9ef5
JB		 2937 /*
2938 * If the ordered extent had an IOERR or something else went
2939 * wrong we need to return the space for this ordered extent
77cef2ec
JB		 2940 * back to the allocator. We only free the extent in the
2941 * truncated case if we didn't write out the extent at all.
0bec9ef5 2942 */
77cef2ec
JB		 2943 if ((ret || !logical_len) &&
2944 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
0bec9ef5
JB		 2945 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2946 btrfs_free_reserved_extent(root, ordered_extent->start,
e570fd27 2947 ordered_extent->disk_len, 1);
0bec9ef5
JB		 2948 }
2949
2950
5fd02043 2951 /*
8bad3c02
LB		 2952 * This needs to be done to make sure anybody waiting knows we are done
2953 * updating everything for this ordered extent.
5fd02043
JB		 2954 */
2955 btrfs_remove_ordered_extent(inode, ordered_extent);
2956
38c227d8 2957 /* for snapshot-aware defrag */
6f519564
LB		 2958 if (new) {
2959 if (ret) {
2960 free_sa_defrag_extent(new);
2961 atomic_dec(&root->fs_info->defrag_running);
2962 } else {
2963 relink_file_extents(new);
2964 }
2965 }
38c227d8 2966
e6dcd2dc
CM		 2967 /* once for us */
2968 btrfs_put_ordered_extent(ordered_extent);
2969 /* once for the tree */
2970 btrfs_put_ordered_extent(ordered_extent);
2971
5fd02043
JB		 2972 return ret;
2973}
2974
2975static void finish_ordered_fn(struct btrfs_work *work)
2976{
2977 struct btrfs_ordered_extent *ordered_extent;
2978 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2979 btrfs_finish_ordered_io(ordered_extent);
e6dcd2dc
CM		 2980}
2981
b2950863 2982static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM		 2983 struct extent_state *state, int uptodate)
2984{
5fd02043
JB		 2985 struct inode *inode = page->mapping->host;
2986 struct btrfs_root *root = BTRFS_I(inode)->root;
2987 struct btrfs_ordered_extent *ordered_extent = NULL;
9e0af237
LB		 2988 struct btrfs_workqueue *wq;
2989 btrfs_work_func_t func;
5fd02043 2990
1abe9b8a 2991 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2992
8b62b72b 2993 ClearPagePrivate2(page);
5fd02043
JB		 2994 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2995 end - start + 1, uptodate))
2996 return 0;
2997
9e0af237
LB		 2998 if (btrfs_is_free_space_inode(inode)) {
2999 wq = root->fs_info->endio_freespace_worker;
3000 func = btrfs_freespace_write_helper;
3001 } else {
3002 wq = root->fs_info->endio_write_workers;
3003 func = btrfs_endio_write_helper;
3004 }
5fd02043 3005
9e0af237
LB		 3006 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3007 NULL);
3008 btrfs_queue_work(wq, &ordered_extent->work);
5fd02043
JB		 3009
3010 return 0;
211f90e6
CM		 3011}
3012
dc380aea
MX		 3013static int __readpage_endio_check(struct inode *inode,
3014 struct btrfs_io_bio *io_bio,
3015 int icsum, struct page *page,
3016 int pgoff, u64 start, size_t len)
3017{
3018 char *kaddr;
3019 u32 csum_expected;
3020 u32 csum = ~(u32)0;
3021 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
3022 DEFAULT_RATELIMIT_BURST);
3023
3024 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3025
3026 kaddr = kmap_atomic(page);
3027 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3028 btrfs_csum_final(csum, (char *)&csum);
3029 if (csum != csum_expected)
3030 goto zeroit;
3031
3032 kunmap_atomic(kaddr);
3033 return 0;
3034zeroit:
3035 if (__ratelimit(&_rs))
f0954c66 3036 btrfs_warn(BTRFS_I(inode)->root->fs_info,
dc380aea
MX		 3037 "csum failed ino %llu off %llu csum %u expected csum %u",
3038 btrfs_ino(inode), start, csum, csum_expected);
3039 memset(kaddr + pgoff, 1, len);
3040 flush_dcache_page(page);
3041 kunmap_atomic(kaddr);
3042 if (csum_expected == 0)
3043 return 0;
3044 return -EIO;
3045}
3046
d352ac68
CM		 3047/*
3048 * when reads are done, we need to check csums to verify the data is correct
4a54c8c1
JS		 3049 * if there's a match, we allow the bio to finish. If not, the code in
3050 * extent_io.c will try to find good copies for us.
d352ac68 3051 */
facc8a22
MX		 3052static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3053 u64 phy_offset, struct page *page,
3054 u64 start, u64 end, int mirror)
07157aac 3055{
4eee4fa4 3056 size_t offset = start - page_offset(page);
07157aac 3057 struct inode *inode = page->mapping->host;
d1310b2e 3058 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
ff79f819 3059 struct btrfs_root *root = BTRFS_I(inode)->root;
d1310b2e 3060
d20f7043
CM		 3061 if (PageChecked(page)) {
3062 ClearPageChecked(page);
dc380aea 3063 return 0;
d20f7043 3064 }
6cbff00f
CH		 3065
3066 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
dc380aea 3067 return 0;
17d217fe
YZ		 3068
3069 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
9655d298 3070 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
17d217fe
YZ		 3071 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
3072 GFP_NOFS);
b6cda9bc 3073 return 0;
17d217fe 3074 }
d20f7043 3075
facc8a22 3076 phy_offset >>= inode->i_sb->s_blocksize_bits;
dc380aea
MX		 3077 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3078 start, (size_t)(end - start + 1));
07157aac 3079}
b888db2b 3080
24bbcf04
YZ		 3081struct delayed_iput {
3082 struct list_head list;
3083 struct inode *inode;
3084};
3085
79787eaa
JM		 3086/* JDM: If this is fs-wide, why can't we add a pointer to
3087 * btrfs_inode instead and avoid the allocation? */
24bbcf04
YZ		 3088void btrfs_add_delayed_iput(struct inode *inode)
3089{
3090 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3091 struct delayed_iput *delayed;
3092
3093 if (atomic_add_unless(&inode->i_count, -1, 1))
3094 return;
3095
3096 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
3097 delayed->inode = inode;
3098
3099 spin_lock(&fs_info->delayed_iput_lock);
3100 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
3101 spin_unlock(&fs_info->delayed_iput_lock);
3102}
3103
3104void btrfs_run_delayed_iputs(struct btrfs_root *root)
3105{
3106 LIST_HEAD(list);
3107 struct btrfs_fs_info *fs_info = root->fs_info;
3108 struct delayed_iput *delayed;
3109 int empty;
3110
3111 spin_lock(&fs_info->delayed_iput_lock);
3112 empty = list_empty(&fs_info->delayed_iputs);
3113 spin_unlock(&fs_info->delayed_iput_lock);
3114 if (empty)
3115 return;
3116
d7c15171
ZL		 3117 down_read(&fs_info->delayed_iput_sem);
3118
24bbcf04
YZ		 3119 spin_lock(&fs_info->delayed_iput_lock);
3120 list_splice_init(&fs_info->delayed_iputs, &list);
3121 spin_unlock(&fs_info->delayed_iput_lock);
3122
3123 while (!list_empty(&list)) {
3124 delayed = list_entry(list.next, struct delayed_iput, list);
3125 list_del(&delayed->list);
3126 iput(delayed->inode);
3127 kfree(delayed);
3128 }
d7c15171
ZL		 3129
3130 up_read(&root->fs_info->delayed_iput_sem);
24bbcf04
YZ		 3131}
3132
d68fc57b 3133/*
42b2aa86 3134 * This is called in transaction commit time. If there are no orphan
d68fc57b
YZ		 3135 * files in the subvolume, it removes orphan item and frees block_rsv
3136 * structure.
3137 */
3138void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3139 struct btrfs_root *root)
3140{
90290e19 3141 struct btrfs_block_rsv *block_rsv;
d68fc57b
YZ		 3142 int ret;
3143
8a35d95f 3144 if (atomic_read(&root->orphan_inodes) ||
d68fc57b
YZ		 3145 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3146 return;
3147
90290e19 3148 spin_lock(&root->orphan_lock);
8a35d95f 3149 if (atomic_read(&root->orphan_inodes)) {
90290e19
JB		 3150 spin_unlock(&root->orphan_lock);
3151 return;
3152 }
3153
3154 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3155 spin_unlock(&root->orphan_lock);
3156 return;
3157 }
3158
3159 block_rsv = root->orphan_block_rsv;
3160 root->orphan_block_rsv = NULL;
3161 spin_unlock(&root->orphan_lock);
3162
27cdeb70 3163 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
d68fc57b
YZ		 3164 btrfs_root_refs(&root->root_item) > 0) {
3165 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3166 root->root_key.objectid);
4ef31a45
JB		 3167 if (ret)
3168 btrfs_abort_transaction(trans, root, ret);
3169 else
27cdeb70
MX		 3170 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3171 &root->state);
d68fc57b
YZ		 3172 }
3173
90290e19
JB		 3174 if (block_rsv) {
3175 WARN_ON(block_rsv->size > 0);
3176 btrfs_free_block_rsv(root, block_rsv);
d68fc57b
YZ		 3177 }
3178}
3179
7b128766
JB		 3180/*
3181 * This creates an orphan entry for the given inode in case something goes
3182 * wrong in the middle of an unlink/truncate.
d68fc57b
YZ		 3183 *
3184 * NOTE: caller of this function should reserve 5 units of metadata for
3185 * this function.
7b128766
JB		 3186 */
3187int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3188{
3189 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 3190 struct btrfs_block_rsv *block_rsv = NULL;
3191 int reserve = 0;
3192 int insert = 0;
3193 int ret;
7b128766 3194
d68fc57b 3195 if (!root->orphan_block_rsv) {
66d8f3dd 3196 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
b532402e
TI		 3197 if (!block_rsv)
3198 return -ENOMEM;
d68fc57b 3199 }
7b128766 3200
d68fc57b
YZ		 3201 spin_lock(&root->orphan_lock);
3202 if (!root->orphan_block_rsv) {
3203 root->orphan_block_rsv = block_rsv;
3204 } else if (block_rsv) {
3205 btrfs_free_block_rsv(root, block_rsv);
3206 block_rsv = NULL;
7b128766 3207 }
7b128766 3208
8a35d95f
JB		 3209 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3210 &BTRFS_I(inode)->runtime_flags)) {
d68fc57b
YZ		 3211#if 0
3212 /*
3213 * For proper ENOSPC handling, we should do orphan
3214 * cleanup when mounting. But this introduces backward
3215 * compatibility issue.
3216 */
3217 if (!xchg(&root->orphan_item_inserted, 1))
3218 insert = 2;
3219 else
3220 insert = 1;
3221#endif
3222 insert = 1;
321f0e70 3223 atomic_inc(&root->orphan_inodes);
7b128766
JB		 3224 }
3225
72ac3c0d
JB		 3226 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3227 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3228 reserve = 1;
d68fc57b 3229 spin_unlock(&root->orphan_lock);
7b128766 3230
d68fc57b
YZ		 3231 /* grab metadata reservation from transaction handle */
3232 if (reserve) {
3233 ret = btrfs_orphan_reserve_metadata(trans, inode);
79787eaa 3234 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
d68fc57b 3235 }
7b128766 3236
d68fc57b
YZ		 3237 /* insert an orphan item to track this unlinked/truncated file */
3238 if (insert >= 1) {
33345d01 3239 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
4ef31a45 3240 if (ret) {
703c88e0 3241 atomic_dec(&root->orphan_inodes);
4ef31a45
JB		 3242 if (reserve) {
3243 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3244 &BTRFS_I(inode)->runtime_flags);
3245 btrfs_orphan_release_metadata(inode);
3246 }
3247 if (ret != -EEXIST) {
e8e7cff6
JB		 3248 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3249 &BTRFS_I(inode)->runtime_flags);
4ef31a45
JB		 3250 btrfs_abort_transaction(trans, root, ret);
3251 return ret;
3252 }
79787eaa
JM		 3253 }
3254 ret = 0;
d68fc57b
YZ		 3255 }
3256
3257 /* insert an orphan item to track subvolume contains orphan files */
3258 if (insert >= 2) {
3259 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3260 root->root_key.objectid);
79787eaa
JM		 3261 if (ret && ret != -EEXIST) {
3262 btrfs_abort_transaction(trans, root, ret);
3263 return ret;
3264 }
d68fc57b
YZ		 3265 }
3266 return 0;
7b128766
JB		 3267}
3268
3269/*
3270 * We have done the truncate/delete so we can go ahead and remove the orphan
3271 * item for this particular inode.
3272 */
48a3b636
ES		 3273static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3274 struct inode *inode)
7b128766
JB		 3275{
3276 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 3277 int delete_item = 0;
3278 int release_rsv = 0;
7b128766
JB		 3279 int ret = 0;
3280
d68fc57b 3281 spin_lock(&root->orphan_lock);
8a35d95f
JB		 3282 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3283 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3284 delete_item = 1;
7b128766 3285
72ac3c0d
JB		 3286 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3287 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3288 release_rsv = 1;
d68fc57b 3289 spin_unlock(&root->orphan_lock);
7b128766 3290
703c88e0 3291 if (delete_item) {
8a35d95f 3292 atomic_dec(&root->orphan_inodes);
703c88e0
FDBM		 3293 if (trans)
3294 ret = btrfs_del_orphan_item(trans, root,
3295 btrfs_ino(inode));
8a35d95f 3296 }
7b128766 3297
703c88e0
FDBM		 3298 if (release_rsv)
3299 btrfs_orphan_release_metadata(inode);
3300
4ef31a45 3301 return ret;
7b128766
JB		 3302}
3303
3304/*
3305 * this cleans up any orphans that may be left on the list from the last use
3306 * of this root.
3307 */
66b4ffd1 3308int btrfs_orphan_cleanup(struct btrfs_root *root)
7b128766
JB		 3309{
3310 struct btrfs_path *path;
3311 struct extent_buffer *leaf;
7b128766
JB		 3312 struct btrfs_key key, found_key;
3313 struct btrfs_trans_handle *trans;
3314 struct inode *inode;
8f6d7f4f 3315 u64 last_objectid = 0;
7b128766
JB		 3316 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3317
d68fc57b 3318 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
66b4ffd1 3319 return 0;
c71bf099
YZ		 3320
3321 path = btrfs_alloc_path();
66b4ffd1
JB		 3322 if (!path) {
3323 ret = -ENOMEM;
3324 goto out;
3325 }
7b128766
JB		 3326 path->reada = -1;
3327
3328 key.objectid = BTRFS_ORPHAN_OBJECTID;
962a298f 3329 key.type = BTRFS_ORPHAN_ITEM_KEY;
7b128766
JB		 3330 key.offset = (u64)-1;
3331
7b128766
JB		 3332 while (1) {
3333 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
66b4ffd1
JB		 3334 if (ret < 0)
3335 goto out;
7b128766
JB		 3336
3337 /*
3338 * if ret == 0 means we found what we were searching for, which
25985edc 3339 * is weird, but possible, so only screw with path if we didn't
7b128766
JB		 3340 * find the key and see if we have stuff that matches
3341 */
3342 if (ret > 0) {
66b4ffd1 3343 ret = 0;
7b128766
JB		 3344 if (path->slots[0] == 0)
3345 break;
3346 path->slots[0]--;
3347 }
3348
3349 /* pull out the item */
3350 leaf = path->nodes[0];
7b128766
JB		 3351 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3352
3353 /* make sure the item matches what we want */
3354 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3355 break;
962a298f 3356 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
7b128766
JB		 3357 break;
3358
3359 /* release the path since we're done with it */
b3b4aa74 3360 btrfs_release_path(path);
7b128766
JB		 3361
3362 /*
3363 * this is where we are basically btrfs_lookup, without the
3364 * crossing root thing. we store the inode number in the
3365 * offset of the orphan item.
3366 */
8f6d7f4f
JB		 3367
3368 if (found_key.offset == last_objectid) {
c2cf52eb
SK		 3369 btrfs_err(root->fs_info,
3370 "Error removing orphan entry, stopping orphan cleanup");
8f6d7f4f
JB		 3371 ret = -EINVAL;
3372 goto out;
3373 }
3374
3375 last_objectid = found_key.offset;
3376
5d4f98a2
YZ		 3377 found_key.objectid = found_key.offset;
3378 found_key.type = BTRFS_INODE_ITEM_KEY;
3379 found_key.offset = 0;
73f73415 3380 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
8c6ffba0 3381 ret = PTR_ERR_OR_ZERO(inode);
a8c9e576 3382 if (ret && ret != -ESTALE)
66b4ffd1 3383 goto out;
7b128766 3384
f8e9e0b0
AJ		 3385 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3386 struct btrfs_root *dead_root;
3387 struct btrfs_fs_info *fs_info = root->fs_info;
3388 int is_dead_root = 0;
3389
3390 /*
3391 * this is an orphan in the tree root. Currently these
3392 * could come from 2 sources:
3393 * a) a snapshot deletion in progress
3394 * b) a free space cache inode
3395 * We need to distinguish those two, as the snapshot
3396 * orphan must not get deleted.
3397 * find_dead_roots already ran before us, so if this
3398 * is a snapshot deletion, we should find the root
3399 * in the dead_roots list
3400 */
3401 spin_lock(&fs_info->trans_lock);
3402 list_for_each_entry(dead_root, &fs_info->dead_roots,
3403 root_list) {
3404 if (dead_root->root_key.objectid ==
3405 found_key.objectid) {
3406 is_dead_root = 1;
3407 break;
3408 }
3409 }
3410 spin_unlock(&fs_info->trans_lock);
3411 if (is_dead_root) {
3412 /* prevent this orphan from being found again */
3413 key.offset = found_key.objectid - 1;
3414 continue;
3415 }
3416 }
7b128766 3417 /*
a8c9e576
JB		 3418 * Inode is already gone but the orphan item is still there,
3419 * kill the orphan item.
7b128766 3420 */
a8c9e576
JB		 3421 if (ret == -ESTALE) {
3422 trans = btrfs_start_transaction(root, 1);
66b4ffd1
JB		 3423 if (IS_ERR(trans)) {
3424 ret = PTR_ERR(trans);
3425 goto out;
3426 }
c2cf52eb
SK		 3427 btrfs_debug(root->fs_info, "auto deleting %Lu",
3428 found_key.objectid);
a8c9e576
JB		 3429 ret = btrfs_del_orphan_item(trans, root,
3430 found_key.objectid);
5b21f2ed 3431 btrfs_end_transaction(trans, root);
4ef31a45
JB		 3432 if (ret)
3433 goto out;
7b128766
JB		 3434 continue;
3435 }
3436
a8c9e576
JB		 3437 /*
3438 * add this inode to the orphan list so btrfs_orphan_del does
3439 * the proper thing when we hit it
3440 */
8a35d95f
JB		 3441 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3442 &BTRFS_I(inode)->runtime_flags);
925396ec 3443 atomic_inc(&root->orphan_inodes);
a8c9e576 3444
7b128766
JB		 3445 /* if we have links, this was a truncate, lets do that */
3446 if (inode->i_nlink) {
fae7f21c 3447 if (WARN_ON(!S_ISREG(inode->i_mode))) {
a41ad394
JB		 3448 iput(inode);
3449 continue;
3450 }
7b128766 3451 nr_truncate++;
f3fe820c
JB		 3452
3453 /* 1 for the orphan item deletion. */
3454 trans = btrfs_start_transaction(root, 1);
3455 if (IS_ERR(trans)) {
c69b26b0 3456 iput(inode);
f3fe820c
JB		 3457 ret = PTR_ERR(trans);
3458 goto out;
3459 }
3460 ret = btrfs_orphan_add(trans, inode);
3461 btrfs_end_transaction(trans, root);
c69b26b0
JB		 3462 if (ret) {
3463 iput(inode);
f3fe820c 3464 goto out;
c69b26b0 3465 }
f3fe820c 3466
66b4ffd1 3467 ret = btrfs_truncate(inode);
4a7d0f68
JB		 3468 if (ret)
3469 btrfs_orphan_del(NULL, inode);
7b128766
JB		 3470 } else {
3471 nr_unlink++;
3472 }
3473
3474 /* this will do delete_inode and everything for us */
3475 iput(inode);
66b4ffd1
JB		 3476 if (ret)
3477 goto out;
7b128766 3478 }
3254c876
MX		 3479 /* release the path since we're done with it */
3480 btrfs_release_path(path);
3481
d68fc57b
YZ		 3482 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3483
3484 if (root->orphan_block_rsv)
3485 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3486 (u64)-1);
3487
27cdeb70
MX		 3488 if (root->orphan_block_rsv ||
3489 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
7a7eaa40 3490 trans = btrfs_join_transaction(root);
66b4ffd1
JB		 3491 if (!IS_ERR(trans))
3492 btrfs_end_transaction(trans, root);
d68fc57b 3493 }
7b128766
JB		 3494
3495 if (nr_unlink)
4884b476 3496 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
7b128766 3497 if (nr_truncate)
4884b476 3498 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
66b4ffd1
JB		 3499
3500out:
3501 if (ret)
68b663d1 3502 btrfs_err(root->fs_info,
c2cf52eb 3503 "could not do orphan cleanup %d", ret);
66b4ffd1
JB		 3504 btrfs_free_path(path);
3505 return ret;
7b128766
JB		 3506}
3507
46a53cca
CM		 3508/*
3509 * very simple check to peek ahead in the leaf looking for xattrs. If we
3510 * don't find any xattrs, we know there can't be any acls.
3511 *
3512 * slot is the slot the inode is in, objectid is the objectid of the inode
3513 */
3514static noinline int acls_after_inode_item(struct extent_buffer *leaf,
63541927
FDBM		 3515 int slot, u64 objectid,
3516 int *first_xattr_slot)
46a53cca
CM		 3517{
3518 u32 nritems = btrfs_header_nritems(leaf);
3519 struct btrfs_key found_key;
f23b5a59
JB		 3520 static u64 xattr_access = 0;
3521 static u64 xattr_default = 0;
46a53cca
CM		 3522 int scanned = 0;
3523
f23b5a59
JB		 3524 if (!xattr_access) {
3525 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3526 strlen(POSIX_ACL_XATTR_ACCESS));
3527 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3528 strlen(POSIX_ACL_XATTR_DEFAULT));
3529 }
3530
46a53cca 3531 slot++;
63541927 3532 *first_xattr_slot = -1;
46a53cca
CM		 3533 while (slot < nritems) {
3534 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3535
3536 /* we found a different objectid, there must not be acls */
3537 if (found_key.objectid != objectid)
3538 return 0;
3539
3540 /* we found an xattr, assume we've got an acl */
f23b5a59 3541 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
63541927
FDBM		 3542 if (*first_xattr_slot == -1)
3543 *first_xattr_slot = slot;
f23b5a59
JB		 3544 if (found_key.offset == xattr_access ||
3545 found_key.offset == xattr_default)
3546 return 1;
3547 }
46a53cca
CM		 3548
3549 /*
3550 * we found a key greater than an xattr key, there can't
3551 * be any acls later on
3552 */
3553 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3554 return 0;
3555
3556 slot++;
3557 scanned++;
3558
3559 /*
3560 * it goes inode, inode backrefs, xattrs, extents,
3561 * so if there are a ton of hard links to an inode there can
3562 * be a lot of backrefs. Don't waste time searching too hard,
3563 * this is just an optimization
3564 */
3565 if (scanned >= 8)
3566 break;
3567 }
3568 /* we hit the end of the leaf before we found an xattr or
3569 * something larger than an xattr. We have to assume the inode
3570 * has acls
3571 */
63541927
FDBM		 3572 if (*first_xattr_slot == -1)
3573 *first_xattr_slot = slot;
46a53cca
CM		 3574 return 1;
3575}
3576
d352ac68
CM		 3577/*
3578 * read an inode from the btree into the in-memory inode
3579 */
5d4f98a2 3580static void btrfs_read_locked_inode(struct inode *inode)
39279cc3
CM		 3581{
3582 struct btrfs_path *path;
5f39d397 3583 struct extent_buffer *leaf;
39279cc3
CM		 3584 struct btrfs_inode_item *inode_item;
3585 struct btrfs_root *root = BTRFS_I(inode)->root;
3586 struct btrfs_key location;
67de1176 3587 unsigned long ptr;
46a53cca 3588 int maybe_acls;
618e21d5 3589 u32 rdev;
39279cc3 3590 int ret;
2f7e33d4 3591 bool filled = false;
63541927 3592 int first_xattr_slot;
2f7e33d4
MX		 3593
3594 ret = btrfs_fill_inode(inode, &rdev);
3595 if (!ret)
3596 filled = true;
39279cc3
CM		 3597
3598 path = btrfs_alloc_path();
1748f843
MF		 3599 if (!path)
3600 goto make_bad;
3601
39279cc3 3602 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 3603
39279cc3 3604 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 3605 if (ret)
39279cc3 3606 goto make_bad;
39279cc3 3607
5f39d397 3608 leaf = path->nodes[0];
2f7e33d4
MX		 3609
3610 if (filled)
67de1176 3611 goto cache_index;
2f7e33d4 3612
5f39d397
CM		 3613 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3614 struct btrfs_inode_item);
5f39d397 3615 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
bfe86848 3616 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2f2f43d3
EB		 3617 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3618 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
dbe674a9 3619 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397 3620
a937b979
DS		 3621 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3622 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
5f39d397 3623
a937b979
DS		 3624 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3625 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
5f39d397 3626
a937b979
DS		 3627 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3628 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
5f39d397 3629
9cc97d64 3630 BTRFS_I(inode)->i_otime.tv_sec =
3631 btrfs_timespec_sec(leaf, &inode_item->otime);
3632 BTRFS_I(inode)->i_otime.tv_nsec =
3633 btrfs_timespec_nsec(leaf, &inode_item->otime);
5f39d397 3634
a76a3cd4 3635 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 3636 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
5dc562c5
JB		 3637 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3638
6e17d30b
YD		 3639 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3640 inode->i_generation = BTRFS_I(inode)->generation;
3641 inode->i_rdev = 0;
3642 rdev = btrfs_inode_rdev(leaf, inode_item);
3643
3644 BTRFS_I(inode)->index_cnt = (u64)-1;
3645 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3646
3647cache_index:
5dc562c5
JB		 3648 /*
3649 * If we were modified in the current generation and evicted from memory
3650 * and then re-read we need to do a full sync since we don't have any
3651 * idea about which extents were modified before we were evicted from
3652 * cache.
6e17d30b
YD		 3653 *
3654 * This is required for both inode re-read from disk and delayed inode
3655 * in delayed_nodes_tree.
5dc562c5
JB		 3656 */
3657 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3658 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3659 &BTRFS_I(inode)->runtime_flags);
3660
67de1176
MX		 3661 path->slots[0]++;
3662 if (inode->i_nlink != 1 ||
3663 path->slots[0] >= btrfs_header_nritems(leaf))
3664 goto cache_acl;
3665
3666 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3667 if (location.objectid != btrfs_ino(inode))
3668 goto cache_acl;
3669
3670 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3671 if (location.type == BTRFS_INODE_REF_KEY) {
3672 struct btrfs_inode_ref *ref;
3673
3674 ref = (struct btrfs_inode_ref *)ptr;
3675 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3676 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3677 struct btrfs_inode_extref *extref;
3678
3679 extref = (struct btrfs_inode_extref *)ptr;
3680 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3681 extref);
3682 }
2f7e33d4 3683cache_acl:
46a53cca
CM		 3684 /*
3685 * try to precache a NULL acl entry for files that don't have
3686 * any xattrs or acls
3687 */
33345d01 3688 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
63541927
FDBM		 3689 btrfs_ino(inode), &first_xattr_slot);
3690 if (first_xattr_slot != -1) {
3691 path->slots[0] = first_xattr_slot;
3692 ret = btrfs_load_inode_props(inode, path);
3693 if (ret)
3694 btrfs_err(root->fs_info,
351fd353 3695 "error loading props for ino %llu (root %llu): %d",
63541927
FDBM		 3696 btrfs_ino(inode),
3697 root->root_key.objectid, ret);
3698 }
3699 btrfs_free_path(path);
3700
72c04902
AV		 3701 if (!maybe_acls)
3702 cache_no_acl(inode);
46a53cca 3703
39279cc3 3704 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 3705 case S_IFREG:
3706 inode->i_mapping->a_ops = &btrfs_aops;
d1310b2e 3707 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 3708 inode->i_fop = &btrfs_file_operations;
3709 inode->i_op = &btrfs_file_inode_operations;
3710 break;
3711 case S_IFDIR:
3712 inode->i_fop = &btrfs_dir_file_operations;
3713 if (root == root->fs_info->tree_root)
3714 inode->i_op = &btrfs_dir_ro_inode_operations;
3715 else
3716 inode->i_op = &btrfs_dir_inode_operations;
3717 break;
3718 case S_IFLNK:
3719 inode->i_op = &btrfs_symlink_inode_operations;
3720 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3721 break;
618e21d5 3722 default:
0279b4cd 3723 inode->i_op = &btrfs_special_inode_operations;
618e21d5
JB		 3724 init_special_inode(inode, inode->i_mode, rdev);
3725 break;
39279cc3 3726 }
6cbff00f
CH		 3727
3728 btrfs_update_iflags(inode);
39279cc3
CM		 3729 return;
3730
3731make_bad:
39279cc3 3732 btrfs_free_path(path);
39279cc3
CM		 3733 make_bad_inode(inode);
3734}
3735
d352ac68
CM		 3736/*
3737 * given a leaf and an inode, copy the inode fields into the leaf
3738 */
e02119d5
CM		 3739static void fill_inode_item(struct btrfs_trans_handle *trans,
3740 struct extent_buffer *leaf,
5f39d397 3741 struct btrfs_inode_item *item,
39279cc3
CM		 3742 struct inode *inode)
3743{
51fab693
LB		 3744 struct btrfs_map_token token;
3745
3746 btrfs_init_map_token(&token);
5f39d397 3747
51fab693
LB		 3748 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3749 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3750 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3751 &token);
3752 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3753 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
5f39d397 3754
a937b979 3755 btrfs_set_token_timespec_sec(leaf, &item->atime,
51fab693 3756 inode->i_atime.tv_sec, &token);
a937b979 3757 btrfs_set_token_timespec_nsec(leaf, &item->atime,
51fab693 3758 inode->i_atime.tv_nsec, &token);
5f39d397 3759
a937b979 3760 btrfs_set_token_timespec_sec(leaf, &item->mtime,
51fab693 3761 inode->i_mtime.tv_sec, &token);
a937b979 3762 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
51fab693 3763 inode->i_mtime.tv_nsec, &token);
5f39d397 3764
a937b979 3765 btrfs_set_token_timespec_sec(leaf, &item->ctime,
51fab693 3766 inode->i_ctime.tv_sec, &token);
a937b979 3767 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
51fab693 3768 inode->i_ctime.tv_nsec, &token);
5f39d397 3769
9cc97d64 3770 btrfs_set_token_timespec_sec(leaf, &item->otime,
3771 BTRFS_I(inode)->i_otime.tv_sec, &token);
3772 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3773 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3774
51fab693
LB		 3775 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3776 &token);
3777 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3778 &token);
3779 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3780 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3781 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3782 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3783 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
39279cc3
CM		 3784}
3785
d352ac68
CM		 3786/*
3787 * copy everything in the in-memory inode into the btree.
3788 */
2115133f 3789static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
d397712b 3790 struct btrfs_root *root, struct inode *inode)
39279cc3
CM		 3791{
3792 struct btrfs_inode_item *inode_item;
3793 struct btrfs_path *path;
5f39d397 3794 struct extent_buffer *leaf;
39279cc3
CM		 3795 int ret;
3796
3797 path = btrfs_alloc_path();
16cdcec7
MX		 3798 if (!path)
3799 return -ENOMEM;
3800
b9473439 3801 path->leave_spinning = 1;
16cdcec7
MX		 3802 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3803 1);
39279cc3
CM		 3804 if (ret) {
3805 if (ret > 0)
3806 ret = -ENOENT;
3807 goto failed;
3808 }
3809
5f39d397
CM		 3810 leaf = path->nodes[0];
3811 inode_item = btrfs_item_ptr(leaf, path->slots[0],
16cdcec7 3812 struct btrfs_inode_item);
39279cc3 3813
e02119d5 3814 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 3815 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 3816 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 3817 ret = 0;
3818failed:
39279cc3
CM		 3819 btrfs_free_path(path);
3820 return ret;
3821}
3822
2115133f
CM		 3823/*
3824 * copy everything in the in-memory inode into the btree.
3825 */
3826noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3827 struct btrfs_root *root, struct inode *inode)
3828{
3829 int ret;
3830
3831 /*
3832 * If the inode is a free space inode, we can deadlock during commit
3833 * if we put it into the delayed code.
3834 *
3835 * The data relocation inode should also be directly updated
3836 * without delay
3837 */
83eea1f1 3838 if (!btrfs_is_free_space_inode(inode)
1d52c78a
JB		 3839 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3840 && !root->fs_info->log_root_recovering) {
8ea05e3a
AB		 3841 btrfs_update_root_times(trans, root);
3842
2115133f
CM		 3843 ret = btrfs_delayed_update_inode(trans, root, inode);
3844 if (!ret)
3845 btrfs_set_inode_last_trans(trans, inode);
3846 return ret;
3847 }
3848
3849 return btrfs_update_inode_item(trans, root, inode);
3850}
3851
be6aef60
JB		 3852noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3853 struct btrfs_root *root,
3854 struct inode *inode)
2115133f
CM		 3855{
3856 int ret;
3857
3858 ret = btrfs_update_inode(trans, root, inode);
3859 if (ret == -ENOSPC)
3860 return btrfs_update_inode_item(trans, root, inode);
3861 return ret;
3862}
3863
d352ac68
CM		 3864/*
3865 * unlink helper that gets used here in inode.c and in the tree logging
3866 * recovery code. It remove a link in a directory with a given name, and
3867 * also drops the back refs in the inode to the directory
3868 */
92986796
AV		 3869static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3870 struct btrfs_root *root,
3871 struct inode *dir, struct inode *inode,
3872 const char *name, int name_len)
39279cc3
CM		 3873{
3874 struct btrfs_path *path;
39279cc3 3875 int ret = 0;
5f39d397 3876 struct extent_buffer *leaf;
39279cc3 3877 struct btrfs_dir_item *di;
5f39d397 3878 struct btrfs_key key;
aec7477b 3879 u64 index;
33345d01
LZ		 3880 u64 ino = btrfs_ino(inode);
3881 u64 dir_ino = btrfs_ino(dir);
39279cc3
CM		 3882
3883 path = btrfs_alloc_path();
54aa1f4d
CM		 3884 if (!path) {
3885 ret = -ENOMEM;
554233a6 3886 goto out;
54aa1f4d
CM		 3887 }
3888
b9473439 3889 path->leave_spinning = 1;
33345d01 3890 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
39279cc3
CM		 3891 name, name_len, -1);
3892 if (IS_ERR(di)) {
3893 ret = PTR_ERR(di);
3894 goto err;
3895 }
3896 if (!di) {
3897 ret = -ENOENT;
3898 goto err;
3899 }
5f39d397
CM		 3900 leaf = path->nodes[0];
3901 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 3902 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 3903 if (ret)
3904 goto err;
b3b4aa74 3905 btrfs_release_path(path);
39279cc3 3906
67de1176
MX		 3907 /*
3908 * If we don't have dir index, we have to get it by looking up
3909 * the inode ref, since we get the inode ref, remove it directly,
3910 * it is unnecessary to do delayed deletion.
3911 *
3912 * But if we have dir index, needn't search inode ref to get it.
3913 * Since the inode ref is close to the inode item, it is better
3914 * that we delay to delete it, and just do this deletion when
3915 * we update the inode item.
3916 */
3917 if (BTRFS_I(inode)->dir_index) {
3918 ret = btrfs_delayed_delete_inode_ref(inode);
3919 if (!ret) {
3920 index = BTRFS_I(inode)->dir_index;
3921 goto skip_backref;
3922 }
3923 }
3924
33345d01
LZ		 3925 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3926 dir_ino, &index);
aec7477b 3927 if (ret) {
c2cf52eb
SK		 3928 btrfs_info(root->fs_info,
3929 "failed to delete reference to %.*s, inode %llu parent %llu",
c1c9ff7c 3930 name_len, name, ino, dir_ino);
79787eaa 3931 btrfs_abort_transaction(trans, root, ret);
aec7477b
JB		 3932 goto err;
3933 }
67de1176 3934skip_backref:
16cdcec7 3935 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 3936 if (ret) {
3937 btrfs_abort_transaction(trans, root, ret);
39279cc3 3938 goto err;
79787eaa 3939 }
39279cc3 3940
e02119d5 3941 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
33345d01 3942 inode, dir_ino);
79787eaa
JM		 3943 if (ret != 0 && ret != -ENOENT) {
3944 btrfs_abort_transaction(trans, root, ret);
3945 goto err;
3946 }
e02119d5
CM		 3947
3948 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3949 dir, index);
6418c961
CM		 3950 if (ret == -ENOENT)
3951 ret = 0;
d4e3991b
ZB		 3952 else if (ret)
3953 btrfs_abort_transaction(trans, root, ret);
39279cc3
CM		 3954err:
3955 btrfs_free_path(path);
e02119d5
CM		 3956 if (ret)
3957 goto out;
3958
3959 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95
JB		 3960 inode_inc_iversion(inode);
3961 inode_inc_iversion(dir);
e02119d5 3962 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
b9959295 3963 ret = btrfs_update_inode(trans, root, dir);
e02119d5 3964out:
39279cc3
CM		 3965 return ret;
3966}
3967
92986796
AV		 3968int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3969 struct btrfs_root *root,
3970 struct inode *dir, struct inode *inode,
3971 const char *name, int name_len)
3972{
3973 int ret;
3974 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3975 if (!ret) {
8b558c5f 3976 drop_nlink(inode);
92986796
AV		 3977 ret = btrfs_update_inode(trans, root, inode);
3978 }
3979 return ret;
3980}
39279cc3 3981
a22285a6
YZ		 3982/*
3983 * helper to start transaction for unlink and rmdir.
3984 *
d52be818
JB		 3985 * unlink and rmdir are special in btrfs, they do not always free space, so
3986 * if we cannot make our reservations the normal way try and see if there is
3987 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3988 * allow the unlink to occur.
a22285a6 3989 */
d52be818 3990static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4df27c4d 3991{
39279cc3 3992 struct btrfs_trans_handle *trans;
a22285a6 3993 struct btrfs_root *root = BTRFS_I(dir)->root;
4df27c4d
YZ		 3994 int ret;
3995
e70bea5f
JB		 3996 /*
3997 * 1 for the possible orphan item
3998 * 1 for the dir item
3999 * 1 for the dir index
4000 * 1 for the inode ref
e70bea5f
JB		 4001 * 1 for the inode
4002 */
6e137ed3 4003 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 4004 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
4005 return trans;
4df27c4d 4006
d52be818
JB		 4007 if (PTR_ERR(trans) == -ENOSPC) {
4008 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4df27c4d 4009
d52be818
JB		 4010 trans = btrfs_start_transaction(root, 0);
4011 if (IS_ERR(trans))
4012 return trans;
4013 ret = btrfs_cond_migrate_bytes(root->fs_info,
4014 &root->fs_info->trans_block_rsv,
4015 num_bytes, 5);
4016 if (ret) {
4017 btrfs_end_transaction(trans, root);
4018 return ERR_PTR(ret);
a22285a6 4019 }
5a77d76c 4020 trans->block_rsv = &root->fs_info->trans_block_rsv;
d52be818 4021 trans->bytes_reserved = num_bytes;
a22285a6 4022 }
d52be818 4023 return trans;
a22285a6
YZ		 4024}
4025
4026static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4027{
4028 struct btrfs_root *root = BTRFS_I(dir)->root;
4029 struct btrfs_trans_handle *trans;
2b0143b5 4030 struct inode *inode = d_inode(dentry);
a22285a6 4031 int ret;
a22285a6 4032
d52be818 4033 trans = __unlink_start_trans(dir);
a22285a6
YZ		 4034 if (IS_ERR(trans))
4035 return PTR_ERR(trans);
5f39d397 4036
2b0143b5 4037 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
12fcfd22 4038
2b0143b5 4039 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
e02119d5 4040 dentry->d_name.name, dentry->d_name.len);
b532402e
TI		 4041 if (ret)
4042 goto out;
7b128766 4043
a22285a6 4044 if (inode->i_nlink == 0) {
7b128766 4045 ret = btrfs_orphan_add(trans, inode);
b532402e
TI		 4046 if (ret)
4047 goto out;
a22285a6 4048 }
7b128766 4049
b532402e 4050out:
d52be818 4051 btrfs_end_transaction(trans, root);
b53d3f5d 4052 btrfs_btree_balance_dirty(root);
39279cc3
CM		 4053 return ret;
4054}
4055
4df27c4d
YZ		 4056int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4057 struct btrfs_root *root,
4058 struct inode *dir, u64 objectid,
4059 const char *name, int name_len)
4060{
4061 struct btrfs_path *path;
4062 struct extent_buffer *leaf;
4063 struct btrfs_dir_item *di;
4064 struct btrfs_key key;
4065 u64 index;
4066 int ret;
33345d01 4067 u64 dir_ino = btrfs_ino(dir);
4df27c4d
YZ		 4068
4069 path = btrfs_alloc_path();
4070 if (!path)
4071 return -ENOMEM;
4072
33345d01 4073 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4df27c4d 4074 name, name_len, -1);
79787eaa
JM		 4075 if (IS_ERR_OR_NULL(di)) {
4076 if (!di)
4077 ret = -ENOENT;
4078 else
4079 ret = PTR_ERR(di);
4080 goto out;
4081 }
4df27c4d
YZ		 4082
4083 leaf = path->nodes[0];
4084 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4085 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4086 ret = btrfs_delete_one_dir_name(trans, root, path, di);
79787eaa
JM		 4087 if (ret) {
4088 btrfs_abort_transaction(trans, root, ret);
4089 goto out;
4090 }
b3b4aa74 4091 btrfs_release_path(path);
4df27c4d
YZ		 4092
4093 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4094 objectid, root->root_key.objectid,
33345d01 4095 dir_ino, &index, name, name_len);
4df27c4d 4096 if (ret < 0) {
79787eaa
JM		 4097 if (ret != -ENOENT) {
4098 btrfs_abort_transaction(trans, root, ret);
4099 goto out;
4100 }
33345d01 4101 di = btrfs_search_dir_index_item(root, path, dir_ino,
4df27c4d 4102 name, name_len);
79787eaa
JM		 4103 if (IS_ERR_OR_NULL(di)) {
4104 if (!di)
4105 ret = -ENOENT;
4106 else
4107 ret = PTR_ERR(di);
4108 btrfs_abort_transaction(trans, root, ret);
4109 goto out;
4110 }
4df27c4d
YZ		 4111
4112 leaf = path->nodes[0];
4113 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
b3b4aa74 4114 btrfs_release_path(path);
4df27c4d
YZ		 4115 index = key.offset;
4116 }
945d8962 4117 btrfs_release_path(path);
4df27c4d 4118
16cdcec7 4119 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 4120 if (ret) {
4121 btrfs_abort_transaction(trans, root, ret);
4122 goto out;
4123 }
4df27c4d
YZ		 4124
4125 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95 4126 inode_inc_iversion(dir);
4df27c4d 4127 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
5a24e84c 4128 ret = btrfs_update_inode_fallback(trans, root, dir);
79787eaa
JM		 4129 if (ret)
4130 btrfs_abort_transaction(trans, root, ret);
4131out:
71d7aed0 4132 btrfs_free_path(path);
79787eaa 4133 return ret;
4df27c4d
YZ		 4134}
4135
39279cc3
CM		 4136static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4137{
2b0143b5 4138 struct inode *inode = d_inode(dentry);
1832a6d5 4139 int err = 0;
39279cc3 4140 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 4141 struct btrfs_trans_handle *trans;
39279cc3 4142
b3ae244e 4143 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
134d4512 4144 return -ENOTEMPTY;
b3ae244e
DS		 4145 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4146 return -EPERM;
134d4512 4147
d52be818 4148 trans = __unlink_start_trans(dir);
a22285a6 4149 if (IS_ERR(trans))
5df6a9f6 4150 return PTR_ERR(trans);
5df6a9f6 4151
33345d01 4152 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4df27c4d
YZ		 4153 err = btrfs_unlink_subvol(trans, root, dir,
4154 BTRFS_I(inode)->location.objectid,
4155 dentry->d_name.name,
4156 dentry->d_name.len);
4157 goto out;
4158 }
4159
7b128766
JB		 4160 err = btrfs_orphan_add(trans, inode);
4161 if (err)
4df27c4d 4162 goto out;
7b128766 4163
39279cc3 4164 /* now the directory is empty */
2b0143b5 4165 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
e02119d5 4166 dentry->d_name.name, dentry->d_name.len);
d397712b 4167 if (!err)
dbe674a9 4168 btrfs_i_size_write(inode, 0);
4df27c4d 4169out:
d52be818 4170 btrfs_end_transaction(trans, root);
b53d3f5d 4171 btrfs_btree_balance_dirty(root);
3954401f 4172
39279cc3
CM		 4173 return err;
4174}
4175
28f75a0e
CM		 4176static int truncate_space_check(struct btrfs_trans_handle *trans,
4177 struct btrfs_root *root,
4178 u64 bytes_deleted)
4179{
4180 int ret;
4181
4182 bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4183 ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4184 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4185 if (!ret)
4186 trans->bytes_reserved += bytes_deleted;
4187 return ret;
4188
4189}
4190
39279cc3
CM		 4191/*
4192 * this can truncate away extent items, csum items and directory items.
4193 * It starts at a high offset and removes keys until it can't find
d352ac68 4194 * any higher than new_size
39279cc3
CM		 4195 *
4196 * csum items that cross the new i_size are truncated to the new size
4197 * as well.
7b128766
JB		 4198 *
4199 * min_type is the minimum key type to truncate down to. If set to 0, this
4200 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 4201 */
8082510e
YZ		 4202int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4203 struct btrfs_root *root,
4204 struct inode *inode,
4205 u64 new_size, u32 min_type)
39279cc3 4206{
39279cc3 4207 struct btrfs_path *path;
5f39d397 4208 struct extent_buffer *leaf;
39279cc3 4209 struct btrfs_file_extent_item *fi;
8082510e
YZ		 4210 struct btrfs_key key;
4211 struct btrfs_key found_key;
39279cc3 4212 u64 extent_start = 0;
db94535d 4213 u64 extent_num_bytes = 0;
5d4f98a2 4214 u64 extent_offset = 0;
39279cc3 4215 u64 item_end = 0;
c1aa4575 4216 u64 last_size = new_size;
8082510e 4217 u32 found_type = (u8)-1;
39279cc3
CM		 4218 int found_extent;
4219 int del_item;
85e21bac
CM		 4220 int pending_del_nr = 0;
4221 int pending_del_slot = 0;
179e29e4 4222 int extent_type = -1;
8082510e
YZ		 4223 int ret;
4224 int err = 0;
33345d01 4225 u64 ino = btrfs_ino(inode);
28ed1345 4226 u64 bytes_deleted = 0;
1262133b
JB		 4227 bool be_nice = 0;
4228 bool should_throttle = 0;
28f75a0e 4229 bool should_end = 0;
8082510e
YZ		 4230
4231 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
39279cc3 4232
28ed1345
CM		 4233 /*
4234 * for non-free space inodes and ref cows, we want to back off from
4235 * time to time
4236 */
4237 if (!btrfs_is_free_space_inode(inode) &&
4238 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4239 be_nice = 1;
4240
0eb0e19c
MF		 4241 path = btrfs_alloc_path();
4242 if (!path)
4243 return -ENOMEM;
4244 path->reada = -1;
4245
5dc562c5
JB		 4246 /*
4247 * We want to drop from the next block forward in case this new size is
4248 * not block aligned since we will be keeping the last block of the
4249 * extent just the way it is.
4250 */
27cdeb70
MX		 4251 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4252 root == root->fs_info->tree_root)
fda2832f
QW		 4253 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4254 root->sectorsize), (u64)-1, 0);
8082510e 4255
16cdcec7
MX		 4256 /*
4257 * This function is also used to drop the items in the log tree before
4258 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4259 * it is used to drop the loged items. So we shouldn't kill the delayed
4260 * items.
4261 */
4262 if (min_type == 0 && root == BTRFS_I(inode)->root)
4263 btrfs_kill_delayed_inode_items(inode);
4264
33345d01 4265 key.objectid = ino;
39279cc3 4266 key.offset = (u64)-1;
5f39d397
CM		 4267 key.type = (u8)-1;
4268
85e21bac 4269search_again:
28ed1345
CM		 4270 /*
4271 * with a 16K leaf size and 128MB extents, you can actually queue
4272 * up a huge file in a single leaf. Most of the time that
4273 * bytes_deleted is > 0, it will be huge by the time we get here
4274 */
4275 if (be_nice && bytes_deleted > 32 * 1024 * 1024) {
4276 if (btrfs_should_end_transaction(trans, root)) {
4277 err = -EAGAIN;
4278 goto error;
4279 }
4280 }
4281
4282
b9473439 4283 path->leave_spinning = 1;
85e21bac 4284 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8082510e
YZ		 4285 if (ret < 0) {
4286 err = ret;
4287 goto out;
4288 }
d397712b 4289
85e21bac 4290 if (ret > 0) {
e02119d5
CM		 4291 /* there are no items in the tree for us to truncate, we're
4292 * done
4293 */
8082510e
YZ		 4294 if (path->slots[0] == 0)
4295 goto out;
85e21bac
CM		 4296 path->slots[0]--;
4297 }
4298
d397712b 4299 while (1) {
39279cc3 4300 fi = NULL;
5f39d397
CM		 4301 leaf = path->nodes[0];
4302 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
962a298f 4303 found_type = found_key.type;
39279cc3 4304
33345d01 4305 if (found_key.objectid != ino)
39279cc3 4306 break;
5f39d397 4307
85e21bac 4308 if (found_type < min_type)
39279cc3
CM		 4309 break;
4310
5f39d397 4311 item_end = found_key.offset;
39279cc3 4312 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 4313 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 4314 struct btrfs_file_extent_item);
179e29e4
CM		 4315 extent_type = btrfs_file_extent_type(leaf, fi);
4316 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 4317 item_end +=
db94535d 4318 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 4319 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 4320 item_end += btrfs_file_extent_inline_len(leaf,
514ac8ad 4321 path->slots[0], fi);
39279cc3 4322 }
008630c1 4323 item_end--;
39279cc3 4324 }
8082510e
YZ		 4325 if (found_type > min_type) {
4326 del_item = 1;
4327 } else {
4328 if (item_end < new_size)
b888db2b 4329 break;
8082510e
YZ		 4330 if (found_key.offset >= new_size)
4331 del_item = 1;
4332 else
4333 del_item = 0;
39279cc3 4334 }
39279cc3 4335 found_extent = 0;
39279cc3 4336 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 4337 if (found_type != BTRFS_EXTENT_DATA_KEY)
4338 goto delete;
4339
7f4f6e0a
JB		 4340 if (del_item)
4341 last_size = found_key.offset;
4342 else
4343 last_size = new_size;
4344
179e29e4 4345 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 4346 u64 num_dec;
db94535d 4347 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
f70a9a6b 4348 if (!del_item) {
db94535d
CM		 4349 u64 orig_num_bytes =
4350 btrfs_file_extent_num_bytes(leaf, fi);
fda2832f
QW		 4351 extent_num_bytes = ALIGN(new_size -
4352 found_key.offset,
4353 root->sectorsize);
db94535d
CM		 4354 btrfs_set_file_extent_num_bytes(leaf, fi,
4355 extent_num_bytes);
4356 num_dec = (orig_num_bytes -
9069218d 4357 extent_num_bytes);
27cdeb70
MX		 4358 if (test_bit(BTRFS_ROOT_REF_COWS,
4359 &root->state) &&
4360 extent_start != 0)
a76a3cd4 4361 inode_sub_bytes(inode, num_dec);
5f39d397 4362 btrfs_mark_buffer_dirty(leaf);
39279cc3 4363 } else {
db94535d
CM		 4364 extent_num_bytes =
4365 btrfs_file_extent_disk_num_bytes(leaf,
4366 fi);
5d4f98a2
YZ		 4367 extent_offset = found_key.offset -
4368 btrfs_file_extent_offset(leaf, fi);
4369
39279cc3 4370 /* FIXME blocksize != 4096 */
9069218d 4371 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 4372 if (extent_start != 0) {
4373 found_extent = 1;
27cdeb70
MX		 4374 if (test_bit(BTRFS_ROOT_REF_COWS,
4375 &root->state))
a76a3cd4 4376 inode_sub_bytes(inode, num_dec);
e02119d5 4377 }
39279cc3 4378 }
9069218d 4379 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 4380 /*
4381 * we can't truncate inline items that have had
4382 * special encodings
4383 */
4384 if (!del_item &&
4385 btrfs_file_extent_compression(leaf, fi) == 0 &&
4386 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4387 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM		 4388 u32 size = new_size - found_key.offset;
4389
27cdeb70 4390 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
a76a3cd4
YZ		 4391 inode_sub_bytes(inode, item_end + 1 -
4392 new_size);
514ac8ad
CM		 4393
4394 /*
4395 * update the ram bytes to properly reflect
4396 * the new size of our item
4397 */
4398 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
e02119d5
CM		 4399 size =
4400 btrfs_file_extent_calc_inline_size(size);
afe5fea7 4401 btrfs_truncate_item(root, path, size, 1);
27cdeb70
MX		 4402 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4403 &root->state)) {
a76a3cd4
YZ		 4404 inode_sub_bytes(inode, item_end + 1 -
4405 found_key.offset);
9069218d 4406 }
39279cc3 4407 }
179e29e4 4408delete:
39279cc3 4409 if (del_item) {
85e21bac
CM		 4410 if (!pending_del_nr) {
4411 /* no pending yet, add ourselves */
4412 pending_del_slot = path->slots[0];
4413 pending_del_nr = 1;
4414 } else if (pending_del_nr &&
4415 path->slots[0] + 1 == pending_del_slot) {
4416 /* hop on the pending chunk */
4417 pending_del_nr++;
4418 pending_del_slot = path->slots[0];
4419 } else {
d397712b 4420 BUG();
85e21bac 4421 }
39279cc3
CM		 4422 } else {
4423 break;
4424 }
28f75a0e
CM		 4425 should_throttle = 0;
4426
27cdeb70
MX		 4427 if (found_extent &&
4428 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4429 root == root->fs_info->tree_root)) {
b9473439 4430 btrfs_set_path_blocking(path);
28ed1345 4431 bytes_deleted += extent_num_bytes;
39279cc3 4432 ret = btrfs_free_extent(trans, root, extent_start,
5d4f98a2
YZ		 4433 extent_num_bytes, 0,
4434 btrfs_header_owner(leaf),
66d7e7f0 4435 ino, extent_offset, 0);
39279cc3 4436 BUG_ON(ret);
1262133b 4437 if (btrfs_should_throttle_delayed_refs(trans, root))
28ed1345
CM		 4438 btrfs_async_run_delayed_refs(root,
4439 trans->delayed_ref_updates * 2, 0);
28f75a0e
CM		 4440 if (be_nice) {
4441 if (truncate_space_check(trans, root,
4442 extent_num_bytes)) {
4443 should_end = 1;
4444 }
4445 if (btrfs_should_throttle_delayed_refs(trans,
4446 root)) {
4447 should_throttle = 1;
4448 }
4449 }
39279cc3 4450 }
85e21bac 4451
8082510e
YZ		 4452 if (found_type == BTRFS_INODE_ITEM_KEY)
4453 break;
4454
4455 if (path->slots[0] == 0 ||
1262133b 4456 path->slots[0] != pending_del_slot ||
28f75a0e 4457 should_throttle || should_end) {
8082510e
YZ		 4458 if (pending_del_nr) {
4459 ret = btrfs_del_items(trans, root, path,
4460 pending_del_slot,
4461 pending_del_nr);
79787eaa
JM		 4462 if (ret) {
4463 btrfs_abort_transaction(trans,
4464 root, ret);
4465 goto error;
4466 }
8082510e
YZ		 4467 pending_del_nr = 0;
4468 }
b3b4aa74 4469 btrfs_release_path(path);
28f75a0e 4470 if (should_throttle) {
1262133b
JB		 4471 unsigned long updates = trans->delayed_ref_updates;
4472 if (updates) {
4473 trans->delayed_ref_updates = 0;
4474 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4475 if (ret && !err)
4476 err = ret;
4477 }
4478 }
28f75a0e
CM		 4479 /*
4480 * if we failed to refill our space rsv, bail out
4481 * and let the transaction restart
4482 */
4483 if (should_end) {
4484 err = -EAGAIN;
4485 goto error;
4486 }
85e21bac 4487 goto search_again;
8082510e
YZ		 4488 } else {
4489 path->slots[0]--;
85e21bac 4490 }
39279cc3 4491 }
8082510e 4492out:
85e21bac
CM		 4493 if (pending_del_nr) {
4494 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4495 pending_del_nr);
79787eaa
JM		 4496 if (ret)
4497 btrfs_abort_transaction(trans, root, ret);
85e21bac 4498 }
79787eaa 4499error:
c1aa4575 4500 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
7f4f6e0a 4501 btrfs_ordered_update_i_size(inode, last_size, NULL);
28ed1345 4502
39279cc3 4503 btrfs_free_path(path);
28ed1345 4504
28f75a0e 4505 if (be_nice && bytes_deleted > 32 * 1024 * 1024) {
28ed1345
CM		 4506 unsigned long updates = trans->delayed_ref_updates;
4507 if (updates) {
4508 trans->delayed_ref_updates = 0;
4509 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4510 if (ret && !err)
4511 err = ret;
4512 }
4513 }
8082510e 4514 return err;
39279cc3
CM		 4515}
4516
4517/*
2aaa6655
JB		 4518 * btrfs_truncate_page - read, zero a chunk and write a page
4519 * @inode - inode that we're zeroing
4520 * @from - the offset to start zeroing
4521 * @len - the length to zero, 0 to zero the entire range respective to the
4522 * offset
4523 * @front - zero up to the offset instead of from the offset on
4524 *
4525 * This will find the page for the "from" offset and cow the page and zero the
4526 * part we want to zero. This is used with truncate and hole punching.
39279cc3 4527 */
2aaa6655
JB		 4528int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4529 int front)
39279cc3 4530{
2aaa6655 4531 struct address_space *mapping = inode->i_mapping;
db94535d 4532 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 4533 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4534 struct btrfs_ordered_extent *ordered;
2ac55d41 4535 struct extent_state *cached_state = NULL;
e6dcd2dc 4536 char *kaddr;
db94535d 4537 u32 blocksize = root->sectorsize;
39279cc3
CM		 4538 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4539 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4540 struct page *page;
3b16a4e3 4541 gfp_t mask = btrfs_alloc_write_mask(mapping);
39279cc3 4542 int ret = 0;
a52d9a80 4543 u64 page_start;
e6dcd2dc 4544 u64 page_end;
39279cc3 4545
2aaa6655
JB		 4546 if ((offset & (blocksize - 1)) == 0 &&
4547 (!len || ((len & (blocksize - 1)) == 0)))
39279cc3 4548 goto out;
0ca1f7ce 4549 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
5d5e103a
JB		 4550 if (ret)
4551 goto out;
39279cc3 4552
211c17f5 4553again:
3b16a4e3 4554 page = find_or_create_page(mapping, index, mask);
5d5e103a 4555 if (!page) {
0ca1f7ce 4556 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
ac6a2b36 4557 ret = -ENOMEM;
39279cc3 4558 goto out;
5d5e103a 4559 }
e6dcd2dc
CM		 4560
4561 page_start = page_offset(page);
4562 page_end = page_start + PAGE_CACHE_SIZE - 1;
4563
39279cc3 4564 if (!PageUptodate(page)) {
9ebefb18 4565 ret = btrfs_readpage(NULL, page);
39279cc3 4566 lock_page(page);
211c17f5
CM		 4567 if (page->mapping != mapping) {
4568 unlock_page(page);
4569 page_cache_release(page);
4570 goto again;
4571 }
39279cc3
CM		 4572 if (!PageUptodate(page)) {
4573 ret = -EIO;
89642229 4574 goto out_unlock;
39279cc3
CM		 4575 }
4576 }
211c17f5 4577 wait_on_page_writeback(page);
e6dcd2dc 4578
d0082371 4579 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 4580 set_page_extent_mapped(page);
4581
4582 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4583 if (ordered) {
2ac55d41
JB		 4584 unlock_extent_cached(io_tree, page_start, page_end,
4585 &cached_state, GFP_NOFS);
e6dcd2dc
CM		 4586 unlock_page(page);
4587 page_cache_release(page);
eb84ae03 4588 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 4589 btrfs_put_ordered_extent(ordered);
4590 goto again;
4591 }
4592
2ac55d41 4593 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 4594 EXTENT_DIRTY | EXTENT_DELALLOC |
4595 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 4596 0, 0, &cached_state, GFP_NOFS);
5d5e103a 4597
2ac55d41
JB		 4598 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4599 &cached_state);
9ed74f2d 4600 if (ret) {
2ac55d41
JB		 4601 unlock_extent_cached(io_tree, page_start, page_end,
4602 &cached_state, GFP_NOFS);
9ed74f2d
JB		 4603 goto out_unlock;
4604 }
4605
e6dcd2dc 4606 if (offset != PAGE_CACHE_SIZE) {
2aaa6655
JB		 4607 if (!len)
4608 len = PAGE_CACHE_SIZE - offset;
e6dcd2dc 4609 kaddr = kmap(page);
2aaa6655
JB		 4610 if (front)
4611 memset(kaddr, 0, offset);
4612 else
4613 memset(kaddr + offset, 0, len);
e6dcd2dc
CM		 4614 flush_dcache_page(page);
4615 kunmap(page);
4616 }
247e743c 4617 ClearPageChecked(page);
e6dcd2dc 4618 set_page_dirty(page);
2ac55d41
JB		 4619 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4620 GFP_NOFS);
39279cc3 4621
89642229 4622out_unlock:
5d5e103a 4623 if (ret)
0ca1f7ce 4624 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
39279cc3
CM		 4625 unlock_page(page);
4626 page_cache_release(page);
4627out:
4628 return ret;
4629}
4630
16e7549f
JB		 4631static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4632 u64 offset, u64 len)
4633{
4634 struct btrfs_trans_handle *trans;
4635 int ret;
4636
4637 /*
4638 * Still need to make sure the inode looks like it's been updated so
4639 * that any holes get logged if we fsync.
4640 */
4641 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4642 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4643 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4644 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4645 return 0;
4646 }
4647
4648 /*
4649 * 1 - for the one we're dropping
4650 * 1 - for the one we're adding
4651 * 1 - for updating the inode.
4652 */
4653 trans = btrfs_start_transaction(root, 3);
4654 if (IS_ERR(trans))
4655 return PTR_ERR(trans);
4656
4657 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4658 if (ret) {
4659 btrfs_abort_transaction(trans, root, ret);
4660 btrfs_end_transaction(trans, root);
4661 return ret;
4662 }
4663
4664 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4665 0, 0, len, 0, len, 0, 0, 0);
4666 if (ret)
4667 btrfs_abort_transaction(trans, root, ret);
4668 else
4669 btrfs_update_inode(trans, root, inode);
4670 btrfs_end_transaction(trans, root);
4671 return ret;
4672}
4673
695a0d0d
JB		 4674/*
4675 * This function puts in dummy file extents for the area we're creating a hole
4676 * for. So if we are truncating this file to a larger size we need to insert
4677 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4678 * the range between oldsize and size
4679 */
a41ad394 4680int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
39279cc3 4681{
9036c102
YZ		 4682 struct btrfs_root *root = BTRFS_I(inode)->root;
4683 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a22285a6 4684 struct extent_map *em = NULL;
2ac55d41 4685 struct extent_state *cached_state = NULL;
5dc562c5 4686 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
fda2832f
QW		 4687 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4688 u64 block_end = ALIGN(size, root->sectorsize);
9036c102
YZ		 4689 u64 last_byte;
4690 u64 cur_offset;
4691 u64 hole_size;
9ed74f2d 4692 int err = 0;
39279cc3 4693
a71754fc
JB		 4694 /*
4695 * If our size started in the middle of a page we need to zero out the
4696 * rest of the page before we expand the i_size, otherwise we could
4697 * expose stale data.
4698 */
4699 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4700 if (err)
4701 return err;
4702
9036c102
YZ		 4703 if (size <= hole_start)
4704 return 0;
4705
9036c102
YZ		 4706 while (1) {
4707 struct btrfs_ordered_extent *ordered;
fa7c1494 4708
2ac55d41 4709 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
d0082371 4710 &cached_state);
fa7c1494
MX		 4711 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4712 block_end - hole_start);
9036c102
YZ		 4713 if (!ordered)
4714 break;
2ac55d41
JB		 4715 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4716 &cached_state, GFP_NOFS);
fa7c1494 4717 btrfs_start_ordered_extent(inode, ordered, 1);
9036c102
YZ		 4718 btrfs_put_ordered_extent(ordered);
4719 }
39279cc3 4720
9036c102
YZ		 4721 cur_offset = hole_start;
4722 while (1) {
4723 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4724 block_end - cur_offset, 0);
79787eaa
JM		 4725 if (IS_ERR(em)) {
4726 err = PTR_ERR(em);
f2767956 4727 em = NULL;
79787eaa
JM		 4728 break;
4729 }
9036c102 4730 last_byte = min(extent_map_end(em), block_end);
fda2832f 4731 last_byte = ALIGN(last_byte , root->sectorsize);
8082510e 4732 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5dc562c5 4733 struct extent_map *hole_em;
9036c102 4734 hole_size = last_byte - cur_offset;
9ed74f2d 4735
16e7549f
JB		 4736 err = maybe_insert_hole(root, inode, cur_offset,
4737 hole_size);
4738 if (err)
3893e33b 4739 break;
5dc562c5
JB		 4740 btrfs_drop_extent_cache(inode, cur_offset,
4741 cur_offset + hole_size - 1, 0);
4742 hole_em = alloc_extent_map();
4743 if (!hole_em) {
4744 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4745 &BTRFS_I(inode)->runtime_flags);
4746 goto next;
4747 }
4748 hole_em->start = cur_offset;
4749 hole_em->len = hole_size;
4750 hole_em->orig_start = cur_offset;
8082510e 4751
5dc562c5
JB		 4752 hole_em->block_start = EXTENT_MAP_HOLE;
4753 hole_em->block_len = 0;
b4939680 4754 hole_em->orig_block_len = 0;
cc95bef6 4755 hole_em->ram_bytes = hole_size;
5dc562c5
JB		 4756 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4757 hole_em->compress_type = BTRFS_COMPRESS_NONE;
16e7549f 4758 hole_em->generation = root->fs_info->generation;
8082510e 4759
5dc562c5
JB		 4760 while (1) {
4761 write_lock(&em_tree->lock);
09a2a8f9 4762 err = add_extent_mapping(em_tree, hole_em, 1);
5dc562c5
JB		 4763 write_unlock(&em_tree->lock);
4764 if (err != -EEXIST)
4765 break;
4766 btrfs_drop_extent_cache(inode, cur_offset,
4767 cur_offset +
4768 hole_size - 1, 0);
4769 }
4770 free_extent_map(hole_em);
9036c102 4771 }
16e7549f 4772next:
9036c102 4773 free_extent_map(em);
a22285a6 4774 em = NULL;
9036c102 4775 cur_offset = last_byte;
8082510e 4776 if (cur_offset >= block_end)
9036c102
YZ		 4777 break;
4778 }
a22285a6 4779 free_extent_map(em);
2ac55d41
JB		 4780 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4781 GFP_NOFS);
9036c102
YZ		 4782 return err;
4783}
39279cc3 4784
9ea24bbe
FM		 4785static int wait_snapshoting_atomic_t(atomic_t *a)
4786{
4787 schedule();
4788 return 0;
4789}
4790
4791static void wait_for_snapshot_creation(struct btrfs_root *root)
4792{
4793 while (true) {
4794 int ret;
4795
4796 ret = btrfs_start_write_no_snapshoting(root);
4797 if (ret)
4798 break;
4799 wait_on_atomic_t(&root->will_be_snapshoted,
4800 wait_snapshoting_atomic_t,
4801 TASK_UNINTERRUPTIBLE);
4802 }
4803}
4804
3972f260 4805static int btrfs_setsize(struct inode *inode, struct iattr *attr)
8082510e 4806{
f4a2f4c5
MX		 4807 struct btrfs_root *root = BTRFS_I(inode)->root;
4808 struct btrfs_trans_handle *trans;
a41ad394 4809 loff_t oldsize = i_size_read(inode);
3972f260
ES		 4810 loff_t newsize = attr->ia_size;
4811 int mask = attr->ia_valid;
8082510e
YZ		 4812 int ret;
4813
3972f260
ES		 4814 /*
4815 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4816 * special case where we need to update the times despite not having
4817 * these flags set. For all other operations the VFS set these flags
4818 * explicitly if it wants a timestamp update.
4819 */
dff6efc3
CH		 4820 if (newsize != oldsize) {
4821 inode_inc_iversion(inode);
4822 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4823 inode->i_ctime = inode->i_mtime =
4824 current_fs_time(inode->i_sb);
4825 }
3972f260 4826
a41ad394 4827 if (newsize > oldsize) {
7caef267 4828 truncate_pagecache(inode, newsize);
9ea24bbe
FM		 4829 /*
4830 * Don't do an expanding truncate while snapshoting is ongoing.
4831 * This is to ensure the snapshot captures a fully consistent
4832 * state of this file - if the snapshot captures this expanding
4833 * truncation, it must capture all writes that happened before
4834 * this truncation.
4835 */
4836 wait_for_snapshot_creation(root);
a41ad394 4837 ret = btrfs_cont_expand(inode, oldsize, newsize);
9ea24bbe
FM		 4838 if (ret) {
4839 btrfs_end_write_no_snapshoting(root);
8082510e 4840 return ret;
9ea24bbe 4841 }
8082510e 4842
f4a2f4c5 4843 trans = btrfs_start_transaction(root, 1);
9ea24bbe
FM		 4844 if (IS_ERR(trans)) {
4845 btrfs_end_write_no_snapshoting(root);
f4a2f4c5 4846 return PTR_ERR(trans);
9ea24bbe 4847 }
f4a2f4c5
MX		 4848
4849 i_size_write(inode, newsize);
4850 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4851 ret = btrfs_update_inode(trans, root, inode);
9ea24bbe 4852 btrfs_end_write_no_snapshoting(root);
7ad85bb7 4853 btrfs_end_transaction(trans, root);
a41ad394 4854 } else {
8082510e 4855
a41ad394
JB		 4856 /*
4857 * We're truncating a file that used to have good data down to
4858 * zero. Make sure it gets into the ordered flush list so that
4859 * any new writes get down to disk quickly.
4860 */
4861 if (newsize == 0)
72ac3c0d
JB		 4862 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4863 &BTRFS_I(inode)->runtime_flags);
8082510e 4864
f3fe820c
JB		 4865 /*
4866 * 1 for the orphan item we're going to add
4867 * 1 for the orphan item deletion.
4868 */
4869 trans = btrfs_start_transaction(root, 2);
4870 if (IS_ERR(trans))
4871 return PTR_ERR(trans);
4872
4873 /*
4874 * We need to do this in case we fail at _any_ point during the
4875 * actual truncate. Once we do the truncate_setsize we could
4876 * invalidate pages which forces any outstanding ordered io to
4877 * be instantly completed which will give us extents that need
4878 * to be truncated. If we fail to get an orphan inode down we
4879 * could have left over extents that were never meant to live,
4880 * so we need to garuntee from this point on that everything
4881 * will be consistent.
4882 */
4883 ret = btrfs_orphan_add(trans, inode);
4884 btrfs_end_transaction(trans, root);
4885 if (ret)
4886 return ret;
4887
a41ad394
JB		 4888 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4889 truncate_setsize(inode, newsize);
2e60a51e
MX		 4890
4891 /* Disable nonlocked read DIO to avoid the end less truncate */
4892 btrfs_inode_block_unlocked_dio(inode);
4893 inode_dio_wait(inode);
4894 btrfs_inode_resume_unlocked_dio(inode);
4895
a41ad394 4896 ret = btrfs_truncate(inode);
7f4f6e0a
JB		 4897 if (ret && inode->i_nlink) {
4898 int err;
4899
4900 /*
4901 * failed to truncate, disk_i_size is only adjusted down
4902 * as we remove extents, so it should represent the true
4903 * size of the inode, so reset the in memory size and
4904 * delete our orphan entry.
4905 */
4906 trans = btrfs_join_transaction(root);
4907 if (IS_ERR(trans)) {
4908 btrfs_orphan_del(NULL, inode);
4909 return ret;
4910 }
4911 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4912 err = btrfs_orphan_del(trans, inode);
4913 if (err)
4914 btrfs_abort_transaction(trans, root, err);
4915 btrfs_end_transaction(trans, root);
4916 }
8082510e
YZ		 4917 }
4918
a41ad394 4919 return ret;
8082510e
YZ		 4920}
4921
9036c102
YZ		 4922static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4923{
2b0143b5 4924 struct inode *inode = d_inode(dentry);
b83cc969 4925 struct btrfs_root *root = BTRFS_I(inode)->root;
9036c102 4926 int err;
39279cc3 4927
b83cc969
LZ		 4928 if (btrfs_root_readonly(root))
4929 return -EROFS;
4930
9036c102
YZ		 4931 err = inode_change_ok(inode, attr);
4932 if (err)
4933 return err;
2bf5a725 4934
5a3f23d5 4935 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3972f260 4936 err = btrfs_setsize(inode, attr);
8082510e
YZ		 4937 if (err)
4938 return err;
39279cc3 4939 }
9036c102 4940
1025774c
CH		 4941 if (attr->ia_valid) {
4942 setattr_copy(inode, attr);
0c4d2d95 4943 inode_inc_iversion(inode);
22c44fe6 4944 err = btrfs_dirty_inode(inode);
1025774c 4945
22c44fe6 4946 if (!err && attr->ia_valid & ATTR_MODE)
996a710d 4947 err = posix_acl_chmod(inode, inode->i_mode);
1025774c 4948 }
33268eaf 4949
39279cc3
CM		 4950 return err;
4951}
61295eb8 4952
131e404a
FDBM		 4953/*
4954 * While truncating the inode pages during eviction, we get the VFS calling
4955 * btrfs_invalidatepage() against each page of the inode. This is slow because
4956 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4957 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4958 * extent_state structures over and over, wasting lots of time.
4959 *
4960 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4961 * those expensive operations on a per page basis and do only the ordered io
4962 * finishing, while we release here the extent_map and extent_state structures,
4963 * without the excessive merging and splitting.
4964 */
4965static void evict_inode_truncate_pages(struct inode *inode)
4966{
4967 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4968 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4969 struct rb_node *node;
4970
4971 ASSERT(inode->i_state & I_FREEING);
91b0abe3 4972 truncate_inode_pages_final(&inode->i_data);
131e404a
FDBM		 4973
4974 write_lock(&map_tree->lock);
4975 while (!RB_EMPTY_ROOT(&map_tree->map)) {
4976 struct extent_map *em;
4977
4978 node = rb_first(&map_tree->map);
4979 em = rb_entry(node, struct extent_map, rb_node);
180589ef
WS		 4980 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4981 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
131e404a
FDBM		 4982 remove_extent_mapping(map_tree, em);
4983 free_extent_map(em);
7064dd5c
FM		 4984 if (need_resched()) {
4985 write_unlock(&map_tree->lock);
4986 cond_resched();
4987 write_lock(&map_tree->lock);
4988 }
131e404a
FDBM		 4989 }
4990 write_unlock(&map_tree->lock);
4991
6ca07097
FM		 4992 /*
4993 * Keep looping until we have no more ranges in the io tree.
4994 * We can have ongoing bios started by readpages (called from readahead)
9c6429d9
FM		 4995 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
4996 * still in progress (unlocked the pages in the bio but did not yet
4997 * unlocked the ranges in the io tree). Therefore this means some
6ca07097
FM		 4998 * ranges can still be locked and eviction started because before
4999 * submitting those bios, which are executed by a separate task (work
5000 * queue kthread), inode references (inode->i_count) were not taken
5001 * (which would be dropped in the end io callback of each bio).
5002 * Therefore here we effectively end up waiting for those bios and
5003 * anyone else holding locked ranges without having bumped the inode's
5004 * reference count - if we don't do it, when they access the inode's
5005 * io_tree to unlock a range it may be too late, leading to an
5006 * use-after-free issue.
5007 */
131e404a
FDBM		 5008 spin_lock(&io_tree->lock);
5009 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5010 struct extent_state *state;
5011 struct extent_state *cached_state = NULL;
6ca07097
FM		 5012 u64 start;
5013 u64 end;
131e404a
FDBM		 5014
5015 node = rb_first(&io_tree->state);
5016 state = rb_entry(node, struct extent_state, rb_node);
6ca07097
FM		 5017 start = state->start;
5018 end = state->end;
131e404a
FDBM		 5019 spin_unlock(&io_tree->lock);
5020
6ca07097
FM		 5021 lock_extent_bits(io_tree, start, end, 0, &cached_state);
5022 clear_extent_bit(io_tree, start, end,
131e404a
FDBM		 5023 EXTENT_LOCKED | EXTENT_DIRTY |
5024 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5025 EXTENT_DEFRAG, 1, 1,
5026 &cached_state, GFP_NOFS);
131e404a 5027
7064dd5c 5028 cond_resched();
131e404a
FDBM		 5029 spin_lock(&io_tree->lock);
5030 }
5031 spin_unlock(&io_tree->lock);
5032}
5033
bd555975 5034void btrfs_evict_inode(struct inode *inode)
39279cc3
CM		 5035{
5036 struct btrfs_trans_handle *trans;
5037 struct btrfs_root *root = BTRFS_I(inode)->root;
726c35fa 5038 struct btrfs_block_rsv *rsv, *global_rsv;
3bce876f 5039 int steal_from_global = 0;
07127184 5040 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3
CM		 5041 int ret;
5042
1abe9b8a 5043 trace_btrfs_inode_evict(inode);
5044
131e404a
FDBM		 5045 evict_inode_truncate_pages(inode);
5046
69e9c6c6
SB		 5047 if (inode->i_nlink &&
5048 ((btrfs_root_refs(&root->root_item) != 0 &&
5049 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5050 btrfs_is_free_space_inode(inode)))
bd555975
AV		 5051 goto no_delete;
5052
39279cc3 5053 if (is_bad_inode(inode)) {
7b128766 5054 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 5055 goto no_delete;
5056 }
bd555975 5057 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4a096752 5058 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 5059
f612496b
MX		 5060 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5061
c71bf099 5062 if (root->fs_info->log_root_recovering) {
6bf02314 5063 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8a35d95f 5064 &BTRFS_I(inode)->runtime_flags));
c71bf099
YZ		 5065 goto no_delete;
5066 }
5067
76dda93c 5068 if (inode->i_nlink > 0) {
69e9c6c6
SB		 5069 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5070 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
76dda93c
YZ		 5071 goto no_delete;
5072 }
5073
0e8c36a9
MX		 5074 ret = btrfs_commit_inode_delayed_inode(inode);
5075 if (ret) {
5076 btrfs_orphan_del(NULL, inode);
5077 goto no_delete;
5078 }
5079
66d8f3dd 5080 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4289a667
JB		 5081 if (!rsv) {
5082 btrfs_orphan_del(NULL, inode);
5083 goto no_delete;
5084 }
4a338542 5085 rsv->size = min_size;
ca7e70f5 5086 rsv->failfast = 1;
726c35fa 5087 global_rsv = &root->fs_info->global_block_rsv;
4289a667 5088
dbe674a9 5089 btrfs_i_size_write(inode, 0);
5f39d397 5090
4289a667 5091 /*
8407aa46
MX		 5092 * This is a bit simpler than btrfs_truncate since we've already
5093 * reserved our space for our orphan item in the unlink, so we just
5094 * need to reserve some slack space in case we add bytes and update
5095 * inode item when doing the truncate.
4289a667 5096 */
8082510e 5097 while (1) {
08e007d2
MX		 5098 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5099 BTRFS_RESERVE_FLUSH_LIMIT);
726c35fa
JB		 5100
5101 /*
5102 * Try and steal from the global reserve since we will
5103 * likely not use this space anyway, we want to try as
5104 * hard as possible to get this to work.
5105 */
5106 if (ret)
3bce876f
JB		 5107 steal_from_global++;
5108 else
5109 steal_from_global = 0;
5110 ret = 0;
d68fc57b 5111
3bce876f
JB		 5112 /*
5113 * steal_from_global == 0: we reserved stuff, hooray!
5114 * steal_from_global == 1: we didn't reserve stuff, boo!
5115 * steal_from_global == 2: we've committed, still not a lot of
5116 * room but maybe we'll have room in the global reserve this
5117 * time.
5118 * steal_from_global == 3: abandon all hope!
5119 */
5120 if (steal_from_global > 2) {
c2cf52eb
SK		 5121 btrfs_warn(root->fs_info,
5122 "Could not get space for a delete, will truncate on mount %d",
5123 ret);
4289a667
JB		 5124 btrfs_orphan_del(NULL, inode);
5125 btrfs_free_block_rsv(root, rsv);
5126 goto no_delete;
d68fc57b 5127 }
7b128766 5128
0e8c36a9 5129 trans = btrfs_join_transaction(root);
4289a667
JB		 5130 if (IS_ERR(trans)) {
5131 btrfs_orphan_del(NULL, inode);
5132 btrfs_free_block_rsv(root, rsv);
5133 goto no_delete;
d68fc57b 5134 }
7b128766 5135
3bce876f
JB		 5136 /*
5137 * We can't just steal from the global reserve, we need tomake
5138 * sure there is room to do it, if not we need to commit and try
5139 * again.
5140 */
5141 if (steal_from_global) {
5142 if (!btrfs_check_space_for_delayed_refs(trans, root))
5143 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5144 min_size);
5145 else
5146 ret = -ENOSPC;
5147 }
5148
5149 /*
5150 * Couldn't steal from the global reserve, we have too much
5151 * pending stuff built up, commit the transaction and try it
5152 * again.
5153 */
5154 if (ret) {
5155 ret = btrfs_commit_transaction(trans, root);
5156 if (ret) {
5157 btrfs_orphan_del(NULL, inode);
5158 btrfs_free_block_rsv(root, rsv);
5159 goto no_delete;
5160 }
5161 continue;
5162 } else {
5163 steal_from_global = 0;
5164 }
5165
4289a667
JB		 5166 trans->block_rsv = rsv;
5167
d68fc57b 5168 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
28ed1345 5169 if (ret != -ENOSPC && ret != -EAGAIN)
8082510e 5170 break;
85e21bac 5171
8407aa46 5172 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e
YZ		 5173 btrfs_end_transaction(trans, root);
5174 trans = NULL;
b53d3f5d 5175 btrfs_btree_balance_dirty(root);
8082510e 5176 }
5f39d397 5177
4289a667
JB		 5178 btrfs_free_block_rsv(root, rsv);
5179
4ef31a45
JB		 5180 /*
5181 * Errors here aren't a big deal, it just means we leave orphan items
5182 * in the tree. They will be cleaned up on the next mount.
5183 */
8082510e 5184 if (ret == 0) {
4289a667 5185 trans->block_rsv = root->orphan_block_rsv;
4ef31a45
JB		 5186 btrfs_orphan_del(trans, inode);
5187 } else {
5188 btrfs_orphan_del(NULL, inode);
8082510e 5189 }
54aa1f4d 5190
4289a667 5191 trans->block_rsv = &root->fs_info->trans_block_rsv;
581bb050
LZ		 5192 if (!(root == root->fs_info->tree_root ||
5193 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
33345d01 5194 btrfs_return_ino(root, btrfs_ino(inode));
581bb050 5195
54aa1f4d 5196 btrfs_end_transaction(trans, root);
b53d3f5d 5197 btrfs_btree_balance_dirty(root);
39279cc3 5198no_delete:
89042e5a 5199 btrfs_remove_delayed_node(inode);
dbd5768f 5200 clear_inode(inode);
8082510e 5201 return;
39279cc3
CM		 5202}
5203
5204/*
5205 * this returns the key found in the dir entry in the location pointer.
5206 * If no dir entries were found, location->objectid is 0.
5207 */
5208static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5209 struct btrfs_key *location)
5210{
5211 const char *name = dentry->d_name.name;
5212 int namelen = dentry->d_name.len;
5213 struct btrfs_dir_item *di;
5214 struct btrfs_path *path;
5215 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 5216 int ret = 0;
39279cc3
CM		 5217
5218 path = btrfs_alloc_path();
d8926bb3
MF		 5219 if (!path)
5220 return -ENOMEM;
3954401f 5221
33345d01 5222 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
39279cc3 5223 namelen, 0);
0d9f7f3e
Y		 5224 if (IS_ERR(di))
5225 ret = PTR_ERR(di);
d397712b 5226
c704005d 5227 if (IS_ERR_OR_NULL(di))
3954401f 5228 goto out_err;
d397712b 5229
5f39d397 5230 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 5231out:
39279cc3
CM		 5232 btrfs_free_path(path);
5233 return ret;
3954401f
CM		 5234out_err:
5235 location->objectid = 0;
5236 goto out;
39279cc3
CM		 5237}
5238
5239/*
5240 * when we hit a tree root in a directory, the btrfs part of the inode
5241 * needs to be changed to reflect the root directory of the tree root. This
5242 * is kind of like crossing a mount point.
5243 */
5244static int fixup_tree_root_location(struct btrfs_root *root,
4df27c4d
YZ		 5245 struct inode *dir,
5246 struct dentry *dentry,
5247 struct btrfs_key *location,
5248 struct btrfs_root **sub_root)
39279cc3 5249{
4df27c4d
YZ		 5250 struct btrfs_path *path;
5251 struct btrfs_root *new_root;
5252 struct btrfs_root_ref *ref;
5253 struct extent_buffer *leaf;
1d4c08e0 5254 struct btrfs_key key;
4df27c4d
YZ		 5255 int ret;
5256 int err = 0;
39279cc3 5257
4df27c4d
YZ		 5258 path = btrfs_alloc_path();
5259 if (!path) {
5260 err = -ENOMEM;
5261 goto out;
5262 }
39279cc3 5263
4df27c4d 5264 err = -ENOENT;
1d4c08e0
DS		 5265 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5266 key.type = BTRFS_ROOT_REF_KEY;
5267 key.offset = location->objectid;
5268
5269 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5270 0, 0);
4df27c4d
YZ		 5271 if (ret) {
5272 if (ret < 0)
5273 err = ret;
5274 goto out;
5275 }
39279cc3 5276
4df27c4d
YZ		 5277 leaf = path->nodes[0];
5278 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
33345d01 5279 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4df27c4d
YZ		 5280 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5281 goto out;
39279cc3 5282
4df27c4d
YZ		 5283 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5284 (unsigned long)(ref + 1),
5285 dentry->d_name.len);
5286 if (ret)
5287 goto out;
5288
b3b4aa74 5289 btrfs_release_path(path);
4df27c4d
YZ		 5290
5291 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5292 if (IS_ERR(new_root)) {
5293 err = PTR_ERR(new_root);
5294 goto out;
5295 }
5296
4df27c4d
YZ		 5297 *sub_root = new_root;
5298 location->objectid = btrfs_root_dirid(&new_root->root_item);
5299 location->type = BTRFS_INODE_ITEM_KEY;
5300 location->offset = 0;
5301 err = 0;
5302out:
5303 btrfs_free_path(path);
5304 return err;
39279cc3
CM		 5305}
5306
5d4f98a2
YZ		 5307static void inode_tree_add(struct inode *inode)
5308{
5309 struct btrfs_root *root = BTRFS_I(inode)->root;
5310 struct btrfs_inode *entry;
03e860bd
FNP		 5311 struct rb_node **p;
5312 struct rb_node *parent;
cef21937 5313 struct rb_node *new = &BTRFS_I(inode)->rb_node;
33345d01 5314 u64 ino = btrfs_ino(inode);
5d4f98a2 5315
1d3382cb 5316 if (inode_unhashed(inode))
76dda93c 5317 return;
e1409cef 5318 parent = NULL;
5d4f98a2 5319 spin_lock(&root->inode_lock);
e1409cef 5320 p = &root->inode_tree.rb_node;
5d4f98a2
YZ		 5321 while (*p) {
5322 parent = *p;
5323 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5324
33345d01 5325 if (ino < btrfs_ino(&entry->vfs_inode))
03e860bd 5326 p = &parent->rb_left;
33345d01 5327 else if (ino > btrfs_ino(&entry->vfs_inode))
03e860bd 5328 p = &parent->rb_right;
5d4f98a2
YZ		 5329 else {
5330 WARN_ON(!(entry->vfs_inode.i_state &
a4ffdde6 5331 (I_WILL_FREE | I_FREEING)));
cef21937 5332 rb_replace_node(parent, new, &root->inode_tree);
03e860bd
FNP		 5333 RB_CLEAR_NODE(parent);
5334 spin_unlock(&root->inode_lock);
cef21937 5335 return;
5d4f98a2
YZ		 5336 }
5337 }
cef21937
FDBM		 5338 rb_link_node(new, parent, p);
5339 rb_insert_color(new, &root->inode_tree);
5d4f98a2
YZ		 5340 spin_unlock(&root->inode_lock);
5341}
5342
5343static void inode_tree_del(struct inode *inode)
5344{
5345 struct btrfs_root *root = BTRFS_I(inode)->root;
76dda93c 5346 int empty = 0;
5d4f98a2 5347
03e860bd 5348 spin_lock(&root->inode_lock);
5d4f98a2 5349 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5d4f98a2 5350 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5d4f98a2 5351 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
76dda93c 5352 empty = RB_EMPTY_ROOT(&root->inode_tree);
5d4f98a2 5353 }
03e860bd 5354 spin_unlock(&root->inode_lock);
76dda93c 5355
69e9c6c6 5356 if (empty && btrfs_root_refs(&root->root_item) == 0) {
76dda93c
YZ		 5357 synchronize_srcu(&root->fs_info->subvol_srcu);
5358 spin_lock(&root->inode_lock);
5359 empty = RB_EMPTY_ROOT(&root->inode_tree);
5360 spin_unlock(&root->inode_lock);
5361 if (empty)
5362 btrfs_add_dead_root(root);
5363 }
5364}
5365
143bede5 5366void btrfs_invalidate_inodes(struct btrfs_root *root)
76dda93c
YZ		 5367{
5368 struct rb_node *node;
5369 struct rb_node *prev;
5370 struct btrfs_inode *entry;
5371 struct inode *inode;
5372 u64 objectid = 0;
5373
7813b3db
LB		 5374 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5375 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
76dda93c
YZ		 5376
5377 spin_lock(&root->inode_lock);
5378again:
5379 node = root->inode_tree.rb_node;
5380 prev = NULL;
5381 while (node) {
5382 prev = node;
5383 entry = rb_entry(node, struct btrfs_inode, rb_node);
5384
33345d01 5385 if (objectid < btrfs_ino(&entry->vfs_inode))
76dda93c 5386 node = node->rb_left;
33345d01 5387 else if (objectid > btrfs_ino(&entry->vfs_inode))
76dda93c
YZ		 5388 node = node->rb_right;
5389 else
5390 break;
5391 }
5392 if (!node) {
5393 while (prev) {
5394 entry = rb_entry(prev, struct btrfs_inode, rb_node);
33345d01 5395 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
76dda93c
YZ		 5396 node = prev;
5397 break;
5398 }
5399 prev = rb_next(prev);
5400 }
5401 }
5402 while (node) {
5403 entry = rb_entry(node, struct btrfs_inode, rb_node);
33345d01 5404 objectid = btrfs_ino(&entry->vfs_inode) + 1;
76dda93c
YZ		 5405 inode = igrab(&entry->vfs_inode);
5406 if (inode) {
5407 spin_unlock(&root->inode_lock);
5408 if (atomic_read(&inode->i_count) > 1)
5409 d_prune_aliases(inode);
5410 /*
45321ac5 5411 * btrfs_drop_inode will have it removed from
76dda93c
YZ		 5412 * the inode cache when its usage count
5413 * hits zero.
5414 */
5415 iput(inode);
5416 cond_resched();
5417 spin_lock(&root->inode_lock);
5418 goto again;
5419 }
5420
5421 if (cond_resched_lock(&root->inode_lock))
5422 goto again;
5423
5424 node = rb_next(node);
5425 }
5426 spin_unlock(&root->inode_lock);
5d4f98a2
YZ		 5427}
5428
e02119d5
CM		 5429static int btrfs_init_locked_inode(struct inode *inode, void *p)
5430{
5431 struct btrfs_iget_args *args = p;
90d3e592
CM		 5432 inode->i_ino = args->location->objectid;
5433 memcpy(&BTRFS_I(inode)->location, args->location,
5434 sizeof(*args->location));
e02119d5 5435 BTRFS_I(inode)->root = args->root;
39279cc3
CM		 5436 return 0;
5437}
5438
5439static int btrfs_find_actor(struct inode *inode, void *opaque)
5440{
5441 struct btrfs_iget_args *args = opaque;
90d3e592 5442 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
d397712b 5443 args->root == BTRFS_I(inode)->root;
39279cc3
CM		 5444}
5445
5d4f98a2 5446static struct inode *btrfs_iget_locked(struct super_block *s,
90d3e592 5447 struct btrfs_key *location,
5d4f98a2 5448 struct btrfs_root *root)
39279cc3
CM		 5449{
5450 struct inode *inode;
5451 struct btrfs_iget_args args;
90d3e592 5452 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
778ba82b 5453
90d3e592 5454 args.location = location;
39279cc3
CM		 5455 args.root = root;
5456
778ba82b 5457 inode = iget5_locked(s, hashval, btrfs_find_actor,
39279cc3
CM		 5458 btrfs_init_locked_inode,
5459 (void *)&args);
5460 return inode;
5461}
5462
1a54ef8c
BR		 5463/* Get an inode object given its location and corresponding root.
5464 * Returns in *is_new if the inode was read from disk
5465 */
5466struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
73f73415 5467 struct btrfs_root *root, int *new)
1a54ef8c
BR		 5468{
5469 struct inode *inode;
5470
90d3e592 5471 inode = btrfs_iget_locked(s, location, root);
1a54ef8c 5472 if (!inode)
5d4f98a2 5473 return ERR_PTR(-ENOMEM);
1a54ef8c
BR		 5474
5475 if (inode->i_state & I_NEW) {
1a54ef8c 5476 btrfs_read_locked_inode(inode);
1748f843
MF		 5477 if (!is_bad_inode(inode)) {
5478 inode_tree_add(inode);
5479 unlock_new_inode(inode);
5480 if (new)
5481 *new = 1;
5482 } else {
e0b6d65b
ST		 5483 unlock_new_inode(inode);
5484 iput(inode);
5485 inode = ERR_PTR(-ESTALE);
1748f843
MF		 5486 }
5487 }
5488
1a54ef8c
BR		 5489 return inode;
5490}
5491
4df27c4d
YZ		 5492static struct inode *new_simple_dir(struct super_block *s,
5493 struct btrfs_key *key,
5494 struct btrfs_root *root)
5495{
5496 struct inode *inode = new_inode(s);
5497
5498 if (!inode)
5499 return ERR_PTR(-ENOMEM);
5500
4df27c4d
YZ		 5501 BTRFS_I(inode)->root = root;
5502 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
72ac3c0d 5503 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4df27c4d
YZ		 5504
5505 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
848cce0d 5506 inode->i_op = &btrfs_dir_ro_inode_operations;
4df27c4d
YZ		 5507 inode->i_fop = &simple_dir_operations;
5508 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
9cc97d64 5509 inode->i_mtime = CURRENT_TIME;
5510 inode->i_atime = inode->i_mtime;
5511 inode->i_ctime = inode->i_mtime;
5512 BTRFS_I(inode)->i_otime = inode->i_mtime;
4df27c4d
YZ		 5513
5514 return inode;
5515}
5516
3de4586c 5517struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3 5518{
d397712b 5519 struct inode *inode;
4df27c4d 5520 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3
CM		 5521 struct btrfs_root *sub_root = root;
5522 struct btrfs_key location;
76dda93c 5523 int index;
b4aff1f8 5524 int ret = 0;
39279cc3
CM		 5525
5526 if (dentry->d_name.len > BTRFS_NAME_LEN)
5527 return ERR_PTR(-ENAMETOOLONG);
5f39d397 5528
39e3c955 5529 ret = btrfs_inode_by_name(dir, dentry, &location);
39279cc3
CM		 5530 if (ret < 0)
5531 return ERR_PTR(ret);
5f39d397 5532
4df27c4d 5533 if (location.objectid == 0)
5662344b 5534 return ERR_PTR(-ENOENT);
4df27c4d
YZ		 5535
5536 if (location.type == BTRFS_INODE_ITEM_KEY) {
73f73415 5537 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4df27c4d
YZ		 5538 return inode;
5539 }
5540
5541 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5542
76dda93c 5543 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4df27c4d
YZ		 5544 ret = fixup_tree_root_location(root, dir, dentry,
5545 &location, &sub_root);
5546 if (ret < 0) {
5547 if (ret != -ENOENT)
5548 inode = ERR_PTR(ret);
5549 else
5550 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5551 } else {
73f73415 5552 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
39279cc3 5553 }
76dda93c
YZ		 5554 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5555
34d19bad 5556 if (!IS_ERR(inode) && root != sub_root) {
c71bf099
YZ		 5557 down_read(&root->fs_info->cleanup_work_sem);
5558 if (!(inode->i_sb->s_flags & MS_RDONLY))
66b4ffd1 5559 ret = btrfs_orphan_cleanup(sub_root);
c71bf099 5560 up_read(&root->fs_info->cleanup_work_sem);
01cd3367
JB		 5561 if (ret) {
5562 iput(inode);
66b4ffd1 5563 inode = ERR_PTR(ret);
01cd3367 5564 }
c71bf099
YZ		 5565 }
5566
3de4586c
CM		 5567 return inode;
5568}
5569
fe15ce44 5570static int btrfs_dentry_delete(const struct dentry *dentry)
76dda93c
YZ		 5571{
5572 struct btrfs_root *root;
2b0143b5 5573 struct inode *inode = d_inode(dentry);
76dda93c 5574
848cce0d 5575 if (!inode && !IS_ROOT(dentry))
2b0143b5 5576 inode = d_inode(dentry->d_parent);
76dda93c 5577
848cce0d
LZ		 5578 if (inode) {
5579 root = BTRFS_I(inode)->root;
efefb143
YZ		 5580 if (btrfs_root_refs(&root->root_item) == 0)
5581 return 1;
848cce0d
LZ		 5582
5583 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5584 return 1;
efefb143 5585 }
76dda93c
YZ		 5586 return 0;
5587}
5588
b4aff1f8
JB		 5589static void btrfs_dentry_release(struct dentry *dentry)
5590{
944a4515 5591 kfree(dentry->d_fsdata);
b4aff1f8
JB		 5592}
5593
3de4586c 5594static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
00cd8dd3 5595 unsigned int flags)
3de4586c 5596{
5662344b 5597 struct inode *inode;
a66e7cc6 5598
5662344b
TI		 5599 inode = btrfs_lookup_dentry(dir, dentry);
5600 if (IS_ERR(inode)) {
5601 if (PTR_ERR(inode) == -ENOENT)
5602 inode = NULL;
5603 else
5604 return ERR_CAST(inode);
5605 }
5606
41d28bca 5607 return d_splice_alias(inode, dentry);
39279cc3
CM		 5608}
5609
16cdcec7 5610unsigned char btrfs_filetype_table[] = {
39279cc3
CM		 5611 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5612};
5613
9cdda8d3 5614static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
39279cc3 5615{
9cdda8d3 5616 struct inode *inode = file_inode(file);
39279cc3
CM		 5617 struct btrfs_root *root = BTRFS_I(inode)->root;
5618 struct btrfs_item *item;
5619 struct btrfs_dir_item *di;
5620 struct btrfs_key key;
5f39d397 5621 struct btrfs_key found_key;
39279cc3 5622 struct btrfs_path *path;
16cdcec7
MX		 5623 struct list_head ins_list;
5624 struct list_head del_list;
39279cc3 5625 int ret;
5f39d397 5626 struct extent_buffer *leaf;
39279cc3 5627 int slot;
39279cc3
CM		 5628 unsigned char d_type;
5629 int over = 0;
5630 u32 di_cur;
5631 u32 di_total;
5632 u32 di_len;
5633 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 5634 char tmp_name[32];
5635 char *name_ptr;
5636 int name_len;
9cdda8d3 5637 int is_curr = 0; /* ctx->pos points to the current index? */
39279cc3
CM		 5638
5639 /* FIXME, use a real flag for deciding about the key type */
5640 if (root->fs_info->tree_root == root)
5641 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 5642
9cdda8d3
AV		 5643 if (!dir_emit_dots(file, ctx))
5644 return 0;
5645
49593bfa 5646 path = btrfs_alloc_path();
16cdcec7
MX		 5647 if (!path)
5648 return -ENOMEM;
ff5714cc 5649
026fd317 5650 path->reada = 1;
49593bfa 5651
16cdcec7
MX		 5652 if (key_type == BTRFS_DIR_INDEX_KEY) {
5653 INIT_LIST_HEAD(&ins_list);
5654 INIT_LIST_HEAD(&del_list);
5655 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5656 }
5657
962a298f 5658 key.type = key_type;
9cdda8d3 5659 key.offset = ctx->pos;
33345d01 5660 key.objectid = btrfs_ino(inode);
5f39d397 5661
39279cc3
CM		 5662 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5663 if (ret < 0)
5664 goto err;
49593bfa
DW		 5665
5666 while (1) {
5f39d397 5667 leaf = path->nodes[0];
39279cc3 5668 slot = path->slots[0];
b9e03af0
LZ		 5669 if (slot >= btrfs_header_nritems(leaf)) {
5670 ret = btrfs_next_leaf(root, path);
5671 if (ret < 0)
5672 goto err;
5673 else if (ret > 0)
5674 break;
5675 continue;
39279cc3 5676 }
3de4586c 5677
dd3cc16b 5678 item = btrfs_item_nr(slot);
5f39d397
CM		 5679 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5680
5681 if (found_key.objectid != key.objectid)
39279cc3 5682 break;
962a298f 5683 if (found_key.type != key_type)
39279cc3 5684 break;
9cdda8d3 5685 if (found_key.offset < ctx->pos)
b9e03af0 5686 goto next;
16cdcec7
MX		 5687 if (key_type == BTRFS_DIR_INDEX_KEY &&
5688 btrfs_should_delete_dir_index(&del_list,
5689 found_key.offset))
5690 goto next;
5f39d397 5691
9cdda8d3 5692 ctx->pos = found_key.offset;
16cdcec7 5693 is_curr = 1;
49593bfa 5694
39279cc3
CM		 5695 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5696 di_cur = 0;
5f39d397 5697 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 5698
5699 while (di_cur < di_total) {
5f39d397
CM		 5700 struct btrfs_key location;
5701
22a94d44
JB		 5702 if (verify_dir_item(root, leaf, di))
5703 break;
5704
5f39d397 5705 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 5706 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 5707 name_ptr = tmp_name;
5708 } else {
5709 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 5710 if (!name_ptr) {
5711 ret = -ENOMEM;
5712 goto err;
5713 }
5f39d397
CM		 5714 }
5715 read_extent_buffer(leaf, name_ptr,
5716 (unsigned long)(di + 1), name_len);
5717
5718 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5719 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c 5720
fede766f 5721
3de4586c 5722 /* is this a reference to our own snapshot? If so
8c9c2bf7
AJ		 5723 * skip it.
5724 *
5725 * In contrast to old kernels, we insert the snapshot's
5726 * dir item and dir index after it has been created, so
5727 * we won't find a reference to our own snapshot. We
5728 * still keep the following code for backward
5729 * compatibility.
3de4586c
CM		 5730 */
5731 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5732 location.objectid == root->root_key.objectid) {
5733 over = 0;
5734 goto skip;
5735 }
9cdda8d3
AV		 5736 over = !dir_emit(ctx, name_ptr, name_len,
5737 location.objectid, d_type);
5f39d397 5738
3de4586c 5739skip:
5f39d397
CM		 5740 if (name_ptr != tmp_name)
5741 kfree(name_ptr);
5742
39279cc3
CM		 5743 if (over)
5744 goto nopos;
5103e947 5745 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 5746 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 5747 di_cur += di_len;
5748 di = (struct btrfs_dir_item *)((char *)di + di_len);
5749 }
b9e03af0
LZ		 5750next:
5751 path->slots[0]++;
39279cc3 5752 }
49593bfa 5753
16cdcec7
MX		 5754 if (key_type == BTRFS_DIR_INDEX_KEY) {
5755 if (is_curr)
9cdda8d3
AV		 5756 ctx->pos++;
5757 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
16cdcec7
MX		 5758 if (ret)
5759 goto nopos;
5760 }
5761
49593bfa 5762 /* Reached end of directory/root. Bump pos past the last item. */
db62efbb
ZB		 5763 ctx->pos++;
5764
5765 /*
5766 * Stop new entries from being returned after we return the last
5767 * entry.
5768 *
5769 * New directory entries are assigned a strictly increasing
5770 * offset. This means that new entries created during readdir
5771 * are *guaranteed* to be seen in the future by that readdir.
5772 * This has broken buggy programs which operate on names as
5773 * they're returned by readdir. Until we re-use freed offsets
5774 * we have this hack to stop new entries from being returned
5775 * under the assumption that they'll never reach this huge
5776 * offset.
5777 *
5778 * This is being careful not to overflow 32bit loff_t unless the
5779 * last entry requires it because doing so has broken 32bit apps
5780 * in the past.
5781 */
5782 if (key_type == BTRFS_DIR_INDEX_KEY) {
5783 if (ctx->pos >= INT_MAX)
5784 ctx->pos = LLONG_MAX;
5785 else
5786 ctx->pos = INT_MAX;
5787 }
39279cc3
CM		 5788nopos:
5789 ret = 0;
5790err:
16cdcec7
MX		 5791 if (key_type == BTRFS_DIR_INDEX_KEY)
5792 btrfs_put_delayed_items(&ins_list, &del_list);
39279cc3 5793 btrfs_free_path(path);
39279cc3
CM		 5794 return ret;
5795}
5796
a9185b41 5797int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
39279cc3
CM		 5798{
5799 struct btrfs_root *root = BTRFS_I(inode)->root;
5800 struct btrfs_trans_handle *trans;
5801 int ret = 0;
0af3d00b 5802 bool nolock = false;
39279cc3 5803
72ac3c0d 5804 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4ca8b41e
CM		 5805 return 0;
5806
83eea1f1 5807 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
82d5902d 5808 nolock = true;
0af3d00b 5809
a9185b41 5810 if (wbc->sync_mode == WB_SYNC_ALL) {
0af3d00b 5811 if (nolock)
7a7eaa40 5812 trans = btrfs_join_transaction_nolock(root);
0af3d00b 5813 else
7a7eaa40 5814 trans = btrfs_join_transaction(root);
3612b495
TI		 5815 if (IS_ERR(trans))
5816 return PTR_ERR(trans);
a698d075 5817 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 5818 }
5819 return ret;
5820}
5821
5822/*
54aa1f4d 5823 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 5824 * inode changes. But, it is most likely to find the inode in cache.
5825 * FIXME, needs more benchmarking...there are no reasons other than performance
5826 * to keep or drop this code.
5827 */
48a3b636 5828static int btrfs_dirty_inode(struct inode *inode)
39279cc3
CM		 5829{
5830 struct btrfs_root *root = BTRFS_I(inode)->root;
5831 struct btrfs_trans_handle *trans;
8929ecfa
YZ		 5832 int ret;
5833
72ac3c0d 5834 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
22c44fe6 5835 return 0;
39279cc3 5836
7a7eaa40 5837 trans = btrfs_join_transaction(root);
22c44fe6
JB		 5838 if (IS_ERR(trans))
5839 return PTR_ERR(trans);
8929ecfa
YZ		 5840
5841 ret = btrfs_update_inode(trans, root, inode);
94b60442
CM		 5842 if (ret && ret == -ENOSPC) {
5843 /* whoops, lets try again with the full transaction */
5844 btrfs_end_transaction(trans, root);
5845 trans = btrfs_start_transaction(root, 1);
22c44fe6
JB		 5846 if (IS_ERR(trans))
5847 return PTR_ERR(trans);
8929ecfa 5848
94b60442 5849 ret = btrfs_update_inode(trans, root, inode);
94b60442 5850 }
39279cc3 5851 btrfs_end_transaction(trans, root);
16cdcec7
MX		 5852 if (BTRFS_I(inode)->delayed_node)
5853 btrfs_balance_delayed_items(root);
22c44fe6
JB		 5854
5855 return ret;
5856}
5857
5858/*
5859 * This is a copy of file_update_time. We need this so we can return error on
5860 * ENOSPC for updating the inode in the case of file write and mmap writes.
5861 */
e41f941a
JB		 5862static int btrfs_update_time(struct inode *inode, struct timespec *now,
5863 int flags)
22c44fe6 5864{
2bc55652
AB		 5865 struct btrfs_root *root = BTRFS_I(inode)->root;
5866
5867 if (btrfs_root_readonly(root))
5868 return -EROFS;
5869
e41f941a 5870 if (flags & S_VERSION)
22c44fe6 5871 inode_inc_iversion(inode);
e41f941a
JB		 5872 if (flags & S_CTIME)
5873 inode->i_ctime = *now;
5874 if (flags & S_MTIME)
5875 inode->i_mtime = *now;
5876 if (flags & S_ATIME)
5877 inode->i_atime = *now;
5878 return btrfs_dirty_inode(inode);
39279cc3
CM		 5879}
5880
d352ac68
CM		 5881/*
5882 * find the highest existing sequence number in a directory
5883 * and then set the in-memory index_cnt variable to reflect
5884 * free sequence numbers
5885 */
aec7477b
JB		 5886static int btrfs_set_inode_index_count(struct inode *inode)
5887{
5888 struct btrfs_root *root = BTRFS_I(inode)->root;
5889 struct btrfs_key key, found_key;
5890 struct btrfs_path *path;
5891 struct extent_buffer *leaf;
5892 int ret;
5893
33345d01 5894 key.objectid = btrfs_ino(inode);
962a298f 5895 key.type = BTRFS_DIR_INDEX_KEY;
aec7477b
JB		 5896 key.offset = (u64)-1;
5897
5898 path = btrfs_alloc_path();
5899 if (!path)
5900 return -ENOMEM;
5901
5902 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5903 if (ret < 0)
5904 goto out;
5905 /* FIXME: we should be able to handle this */
5906 if (ret == 0)
5907 goto out;
5908 ret = 0;
5909
5910 /*
5911 * MAGIC NUMBER EXPLANATION:
5912 * since we search a directory based on f_pos we have to start at 2
5913 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5914 * else has to start at 2
5915 */
5916 if (path->slots[0] == 0) {
5917 BTRFS_I(inode)->index_cnt = 2;
5918 goto out;
5919 }
5920
5921 path->slots[0]--;
5922
5923 leaf = path->nodes[0];
5924 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5925
33345d01 5926 if (found_key.objectid != btrfs_ino(inode) ||
962a298f 5927 found_key.type != BTRFS_DIR_INDEX_KEY) {
aec7477b
JB		 5928 BTRFS_I(inode)->index_cnt = 2;
5929 goto out;
5930 }
5931
5932 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5933out:
5934 btrfs_free_path(path);
5935 return ret;
5936}
5937
d352ac68
CM		 5938/*
5939 * helper to find a free sequence number in a given directory. This current
5940 * code is very simple, later versions will do smarter things in the btree
5941 */
3de4586c 5942int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 5943{
5944 int ret = 0;
5945
5946 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
16cdcec7
MX		 5947 ret = btrfs_inode_delayed_dir_index_count(dir);
5948 if (ret) {
5949 ret = btrfs_set_inode_index_count(dir);
5950 if (ret)
5951 return ret;
5952 }
aec7477b
JB		 5953 }
5954
00e4e6b3 5955 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 5956 BTRFS_I(dir)->index_cnt++;
5957
5958 return ret;
5959}
5960
b0d5d10f
CM		 5961static int btrfs_insert_inode_locked(struct inode *inode)
5962{
5963 struct btrfs_iget_args args;
5964 args.location = &BTRFS_I(inode)->location;
5965 args.root = BTRFS_I(inode)->root;
5966
5967 return insert_inode_locked4(inode,
5968 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
5969 btrfs_find_actor, &args);
5970}
5971
39279cc3
CM		 5972static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5973 struct btrfs_root *root,
aec7477b 5974 struct inode *dir,
9c58309d 5975 const char *name, int name_len,
175a4eb7
AV		 5976 u64 ref_objectid, u64 objectid,
5977 umode_t mode, u64 *index)
39279cc3
CM		 5978{
5979 struct inode *inode;
5f39d397 5980 struct btrfs_inode_item *inode_item;
39279cc3 5981 struct btrfs_key *location;
5f39d397 5982 struct btrfs_path *path;
9c58309d
CM		 5983 struct btrfs_inode_ref *ref;
5984 struct btrfs_key key[2];
5985 u32 sizes[2];
ef3b9af5 5986 int nitems = name ? 2 : 1;
9c58309d 5987 unsigned long ptr;
39279cc3 5988 int ret;
39279cc3 5989
5f39d397 5990 path = btrfs_alloc_path();
d8926bb3
MF		 5991 if (!path)
5992 return ERR_PTR(-ENOMEM);
5f39d397 5993
39279cc3 5994 inode = new_inode(root->fs_info->sb);
8fb27640
YS		 5995 if (!inode) {
5996 btrfs_free_path(path);
39279cc3 5997 return ERR_PTR(-ENOMEM);
8fb27640 5998 }
39279cc3 5999
5762b5c9
FM		 6000 /*
6001 * O_TMPFILE, set link count to 0, so that after this point,
6002 * we fill in an inode item with the correct link count.
6003 */
6004 if (!name)
6005 set_nlink(inode, 0);
6006
581bb050
LZ		 6007 /*
6008 * we have to initialize this early, so we can reclaim the inode
6009 * number if we fail afterwards in this function.
6010 */
6011 inode->i_ino = objectid;
6012
ef3b9af5 6013 if (dir && name) {
1abe9b8a 6014 trace_btrfs_inode_request(dir);
6015
3de4586c 6016 ret = btrfs_set_inode_index(dir, index);
09771430 6017 if (ret) {
8fb27640 6018 btrfs_free_path(path);
09771430 6019 iput(inode);
aec7477b 6020 return ERR_PTR(ret);
09771430 6021 }
ef3b9af5
FM		 6022 } else if (dir) {
6023 *index = 0;
aec7477b
JB		 6024 }
6025 /*
6026 * index_cnt is ignored for everything but a dir,
6027 * btrfs_get_inode_index_count has an explanation for the magic
6028 * number
6029 */
6030 BTRFS_I(inode)->index_cnt = 2;
67de1176 6031 BTRFS_I(inode)->dir_index = *index;
39279cc3 6032 BTRFS_I(inode)->root = root;
e02119d5 6033 BTRFS_I(inode)->generation = trans->transid;
76195853 6034 inode->i_generation = BTRFS_I(inode)->generation;
b888db2b 6035
5dc562c5
JB		 6036 /*
6037 * We could have gotten an inode number from somebody who was fsynced
6038 * and then removed in this same transaction, so let's just set full
6039 * sync since it will be a full sync anyway and this will blow away the
6040 * old info in the log.
6041 */
6042 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6043
9c58309d 6044 key[0].objectid = objectid;
962a298f 6045 key[0].type = BTRFS_INODE_ITEM_KEY;
9c58309d
CM		 6046 key[0].offset = 0;
6047
9c58309d 6048 sizes[0] = sizeof(struct btrfs_inode_item);
ef3b9af5
FM		 6049
6050 if (name) {
6051 /*
6052 * Start new inodes with an inode_ref. This is slightly more
6053 * efficient for small numbers of hard links since they will
6054 * be packed into one item. Extended refs will kick in if we
6055 * add more hard links than can fit in the ref item.
6056 */
6057 key[1].objectid = objectid;
962a298f 6058 key[1].type = BTRFS_INODE_REF_KEY;
ef3b9af5
FM		 6059 key[1].offset = ref_objectid;
6060
6061 sizes[1] = name_len + sizeof(*ref);
6062 }
9c58309d 6063
b0d5d10f
CM		 6064 location = &BTRFS_I(inode)->location;
6065 location->objectid = objectid;
6066 location->offset = 0;
962a298f 6067 location->type = BTRFS_INODE_ITEM_KEY;
b0d5d10f
CM		 6068
6069 ret = btrfs_insert_inode_locked(inode);
6070 if (ret < 0)
6071 goto fail;
6072
b9473439 6073 path->leave_spinning = 1;
ef3b9af5 6074 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
9c58309d 6075 if (ret != 0)
b0d5d10f 6076 goto fail_unlock;
5f39d397 6077
ecc11fab 6078 inode_init_owner(inode, dir, mode);
a76a3cd4 6079 inode_set_bytes(inode, 0);
9cc97d64 6080
6081 inode->i_mtime = CURRENT_TIME;
6082 inode->i_atime = inode->i_mtime;
6083 inode->i_ctime = inode->i_mtime;
6084 BTRFS_I(inode)->i_otime = inode->i_mtime;
6085
5f39d397
CM		 6086 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6087 struct btrfs_inode_item);
293f7e07
LZ		 6088 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6089 sizeof(*inode_item));
e02119d5 6090 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d 6091
ef3b9af5
FM		 6092 if (name) {
6093 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6094 struct btrfs_inode_ref);
6095 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6096 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6097 ptr = (unsigned long)(ref + 1);
6098 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6099 }
9c58309d 6100
5f39d397
CM		 6101 btrfs_mark_buffer_dirty(path->nodes[0]);
6102 btrfs_free_path(path);
6103
6cbff00f
CH		 6104 btrfs_inherit_iflags(inode, dir);
6105
569254b0 6106 if (S_ISREG(mode)) {
94272164
CM		 6107 if (btrfs_test_opt(root, NODATASUM))
6108 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
213490b3 6109 if (btrfs_test_opt(root, NODATACOW))
f2bdf9a8
JB		 6110 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6111 BTRFS_INODE_NODATASUM;
94272164
CM		 6112 }
6113
5d4f98a2 6114 inode_tree_add(inode);
1abe9b8a 6115
6116 trace_btrfs_inode_new(inode);
1973f0fa 6117 btrfs_set_inode_last_trans(trans, inode);
1abe9b8a 6118
8ea05e3a
AB		 6119 btrfs_update_root_times(trans, root);
6120
63541927
FDBM		 6121 ret = btrfs_inode_inherit_props(trans, inode, dir);
6122 if (ret)
6123 btrfs_err(root->fs_info,
6124 "error inheriting props for ino %llu (root %llu): %d",
6125 btrfs_ino(inode), root->root_key.objectid, ret);
6126
39279cc3 6127 return inode;
b0d5d10f
CM		 6128
6129fail_unlock:
6130 unlock_new_inode(inode);
5f39d397 6131fail:
ef3b9af5 6132 if (dir && name)
aec7477b 6133 BTRFS_I(dir)->index_cnt--;
5f39d397 6134 btrfs_free_path(path);
09771430 6135 iput(inode);
5f39d397 6136 return ERR_PTR(ret);
39279cc3
CM		 6137}
6138
6139static inline u8 btrfs_inode_type(struct inode *inode)
6140{
6141 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6142}
6143
d352ac68
CM		 6144/*
6145 * utility function to add 'inode' into 'parent_inode' with
6146 * a give name and a given sequence number.
6147 * if 'add_backref' is true, also insert a backref from the
6148 * inode to the parent directory.
6149 */
e02119d5
CM		 6150int btrfs_add_link(struct btrfs_trans_handle *trans,
6151 struct inode *parent_inode, struct inode *inode,
6152 const char *name, int name_len, int add_backref, u64 index)
39279cc3 6153{
4df27c4d 6154 int ret = 0;
39279cc3 6155 struct btrfs_key key;
e02119d5 6156 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
33345d01
LZ		 6157 u64 ino = btrfs_ino(inode);
6158 u64 parent_ino = btrfs_ino(parent_inode);
5f39d397 6159
33345d01 6160 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 6161 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6162 } else {
33345d01 6163 key.objectid = ino;
962a298f 6164 key.type = BTRFS_INODE_ITEM_KEY;
4df27c4d
YZ		 6165 key.offset = 0;
6166 }
6167
33345d01 6168 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 6169 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6170 key.objectid, root->root_key.objectid,
33345d01 6171 parent_ino, index, name, name_len);
4df27c4d 6172 } else if (add_backref) {
33345d01
LZ		 6173 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6174 parent_ino, index);
4df27c4d 6175 }
39279cc3 6176
79787eaa
JM		 6177 /* Nothing to clean up yet */
6178 if (ret)
6179 return ret;
4df27c4d 6180
79787eaa
JM		 6181 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6182 parent_inode, &key,
6183 btrfs_inode_type(inode), index);
9c52057c 6184 if (ret == -EEXIST || ret == -EOVERFLOW)
79787eaa
JM		 6185 goto fail_dir_item;
6186 else if (ret) {
6187 btrfs_abort_transaction(trans, root, ret);
6188 return ret;
39279cc3 6189 }
79787eaa
JM		 6190
6191 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6192 name_len * 2);
0c4d2d95 6193 inode_inc_iversion(parent_inode);
79787eaa
JM		 6194 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
6195 ret = btrfs_update_inode(trans, root, parent_inode);
6196 if (ret)
6197 btrfs_abort_transaction(trans, root, ret);
39279cc3 6198 return ret;
fe66a05a
CM		 6199
6200fail_dir_item:
6201 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6202 u64 local_index;
6203 int err;
6204 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6205 key.objectid, root->root_key.objectid,
6206 parent_ino, &local_index, name, name_len);
6207
6208 } else if (add_backref) {
6209 u64 local_index;
6210 int err;
6211
6212 err = btrfs_del_inode_ref(trans, root, name, name_len,
6213 ino, parent_ino, &local_index);
6214 }
6215 return ret;
39279cc3
CM		 6216}
6217
6218static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
a1b075d2
JB		 6219 struct inode *dir, struct dentry *dentry,
6220 struct inode *inode, int backref, u64 index)
39279cc3 6221{
a1b075d2
JB		 6222 int err = btrfs_add_link(trans, dir, inode,
6223 dentry->d_name.name, dentry->d_name.len,
6224 backref, index);
39279cc3
CM		 6225 if (err > 0)
6226 err = -EEXIST;
6227 return err;
6228}
6229
618e21d5 6230static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1a67aafb 6231 umode_t mode, dev_t rdev)
618e21d5
JB		 6232{
6233 struct btrfs_trans_handle *trans;
6234 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 6235 struct inode *inode = NULL;
618e21d5
JB		 6236 int err;
6237 int drop_inode = 0;
6238 u64 objectid;
00e4e6b3 6239 u64 index = 0;
618e21d5
JB		 6240
6241 if (!new_valid_dev(rdev))
6242 return -EINVAL;
6243
9ed74f2d
JB		 6244 /*
6245 * 2 for inode item and ref
6246 * 2 for dir items
6247 * 1 for xattr if selinux is on
6248 */
a22285a6
YZ		 6249 trans = btrfs_start_transaction(root, 5);
6250 if (IS_ERR(trans))
6251 return PTR_ERR(trans);
1832a6d5 6252
581bb050
LZ		 6253 err = btrfs_find_free_ino(root, &objectid);
6254 if (err)
6255 goto out_unlock;
6256
aec7477b 6257 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 6258 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 6259 mode, &index);
7cf96da3
TI		 6260 if (IS_ERR(inode)) {
6261 err = PTR_ERR(inode);
618e21d5 6262 goto out_unlock;
7cf96da3 6263 }
618e21d5 6264
ad19db71
CS		 6265 /*
6266 * If the active LSM wants to access the inode during
6267 * d_instantiate it needs these. Smack checks to see
6268 * if the filesystem supports xattrs by looking at the
6269 * ops vector.
6270 */
ad19db71 6271 inode->i_op = &btrfs_special_inode_operations;
b0d5d10f
CM		 6272 init_special_inode(inode, inode->i_mode, rdev);
6273
6274 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
618e21d5 6275 if (err)
b0d5d10f
CM		 6276 goto out_unlock_inode;
6277
6278 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6279 if (err) {
6280 goto out_unlock_inode;
6281 } else {
1b4ab1bb 6282 btrfs_update_inode(trans, root, inode);
b0d5d10f 6283 unlock_new_inode(inode);
08c422c2 6284 d_instantiate(dentry, inode);
618e21d5 6285 }
b0d5d10f 6286
618e21d5 6287out_unlock:
7ad85bb7 6288 btrfs_end_transaction(trans, root);
c581afc8 6289 btrfs_balance_delayed_items(root);
b53d3f5d 6290 btrfs_btree_balance_dirty(root);
618e21d5
JB		 6291 if (drop_inode) {
6292 inode_dec_link_count(inode);
6293 iput(inode);
6294 }
618e21d5 6295 return err;
b0d5d10f
CM		 6296
6297out_unlock_inode:
6298 drop_inode = 1;
6299 unlock_new_inode(inode);
6300 goto out_unlock;
6301
618e21d5
JB		 6302}
6303
39279cc3 6304static int btrfs_create(struct inode *dir, struct dentry *dentry,
ebfc3b49 6305 umode_t mode, bool excl)
39279cc3
CM		 6306{
6307 struct btrfs_trans_handle *trans;
6308 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 6309 struct inode *inode = NULL;
43baa579 6310 int drop_inode_on_err = 0;
a22285a6 6311 int err;
39279cc3 6312 u64 objectid;
00e4e6b3 6313 u64 index = 0;
39279cc3 6314
9ed74f2d
JB		 6315 /*
6316 * 2 for inode item and ref
6317 * 2 for dir items
6318 * 1 for xattr if selinux is on
6319 */
a22285a6
YZ		 6320 trans = btrfs_start_transaction(root, 5);
6321 if (IS_ERR(trans))
6322 return PTR_ERR(trans);
9ed74f2d 6323
581bb050
LZ		 6324 err = btrfs_find_free_ino(root, &objectid);
6325 if (err)
6326 goto out_unlock;
6327
aec7477b 6328 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 6329 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 6330 mode, &index);
7cf96da3
TI		 6331 if (IS_ERR(inode)) {
6332 err = PTR_ERR(inode);
39279cc3 6333 goto out_unlock;
7cf96da3 6334 }
43baa579 6335 drop_inode_on_err = 1;
ad19db71
CS		 6336 /*
6337 * If the active LSM wants to access the inode during
6338 * d_instantiate it needs these. Smack checks to see
6339 * if the filesystem supports xattrs by looking at the
6340 * ops vector.
6341 */
6342 inode->i_fop = &btrfs_file_operations;
6343 inode->i_op = &btrfs_file_inode_operations;
b0d5d10f 6344 inode->i_mapping->a_ops = &btrfs_aops;
b0d5d10f
CM		 6345
6346 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6347 if (err)
6348 goto out_unlock_inode;
6349
6350 err = btrfs_update_inode(trans, root, inode);
6351 if (err)
6352 goto out_unlock_inode;
ad19db71 6353
a1b075d2 6354 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3 6355 if (err)
b0d5d10f 6356 goto out_unlock_inode;
43baa579 6357
43baa579 6358 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
b0d5d10f 6359 unlock_new_inode(inode);
43baa579
FB		 6360 d_instantiate(dentry, inode);
6361
39279cc3 6362out_unlock:
7ad85bb7 6363 btrfs_end_transaction(trans, root);
43baa579 6364 if (err && drop_inode_on_err) {
39279cc3
CM		 6365 inode_dec_link_count(inode);
6366 iput(inode);
6367 }
c581afc8 6368 btrfs_balance_delayed_items(root);
b53d3f5d 6369 btrfs_btree_balance_dirty(root);
39279cc3 6370 return err;
b0d5d10f
CM		 6371
6372out_unlock_inode:
6373 unlock_new_inode(inode);
6374 goto out_unlock;
6375
39279cc3
CM		 6376}
6377
6378static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6379 struct dentry *dentry)
6380{
6381 struct btrfs_trans_handle *trans;
6382 struct btrfs_root *root = BTRFS_I(dir)->root;
2b0143b5 6383 struct inode *inode = d_inode(old_dentry);
00e4e6b3 6384 u64 index;
39279cc3
CM		 6385 int err;
6386 int drop_inode = 0;
6387
4a8be425
TH		 6388 /* do not allow sys_link's with other subvols of the same device */
6389 if (root->objectid != BTRFS_I(inode)->root->objectid)
3ab3564f 6390 return -EXDEV;
4a8be425 6391
f186373f 6392 if (inode->i_nlink >= BTRFS_LINK_MAX)
c055e99e 6393 return -EMLINK;
4a8be425 6394
3de4586c 6395 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 6396 if (err)
6397 goto fail;
6398
a22285a6 6399 /*
7e6b6465 6400 * 2 items for inode and inode ref
a22285a6 6401 * 2 items for dir items
7e6b6465 6402 * 1 item for parent inode
a22285a6 6403 */
7e6b6465 6404 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 6405 if (IS_ERR(trans)) {
6406 err = PTR_ERR(trans);
6407 goto fail;
6408 }
5f39d397 6409
67de1176
MX		 6410 /* There are several dir indexes for this inode, clear the cache. */
6411 BTRFS_I(inode)->dir_index = 0ULL;
8b558c5f 6412 inc_nlink(inode);
0c4d2d95 6413 inode_inc_iversion(inode);
3153495d 6414 inode->i_ctime = CURRENT_TIME;
7de9c6ee 6415 ihold(inode);
e9976151 6416 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
aec7477b 6417
a1b075d2 6418 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5f39d397 6419
a5719521 6420 if (err) {
54aa1f4d 6421 drop_inode = 1;
a5719521 6422 } else {
10d9f309 6423 struct dentry *parent = dentry->d_parent;
a5719521 6424 err = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 6425 if (err)
6426 goto fail;
ef3b9af5
FM		 6427 if (inode->i_nlink == 1) {
6428 /*
6429 * If new hard link count is 1, it's a file created
6430 * with open(2) O_TMPFILE flag.
6431 */
6432 err = btrfs_orphan_del(trans, inode);
6433 if (err)
6434 goto fail;
6435 }
08c422c2 6436 d_instantiate(dentry, inode);
6a912213 6437 btrfs_log_new_name(trans, inode, NULL, parent);
a5719521 6438 }
39279cc3 6439
7ad85bb7 6440 btrfs_end_transaction(trans, root);
c581afc8 6441 btrfs_balance_delayed_items(root);
1832a6d5 6442fail:
39279cc3
CM		 6443 if (drop_inode) {
6444 inode_dec_link_count(inode);
6445 iput(inode);
6446 }
b53d3f5d 6447 btrfs_btree_balance_dirty(root);
39279cc3
CM		 6448 return err;
6449}
6450
18bb1db3 6451static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
39279cc3 6452{
b9d86667 6453 struct inode *inode = NULL;
39279cc3
CM		 6454 struct btrfs_trans_handle *trans;
6455 struct btrfs_root *root = BTRFS_I(dir)->root;
6456 int err = 0;
6457 int drop_on_err = 0;
b9d86667 6458 u64 objectid = 0;
00e4e6b3 6459 u64 index = 0;
39279cc3 6460
9ed74f2d
JB		 6461 /*
6462 * 2 items for inode and ref
6463 * 2 items for dir items
6464 * 1 for xattr if selinux is on
6465 */
a22285a6
YZ		 6466 trans = btrfs_start_transaction(root, 5);
6467 if (IS_ERR(trans))
6468 return PTR_ERR(trans);
39279cc3 6469
581bb050
LZ		 6470 err = btrfs_find_free_ino(root, &objectid);
6471 if (err)
6472 goto out_fail;
6473
aec7477b 6474 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 6475 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 6476 S_IFDIR | mode, &index);
39279cc3
CM		 6477 if (IS_ERR(inode)) {
6478 err = PTR_ERR(inode);
6479 goto out_fail;
6480 }
5f39d397 6481
39279cc3 6482 drop_on_err = 1;
b0d5d10f
CM		 6483 /* these must be set before we unlock the inode */
6484 inode->i_op = &btrfs_dir_inode_operations;
6485 inode->i_fop = &btrfs_dir_file_operations;
33268eaf 6486
2a7dba39 6487 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf 6488 if (err)
b0d5d10f 6489 goto out_fail_inode;
39279cc3 6490
dbe674a9 6491 btrfs_i_size_write(inode, 0);
39279cc3
CM		 6492 err = btrfs_update_inode(trans, root, inode);
6493 if (err)
b0d5d10f 6494 goto out_fail_inode;
5f39d397 6495
a1b075d2
JB		 6496 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6497 dentry->d_name.len, 0, index);
39279cc3 6498 if (err)
b0d5d10f 6499 goto out_fail_inode;
5f39d397 6500
39279cc3 6501 d_instantiate(dentry, inode);
b0d5d10f
CM		 6502 /*
6503 * mkdir is special. We're unlocking after we call d_instantiate
6504 * to avoid a race with nfsd calling d_instantiate.
6505 */
6506 unlock_new_inode(inode);
39279cc3 6507 drop_on_err = 0;
39279cc3
CM		 6508
6509out_fail:
7ad85bb7 6510 btrfs_end_transaction(trans, root);
c7cfb8a5
WS		 6511 if (drop_on_err) {
6512 inode_dec_link_count(inode);
39279cc3 6513 iput(inode);
c7cfb8a5 6514 }
c581afc8 6515 btrfs_balance_delayed_items(root);
b53d3f5d 6516 btrfs_btree_balance_dirty(root);
39279cc3 6517 return err;
b0d5d10f
CM		 6518
6519out_fail_inode:
6520 unlock_new_inode(inode);
6521 goto out_fail;
39279cc3
CM		 6522}
6523
e6c4efd8
QW		 6524/* Find next extent map of a given extent map, caller needs to ensure locks */
6525static struct extent_map *next_extent_map(struct extent_map *em)
6526{
6527 struct rb_node *next;
6528
6529 next = rb_next(&em->rb_node);
6530 if (!next)
6531 return NULL;
6532 return container_of(next, struct extent_map, rb_node);
6533}
6534
6535static struct extent_map *prev_extent_map(struct extent_map *em)
6536{
6537 struct rb_node *prev;
6538
6539 prev = rb_prev(&em->rb_node);
6540 if (!prev)
6541 return NULL;
6542 return container_of(prev, struct extent_map, rb_node);
6543}
6544
d352ac68 6545/* helper for btfs_get_extent. Given an existing extent in the tree,
e6c4efd8 6546 * the existing extent is the nearest extent to map_start,
d352ac68 6547 * and an extent that you want to insert, deal with overlap and insert
e6c4efd8 6548 * the best fitted new extent into the tree.
d352ac68 6549 */
3b951516
CM		 6550static int merge_extent_mapping(struct extent_map_tree *em_tree,
6551 struct extent_map *existing,
e6dcd2dc 6552 struct extent_map *em,
51f395ad 6553 u64 map_start)
3b951516 6554{
e6c4efd8
QW		 6555 struct extent_map *prev;
6556 struct extent_map *next;
6557 u64 start;
6558 u64 end;
3b951516 6559 u64 start_diff;
3b951516 6560
e6dcd2dc 6561 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
e6c4efd8
QW		 6562
6563 if (existing->start > map_start) {
6564 next = existing;
6565 prev = prev_extent_map(next);
6566 } else {
6567 prev = existing;
6568 next = next_extent_map(prev);
6569 }
6570
6571 start = prev ? extent_map_end(prev) : em->start;
6572 start = max_t(u64, start, em->start);
6573 end = next ? next->start : extent_map_end(em);
6574 end = min_t(u64, end, extent_map_end(em));
6575 start_diff = start - em->start;
6576 em->start = start;
6577 em->len = end - start;
c8b97818
CM		 6578 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6579 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 6580 em->block_start += start_diff;
c8b97818
CM		 6581 em->block_len -= start_diff;
6582 }
09a2a8f9 6583 return add_extent_mapping(em_tree, em, 0);
3b951516
CM		 6584}
6585
c8b97818
CM		 6586static noinline int uncompress_inline(struct btrfs_path *path,
6587 struct inode *inode, struct page *page,
6588 size_t pg_offset, u64 extent_offset,
6589 struct btrfs_file_extent_item *item)
6590{
6591 int ret;
6592 struct extent_buffer *leaf = path->nodes[0];
6593 char *tmp;
6594 size_t max_size;
6595 unsigned long inline_size;
6596 unsigned long ptr;
261507a0 6597 int compress_type;
c8b97818
CM		 6598
6599 WARN_ON(pg_offset != 0);
261507a0 6600 compress_type = btrfs_file_extent_compression(leaf, item);
c8b97818
CM		 6601 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6602 inline_size = btrfs_file_extent_inline_item_len(leaf,
dd3cc16b 6603 btrfs_item_nr(path->slots[0]));
c8b97818 6604 tmp = kmalloc(inline_size, GFP_NOFS);
8d413713
TI		 6605 if (!tmp)
6606 return -ENOMEM;
c8b97818
CM		 6607 ptr = btrfs_file_extent_inline_start(item);
6608
6609 read_extent_buffer(leaf, tmp, ptr, inline_size);
6610
5b050f04 6611 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
261507a0
LZ		 6612 ret = btrfs_decompress(compress_type, tmp, page,
6613 extent_offset, inline_size, max_size);
c8b97818 6614 kfree(tmp);
166ae5a4 6615 return ret;
c8b97818
CM		 6616}
6617
d352ac68
CM		 6618/*
6619 * a bit scary, this does extent mapping from logical file offset to the disk.
d397712b
CM		 6620 * the ugly parts come from merging extents from the disk with the in-ram
6621 * representation. This gets more complex because of the data=ordered code,
d352ac68
CM		 6622 * where the in-ram extents might be locked pending data=ordered completion.
6623 *
6624 * This also copies inline extents directly into the page.
6625 */
d397712b 6626
a52d9a80 6627struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 6628 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 6629 int create)
6630{
6631 int ret;
6632 int err = 0;
a52d9a80
CM		 6633 u64 extent_start = 0;
6634 u64 extent_end = 0;
33345d01 6635 u64 objectid = btrfs_ino(inode);
a52d9a80 6636 u32 found_type;
f421950f 6637 struct btrfs_path *path = NULL;
a52d9a80
CM		 6638 struct btrfs_root *root = BTRFS_I(inode)->root;
6639 struct btrfs_file_extent_item *item;
5f39d397
CM		 6640 struct extent_buffer *leaf;
6641 struct btrfs_key found_key;
a52d9a80
CM		 6642 struct extent_map *em = NULL;
6643 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 6644 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 6645 struct btrfs_trans_handle *trans = NULL;
7ffbb598 6646 const bool new_inline = !page || create;
a52d9a80 6647
a52d9a80 6648again:
890871be 6649 read_lock(&em_tree->lock);
d1310b2e 6650 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 6651 if (em)
6652 em->bdev = root->fs_info->fs_devices->latest_bdev;
890871be 6653 read_unlock(&em_tree->lock);
d1310b2e 6654
a52d9a80 6655 if (em) {
e1c4b745
CM		 6656 if (em->start > start || em->start + em->len <= start)
6657 free_extent_map(em);
6658 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 6659 free_extent_map(em);
6660 else
6661 goto out;
a52d9a80 6662 }
172ddd60 6663 em = alloc_extent_map();
a52d9a80 6664 if (!em) {
d1310b2e
CM		 6665 err = -ENOMEM;
6666 goto out;
a52d9a80 6667 }
e6dcd2dc 6668 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 6669 em->start = EXTENT_MAP_HOLE;
445a6944 6670 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 6671 em->len = (u64)-1;
c8b97818 6672 em->block_len = (u64)-1;
f421950f
CM		 6673
6674 if (!path) {
6675 path = btrfs_alloc_path();
026fd317
JB		 6676 if (!path) {
6677 err = -ENOMEM;
6678 goto out;
6679 }
6680 /*
6681 * Chances are we'll be called again, so go ahead and do
6682 * readahead
6683 */
6684 path->reada = 1;
f421950f
CM		 6685 }
6686
179e29e4
CM		 6687 ret = btrfs_lookup_file_extent(trans, root, path,
6688 objectid, start, trans != NULL);
a52d9a80
CM		 6689 if (ret < 0) {
6690 err = ret;
6691 goto out;
6692 }
6693
6694 if (ret != 0) {
6695 if (path->slots[0] == 0)
6696 goto not_found;
6697 path->slots[0]--;
6698 }
6699
5f39d397
CM		 6700 leaf = path->nodes[0];
6701 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 6702 struct btrfs_file_extent_item);
a52d9a80 6703 /* are we inside the extent that was found? */
5f39d397 6704 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
962a298f 6705 found_type = found_key.type;
5f39d397 6706 if (found_key.objectid != objectid ||
a52d9a80 6707 found_type != BTRFS_EXTENT_DATA_KEY) {
25a50341
JB		 6708 /*
6709 * If we backup past the first extent we want to move forward
6710 * and see if there is an extent in front of us, otherwise we'll
6711 * say there is a hole for our whole search range which can
6712 * cause problems.
6713 */
6714 extent_end = start;
6715 goto next;
a52d9a80
CM		 6716 }
6717
5f39d397
CM		 6718 found_type = btrfs_file_extent_type(leaf, item);
6719 extent_start = found_key.offset;
d899e052
YZ		 6720 if (found_type == BTRFS_FILE_EXTENT_REG ||
6721 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 6722 extent_end = extent_start +
db94535d 6723 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 6724 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6725 size_t size;
514ac8ad 6726 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
fda2832f 6727 extent_end = ALIGN(extent_start + size, root->sectorsize);
9036c102 6728 }
25a50341 6729next:
9036c102
YZ		 6730 if (start >= extent_end) {
6731 path->slots[0]++;
6732 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6733 ret = btrfs_next_leaf(root, path);
6734 if (ret < 0) {
6735 err = ret;
6736 goto out;
a52d9a80 6737 }
9036c102
YZ		 6738 if (ret > 0)
6739 goto not_found;
6740 leaf = path->nodes[0];
a52d9a80 6741 }
9036c102
YZ		 6742 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6743 if (found_key.objectid != objectid ||
6744 found_key.type != BTRFS_EXTENT_DATA_KEY)
6745 goto not_found;
6746 if (start + len <= found_key.offset)
6747 goto not_found;
e2eca69d
WS		 6748 if (start > found_key.offset)
6749 goto next;
9036c102 6750 em->start = start;
70c8a91c 6751 em->orig_start = start;
9036c102
YZ		 6752 em->len = found_key.offset - start;
6753 goto not_found_em;
6754 }
6755
7ffbb598
FM		 6756 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6757
d899e052
YZ		 6758 if (found_type == BTRFS_FILE_EXTENT_REG ||
6759 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80
CM		 6760 goto insert;
6761 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 6762 unsigned long ptr;
a52d9a80 6763 char *map;
3326d1b0
CM		 6764 size_t size;
6765 size_t extent_offset;
6766 size_t copy_size;
a52d9a80 6767
7ffbb598 6768 if (new_inline)
689f9346 6769 goto out;
5f39d397 6770
514ac8ad 6771 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
9036c102 6772 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 6773 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 6774 size - extent_offset);
3326d1b0 6775 em->start = extent_start + extent_offset;
fda2832f 6776 em->len = ALIGN(copy_size, root->sectorsize);
b4939680 6777 em->orig_block_len = em->len;
70c8a91c 6778 em->orig_start = em->start;
689f9346 6779 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 6780 if (create == 0 && !PageUptodate(page)) {
261507a0
LZ		 6781 if (btrfs_file_extent_compression(leaf, item) !=
6782 BTRFS_COMPRESS_NONE) {
c8b97818
CM		 6783 ret = uncompress_inline(path, inode, page,
6784 pg_offset,
6785 extent_offset, item);
166ae5a4
ZB		 6786 if (ret) {
6787 err = ret;
6788 goto out;
6789 }
c8b97818
CM		 6790 } else {
6791 map = kmap(page);
6792 read_extent_buffer(leaf, map + pg_offset, ptr,
6793 copy_size);
93c82d57
CM		 6794 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6795 memset(map + pg_offset + copy_size, 0,
6796 PAGE_CACHE_SIZE - pg_offset -
6797 copy_size);
6798 }
c8b97818
CM		 6799 kunmap(page);
6800 }
179e29e4
CM		 6801 flush_dcache_page(page);
6802 } else if (create && PageUptodate(page)) {
6bf7e080 6803 BUG();
179e29e4
CM		 6804 if (!trans) {
6805 kunmap(page);
6806 free_extent_map(em);
6807 em = NULL;
ff5714cc 6808
b3b4aa74 6809 btrfs_release_path(path);
7a7eaa40 6810 trans = btrfs_join_transaction(root);
ff5714cc 6811
3612b495
TI		 6812 if (IS_ERR(trans))
6813 return ERR_CAST(trans);
179e29e4
CM		 6814 goto again;
6815 }
c8b97818 6816 map = kmap(page);
70dec807 6817 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 6818 copy_size);
c8b97818 6819 kunmap(page);
179e29e4 6820 btrfs_mark_buffer_dirty(leaf);
a52d9a80 6821 }
d1310b2e 6822 set_extent_uptodate(io_tree, em->start,
507903b8 6823 extent_map_end(em) - 1, NULL, GFP_NOFS);
a52d9a80 6824 goto insert;
a52d9a80
CM		 6825 }
6826not_found:
6827 em->start = start;
70c8a91c 6828 em->orig_start = start;
d1310b2e 6829 em->len = len;
a52d9a80 6830not_found_em:
5f39d397 6831 em->block_start = EXTENT_MAP_HOLE;
9036c102 6832 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80 6833insert:
b3b4aa74 6834 btrfs_release_path(path);
d1310b2e 6835 if (em->start > start || extent_map_end(em) <= start) {
c2cf52eb 6836 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
c1c9ff7c 6837 em->start, em->len, start, len);
a52d9a80
CM		 6838 err = -EIO;
6839 goto out;
6840 }
d1310b2e
CM		 6841
6842 err = 0;
890871be 6843 write_lock(&em_tree->lock);
09a2a8f9 6844 ret = add_extent_mapping(em_tree, em, 0);
3b951516
CM		 6845 /* it is possible that someone inserted the extent into the tree
6846 * while we had the lock dropped. It is also possible that
6847 * an overlapping map exists in the tree
6848 */
a52d9a80 6849 if (ret == -EEXIST) {
3b951516 6850 struct extent_map *existing;
e6dcd2dc
CM		 6851
6852 ret = 0;
6853
e6c4efd8
QW		 6854 existing = search_extent_mapping(em_tree, start, len);
6855 /*
6856 * existing will always be non-NULL, since there must be
6857 * extent causing the -EEXIST.
6858 */
6859 if (start >= extent_map_end(existing) ||
32be3a1a 6860 start <= existing->start) {
e6c4efd8
QW		 6861 /*
6862 * The existing extent map is the one nearest to
6863 * the [start, start + len) range which overlaps
6864 */
6865 err = merge_extent_mapping(em_tree, existing,
6866 em, start);
e1c4b745 6867 free_extent_map(existing);
e6c4efd8 6868 if (err) {
3b951516
CM		 6869 free_extent_map(em);
6870 em = NULL;
6871 }
6872 } else {
6873 free_extent_map(em);
6874 em = existing;
e6dcd2dc 6875 err = 0;
a52d9a80 6876 }
a52d9a80 6877 }
890871be 6878 write_unlock(&em_tree->lock);
a52d9a80 6879out:
1abe9b8a 6880
4cd8587c 6881 trace_btrfs_get_extent(root, em);
1abe9b8a 6882
f421950f
CM		 6883 if (path)
6884 btrfs_free_path(path);
a52d9a80
CM		 6885 if (trans) {
6886 ret = btrfs_end_transaction(trans, root);
d397712b 6887 if (!err)
a52d9a80
CM		 6888 err = ret;
6889 }
a52d9a80
CM		 6890 if (err) {
6891 free_extent_map(em);
a52d9a80
CM		 6892 return ERR_PTR(err);
6893 }
79787eaa 6894 BUG_ON(!em); /* Error is always set */
a52d9a80
CM		 6895 return em;
6896}
6897
ec29ed5b
CM		 6898struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6899 size_t pg_offset, u64 start, u64 len,
6900 int create)
6901{
6902 struct extent_map *em;
6903 struct extent_map *hole_em = NULL;
6904 u64 range_start = start;
6905 u64 end;
6906 u64 found;
6907 u64 found_end;
6908 int err = 0;
6909
6910 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6911 if (IS_ERR(em))
6912 return em;
6913 if (em) {
6914 /*
f9e4fb53
LB		 6915 * if our em maps to
6916 * - a hole or
6917 * - a pre-alloc extent,
6918 * there might actually be delalloc bytes behind it.
ec29ed5b 6919 */
f9e4fb53
LB		 6920 if (em->block_start != EXTENT_MAP_HOLE &&
6921 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
ec29ed5b
CM		 6922 return em;
6923 else
6924 hole_em = em;
6925 }
6926
6927 /* check to see if we've wrapped (len == -1 or similar) */
6928 end = start + len;
6929 if (end < start)
6930 end = (u64)-1;
6931 else
6932 end -= 1;
6933
6934 em = NULL;
6935
6936 /* ok, we didn't find anything, lets look for delalloc */
6937 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6938 end, len, EXTENT_DELALLOC, 1);
6939 found_end = range_start + found;
6940 if (found_end < range_start)
6941 found_end = (u64)-1;
6942
6943 /*
6944 * we didn't find anything useful, return
6945 * the original results from get_extent()
6946 */
6947 if (range_start > end || found_end <= start) {
6948 em = hole_em;
6949 hole_em = NULL;
6950 goto out;
6951 }
6952
6953 /* adjust the range_start to make sure it doesn't
6954 * go backwards from the start they passed in
6955 */
67871254 6956 range_start = max(start, range_start);
ec29ed5b
CM		 6957 found = found_end - range_start;
6958
6959 if (found > 0) {
6960 u64 hole_start = start;
6961 u64 hole_len = len;
6962
172ddd60 6963 em = alloc_extent_map();
ec29ed5b
CM		 6964 if (!em) {
6965 err = -ENOMEM;
6966 goto out;
6967 }
6968 /*
6969 * when btrfs_get_extent can't find anything it
6970 * returns one huge hole
6971 *
6972 * make sure what it found really fits our range, and
6973 * adjust to make sure it is based on the start from
6974 * the caller
6975 */
6976 if (hole_em) {
6977 u64 calc_end = extent_map_end(hole_em);
6978
6979 if (calc_end <= start || (hole_em->start > end)) {
6980 free_extent_map(hole_em);
6981 hole_em = NULL;
6982 } else {
6983 hole_start = max(hole_em->start, start);
6984 hole_len = calc_end - hole_start;
6985 }
6986 }
6987 em->bdev = NULL;
6988 if (hole_em && range_start > hole_start) {
6989 /* our hole starts before our delalloc, so we
6990 * have to return just the parts of the hole
6991 * that go until the delalloc starts
6992 */
6993 em->len = min(hole_len,
6994 range_start - hole_start);
6995 em->start = hole_start;
6996 em->orig_start = hole_start;
6997 /*
6998 * don't adjust block start at all,
6999 * it is fixed at EXTENT_MAP_HOLE
7000 */
7001 em->block_start = hole_em->block_start;
7002 em->block_len = hole_len;
f9e4fb53
LB		 7003 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7004 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
ec29ed5b
CM		 7005 } else {
7006 em->start = range_start;
7007 em->len = found;
7008 em->orig_start = range_start;
7009 em->block_start = EXTENT_MAP_DELALLOC;
7010 em->block_len = found;
7011 }
7012 } else if (hole_em) {
7013 return hole_em;
7014 }
7015out:
7016
7017 free_extent_map(hole_em);
7018 if (err) {
7019 free_extent_map(em);
7020 return ERR_PTR(err);
7021 }
7022 return em;
7023}
7024
4b46fce2
JB		 7025static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7026 u64 start, u64 len)
7027{
7028 struct btrfs_root *root = BTRFS_I(inode)->root;
70c8a91c 7029 struct extent_map *em;
4b46fce2
JB		 7030 struct btrfs_key ins;
7031 u64 alloc_hint;
7032 int ret;
4b46fce2 7033
4b46fce2 7034 alloc_hint = get_extent_allocation_hint(inode, start, len);
00361589 7035 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
e570fd27 7036 alloc_hint, &ins, 1, 1);
00361589
JB		 7037 if (ret)
7038 return ERR_PTR(ret);
4b46fce2 7039
70c8a91c 7040 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
cc95bef6 7041 ins.offset, ins.offset, ins.offset, 0);
00361589 7042 if (IS_ERR(em)) {
e570fd27 7043 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
00361589
JB		 7044 return em;
7045 }
4b46fce2
JB		 7046
7047 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
7048 ins.offset, ins.offset, 0);
7049 if (ret) {
e570fd27 7050 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
00361589
JB		 7051 free_extent_map(em);
7052 return ERR_PTR(ret);
4b46fce2 7053 }
00361589 7054
4b46fce2
JB		 7055 return em;
7056}
7057
46bfbb5c
CM		 7058/*
7059 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7060 * block must be cow'd
7061 */
00361589 7062noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7ee9e440
JB		 7063 u64 *orig_start, u64 *orig_block_len,
7064 u64 *ram_bytes)
46bfbb5c 7065{
00361589 7066 struct btrfs_trans_handle *trans;
46bfbb5c
CM		 7067 struct btrfs_path *path;
7068 int ret;
7069 struct extent_buffer *leaf;
7070 struct btrfs_root *root = BTRFS_I(inode)->root;
7b2b7085 7071 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
46bfbb5c
CM		 7072 struct btrfs_file_extent_item *fi;
7073 struct btrfs_key key;
7074 u64 disk_bytenr;
7075 u64 backref_offset;
7076 u64 extent_end;
7077 u64 num_bytes;
7078 int slot;
7079 int found_type;
7ee9e440 7080 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
e77751aa 7081
46bfbb5c
CM		 7082 path = btrfs_alloc_path();
7083 if (!path)
7084 return -ENOMEM;
7085
00361589 7086 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
46bfbb5c
CM		 7087 offset, 0);
7088 if (ret < 0)
7089 goto out;
7090
7091 slot = path->slots[0];
7092 if (ret == 1) {
7093 if (slot == 0) {
7094 /* can't find the item, must cow */
7095 ret = 0;
7096 goto out;
7097 }
7098 slot--;
7099 }
7100 ret = 0;
7101 leaf = path->nodes[0];
7102 btrfs_item_key_to_cpu(leaf, &key, slot);
33345d01 7103 if (key.objectid != btrfs_ino(inode) ||
46bfbb5c
CM		 7104 key.type != BTRFS_EXTENT_DATA_KEY) {
7105 /* not our file or wrong item type, must cow */
7106 goto out;
7107 }
7108
7109 if (key.offset > offset) {
7110 /* Wrong offset, must cow */
7111 goto out;
7112 }
7113
7114 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7115 found_type = btrfs_file_extent_type(leaf, fi);
7116 if (found_type != BTRFS_FILE_EXTENT_REG &&
7117 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7118 /* not a regular extent, must cow */
7119 goto out;
7120 }
7ee9e440
JB		 7121
7122 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7123 goto out;
7124
e77751aa
MX		 7125 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7126 if (extent_end <= offset)
7127 goto out;
7128
46bfbb5c 7129 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7ee9e440
JB		 7130 if (disk_bytenr == 0)
7131 goto out;
7132
7133 if (btrfs_file_extent_compression(leaf, fi) ||
7134 btrfs_file_extent_encryption(leaf, fi) ||
7135 btrfs_file_extent_other_encoding(leaf, fi))
7136 goto out;
7137
46bfbb5c
CM		 7138 backref_offset = btrfs_file_extent_offset(leaf, fi);
7139
7ee9e440
JB		 7140 if (orig_start) {
7141 *orig_start = key.offset - backref_offset;
7142 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7143 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7144 }
eb384b55 7145
46bfbb5c
CM		 7146 if (btrfs_extent_readonly(root, disk_bytenr))
7147 goto out;
7b2b7085
MX		 7148
7149 num_bytes = min(offset + *len, extent_end) - offset;
7150 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7151 u64 range_end;
7152
7153 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7154 ret = test_range_bit(io_tree, offset, range_end,
7155 EXTENT_DELALLOC, 0, NULL);
7156 if (ret) {
7157 ret = -EAGAIN;
7158 goto out;
7159 }
7160 }
7161
1bda19eb 7162 btrfs_release_path(path);
46bfbb5c
CM		 7163
7164 /*
7165 * look for other files referencing this extent, if we
7166 * find any we must cow
7167 */
00361589
JB		 7168 trans = btrfs_join_transaction(root);
7169 if (IS_ERR(trans)) {
7170 ret = 0;
46bfbb5c 7171 goto out;
00361589
JB		 7172 }
7173
7174 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7175 key.offset - backref_offset, disk_bytenr);
7176 btrfs_end_transaction(trans, root);
7177 if (ret) {
7178 ret = 0;
7179 goto out;
7180 }
46bfbb5c
CM		 7181
7182 /*
7183 * adjust disk_bytenr and num_bytes to cover just the bytes
7184 * in this extent we are about to write. If there
7185 * are any csums in that range we have to cow in order
7186 * to keep the csums correct
7187 */
7188 disk_bytenr += backref_offset;
7189 disk_bytenr += offset - key.offset;
46bfbb5c
CM		 7190 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7191 goto out;
7192 /*
7193 * all of the above have passed, it is safe to overwrite this extent
7194 * without cow
7195 */
eb384b55 7196 *len = num_bytes;
46bfbb5c
CM		 7197 ret = 1;
7198out:
7199 btrfs_free_path(path);
7200 return ret;
7201}
7202
fc4adbff
AG		 7203bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7204{
7205 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7206 int found = false;
7207 void **pagep = NULL;
7208 struct page *page = NULL;
7209 int start_idx;
7210 int end_idx;
7211
7212 start_idx = start >> PAGE_CACHE_SHIFT;
7213
7214 /*
7215 * end is the last byte in the last page. end == start is legal
7216 */
7217 end_idx = end >> PAGE_CACHE_SHIFT;
7218
7219 rcu_read_lock();
7220
7221 /* Most of the code in this while loop is lifted from
7222 * find_get_page. It's been modified to begin searching from a
7223 * page and return just the first page found in that range. If the
7224 * found idx is less than or equal to the end idx then we know that
7225 * a page exists. If no pages are found or if those pages are
7226 * outside of the range then we're fine (yay!) */
7227 while (page == NULL &&
7228 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7229 page = radix_tree_deref_slot(pagep);
7230 if (unlikely(!page))
7231 break;
7232
7233 if (radix_tree_exception(page)) {
809f9016
FM		 7234 if (radix_tree_deref_retry(page)) {
7235 page = NULL;
fc4adbff 7236 continue;
809f9016 7237 }
fc4adbff
AG		 7238 /*
7239 * Otherwise, shmem/tmpfs must be storing a swap entry
7240 * here as an exceptional entry: so return it without
7241 * attempting to raise page count.
7242 */
6fdef6d4 7243 page = NULL;
fc4adbff
AG		 7244 break; /* TODO: Is this relevant for this use case? */
7245 }
7246
91405151
FM		 7247 if (!page_cache_get_speculative(page)) {
7248 page = NULL;
fc4adbff 7249 continue;
91405151 7250 }
fc4adbff
AG		 7251
7252 /*
7253 * Has the page moved?
7254 * This is part of the lockless pagecache protocol. See
7255 * include/linux/pagemap.h for details.
7256 */
7257 if (unlikely(page != *pagep)) {
7258 page_cache_release(page);
7259 page = NULL;
7260 }
7261 }
7262
7263 if (page) {
7264 if (page->index <= end_idx)
7265 found = true;
7266 page_cache_release(page);
7267 }
7268
7269 rcu_read_unlock();
7270 return found;
7271}
7272
eb838e73
JB		 7273static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7274 struct extent_state **cached_state, int writing)
7275{
7276 struct btrfs_ordered_extent *ordered;
7277 int ret = 0;
7278
7279 while (1) {
7280 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7281 0, cached_state);
7282 /*
7283 * We're concerned with the entire range that we're going to be
7284 * doing DIO to, so we need to make sure theres no ordered
7285 * extents in this range.
7286 */
7287 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7288 lockend - lockstart + 1);
7289
7290 /*
7291 * We need to make sure there are no buffered pages in this
7292 * range either, we could have raced between the invalidate in
7293 * generic_file_direct_write and locking the extent. The
7294 * invalidate needs to happen so that reads after a write do not
7295 * get stale data.
7296 */
fc4adbff
AG		 7297 if (!ordered &&
7298 (!writing ||
7299 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
eb838e73
JB		 7300 break;
7301
7302 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7303 cached_state, GFP_NOFS);
7304
7305 if (ordered) {
7306 btrfs_start_ordered_extent(inode, ordered, 1);
7307 btrfs_put_ordered_extent(ordered);
7308 } else {
7309 /* Screw you mmap */
728404da 7310 ret = btrfs_fdatawrite_range(inode, lockstart, lockend);
075bdbdb
FM		 7311 if (ret)
7312 break;
7313 ret = filemap_fdatawait_range(inode->i_mapping,
7314 lockstart,
7315 lockend);
eb838e73
JB		 7316 if (ret)
7317 break;
7318
7319 /*
7320 * If we found a page that couldn't be invalidated just
7321 * fall back to buffered.
7322 */
7323 ret = invalidate_inode_pages2_range(inode->i_mapping,
7324 lockstart >> PAGE_CACHE_SHIFT,
7325 lockend >> PAGE_CACHE_SHIFT);
7326 if (ret)
7327 break;
7328 }
7329
7330 cond_resched();
7331 }
7332
7333 return ret;
7334}
7335
69ffb543
JB		 7336static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7337 u64 len, u64 orig_start,
7338 u64 block_start, u64 block_len,
cc95bef6
JB		 7339 u64 orig_block_len, u64 ram_bytes,
7340 int type)
69ffb543
JB		 7341{
7342 struct extent_map_tree *em_tree;
7343 struct extent_map *em;
7344 struct btrfs_root *root = BTRFS_I(inode)->root;
7345 int ret;
7346
7347 em_tree = &BTRFS_I(inode)->extent_tree;
7348 em = alloc_extent_map();
7349 if (!em)
7350 return ERR_PTR(-ENOMEM);
7351
7352 em->start = start;
7353 em->orig_start = orig_start;
2ab28f32
JB		 7354 em->mod_start = start;
7355 em->mod_len = len;
69ffb543
JB		 7356 em->len = len;
7357 em->block_len = block_len;
7358 em->block_start = block_start;
7359 em->bdev = root->fs_info->fs_devices->latest_bdev;
b4939680 7360 em->orig_block_len = orig_block_len;
cc95bef6 7361 em->ram_bytes = ram_bytes;
70c8a91c 7362 em->generation = -1;
69ffb543
JB		 7363 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7364 if (type == BTRFS_ORDERED_PREALLOC)
b11e234d 7365 set_bit(EXTENT_FLAG_FILLING, &em->flags);
69ffb543
JB		 7366
7367 do {
7368 btrfs_drop_extent_cache(inode, em->start,
7369 em->start + em->len - 1, 0);
7370 write_lock(&em_tree->lock);
09a2a8f9 7371 ret = add_extent_mapping(em_tree, em, 1);
69ffb543
JB		 7372 write_unlock(&em_tree->lock);
7373 } while (ret == -EEXIST);
7374
7375 if (ret) {
7376 free_extent_map(em);
7377 return ERR_PTR(ret);
7378 }
7379
7380 return em;
7381}
7382
7383
4b46fce2
JB		 7384static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7385 struct buffer_head *bh_result, int create)
7386{
7387 struct extent_map *em;
7388 struct btrfs_root *root = BTRFS_I(inode)->root;
eb838e73 7389 struct extent_state *cached_state = NULL;
4b46fce2 7390 u64 start = iblock << inode->i_blkbits;
eb838e73 7391 u64 lockstart, lockend;
4b46fce2 7392 u64 len = bh_result->b_size;
e1cbbfa5 7393 u64 *outstanding_extents = NULL;
eb838e73 7394 int unlock_bits = EXTENT_LOCKED;
0934856d 7395 int ret = 0;
eb838e73 7396
172a5049 7397 if (create)
3266789f 7398 unlock_bits |= EXTENT_DIRTY;
172a5049 7399 else
c329861d 7400 len = min_t(u64, len, root->sectorsize);
eb838e73 7401
c329861d
JB		 7402 lockstart = start;
7403 lockend = start + len - 1;
7404
e1cbbfa5
JB		 7405 if (current->journal_info) {
7406 /*
7407 * Need to pull our outstanding extents and set journal_info to NULL so
7408 * that anything that needs to check if there's a transction doesn't get
7409 * confused.
7410 */
7411 outstanding_extents = current->journal_info;
7412 current->journal_info = NULL;
7413 }
7414
eb838e73
JB		 7415 /*
7416 * If this errors out it's because we couldn't invalidate pagecache for
7417 * this range and we need to fallback to buffered.
7418 */
7419 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
7420 return -ENOTBLK;
7421
4b46fce2 7422 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
eb838e73
JB		 7423 if (IS_ERR(em)) {
7424 ret = PTR_ERR(em);
7425 goto unlock_err;
7426 }
4b46fce2
JB		 7427
7428 /*
7429 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7430 * io. INLINE is special, and we could probably kludge it in here, but
7431 * it's still buffered so for safety lets just fall back to the generic
7432 * buffered path.
7433 *
7434 * For COMPRESSED we _have_ to read the entire extent in so we can
7435 * decompress it, so there will be buffering required no matter what we
7436 * do, so go ahead and fallback to buffered.
7437 *
7438 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7439 * to buffered IO. Don't blame me, this is the price we pay for using
7440 * the generic code.
7441 */
7442 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7443 em->block_start == EXTENT_MAP_INLINE) {
7444 free_extent_map(em);
eb838e73
JB		 7445 ret = -ENOTBLK;
7446 goto unlock_err;
4b46fce2
JB		 7447 }
7448
7449 /* Just a good old fashioned hole, return */
7450 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7451 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7452 free_extent_map(em);
eb838e73 7453 goto unlock_err;
4b46fce2
JB		 7454 }
7455
7456 /*
7457 * We don't allocate a new extent in the following cases
7458 *
7459 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7460 * existing extent.
7461 * 2) The extent is marked as PREALLOC. We're good to go here and can
7462 * just use the extent.
7463 *
7464 */
46bfbb5c 7465 if (!create) {
eb838e73
JB		 7466 len = min(len, em->len - (start - em->start));
7467 lockstart = start + len;
7468 goto unlock;
46bfbb5c 7469 }
4b46fce2
JB		 7470
7471 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7472 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7473 em->block_start != EXTENT_MAP_HOLE)) {
4b46fce2 7474 int type;
eb384b55 7475 u64 block_start, orig_start, orig_block_len, ram_bytes;
4b46fce2
JB		 7476
7477 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7478 type = BTRFS_ORDERED_PREALLOC;
7479 else
7480 type = BTRFS_ORDERED_NOCOW;
46bfbb5c 7481 len = min(len, em->len - (start - em->start));
4b46fce2 7482 block_start = em->block_start + (start - em->start);
46bfbb5c 7483
00361589 7484 if (can_nocow_extent(inode, start, &len, &orig_start,
7ee9e440 7485 &orig_block_len, &ram_bytes) == 1) {
69ffb543
JB		 7486 if (type == BTRFS_ORDERED_PREALLOC) {
7487 free_extent_map(em);
7488 em = create_pinned_em(inode, start, len,
7489 orig_start,
b4939680 7490 block_start, len,
cc95bef6
JB		 7491 orig_block_len,
7492 ram_bytes, type);
555e1286
FM		 7493 if (IS_ERR(em)) {
7494 ret = PTR_ERR(em);
69ffb543 7495 goto unlock_err;
555e1286 7496 }
69ffb543
JB		 7497 }
7498
46bfbb5c
CM		 7499 ret = btrfs_add_ordered_extent_dio(inode, start,
7500 block_start, len, len, type);
46bfbb5c
CM		 7501 if (ret) {
7502 free_extent_map(em);
eb838e73 7503 goto unlock_err;
46bfbb5c
CM		 7504 }
7505 goto unlock;
4b46fce2 7506 }
4b46fce2 7507 }
00361589 7508
46bfbb5c
CM		 7509 /*
7510 * this will cow the extent, reset the len in case we changed
7511 * it above
7512 */
7513 len = bh_result->b_size;
70c8a91c
JB		 7514 free_extent_map(em);
7515 em = btrfs_new_extent_direct(inode, start, len);
eb838e73
JB		 7516 if (IS_ERR(em)) {
7517 ret = PTR_ERR(em);
7518 goto unlock_err;
7519 }
46bfbb5c
CM		 7520 len = min(len, em->len - (start - em->start));
7521unlock:
4b46fce2
JB		 7522 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7523 inode->i_blkbits;
46bfbb5c 7524 bh_result->b_size = len;
4b46fce2
JB		 7525 bh_result->b_bdev = em->bdev;
7526 set_buffer_mapped(bh_result);
c3473e83
JB		 7527 if (create) {
7528 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7529 set_buffer_new(bh_result);
7530
7531 /*
7532 * Need to update the i_size under the extent lock so buffered
7533 * readers will get the updated i_size when we unlock.
7534 */
7535 if (start + len > i_size_read(inode))
7536 i_size_write(inode, start + len);
0934856d 7537
e1cbbfa5
JB		 7538 /*
7539 * If we have an outstanding_extents count still set then we're
7540 * within our reservation, otherwise we need to adjust our inode
7541 * counter appropriately.
7542 */
7543 if (*outstanding_extents) {
7544 (*outstanding_extents)--;
7545 } else {
3e05bde8
JB		 7546 spin_lock(&BTRFS_I(inode)->lock);
7547 BTRFS_I(inode)->outstanding_extents++;
7548 spin_unlock(&BTRFS_I(inode)->lock);
7549 }
e1cbbfa5
JB		 7550
7551 current->journal_info = outstanding_extents;
3266789f 7552 btrfs_free_reserved_data_space(inode, len);
ddba1bfc 7553 set_bit(BTRFS_INODE_DIO_READY, &BTRFS_I(inode)->runtime_flags);
c3473e83 7554 }
4b46fce2 7555
eb838e73
JB		 7556 /*
7557 * In the case of write we need to clear and unlock the entire range,
7558 * in the case of read we need to unlock only the end area that we
7559 * aren't using if there is any left over space.
7560 */
24c03fa5 7561 if (lockstart < lockend) {
0934856d
MX		 7562 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7563 lockend, unlock_bits, 1, 0,
7564 &cached_state, GFP_NOFS);
24c03fa5 7565 } else {
eb838e73 7566 free_extent_state(cached_state);
24c03fa5 7567 }
eb838e73 7568
4b46fce2
JB		 7569 free_extent_map(em);
7570
7571 return 0;
eb838e73
JB		 7572
7573unlock_err:
eb838e73
JB		 7574 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7575 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
e1cbbfa5
JB		 7576 if (outstanding_extents)
7577 current->journal_info = outstanding_extents;
eb838e73 7578 return ret;
4b46fce2
JB		 7579}
7580
8b110e39
MX		 7581static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7582 int rw, int mirror_num)
7583{
7584 struct btrfs_root *root = BTRFS_I(inode)->root;
7585 int ret;
7586
7587 BUG_ON(rw & REQ_WRITE);
7588
7589 bio_get(bio);
7590
7591 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7592 BTRFS_WQ_ENDIO_DIO_REPAIR);
7593 if (ret)
7594 goto err;
7595
7596 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7597err:
7598 bio_put(bio);
7599 return ret;
7600}
7601
7602static int btrfs_check_dio_repairable(struct inode *inode,
7603 struct bio *failed_bio,
7604 struct io_failure_record *failrec,
7605 int failed_mirror)
7606{
7607 int num_copies;
7608
7609 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7610 failrec->logical, failrec->len);
7611 if (num_copies == 1) {
7612 /*
7613 * we only have a single copy of the data, so don't bother with
7614 * all the retry and error correction code that follows. no
7615 * matter what the error is, it is very likely to persist.
7616 */
7617 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7618 num_copies, failrec->this_mirror, failed_mirror);
7619 return 0;
7620 }
7621
7622 failrec->failed_mirror = failed_mirror;
7623 failrec->this_mirror++;
7624 if (failrec->this_mirror == failed_mirror)
7625 failrec->this_mirror++;
7626
7627 if (failrec->this_mirror > num_copies) {
7628 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7629 num_copies, failrec->this_mirror, failed_mirror);
7630 return 0;
7631 }
7632
7633 return 1;
7634}
7635
7636static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7637 struct page *page, u64 start, u64 end,
7638 int failed_mirror, bio_end_io_t *repair_endio,
7639 void *repair_arg)
7640{
7641 struct io_failure_record *failrec;
7642 struct bio *bio;
7643 int isector;
7644 int read_mode;
7645 int ret;
7646
7647 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7648
7649 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7650 if (ret)
7651 return ret;
7652
7653 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7654 failed_mirror);
7655 if (!ret) {
7656 free_io_failure(inode, failrec);
7657 return -EIO;
7658 }
7659
7660 if (failed_bio->bi_vcnt > 1)
7661 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7662 else
7663 read_mode = READ_SYNC;
7664
7665 isector = start - btrfs_io_bio(failed_bio)->logical;
7666 isector >>= inode->i_sb->s_blocksize_bits;
7667 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7668 0, isector, repair_endio, repair_arg);
7669 if (!bio) {
7670 free_io_failure(inode, failrec);
7671 return -EIO;
7672 }
7673
7674 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7675 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7676 read_mode, failrec->this_mirror, failrec->in_validation);
7677
7678 ret = submit_dio_repair_bio(inode, bio, read_mode,
7679 failrec->this_mirror);
7680 if (ret) {
7681 free_io_failure(inode, failrec);
7682 bio_put(bio);
7683 }
7684
7685 return ret;
7686}
7687
7688struct btrfs_retry_complete {
7689 struct completion done;
7690 struct inode *inode;
7691 u64 start;
7692 int uptodate;
7693};
7694
4246a0b6 7695static void btrfs_retry_endio_nocsum(struct bio *bio)
8b110e39
MX		 7696{
7697 struct btrfs_retry_complete *done = bio->bi_private;
7698 struct bio_vec *bvec;
7699 int i;
7700
4246a0b6 7701 if (bio->bi_error)
8b110e39
MX		 7702 goto end;
7703
7704 done->uptodate = 1;
7705 bio_for_each_segment_all(bvec, bio, i)
7706 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7707end:
7708 complete(&done->done);
7709 bio_put(bio);
7710}
7711
7712static int __btrfs_correct_data_nocsum(struct inode *inode,
7713 struct btrfs_io_bio *io_bio)
4b46fce2 7714{
2c30c71b 7715 struct bio_vec *bvec;
8b110e39 7716 struct btrfs_retry_complete done;
4b46fce2 7717 u64 start;
2c30c71b 7718 int i;
c1dc0896 7719 int ret;
4b46fce2 7720
8b110e39
MX		 7721 start = io_bio->logical;
7722 done.inode = inode;
7723
7724 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7725try_again:
7726 done.uptodate = 0;
7727 done.start = start;
7728 init_completion(&done.done);
7729
7730 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7731 start + bvec->bv_len - 1,
7732 io_bio->mirror_num,
7733 btrfs_retry_endio_nocsum, &done);
7734 if (ret)
7735 return ret;
7736
7737 wait_for_completion(&done.done);
7738
7739 if (!done.uptodate) {
7740 /* We might have another mirror, so try again */
7741 goto try_again;
7742 }
7743
7744 start += bvec->bv_len;
7745 }
7746
7747 return 0;
7748}
7749
4246a0b6 7750static void btrfs_retry_endio(struct bio *bio)
8b110e39
MX		 7751{
7752 struct btrfs_retry_complete *done = bio->bi_private;
7753 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7754 struct bio_vec *bvec;
7755 int uptodate;
7756 int ret;
7757 int i;
7758
4246a0b6 7759 if (bio->bi_error)
8b110e39
MX		 7760 goto end;
7761
7762 uptodate = 1;
7763 bio_for_each_segment_all(bvec, bio, i) {
7764 ret = __readpage_endio_check(done->inode, io_bio, i,
7765 bvec->bv_page, 0,
7766 done->start, bvec->bv_len);
7767 if (!ret)
7768 clean_io_failure(done->inode, done->start,
7769 bvec->bv_page, 0);
7770 else
7771 uptodate = 0;
7772 }
7773
7774 done->uptodate = uptodate;
7775end:
7776 complete(&done->done);
7777 bio_put(bio);
7778}
7779
7780static int __btrfs_subio_endio_read(struct inode *inode,
7781 struct btrfs_io_bio *io_bio, int err)
7782{
7783 struct bio_vec *bvec;
7784 struct btrfs_retry_complete done;
7785 u64 start;
7786 u64 offset = 0;
7787 int i;
7788 int ret;
dc380aea 7789
8b110e39 7790 err = 0;
c1dc0896 7791 start = io_bio->logical;
8b110e39
MX		 7792 done.inode = inode;
7793
c1dc0896 7794 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
dc380aea
MX		 7795 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7796 0, start, bvec->bv_len);
8b110e39
MX		 7797 if (likely(!ret))
7798 goto next;
7799try_again:
7800 done.uptodate = 0;
7801 done.start = start;
7802 init_completion(&done.done);
7803
7804 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7805 start + bvec->bv_len - 1,
7806 io_bio->mirror_num,
7807 btrfs_retry_endio, &done);
7808 if (ret) {
7809 err = ret;
7810 goto next;
7811 }
7812
7813 wait_for_completion(&done.done);
7814
7815 if (!done.uptodate) {
7816 /* We might have another mirror, so try again */
7817 goto try_again;
7818 }
7819next:
7820 offset += bvec->bv_len;
4b46fce2 7821 start += bvec->bv_len;
2c30c71b 7822 }
c1dc0896
MX		 7823
7824 return err;
7825}
7826
8b110e39
MX		 7827static int btrfs_subio_endio_read(struct inode *inode,
7828 struct btrfs_io_bio *io_bio, int err)
7829{
7830 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7831
7832 if (skip_csum) {
7833 if (unlikely(err))
7834 return __btrfs_correct_data_nocsum(inode, io_bio);
7835 else
7836 return 0;
7837 } else {
7838 return __btrfs_subio_endio_read(inode, io_bio, err);
7839 }
7840}
7841
4246a0b6 7842static void btrfs_endio_direct_read(struct bio *bio)
c1dc0896
MX		 7843{
7844 struct btrfs_dio_private *dip = bio->bi_private;
7845 struct inode *inode = dip->inode;
7846 struct bio *dio_bio;
7847 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
4246a0b6 7848 int err = bio->bi_error;
c1dc0896 7849
8b110e39
MX		 7850 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
7851 err = btrfs_subio_endio_read(inode, io_bio, err);
c1dc0896 7852
4b46fce2 7853 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
d0082371 7854 dip->logical_offset + dip->bytes - 1);
9be3395b 7855 dio_bio = dip->dio_bio;
4b46fce2 7856
4b46fce2 7857 kfree(dip);
c0da7aa1 7858
4246a0b6 7859 dio_end_io(dio_bio, bio->bi_error);
23ea8e5a
MX		 7860
7861 if (io_bio->end_io)
7862 io_bio->end_io(io_bio, err);
9be3395b 7863 bio_put(bio);
4b46fce2
JB		 7864}
7865
4246a0b6 7866static void btrfs_endio_direct_write(struct bio *bio)
4b46fce2
JB		 7867{
7868 struct btrfs_dio_private *dip = bio->bi_private;
7869 struct inode *inode = dip->inode;
7870 struct btrfs_root *root = BTRFS_I(inode)->root;
4b46fce2 7871 struct btrfs_ordered_extent *ordered = NULL;
163cf09c
CM		 7872 u64 ordered_offset = dip->logical_offset;
7873 u64 ordered_bytes = dip->bytes;
9be3395b 7874 struct bio *dio_bio;
4b46fce2
JB		 7875 int ret;
7876
163cf09c
CM		 7877again:
7878 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7879 &ordered_offset,
4246a0b6
CH		 7880 ordered_bytes,
7881 !bio->bi_error);
4b46fce2 7882 if (!ret)
163cf09c 7883 goto out_test;
4b46fce2 7884
9e0af237
LB		 7885 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
7886 finish_ordered_fn, NULL, NULL);
fccb5d86
QW		 7887 btrfs_queue_work(root->fs_info->endio_write_workers,
7888 &ordered->work);
163cf09c
CM		 7889out_test:
7890 /*
7891 * our bio might span multiple ordered extents. If we haven't
7892 * completed the accounting for the whole dio, go back and try again
7893 */
7894 if (ordered_offset < dip->logical_offset + dip->bytes) {
7895 ordered_bytes = dip->logical_offset + dip->bytes -
7896 ordered_offset;
5fd02043 7897 ordered = NULL;
163cf09c
CM		 7898 goto again;
7899 }
9be3395b 7900 dio_bio = dip->dio_bio;
4b46fce2 7901
4b46fce2 7902 kfree(dip);
c0da7aa1 7903
4246a0b6 7904 dio_end_io(dio_bio, bio->bi_error);
9be3395b 7905 bio_put(bio);
4b46fce2
JB		 7906}
7907
eaf25d93
CM		 7908static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7909 struct bio *bio, int mirror_num,
7910 unsigned long bio_flags, u64 offset)
7911{
7912 int ret;
7913 struct btrfs_root *root = BTRFS_I(inode)->root;
7914 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
79787eaa 7915 BUG_ON(ret); /* -ENOMEM */
eaf25d93
CM		 7916 return 0;
7917}
7918
4246a0b6 7919static void btrfs_end_dio_bio(struct bio *bio)
e65e1535
MX		 7920{
7921 struct btrfs_dio_private *dip = bio->bi_private;
4246a0b6 7922 int err = bio->bi_error;
e65e1535 7923
8b110e39
MX		 7924 if (err)
7925 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7926 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7927 btrfs_ino(dip->inode), bio->bi_rw,
7928 (unsigned long long)bio->bi_iter.bi_sector,
7929 bio->bi_iter.bi_size, err);
7930
7931 if (dip->subio_endio)
7932 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
c1dc0896
MX		 7933
7934 if (err) {
e65e1535
MX		 7935 dip->errors = 1;
7936
7937 /*
7938 * before atomic variable goto zero, we must make sure
7939 * dip->errors is perceived to be set.
7940 */
4e857c58 7941 smp_mb__before_atomic();
e65e1535
MX		 7942 }
7943
7944 /* if there are more bios still pending for this dio, just exit */
7945 if (!atomic_dec_and_test(&dip->pending_bios))
7946 goto out;
7947
9be3395b 7948 if (dip->errors) {
e65e1535 7949 bio_io_error(dip->orig_bio);
9be3395b 7950 } else {
4246a0b6
CH		 7951 dip->dio_bio->bi_error = 0;
7952 bio_endio(dip->orig_bio);
e65e1535
MX		 7953 }
7954out:
7955 bio_put(bio);
7956}
7957
7958static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7959 u64 first_sector, gfp_t gfp_flags)
7960{
b54ffb73 7961 return btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
e65e1535
MX		 7962}
7963
c1dc0896
MX		 7964static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
7965 struct inode *inode,
7966 struct btrfs_dio_private *dip,
7967 struct bio *bio,
7968 u64 file_offset)
7969{
7970 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7971 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
7972 int ret;
7973
7974 /*
7975 * We load all the csum data we need when we submit
7976 * the first bio to reduce the csum tree search and
7977 * contention.
7978 */
7979 if (dip->logical_offset == file_offset) {
7980 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
7981 file_offset);
7982 if (ret)
7983 return ret;
7984 }
7985
7986 if (bio == dip->orig_bio)
7987 return 0;
7988
7989 file_offset -= dip->logical_offset;
7990 file_offset >>= inode->i_sb->s_blocksize_bits;
7991 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
7992
7993 return 0;
7994}
7995
e65e1535
MX		 7996static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7997 int rw, u64 file_offset, int skip_sum,
c329861d 7998 int async_submit)
e65e1535 7999{
facc8a22 8000 struct btrfs_dio_private *dip = bio->bi_private;
e65e1535
MX		 8001 int write = rw & REQ_WRITE;
8002 struct btrfs_root *root = BTRFS_I(inode)->root;
8003 int ret;
8004
b812ce28
JB		 8005 if (async_submit)
8006 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8007
e65e1535 8008 bio_get(bio);
5fd02043
JB		 8009
8010 if (!write) {
bfebd8b5
DS		 8011 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8012 BTRFS_WQ_ENDIO_DATA);
5fd02043
JB		 8013 if (ret)
8014 goto err;
8015 }
e65e1535 8016
1ae39938
JB		 8017 if (skip_sum)
8018 goto map;
8019
8020 if (write && async_submit) {
e65e1535
MX		 8021 ret = btrfs_wq_submit_bio(root->fs_info,
8022 inode, rw, bio, 0, 0,
8023 file_offset,
8024 __btrfs_submit_bio_start_direct_io,
8025 __btrfs_submit_bio_done);
8026 goto err;
1ae39938
JB		 8027 } else if (write) {
8028 /*
8029 * If we aren't doing async submit, calculate the csum of the
8030 * bio now.
8031 */
8032 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8033 if (ret)
8034 goto err;
23ea8e5a 8035 } else {
c1dc0896
MX		 8036 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8037 file_offset);
c2db1073
TI		 8038 if (ret)
8039 goto err;
8040 }
1ae39938
JB		 8041map:
8042 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
e65e1535
MX		 8043err:
8044 bio_put(bio);
8045 return ret;
8046}
8047
8048static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
8049 int skip_sum)
8050{
8051 struct inode *inode = dip->inode;
8052 struct btrfs_root *root = BTRFS_I(inode)->root;
e65e1535
MX		 8053 struct bio *bio;
8054 struct bio *orig_bio = dip->orig_bio;
8055 struct bio_vec *bvec = orig_bio->bi_io_vec;
4f024f37 8056 u64 start_sector = orig_bio->bi_iter.bi_sector;
e65e1535
MX		 8057 u64 file_offset = dip->logical_offset;
8058 u64 submit_len = 0;
8059 u64 map_length;
8060 int nr_pages = 0;
23ea8e5a 8061 int ret;
1ae39938 8062 int async_submit = 0;
e65e1535 8063
4f024f37 8064 map_length = orig_bio->bi_iter.bi_size;
53b381b3 8065 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
e65e1535 8066 &map_length, NULL, 0);
7a5c3c9b 8067 if (ret)
e65e1535 8068 return -EIO;
facc8a22 8069
4f024f37 8070 if (map_length >= orig_bio->bi_iter.bi_size) {
02f57c7a 8071 bio = orig_bio;
c1dc0896 8072 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
02f57c7a
JB		 8073 goto submit;
8074 }
8075
53b381b3 8076 /* async crcs make it difficult to collect full stripe writes. */
ffe2d203 8077 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
53b381b3
DW		 8078 async_submit = 0;
8079 else
8080 async_submit = 1;
8081
02f57c7a
JB		 8082 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8083 if (!bio)
8084 return -ENOMEM;
7a5c3c9b 8085
02f57c7a
JB		 8086 bio->bi_private = dip;
8087 bio->bi_end_io = btrfs_end_dio_bio;
c1dc0896 8088 btrfs_io_bio(bio)->logical = file_offset;
02f57c7a
JB		 8089 atomic_inc(&dip->pending_bios);
8090
e65e1535 8091 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
ee39b432 8092 if (map_length < submit_len + bvec->bv_len ||
e65e1535 8093 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
ee39b432 8094 bvec->bv_offset) < bvec->bv_len) {
e65e1535
MX		 8095 /*
8096 * inc the count before we submit the bio so
8097 * we know the end IO handler won't happen before
8098 * we inc the count. Otherwise, the dip might get freed
8099 * before we're done setting it up
8100 */
8101 atomic_inc(&dip->pending_bios);
8102 ret = __btrfs_submit_dio_bio(bio, inode, rw,
8103 file_offset, skip_sum,
c329861d 8104 async_submit);
e65e1535
MX		 8105 if (ret) {
8106 bio_put(bio);
8107 atomic_dec(&dip->pending_bios);
8108 goto out_err;
8109 }
8110
e65e1535
MX		 8111 start_sector += submit_len >> 9;
8112 file_offset += submit_len;
8113
8114 submit_len = 0;
8115 nr_pages = 0;
8116
8117 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8118 start_sector, GFP_NOFS);
8119 if (!bio)
8120 goto out_err;
8121 bio->bi_private = dip;
8122 bio->bi_end_io = btrfs_end_dio_bio;
c1dc0896 8123 btrfs_io_bio(bio)->logical = file_offset;
e65e1535 8124
4f024f37 8125 map_length = orig_bio->bi_iter.bi_size;
53b381b3 8126 ret = btrfs_map_block(root->fs_info, rw,
3ec706c8 8127 start_sector << 9,
e65e1535
MX		 8128 &map_length, NULL, 0);
8129 if (ret) {
8130 bio_put(bio);
8131 goto out_err;
8132 }
8133 } else {
8134 submit_len += bvec->bv_len;
67871254 8135 nr_pages++;
e65e1535
MX		 8136 bvec++;
8137 }
8138 }
8139
02f57c7a 8140submit:
e65e1535 8141 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
c329861d 8142 async_submit);
e65e1535
MX		 8143 if (!ret)
8144 return 0;
8145
8146 bio_put(bio);
8147out_err:
8148 dip->errors = 1;
8149 /*
8150 * before atomic variable goto zero, we must
8151 * make sure dip->errors is perceived to be set.
8152 */
4e857c58 8153 smp_mb__before_atomic();
e65e1535
MX		 8154 if (atomic_dec_and_test(&dip->pending_bios))
8155 bio_io_error(dip->orig_bio);
8156
8157 /* bio_end_io() will handle error, so we needn't return it */
8158 return 0;
8159}
8160
9be3395b
CM		 8161static void btrfs_submit_direct(int rw, struct bio *dio_bio,
8162 struct inode *inode, loff_t file_offset)
4b46fce2 8163{
61de718f
FM		 8164 struct btrfs_dio_private *dip = NULL;
8165 struct bio *io_bio = NULL;
23ea8e5a 8166 struct btrfs_io_bio *btrfs_bio;
4b46fce2 8167 int skip_sum;
7b6d91da 8168 int write = rw & REQ_WRITE;
4b46fce2
JB		 8169 int ret = 0;
8170
8171 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8172
9be3395b 8173 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
9be3395b
CM		 8174 if (!io_bio) {
8175 ret = -ENOMEM;
8176 goto free_ordered;
8177 }
8178
c1dc0896 8179 dip = kzalloc(sizeof(*dip), GFP_NOFS);
4b46fce2
JB		 8180 if (!dip) {
8181 ret = -ENOMEM;
61de718f 8182 goto free_ordered;
4b46fce2 8183 }
4b46fce2 8184
9be3395b 8185 dip->private = dio_bio->bi_private;
4b46fce2
JB		 8186 dip->inode = inode;
8187 dip->logical_offset = file_offset;
4f024f37
KO		 8188 dip->bytes = dio_bio->bi_iter.bi_size;
8189 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
9be3395b 8190 io_bio->bi_private = dip;
9be3395b
CM		 8191 dip->orig_bio = io_bio;
8192 dip->dio_bio = dio_bio;
e65e1535 8193 atomic_set(&dip->pending_bios, 0);
c1dc0896
MX		 8194 btrfs_bio = btrfs_io_bio(io_bio);
8195 btrfs_bio->logical = file_offset;
4b46fce2 8196
c1dc0896 8197 if (write) {
9be3395b 8198 io_bio->bi_end_io = btrfs_endio_direct_write;
c1dc0896 8199 } else {
9be3395b 8200 io_bio->bi_end_io = btrfs_endio_direct_read;
c1dc0896
MX		 8201 dip->subio_endio = btrfs_subio_endio_read;
8202 }
4b46fce2 8203
e65e1535
MX		 8204 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8205 if (!ret)
eaf25d93 8206 return;
9be3395b 8207
23ea8e5a
MX		 8208 if (btrfs_bio->end_io)
8209 btrfs_bio->end_io(btrfs_bio, ret);
9be3395b 8210
4b46fce2
JB		 8211free_ordered:
8212 /*
61de718f
FM		 8213 * If we arrived here it means either we failed to submit the dip
8214 * or we either failed to clone the dio_bio or failed to allocate the
8215 * dip. If we cloned the dio_bio and allocated the dip, we can just
8216 * call bio_endio against our io_bio so that we get proper resource
8217 * cleanup if we fail to submit the dip, otherwise, we must do the
8218 * same as btrfs_endio_direct_[write|read] because we can't call these
8219 * callbacks - they require an allocated dip and a clone of dio_bio.
4b46fce2 8220 */
61de718f 8221 if (io_bio && dip) {
4246a0b6
CH		 8222 io_bio->bi_error = -EIO;
8223 bio_endio(io_bio);
61de718f
FM		 8224 /*
8225 * The end io callbacks free our dip, do the final put on io_bio
8226 * and all the cleanup and final put for dio_bio (through
8227 * dio_end_io()).
8228 */
8229 dip = NULL;
8230 io_bio = NULL;
8231 } else {
8232 if (write) {
8233 struct btrfs_ordered_extent *ordered;
8234
8235 ordered = btrfs_lookup_ordered_extent(inode,
8236 file_offset);
8237 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
8238 /*
8239 * Decrements our ref on the ordered extent and removes
8240 * the ordered extent from the inode's ordered tree,
8241 * doing all the proper resource cleanup such as for the
8242 * reserved space and waking up any waiters for this
8243 * ordered extent (through btrfs_remove_ordered_extent).
8244 */
8245 btrfs_finish_ordered_io(ordered);
8246 } else {
8247 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8248 file_offset + dio_bio->bi_iter.bi_size - 1);
8249 }
4246a0b6 8250 dio_bio->bi_error = -EIO;
61de718f
FM		 8251 /*
8252 * Releases and cleans up our dio_bio, no need to bio_put()
8253 * nor bio_endio()/bio_io_error() against dio_bio.
8254 */
8255 dio_end_io(dio_bio, ret);
4b46fce2 8256 }
61de718f
FM		 8257 if (io_bio)
8258 bio_put(io_bio);
8259 kfree(dip);
4b46fce2
JB		 8260}
8261
6f673763 8262static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
28060d5d 8263 const struct iov_iter *iter, loff_t offset)
5a5f79b5
CM		 8264{
8265 int seg;
a1b75f7d 8266 int i;
5a5f79b5
CM		 8267 unsigned blocksize_mask = root->sectorsize - 1;
8268 ssize_t retval = -EINVAL;
5a5f79b5
CM		 8269
8270 if (offset & blocksize_mask)
8271 goto out;
8272
28060d5d
AV		 8273 if (iov_iter_alignment(iter) & blocksize_mask)
8274 goto out;
a1b75f7d 8275
28060d5d 8276 /* If this is a write we don't need to check anymore */
6f673763 8277 if (iov_iter_rw(iter) == WRITE)
28060d5d
AV		 8278 return 0;
8279 /*
8280 * Check to make sure we don't have duplicate iov_base's in this
8281 * iovec, if so return EINVAL, otherwise we'll get csum errors
8282 * when reading back.
8283 */
8284 for (seg = 0; seg < iter->nr_segs; seg++) {
8285 for (i = seg + 1; i < iter->nr_segs; i++) {
8286 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
a1b75f7d
JB		 8287 goto out;
8288 }
5a5f79b5
CM		 8289 }
8290 retval = 0;
8291out:
8292 return retval;
8293}
eb838e73 8294
22c6186e
OS		 8295static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
8296 loff_t offset)
16432985 8297{
4b46fce2
JB		 8298 struct file *file = iocb->ki_filp;
8299 struct inode *inode = file->f_mapping->host;
e1cbbfa5 8300 u64 outstanding_extents = 0;
0934856d 8301 size_t count = 0;
2e60a51e 8302 int flags = 0;
38851cc1
MX		 8303 bool wakeup = true;
8304 bool relock = false;
0934856d 8305 ssize_t ret;
4b46fce2 8306
6f673763 8307 if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
5a5f79b5 8308 return 0;
3f7c579c 8309
fe0f07d0 8310 inode_dio_begin(inode);
4e857c58 8311 smp_mb__after_atomic();
38851cc1 8312
0e267c44 8313 /*
41bd9ca4
MX		 8314 * The generic stuff only does filemap_write_and_wait_range, which
8315 * isn't enough if we've written compressed pages to this area, so
8316 * we need to flush the dirty pages again to make absolutely sure
8317 * that any outstanding dirty pages are on disk.
0e267c44 8318 */
a6cbcd4a 8319 count = iov_iter_count(iter);
41bd9ca4
MX		 8320 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8321 &BTRFS_I(inode)->runtime_flags))
9a025a08
WS		 8322 filemap_fdatawrite_range(inode->i_mapping, offset,
8323 offset + count - 1);
0e267c44 8324
6f673763 8325 if (iov_iter_rw(iter) == WRITE) {
38851cc1
MX		 8326 /*
8327 * If the write DIO is beyond the EOF, we need update
8328 * the isize, but it is protected by i_mutex. So we can
8329 * not unlock the i_mutex at this case.
8330 */
8331 if (offset + count <= inode->i_size) {
8332 mutex_unlock(&inode->i_mutex);
8333 relock = true;
8334 }
0934856d
MX		 8335 ret = btrfs_delalloc_reserve_space(inode, count);
8336 if (ret)
38851cc1 8337 goto out;
e1cbbfa5
JB		 8338 outstanding_extents = div64_u64(count +
8339 BTRFS_MAX_EXTENT_SIZE - 1,
8340 BTRFS_MAX_EXTENT_SIZE);
8341
8342 /*
8343 * We need to know how many extents we reserved so that we can
8344 * do the accounting properly if we go over the number we
8345 * originally calculated. Abuse current->journal_info for this.
8346 */
8347 current->journal_info = &outstanding_extents;
ee39b432
DS		 8348 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8349 &BTRFS_I(inode)->runtime_flags)) {
fe0f07d0 8350 inode_dio_end(inode);
38851cc1
MX		 8351 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8352 wakeup = false;
0934856d
MX		 8353 }
8354
17f8c842
OS		 8355 ret = __blockdev_direct_IO(iocb, inode,
8356 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8357 iter, offset, btrfs_get_blocks_direct, NULL,
8358 btrfs_submit_direct, flags);
6f673763 8359 if (iov_iter_rw(iter) == WRITE) {
e1cbbfa5 8360 current->journal_info = NULL;
ddba1bfc
LB		 8361 if (ret < 0 && ret != -EIOCBQUEUED) {
8362 /*
8363 * If the error comes from submitting stage,
8364 * btrfs_get_blocsk_direct() has free'd data space,
8365 * and metadata space will be handled by
8366 * finish_ordered_fn, don't do that again to make
8367 * sure bytes_may_use is correct.
8368 */
8369 if (!test_and_clear_bit(BTRFS_INODE_DIO_READY,
8370 &BTRFS_I(inode)->runtime_flags))
8371 btrfs_delalloc_release_space(inode, count);
8372 } else if (ret >= 0 && (size_t)ret < count)
0934856d
MX		 8373 btrfs_delalloc_release_space(inode,
8374 count - (size_t)ret);
0934856d 8375 }
38851cc1 8376out:
2e60a51e 8377 if (wakeup)
fe0f07d0 8378 inode_dio_end(inode);
38851cc1
MX		 8379 if (relock)
8380 mutex_lock(&inode->i_mutex);
0934856d
MX		 8381
8382 return ret;
16432985
CM		 8383}
8384
05dadc09
TI		 8385#define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8386
1506fcc8
YS		 8387static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8388 __u64 start, __u64 len)
8389{
05dadc09
TI		 8390 int ret;
8391
8392 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8393 if (ret)
8394 return ret;
8395
ec29ed5b 8396 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
1506fcc8
YS		 8397}
8398
a52d9a80 8399int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 8400{
d1310b2e
CM		 8401 struct extent_io_tree *tree;
8402 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 8403 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
9ebefb18 8404}
1832a6d5 8405
a52d9a80 8406static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 8407{
d1310b2e 8408 struct extent_io_tree *tree;
b888db2b
CM		 8409
8410
8411 if (current->flags & PF_MEMALLOC) {
8412 redirty_page_for_writepage(wbc, page);
8413 unlock_page(page);
8414 return 0;
8415 }
d1310b2e 8416 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 8417 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM		 8418}
8419
48a3b636
ES		 8420static int btrfs_writepages(struct address_space *mapping,
8421 struct writeback_control *wbc)
b293f02e 8422{
d1310b2e 8423 struct extent_io_tree *tree;
771ed689 8424
d1310b2e 8425 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 8426 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8427}
8428
3ab2fb5a
CM		 8429static int
8430btrfs_readpages(struct file *file, struct address_space *mapping,
8431 struct list_head *pages, unsigned nr_pages)
8432{
d1310b2e
CM		 8433 struct extent_io_tree *tree;
8434 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 8435 return extent_readpages(tree, mapping, pages, nr_pages,
8436 btrfs_get_extent);
8437}
e6dcd2dc 8438static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 8439{
d1310b2e
CM		 8440 struct extent_io_tree *tree;
8441 struct extent_map_tree *map;
a52d9a80 8442 int ret;
8c2383c3 8443
d1310b2e
CM		 8444 tree = &BTRFS_I(page->mapping->host)->io_tree;
8445 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 8446 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 8447 if (ret == 1) {
8448 ClearPagePrivate(page);
8449 set_page_private(page, 0);
8450 page_cache_release(page);
39279cc3 8451 }
a52d9a80 8452 return ret;
39279cc3
CM		 8453}
8454
e6dcd2dc
CM		 8455static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8456{
98509cfc
CM		 8457 if (PageWriteback(page) || PageDirty(page))
8458 return 0;
b335b003 8459 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
e6dcd2dc
CM		 8460}
8461
d47992f8
LC		 8462static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8463 unsigned int length)
39279cc3 8464{
5fd02043 8465 struct inode *inode = page->mapping->host;
d1310b2e 8466 struct extent_io_tree *tree;
e6dcd2dc 8467 struct btrfs_ordered_extent *ordered;
2ac55d41 8468 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 8469 u64 page_start = page_offset(page);
8470 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
131e404a 8471 int inode_evicting = inode->i_state & I_FREEING;
39279cc3 8472
8b62b72b
CM		 8473 /*
8474 * we have the page locked, so new writeback can't start,
8475 * and the dirty bit won't be cleared while we are here.
8476 *
8477 * Wait for IO on this page so that we can safely clear
8478 * the PagePrivate2 bit and do ordered accounting
8479 */
e6dcd2dc 8480 wait_on_page_writeback(page);
8b62b72b 8481
5fd02043 8482 tree = &BTRFS_I(inode)->io_tree;
e6dcd2dc
CM		 8483 if (offset) {
8484 btrfs_releasepage(page, GFP_NOFS);
8485 return;
8486 }
131e404a
FDBM		 8487
8488 if (!inode_evicting)
8489 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
8490 ordered = btrfs_lookup_ordered_extent(inode, page_start);
e6dcd2dc 8491 if (ordered) {
eb84ae03
CM		 8492 /*
8493 * IO on this page will never be started, so we need
8494 * to account for any ordered extents now
8495 */
131e404a
FDBM		 8496 if (!inode_evicting)
8497 clear_extent_bit(tree, page_start, page_end,
8498 EXTENT_DIRTY | EXTENT_DELALLOC |
8499 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8500 EXTENT_DEFRAG, 1, 0, &cached_state,
8501 GFP_NOFS);
8b62b72b
CM		 8502 /*
8503 * whoever cleared the private bit is responsible
8504 * for the finish_ordered_io
8505 */
77cef2ec
JB		 8506 if (TestClearPagePrivate2(page)) {
8507 struct btrfs_ordered_inode_tree *tree;
8508 u64 new_len;
8509
8510 tree = &BTRFS_I(inode)->ordered_tree;
8511
8512 spin_lock_irq(&tree->lock);
8513 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8514 new_len = page_start - ordered->file_offset;
8515 if (new_len < ordered->truncated_len)
8516 ordered->truncated_len = new_len;
8517 spin_unlock_irq(&tree->lock);
8518
8519 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8520 page_start,
8521 PAGE_CACHE_SIZE, 1))
8522 btrfs_finish_ordered_io(ordered);
8b62b72b 8523 }
e6dcd2dc 8524 btrfs_put_ordered_extent(ordered);
131e404a
FDBM		 8525 if (!inode_evicting) {
8526 cached_state = NULL;
8527 lock_extent_bits(tree, page_start, page_end, 0,
8528 &cached_state);
8529 }
8530 }
8531
8532 if (!inode_evicting) {
8533 clear_extent_bit(tree, page_start, page_end,
8534 EXTENT_LOCKED | EXTENT_DIRTY |
8535 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8536 EXTENT_DEFRAG, 1, 1,
8537 &cached_state, GFP_NOFS);
8538
8539 __btrfs_releasepage(page, GFP_NOFS);
e6dcd2dc 8540 }
e6dcd2dc 8541
4a096752 8542 ClearPageChecked(page);
9ad6b7bc 8543 if (PagePrivate(page)) {
9ad6b7bc
CM		 8544 ClearPagePrivate(page);
8545 set_page_private(page, 0);
8546 page_cache_release(page);
8547 }
39279cc3
CM		 8548}
8549
9ebefb18
CM		 8550/*
8551 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8552 * called from a page fault handler when a page is first dirtied. Hence we must
8553 * be careful to check for EOF conditions here. We set the page up correctly
8554 * for a written page which means we get ENOSPC checking when writing into
8555 * holes and correct delalloc and unwritten extent mapping on filesystems that
8556 * support these features.
8557 *
8558 * We are not allowed to take the i_mutex here so we have to play games to
8559 * protect against truncate races as the page could now be beyond EOF. Because
8560 * vmtruncate() writes the inode size before removing pages, once we have the
8561 * page lock we can determine safely if the page is beyond EOF. If it is not
8562 * beyond EOF, then the page is guaranteed safe against truncation until we
8563 * unlock the page.
8564 */
c2ec175c 8565int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
9ebefb18 8566{
c2ec175c 8567 struct page *page = vmf->page;
496ad9aa 8568 struct inode *inode = file_inode(vma->vm_file);
1832a6d5 8569 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 8570 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8571 struct btrfs_ordered_extent *ordered;
2ac55d41 8572 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 8573 char *kaddr;
8574 unsigned long zero_start;
9ebefb18 8575 loff_t size;
1832a6d5 8576 int ret;
9998eb70 8577 int reserved = 0;
a52d9a80 8578 u64 page_start;
e6dcd2dc 8579 u64 page_end;
9ebefb18 8580
b2b5ef5c 8581 sb_start_pagefault(inode->i_sb);
0ca1f7ce 8582 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
9998eb70 8583 if (!ret) {
e41f941a 8584 ret = file_update_time(vma->vm_file);
9998eb70
CM		 8585 reserved = 1;
8586 }
56a76f82
NP		 8587 if (ret) {
8588 if (ret == -ENOMEM)
8589 ret = VM_FAULT_OOM;
8590 else /* -ENOSPC, -EIO, etc */
8591 ret = VM_FAULT_SIGBUS;
9998eb70
CM		 8592 if (reserved)
8593 goto out;
8594 goto out_noreserve;
56a76f82 8595 }
1832a6d5 8596
56a76f82 8597 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
e6dcd2dc 8598again:
9ebefb18 8599 lock_page(page);
9ebefb18 8600 size = i_size_read(inode);
e6dcd2dc
CM		 8601 page_start = page_offset(page);
8602 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 8603
9ebefb18 8604 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 8605 (page_start >= size)) {
9ebefb18
CM		 8606 /* page got truncated out from underneath us */
8607 goto out_unlock;
8608 }
e6dcd2dc
CM		 8609 wait_on_page_writeback(page);
8610
d0082371 8611 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 8612 set_page_extent_mapped(page);
8613
eb84ae03
CM		 8614 /*
8615 * we can't set the delalloc bits if there are pending ordered
8616 * extents. Drop our locks and wait for them to finish
8617 */
e6dcd2dc
CM		 8618 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8619 if (ordered) {
2ac55d41
JB		 8620 unlock_extent_cached(io_tree, page_start, page_end,
8621 &cached_state, GFP_NOFS);
e6dcd2dc 8622 unlock_page(page);
eb84ae03 8623 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 8624 btrfs_put_ordered_extent(ordered);
8625 goto again;
8626 }
8627
fbf19087
JB		 8628 /*
8629 * XXX - page_mkwrite gets called every time the page is dirtied, even
8630 * if it was already dirty, so for space accounting reasons we need to
8631 * clear any delalloc bits for the range we are fixing to save. There
8632 * is probably a better way to do this, but for now keep consistent with
8633 * prepare_pages in the normal write path.
8634 */
2ac55d41 8635 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 8636 EXTENT_DIRTY | EXTENT_DELALLOC |
8637 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 8638 0, 0, &cached_state, GFP_NOFS);
fbf19087 8639
2ac55d41
JB		 8640 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
8641 &cached_state);
9ed74f2d 8642 if (ret) {
2ac55d41
JB		 8643 unlock_extent_cached(io_tree, page_start, page_end,
8644 &cached_state, GFP_NOFS);
9ed74f2d
JB		 8645 ret = VM_FAULT_SIGBUS;
8646 goto out_unlock;
8647 }
e6dcd2dc 8648 ret = 0;
9ebefb18
CM		 8649
8650 /* page is wholly or partially inside EOF */
a52d9a80 8651 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 8652 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 8653 else
e6dcd2dc 8654 zero_start = PAGE_CACHE_SIZE;
9ebefb18 8655
e6dcd2dc
CM		 8656 if (zero_start != PAGE_CACHE_SIZE) {
8657 kaddr = kmap(page);
8658 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
8659 flush_dcache_page(page);
8660 kunmap(page);
8661 }
247e743c 8662 ClearPageChecked(page);
e6dcd2dc 8663 set_page_dirty(page);
50a9b214 8664 SetPageUptodate(page);
5a3f23d5 8665
257c62e1
CM		 8666 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8667 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
46d8bc34 8668 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
257c62e1 8669
2ac55d41 8670 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9ebefb18
CM		 8671
8672out_unlock:
b2b5ef5c
JK		 8673 if (!ret) {
8674 sb_end_pagefault(inode->i_sb);
50a9b214 8675 return VM_FAULT_LOCKED;
b2b5ef5c 8676 }
9ebefb18 8677 unlock_page(page);
1832a6d5 8678out:
ec39e180 8679 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
9998eb70 8680out_noreserve:
b2b5ef5c 8681 sb_end_pagefault(inode->i_sb);
9ebefb18
CM		 8682 return ret;
8683}
8684
a41ad394 8685static int btrfs_truncate(struct inode *inode)
39279cc3
CM		 8686{
8687 struct btrfs_root *root = BTRFS_I(inode)->root;
fcb80c2a 8688 struct btrfs_block_rsv *rsv;
a71754fc 8689 int ret = 0;
3893e33b 8690 int err = 0;
39279cc3 8691 struct btrfs_trans_handle *trans;
dbe674a9 8692 u64 mask = root->sectorsize - 1;
07127184 8693 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3 8694
0ef8b726
JB		 8695 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8696 (u64)-1);
8697 if (ret)
8698 return ret;
39279cc3 8699
fcb80c2a
JB		 8700 /*
8701 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8702 * 3 things going on here
8703 *
8704 * 1) We need to reserve space for our orphan item and the space to
8705 * delete our orphan item. Lord knows we don't want to have a dangling
8706 * orphan item because we didn't reserve space to remove it.
8707 *
8708 * 2) We need to reserve space to update our inode.
8709 *
8710 * 3) We need to have something to cache all the space that is going to
8711 * be free'd up by the truncate operation, but also have some slack
8712 * space reserved in case it uses space during the truncate (thank you
8713 * very much snapshotting).
8714 *
8715 * And we need these to all be seperate. The fact is we can use alot of
8716 * space doing the truncate, and we have no earthly idea how much space
8717 * we will use, so we need the truncate reservation to be seperate so it
8718 * doesn't end up using space reserved for updating the inode or
8719 * removing the orphan item. We also need to be able to stop the
8720 * transaction and start a new one, which means we need to be able to
8721 * update the inode several times, and we have no idea of knowing how
8722 * many times that will be, so we can't just reserve 1 item for the
8723 * entirety of the opration, so that has to be done seperately as well.
8724 * Then there is the orphan item, which does indeed need to be held on
8725 * to for the whole operation, and we need nobody to touch this reserved
8726 * space except the orphan code.
8727 *
8728 * So that leaves us with
8729 *
8730 * 1) root->orphan_block_rsv - for the orphan deletion.
8731 * 2) rsv - for the truncate reservation, which we will steal from the
8732 * transaction reservation.
8733 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8734 * updating the inode.
8735 */
66d8f3dd 8736 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
fcb80c2a
JB		 8737 if (!rsv)
8738 return -ENOMEM;
4a338542 8739 rsv->size = min_size;
ca7e70f5 8740 rsv->failfast = 1;
f0cd846e 8741
907cbceb 8742 /*
07127184 8743 * 1 for the truncate slack space
907cbceb
JB		 8744 * 1 for updating the inode.
8745 */
f3fe820c 8746 trans = btrfs_start_transaction(root, 2);
fcb80c2a
JB		 8747 if (IS_ERR(trans)) {
8748 err = PTR_ERR(trans);
8749 goto out;
8750 }
f0cd846e 8751
907cbceb
JB		 8752 /* Migrate the slack space for the truncate to our reserve */
8753 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8754 min_size);
fcb80c2a 8755 BUG_ON(ret);
f0cd846e 8756
5dc562c5
JB		 8757 /*
8758 * So if we truncate and then write and fsync we normally would just
8759 * write the extents that changed, which is a problem if we need to
8760 * first truncate that entire inode. So set this flag so we write out
8761 * all of the extents in the inode to the sync log so we're completely
8762 * safe.
8763 */
8764 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
ca7e70f5 8765 trans->block_rsv = rsv;
907cbceb 8766
8082510e
YZ		 8767 while (1) {
8768 ret = btrfs_truncate_inode_items(trans, root, inode,
8769 inode->i_size,
8770 BTRFS_EXTENT_DATA_KEY);
28ed1345 8771 if (ret != -ENOSPC && ret != -EAGAIN) {
3893e33b 8772 err = ret;
8082510e 8773 break;
3893e33b 8774 }
39279cc3 8775
fcb80c2a 8776 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e 8777 ret = btrfs_update_inode(trans, root, inode);
3893e33b
JB		 8778 if (ret) {
8779 err = ret;
8780 break;
8781 }
ca7e70f5 8782
8082510e 8783 btrfs_end_transaction(trans, root);
b53d3f5d 8784 btrfs_btree_balance_dirty(root);
ca7e70f5
JB		 8785
8786 trans = btrfs_start_transaction(root, 2);
8787 if (IS_ERR(trans)) {
8788 ret = err = PTR_ERR(trans);
8789 trans = NULL;
8790 break;
8791 }
8792
8793 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
8794 rsv, min_size);
8795 BUG_ON(ret); /* shouldn't happen */
8796 trans->block_rsv = rsv;
8082510e
YZ		 8797 }
8798
8799 if (ret == 0 && inode->i_nlink > 0) {
fcb80c2a 8800 trans->block_rsv = root->orphan_block_rsv;
8082510e 8801 ret = btrfs_orphan_del(trans, inode);
3893e33b
JB		 8802 if (ret)
8803 err = ret;
8082510e
YZ		 8804 }
8805
917c16b2
CM		 8806 if (trans) {
8807 trans->block_rsv = &root->fs_info->trans_block_rsv;
8808 ret = btrfs_update_inode(trans, root, inode);
8809 if (ret && !err)
8810 err = ret;
7b128766 8811
7ad85bb7 8812 ret = btrfs_end_transaction(trans, root);
b53d3f5d 8813 btrfs_btree_balance_dirty(root);
917c16b2 8814 }
fcb80c2a
JB		 8815
8816out:
8817 btrfs_free_block_rsv(root, rsv);
8818
3893e33b
JB		 8819 if (ret && !err)
8820 err = ret;
a41ad394 8821
3893e33b 8822 return err;
39279cc3
CM		 8823}
8824
d352ac68
CM		 8825/*
8826 * create a new subvolume directory/inode (helper for the ioctl).
8827 */
d2fb3437 8828int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
63541927
FDBM		 8829 struct btrfs_root *new_root,
8830 struct btrfs_root *parent_root,
8831 u64 new_dirid)
39279cc3 8832{
39279cc3 8833 struct inode *inode;
76dda93c 8834 int err;
00e4e6b3 8835 u64 index = 0;
39279cc3 8836
12fc9d09
FA		 8837 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8838 new_dirid, new_dirid,
8839 S_IFDIR | (~current_umask() & S_IRWXUGO),
8840 &index);
54aa1f4d 8841 if (IS_ERR(inode))
f46b5a66 8842 return PTR_ERR(inode);
39279cc3
CM		 8843 inode->i_op = &btrfs_dir_inode_operations;
8844 inode->i_fop = &btrfs_dir_file_operations;
8845
bfe86848 8846 set_nlink(inode, 1);
dbe674a9 8847 btrfs_i_size_write(inode, 0);
b0d5d10f 8848 unlock_new_inode(inode);
3b96362c 8849
63541927
FDBM		 8850 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8851 if (err)
8852 btrfs_err(new_root->fs_info,
351fd353 8853 "error inheriting subvolume %llu properties: %d",
63541927
FDBM		 8854 new_root->root_key.objectid, err);
8855
76dda93c 8856 err = btrfs_update_inode(trans, new_root, inode);
cb8e7090 8857
76dda93c 8858 iput(inode);
ce598979 8859 return err;
39279cc3
CM		 8860}
8861
39279cc3
CM		 8862struct inode *btrfs_alloc_inode(struct super_block *sb)
8863{
8864 struct btrfs_inode *ei;
2ead6ae7 8865 struct inode *inode;
39279cc3
CM		 8866
8867 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
8868 if (!ei)
8869 return NULL;
2ead6ae7
YZ		 8870
8871 ei->root = NULL;
2ead6ae7 8872 ei->generation = 0;
15ee9bc7 8873 ei->last_trans = 0;
257c62e1 8874 ei->last_sub_trans = 0;
e02119d5 8875 ei->logged_trans = 0;
2ead6ae7 8876 ei->delalloc_bytes = 0;
47059d93 8877 ei->defrag_bytes = 0;
2ead6ae7
YZ		 8878 ei->disk_i_size = 0;
8879 ei->flags = 0;
7709cde3 8880 ei->csum_bytes = 0;
2ead6ae7 8881 ei->index_cnt = (u64)-1;
67de1176 8882 ei->dir_index = 0;
2ead6ae7 8883 ei->last_unlink_trans = 0;
46d8bc34 8884 ei->last_log_commit = 0;
2ead6ae7 8885
9e0baf60
JB		 8886 spin_lock_init(&ei->lock);
8887 ei->outstanding_extents = 0;
8888 ei->reserved_extents = 0;
2ead6ae7 8889
72ac3c0d 8890 ei->runtime_flags = 0;
261507a0 8891 ei->force_compress = BTRFS_COMPRESS_NONE;
2ead6ae7 8892
16cdcec7
MX		 8893 ei->delayed_node = NULL;
8894
9cc97d64 8895 ei->i_otime.tv_sec = 0;
8896 ei->i_otime.tv_nsec = 0;
8897
2ead6ae7 8898 inode = &ei->vfs_inode;
a8067e02 8899 extent_map_tree_init(&ei->extent_tree);
f993c883
DS		 8900 extent_io_tree_init(&ei->io_tree, &inode->i_data);
8901 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
0b32f4bb
JB		 8902 ei->io_tree.track_uptodate = 1;
8903 ei->io_failure_tree.track_uptodate = 1;
b812ce28 8904 atomic_set(&ei->sync_writers, 0);
2ead6ae7 8905 mutex_init(&ei->log_mutex);
f248679e 8906 mutex_init(&ei->delalloc_mutex);
e6dcd2dc 8907 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
2ead6ae7 8908 INIT_LIST_HEAD(&ei->delalloc_inodes);
2ead6ae7
YZ		 8909 RB_CLEAR_NODE(&ei->rb_node);
8910
8911 return inode;
39279cc3
CM		 8912}
8913
aaedb55b
JB		 8914#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8915void btrfs_test_destroy_inode(struct inode *inode)
8916{
8917 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8918 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8919}
8920#endif
8921
fa0d7e3d
NP		 8922static void btrfs_i_callback(struct rcu_head *head)
8923{
8924 struct inode *inode = container_of(head, struct inode, i_rcu);
fa0d7e3d
NP		 8925 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8926}
8927
39279cc3
CM		 8928void btrfs_destroy_inode(struct inode *inode)
8929{
e6dcd2dc 8930 struct btrfs_ordered_extent *ordered;
5a3f23d5
CM		 8931 struct btrfs_root *root = BTRFS_I(inode)->root;
8932
b3d9b7a3 8933 WARN_ON(!hlist_empty(&inode->i_dentry));
39279cc3 8934 WARN_ON(inode->i_data.nrpages);
9e0baf60
JB		 8935 WARN_ON(BTRFS_I(inode)->outstanding_extents);
8936 WARN_ON(BTRFS_I(inode)->reserved_extents);
7709cde3
JB		 8937 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8938 WARN_ON(BTRFS_I(inode)->csum_bytes);
47059d93 8939 WARN_ON(BTRFS_I(inode)->defrag_bytes);
39279cc3 8940
a6dbd429
JB		 8941 /*
8942 * This can happen where we create an inode, but somebody else also
8943 * created the same inode and we need to destroy the one we already
8944 * created.
8945 */
8946 if (!root)
8947 goto free;
8948
8a35d95f
JB		 8949 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8950 &BTRFS_I(inode)->runtime_flags)) {
c2cf52eb 8951 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
c1c9ff7c 8952 btrfs_ino(inode));
8a35d95f 8953 atomic_dec(&root->orphan_inodes);
7b128766 8954 }
7b128766 8955
d397712b 8956 while (1) {
e6dcd2dc
CM		 8957 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8958 if (!ordered)
8959 break;
8960 else {
c2cf52eb 8961 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
c1c9ff7c 8962 ordered->file_offset, ordered->len);
e6dcd2dc
CM		 8963 btrfs_remove_ordered_extent(inode, ordered);
8964 btrfs_put_ordered_extent(ordered);
8965 btrfs_put_ordered_extent(ordered);
8966 }
8967 }
5d4f98a2 8968 inode_tree_del(inode);
5b21f2ed 8969 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
a6dbd429 8970free:
fa0d7e3d 8971 call_rcu(&inode->i_rcu, btrfs_i_callback);
39279cc3
CM		 8972}
8973
45321ac5 8974int btrfs_drop_inode(struct inode *inode)
76dda93c
YZ		 8975{
8976 struct btrfs_root *root = BTRFS_I(inode)->root;
45321ac5 8977
6379ef9f
NA		 8978 if (root == NULL)
8979 return 1;
8980
fa6ac876 8981 /* the snap/subvol tree is on deleting */
69e9c6c6 8982 if (btrfs_root_refs(&root->root_item) == 0)
45321ac5 8983 return 1;
76dda93c 8984 else
45321ac5 8985 return generic_drop_inode(inode);
76dda93c
YZ		 8986}
8987
0ee0fda0 8988static void init_once(void *foo)
39279cc3
CM		 8989{
8990 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8991
8992 inode_init_once(&ei->vfs_inode);
8993}
8994
8995void btrfs_destroy_cachep(void)
8996{
8c0a8537
KS		 8997 /*
8998 * Make sure all delayed rcu free inodes are flushed before we
8999 * destroy cache.
9000 */
9001 rcu_barrier();
39279cc3
CM		 9002 if (btrfs_inode_cachep)
9003 kmem_cache_destroy(btrfs_inode_cachep);
9004 if (btrfs_trans_handle_cachep)
9005 kmem_cache_destroy(btrfs_trans_handle_cachep);
9006 if (btrfs_transaction_cachep)
9007 kmem_cache_destroy(btrfs_transaction_cachep);
39279cc3
CM		 9008 if (btrfs_path_cachep)
9009 kmem_cache_destroy(btrfs_path_cachep);
dc89e982
JB		 9010 if (btrfs_free_space_cachep)
9011 kmem_cache_destroy(btrfs_free_space_cachep);
8ccf6f19
MX		 9012 if (btrfs_delalloc_work_cachep)
9013 kmem_cache_destroy(btrfs_delalloc_work_cachep);
39279cc3
CM		 9014}
9015
9016int btrfs_init_cachep(void)
9017{
837e1972 9018 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9601e3f6
CH		 9019 sizeof(struct btrfs_inode), 0,
9020 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
39279cc3
CM		 9021 if (!btrfs_inode_cachep)
9022 goto fail;
9601e3f6 9023
837e1972 9024 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9601e3f6
CH		 9025 sizeof(struct btrfs_trans_handle), 0,
9026 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 9027 if (!btrfs_trans_handle_cachep)
9028 goto fail;
9601e3f6 9029
837e1972 9030 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9601e3f6
CH		 9031 sizeof(struct btrfs_transaction), 0,
9032 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 9033 if (!btrfs_transaction_cachep)
9034 goto fail;
9601e3f6 9035
837e1972 9036 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9601e3f6
CH		 9037 sizeof(struct btrfs_path), 0,
9038 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 9039 if (!btrfs_path_cachep)
9040 goto fail;
9601e3f6 9041
837e1972 9042 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
dc89e982
JB		 9043 sizeof(struct btrfs_free_space), 0,
9044 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9045 if (!btrfs_free_space_cachep)
9046 goto fail;
9047
8ccf6f19
MX		 9048 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
9049 sizeof(struct btrfs_delalloc_work), 0,
9050 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
9051 NULL);
9052 if (!btrfs_delalloc_work_cachep)
9053 goto fail;
9054
39279cc3
CM		 9055 return 0;
9056fail:
9057 btrfs_destroy_cachep();
9058 return -ENOMEM;
9059}
9060
9061static int btrfs_getattr(struct vfsmount *mnt,
9062 struct dentry *dentry, struct kstat *stat)
9063{
df0af1a5 9064 u64 delalloc_bytes;
2b0143b5 9065 struct inode *inode = d_inode(dentry);
fadc0d8b
DS		 9066 u32 blocksize = inode->i_sb->s_blocksize;
9067
39279cc3 9068 generic_fillattr(inode, stat);
0ee5dc67 9069 stat->dev = BTRFS_I(inode)->root->anon_dev;
d6667462 9070 stat->blksize = PAGE_CACHE_SIZE;
df0af1a5
MX		 9071
9072 spin_lock(&BTRFS_I(inode)->lock);
9073 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9074 spin_unlock(&BTRFS_I(inode)->lock);
fadc0d8b 9075 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
df0af1a5 9076 ALIGN(delalloc_bytes, blocksize)) >> 9;
39279cc3
CM		 9077 return 0;
9078}
9079
d397712b
CM		 9080static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9081 struct inode *new_dir, struct dentry *new_dentry)
39279cc3
CM		 9082{
9083 struct btrfs_trans_handle *trans;
9084 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4df27c4d 9085 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
2b0143b5
DH		 9086 struct inode *new_inode = d_inode(new_dentry);
9087 struct inode *old_inode = d_inode(old_dentry);
39279cc3 9088 struct timespec ctime = CURRENT_TIME;
00e4e6b3 9089 u64 index = 0;
4df27c4d 9090 u64 root_objectid;
39279cc3 9091 int ret;
33345d01 9092 u64 old_ino = btrfs_ino(old_inode);
39279cc3 9093
33345d01 9094 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
f679a840
YZ		 9095 return -EPERM;
9096
4df27c4d 9097 /* we only allow rename subvolume link between subvolumes */
33345d01 9098 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
3394e160
CM		 9099 return -EXDEV;
9100
33345d01
LZ		 9101 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9102 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
39279cc3 9103 return -ENOTEMPTY;
5f39d397 9104
4df27c4d
YZ		 9105 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9106 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9107 return -ENOTEMPTY;
9c52057c
CM		 9108
9109
9110 /* check for collisions, even if the name isn't there */
4871c158 9111 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9c52057c
CM		 9112 new_dentry->d_name.name,
9113 new_dentry->d_name.len);
9114
9115 if (ret) {
9116 if (ret == -EEXIST) {
9117 /* we shouldn't get
9118 * eexist without a new_inode */
fae7f21c 9119 if (WARN_ON(!new_inode)) {
9c52057c
CM		 9120 return ret;
9121 }
9122 } else {
9123 /* maybe -EOVERFLOW */
9124 return ret;
9125 }
9126 }
9127 ret = 0;
9128
5a3f23d5 9129 /*
8d875f95
CM		 9130 * we're using rename to replace one file with another. Start IO on it
9131 * now so we don't add too much work to the end of the transaction
5a3f23d5 9132 */
8d875f95 9133 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
5a3f23d5
CM		 9134 filemap_flush(old_inode->i_mapping);
9135
76dda93c 9136 /* close the racy window with snapshot create/destroy ioctl */
33345d01 9137 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 9138 down_read(&root->fs_info->subvol_sem);
a22285a6
YZ		 9139 /*
9140 * We want to reserve the absolute worst case amount of items. So if
9141 * both inodes are subvols and we need to unlink them then that would
9142 * require 4 item modifications, but if they are both normal inodes it
9143 * would require 5 item modifications, so we'll assume their normal
9144 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9145 * should cover the worst case number of items we'll modify.
9146 */
6e137ed3 9147 trans = btrfs_start_transaction(root, 11);
b44c59a8
JL		 9148 if (IS_ERR(trans)) {
9149 ret = PTR_ERR(trans);
9150 goto out_notrans;
9151 }
76dda93c 9152
4df27c4d
YZ		 9153 if (dest != root)
9154 btrfs_record_root_in_trans(trans, dest);
5f39d397 9155
a5719521
YZ		 9156 ret = btrfs_set_inode_index(new_dir, &index);
9157 if (ret)
9158 goto out_fail;
5a3f23d5 9159
67de1176 9160 BTRFS_I(old_inode)->dir_index = 0ULL;
33345d01 9161 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d 9162 /* force full log commit if subvolume involved. */
995946dd 9163 btrfs_set_log_full_commit(root->fs_info, trans);
4df27c4d 9164 } else {
a5719521
YZ		 9165 ret = btrfs_insert_inode_ref(trans, dest,
9166 new_dentry->d_name.name,
9167 new_dentry->d_name.len,
33345d01
LZ		 9168 old_ino,
9169 btrfs_ino(new_dir), index);
a5719521
YZ		 9170 if (ret)
9171 goto out_fail;
4df27c4d
YZ		 9172 /*
9173 * this is an ugly little race, but the rename is required
9174 * to make sure that if we crash, the inode is either at the
9175 * old name or the new one. pinning the log transaction lets
9176 * us make sure we don't allow a log commit to come in after
9177 * we unlink the name but before we add the new name back in.
9178 */
9179 btrfs_pin_log_trans(root);
9180 }
5a3f23d5 9181
0c4d2d95
JB		 9182 inode_inc_iversion(old_dir);
9183 inode_inc_iversion(new_dir);
9184 inode_inc_iversion(old_inode);
39279cc3
CM		 9185 old_dir->i_ctime = old_dir->i_mtime = ctime;
9186 new_dir->i_ctime = new_dir->i_mtime = ctime;
9187 old_inode->i_ctime = ctime;
5f39d397 9188
12fcfd22
CM		 9189 if (old_dentry->d_parent != new_dentry->d_parent)
9190 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9191
33345d01 9192 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 9193 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9194 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9195 old_dentry->d_name.name,
9196 old_dentry->d_name.len);
9197 } else {
92986796 9198 ret = __btrfs_unlink_inode(trans, root, old_dir,
2b0143b5 9199 d_inode(old_dentry),
92986796
AV		 9200 old_dentry->d_name.name,
9201 old_dentry->d_name.len);
9202 if (!ret)
9203 ret = btrfs_update_inode(trans, root, old_inode);
4df27c4d 9204 }
79787eaa
JM		 9205 if (ret) {
9206 btrfs_abort_transaction(trans, root, ret);
9207 goto out_fail;
9208 }
39279cc3
CM		 9209
9210 if (new_inode) {
0c4d2d95 9211 inode_inc_iversion(new_inode);
39279cc3 9212 new_inode->i_ctime = CURRENT_TIME;
33345d01 9213 if (unlikely(btrfs_ino(new_inode) ==
4df27c4d
YZ		 9214 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9215 root_objectid = BTRFS_I(new_inode)->location.objectid;
9216 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9217 root_objectid,
9218 new_dentry->d_name.name,
9219 new_dentry->d_name.len);
9220 BUG_ON(new_inode->i_nlink == 0);
9221 } else {
9222 ret = btrfs_unlink_inode(trans, dest, new_dir,
2b0143b5 9223 d_inode(new_dentry),
4df27c4d
YZ		 9224 new_dentry->d_name.name,
9225 new_dentry->d_name.len);
9226 }
4ef31a45 9227 if (!ret && new_inode->i_nlink == 0)
2b0143b5 9228 ret = btrfs_orphan_add(trans, d_inode(new_dentry));
79787eaa
JM		 9229 if (ret) {
9230 btrfs_abort_transaction(trans, root, ret);
9231 goto out_fail;
9232 }
39279cc3 9233 }
aec7477b 9234
4df27c4d
YZ		 9235 ret = btrfs_add_link(trans, new_dir, old_inode,
9236 new_dentry->d_name.name,
a5719521 9237 new_dentry->d_name.len, 0, index);
79787eaa
JM		 9238 if (ret) {
9239 btrfs_abort_transaction(trans, root, ret);
9240 goto out_fail;
9241 }
39279cc3 9242
67de1176
MX		 9243 if (old_inode->i_nlink == 1)
9244 BTRFS_I(old_inode)->dir_index = index;
9245
33345d01 9246 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
10d9f309 9247 struct dentry *parent = new_dentry->d_parent;
6a912213 9248 btrfs_log_new_name(trans, old_inode, old_dir, parent);
4df27c4d
YZ		 9249 btrfs_end_log_trans(root);
9250 }
39279cc3 9251out_fail:
7ad85bb7 9252 btrfs_end_transaction(trans, root);
b44c59a8 9253out_notrans:
33345d01 9254 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 9255 up_read(&root->fs_info->subvol_sem);
9ed74f2d 9256
39279cc3
CM		 9257 return ret;
9258}
9259
80ace85c
MS		 9260static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9261 struct inode *new_dir, struct dentry *new_dentry,
9262 unsigned int flags)
9263{
9264 if (flags & ~RENAME_NOREPLACE)
9265 return -EINVAL;
9266
9267 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
9268}
9269
8ccf6f19
MX		 9270static void btrfs_run_delalloc_work(struct btrfs_work *work)
9271{
9272 struct btrfs_delalloc_work *delalloc_work;
9f23e289 9273 struct inode *inode;
8ccf6f19
MX		 9274
9275 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9276 work);
9f23e289
JB		 9277 inode = delalloc_work->inode;
9278 if (delalloc_work->wait) {
9279 btrfs_wait_ordered_range(inode, 0, (u64)-1);
9280 } else {
9281 filemap_flush(inode->i_mapping);
9282 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9283 &BTRFS_I(inode)->runtime_flags))
9284 filemap_flush(inode->i_mapping);
9285 }
8ccf6f19
MX		 9286
9287 if (delalloc_work->delay_iput)
9f23e289 9288 btrfs_add_delayed_iput(inode);
8ccf6f19 9289 else
9f23e289 9290 iput(inode);
8ccf6f19
MX		 9291 complete(&delalloc_work->completion);
9292}
9293
9294struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9295 int wait, int delay_iput)
9296{
9297 struct btrfs_delalloc_work *work;
9298
9299 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
9300 if (!work)
9301 return NULL;
9302
9303 init_completion(&work->completion);
9304 INIT_LIST_HEAD(&work->list);
9305 work->inode = inode;
9306 work->wait = wait;
9307 work->delay_iput = delay_iput;
9e0af237
LB		 9308 WARN_ON_ONCE(!inode);
9309 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9310 btrfs_run_delalloc_work, NULL, NULL);
8ccf6f19
MX		 9311
9312 return work;
9313}
9314
9315void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9316{
9317 wait_for_completion(&work->completion);
9318 kmem_cache_free(btrfs_delalloc_work_cachep, work);
9319}
9320
d352ac68
CM		 9321/*
9322 * some fairly slow code that needs optimization. This walks the list
9323 * of all the inodes with pending delalloc and forces them to disk.
9324 */
6c255e67
MX		 9325static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9326 int nr)
ea8c2819 9327{
ea8c2819 9328 struct btrfs_inode *binode;
5b21f2ed 9329 struct inode *inode;
8ccf6f19
MX		 9330 struct btrfs_delalloc_work *work, *next;
9331 struct list_head works;
1eafa6c7 9332 struct list_head splice;
8ccf6f19 9333 int ret = 0;
ea8c2819 9334
8ccf6f19 9335 INIT_LIST_HEAD(&works);
1eafa6c7 9336 INIT_LIST_HEAD(&splice);
63607cc8 9337
573bfb72 9338 mutex_lock(&root->delalloc_mutex);
eb73c1b7
MX		 9339 spin_lock(&root->delalloc_lock);
9340 list_splice_init(&root->delalloc_inodes, &splice);
1eafa6c7
MX		 9341 while (!list_empty(&splice)) {
9342 binode = list_entry(splice.next, struct btrfs_inode,
ea8c2819 9343 delalloc_inodes);
1eafa6c7 9344
eb73c1b7
MX		 9345 list_move_tail(&binode->delalloc_inodes,
9346 &root->delalloc_inodes);
5b21f2ed 9347 inode = igrab(&binode->vfs_inode);
df0af1a5 9348 if (!inode) {
eb73c1b7 9349 cond_resched_lock(&root->delalloc_lock);
1eafa6c7 9350 continue;
df0af1a5 9351 }
eb73c1b7 9352 spin_unlock(&root->delalloc_lock);
1eafa6c7
MX		 9353
9354 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
5d99a998 9355 if (!work) {
f4ab9ea7
JB		 9356 if (delay_iput)
9357 btrfs_add_delayed_iput(inode);
9358 else
9359 iput(inode);
1eafa6c7 9360 ret = -ENOMEM;
a1ecaabb 9361 goto out;
5b21f2ed 9362 }
1eafa6c7 9363 list_add_tail(&work->list, &works);
a44903ab
QW		 9364 btrfs_queue_work(root->fs_info->flush_workers,
9365 &work->work);
6c255e67
MX		 9366 ret++;
9367 if (nr != -1 && ret >= nr)
a1ecaabb 9368 goto out;
5b21f2ed 9369 cond_resched();
eb73c1b7 9370 spin_lock(&root->delalloc_lock);
ea8c2819 9371 }
eb73c1b7 9372 spin_unlock(&root->delalloc_lock);
8c8bee1d 9373
a1ecaabb 9374out:
eb73c1b7
MX		 9375 list_for_each_entry_safe(work, next, &works, list) {
9376 list_del_init(&work->list);
9377 btrfs_wait_and_free_delalloc_work(work);
9378 }
9379
9380 if (!list_empty_careful(&splice)) {
9381 spin_lock(&root->delalloc_lock);
9382 list_splice_tail(&splice, &root->delalloc_inodes);
9383 spin_unlock(&root->delalloc_lock);
9384 }
573bfb72 9385 mutex_unlock(&root->delalloc_mutex);
eb73c1b7
MX		 9386 return ret;
9387}
1eafa6c7 9388
eb73c1b7
MX		 9389int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9390{
9391 int ret;
1eafa6c7 9392
2c21b4d7 9393 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
eb73c1b7
MX		 9394 return -EROFS;
9395
6c255e67
MX		 9396 ret = __start_delalloc_inodes(root, delay_iput, -1);
9397 if (ret > 0)
9398 ret = 0;
eb73c1b7
MX		 9399 /*
9400 * the filemap_flush will queue IO into the worker threads, but
8c8bee1d
CM		 9401 * we have to make sure the IO is actually started and that
9402 * ordered extents get created before we return
9403 */
9404 atomic_inc(&root->fs_info->async_submit_draining);
d397712b 9405 while (atomic_read(&root->fs_info->nr_async_submits) ||
771ed689 9406 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 9407 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 9408 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9409 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 9410 }
9411 atomic_dec(&root->fs_info->async_submit_draining);
eb73c1b7
MX		 9412 return ret;
9413}
9414
6c255e67
MX		 9415int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9416 int nr)
eb73c1b7
MX		 9417{
9418 struct btrfs_root *root;
9419 struct list_head splice;
9420 int ret;
9421
2c21b4d7 9422 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
eb73c1b7
MX		 9423 return -EROFS;
9424
9425 INIT_LIST_HEAD(&splice);
9426
573bfb72 9427 mutex_lock(&fs_info->delalloc_root_mutex);
eb73c1b7
MX		 9428 spin_lock(&fs_info->delalloc_root_lock);
9429 list_splice_init(&fs_info->delalloc_roots, &splice);
6c255e67 9430 while (!list_empty(&splice) && nr) {
eb73c1b7
MX		 9431 root = list_first_entry(&splice, struct btrfs_root,
9432 delalloc_root);
9433 root = btrfs_grab_fs_root(root);
9434 BUG_ON(!root);
9435 list_move_tail(&root->delalloc_root,
9436 &fs_info->delalloc_roots);
9437 spin_unlock(&fs_info->delalloc_root_lock);
9438
6c255e67 9439 ret = __start_delalloc_inodes(root, delay_iput, nr);
eb73c1b7 9440 btrfs_put_fs_root(root);
6c255e67 9441 if (ret < 0)
eb73c1b7
MX		 9442 goto out;
9443
6c255e67
MX		 9444 if (nr != -1) {
9445 nr -= ret;
9446 WARN_ON(nr < 0);
9447 }
eb73c1b7 9448 spin_lock(&fs_info->delalloc_root_lock);
8ccf6f19 9449 }
eb73c1b7 9450 spin_unlock(&fs_info->delalloc_root_lock);
1eafa6c7 9451
6c255e67 9452 ret = 0;
eb73c1b7
MX		 9453 atomic_inc(&fs_info->async_submit_draining);
9454 while (atomic_read(&fs_info->nr_async_submits) ||
9455 atomic_read(&fs_info->async_delalloc_pages)) {
9456 wait_event(fs_info->async_submit_wait,
9457 (atomic_read(&fs_info->nr_async_submits) == 0 &&
9458 atomic_read(&fs_info->async_delalloc_pages) == 0));
9459 }
9460 atomic_dec(&fs_info->async_submit_draining);
eb73c1b7 9461out:
1eafa6c7 9462 if (!list_empty_careful(&splice)) {
eb73c1b7
MX		 9463 spin_lock(&fs_info->delalloc_root_lock);
9464 list_splice_tail(&splice, &fs_info->delalloc_roots);
9465 spin_unlock(&fs_info->delalloc_root_lock);
1eafa6c7 9466 }
573bfb72 9467 mutex_unlock(&fs_info->delalloc_root_mutex);
8ccf6f19 9468 return ret;
ea8c2819
CM		 9469}
9470
39279cc3
CM		 9471static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9472 const char *symname)
9473{
9474 struct btrfs_trans_handle *trans;
9475 struct btrfs_root *root = BTRFS_I(dir)->root;
9476 struct btrfs_path *path;
9477 struct btrfs_key key;
1832a6d5 9478 struct inode *inode = NULL;
39279cc3
CM		 9479 int err;
9480 int drop_inode = 0;
9481 u64 objectid;
67871254 9482 u64 index = 0;
39279cc3
CM		 9483 int name_len;
9484 int datasize;
5f39d397 9485 unsigned long ptr;
39279cc3 9486 struct btrfs_file_extent_item *ei;
5f39d397 9487 struct extent_buffer *leaf;
39279cc3 9488
f06becc4 9489 name_len = strlen(symname);
39279cc3
CM		 9490 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9491 return -ENAMETOOLONG;
1832a6d5 9492
9ed74f2d
JB		 9493 /*
9494 * 2 items for inode item and ref
9495 * 2 items for dir items
9496 * 1 item for xattr if selinux is on
9497 */
a22285a6
YZ		 9498 trans = btrfs_start_transaction(root, 5);
9499 if (IS_ERR(trans))
9500 return PTR_ERR(trans);
1832a6d5 9501
581bb050
LZ		 9502 err = btrfs_find_free_ino(root, &objectid);
9503 if (err)
9504 goto out_unlock;
9505
aec7477b 9506 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 9507 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 9508 S_IFLNK|S_IRWXUGO, &index);
7cf96da3
TI		 9509 if (IS_ERR(inode)) {
9510 err = PTR_ERR(inode);
39279cc3 9511 goto out_unlock;
7cf96da3 9512 }
39279cc3 9513
ad19db71
CS		 9514 /*
9515 * If the active LSM wants to access the inode during
9516 * d_instantiate it needs these. Smack checks to see
9517 * if the filesystem supports xattrs by looking at the
9518 * ops vector.
9519 */
9520 inode->i_fop = &btrfs_file_operations;
9521 inode->i_op = &btrfs_file_inode_operations;
b0d5d10f 9522 inode->i_mapping->a_ops = &btrfs_aops;
b0d5d10f
CM		 9523 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9524
9525 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9526 if (err)
9527 goto out_unlock_inode;
ad19db71 9528
a1b075d2 9529 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3 9530 if (err)
b0d5d10f 9531 goto out_unlock_inode;
39279cc3
CM		 9532
9533 path = btrfs_alloc_path();
d8926bb3
MF		 9534 if (!path) {
9535 err = -ENOMEM;
b0d5d10f 9536 goto out_unlock_inode;
d8926bb3 9537 }
33345d01 9538 key.objectid = btrfs_ino(inode);
39279cc3 9539 key.offset = 0;
962a298f 9540 key.type = BTRFS_EXTENT_DATA_KEY;
39279cc3
CM		 9541 datasize = btrfs_file_extent_calc_inline_size(name_len);
9542 err = btrfs_insert_empty_item(trans, root, path, &key,
9543 datasize);
54aa1f4d 9544 if (err) {
b0839166 9545 btrfs_free_path(path);
b0d5d10f 9546 goto out_unlock_inode;
54aa1f4d 9547 }
5f39d397
CM		 9548 leaf = path->nodes[0];
9549 ei = btrfs_item_ptr(leaf, path->slots[0],
9550 struct btrfs_file_extent_item);
9551 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9552 btrfs_set_file_extent_type(leaf, ei,
39279cc3 9553 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 9554 btrfs_set_file_extent_encryption(leaf, ei, 0);
9555 btrfs_set_file_extent_compression(leaf, ei, 0);
9556 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9557 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9558
39279cc3 9559 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 9560 write_extent_buffer(leaf, symname, ptr, name_len);
9561 btrfs_mark_buffer_dirty(leaf);
39279cc3 9562 btrfs_free_path(path);
5f39d397 9563
39279cc3
CM		 9564 inode->i_op = &btrfs_symlink_inode_operations;
9565 inode->i_mapping->a_ops = &btrfs_symlink_aops;
d899e052 9566 inode_set_bytes(inode, name_len);
f06becc4 9567 btrfs_i_size_write(inode, name_len);
54aa1f4d 9568 err = btrfs_update_inode(trans, root, inode);
b0d5d10f 9569 if (err) {
54aa1f4d 9570 drop_inode = 1;
b0d5d10f
CM		 9571 goto out_unlock_inode;
9572 }
9573
9574 unlock_new_inode(inode);
9575 d_instantiate(dentry, inode);
39279cc3
CM		 9576
9577out_unlock:
7ad85bb7 9578 btrfs_end_transaction(trans, root);
39279cc3
CM		 9579 if (drop_inode) {
9580 inode_dec_link_count(inode);
9581 iput(inode);
9582 }
b53d3f5d 9583 btrfs_btree_balance_dirty(root);
39279cc3 9584 return err;
b0d5d10f
CM		 9585
9586out_unlock_inode:
9587 drop_inode = 1;
9588 unlock_new_inode(inode);
9589 goto out_unlock;
39279cc3 9590}
16432985 9591
0af3d00b
JB		 9592static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9593 u64 start, u64 num_bytes, u64 min_size,
9594 loff_t actual_len, u64 *alloc_hint,
9595 struct btrfs_trans_handle *trans)
d899e052 9596{
5dc562c5
JB		 9597 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9598 struct extent_map *em;
d899e052
YZ		 9599 struct btrfs_root *root = BTRFS_I(inode)->root;
9600 struct btrfs_key ins;
d899e052 9601 u64 cur_offset = start;
55a61d1d 9602 u64 i_size;
154ea289 9603 u64 cur_bytes;
d899e052 9604 int ret = 0;
0af3d00b 9605 bool own_trans = true;
d899e052 9606
0af3d00b
JB		 9607 if (trans)
9608 own_trans = false;
d899e052 9609 while (num_bytes > 0) {
0af3d00b
JB		 9610 if (own_trans) {
9611 trans = btrfs_start_transaction(root, 3);
9612 if (IS_ERR(trans)) {
9613 ret = PTR_ERR(trans);
9614 break;
9615 }
5a303d5d
YZ		 9616 }
9617
154ea289
CM		 9618 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
9619 cur_bytes = max(cur_bytes, min_size);
00361589 9620 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
e570fd27 9621 *alloc_hint, &ins, 1, 0);
5a303d5d 9622 if (ret) {
0af3d00b
JB		 9623 if (own_trans)
9624 btrfs_end_transaction(trans, root);
a22285a6 9625 break;
d899e052 9626 }
5a303d5d 9627
d899e052
YZ		 9628 ret = insert_reserved_file_extent(trans, inode,
9629 cur_offset, ins.objectid,
9630 ins.offset, ins.offset,
920bbbfb 9631 ins.offset, 0, 0, 0,
d899e052 9632 BTRFS_FILE_EXTENT_PREALLOC);
79787eaa 9633 if (ret) {
857cc2fc 9634 btrfs_free_reserved_extent(root, ins.objectid,
e570fd27 9635 ins.offset, 0);
79787eaa
JM		 9636 btrfs_abort_transaction(trans, root, ret);
9637 if (own_trans)
9638 btrfs_end_transaction(trans, root);
9639 break;
9640 }
31193213 9641
a1ed835e
CM		 9642 btrfs_drop_extent_cache(inode, cur_offset,
9643 cur_offset + ins.offset -1, 0);
5a303d5d 9644
5dc562c5
JB		 9645 em = alloc_extent_map();
9646 if (!em) {
9647 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9648 &BTRFS_I(inode)->runtime_flags);
9649 goto next;
9650 }
9651
9652 em->start = cur_offset;
9653 em->orig_start = cur_offset;
9654 em->len = ins.offset;
9655 em->block_start = ins.objectid;
9656 em->block_len = ins.offset;
b4939680 9657 em->orig_block_len = ins.offset;
cc95bef6 9658 em->ram_bytes = ins.offset;
5dc562c5
JB		 9659 em->bdev = root->fs_info->fs_devices->latest_bdev;
9660 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9661 em->generation = trans->transid;
9662
9663 while (1) {
9664 write_lock(&em_tree->lock);
09a2a8f9 9665 ret = add_extent_mapping(em_tree, em, 1);
5dc562c5
JB		 9666 write_unlock(&em_tree->lock);
9667 if (ret != -EEXIST)
9668 break;
9669 btrfs_drop_extent_cache(inode, cur_offset,
9670 cur_offset + ins.offset - 1,
9671 0);
9672 }
9673 free_extent_map(em);
9674next:
d899e052
YZ		 9675 num_bytes -= ins.offset;
9676 cur_offset += ins.offset;
efa56464 9677 *alloc_hint = ins.objectid + ins.offset;
5a303d5d 9678
0c4d2d95 9679 inode_inc_iversion(inode);
d899e052 9680 inode->i_ctime = CURRENT_TIME;
6cbff00f 9681 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
d899e052 9682 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
efa56464
YZ		 9683 (actual_len > inode->i_size) &&
9684 (cur_offset > inode->i_size)) {
d1ea6a61 9685 if (cur_offset > actual_len)
55a61d1d 9686 i_size = actual_len;
d1ea6a61 9687 else
55a61d1d
JB		 9688 i_size = cur_offset;
9689 i_size_write(inode, i_size);
9690 btrfs_ordered_update_i_size(inode, i_size, NULL);
5a303d5d
YZ		 9691 }
9692
d899e052 9693 ret = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 9694
9695 if (ret) {
9696 btrfs_abort_transaction(trans, root, ret);
9697 if (own_trans)
9698 btrfs_end_transaction(trans, root);
9699 break;
9700 }
d899e052 9701
0af3d00b
JB		 9702 if (own_trans)
9703 btrfs_end_transaction(trans, root);
5a303d5d 9704 }
d899e052
YZ		 9705 return ret;
9706}
9707
0af3d00b
JB		 9708int btrfs_prealloc_file_range(struct inode *inode, int mode,
9709 u64 start, u64 num_bytes, u64 min_size,
9710 loff_t actual_len, u64 *alloc_hint)
9711{
9712 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9713 min_size, actual_len, alloc_hint,
9714 NULL);
9715}
9716
9717int btrfs_prealloc_file_range_trans(struct inode *inode,
9718 struct btrfs_trans_handle *trans, int mode,
9719 u64 start, u64 num_bytes, u64 min_size,
9720 loff_t actual_len, u64 *alloc_hint)
9721{
9722 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9723 min_size, actual_len, alloc_hint, trans);
9724}
9725
e6dcd2dc
CM		 9726static int btrfs_set_page_dirty(struct page *page)
9727{
e6dcd2dc
CM		 9728 return __set_page_dirty_nobuffers(page);
9729}
9730
10556cb2 9731static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd 9732{
b83cc969 9733 struct btrfs_root *root = BTRFS_I(inode)->root;
cb6db4e5 9734 umode_t mode = inode->i_mode;
b83cc969 9735
cb6db4e5
JM		 9736 if (mask & MAY_WRITE &&
9737 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9738 if (btrfs_root_readonly(root))
9739 return -EROFS;
9740 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9741 return -EACCES;
9742 }
2830ba7f 9743 return generic_permission(inode, mask);
fdebe2bd 9744}
39279cc3 9745
ef3b9af5
FM		 9746static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9747{
9748 struct btrfs_trans_handle *trans;
9749 struct btrfs_root *root = BTRFS_I(dir)->root;
9750 struct inode *inode = NULL;
9751 u64 objectid;
9752 u64 index;
9753 int ret = 0;
9754
9755 /*
9756 * 5 units required for adding orphan entry
9757 */
9758 trans = btrfs_start_transaction(root, 5);
9759 if (IS_ERR(trans))
9760 return PTR_ERR(trans);
9761
9762 ret = btrfs_find_free_ino(root, &objectid);
9763 if (ret)
9764 goto out;
9765
9766 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9767 btrfs_ino(dir), objectid, mode, &index);
9768 if (IS_ERR(inode)) {
9769 ret = PTR_ERR(inode);
9770 inode = NULL;
9771 goto out;
9772 }
9773
ef3b9af5
FM		 9774 inode->i_fop = &btrfs_file_operations;
9775 inode->i_op = &btrfs_file_inode_operations;
9776
9777 inode->i_mapping->a_ops = &btrfs_aops;
ef3b9af5
FM		 9778 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9779
b0d5d10f
CM		 9780 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9781 if (ret)
9782 goto out_inode;
9783
9784 ret = btrfs_update_inode(trans, root, inode);
9785 if (ret)
9786 goto out_inode;
ef3b9af5
FM		 9787 ret = btrfs_orphan_add(trans, inode);
9788 if (ret)
b0d5d10f 9789 goto out_inode;
ef3b9af5 9790
5762b5c9
FM		 9791 /*
9792 * We set number of links to 0 in btrfs_new_inode(), and here we set
9793 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9794 * through:
9795 *
9796 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9797 */
9798 set_nlink(inode, 1);
b0d5d10f 9799 unlock_new_inode(inode);
ef3b9af5
FM		 9800 d_tmpfile(dentry, inode);
9801 mark_inode_dirty(inode);
9802
9803out:
9804 btrfs_end_transaction(trans, root);
9805 if (ret)
9806 iput(inode);
9807 btrfs_balance_delayed_items(root);
9808 btrfs_btree_balance_dirty(root);
ef3b9af5 9809 return ret;
b0d5d10f
CM		 9810
9811out_inode:
9812 unlock_new_inode(inode);
9813 goto out;
9814
ef3b9af5
FM		 9815}
9816
b38ef71c
FM		 9817/* Inspired by filemap_check_errors() */
9818int btrfs_inode_check_errors(struct inode *inode)
9819{
9820 int ret = 0;
9821
9822 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
9823 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
9824 ret = -ENOSPC;
9825 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
9826 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
9827 ret = -EIO;
9828
9829 return ret;
9830}
9831
6e1d5dcc 9832static const struct inode_operations btrfs_dir_inode_operations = {
3394e160 9833 .getattr = btrfs_getattr,
39279cc3
CM		 9834 .lookup = btrfs_lookup,
9835 .create = btrfs_create,
9836 .unlink = btrfs_unlink,
9837 .link = btrfs_link,
9838 .mkdir = btrfs_mkdir,
9839 .rmdir = btrfs_rmdir,
80ace85c 9840 .rename2 = btrfs_rename2,
39279cc3
CM		 9841 .symlink = btrfs_symlink,
9842 .setattr = btrfs_setattr,
618e21d5 9843 .mknod = btrfs_mknod,
95819c05
CH		 9844 .setxattr = btrfs_setxattr,
9845 .getxattr = btrfs_getxattr,
5103e947 9846 .listxattr = btrfs_listxattr,
95819c05 9847 .removexattr = btrfs_removexattr,
fdebe2bd 9848 .permission = btrfs_permission,
4e34e719 9849 .get_acl = btrfs_get_acl,
996a710d 9850 .set_acl = btrfs_set_acl,
93fd63c2 9851 .update_time = btrfs_update_time,
ef3b9af5 9852 .tmpfile = btrfs_tmpfile,
39279cc3 9853};
6e1d5dcc 9854static const struct inode_operations btrfs_dir_ro_inode_operations = {
39279cc3 9855 .lookup = btrfs_lookup,
fdebe2bd 9856 .permission = btrfs_permission,
4e34e719 9857 .get_acl = btrfs_get_acl,
996a710d 9858 .set_acl = btrfs_set_acl,
93fd63c2 9859 .update_time = btrfs_update_time,
39279cc3 9860};
76dda93c 9861
828c0950 9862static const struct file_operations btrfs_dir_file_operations = {
39279cc3
CM		 9863 .llseek = generic_file_llseek,
9864 .read = generic_read_dir,
9cdda8d3 9865 .iterate = btrfs_real_readdir,
34287aa3 9866 .unlocked_ioctl = btrfs_ioctl,
39279cc3 9867#ifdef CONFIG_COMPAT
34287aa3 9868 .compat_ioctl = btrfs_ioctl,
39279cc3 9869#endif
6bf13c0c 9870 .release = btrfs_release_file,
e02119d5 9871 .fsync = btrfs_sync_file,
39279cc3
CM		 9872};
9873
d1310b2e 9874static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 9875 .fill_delalloc = run_delalloc_range,
065631f6 9876 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 9877 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 9878 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 9879 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 9880 .writepage_start_hook = btrfs_writepage_start_hook,
b0c68f8b
CM		 9881 .set_bit_hook = btrfs_set_bit_hook,
9882 .clear_bit_hook = btrfs_clear_bit_hook,
9ed74f2d
JB		 9883 .merge_extent_hook = btrfs_merge_extent_hook,
9884 .split_extent_hook = btrfs_split_extent_hook,
07157aac
CM		 9885};
9886
35054394
CM		 9887/*
9888 * btrfs doesn't support the bmap operation because swapfiles
9889 * use bmap to make a mapping of extents in the file. They assume
9890 * these extents won't change over the life of the file and they
9891 * use the bmap result to do IO directly to the drive.
9892 *
9893 * the btrfs bmap call would return logical addresses that aren't
9894 * suitable for IO and they also will change frequently as COW
9895 * operations happen. So, swapfile + btrfs == corruption.
9896 *
9897 * For now we're avoiding this by dropping bmap.
9898 */
7f09410b 9899static const struct address_space_operations btrfs_aops = {
39279cc3
CM		 9900 .readpage = btrfs_readpage,
9901 .writepage = btrfs_writepage,
b293f02e 9902 .writepages = btrfs_writepages,
3ab2fb5a 9903 .readpages = btrfs_readpages,
16432985 9904 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 9905 .invalidatepage = btrfs_invalidatepage,
9906 .releasepage = btrfs_releasepage,
e6dcd2dc 9907 .set_page_dirty = btrfs_set_page_dirty,
465fdd97 9908 .error_remove_page = generic_error_remove_page,
39279cc3
CM		 9909};
9910
7f09410b 9911static const struct address_space_operations btrfs_symlink_aops = {
39279cc3
CM		 9912 .readpage = btrfs_readpage,
9913 .writepage = btrfs_writepage,
2bf5a725
CM		 9914 .invalidatepage = btrfs_invalidatepage,
9915 .releasepage = btrfs_releasepage,
39279cc3
CM		 9916};
9917
6e1d5dcc 9918static const struct inode_operations btrfs_file_inode_operations = {
39279cc3
CM		 9919 .getattr = btrfs_getattr,
9920 .setattr = btrfs_setattr,
95819c05
CH		 9921 .setxattr = btrfs_setxattr,
9922 .getxattr = btrfs_getxattr,
5103e947 9923 .listxattr = btrfs_listxattr,
95819c05 9924 .removexattr = btrfs_removexattr,
fdebe2bd 9925 .permission = btrfs_permission,
1506fcc8 9926 .fiemap = btrfs_fiemap,
4e34e719 9927 .get_acl = btrfs_get_acl,
996a710d 9928 .set_acl = btrfs_set_acl,
e41f941a 9929 .update_time = btrfs_update_time,
39279cc3 9930};
6e1d5dcc 9931static const struct inode_operations btrfs_special_inode_operations = {
618e21d5
JB		 9932 .getattr = btrfs_getattr,
9933 .setattr = btrfs_setattr,
fdebe2bd 9934 .permission = btrfs_permission,
95819c05
CH		 9935 .setxattr = btrfs_setxattr,
9936 .getxattr = btrfs_getxattr,
33268eaf 9937 .listxattr = btrfs_listxattr,
95819c05 9938 .removexattr = btrfs_removexattr,
4e34e719 9939 .get_acl = btrfs_get_acl,
996a710d 9940 .set_acl = btrfs_set_acl,
e41f941a 9941 .update_time = btrfs_update_time,
618e21d5 9942};
6e1d5dcc 9943static const struct inode_operations btrfs_symlink_inode_operations = {
39279cc3
CM		 9944 .readlink = generic_readlink,
9945 .follow_link = page_follow_link_light,
9946 .put_link = page_put_link,
f209561a 9947 .getattr = btrfs_getattr,
22c44fe6 9948 .setattr = btrfs_setattr,
fdebe2bd 9949 .permission = btrfs_permission,
0279b4cd
JO		 9950 .setxattr = btrfs_setxattr,
9951 .getxattr = btrfs_getxattr,
9952 .listxattr = btrfs_listxattr,
9953 .removexattr = btrfs_removexattr,
e41f941a 9954 .update_time = btrfs_update_time,
39279cc3 9955};
76dda93c 9956
82d339d9 9957const struct dentry_operations btrfs_dentry_operations = {
76dda93c 9958 .d_delete = btrfs_dentry_delete,
b4aff1f8 9959 .d_release = btrfs_dentry_release,
76dda93c 9960};
Linux 6.x block layer and io_uring tree(s)