projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
btrfs: drop duplicate prefix from scrub workqueues
[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:
94ed938a
QW		 313 /*
314 * Don't forget to free the reserved space, as for inlined extent
315 * it won't count as data extent, free them directly here.
316 * And at reserve time, it's always aligned to page size, so
317 * just free one page here.
318 */
319 btrfs_qgroup_free_data(inode, 0, PAGE_CACHE_SIZE);
1acae57b 320 btrfs_free_path(path);
00361589
JB		 321 btrfs_end_transaction(trans, root);
322 return ret;
c8b97818
CM		 323}
324
771ed689
CM		 325struct async_extent {
326 u64 start;
327 u64 ram_size;
328 u64 compressed_size;
329 struct page **pages;
330 unsigned long nr_pages;
261507a0 331 int compress_type;
771ed689
CM		 332 struct list_head list;
333};
334
335struct async_cow {
336 struct inode *inode;
337 struct btrfs_root *root;
338 struct page *locked_page;
339 u64 start;
340 u64 end;
341 struct list_head extents;
342 struct btrfs_work work;
343};
344
345static noinline int add_async_extent(struct async_cow *cow,
346 u64 start, u64 ram_size,
347 u64 compressed_size,
348 struct page **pages,
261507a0
LZ		 349 unsigned long nr_pages,
350 int compress_type)
771ed689
CM		 351{
352 struct async_extent *async_extent;
353
354 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
79787eaa 355 BUG_ON(!async_extent); /* -ENOMEM */
771ed689
CM		 356 async_extent->start = start;
357 async_extent->ram_size = ram_size;
358 async_extent->compressed_size = compressed_size;
359 async_extent->pages = pages;
360 async_extent->nr_pages = nr_pages;
261507a0 361 async_extent->compress_type = compress_type;
771ed689
CM		 362 list_add_tail(&async_extent->list, &cow->extents);
363 return 0;
364}
365
f79707b0
WS		 366static inline int inode_need_compress(struct inode *inode)
367{
368 struct btrfs_root *root = BTRFS_I(inode)->root;
369
370 /* force compress */
371 if (btrfs_test_opt(root, FORCE_COMPRESS))
372 return 1;
373 /* bad compression ratios */
374 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
375 return 0;
376 if (btrfs_test_opt(root, COMPRESS) ||
377 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
378 BTRFS_I(inode)->force_compress)
379 return 1;
380 return 0;
381}
382
d352ac68 383/*
771ed689
CM		 384 * we create compressed extents in two phases. The first
385 * phase compresses a range of pages that have already been
386 * locked (both pages and state bits are locked).
c8b97818 387 *
771ed689
CM		 388 * This is done inside an ordered work queue, and the compression
389 * is spread across many cpus. The actual IO submission is step
390 * two, and the ordered work queue takes care of making sure that
391 * happens in the same order things were put onto the queue by
392 * writepages and friends.
c8b97818 393 *
771ed689
CM		 394 * If this code finds it can't get good compression, it puts an
395 * entry onto the work queue to write the uncompressed bytes. This
396 * makes sure that both compressed inodes and uncompressed inodes
b2570314
AB		 397 * are written in the same order that the flusher thread sent them
398 * down.
d352ac68 399 */
c44f649e 400static noinline void compress_file_range(struct inode *inode,
771ed689
CM		 401 struct page *locked_page,
402 u64 start, u64 end,
403 struct async_cow *async_cow,
404 int *num_added)
b888db2b
CM		 405{
406 struct btrfs_root *root = BTRFS_I(inode)->root;
db94535d 407 u64 num_bytes;
db94535d 408 u64 blocksize = root->sectorsize;
c8b97818 409 u64 actual_end;
42dc7bab 410 u64 isize = i_size_read(inode);
e6dcd2dc 411 int ret = 0;
c8b97818
CM		 412 struct page **pages = NULL;
413 unsigned long nr_pages;
414 unsigned long nr_pages_ret = 0;
415 unsigned long total_compressed = 0;
416 unsigned long total_in = 0;
417 unsigned long max_compressed = 128 * 1024;
771ed689 418 unsigned long max_uncompressed = 128 * 1024;
c8b97818
CM		 419 int i;
420 int will_compress;
261507a0 421 int compress_type = root->fs_info->compress_type;
4adaa611 422 int redirty = 0;
b888db2b 423
4cb13e5d
LB		 424 /* if this is a small write inside eof, kick off a defrag */
425 if ((end - start + 1) < 16 * 1024 &&
426 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
4cb5300b
CM		 427 btrfs_add_inode_defrag(NULL, inode);
428
42dc7bab 429 actual_end = min_t(u64, isize, end + 1);
c8b97818
CM		 430again:
431 will_compress = 0;
432 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
433 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
be20aa9d 434
f03d9301
CM		 435 /*
436 * we don't want to send crud past the end of i_size through
437 * compression, that's just a waste of CPU time. So, if the
438 * end of the file is before the start of our current
439 * requested range of bytes, we bail out to the uncompressed
440 * cleanup code that can deal with all of this.
441 *
442 * It isn't really the fastest way to fix things, but this is a
443 * very uncommon corner.
444 */
445 if (actual_end <= start)
446 goto cleanup_and_bail_uncompressed;
447
c8b97818
CM		 448 total_compressed = actual_end - start;
449
4bcbb332
SW		 450 /*
451 * skip compression for a small file range(<=blocksize) that
452 * isn't an inline extent, since it dosen't save disk space at all.
453 */
454 if (total_compressed <= blocksize &&
455 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
456 goto cleanup_and_bail_uncompressed;
457
c8b97818
CM		 458 /* we want to make sure that amount of ram required to uncompress
459 * an extent is reasonable, so we limit the total size in ram
771ed689
CM		 460 * of a compressed extent to 128k. This is a crucial number
461 * because it also controls how easily we can spread reads across
462 * cpus for decompression.
463 *
464 * We also want to make sure the amount of IO required to do
465 * a random read is reasonably small, so we limit the size of
466 * a compressed extent to 128k.
c8b97818
CM		 467 */
468 total_compressed = min(total_compressed, max_uncompressed);
fda2832f 469 num_bytes = ALIGN(end - start + 1, blocksize);
be20aa9d 470 num_bytes = max(blocksize, num_bytes);
c8b97818
CM		 471 total_in = 0;
472 ret = 0;
db94535d 473
771ed689
CM		 474 /*
475 * we do compression for mount -o compress and when the
476 * inode has not been flagged as nocompress. This flag can
477 * change at any time if we discover bad compression ratios.
c8b97818 478 */
f79707b0 479 if (inode_need_compress(inode)) {
c8b97818 480 WARN_ON(pages);
31e818fe 481 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
560f7d75
LZ		 482 if (!pages) {
483 /* just bail out to the uncompressed code */
484 goto cont;
485 }
c8b97818 486
261507a0
LZ		 487 if (BTRFS_I(inode)->force_compress)
488 compress_type = BTRFS_I(inode)->force_compress;
489
4adaa611
CM		 490 /*
491 * we need to call clear_page_dirty_for_io on each
492 * page in the range. Otherwise applications with the file
493 * mmap'd can wander in and change the page contents while
494 * we are compressing them.
495 *
496 * If the compression fails for any reason, we set the pages
497 * dirty again later on.
498 */
499 extent_range_clear_dirty_for_io(inode, start, end);
500 redirty = 1;
261507a0
LZ		 501 ret = btrfs_compress_pages(compress_type,
502 inode->i_mapping, start,
503 total_compressed, pages,
504 nr_pages, &nr_pages_ret,
505 &total_in,
506 &total_compressed,
507 max_compressed);
c8b97818
CM		 508
509 if (!ret) {
510 unsigned long offset = total_compressed &
511 (PAGE_CACHE_SIZE - 1);
512 struct page *page = pages[nr_pages_ret - 1];
513 char *kaddr;
514
515 /* zero the tail end of the last page, we might be
516 * sending it down to disk
517 */
518 if (offset) {
7ac687d9 519 kaddr = kmap_atomic(page);
c8b97818
CM		 520 memset(kaddr + offset, 0,
521 PAGE_CACHE_SIZE - offset);
7ac687d9 522 kunmap_atomic(kaddr);
c8b97818
CM		 523 }
524 will_compress = 1;
525 }
526 }
560f7d75 527cont:
c8b97818
CM		 528 if (start == 0) {
529 /* lets try to make an inline extent */
771ed689 530 if (ret || total_in < (actual_end - start)) {
c8b97818 531 /* we didn't compress the entire range, try
771ed689 532 * to make an uncompressed inline extent.
c8b97818 533 */
00361589
JB		 534 ret = cow_file_range_inline(root, inode, start, end,
535 0, 0, NULL);
c8b97818 536 } else {
771ed689 537 /* try making a compressed inline extent */
00361589 538 ret = cow_file_range_inline(root, inode, start, end,
fe3f566c
LZ		 539 total_compressed,
540 compress_type, pages);
c8b97818 541 }
79787eaa 542 if (ret <= 0) {
151a41bc
JB		 543 unsigned long clear_flags = EXTENT_DELALLOC |
544 EXTENT_DEFRAG;
e6eb4314
FM		 545 unsigned long page_error_op;
546
151a41bc 547 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
e6eb4314 548 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
151a41bc 549
771ed689 550 /*
79787eaa
JM		 551 * inline extent creation worked or returned error,
552 * we don't need to create any more async work items.
553 * Unlock and free up our temp pages.
771ed689 554 */
c2790a2e 555 extent_clear_unlock_delalloc(inode, start, end, NULL,
151a41bc 556 clear_flags, PAGE_UNLOCK |
c2790a2e
JB		 557 PAGE_CLEAR_DIRTY |
558 PAGE_SET_WRITEBACK |
e6eb4314 559 page_error_op |
c2790a2e 560 PAGE_END_WRITEBACK);
c8b97818
CM		 561 goto free_pages_out;
562 }
563 }
564
565 if (will_compress) {
566 /*
567 * we aren't doing an inline extent round the compressed size
568 * up to a block size boundary so the allocator does sane
569 * things
570 */
fda2832f 571 total_compressed = ALIGN(total_compressed, blocksize);
c8b97818
CM		 572
573 /*
574 * one last check to make sure the compression is really a
575 * win, compare the page count read with the blocks on disk
576 */
fda2832f 577 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
c8b97818
CM		 578 if (total_compressed >= total_in) {
579 will_compress = 0;
580 } else {
c8b97818
CM		 581 num_bytes = total_in;
582 }
583 }
584 if (!will_compress && pages) {
585 /*
586 * the compression code ran but failed to make things smaller,
587 * free any pages it allocated and our page pointer array
588 */
589 for (i = 0; i < nr_pages_ret; i++) {
70b99e69 590 WARN_ON(pages[i]->mapping);
c8b97818
CM		 591 page_cache_release(pages[i]);
592 }
593 kfree(pages);
594 pages = NULL;
595 total_compressed = 0;
596 nr_pages_ret = 0;
597
598 /* flag the file so we don't compress in the future */
1e701a32
CM		 599 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
600 !(BTRFS_I(inode)->force_compress)) {
a555f810 601 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
1e701a32 602 }
c8b97818 603 }
771ed689
CM		 604 if (will_compress) {
605 *num_added += 1;
c8b97818 606
771ed689
CM		 607 /* the async work queues will take care of doing actual
608 * allocation on disk for these compressed pages,
609 * and will submit them to the elevator.
610 */
611 add_async_extent(async_cow, start, num_bytes,
261507a0
LZ		 612 total_compressed, pages, nr_pages_ret,
613 compress_type);
179e29e4 614
24ae6365 615 if (start + num_bytes < end) {
771ed689
CM		 616 start += num_bytes;
617 pages = NULL;
618 cond_resched();
619 goto again;
620 }
621 } else {
f03d9301 622cleanup_and_bail_uncompressed:
771ed689
CM		 623 /*
624 * No compression, but we still need to write the pages in
625 * the file we've been given so far. redirty the locked
626 * page if it corresponds to our extent and set things up
627 * for the async work queue to run cow_file_range to do
628 * the normal delalloc dance
629 */
630 if (page_offset(locked_page) >= start &&
631 page_offset(locked_page) <= end) {
632 __set_page_dirty_nobuffers(locked_page);
633 /* unlocked later on in the async handlers */
634 }
4adaa611
CM		 635 if (redirty)
636 extent_range_redirty_for_io(inode, start, end);
261507a0
LZ		 637 add_async_extent(async_cow, start, end - start + 1,
638 0, NULL, 0, BTRFS_COMPRESS_NONE);
771ed689
CM		 639 *num_added += 1;
640 }
3b951516 641
c44f649e 642 return;
771ed689
CM		 643
644free_pages_out:
645 for (i = 0; i < nr_pages_ret; i++) {
646 WARN_ON(pages[i]->mapping);
647 page_cache_release(pages[i]);
648 }
d397712b 649 kfree(pages);
771ed689 650}
771ed689 651
40ae837b
FM		 652static void free_async_extent_pages(struct async_extent *async_extent)
653{
654 int i;
655
656 if (!async_extent->pages)
657 return;
658
659 for (i = 0; i < async_extent->nr_pages; i++) {
660 WARN_ON(async_extent->pages[i]->mapping);
661 page_cache_release(async_extent->pages[i]);
662 }
663 kfree(async_extent->pages);
664 async_extent->nr_pages = 0;
665 async_extent->pages = NULL;
771ed689
CM		 666}
667
668/*
669 * phase two of compressed writeback. This is the ordered portion
670 * of the code, which only gets called in the order the work was
671 * queued. We walk all the async extents created by compress_file_range
672 * and send them down to the disk.
673 */
dec8f175 674static noinline void submit_compressed_extents(struct inode *inode,
771ed689
CM		 675 struct async_cow *async_cow)
676{
677 struct async_extent *async_extent;
678 u64 alloc_hint = 0;
771ed689
CM		 679 struct btrfs_key ins;
680 struct extent_map *em;
681 struct btrfs_root *root = BTRFS_I(inode)->root;
682 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
683 struct extent_io_tree *io_tree;
f5a84ee3 684 int ret = 0;
771ed689 685
3e04e7f1 686again:
d397712b 687 while (!list_empty(&async_cow->extents)) {
771ed689
CM		 688 async_extent = list_entry(async_cow->extents.next,
689 struct async_extent, list);
690 list_del(&async_extent->list);
c8b97818 691
771ed689
CM		 692 io_tree = &BTRFS_I(inode)->io_tree;
693
f5a84ee3 694retry:
771ed689
CM		 695 /* did the compression code fall back to uncompressed IO? */
696 if (!async_extent->pages) {
697 int page_started = 0;
698 unsigned long nr_written = 0;
699
700 lock_extent(io_tree, async_extent->start,
2ac55d41 701 async_extent->start +
d0082371 702 async_extent->ram_size - 1);
771ed689
CM		 703
704 /* allocate blocks */
f5a84ee3
JB		 705 ret = cow_file_range(inode, async_cow->locked_page,
706 async_extent->start,
707 async_extent->start +
708 async_extent->ram_size - 1,
709 &page_started, &nr_written, 0);
771ed689 710
79787eaa
JM		 711 /* JDM XXX */
712
771ed689
CM		 713 /*
714 * if page_started, cow_file_range inserted an
715 * inline extent and took care of all the unlocking
716 * and IO for us. Otherwise, we need to submit
717 * all those pages down to the drive.
718 */
f5a84ee3 719 if (!page_started && !ret)
771ed689
CM		 720 extent_write_locked_range(io_tree,
721 inode, async_extent->start,
d397712b 722 async_extent->start +
771ed689
CM		 723 async_extent->ram_size - 1,
724 btrfs_get_extent,
725 WB_SYNC_ALL);
3e04e7f1
JB		 726 else if (ret)
727 unlock_page(async_cow->locked_page);
771ed689
CM		 728 kfree(async_extent);
729 cond_resched();
730 continue;
731 }
732
733 lock_extent(io_tree, async_extent->start,
d0082371 734 async_extent->start + async_extent->ram_size - 1);
771ed689 735
00361589 736 ret = btrfs_reserve_extent(root,
771ed689
CM		 737 async_extent->compressed_size,
738 async_extent->compressed_size,
e570fd27 739 0, alloc_hint, &ins, 1, 1);
f5a84ee3 740 if (ret) {
40ae837b 741 free_async_extent_pages(async_extent);
3e04e7f1 742
fdf8e2ea
JB		 743 if (ret == -ENOSPC) {
744 unlock_extent(io_tree, async_extent->start,
745 async_extent->start +
746 async_extent->ram_size - 1);
ce62003f
LB		 747
748 /*
749 * we need to redirty the pages if we decide to
750 * fallback to uncompressed IO, otherwise we
751 * will not submit these pages down to lower
752 * layers.
753 */
754 extent_range_redirty_for_io(inode,
755 async_extent->start,
756 async_extent->start +
757 async_extent->ram_size - 1);
758
79787eaa 759 goto retry;
fdf8e2ea 760 }
3e04e7f1 761 goto out_free;
f5a84ee3 762 }
c2167754
YZ		 763 /*
764 * here we're doing allocation and writeback of the
765 * compressed pages
766 */
767 btrfs_drop_extent_cache(inode, async_extent->start,
768 async_extent->start +
769 async_extent->ram_size - 1, 0);
770
172ddd60 771 em = alloc_extent_map();
b9aa55be
LB		 772 if (!em) {
773 ret = -ENOMEM;
3e04e7f1 774 goto out_free_reserve;
b9aa55be 775 }
771ed689
CM		 776 em->start = async_extent->start;
777 em->len = async_extent->ram_size;
445a6944 778 em->orig_start = em->start;
2ab28f32
JB		 779 em->mod_start = em->start;
780 em->mod_len = em->len;
c8b97818 781
771ed689
CM		 782 em->block_start = ins.objectid;
783 em->block_len = ins.offset;
b4939680 784 em->orig_block_len = ins.offset;
cc95bef6 785 em->ram_bytes = async_extent->ram_size;
771ed689 786 em->bdev = root->fs_info->fs_devices->latest_bdev;
261507a0 787 em->compress_type = async_extent->compress_type;
771ed689
CM		 788 set_bit(EXTENT_FLAG_PINNED, &em->flags);
789 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
70c8a91c 790 em->generation = -1;
771ed689 791
d397712b 792 while (1) {
890871be 793 write_lock(&em_tree->lock);
09a2a8f9 794 ret = add_extent_mapping(em_tree, em, 1);
890871be 795 write_unlock(&em_tree->lock);
771ed689
CM		 796 if (ret != -EEXIST) {
797 free_extent_map(em);
798 break;
799 }
800 btrfs_drop_extent_cache(inode, async_extent->start,
801 async_extent->start +
802 async_extent->ram_size - 1, 0);
803 }
804
3e04e7f1
JB		 805 if (ret)
806 goto out_free_reserve;
807
261507a0
LZ		 808 ret = btrfs_add_ordered_extent_compress(inode,
809 async_extent->start,
810 ins.objectid,
811 async_extent->ram_size,
812 ins.offset,
813 BTRFS_ORDERED_COMPRESSED,
814 async_extent->compress_type);
d9f85963
FM		 815 if (ret) {
816 btrfs_drop_extent_cache(inode, async_extent->start,
817 async_extent->start +
818 async_extent->ram_size - 1, 0);
3e04e7f1 819 goto out_free_reserve;
d9f85963 820 }
771ed689 821
771ed689
CM		 822 /*
823 * clear dirty, set writeback and unlock the pages.
824 */
c2790a2e 825 extent_clear_unlock_delalloc(inode, async_extent->start,
a791e35e
CM		 826 async_extent->start +
827 async_extent->ram_size - 1,
151a41bc
JB		 828 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
829 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
c2790a2e 830 PAGE_SET_WRITEBACK);
771ed689 831 ret = btrfs_submit_compressed_write(inode,
d397712b
CM		 832 async_extent->start,
833 async_extent->ram_size,
834 ins.objectid,
835 ins.offset, async_extent->pages,
836 async_extent->nr_pages);
fce2a4e6
FM		 837 if (ret) {
838 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
839 struct page *p = async_extent->pages[0];
840 const u64 start = async_extent->start;
841 const u64 end = start + async_extent->ram_size - 1;
842
843 p->mapping = inode->i_mapping;
844 tree->ops->writepage_end_io_hook(p, start, end,
845 NULL, 0);
846 p->mapping = NULL;
847 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
848 PAGE_END_WRITEBACK |
849 PAGE_SET_ERROR);
40ae837b 850 free_async_extent_pages(async_extent);
fce2a4e6 851 }
771ed689
CM		 852 alloc_hint = ins.objectid + ins.offset;
853 kfree(async_extent);
854 cond_resched();
855 }
dec8f175 856 return;
3e04e7f1 857out_free_reserve:
e570fd27 858 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
79787eaa 859out_free:
c2790a2e 860 extent_clear_unlock_delalloc(inode, async_extent->start,
3e04e7f1
JB		 861 async_extent->start +
862 async_extent->ram_size - 1,
c2790a2e 863 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc
JB		 864 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
865 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
704de49d
FM		 866 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
867 PAGE_SET_ERROR);
40ae837b 868 free_async_extent_pages(async_extent);
79787eaa 869 kfree(async_extent);
3e04e7f1 870 goto again;
771ed689
CM		 871}
872
4b46fce2
JB		 873static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
874 u64 num_bytes)
875{
876 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
877 struct extent_map *em;
878 u64 alloc_hint = 0;
879
880 read_lock(&em_tree->lock);
881 em = search_extent_mapping(em_tree, start, num_bytes);
882 if (em) {
883 /*
884 * if block start isn't an actual block number then find the
885 * first block in this inode and use that as a hint. If that
886 * block is also bogus then just don't worry about it.
887 */
888 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
889 free_extent_map(em);
890 em = search_extent_mapping(em_tree, 0, 0);
891 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
892 alloc_hint = em->block_start;
893 if (em)
894 free_extent_map(em);
895 } else {
896 alloc_hint = em->block_start;
897 free_extent_map(em);
898 }
899 }
900 read_unlock(&em_tree->lock);
901
902 return alloc_hint;
903}
904
771ed689
CM		 905/*
906 * when extent_io.c finds a delayed allocation range in the file,
907 * the call backs end up in this code. The basic idea is to
908 * allocate extents on disk for the range, and create ordered data structs
909 * in ram to track those extents.
910 *
911 * locked_page is the page that writepage had locked already. We use
912 * it to make sure we don't do extra locks or unlocks.
913 *
914 * *page_started is set to one if we unlock locked_page and do everything
915 * required to start IO on it. It may be clean and already done with
916 * IO when we return.
917 */
00361589
JB		 918static noinline int cow_file_range(struct inode *inode,
919 struct page *locked_page,
920 u64 start, u64 end, int *page_started,
921 unsigned long *nr_written,
922 int unlock)
771ed689 923{
00361589 924 struct btrfs_root *root = BTRFS_I(inode)->root;
771ed689
CM		 925 u64 alloc_hint = 0;
926 u64 num_bytes;
927 unsigned long ram_size;
928 u64 disk_num_bytes;
929 u64 cur_alloc_size;
930 u64 blocksize = root->sectorsize;
771ed689
CM		 931 struct btrfs_key ins;
932 struct extent_map *em;
933 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
934 int ret = 0;
935
02ecd2c2
JB		 936 if (btrfs_is_free_space_inode(inode)) {
937 WARN_ON_ONCE(1);
29bce2f3
JB		 938 ret = -EINVAL;
939 goto out_unlock;
02ecd2c2 940 }
771ed689 941
fda2832f 942 num_bytes = ALIGN(end - start + 1, blocksize);
771ed689
CM		 943 num_bytes = max(blocksize, num_bytes);
944 disk_num_bytes = num_bytes;
771ed689 945
4cb5300b 946 /* if this is a small write inside eof, kick off defrag */
4cb13e5d
LB		 947 if (num_bytes < 64 * 1024 &&
948 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
00361589 949 btrfs_add_inode_defrag(NULL, inode);
4cb5300b 950
771ed689
CM		 951 if (start == 0) {
952 /* lets try to make an inline extent */
00361589
JB		 953 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
954 NULL);
771ed689 955 if (ret == 0) {
c2790a2e
JB		 956 extent_clear_unlock_delalloc(inode, start, end, NULL,
957 EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc 958 EXTENT_DEFRAG, PAGE_UNLOCK |
c2790a2e
JB		 959 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
960 PAGE_END_WRITEBACK);
c2167754 961
771ed689
CM		 962 *nr_written = *nr_written +
963 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
964 *page_started = 1;
771ed689 965 goto out;
79787eaa 966 } else if (ret < 0) {
79787eaa 967 goto out_unlock;
771ed689
CM		 968 }
969 }
970
971 BUG_ON(disk_num_bytes >
6c41761f 972 btrfs_super_total_bytes(root->fs_info->super_copy));
771ed689 973
4b46fce2 974 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
771ed689
CM		 975 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
976
d397712b 977 while (disk_num_bytes > 0) {
a791e35e
CM		 978 unsigned long op;
979
287a0ab9 980 cur_alloc_size = disk_num_bytes;
00361589 981 ret = btrfs_reserve_extent(root, cur_alloc_size,
771ed689 982 root->sectorsize, 0, alloc_hint,
e570fd27 983 &ins, 1, 1);
00361589 984 if (ret < 0)
79787eaa 985 goto out_unlock;
d397712b 986
172ddd60 987 em = alloc_extent_map();
b9aa55be
LB		 988 if (!em) {
989 ret = -ENOMEM;
ace68bac 990 goto out_reserve;
b9aa55be 991 }
e6dcd2dc 992 em->start = start;
445a6944 993 em->orig_start = em->start;
771ed689
CM		 994 ram_size = ins.offset;
995 em->len = ins.offset;
2ab28f32
JB		 996 em->mod_start = em->start;
997 em->mod_len = em->len;
c8b97818 998
e6dcd2dc 999 em->block_start = ins.objectid;
c8b97818 1000 em->block_len = ins.offset;
b4939680 1001 em->orig_block_len = ins.offset;
cc95bef6 1002 em->ram_bytes = ram_size;
e6dcd2dc 1003 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 1004 set_bit(EXTENT_FLAG_PINNED, &em->flags);
70c8a91c 1005 em->generation = -1;
c8b97818 1006
d397712b 1007 while (1) {
890871be 1008 write_lock(&em_tree->lock);
09a2a8f9 1009 ret = add_extent_mapping(em_tree, em, 1);
890871be 1010 write_unlock(&em_tree->lock);
e6dcd2dc
CM		 1011 if (ret != -EEXIST) {
1012 free_extent_map(em);
1013 break;
1014 }
1015 btrfs_drop_extent_cache(inode, start,
c8b97818 1016 start + ram_size - 1, 0);
e6dcd2dc 1017 }
ace68bac
LB		 1018 if (ret)
1019 goto out_reserve;
e6dcd2dc 1020
98d20f67 1021 cur_alloc_size = ins.offset;
e6dcd2dc 1022 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771ed689 1023 ram_size, cur_alloc_size, 0);
ace68bac 1024 if (ret)
d9f85963 1025 goto out_drop_extent_cache;
c8b97818 1026
17d217fe
YZ		 1027 if (root->root_key.objectid ==
1028 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1029 ret = btrfs_reloc_clone_csums(inode, start,
1030 cur_alloc_size);
00361589 1031 if (ret)
d9f85963 1032 goto out_drop_extent_cache;
17d217fe
YZ		 1033 }
1034
d397712b 1035 if (disk_num_bytes < cur_alloc_size)
3b951516 1036 break;
d397712b 1037
c8b97818
CM		 1038 /* we're not doing compressed IO, don't unlock the first
1039 * page (which the caller expects to stay locked), don't
1040 * clear any dirty bits and don't set any writeback bits
8b62b72b
CM		 1041 *
1042 * Do set the Private2 bit so we know this page was properly
1043 * setup for writepage
c8b97818 1044 */
c2790a2e
JB		 1045 op = unlock ? PAGE_UNLOCK : 0;
1046 op |= PAGE_SET_PRIVATE2;
a791e35e 1047
c2790a2e
JB		 1048 extent_clear_unlock_delalloc(inode, start,
1049 start + ram_size - 1, locked_page,
1050 EXTENT_LOCKED | EXTENT_DELALLOC,
1051 op);
c8b97818 1052 disk_num_bytes -= cur_alloc_size;
c59f8951
CM		 1053 num_bytes -= cur_alloc_size;
1054 alloc_hint = ins.objectid + ins.offset;
1055 start += cur_alloc_size;
b888db2b 1056 }
79787eaa 1057out:
be20aa9d 1058 return ret;
b7d5b0a8 1059
d9f85963
FM		 1060out_drop_extent_cache:
1061 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
ace68bac 1062out_reserve:
e570fd27 1063 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
79787eaa 1064out_unlock:
c2790a2e 1065 extent_clear_unlock_delalloc(inode, start, end, locked_page,
151a41bc
JB		 1066 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1067 EXTENT_DELALLOC | EXTENT_DEFRAG,
1068 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1069 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
79787eaa 1070 goto out;
771ed689 1071}
c8b97818 1072
771ed689
CM		 1073/*
1074 * work queue call back to started compression on a file and pages
1075 */
1076static noinline void async_cow_start(struct btrfs_work *work)
1077{
1078 struct async_cow *async_cow;
1079 int num_added = 0;
1080 async_cow = container_of(work, struct async_cow, work);
1081
1082 compress_file_range(async_cow->inode, async_cow->locked_page,
1083 async_cow->start, async_cow->end, async_cow,
1084 &num_added);
8180ef88 1085 if (num_added == 0) {
cb77fcd8 1086 btrfs_add_delayed_iput(async_cow->inode);
771ed689 1087 async_cow->inode = NULL;
8180ef88 1088 }
771ed689
CM		 1089}
1090
1091/*
1092 * work queue call back to submit previously compressed pages
1093 */
1094static noinline void async_cow_submit(struct btrfs_work *work)
1095{
1096 struct async_cow *async_cow;
1097 struct btrfs_root *root;
1098 unsigned long nr_pages;
1099
1100 async_cow = container_of(work, struct async_cow, work);
1101
1102 root = async_cow->root;
1103 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1104 PAGE_CACHE_SHIFT;
1105
ee863954
DS		 1106 /*
1107 * atomic_sub_return implies a barrier for waitqueue_active
1108 */
66657b31 1109 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
287082b0 1110 5 * 1024 * 1024 &&
771ed689
CM		 1111 waitqueue_active(&root->fs_info->async_submit_wait))
1112 wake_up(&root->fs_info->async_submit_wait);
1113
d397712b 1114 if (async_cow->inode)
771ed689 1115 submit_compressed_extents(async_cow->inode, async_cow);
771ed689 1116}
c8b97818 1117
771ed689
CM		 1118static noinline void async_cow_free(struct btrfs_work *work)
1119{
1120 struct async_cow *async_cow;
1121 async_cow = container_of(work, struct async_cow, work);
8180ef88 1122 if (async_cow->inode)
cb77fcd8 1123 btrfs_add_delayed_iput(async_cow->inode);
771ed689
CM		 1124 kfree(async_cow);
1125}
1126
1127static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1128 u64 start, u64 end, int *page_started,
1129 unsigned long *nr_written)
1130{
1131 struct async_cow *async_cow;
1132 struct btrfs_root *root = BTRFS_I(inode)->root;
1133 unsigned long nr_pages;
1134 u64 cur_end;
287082b0 1135 int limit = 10 * 1024 * 1024;
771ed689 1136
a3429ab7
CM		 1137 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1138 1, 0, NULL, GFP_NOFS);
d397712b 1139 while (start < end) {
771ed689 1140 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
79787eaa 1141 BUG_ON(!async_cow); /* -ENOMEM */
8180ef88 1142 async_cow->inode = igrab(inode);
771ed689
CM		 1143 async_cow->root = root;
1144 async_cow->locked_page = locked_page;
1145 async_cow->start = start;
1146
f79707b0
WS		 1147 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1148 !btrfs_test_opt(root, FORCE_COMPRESS))
771ed689
CM		 1149 cur_end = end;
1150 else
1151 cur_end = min(end, start + 512 * 1024 - 1);
1152
1153 async_cow->end = cur_end;
1154 INIT_LIST_HEAD(&async_cow->extents);
1155
9e0af237
LB		 1156 btrfs_init_work(&async_cow->work,
1157 btrfs_delalloc_helper,
1158 async_cow_start, async_cow_submit,
1159 async_cow_free);
771ed689 1160
771ed689
CM		 1161 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1162 PAGE_CACHE_SHIFT;
1163 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1164
afe3d242
QW		 1165 btrfs_queue_work(root->fs_info->delalloc_workers,
1166 &async_cow->work);
771ed689
CM		 1167
1168 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1169 wait_event(root->fs_info->async_submit_wait,
1170 (atomic_read(&root->fs_info->async_delalloc_pages) <
1171 limit));
1172 }
1173
d397712b 1174 while (atomic_read(&root->fs_info->async_submit_draining) &&
771ed689
CM		 1175 atomic_read(&root->fs_info->async_delalloc_pages)) {
1176 wait_event(root->fs_info->async_submit_wait,
1177 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1178 0));
1179 }
1180
1181 *nr_written += nr_pages;
1182 start = cur_end + 1;
1183 }
1184 *page_started = 1;
1185 return 0;
be20aa9d
CM		 1186}
1187
d397712b 1188static noinline int csum_exist_in_range(struct btrfs_root *root,
17d217fe
YZ		 1189 u64 bytenr, u64 num_bytes)
1190{
1191 int ret;
1192 struct btrfs_ordered_sum *sums;
1193 LIST_HEAD(list);
1194
07d400a6 1195 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
a2de733c 1196 bytenr + num_bytes - 1, &list, 0);
17d217fe
YZ		 1197 if (ret == 0 && list_empty(&list))
1198 return 0;
1199
1200 while (!list_empty(&list)) {
1201 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1202 list_del(&sums->list);
1203 kfree(sums);
1204 }
1205 return 1;
1206}
1207
d352ac68
CM		 1208/*
1209 * when nowcow writeback call back. This checks for snapshots or COW copies
1210 * of the extents that exist in the file, and COWs the file as required.
1211 *
1212 * If no cow copies or snapshots exist, we write directly to the existing
1213 * blocks on disk
1214 */
7f366cfe
CM		 1215static noinline int run_delalloc_nocow(struct inode *inode,
1216 struct page *locked_page,
771ed689
CM		 1217 u64 start, u64 end, int *page_started, int force,
1218 unsigned long *nr_written)
be20aa9d 1219{
be20aa9d 1220 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 1221 struct btrfs_trans_handle *trans;
be20aa9d 1222 struct extent_buffer *leaf;
be20aa9d 1223 struct btrfs_path *path;
80ff3856 1224 struct btrfs_file_extent_item *fi;
be20aa9d 1225 struct btrfs_key found_key;
80ff3856
YZ		 1226 u64 cow_start;
1227 u64 cur_offset;
1228 u64 extent_end;
5d4f98a2 1229 u64 extent_offset;
80ff3856
YZ		 1230 u64 disk_bytenr;
1231 u64 num_bytes;
b4939680 1232 u64 disk_num_bytes;
cc95bef6 1233 u64 ram_bytes;
80ff3856 1234 int extent_type;
79787eaa 1235 int ret, err;
d899e052 1236 int type;
80ff3856
YZ		 1237 int nocow;
1238 int check_prev = 1;
82d5902d 1239 bool nolock;
33345d01 1240 u64 ino = btrfs_ino(inode);
be20aa9d
CM		 1241
1242 path = btrfs_alloc_path();
17ca04af 1243 if (!path) {
c2790a2e
JB		 1244 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1245 EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc
JB		 1246 EXTENT_DO_ACCOUNTING |
1247 EXTENT_DEFRAG, PAGE_UNLOCK |
c2790a2e
JB		 1248 PAGE_CLEAR_DIRTY |
1249 PAGE_SET_WRITEBACK |
1250 PAGE_END_WRITEBACK);
d8926bb3 1251 return -ENOMEM;
17ca04af 1252 }
82d5902d 1253
83eea1f1 1254 nolock = btrfs_is_free_space_inode(inode);
82d5902d
LZ		 1255
1256 if (nolock)
7a7eaa40 1257 trans = btrfs_join_transaction_nolock(root);
82d5902d 1258 else
7a7eaa40 1259 trans = btrfs_join_transaction(root);
ff5714cc 1260
79787eaa 1261 if (IS_ERR(trans)) {
c2790a2e
JB		 1262 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1263 EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc
JB		 1264 EXTENT_DO_ACCOUNTING |
1265 EXTENT_DEFRAG, PAGE_UNLOCK |
c2790a2e
JB		 1266 PAGE_CLEAR_DIRTY |
1267 PAGE_SET_WRITEBACK |
1268 PAGE_END_WRITEBACK);
79787eaa
JM		 1269 btrfs_free_path(path);
1270 return PTR_ERR(trans);
1271 }
1272
74b21075 1273 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
be20aa9d 1274
80ff3856
YZ		 1275 cow_start = (u64)-1;
1276 cur_offset = start;
1277 while (1) {
33345d01 1278 ret = btrfs_lookup_file_extent(trans, root, path, ino,
80ff3856 1279 cur_offset, 0);
d788a349 1280 if (ret < 0)
79787eaa 1281 goto error;
80ff3856
YZ		 1282 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1283 leaf = path->nodes[0];
1284 btrfs_item_key_to_cpu(leaf, &found_key,
1285 path->slots[0] - 1);
33345d01 1286 if (found_key.objectid == ino &&
80ff3856
YZ		 1287 found_key.type == BTRFS_EXTENT_DATA_KEY)
1288 path->slots[0]--;
1289 }
1290 check_prev = 0;
1291next_slot:
1292 leaf = path->nodes[0];
1293 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1294 ret = btrfs_next_leaf(root, path);
d788a349 1295 if (ret < 0)
79787eaa 1296 goto error;
80ff3856
YZ		 1297 if (ret > 0)
1298 break;
1299 leaf = path->nodes[0];
1300 }
be20aa9d 1301
80ff3856
YZ		 1302 nocow = 0;
1303 disk_bytenr = 0;
17d217fe 1304 num_bytes = 0;
80ff3856
YZ		 1305 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1306
1d512cb7
FM		 1307 if (found_key.objectid > ino)
1308 break;
1309 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1310 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1311 path->slots[0]++;
1312 goto next_slot;
1313 }
1314 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
80ff3856
YZ		 1315 found_key.offset > end)
1316 break;
1317
1318 if (found_key.offset > cur_offset) {
1319 extent_end = found_key.offset;
e9061e21 1320 extent_type = 0;
80ff3856
YZ		 1321 goto out_check;
1322 }
1323
1324 fi = btrfs_item_ptr(leaf, path->slots[0],
1325 struct btrfs_file_extent_item);
1326 extent_type = btrfs_file_extent_type(leaf, fi);
1327
cc95bef6 1328 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
d899e052
YZ		 1329 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1330 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856 1331 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5d4f98a2 1332 extent_offset = btrfs_file_extent_offset(leaf, fi);
80ff3856
YZ		 1333 extent_end = found_key.offset +
1334 btrfs_file_extent_num_bytes(leaf, fi);
b4939680
JB		 1335 disk_num_bytes =
1336 btrfs_file_extent_disk_num_bytes(leaf, fi);
80ff3856
YZ		 1337 if (extent_end <= start) {
1338 path->slots[0]++;
1339 goto next_slot;
1340 }
17d217fe
YZ		 1341 if (disk_bytenr == 0)
1342 goto out_check;
80ff3856
YZ		 1343 if (btrfs_file_extent_compression(leaf, fi) ||
1344 btrfs_file_extent_encryption(leaf, fi) ||
1345 btrfs_file_extent_other_encoding(leaf, fi))
1346 goto out_check;
d899e052
YZ		 1347 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1348 goto out_check;
d2fb3437 1349 if (btrfs_extent_readonly(root, disk_bytenr))
80ff3856 1350 goto out_check;
33345d01 1351 if (btrfs_cross_ref_exist(trans, root, ino,
5d4f98a2
YZ		 1352 found_key.offset -
1353 extent_offset, disk_bytenr))
17d217fe 1354 goto out_check;
5d4f98a2 1355 disk_bytenr += extent_offset;
17d217fe
YZ		 1356 disk_bytenr += cur_offset - found_key.offset;
1357 num_bytes = min(end + 1, extent_end) - cur_offset;
e9894fd3
WS		 1358 /*
1359 * if there are pending snapshots for this root,
1360 * we fall into common COW way.
1361 */
1362 if (!nolock) {
9ea24bbe 1363 err = btrfs_start_write_no_snapshoting(root);
e9894fd3
WS		 1364 if (!err)
1365 goto out_check;
1366 }
17d217fe
YZ		 1367 /*
1368 * force cow if csum exists in the range.
1369 * this ensure that csum for a given extent are
1370 * either valid or do not exist.
1371 */
1372 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1373 goto out_check;
80ff3856
YZ		 1374 nocow = 1;
1375 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1376 extent_end = found_key.offset +
514ac8ad
CM		 1377 btrfs_file_extent_inline_len(leaf,
1378 path->slots[0], fi);
80ff3856
YZ		 1379 extent_end = ALIGN(extent_end, root->sectorsize);
1380 } else {
1381 BUG_ON(1);
1382 }
1383out_check:
1384 if (extent_end <= start) {
1385 path->slots[0]++;
e9894fd3 1386 if (!nolock && nocow)
9ea24bbe 1387 btrfs_end_write_no_snapshoting(root);
80ff3856
YZ		 1388 goto next_slot;
1389 }
1390 if (!nocow) {
1391 if (cow_start == (u64)-1)
1392 cow_start = cur_offset;
1393 cur_offset = extent_end;
1394 if (cur_offset > end)
1395 break;
1396 path->slots[0]++;
1397 goto next_slot;
7ea394f1
YZ		 1398 }
1399
b3b4aa74 1400 btrfs_release_path(path);
80ff3856 1401 if (cow_start != (u64)-1) {
00361589
JB		 1402 ret = cow_file_range(inode, locked_page,
1403 cow_start, found_key.offset - 1,
1404 page_started, nr_written, 1);
e9894fd3
WS		 1405 if (ret) {
1406 if (!nolock && nocow)
9ea24bbe 1407 btrfs_end_write_no_snapshoting(root);
79787eaa 1408 goto error;
e9894fd3 1409 }
80ff3856 1410 cow_start = (u64)-1;
7ea394f1 1411 }
80ff3856 1412
d899e052
YZ		 1413 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1414 struct extent_map *em;
1415 struct extent_map_tree *em_tree;
1416 em_tree = &BTRFS_I(inode)->extent_tree;
172ddd60 1417 em = alloc_extent_map();
79787eaa 1418 BUG_ON(!em); /* -ENOMEM */
d899e052 1419 em->start = cur_offset;
70c8a91c 1420 em->orig_start = found_key.offset - extent_offset;
d899e052
YZ		 1421 em->len = num_bytes;
1422 em->block_len = num_bytes;
1423 em->block_start = disk_bytenr;
b4939680 1424 em->orig_block_len = disk_num_bytes;
cc95bef6 1425 em->ram_bytes = ram_bytes;
d899e052 1426 em->bdev = root->fs_info->fs_devices->latest_bdev;
2ab28f32
JB		 1427 em->mod_start = em->start;
1428 em->mod_len = em->len;
d899e052 1429 set_bit(EXTENT_FLAG_PINNED, &em->flags);
b11e234d 1430 set_bit(EXTENT_FLAG_FILLING, &em->flags);
70c8a91c 1431 em->generation = -1;
d899e052 1432 while (1) {
890871be 1433 write_lock(&em_tree->lock);
09a2a8f9 1434 ret = add_extent_mapping(em_tree, em, 1);
890871be 1435 write_unlock(&em_tree->lock);
d899e052
YZ		 1436 if (ret != -EEXIST) {
1437 free_extent_map(em);
1438 break;
1439 }
1440 btrfs_drop_extent_cache(inode, em->start,
1441 em->start + em->len - 1, 0);
1442 }
1443 type = BTRFS_ORDERED_PREALLOC;
1444 } else {
1445 type = BTRFS_ORDERED_NOCOW;
1446 }
80ff3856
YZ		 1447
1448 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052 1449 num_bytes, num_bytes, type);
79787eaa 1450 BUG_ON(ret); /* -ENOMEM */
771ed689 1451
efa56464
YZ		 1452 if (root->root_key.objectid ==
1453 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1454 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1455 num_bytes);
e9894fd3
WS		 1456 if (ret) {
1457 if (!nolock && nocow)
9ea24bbe 1458 btrfs_end_write_no_snapshoting(root);
79787eaa 1459 goto error;
e9894fd3 1460 }
efa56464
YZ		 1461 }
1462
c2790a2e
JB		 1463 extent_clear_unlock_delalloc(inode, cur_offset,
1464 cur_offset + num_bytes - 1,
1465 locked_page, EXTENT_LOCKED |
1466 EXTENT_DELALLOC, PAGE_UNLOCK |
1467 PAGE_SET_PRIVATE2);
e9894fd3 1468 if (!nolock && nocow)
9ea24bbe 1469 btrfs_end_write_no_snapshoting(root);
80ff3856
YZ		 1470 cur_offset = extent_end;
1471 if (cur_offset > end)
1472 break;
be20aa9d 1473 }
b3b4aa74 1474 btrfs_release_path(path);
80ff3856 1475
17ca04af 1476 if (cur_offset <= end && cow_start == (u64)-1) {
80ff3856 1477 cow_start = cur_offset;
17ca04af
JB		 1478 cur_offset = end;
1479 }
1480
80ff3856 1481 if (cow_start != (u64)-1) {
00361589
JB		 1482 ret = cow_file_range(inode, locked_page, cow_start, end,
1483 page_started, nr_written, 1);
d788a349 1484 if (ret)
79787eaa 1485 goto error;
80ff3856
YZ		 1486 }
1487
79787eaa 1488error:
a698d075 1489 err = btrfs_end_transaction(trans, root);
79787eaa
JM		 1490 if (!ret)
1491 ret = err;
1492
17ca04af 1493 if (ret && cur_offset < end)
c2790a2e
JB		 1494 extent_clear_unlock_delalloc(inode, cur_offset, end,
1495 locked_page, EXTENT_LOCKED |
151a41bc
JB		 1496 EXTENT_DELALLOC | EXTENT_DEFRAG |
1497 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1498 PAGE_CLEAR_DIRTY |
c2790a2e
JB		 1499 PAGE_SET_WRITEBACK |
1500 PAGE_END_WRITEBACK);
7ea394f1 1501 btrfs_free_path(path);
79787eaa 1502 return ret;
be20aa9d
CM		 1503}
1504
47059d93
WS		 1505static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1506{
1507
1508 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1509 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1510 return 0;
1511
1512 /*
1513 * @defrag_bytes is a hint value, no spinlock held here,
1514 * if is not zero, it means the file is defragging.
1515 * Force cow if given extent needs to be defragged.
1516 */
1517 if (BTRFS_I(inode)->defrag_bytes &&
1518 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1519 EXTENT_DEFRAG, 0, NULL))
1520 return 1;
1521
1522 return 0;
1523}
1524
d352ac68
CM		 1525/*
1526 * extent_io.c call back to do delayed allocation processing
1527 */
c8b97818 1528static int run_delalloc_range(struct inode *inode, struct page *locked_page,
771ed689
CM		 1529 u64 start, u64 end, int *page_started,
1530 unsigned long *nr_written)
be20aa9d 1531{
be20aa9d 1532 int ret;
47059d93 1533 int force_cow = need_force_cow(inode, start, end);
a2135011 1534
47059d93 1535 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
c8b97818 1536 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1537 page_started, 1, nr_written);
47059d93 1538 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
d899e052 1539 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1540 page_started, 0, nr_written);
7816030e 1541 } else if (!inode_need_compress(inode)) {
7f366cfe
CM		 1542 ret = cow_file_range(inode, locked_page, start, end,
1543 page_started, nr_written, 1);
7ddf5a42
JB		 1544 } else {
1545 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1546 &BTRFS_I(inode)->runtime_flags);
771ed689 1547 ret = cow_file_range_async(inode, locked_page, start, end,
d397712b 1548 page_started, nr_written);
7ddf5a42 1549 }
b888db2b
CM		 1550 return ret;
1551}
1552
1bf85046
JM		 1553static void btrfs_split_extent_hook(struct inode *inode,
1554 struct extent_state *orig, u64 split)
9ed74f2d 1555{
dcab6a3b
JB		 1556 u64 size;
1557
0ca1f7ce 1558 /* not delalloc, ignore it */
9ed74f2d 1559 if (!(orig->state & EXTENT_DELALLOC))
1bf85046 1560 return;
9ed74f2d 1561
dcab6a3b
JB		 1562 size = orig->end - orig->start + 1;
1563 if (size > BTRFS_MAX_EXTENT_SIZE) {
1564 u64 num_extents;
1565 u64 new_size;
1566
1567 /*
ba117213
JB		 1568 * See the explanation in btrfs_merge_extent_hook, the same
1569 * applies here, just in reverse.
dcab6a3b
JB		 1570 */
1571 new_size = orig->end - split + 1;
ba117213 1572 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
dcab6a3b 1573 BTRFS_MAX_EXTENT_SIZE);
ba117213
JB		 1574 new_size = split - orig->start;
1575 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1576 BTRFS_MAX_EXTENT_SIZE);
1577 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1578 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
dcab6a3b
JB		 1579 return;
1580 }
1581
9e0baf60
JB		 1582 spin_lock(&BTRFS_I(inode)->lock);
1583 BTRFS_I(inode)->outstanding_extents++;
1584 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1585}
1586
1587/*
1588 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1589 * extents so we can keep track of new extents that are just merged onto old
1590 * extents, such as when we are doing sequential writes, so we can properly
1591 * account for the metadata space we'll need.
1592 */
1bf85046
JM		 1593static void btrfs_merge_extent_hook(struct inode *inode,
1594 struct extent_state *new,
1595 struct extent_state *other)
9ed74f2d 1596{
dcab6a3b
JB		 1597 u64 new_size, old_size;
1598 u64 num_extents;
1599
9ed74f2d
JB		 1600 /* not delalloc, ignore it */
1601 if (!(other->state & EXTENT_DELALLOC))
1bf85046 1602 return;
9ed74f2d 1603
8461a3de
JB		 1604 if (new->start > other->start)
1605 new_size = new->end - other->start + 1;
1606 else
1607 new_size = other->end - new->start + 1;
dcab6a3b
JB		 1608
1609 /* we're not bigger than the max, unreserve the space and go */
1610 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1611 spin_lock(&BTRFS_I(inode)->lock);
1612 BTRFS_I(inode)->outstanding_extents--;
1613 spin_unlock(&BTRFS_I(inode)->lock);
1614 return;
1615 }
1616
1617 /*
ba117213
JB		 1618 * We have to add up either side to figure out how many extents were
1619 * accounted for before we merged into one big extent. If the number of
1620 * extents we accounted for is <= the amount we need for the new range
1621 * then we can return, otherwise drop. Think of it like this
1622 *
1623 * [ 4k][MAX_SIZE]
1624 *
1625 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1626 * need 2 outstanding extents, on one side we have 1 and the other side
1627 * we have 1 so they are == and we can return. But in this case
1628 *
1629 * [MAX_SIZE+4k][MAX_SIZE+4k]
1630 *
1631 * Each range on their own accounts for 2 extents, but merged together
1632 * they are only 3 extents worth of accounting, so we need to drop in
1633 * this case.
dcab6a3b 1634 */
ba117213 1635 old_size = other->end - other->start + 1;
dcab6a3b
JB		 1636 num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1637 BTRFS_MAX_EXTENT_SIZE);
ba117213
JB		 1638 old_size = new->end - new->start + 1;
1639 num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1640 BTRFS_MAX_EXTENT_SIZE);
1641
dcab6a3b 1642 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
ba117213 1643 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
dcab6a3b
JB		 1644 return;
1645
9e0baf60
JB		 1646 spin_lock(&BTRFS_I(inode)->lock);
1647 BTRFS_I(inode)->outstanding_extents--;
1648 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1649}
1650
eb73c1b7
MX		 1651static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1652 struct inode *inode)
1653{
1654 spin_lock(&root->delalloc_lock);
1655 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1656 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1657 &root->delalloc_inodes);
1658 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1659 &BTRFS_I(inode)->runtime_flags);
1660 root->nr_delalloc_inodes++;
1661 if (root->nr_delalloc_inodes == 1) {
1662 spin_lock(&root->fs_info->delalloc_root_lock);
1663 BUG_ON(!list_empty(&root->delalloc_root));
1664 list_add_tail(&root->delalloc_root,
1665 &root->fs_info->delalloc_roots);
1666 spin_unlock(&root->fs_info->delalloc_root_lock);
1667 }
1668 }
1669 spin_unlock(&root->delalloc_lock);
1670}
1671
1672static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1673 struct inode *inode)
1674{
1675 spin_lock(&root->delalloc_lock);
1676 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1677 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1678 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1679 &BTRFS_I(inode)->runtime_flags);
1680 root->nr_delalloc_inodes--;
1681 if (!root->nr_delalloc_inodes) {
1682 spin_lock(&root->fs_info->delalloc_root_lock);
1683 BUG_ON(list_empty(&root->delalloc_root));
1684 list_del_init(&root->delalloc_root);
1685 spin_unlock(&root->fs_info->delalloc_root_lock);
1686 }
1687 }
1688 spin_unlock(&root->delalloc_lock);
1689}
1690
d352ac68
CM		 1691/*
1692 * extent_io.c set_bit_hook, used to track delayed allocation
1693 * bytes in this file, and to maintain the list of inodes that
1694 * have pending delalloc work to be done.
1695 */
1bf85046 1696static void btrfs_set_bit_hook(struct inode *inode,
9ee49a04 1697 struct extent_state *state, unsigned *bits)
291d673e 1698{
9ed74f2d 1699
47059d93
WS		 1700 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1701 WARN_ON(1);
75eff68e
CM		 1702 /*
1703 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1704 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1705 * bit, which is only set or cleared with irqs on
1706 */
0ca1f7ce 1707 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1708 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1709 u64 len = state->end + 1 - state->start;
83eea1f1 1710 bool do_list = !btrfs_is_free_space_inode(inode);
9ed74f2d 1711
9e0baf60 1712 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1713 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1714 } else {
1715 spin_lock(&BTRFS_I(inode)->lock);
1716 BTRFS_I(inode)->outstanding_extents++;
1717 spin_unlock(&BTRFS_I(inode)->lock);
1718 }
287a0ab9 1719
6a3891c5
JB		 1720 /* For sanity tests */
1721 if (btrfs_test_is_dummy_root(root))
1722 return;
1723
963d678b
MX		 1724 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1725 root->fs_info->delalloc_batch);
df0af1a5 1726 spin_lock(&BTRFS_I(inode)->lock);
0ca1f7ce 1727 BTRFS_I(inode)->delalloc_bytes += len;
47059d93
WS		 1728 if (*bits & EXTENT_DEFRAG)
1729 BTRFS_I(inode)->defrag_bytes += len;
df0af1a5 1730 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
eb73c1b7
MX		 1731 &BTRFS_I(inode)->runtime_flags))
1732 btrfs_add_delalloc_inodes(root, inode);
df0af1a5 1733 spin_unlock(&BTRFS_I(inode)->lock);
291d673e 1734 }
291d673e
CM		 1735}
1736
d352ac68
CM		 1737/*
1738 * extent_io.c clear_bit_hook, see set_bit_hook for why
1739 */
1bf85046 1740static void btrfs_clear_bit_hook(struct inode *inode,
41074888 1741 struct extent_state *state,
9ee49a04 1742 unsigned *bits)
291d673e 1743{
47059d93 1744 u64 len = state->end + 1 - state->start;
dcab6a3b
JB		 1745 u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1746 BTRFS_MAX_EXTENT_SIZE);
47059d93
WS		 1747
1748 spin_lock(&BTRFS_I(inode)->lock);
1749 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1750 BTRFS_I(inode)->defrag_bytes -= len;
1751 spin_unlock(&BTRFS_I(inode)->lock);
1752
75eff68e
CM		 1753 /*
1754 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1755 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1756 * bit, which is only set or cleared with irqs on
1757 */
0ca1f7ce 1758 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1759 struct btrfs_root *root = BTRFS_I(inode)->root;
83eea1f1 1760 bool do_list = !btrfs_is_free_space_inode(inode);
bcbfce8a 1761
9e0baf60 1762 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1763 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1764 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1765 spin_lock(&BTRFS_I(inode)->lock);
dcab6a3b 1766 BTRFS_I(inode)->outstanding_extents -= num_extents;
9e0baf60
JB		 1767 spin_unlock(&BTRFS_I(inode)->lock);
1768 }
0ca1f7ce 1769
b6d08f06
JB		 1770 /*
1771 * We don't reserve metadata space for space cache inodes so we
1772 * don't need to call dellalloc_release_metadata if there is an
1773 * error.
1774 */
1775 if (*bits & EXTENT_DO_ACCOUNTING &&
1776 root != root->fs_info->tree_root)
0ca1f7ce
YZ		 1777 btrfs_delalloc_release_metadata(inode, len);
1778
6a3891c5
JB		 1779 /* For sanity tests. */
1780 if (btrfs_test_is_dummy_root(root))
1781 return;
1782
0cb59c99 1783 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
7ee9e440 1784 && do_list && !(state->state & EXTENT_NORESERVE))
51773bec
QW		 1785 btrfs_free_reserved_data_space_noquota(inode,
1786 state->start, len);
9ed74f2d 1787
963d678b
MX		 1788 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1789 root->fs_info->delalloc_batch);
df0af1a5 1790 spin_lock(&BTRFS_I(inode)->lock);
0ca1f7ce 1791 BTRFS_I(inode)->delalloc_bytes -= len;
0cb59c99 1792 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
df0af1a5 1793 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
eb73c1b7
MX		 1794 &BTRFS_I(inode)->runtime_flags))
1795 btrfs_del_delalloc_inode(root, inode);
df0af1a5 1796 spin_unlock(&BTRFS_I(inode)->lock);
291d673e 1797 }
291d673e
CM		 1798}
1799
d352ac68
CM		 1800/*
1801 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1802 * we don't create bios that span stripes or chunks
1803 */
64a16701 1804int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
c8b97818
CM		 1805 size_t size, struct bio *bio,
1806 unsigned long bio_flags)
239b14b3
CM		 1807{
1808 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
4f024f37 1809 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
239b14b3
CM		 1810 u64 length = 0;
1811 u64 map_length;
239b14b3
CM		 1812 int ret;
1813
771ed689
CM		 1814 if (bio_flags & EXTENT_BIO_COMPRESSED)
1815 return 0;
1816
4f024f37 1817 length = bio->bi_iter.bi_size;
239b14b3 1818 map_length = length;
64a16701 1819 ret = btrfs_map_block(root->fs_info, rw, logical,
f188591e 1820 &map_length, NULL, 0);
3ec706c8 1821 /* Will always return 0 with map_multi == NULL */
3444a972 1822 BUG_ON(ret < 0);
d397712b 1823 if (map_length < length + size)
239b14b3 1824 return 1;
3444a972 1825 return 0;
239b14b3
CM		 1826}
1827
d352ac68
CM		 1828/*
1829 * in order to insert checksums into the metadata in large chunks,
1830 * we wait until bio submission time. All the pages in the bio are
1831 * checksummed and sums are attached onto the ordered extent record.
1832 *
1833 * At IO completion time the cums attached on the ordered extent record
1834 * are inserted into the btree
1835 */
d397712b
CM		 1836static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1837 struct bio *bio, int mirror_num,
eaf25d93
CM		 1838 unsigned long bio_flags,
1839 u64 bio_offset)
065631f6 1840{
065631f6 1841 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 1842 int ret = 0;
e015640f 1843
d20f7043 1844 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
79787eaa 1845 BUG_ON(ret); /* -ENOMEM */
4a69a410
CM		 1846 return 0;
1847}
e015640f 1848
4a69a410
CM		 1849/*
1850 * in order to insert checksums into the metadata in large chunks,
1851 * we wait until bio submission time. All the pages in the bio are
1852 * checksummed and sums are attached onto the ordered extent record.
1853 *
1854 * At IO completion time the cums attached on the ordered extent record
1855 * are inserted into the btree
1856 */
b2950863 1857static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1858 int mirror_num, unsigned long bio_flags,
1859 u64 bio_offset)
4a69a410
CM		 1860{
1861 struct btrfs_root *root = BTRFS_I(inode)->root;
61891923
SB		 1862 int ret;
1863
1864 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
4246a0b6
CH		 1865 if (ret) {
1866 bio->bi_error = ret;
1867 bio_endio(bio);
1868 }
61891923 1869 return ret;
44b8bd7e
CM		 1870}
1871
d352ac68 1872/*
cad321ad
CM		 1873 * extent_io.c submission hook. This does the right thing for csum calculation
1874 * on write, or reading the csums from the tree before a read
d352ac68 1875 */
b2950863 1876static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1877 int mirror_num, unsigned long bio_flags,
1878 u64 bio_offset)
44b8bd7e
CM		 1879{
1880 struct btrfs_root *root = BTRFS_I(inode)->root;
0d51e28a 1881 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
44b8bd7e 1882 int ret = 0;
19b9bdb0 1883 int skip_sum;
b812ce28 1884 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
44b8bd7e 1885
6cbff00f 1886 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
cad321ad 1887
83eea1f1 1888 if (btrfs_is_free_space_inode(inode))
0d51e28a 1889 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
0417341e 1890
7b6d91da 1891 if (!(rw & REQ_WRITE)) {
5fd02043
JB		 1892 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1893 if (ret)
61891923 1894 goto out;
5fd02043 1895
d20f7043 1896 if (bio_flags & EXTENT_BIO_COMPRESSED) {
61891923
SB		 1897 ret = btrfs_submit_compressed_read(inode, bio,
1898 mirror_num,
1899 bio_flags);
1900 goto out;
c2db1073
TI		 1901 } else if (!skip_sum) {
1902 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1903 if (ret)
61891923 1904 goto out;
c2db1073 1905 }
4d1b5fb4 1906 goto mapit;
b812ce28 1907 } else if (async && !skip_sum) {
17d217fe
YZ		 1908 /* csum items have already been cloned */
1909 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1910 goto mapit;
19b9bdb0 1911 /* we're doing a write, do the async checksumming */
61891923 1912 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 1913 inode, rw, bio, mirror_num,
eaf25d93
CM		 1914 bio_flags, bio_offset,
1915 __btrfs_submit_bio_start,
4a69a410 1916 __btrfs_submit_bio_done);
61891923 1917 goto out;
b812ce28
JB		 1918 } else if (!skip_sum) {
1919 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1920 if (ret)
1921 goto out;
19b9bdb0
CM		 1922 }
1923
0b86a832 1924mapit:
61891923
SB		 1925 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1926
1927out:
4246a0b6
CH		 1928 if (ret < 0) {
1929 bio->bi_error = ret;
1930 bio_endio(bio);
1931 }
61891923 1932 return ret;
065631f6 1933}
6885f308 1934
d352ac68
CM		 1935/*
1936 * given a list of ordered sums record them in the inode. This happens
1937 * at IO completion time based on sums calculated at bio submission time.
1938 */
ba1da2f4 1939static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM		 1940 struct inode *inode, u64 file_offset,
1941 struct list_head *list)
1942{
e6dcd2dc
CM		 1943 struct btrfs_ordered_sum *sum;
1944
c6e30871 1945 list_for_each_entry(sum, list, list) {
39847c4d 1946 trans->adding_csums = 1;
d20f7043
CM		 1947 btrfs_csum_file_blocks(trans,
1948 BTRFS_I(inode)->root->fs_info->csum_root, sum);
39847c4d 1949 trans->adding_csums = 0;
e6dcd2dc
CM		 1950 }
1951 return 0;
1952}
1953
2ac55d41
JB		 1954int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1955 struct extent_state **cached_state)
ea8c2819 1956{
6c1500f2 1957 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
ea8c2819 1958 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2ac55d41 1959 cached_state, GFP_NOFS);
ea8c2819
CM		 1960}
1961
d352ac68 1962/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM		 1963struct btrfs_writepage_fixup {
1964 struct page *page;
1965 struct btrfs_work work;
1966};
1967
b2950863 1968static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
247e743c
CM		 1969{
1970 struct btrfs_writepage_fixup *fixup;
1971 struct btrfs_ordered_extent *ordered;
2ac55d41 1972 struct extent_state *cached_state = NULL;
247e743c
CM		 1973 struct page *page;
1974 struct inode *inode;
1975 u64 page_start;
1976 u64 page_end;
87826df0 1977 int ret;
247e743c
CM		 1978
1979 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1980 page = fixup->page;
4a096752 1981again:
247e743c
CM		 1982 lock_page(page);
1983 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1984 ClearPageChecked(page);
1985 goto out_page;
1986 }
1987
1988 inode = page->mapping->host;
1989 page_start = page_offset(page);
1990 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1991
2ac55d41 1992 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
d0082371 1993 &cached_state);
4a096752
CM		 1994
1995 /* already ordered? We're done */
8b62b72b 1996 if (PagePrivate2(page))
247e743c 1997 goto out;
4a096752
CM		 1998
1999 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2000 if (ordered) {
2ac55d41
JB		 2001 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2002 page_end, &cached_state, GFP_NOFS);
4a096752
CM		 2003 unlock_page(page);
2004 btrfs_start_ordered_extent(inode, ordered, 1);
87826df0 2005 btrfs_put_ordered_extent(ordered);
4a096752
CM		 2006 goto again;
2007 }
247e743c 2008
7cf5b976
QW		 2009 ret = btrfs_delalloc_reserve_space(inode, page_start,
2010 PAGE_CACHE_SIZE);
87826df0
JM		 2011 if (ret) {
2012 mapping_set_error(page->mapping, ret);
2013 end_extent_writepage(page, ret, page_start, page_end);
2014 ClearPageChecked(page);
2015 goto out;
2016 }
2017
2ac55d41 2018 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
247e743c 2019 ClearPageChecked(page);
87826df0 2020 set_page_dirty(page);
247e743c 2021out:
2ac55d41
JB		 2022 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2023 &cached_state, GFP_NOFS);
247e743c
CM		 2024out_page:
2025 unlock_page(page);
2026 page_cache_release(page);
b897abec 2027 kfree(fixup);
247e743c
CM		 2028}
2029
2030/*
2031 * There are a few paths in the higher layers of the kernel that directly
2032 * set the page dirty bit without asking the filesystem if it is a
2033 * good idea. This causes problems because we want to make sure COW
2034 * properly happens and the data=ordered rules are followed.
2035 *
c8b97818 2036 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 2037 * hasn't been properly setup for IO. We kick off an async process
2038 * to fix it up. The async helper will wait for ordered extents, set
2039 * the delalloc bit and make it safe to write the page.
2040 */
b2950863 2041static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
247e743c
CM		 2042{
2043 struct inode *inode = page->mapping->host;
2044 struct btrfs_writepage_fixup *fixup;
2045 struct btrfs_root *root = BTRFS_I(inode)->root;
247e743c 2046
8b62b72b
CM		 2047 /* this page is properly in the ordered list */
2048 if (TestClearPagePrivate2(page))
247e743c
CM		 2049 return 0;
2050
2051 if (PageChecked(page))
2052 return -EAGAIN;
2053
2054 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2055 if (!fixup)
2056 return -EAGAIN;
f421950f 2057
247e743c
CM		 2058 SetPageChecked(page);
2059 page_cache_get(page);
9e0af237
LB		 2060 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2061 btrfs_writepage_fixup_worker, NULL, NULL);
247e743c 2062 fixup->page = page;
dc6e3209 2063 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
87826df0 2064 return -EBUSY;
247e743c
CM		 2065}
2066
d899e052
YZ		 2067static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2068 struct inode *inode, u64 file_pos,
2069 u64 disk_bytenr, u64 disk_num_bytes,
2070 u64 num_bytes, u64 ram_bytes,
2071 u8 compression, u8 encryption,
2072 u16 other_encoding, int extent_type)
2073{
2074 struct btrfs_root *root = BTRFS_I(inode)->root;
2075 struct btrfs_file_extent_item *fi;
2076 struct btrfs_path *path;
2077 struct extent_buffer *leaf;
2078 struct btrfs_key ins;
1acae57b 2079 int extent_inserted = 0;
d899e052
YZ		 2080 int ret;
2081
2082 path = btrfs_alloc_path();
d8926bb3
MF		 2083 if (!path)
2084 return -ENOMEM;
d899e052 2085
a1ed835e
CM		 2086 /*
2087 * we may be replacing one extent in the tree with another.
2088 * The new extent is pinned in the extent map, and we don't want
2089 * to drop it from the cache until it is completely in the btree.
2090 *
2091 * So, tell btrfs_drop_extents to leave this extent in the cache.
2092 * the caller is expected to unpin it and allow it to be merged
2093 * with the others.
2094 */
1acae57b
FDBM		 2095 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2096 file_pos + num_bytes, NULL, 0,
2097 1, sizeof(*fi), &extent_inserted);
79787eaa
JM		 2098 if (ret)
2099 goto out;
d899e052 2100
1acae57b
FDBM		 2101 if (!extent_inserted) {
2102 ins.objectid = btrfs_ino(inode);
2103 ins.offset = file_pos;
2104 ins.type = BTRFS_EXTENT_DATA_KEY;
2105
2106 path->leave_spinning = 1;
2107 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2108 sizeof(*fi));
2109 if (ret)
2110 goto out;
2111 }
d899e052
YZ		 2112 leaf = path->nodes[0];
2113 fi = btrfs_item_ptr(leaf, path->slots[0],
2114 struct btrfs_file_extent_item);
2115 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2116 btrfs_set_file_extent_type(leaf, fi, extent_type);
2117 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2118 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2119 btrfs_set_file_extent_offset(leaf, fi, 0);
2120 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2121 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2122 btrfs_set_file_extent_compression(leaf, fi, compression);
2123 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2124 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
b9473439 2125
d899e052 2126 btrfs_mark_buffer_dirty(leaf);
ce195332 2127 btrfs_release_path(path);
d899e052
YZ		 2128
2129 inode_add_bytes(inode, num_bytes);
d899e052
YZ		 2130
2131 ins.objectid = disk_bytenr;
2132 ins.offset = disk_num_bytes;
2133 ins.type = BTRFS_EXTENT_ITEM_KEY;
5d4f98a2
YZ		 2134 ret = btrfs_alloc_reserved_file_extent(trans, root,
2135 root->root_key.objectid,
5846a3c2
QW		 2136 btrfs_ino(inode), file_pos,
2137 ram_bytes, &ins);
297d750b 2138 /*
5846a3c2
QW		 2139 * Release the reserved range from inode dirty range map, as it is
2140 * already moved into delayed_ref_head
297d750b
QW		 2141 */
2142 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
79787eaa 2143out:
d899e052 2144 btrfs_free_path(path);
b9473439 2145
79787eaa 2146 return ret;
d899e052
YZ		 2147}
2148
38c227d8
LB		 2149/* snapshot-aware defrag */
2150struct sa_defrag_extent_backref {
2151 struct rb_node node;
2152 struct old_sa_defrag_extent *old;
2153 u64 root_id;
2154 u64 inum;
2155 u64 file_pos;
2156 u64 extent_offset;
2157 u64 num_bytes;
2158 u64 generation;
2159};
2160
2161struct old_sa_defrag_extent {
2162 struct list_head list;
2163 struct new_sa_defrag_extent *new;
2164
2165 u64 extent_offset;
2166 u64 bytenr;
2167 u64 offset;
2168 u64 len;
2169 int count;
2170};
2171
2172struct new_sa_defrag_extent {
2173 struct rb_root root;
2174 struct list_head head;
2175 struct btrfs_path *path;
2176 struct inode *inode;
2177 u64 file_pos;
2178 u64 len;
2179 u64 bytenr;
2180 u64 disk_len;
2181 u8 compress_type;
2182};
2183
2184static int backref_comp(struct sa_defrag_extent_backref *b1,
2185 struct sa_defrag_extent_backref *b2)
2186{
2187 if (b1->root_id < b2->root_id)
2188 return -1;
2189 else if (b1->root_id > b2->root_id)
2190 return 1;
2191
2192 if (b1->inum < b2->inum)
2193 return -1;
2194 else if (b1->inum > b2->inum)
2195 return 1;
2196
2197 if (b1->file_pos < b2->file_pos)
2198 return -1;
2199 else if (b1->file_pos > b2->file_pos)
2200 return 1;
2201
2202 /*
2203 * [------------------------------] ===> (a range of space)
2204 * |<--->| |<---->| =============> (fs/file tree A)
2205 * |<---------------------------->| ===> (fs/file tree B)
2206 *
2207 * A range of space can refer to two file extents in one tree while
2208 * refer to only one file extent in another tree.
2209 *
2210 * So we may process a disk offset more than one time(two extents in A)
2211 * and locate at the same extent(one extent in B), then insert two same
2212 * backrefs(both refer to the extent in B).
2213 */
2214 return 0;
2215}
2216
2217static void backref_insert(struct rb_root *root,
2218 struct sa_defrag_extent_backref *backref)
2219{
2220 struct rb_node **p = &root->rb_node;
2221 struct rb_node *parent = NULL;
2222 struct sa_defrag_extent_backref *entry;
2223 int ret;
2224
2225 while (*p) {
2226 parent = *p;
2227 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2228
2229 ret = backref_comp(backref, entry);
2230 if (ret < 0)
2231 p = &(*p)->rb_left;
2232 else
2233 p = &(*p)->rb_right;
2234 }
2235
2236 rb_link_node(&backref->node, parent, p);
2237 rb_insert_color(&backref->node, root);
2238}
2239
2240/*
2241 * Note the backref might has changed, and in this case we just return 0.
2242 */
2243static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2244 void *ctx)
2245{
2246 struct btrfs_file_extent_item *extent;
2247 struct btrfs_fs_info *fs_info;
2248 struct old_sa_defrag_extent *old = ctx;
2249 struct new_sa_defrag_extent *new = old->new;
2250 struct btrfs_path *path = new->path;
2251 struct btrfs_key key;
2252 struct btrfs_root *root;
2253 struct sa_defrag_extent_backref *backref;
2254 struct extent_buffer *leaf;
2255 struct inode *inode = new->inode;
2256 int slot;
2257 int ret;
2258 u64 extent_offset;
2259 u64 num_bytes;
2260
2261 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2262 inum == btrfs_ino(inode))
2263 return 0;
2264
2265 key.objectid = root_id;
2266 key.type = BTRFS_ROOT_ITEM_KEY;
2267 key.offset = (u64)-1;
2268
2269 fs_info = BTRFS_I(inode)->root->fs_info;
2270 root = btrfs_read_fs_root_no_name(fs_info, &key);
2271 if (IS_ERR(root)) {
2272 if (PTR_ERR(root) == -ENOENT)
2273 return 0;
2274 WARN_ON(1);
2275 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2276 inum, offset, root_id);
2277 return PTR_ERR(root);
2278 }
2279
2280 key.objectid = inum;
2281 key.type = BTRFS_EXTENT_DATA_KEY;
2282 if (offset > (u64)-1 << 32)
2283 key.offset = 0;
2284 else
2285 key.offset = offset;
2286
2287 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
fae7f21c 2288 if (WARN_ON(ret < 0))
38c227d8 2289 return ret;
50f1319c 2290 ret = 0;
38c227d8
LB		 2291
2292 while (1) {
2293 cond_resched();
2294
2295 leaf = path->nodes[0];
2296 slot = path->slots[0];
2297
2298 if (slot >= btrfs_header_nritems(leaf)) {
2299 ret = btrfs_next_leaf(root, path);
2300 if (ret < 0) {
2301 goto out;
2302 } else if (ret > 0) {
2303 ret = 0;
2304 goto out;
2305 }
2306 continue;
2307 }
2308
2309 path->slots[0]++;
2310
2311 btrfs_item_key_to_cpu(leaf, &key, slot);
2312
2313 if (key.objectid > inum)
2314 goto out;
2315
2316 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2317 continue;
2318
2319 extent = btrfs_item_ptr(leaf, slot,
2320 struct btrfs_file_extent_item);
2321
2322 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2323 continue;
2324
e68afa49
LB		 2325 /*
2326 * 'offset' refers to the exact key.offset,
2327 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2328 * (key.offset - extent_offset).
2329 */
2330 if (key.offset != offset)
38c227d8
LB		 2331 continue;
2332
e68afa49 2333 extent_offset = btrfs_file_extent_offset(leaf, extent);
38c227d8 2334 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
e68afa49 2335
38c227d8
LB		 2336 if (extent_offset >= old->extent_offset + old->offset +
2337 old->len || extent_offset + num_bytes <=
2338 old->extent_offset + old->offset)
2339 continue;
38c227d8
LB		 2340 break;
2341 }
2342
2343 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2344 if (!backref) {
2345 ret = -ENOENT;
2346 goto out;
2347 }
2348
2349 backref->root_id = root_id;
2350 backref->inum = inum;
e68afa49 2351 backref->file_pos = offset;
38c227d8
LB		 2352 backref->num_bytes = num_bytes;
2353 backref->extent_offset = extent_offset;
2354 backref->generation = btrfs_file_extent_generation(leaf, extent);
2355 backref->old = old;
2356 backref_insert(&new->root, backref);
2357 old->count++;
2358out:
2359 btrfs_release_path(path);
2360 WARN_ON(ret);
2361 return ret;
2362}
2363
2364static noinline bool record_extent_backrefs(struct btrfs_path *path,
2365 struct new_sa_defrag_extent *new)
2366{
2367 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2368 struct old_sa_defrag_extent *old, *tmp;
2369 int ret;
2370
2371 new->path = path;
2372
2373 list_for_each_entry_safe(old, tmp, &new->head, list) {
e68afa49
LB		 2374 ret = iterate_inodes_from_logical(old->bytenr +
2375 old->extent_offset, fs_info,
38c227d8
LB		 2376 path, record_one_backref,
2377 old);
4724b106
JB		 2378 if (ret < 0 && ret != -ENOENT)
2379 return false;
38c227d8
LB		 2380
2381 /* no backref to be processed for this extent */
2382 if (!old->count) {
2383 list_del(&old->list);
2384 kfree(old);
2385 }
2386 }
2387
2388 if (list_empty(&new->head))
2389 return false;
2390
2391 return true;
2392}
2393
2394static int relink_is_mergable(struct extent_buffer *leaf,
2395 struct btrfs_file_extent_item *fi,
116e0024 2396 struct new_sa_defrag_extent *new)
38c227d8 2397{
116e0024 2398 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
38c227d8
LB		 2399 return 0;
2400
2401 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2402 return 0;
2403
116e0024
LB		 2404 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2405 return 0;
2406
2407 if (btrfs_file_extent_encryption(leaf, fi) ||
38c227d8
LB		 2408 btrfs_file_extent_other_encoding(leaf, fi))
2409 return 0;
2410
2411 return 1;
2412}
2413
2414/*
2415 * Note the backref might has changed, and in this case we just return 0.
2416 */
2417static noinline int relink_extent_backref(struct btrfs_path *path,
2418 struct sa_defrag_extent_backref *prev,
2419 struct sa_defrag_extent_backref *backref)
2420{
2421 struct btrfs_file_extent_item *extent;
2422 struct btrfs_file_extent_item *item;
2423 struct btrfs_ordered_extent *ordered;
2424 struct btrfs_trans_handle *trans;
2425 struct btrfs_fs_info *fs_info;
2426 struct btrfs_root *root;
2427 struct btrfs_key key;
2428 struct extent_buffer *leaf;
2429 struct old_sa_defrag_extent *old = backref->old;
2430 struct new_sa_defrag_extent *new = old->new;
2431 struct inode *src_inode = new->inode;
2432 struct inode *inode;
2433 struct extent_state *cached = NULL;
2434 int ret = 0;
2435 u64 start;
2436 u64 len;
2437 u64 lock_start;
2438 u64 lock_end;
2439 bool merge = false;
2440 int index;
2441
2442 if (prev && prev->root_id == backref->root_id &&
2443 prev->inum == backref->inum &&
2444 prev->file_pos + prev->num_bytes == backref->file_pos)
2445 merge = true;
2446
2447 /* step 1: get root */
2448 key.objectid = backref->root_id;
2449 key.type = BTRFS_ROOT_ITEM_KEY;
2450 key.offset = (u64)-1;
2451
2452 fs_info = BTRFS_I(src_inode)->root->fs_info;
2453 index = srcu_read_lock(&fs_info->subvol_srcu);
2454
2455 root = btrfs_read_fs_root_no_name(fs_info, &key);
2456 if (IS_ERR(root)) {
2457 srcu_read_unlock(&fs_info->subvol_srcu, index);
2458 if (PTR_ERR(root) == -ENOENT)
2459 return 0;
2460 return PTR_ERR(root);
2461 }
38c227d8 2462
bcbba5e6
WS		 2463 if (btrfs_root_readonly(root)) {
2464 srcu_read_unlock(&fs_info->subvol_srcu, index);
2465 return 0;
2466 }
2467
38c227d8
LB		 2468 /* step 2: get inode */
2469 key.objectid = backref->inum;
2470 key.type = BTRFS_INODE_ITEM_KEY;
2471 key.offset = 0;
2472
2473 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2474 if (IS_ERR(inode)) {
2475 srcu_read_unlock(&fs_info->subvol_srcu, index);
2476 return 0;
2477 }
2478
2479 srcu_read_unlock(&fs_info->subvol_srcu, index);
2480
2481 /* step 3: relink backref */
2482 lock_start = backref->file_pos;
2483 lock_end = backref->file_pos + backref->num_bytes - 1;
2484 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2485 0, &cached);
2486
2487 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2488 if (ordered) {
2489 btrfs_put_ordered_extent(ordered);
2490 goto out_unlock;
2491 }
2492
2493 trans = btrfs_join_transaction(root);
2494 if (IS_ERR(trans)) {
2495 ret = PTR_ERR(trans);
2496 goto out_unlock;
2497 }
2498
2499 key.objectid = backref->inum;
2500 key.type = BTRFS_EXTENT_DATA_KEY;
2501 key.offset = backref->file_pos;
2502
2503 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2504 if (ret < 0) {
2505 goto out_free_path;
2506 } else if (ret > 0) {
2507 ret = 0;
2508 goto out_free_path;
2509 }
2510
2511 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2512 struct btrfs_file_extent_item);
2513
2514 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2515 backref->generation)
2516 goto out_free_path;
2517
2518 btrfs_release_path(path);
2519
2520 start = backref->file_pos;
2521 if (backref->extent_offset < old->extent_offset + old->offset)
2522 start += old->extent_offset + old->offset -
2523 backref->extent_offset;
2524
2525 len = min(backref->extent_offset + backref->num_bytes,
2526 old->extent_offset + old->offset + old->len);
2527 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2528
2529 ret = btrfs_drop_extents(trans, root, inode, start,
2530 start + len, 1);
2531 if (ret)
2532 goto out_free_path;
2533again:
2534 key.objectid = btrfs_ino(inode);
2535 key.type = BTRFS_EXTENT_DATA_KEY;
2536 key.offset = start;
2537
a09a0a70 2538 path->leave_spinning = 1;
38c227d8
LB		 2539 if (merge) {
2540 struct btrfs_file_extent_item *fi;
2541 u64 extent_len;
2542 struct btrfs_key found_key;
2543
3c9665df 2544 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
38c227d8
LB		 2545 if (ret < 0)
2546 goto out_free_path;
2547
2548 path->slots[0]--;
2549 leaf = path->nodes[0];
2550 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2551
2552 fi = btrfs_item_ptr(leaf, path->slots[0],
2553 struct btrfs_file_extent_item);
2554 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2555
116e0024
LB		 2556 if (extent_len + found_key.offset == start &&
2557 relink_is_mergable(leaf, fi, new)) {
38c227d8
LB		 2558 btrfs_set_file_extent_num_bytes(leaf, fi,
2559 extent_len + len);
2560 btrfs_mark_buffer_dirty(leaf);
2561 inode_add_bytes(inode, len);
2562
2563 ret = 1;
2564 goto out_free_path;
2565 } else {
2566 merge = false;
2567 btrfs_release_path(path);
2568 goto again;
2569 }
2570 }
2571
2572 ret = btrfs_insert_empty_item(trans, root, path, &key,
2573 sizeof(*extent));
2574 if (ret) {
2575 btrfs_abort_transaction(trans, root, ret);
2576 goto out_free_path;
2577 }
2578
2579 leaf = path->nodes[0];
2580 item = btrfs_item_ptr(leaf, path->slots[0],
2581 struct btrfs_file_extent_item);
2582 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2583 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2584 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2585 btrfs_set_file_extent_num_bytes(leaf, item, len);
2586 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2587 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2588 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2589 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2590 btrfs_set_file_extent_encryption(leaf, item, 0);
2591 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2592
2593 btrfs_mark_buffer_dirty(leaf);
2594 inode_add_bytes(inode, len);
a09a0a70 2595 btrfs_release_path(path);
38c227d8
LB		 2596
2597 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2598 new->disk_len, 0,
2599 backref->root_id, backref->inum,
b06c4bf5 2600 new->file_pos); /* start - extent_offset */
38c227d8
LB		 2601 if (ret) {
2602 btrfs_abort_transaction(trans, root, ret);
2603 goto out_free_path;
2604 }
2605
2606 ret = 1;
2607out_free_path:
2608 btrfs_release_path(path);
a09a0a70 2609 path->leave_spinning = 0;
38c227d8
LB		 2610 btrfs_end_transaction(trans, root);
2611out_unlock:
2612 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2613 &cached, GFP_NOFS);
2614 iput(inode);
2615 return ret;
2616}
2617
6f519564
LB		 2618static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2619{
2620 struct old_sa_defrag_extent *old, *tmp;
2621
2622 if (!new)
2623 return;
2624
2625 list_for_each_entry_safe(old, tmp, &new->head, list) {
6f519564
LB		 2626 kfree(old);
2627 }
2628 kfree(new);
2629}
2630
38c227d8
LB		 2631static void relink_file_extents(struct new_sa_defrag_extent *new)
2632{
2633 struct btrfs_path *path;
38c227d8
LB		 2634 struct sa_defrag_extent_backref *backref;
2635 struct sa_defrag_extent_backref *prev = NULL;
2636 struct inode *inode;
2637 struct btrfs_root *root;
2638 struct rb_node *node;
2639 int ret;
2640
2641 inode = new->inode;
2642 root = BTRFS_I(inode)->root;
2643
2644 path = btrfs_alloc_path();
2645 if (!path)
2646 return;
2647
2648 if (!record_extent_backrefs(path, new)) {
2649 btrfs_free_path(path);
2650 goto out;
2651 }
2652 btrfs_release_path(path);
2653
2654 while (1) {
2655 node = rb_first(&new->root);
2656 if (!node)
2657 break;
2658 rb_erase(node, &new->root);
2659
2660 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2661
2662 ret = relink_extent_backref(path, prev, backref);
2663 WARN_ON(ret < 0);
2664
2665 kfree(prev);
2666
2667 if (ret == 1)
2668 prev = backref;
2669 else
2670 prev = NULL;
2671 cond_resched();
2672 }
2673 kfree(prev);
2674
2675 btrfs_free_path(path);
38c227d8 2676out:
6f519564
LB		 2677 free_sa_defrag_extent(new);
2678
38c227d8
LB		 2679 atomic_dec(&root->fs_info->defrag_running);
2680 wake_up(&root->fs_info->transaction_wait);
38c227d8
LB		 2681}
2682
2683static struct new_sa_defrag_extent *
2684record_old_file_extents(struct inode *inode,
2685 struct btrfs_ordered_extent *ordered)
2686{
2687 struct btrfs_root *root = BTRFS_I(inode)->root;
2688 struct btrfs_path *path;
2689 struct btrfs_key key;
6f519564 2690 struct old_sa_defrag_extent *old;
38c227d8
LB		 2691 struct new_sa_defrag_extent *new;
2692 int ret;
2693
2694 new = kmalloc(sizeof(*new), GFP_NOFS);
2695 if (!new)
2696 return NULL;
2697
2698 new->inode = inode;
2699 new->file_pos = ordered->file_offset;
2700 new->len = ordered->len;
2701 new->bytenr = ordered->start;
2702 new->disk_len = ordered->disk_len;
2703 new->compress_type = ordered->compress_type;
2704 new->root = RB_ROOT;
2705 INIT_LIST_HEAD(&new->head);
2706
2707 path = btrfs_alloc_path();
2708 if (!path)
2709 goto out_kfree;
2710
2711 key.objectid = btrfs_ino(inode);
2712 key.type = BTRFS_EXTENT_DATA_KEY;
2713 key.offset = new->file_pos;
2714
2715 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2716 if (ret < 0)
2717 goto out_free_path;
2718 if (ret > 0 && path->slots[0] > 0)
2719 path->slots[0]--;
2720
2721 /* find out all the old extents for the file range */
2722 while (1) {
2723 struct btrfs_file_extent_item *extent;
2724 struct extent_buffer *l;
2725 int slot;
2726 u64 num_bytes;
2727 u64 offset;
2728 u64 end;
2729 u64 disk_bytenr;
2730 u64 extent_offset;
2731
2732 l = path->nodes[0];
2733 slot = path->slots[0];
2734
2735 if (slot >= btrfs_header_nritems(l)) {
2736 ret = btrfs_next_leaf(root, path);
2737 if (ret < 0)
6f519564 2738 goto out_free_path;
38c227d8
LB		 2739 else if (ret > 0)
2740 break;
2741 continue;
2742 }
2743
2744 btrfs_item_key_to_cpu(l, &key, slot);
2745
2746 if (key.objectid != btrfs_ino(inode))
2747 break;
2748 if (key.type != BTRFS_EXTENT_DATA_KEY)
2749 break;
2750 if (key.offset >= new->file_pos + new->len)
2751 break;
2752
2753 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2754
2755 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2756 if (key.offset + num_bytes < new->file_pos)
2757 goto next;
2758
2759 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2760 if (!disk_bytenr)
2761 goto next;
2762
2763 extent_offset = btrfs_file_extent_offset(l, extent);
2764
2765 old = kmalloc(sizeof(*old), GFP_NOFS);
2766 if (!old)
6f519564 2767 goto out_free_path;
38c227d8
LB		 2768
2769 offset = max(new->file_pos, key.offset);
2770 end = min(new->file_pos + new->len, key.offset + num_bytes);
2771
2772 old->bytenr = disk_bytenr;
2773 old->extent_offset = extent_offset;
2774 old->offset = offset - key.offset;
2775 old->len = end - offset;
2776 old->new = new;
2777 old->count = 0;
2778 list_add_tail(&old->list, &new->head);
2779next:
2780 path->slots[0]++;
2781 cond_resched();
2782 }
2783
2784 btrfs_free_path(path);
2785 atomic_inc(&root->fs_info->defrag_running);
2786
2787 return new;
2788
38c227d8
LB		 2789out_free_path:
2790 btrfs_free_path(path);
2791out_kfree:
6f519564 2792 free_sa_defrag_extent(new);
38c227d8
LB		 2793 return NULL;
2794}
2795
e570fd27
MX		 2796static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2797 u64 start, u64 len)
2798{
2799 struct btrfs_block_group_cache *cache;
2800
2801 cache = btrfs_lookup_block_group(root->fs_info, start);
2802 ASSERT(cache);
2803
2804 spin_lock(&cache->lock);
2805 cache->delalloc_bytes -= len;
2806 spin_unlock(&cache->lock);
2807
2808 btrfs_put_block_group(cache);
2809}
2810
d352ac68
CM		 2811/* as ordered data IO finishes, this gets called so we can finish
2812 * an ordered extent if the range of bytes in the file it covers are
2813 * fully written.
2814 */
5fd02043 2815static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
e6dcd2dc 2816{
5fd02043 2817 struct inode *inode = ordered_extent->inode;
e6dcd2dc 2818 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 2819 struct btrfs_trans_handle *trans = NULL;
e6dcd2dc 2820 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2ac55d41 2821 struct extent_state *cached_state = NULL;
38c227d8 2822 struct new_sa_defrag_extent *new = NULL;
261507a0 2823 int compress_type = 0;
77cef2ec
JB		 2824 int ret = 0;
2825 u64 logical_len = ordered_extent->len;
82d5902d 2826 bool nolock;
77cef2ec 2827 bool truncated = false;
e6dcd2dc 2828
83eea1f1 2829 nolock = btrfs_is_free_space_inode(inode);
0cb59c99 2830
5fd02043
JB		 2831 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2832 ret = -EIO;
2833 goto out;
2834 }
2835
f612496b
MX		 2836 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2837 ordered_extent->file_offset +
2838 ordered_extent->len - 1);
2839
77cef2ec
JB		 2840 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2841 truncated = true;
2842 logical_len = ordered_extent->truncated_len;
2843 /* Truncated the entire extent, don't bother adding */
2844 if (!logical_len)
2845 goto out;
2846 }
2847
c2167754 2848 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
79787eaa 2849 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
94ed938a
QW		 2850
2851 /*
2852 * For mwrite(mmap + memset to write) case, we still reserve
2853 * space for NOCOW range.
2854 * As NOCOW won't cause a new delayed ref, just free the space
2855 */
2856 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2857 ordered_extent->len);
6c760c07
JB		 2858 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2859 if (nolock)
2860 trans = btrfs_join_transaction_nolock(root);
2861 else
2862 trans = btrfs_join_transaction(root);
2863 if (IS_ERR(trans)) {
2864 ret = PTR_ERR(trans);
2865 trans = NULL;
2866 goto out;
c2167754 2867 }
6c760c07
JB		 2868 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2869 ret = btrfs_update_inode_fallback(trans, root, inode);
2870 if (ret) /* -ENOMEM or corruption */
2871 btrfs_abort_transaction(trans, root, ret);
c2167754
YZ		 2872 goto out;
2873 }
e6dcd2dc 2874
2ac55d41
JB		 2875 lock_extent_bits(io_tree, ordered_extent->file_offset,
2876 ordered_extent->file_offset + ordered_extent->len - 1,
d0082371 2877 0, &cached_state);
e6dcd2dc 2878
38c227d8
LB		 2879 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2880 ordered_extent->file_offset + ordered_extent->len - 1,
2881 EXTENT_DEFRAG, 1, cached_state);
2882 if (ret) {
2883 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
8101c8db 2884 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
38c227d8
LB		 2885 /* the inode is shared */
2886 new = record_old_file_extents(inode, ordered_extent);
2887
2888 clear_extent_bit(io_tree, ordered_extent->file_offset,
2889 ordered_extent->file_offset + ordered_extent->len - 1,
2890 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2891 }
2892
0cb59c99 2893 if (nolock)
7a7eaa40 2894 trans = btrfs_join_transaction_nolock(root);
0cb59c99 2895 else
7a7eaa40 2896 trans = btrfs_join_transaction(root);
79787eaa
JM		 2897 if (IS_ERR(trans)) {
2898 ret = PTR_ERR(trans);
2899 trans = NULL;
2900 goto out_unlock;
2901 }
a79b7d4b 2902
0ca1f7ce 2903 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
c2167754 2904
c8b97818 2905 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
261507a0 2906 compress_type = ordered_extent->compress_type;
d899e052 2907 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
261507a0 2908 BUG_ON(compress_type);
920bbbfb 2909 ret = btrfs_mark_extent_written(trans, inode,
d899e052
YZ		 2910 ordered_extent->file_offset,
2911 ordered_extent->file_offset +
77cef2ec 2912 logical_len);
d899e052 2913 } else {
0af3d00b 2914 BUG_ON(root == root->fs_info->tree_root);
d899e052
YZ		 2915 ret = insert_reserved_file_extent(trans, inode,
2916 ordered_extent->file_offset,
2917 ordered_extent->start,
2918 ordered_extent->disk_len,
77cef2ec 2919 logical_len, logical_len,
261507a0 2920 compress_type, 0, 0,
d899e052 2921 BTRFS_FILE_EXTENT_REG);
e570fd27
MX		 2922 if (!ret)
2923 btrfs_release_delalloc_bytes(root,
2924 ordered_extent->start,
2925 ordered_extent->disk_len);
d899e052 2926 }
5dc562c5
JB		 2927 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2928 ordered_extent->file_offset, ordered_extent->len,
2929 trans->transid);
79787eaa
JM		 2930 if (ret < 0) {
2931 btrfs_abort_transaction(trans, root, ret);
5fd02043 2932 goto out_unlock;
79787eaa 2933 }
2ac55d41 2934
e6dcd2dc
CM		 2935 add_pending_csums(trans, inode, ordered_extent->file_offset,
2936 &ordered_extent->list);
2937
6c760c07
JB		 2938 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2939 ret = btrfs_update_inode_fallback(trans, root, inode);
2940 if (ret) { /* -ENOMEM or corruption */
2941 btrfs_abort_transaction(trans, root, ret);
2942 goto out_unlock;
1ef30be1
JB		 2943 }
2944 ret = 0;
5fd02043
JB		 2945out_unlock:
2946 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2947 ordered_extent->file_offset +
2948 ordered_extent->len - 1, &cached_state, GFP_NOFS);
c2167754 2949out:
5b0e95bf 2950 if (root != root->fs_info->tree_root)
0cb59c99 2951 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
a698d075
MX		 2952 if (trans)
2953 btrfs_end_transaction(trans, root);
0cb59c99 2954
77cef2ec
JB		 2955 if (ret || truncated) {
2956 u64 start, end;
2957
2958 if (truncated)
2959 start = ordered_extent->file_offset + logical_len;
2960 else
2961 start = ordered_extent->file_offset;
2962 end = ordered_extent->file_offset + ordered_extent->len - 1;
2963 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2964
2965 /* Drop the cache for the part of the extent we didn't write. */
2966 btrfs_drop_extent_cache(inode, start, end, 0);
5fd02043 2967
0bec9ef5
JB		 2968 /*
2969 * If the ordered extent had an IOERR or something else went
2970 * wrong we need to return the space for this ordered extent
77cef2ec
JB		 2971 * back to the allocator. We only free the extent in the
2972 * truncated case if we didn't write out the extent at all.
0bec9ef5 2973 */
77cef2ec
JB		 2974 if ((ret || !logical_len) &&
2975 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
0bec9ef5
JB		 2976 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2977 btrfs_free_reserved_extent(root, ordered_extent->start,
e570fd27 2978 ordered_extent->disk_len, 1);
0bec9ef5
JB		 2979 }
2980
2981
5fd02043 2982 /*
8bad3c02
LB		 2983 * This needs to be done to make sure anybody waiting knows we are done
2984 * updating everything for this ordered extent.
5fd02043
JB		 2985 */
2986 btrfs_remove_ordered_extent(inode, ordered_extent);
2987
38c227d8 2988 /* for snapshot-aware defrag */
6f519564
LB		 2989 if (new) {
2990 if (ret) {
2991 free_sa_defrag_extent(new);
2992 atomic_dec(&root->fs_info->defrag_running);
2993 } else {
2994 relink_file_extents(new);
2995 }
2996 }
38c227d8 2997
e6dcd2dc
CM		 2998 /* once for us */
2999 btrfs_put_ordered_extent(ordered_extent);
3000 /* once for the tree */
3001 btrfs_put_ordered_extent(ordered_extent);
3002
5fd02043
JB		 3003 return ret;
3004}
3005
3006static void finish_ordered_fn(struct btrfs_work *work)
3007{
3008 struct btrfs_ordered_extent *ordered_extent;
3009 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3010 btrfs_finish_ordered_io(ordered_extent);
e6dcd2dc
CM		 3011}
3012
b2950863 3013static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM		 3014 struct extent_state *state, int uptodate)
3015{
5fd02043
JB		 3016 struct inode *inode = page->mapping->host;
3017 struct btrfs_root *root = BTRFS_I(inode)->root;
3018 struct btrfs_ordered_extent *ordered_extent = NULL;
9e0af237
LB		 3019 struct btrfs_workqueue *wq;
3020 btrfs_work_func_t func;
5fd02043 3021
1abe9b8a 3022 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3023
8b62b72b 3024 ClearPagePrivate2(page);
5fd02043
JB		 3025 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3026 end - start + 1, uptodate))
3027 return 0;
3028
9e0af237
LB		 3029 if (btrfs_is_free_space_inode(inode)) {
3030 wq = root->fs_info->endio_freespace_worker;
3031 func = btrfs_freespace_write_helper;
3032 } else {
3033 wq = root->fs_info->endio_write_workers;
3034 func = btrfs_endio_write_helper;
3035 }
5fd02043 3036
9e0af237
LB		 3037 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3038 NULL);
3039 btrfs_queue_work(wq, &ordered_extent->work);
5fd02043
JB		 3040
3041 return 0;
211f90e6
CM		 3042}
3043
dc380aea
MX		 3044static int __readpage_endio_check(struct inode *inode,
3045 struct btrfs_io_bio *io_bio,
3046 int icsum, struct page *page,
3047 int pgoff, u64 start, size_t len)
3048{
3049 char *kaddr;
3050 u32 csum_expected;
3051 u32 csum = ~(u32)0;
dc380aea
MX		 3052
3053 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3054
3055 kaddr = kmap_atomic(page);
3056 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3057 btrfs_csum_final(csum, (char *)&csum);
3058 if (csum != csum_expected)
3059 goto zeroit;
3060
3061 kunmap_atomic(kaddr);
3062 return 0;
3063zeroit:
94647322
DS		 3064 btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3065 "csum failed ino %llu off %llu csum %u expected csum %u",
dc380aea
MX		 3066 btrfs_ino(inode), start, csum, csum_expected);
3067 memset(kaddr + pgoff, 1, len);
3068 flush_dcache_page(page);
3069 kunmap_atomic(kaddr);
3070 if (csum_expected == 0)
3071 return 0;
3072 return -EIO;
3073}
3074
d352ac68
CM		 3075/*
3076 * when reads are done, we need to check csums to verify the data is correct
4a54c8c1
JS		 3077 * if there's a match, we allow the bio to finish. If not, the code in
3078 * extent_io.c will try to find good copies for us.
d352ac68 3079 */
facc8a22
MX		 3080static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3081 u64 phy_offset, struct page *page,
3082 u64 start, u64 end, int mirror)
07157aac 3083{
4eee4fa4 3084 size_t offset = start - page_offset(page);
07157aac 3085 struct inode *inode = page->mapping->host;
d1310b2e 3086 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
ff79f819 3087 struct btrfs_root *root = BTRFS_I(inode)->root;
d1310b2e 3088
d20f7043
CM		 3089 if (PageChecked(page)) {
3090 ClearPageChecked(page);
dc380aea 3091 return 0;
d20f7043 3092 }
6cbff00f
CH		 3093
3094 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
dc380aea 3095 return 0;
17d217fe
YZ		 3096
3097 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
9655d298 3098 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
17d217fe
YZ		 3099 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
3100 GFP_NOFS);
b6cda9bc 3101 return 0;
17d217fe 3102 }
d20f7043 3103
facc8a22 3104 phy_offset >>= inode->i_sb->s_blocksize_bits;
dc380aea
MX		 3105 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3106 start, (size_t)(end - start + 1));
07157aac 3107}
b888db2b 3108
24bbcf04
YZ		 3109void btrfs_add_delayed_iput(struct inode *inode)
3110{
3111 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
8089fe62 3112 struct btrfs_inode *binode = BTRFS_I(inode);
24bbcf04
YZ		 3113
3114 if (atomic_add_unless(&inode->i_count, -1, 1))
3115 return;
3116
24bbcf04 3117 spin_lock(&fs_info->delayed_iput_lock);
8089fe62
DS		 3118 if (binode->delayed_iput_count == 0) {
3119 ASSERT(list_empty(&binode->delayed_iput));
3120 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3121 } else {
3122 binode->delayed_iput_count++;
3123 }
24bbcf04
YZ		 3124 spin_unlock(&fs_info->delayed_iput_lock);
3125}
3126
3127void btrfs_run_delayed_iputs(struct btrfs_root *root)
3128{
24bbcf04 3129 struct btrfs_fs_info *fs_info = root->fs_info;
24bbcf04 3130
d7c15171 3131 down_read(&fs_info->delayed_iput_sem);
24bbcf04 3132 spin_lock(&fs_info->delayed_iput_lock);
8089fe62
DS		 3133 while (!list_empty(&fs_info->delayed_iputs)) {
3134 struct btrfs_inode *inode;
3135
3136 inode = list_first_entry(&fs_info->delayed_iputs,
3137 struct btrfs_inode, delayed_iput);
3138 if (inode->delayed_iput_count) {
3139 inode->delayed_iput_count--;
3140 list_move_tail(&inode->delayed_iput,
3141 &fs_info->delayed_iputs);
3142 } else {
3143 list_del_init(&inode->delayed_iput);
3144 }
3145 spin_unlock(&fs_info->delayed_iput_lock);
3146 iput(&inode->vfs_inode);
3147 spin_lock(&fs_info->delayed_iput_lock);
24bbcf04 3148 }
8089fe62 3149 spin_unlock(&fs_info->delayed_iput_lock);
d7c15171 3150 up_read(&root->fs_info->delayed_iput_sem);
24bbcf04
YZ		 3151}
3152
d68fc57b 3153/*
42b2aa86 3154 * This is called in transaction commit time. If there are no orphan
d68fc57b
YZ		 3155 * files in the subvolume, it removes orphan item and frees block_rsv
3156 * structure.
3157 */
3158void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3159 struct btrfs_root *root)
3160{
90290e19 3161 struct btrfs_block_rsv *block_rsv;
d68fc57b
YZ		 3162 int ret;
3163
8a35d95f 3164 if (atomic_read(&root->orphan_inodes) ||
d68fc57b
YZ		 3165 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3166 return;
3167
90290e19 3168 spin_lock(&root->orphan_lock);
8a35d95f 3169 if (atomic_read(&root->orphan_inodes)) {
90290e19
JB		 3170 spin_unlock(&root->orphan_lock);
3171 return;
3172 }
3173
3174 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3175 spin_unlock(&root->orphan_lock);
3176 return;
3177 }
3178
3179 block_rsv = root->orphan_block_rsv;
3180 root->orphan_block_rsv = NULL;
3181 spin_unlock(&root->orphan_lock);
3182
27cdeb70 3183 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
d68fc57b
YZ		 3184 btrfs_root_refs(&root->root_item) > 0) {
3185 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3186 root->root_key.objectid);
4ef31a45
JB		 3187 if (ret)
3188 btrfs_abort_transaction(trans, root, ret);
3189 else
27cdeb70
MX		 3190 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3191 &root->state);
d68fc57b
YZ		 3192 }
3193
90290e19
JB		 3194 if (block_rsv) {
3195 WARN_ON(block_rsv->size > 0);
3196 btrfs_free_block_rsv(root, block_rsv);
d68fc57b
YZ		 3197 }
3198}
3199
7b128766
JB		 3200/*
3201 * This creates an orphan entry for the given inode in case something goes
3202 * wrong in the middle of an unlink/truncate.
d68fc57b
YZ		 3203 *
3204 * NOTE: caller of this function should reserve 5 units of metadata for
3205 * this function.
7b128766
JB		 3206 */
3207int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3208{
3209 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 3210 struct btrfs_block_rsv *block_rsv = NULL;
3211 int reserve = 0;
3212 int insert = 0;
3213 int ret;
7b128766 3214
d68fc57b 3215 if (!root->orphan_block_rsv) {
66d8f3dd 3216 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
b532402e
TI		 3217 if (!block_rsv)
3218 return -ENOMEM;
d68fc57b 3219 }
7b128766 3220
d68fc57b
YZ		 3221 spin_lock(&root->orphan_lock);
3222 if (!root->orphan_block_rsv) {
3223 root->orphan_block_rsv = block_rsv;
3224 } else if (block_rsv) {
3225 btrfs_free_block_rsv(root, block_rsv);
3226 block_rsv = NULL;
7b128766 3227 }
7b128766 3228
8a35d95f
JB		 3229 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3230 &BTRFS_I(inode)->runtime_flags)) {
d68fc57b
YZ		 3231#if 0
3232 /*
3233 * For proper ENOSPC handling, we should do orphan
3234 * cleanup when mounting. But this introduces backward
3235 * compatibility issue.
3236 */
3237 if (!xchg(&root->orphan_item_inserted, 1))
3238 insert = 2;
3239 else
3240 insert = 1;
3241#endif
3242 insert = 1;
321f0e70 3243 atomic_inc(&root->orphan_inodes);
7b128766
JB		 3244 }
3245
72ac3c0d
JB		 3246 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3247 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3248 reserve = 1;
d68fc57b 3249 spin_unlock(&root->orphan_lock);
7b128766 3250
d68fc57b
YZ		 3251 /* grab metadata reservation from transaction handle */
3252 if (reserve) {
3253 ret = btrfs_orphan_reserve_metadata(trans, inode);
79787eaa 3254 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
d68fc57b 3255 }
7b128766 3256
d68fc57b
YZ		 3257 /* insert an orphan item to track this unlinked/truncated file */
3258 if (insert >= 1) {
33345d01 3259 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
4ef31a45 3260 if (ret) {
703c88e0 3261 atomic_dec(&root->orphan_inodes);
4ef31a45
JB		 3262 if (reserve) {
3263 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3264 &BTRFS_I(inode)->runtime_flags);
3265 btrfs_orphan_release_metadata(inode);
3266 }
3267 if (ret != -EEXIST) {
e8e7cff6
JB		 3268 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3269 &BTRFS_I(inode)->runtime_flags);
4ef31a45
JB		 3270 btrfs_abort_transaction(trans, root, ret);
3271 return ret;
3272 }
79787eaa
JM		 3273 }
3274 ret = 0;
d68fc57b
YZ		 3275 }
3276
3277 /* insert an orphan item to track subvolume contains orphan files */
3278 if (insert >= 2) {
3279 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3280 root->root_key.objectid);
79787eaa
JM		 3281 if (ret && ret != -EEXIST) {
3282 btrfs_abort_transaction(trans, root, ret);
3283 return ret;
3284 }
d68fc57b
YZ		 3285 }
3286 return 0;
7b128766
JB		 3287}
3288
3289/*
3290 * We have done the truncate/delete so we can go ahead and remove the orphan
3291 * item for this particular inode.
3292 */
48a3b636
ES		 3293static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3294 struct inode *inode)
7b128766
JB		 3295{
3296 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 3297 int delete_item = 0;
3298 int release_rsv = 0;
7b128766
JB		 3299 int ret = 0;
3300
d68fc57b 3301 spin_lock(&root->orphan_lock);
8a35d95f
JB		 3302 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3303 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3304 delete_item = 1;
7b128766 3305
72ac3c0d
JB		 3306 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3307 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3308 release_rsv = 1;
d68fc57b 3309 spin_unlock(&root->orphan_lock);
7b128766 3310
703c88e0 3311 if (delete_item) {
8a35d95f 3312 atomic_dec(&root->orphan_inodes);
703c88e0
FDBM		 3313 if (trans)
3314 ret = btrfs_del_orphan_item(trans, root,
3315 btrfs_ino(inode));
8a35d95f 3316 }
7b128766 3317
703c88e0
FDBM		 3318 if (release_rsv)
3319 btrfs_orphan_release_metadata(inode);
3320
4ef31a45 3321 return ret;
7b128766
JB		 3322}
3323
3324/*
3325 * this cleans up any orphans that may be left on the list from the last use
3326 * of this root.
3327 */
66b4ffd1 3328int btrfs_orphan_cleanup(struct btrfs_root *root)
7b128766
JB		 3329{
3330 struct btrfs_path *path;
3331 struct extent_buffer *leaf;
7b128766
JB		 3332 struct btrfs_key key, found_key;
3333 struct btrfs_trans_handle *trans;
3334 struct inode *inode;
8f6d7f4f 3335 u64 last_objectid = 0;
7b128766
JB		 3336 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3337
d68fc57b 3338 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
66b4ffd1 3339 return 0;
c71bf099
YZ		 3340
3341 path = btrfs_alloc_path();
66b4ffd1
JB		 3342 if (!path) {
3343 ret = -ENOMEM;
3344 goto out;
3345 }
7b128766
JB		 3346 path->reada = -1;
3347
3348 key.objectid = BTRFS_ORPHAN_OBJECTID;
962a298f 3349 key.type = BTRFS_ORPHAN_ITEM_KEY;
7b128766
JB		 3350 key.offset = (u64)-1;
3351
7b128766
JB		 3352 while (1) {
3353 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
66b4ffd1
JB		 3354 if (ret < 0)
3355 goto out;
7b128766
JB		 3356
3357 /*
3358 * if ret == 0 means we found what we were searching for, which
25985edc 3359 * is weird, but possible, so only screw with path if we didn't
7b128766
JB		 3360 * find the key and see if we have stuff that matches
3361 */
3362 if (ret > 0) {
66b4ffd1 3363 ret = 0;
7b128766
JB		 3364 if (path->slots[0] == 0)
3365 break;
3366 path->slots[0]--;
3367 }
3368
3369 /* pull out the item */
3370 leaf = path->nodes[0];
7b128766
JB		 3371 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3372
3373 /* make sure the item matches what we want */
3374 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3375 break;
962a298f 3376 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
7b128766
JB		 3377 break;
3378
3379 /* release the path since we're done with it */
b3b4aa74 3380 btrfs_release_path(path);
7b128766
JB		 3381
3382 /*
3383 * this is where we are basically btrfs_lookup, without the
3384 * crossing root thing. we store the inode number in the
3385 * offset of the orphan item.
3386 */
8f6d7f4f
JB		 3387
3388 if (found_key.offset == last_objectid) {
c2cf52eb
SK		 3389 btrfs_err(root->fs_info,
3390 "Error removing orphan entry, stopping orphan cleanup");
8f6d7f4f
JB		 3391 ret = -EINVAL;
3392 goto out;
3393 }
3394
3395 last_objectid = found_key.offset;
3396
5d4f98a2
YZ		 3397 found_key.objectid = found_key.offset;
3398 found_key.type = BTRFS_INODE_ITEM_KEY;
3399 found_key.offset = 0;
73f73415 3400 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
8c6ffba0 3401 ret = PTR_ERR_OR_ZERO(inode);
a8c9e576 3402 if (ret && ret != -ESTALE)
66b4ffd1 3403 goto out;
7b128766 3404
f8e9e0b0
AJ		 3405 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3406 struct btrfs_root *dead_root;
3407 struct btrfs_fs_info *fs_info = root->fs_info;
3408 int is_dead_root = 0;
3409
3410 /*
3411 * this is an orphan in the tree root. Currently these
3412 * could come from 2 sources:
3413 * a) a snapshot deletion in progress
3414 * b) a free space cache inode
3415 * We need to distinguish those two, as the snapshot
3416 * orphan must not get deleted.
3417 * find_dead_roots already ran before us, so if this
3418 * is a snapshot deletion, we should find the root
3419 * in the dead_roots list
3420 */
3421 spin_lock(&fs_info->trans_lock);
3422 list_for_each_entry(dead_root, &fs_info->dead_roots,
3423 root_list) {
3424 if (dead_root->root_key.objectid ==
3425 found_key.objectid) {
3426 is_dead_root = 1;
3427 break;
3428 }
3429 }
3430 spin_unlock(&fs_info->trans_lock);
3431 if (is_dead_root) {
3432 /* prevent this orphan from being found again */
3433 key.offset = found_key.objectid - 1;
3434 continue;
3435 }
3436 }
7b128766 3437 /*
a8c9e576
JB		 3438 * Inode is already gone but the orphan item is still there,
3439 * kill the orphan item.
7b128766 3440 */
a8c9e576
JB		 3441 if (ret == -ESTALE) {
3442 trans = btrfs_start_transaction(root, 1);
66b4ffd1
JB		 3443 if (IS_ERR(trans)) {
3444 ret = PTR_ERR(trans);
3445 goto out;
3446 }
c2cf52eb
SK		 3447 btrfs_debug(root->fs_info, "auto deleting %Lu",
3448 found_key.objectid);
a8c9e576
JB		 3449 ret = btrfs_del_orphan_item(trans, root,
3450 found_key.objectid);
5b21f2ed 3451 btrfs_end_transaction(trans, root);
4ef31a45
JB		 3452 if (ret)
3453 goto out;
7b128766
JB		 3454 continue;
3455 }
3456
a8c9e576
JB		 3457 /*
3458 * add this inode to the orphan list so btrfs_orphan_del does
3459 * the proper thing when we hit it
3460 */
8a35d95f
JB		 3461 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3462 &BTRFS_I(inode)->runtime_flags);
925396ec 3463 atomic_inc(&root->orphan_inodes);
a8c9e576 3464
7b128766
JB		 3465 /* if we have links, this was a truncate, lets do that */
3466 if (inode->i_nlink) {
fae7f21c 3467 if (WARN_ON(!S_ISREG(inode->i_mode))) {
a41ad394
JB		 3468 iput(inode);
3469 continue;
3470 }
7b128766 3471 nr_truncate++;
f3fe820c
JB		 3472
3473 /* 1 for the orphan item deletion. */
3474 trans = btrfs_start_transaction(root, 1);
3475 if (IS_ERR(trans)) {
c69b26b0 3476 iput(inode);
f3fe820c
JB		 3477 ret = PTR_ERR(trans);
3478 goto out;
3479 }
3480 ret = btrfs_orphan_add(trans, inode);
3481 btrfs_end_transaction(trans, root);
c69b26b0
JB		 3482 if (ret) {
3483 iput(inode);
f3fe820c 3484 goto out;
c69b26b0 3485 }
f3fe820c 3486
66b4ffd1 3487 ret = btrfs_truncate(inode);
4a7d0f68
JB		 3488 if (ret)
3489 btrfs_orphan_del(NULL, inode);
7b128766
JB		 3490 } else {
3491 nr_unlink++;
3492 }
3493
3494 /* this will do delete_inode and everything for us */
3495 iput(inode);
66b4ffd1
JB		 3496 if (ret)
3497 goto out;
7b128766 3498 }
3254c876
MX		 3499 /* release the path since we're done with it */
3500 btrfs_release_path(path);
3501
d68fc57b
YZ		 3502 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3503
3504 if (root->orphan_block_rsv)
3505 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3506 (u64)-1);
3507
27cdeb70
MX		 3508 if (root->orphan_block_rsv ||
3509 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
7a7eaa40 3510 trans = btrfs_join_transaction(root);
66b4ffd1
JB		 3511 if (!IS_ERR(trans))
3512 btrfs_end_transaction(trans, root);
d68fc57b 3513 }
7b128766
JB		 3514
3515 if (nr_unlink)
4884b476 3516 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
7b128766 3517 if (nr_truncate)
4884b476 3518 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
66b4ffd1
JB		 3519
3520out:
3521 if (ret)
68b663d1 3522 btrfs_err(root->fs_info,
c2cf52eb 3523 "could not do orphan cleanup %d", ret);
66b4ffd1
JB		 3524 btrfs_free_path(path);
3525 return ret;
7b128766
JB		 3526}
3527
46a53cca
CM		 3528/*
3529 * very simple check to peek ahead in the leaf looking for xattrs. If we
3530 * don't find any xattrs, we know there can't be any acls.
3531 *
3532 * slot is the slot the inode is in, objectid is the objectid of the inode
3533 */
3534static noinline int acls_after_inode_item(struct extent_buffer *leaf,
63541927
FDBM		 3535 int slot, u64 objectid,
3536 int *first_xattr_slot)
46a53cca
CM		 3537{
3538 u32 nritems = btrfs_header_nritems(leaf);
3539 struct btrfs_key found_key;
f23b5a59
JB		 3540 static u64 xattr_access = 0;
3541 static u64 xattr_default = 0;
46a53cca
CM		 3542 int scanned = 0;
3543
f23b5a59
JB		 3544 if (!xattr_access) {
3545 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3546 strlen(POSIX_ACL_XATTR_ACCESS));
3547 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3548 strlen(POSIX_ACL_XATTR_DEFAULT));
3549 }
3550
46a53cca 3551 slot++;
63541927 3552 *first_xattr_slot = -1;
46a53cca
CM		 3553 while (slot < nritems) {
3554 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3555
3556 /* we found a different objectid, there must not be acls */
3557 if (found_key.objectid != objectid)
3558 return 0;
3559
3560 /* we found an xattr, assume we've got an acl */
f23b5a59 3561 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
63541927
FDBM		 3562 if (*first_xattr_slot == -1)
3563 *first_xattr_slot = slot;
f23b5a59
JB		 3564 if (found_key.offset == xattr_access ||
3565 found_key.offset == xattr_default)
3566 return 1;
3567 }
46a53cca
CM		 3568
3569 /*
3570 * we found a key greater than an xattr key, there can't
3571 * be any acls later on
3572 */
3573 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3574 return 0;
3575
3576 slot++;
3577 scanned++;
3578
3579 /*
3580 * it goes inode, inode backrefs, xattrs, extents,
3581 * so if there are a ton of hard links to an inode there can
3582 * be a lot of backrefs. Don't waste time searching too hard,
3583 * this is just an optimization
3584 */
3585 if (scanned >= 8)
3586 break;
3587 }
3588 /* we hit the end of the leaf before we found an xattr or
3589 * something larger than an xattr. We have to assume the inode
3590 * has acls
3591 */
63541927
FDBM		 3592 if (*first_xattr_slot == -1)
3593 *first_xattr_slot = slot;
46a53cca
CM		 3594 return 1;
3595}
3596
d352ac68
CM		 3597/*
3598 * read an inode from the btree into the in-memory inode
3599 */
5d4f98a2 3600static void btrfs_read_locked_inode(struct inode *inode)
39279cc3
CM		 3601{
3602 struct btrfs_path *path;
5f39d397 3603 struct extent_buffer *leaf;
39279cc3
CM		 3604 struct btrfs_inode_item *inode_item;
3605 struct btrfs_root *root = BTRFS_I(inode)->root;
3606 struct btrfs_key location;
67de1176 3607 unsigned long ptr;
46a53cca 3608 int maybe_acls;
618e21d5 3609 u32 rdev;
39279cc3 3610 int ret;
2f7e33d4 3611 bool filled = false;
63541927 3612 int first_xattr_slot;
2f7e33d4
MX		 3613
3614 ret = btrfs_fill_inode(inode, &rdev);
3615 if (!ret)
3616 filled = true;
39279cc3
CM		 3617
3618 path = btrfs_alloc_path();
1748f843
MF		 3619 if (!path)
3620 goto make_bad;
3621
39279cc3 3622 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 3623
39279cc3 3624 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 3625 if (ret)
39279cc3 3626 goto make_bad;
39279cc3 3627
5f39d397 3628 leaf = path->nodes[0];
2f7e33d4
MX		 3629
3630 if (filled)
67de1176 3631 goto cache_index;
2f7e33d4 3632
5f39d397
CM		 3633 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3634 struct btrfs_inode_item);
5f39d397 3635 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
bfe86848 3636 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2f2f43d3
EB		 3637 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3638 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
dbe674a9 3639 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397 3640
a937b979
DS		 3641 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3642 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
5f39d397 3643
a937b979
DS		 3644 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3645 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
5f39d397 3646
a937b979
DS		 3647 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3648 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
5f39d397 3649
9cc97d64 3650 BTRFS_I(inode)->i_otime.tv_sec =
3651 btrfs_timespec_sec(leaf, &inode_item->otime);
3652 BTRFS_I(inode)->i_otime.tv_nsec =
3653 btrfs_timespec_nsec(leaf, &inode_item->otime);
5f39d397 3654
a76a3cd4 3655 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 3656 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
5dc562c5
JB		 3657 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3658
6e17d30b
YD		 3659 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3660 inode->i_generation = BTRFS_I(inode)->generation;
3661 inode->i_rdev = 0;
3662 rdev = btrfs_inode_rdev(leaf, inode_item);
3663
3664 BTRFS_I(inode)->index_cnt = (u64)-1;
3665 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3666
3667cache_index:
5dc562c5
JB		 3668 /*
3669 * If we were modified in the current generation and evicted from memory
3670 * and then re-read we need to do a full sync since we don't have any
3671 * idea about which extents were modified before we were evicted from
3672 * cache.
6e17d30b
YD		 3673 *
3674 * This is required for both inode re-read from disk and delayed inode
3675 * in delayed_nodes_tree.
5dc562c5
JB		 3676 */
3677 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3678 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3679 &BTRFS_I(inode)->runtime_flags);
3680
bde6c242
FM		 3681 /*
3682 * We don't persist the id of the transaction where an unlink operation
3683 * against the inode was last made. So here we assume the inode might
3684 * have been evicted, and therefore the exact value of last_unlink_trans
3685 * lost, and set it to last_trans to avoid metadata inconsistencies
3686 * between the inode and its parent if the inode is fsync'ed and the log
3687 * replayed. For example, in the scenario:
3688 *
3689 * touch mydir/foo
3690 * ln mydir/foo mydir/bar
3691 * sync
3692 * unlink mydir/bar
3693 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3694 * xfs_io -c fsync mydir/foo
3695 * <power failure>
3696 * mount fs, triggers fsync log replay
3697 *
3698 * We must make sure that when we fsync our inode foo we also log its
3699 * parent inode, otherwise after log replay the parent still has the
3700 * dentry with the "bar" name but our inode foo has a link count of 1
3701 * and doesn't have an inode ref with the name "bar" anymore.
3702 *
3703 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3704 * but it guarantees correctness at the expense of ocassional full
3705 * transaction commits on fsync if our inode is a directory, or if our
3706 * inode is not a directory, logging its parent unnecessarily.
3707 */
3708 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3709
67de1176
MX		 3710 path->slots[0]++;
3711 if (inode->i_nlink != 1 ||
3712 path->slots[0] >= btrfs_header_nritems(leaf))
3713 goto cache_acl;
3714
3715 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3716 if (location.objectid != btrfs_ino(inode))
3717 goto cache_acl;
3718
3719 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3720 if (location.type == BTRFS_INODE_REF_KEY) {
3721 struct btrfs_inode_ref *ref;
3722
3723 ref = (struct btrfs_inode_ref *)ptr;
3724 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3725 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3726 struct btrfs_inode_extref *extref;
3727
3728 extref = (struct btrfs_inode_extref *)ptr;
3729 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3730 extref);
3731 }
2f7e33d4 3732cache_acl:
46a53cca
CM		 3733 /*
3734 * try to precache a NULL acl entry for files that don't have
3735 * any xattrs or acls
3736 */
33345d01 3737 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
63541927
FDBM		 3738 btrfs_ino(inode), &first_xattr_slot);
3739 if (first_xattr_slot != -1) {
3740 path->slots[0] = first_xattr_slot;
3741 ret = btrfs_load_inode_props(inode, path);
3742 if (ret)
3743 btrfs_err(root->fs_info,
351fd353 3744 "error loading props for ino %llu (root %llu): %d",
63541927
FDBM		 3745 btrfs_ino(inode),
3746 root->root_key.objectid, ret);
3747 }
3748 btrfs_free_path(path);
3749
72c04902
AV		 3750 if (!maybe_acls)
3751 cache_no_acl(inode);
46a53cca 3752
39279cc3 3753 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 3754 case S_IFREG:
3755 inode->i_mapping->a_ops = &btrfs_aops;
d1310b2e 3756 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 3757 inode->i_fop = &btrfs_file_operations;
3758 inode->i_op = &btrfs_file_inode_operations;
3759 break;
3760 case S_IFDIR:
3761 inode->i_fop = &btrfs_dir_file_operations;
3762 if (root == root->fs_info->tree_root)
3763 inode->i_op = &btrfs_dir_ro_inode_operations;
3764 else
3765 inode->i_op = &btrfs_dir_inode_operations;
3766 break;
3767 case S_IFLNK:
3768 inode->i_op = &btrfs_symlink_inode_operations;
3769 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3770 break;
618e21d5 3771 default:
0279b4cd 3772 inode->i_op = &btrfs_special_inode_operations;
618e21d5
JB		 3773 init_special_inode(inode, inode->i_mode, rdev);
3774 break;
39279cc3 3775 }
6cbff00f
CH		 3776
3777 btrfs_update_iflags(inode);
39279cc3
CM		 3778 return;
3779
3780make_bad:
39279cc3 3781 btrfs_free_path(path);
39279cc3
CM		 3782 make_bad_inode(inode);
3783}
3784
d352ac68
CM		 3785/*
3786 * given a leaf and an inode, copy the inode fields into the leaf
3787 */
e02119d5
CM		 3788static void fill_inode_item(struct btrfs_trans_handle *trans,
3789 struct extent_buffer *leaf,
5f39d397 3790 struct btrfs_inode_item *item,
39279cc3
CM		 3791 struct inode *inode)
3792{
51fab693
LB		 3793 struct btrfs_map_token token;
3794
3795 btrfs_init_map_token(&token);
5f39d397 3796
51fab693
LB		 3797 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3798 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3799 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3800 &token);
3801 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3802 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
5f39d397 3803
a937b979 3804 btrfs_set_token_timespec_sec(leaf, &item->atime,
51fab693 3805 inode->i_atime.tv_sec, &token);
a937b979 3806 btrfs_set_token_timespec_nsec(leaf, &item->atime,
51fab693 3807 inode->i_atime.tv_nsec, &token);
5f39d397 3808
a937b979 3809 btrfs_set_token_timespec_sec(leaf, &item->mtime,
51fab693 3810 inode->i_mtime.tv_sec, &token);
a937b979 3811 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
51fab693 3812 inode->i_mtime.tv_nsec, &token);
5f39d397 3813
a937b979 3814 btrfs_set_token_timespec_sec(leaf, &item->ctime,
51fab693 3815 inode->i_ctime.tv_sec, &token);
a937b979 3816 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
51fab693 3817 inode->i_ctime.tv_nsec, &token);
5f39d397 3818
9cc97d64 3819 btrfs_set_token_timespec_sec(leaf, &item->otime,
3820 BTRFS_I(inode)->i_otime.tv_sec, &token);
3821 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3822 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3823
51fab693
LB		 3824 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3825 &token);
3826 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3827 &token);
3828 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3829 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3830 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3831 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3832 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
39279cc3
CM		 3833}
3834
d352ac68
CM		 3835/*
3836 * copy everything in the in-memory inode into the btree.
3837 */
2115133f 3838static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
d397712b 3839 struct btrfs_root *root, struct inode *inode)
39279cc3
CM		 3840{
3841 struct btrfs_inode_item *inode_item;
3842 struct btrfs_path *path;
5f39d397 3843 struct extent_buffer *leaf;
39279cc3
CM		 3844 int ret;
3845
3846 path = btrfs_alloc_path();
16cdcec7
MX		 3847 if (!path)
3848 return -ENOMEM;
3849
b9473439 3850 path->leave_spinning = 1;
16cdcec7
MX		 3851 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3852 1);
39279cc3
CM		 3853 if (ret) {
3854 if (ret > 0)
3855 ret = -ENOENT;
3856 goto failed;
3857 }
3858
5f39d397
CM		 3859 leaf = path->nodes[0];
3860 inode_item = btrfs_item_ptr(leaf, path->slots[0],
16cdcec7 3861 struct btrfs_inode_item);
39279cc3 3862
e02119d5 3863 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 3864 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 3865 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 3866 ret = 0;
3867failed:
39279cc3
CM		 3868 btrfs_free_path(path);
3869 return ret;
3870}
3871
2115133f
CM		 3872/*
3873 * copy everything in the in-memory inode into the btree.
3874 */
3875noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3876 struct btrfs_root *root, struct inode *inode)
3877{
3878 int ret;
3879
3880 /*
3881 * If the inode is a free space inode, we can deadlock during commit
3882 * if we put it into the delayed code.
3883 *
3884 * The data relocation inode should also be directly updated
3885 * without delay
3886 */
83eea1f1 3887 if (!btrfs_is_free_space_inode(inode)
1d52c78a
JB		 3888 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3889 && !root->fs_info->log_root_recovering) {
8ea05e3a
AB		 3890 btrfs_update_root_times(trans, root);
3891
2115133f
CM		 3892 ret = btrfs_delayed_update_inode(trans, root, inode);
3893 if (!ret)
3894 btrfs_set_inode_last_trans(trans, inode);
3895 return ret;
3896 }
3897
3898 return btrfs_update_inode_item(trans, root, inode);
3899}
3900
be6aef60
JB		 3901noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3902 struct btrfs_root *root,
3903 struct inode *inode)
2115133f
CM		 3904{
3905 int ret;
3906
3907 ret = btrfs_update_inode(trans, root, inode);
3908 if (ret == -ENOSPC)
3909 return btrfs_update_inode_item(trans, root, inode);
3910 return ret;
3911}
3912
d352ac68
CM		 3913/*
3914 * unlink helper that gets used here in inode.c and in the tree logging
3915 * recovery code. It remove a link in a directory with a given name, and
3916 * also drops the back refs in the inode to the directory
3917 */
92986796
AV		 3918static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3919 struct btrfs_root *root,
3920 struct inode *dir, struct inode *inode,
3921 const char *name, int name_len)
39279cc3
CM		 3922{
3923 struct btrfs_path *path;
39279cc3 3924 int ret = 0;
5f39d397 3925 struct extent_buffer *leaf;
39279cc3 3926 struct btrfs_dir_item *di;
5f39d397 3927 struct btrfs_key key;
aec7477b 3928 u64 index;
33345d01
LZ		 3929 u64 ino = btrfs_ino(inode);
3930 u64 dir_ino = btrfs_ino(dir);
39279cc3
CM		 3931
3932 path = btrfs_alloc_path();
54aa1f4d
CM		 3933 if (!path) {
3934 ret = -ENOMEM;
554233a6 3935 goto out;
54aa1f4d
CM		 3936 }
3937
b9473439 3938 path->leave_spinning = 1;
33345d01 3939 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
39279cc3
CM		 3940 name, name_len, -1);
3941 if (IS_ERR(di)) {
3942 ret = PTR_ERR(di);
3943 goto err;
3944 }
3945 if (!di) {
3946 ret = -ENOENT;
3947 goto err;
3948 }
5f39d397
CM		 3949 leaf = path->nodes[0];
3950 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 3951 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 3952 if (ret)
3953 goto err;
b3b4aa74 3954 btrfs_release_path(path);
39279cc3 3955
67de1176
MX		 3956 /*
3957 * If we don't have dir index, we have to get it by looking up
3958 * the inode ref, since we get the inode ref, remove it directly,
3959 * it is unnecessary to do delayed deletion.
3960 *
3961 * But if we have dir index, needn't search inode ref to get it.
3962 * Since the inode ref is close to the inode item, it is better
3963 * that we delay to delete it, and just do this deletion when
3964 * we update the inode item.
3965 */
3966 if (BTRFS_I(inode)->dir_index) {
3967 ret = btrfs_delayed_delete_inode_ref(inode);
3968 if (!ret) {
3969 index = BTRFS_I(inode)->dir_index;
3970 goto skip_backref;
3971 }
3972 }
3973
33345d01
LZ		 3974 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3975 dir_ino, &index);
aec7477b 3976 if (ret) {
c2cf52eb
SK		 3977 btrfs_info(root->fs_info,
3978 "failed to delete reference to %.*s, inode %llu parent %llu",
c1c9ff7c 3979 name_len, name, ino, dir_ino);
79787eaa 3980 btrfs_abort_transaction(trans, root, ret);
aec7477b
JB		 3981 goto err;
3982 }
67de1176 3983skip_backref:
16cdcec7 3984 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 3985 if (ret) {
3986 btrfs_abort_transaction(trans, root, ret);
39279cc3 3987 goto err;
79787eaa 3988 }
39279cc3 3989
e02119d5 3990 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
33345d01 3991 inode, dir_ino);
79787eaa
JM		 3992 if (ret != 0 && ret != -ENOENT) {
3993 btrfs_abort_transaction(trans, root, ret);
3994 goto err;
3995 }
e02119d5
CM		 3996
3997 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3998 dir, index);
6418c961
CM		 3999 if (ret == -ENOENT)
4000 ret = 0;
d4e3991b
ZB		 4001 else if (ret)
4002 btrfs_abort_transaction(trans, root, ret);
39279cc3
CM		 4003err:
4004 btrfs_free_path(path);
e02119d5
CM		 4005 if (ret)
4006 goto out;
4007
4008 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95
JB		 4009 inode_inc_iversion(inode);
4010 inode_inc_iversion(dir);
e02119d5 4011 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
b9959295 4012 ret = btrfs_update_inode(trans, root, dir);
e02119d5 4013out:
39279cc3
CM		 4014 return ret;
4015}
4016
92986796
AV		 4017int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4018 struct btrfs_root *root,
4019 struct inode *dir, struct inode *inode,
4020 const char *name, int name_len)
4021{
4022 int ret;
4023 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4024 if (!ret) {
8b558c5f 4025 drop_nlink(inode);
92986796
AV		 4026 ret = btrfs_update_inode(trans, root, inode);
4027 }
4028 return ret;
4029}
39279cc3 4030
a22285a6
YZ		 4031/*
4032 * helper to start transaction for unlink and rmdir.
4033 *
d52be818
JB		 4034 * unlink and rmdir are special in btrfs, they do not always free space, so
4035 * if we cannot make our reservations the normal way try and see if there is
4036 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4037 * allow the unlink to occur.
a22285a6 4038 */
d52be818 4039static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4df27c4d 4040{
a22285a6 4041 struct btrfs_root *root = BTRFS_I(dir)->root;
4df27c4d 4042
e70bea5f
JB		 4043 /*
4044 * 1 for the possible orphan item
4045 * 1 for the dir item
4046 * 1 for the dir index
4047 * 1 for the inode ref
e70bea5f
JB		 4048 * 1 for the inode
4049 */
8eab77ff 4050 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
a22285a6
YZ		 4051}
4052
4053static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4054{
4055 struct btrfs_root *root = BTRFS_I(dir)->root;
4056 struct btrfs_trans_handle *trans;
2b0143b5 4057 struct inode *inode = d_inode(dentry);
a22285a6 4058 int ret;
a22285a6 4059
d52be818 4060 trans = __unlink_start_trans(dir);
a22285a6
YZ		 4061 if (IS_ERR(trans))
4062 return PTR_ERR(trans);
5f39d397 4063
2b0143b5 4064 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
12fcfd22 4065
2b0143b5 4066 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
e02119d5 4067 dentry->d_name.name, dentry->d_name.len);
b532402e
TI		 4068 if (ret)
4069 goto out;
7b128766 4070
a22285a6 4071 if (inode->i_nlink == 0) {
7b128766 4072 ret = btrfs_orphan_add(trans, inode);
b532402e
TI		 4073 if (ret)
4074 goto out;
a22285a6 4075 }
7b128766 4076
b532402e 4077out:
d52be818 4078 btrfs_end_transaction(trans, root);
b53d3f5d 4079 btrfs_btree_balance_dirty(root);
39279cc3
CM		 4080 return ret;
4081}
4082
4df27c4d
YZ		 4083int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4084 struct btrfs_root *root,
4085 struct inode *dir, u64 objectid,
4086 const char *name, int name_len)
4087{
4088 struct btrfs_path *path;
4089 struct extent_buffer *leaf;
4090 struct btrfs_dir_item *di;
4091 struct btrfs_key key;
4092 u64 index;
4093 int ret;
33345d01 4094 u64 dir_ino = btrfs_ino(dir);
4df27c4d
YZ		 4095
4096 path = btrfs_alloc_path();
4097 if (!path)
4098 return -ENOMEM;
4099
33345d01 4100 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4df27c4d 4101 name, name_len, -1);
79787eaa
JM		 4102 if (IS_ERR_OR_NULL(di)) {
4103 if (!di)
4104 ret = -ENOENT;
4105 else
4106 ret = PTR_ERR(di);
4107 goto out;
4108 }
4df27c4d
YZ		 4109
4110 leaf = path->nodes[0];
4111 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4112 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4113 ret = btrfs_delete_one_dir_name(trans, root, path, di);
79787eaa
JM		 4114 if (ret) {
4115 btrfs_abort_transaction(trans, root, ret);
4116 goto out;
4117 }
b3b4aa74 4118 btrfs_release_path(path);
4df27c4d
YZ		 4119
4120 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4121 objectid, root->root_key.objectid,
33345d01 4122 dir_ino, &index, name, name_len);
4df27c4d 4123 if (ret < 0) {
79787eaa
JM		 4124 if (ret != -ENOENT) {
4125 btrfs_abort_transaction(trans, root, ret);
4126 goto out;
4127 }
33345d01 4128 di = btrfs_search_dir_index_item(root, path, dir_ino,
4df27c4d 4129 name, name_len);
79787eaa
JM		 4130 if (IS_ERR_OR_NULL(di)) {
4131 if (!di)
4132 ret = -ENOENT;
4133 else
4134 ret = PTR_ERR(di);
4135 btrfs_abort_transaction(trans, root, ret);
4136 goto out;
4137 }
4df27c4d
YZ		 4138
4139 leaf = path->nodes[0];
4140 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
b3b4aa74 4141 btrfs_release_path(path);
4df27c4d
YZ		 4142 index = key.offset;
4143 }
945d8962 4144 btrfs_release_path(path);
4df27c4d 4145
16cdcec7 4146 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 4147 if (ret) {
4148 btrfs_abort_transaction(trans, root, ret);
4149 goto out;
4150 }
4df27c4d
YZ		 4151
4152 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95 4153 inode_inc_iversion(dir);
4df27c4d 4154 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
5a24e84c 4155 ret = btrfs_update_inode_fallback(trans, root, dir);
79787eaa
JM		 4156 if (ret)
4157 btrfs_abort_transaction(trans, root, ret);
4158out:
71d7aed0 4159 btrfs_free_path(path);
79787eaa 4160 return ret;
4df27c4d
YZ		 4161}
4162
39279cc3
CM		 4163static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4164{
2b0143b5 4165 struct inode *inode = d_inode(dentry);
1832a6d5 4166 int err = 0;
39279cc3 4167 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 4168 struct btrfs_trans_handle *trans;
39279cc3 4169
b3ae244e 4170 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
134d4512 4171 return -ENOTEMPTY;
b3ae244e
DS		 4172 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4173 return -EPERM;
134d4512 4174
d52be818 4175 trans = __unlink_start_trans(dir);
a22285a6 4176 if (IS_ERR(trans))
5df6a9f6 4177 return PTR_ERR(trans);
5df6a9f6 4178
33345d01 4179 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4df27c4d
YZ		 4180 err = btrfs_unlink_subvol(trans, root, dir,
4181 BTRFS_I(inode)->location.objectid,
4182 dentry->d_name.name,
4183 dentry->d_name.len);
4184 goto out;
4185 }
4186
7b128766
JB		 4187 err = btrfs_orphan_add(trans, inode);
4188 if (err)
4df27c4d 4189 goto out;
7b128766 4190
39279cc3 4191 /* now the directory is empty */
2b0143b5 4192 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
e02119d5 4193 dentry->d_name.name, dentry->d_name.len);
d397712b 4194 if (!err)
dbe674a9 4195 btrfs_i_size_write(inode, 0);
4df27c4d 4196out:
d52be818 4197 btrfs_end_transaction(trans, root);
b53d3f5d 4198 btrfs_btree_balance_dirty(root);
3954401f 4199
39279cc3
CM		 4200 return err;
4201}
4202
28f75a0e
CM		 4203static int truncate_space_check(struct btrfs_trans_handle *trans,
4204 struct btrfs_root *root,
4205 u64 bytes_deleted)
4206{
4207 int ret;
4208
4209 bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4210 ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4211 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4212 if (!ret)
4213 trans->bytes_reserved += bytes_deleted;
4214 return ret;
4215
4216}
4217
0305cd5f
FM		 4218static int truncate_inline_extent(struct inode *inode,
4219 struct btrfs_path *path,
4220 struct btrfs_key *found_key,
4221 const u64 item_end,
4222 const u64 new_size)
4223{
4224 struct extent_buffer *leaf = path->nodes[0];
4225 int slot = path->slots[0];
4226 struct btrfs_file_extent_item *fi;
4227 u32 size = (u32)(new_size - found_key->offset);
4228 struct btrfs_root *root = BTRFS_I(inode)->root;
4229
4230 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4231
4232 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4233 loff_t offset = new_size;
4234 loff_t page_end = ALIGN(offset, PAGE_CACHE_SIZE);
4235
4236 /*
4237 * Zero out the remaining of the last page of our inline extent,
4238 * instead of directly truncating our inline extent here - that
4239 * would be much more complex (decompressing all the data, then
4240 * compressing the truncated data, which might be bigger than
4241 * the size of the inline extent, resize the extent, etc).
4242 * We release the path because to get the page we might need to
4243 * read the extent item from disk (data not in the page cache).
4244 */
4245 btrfs_release_path(path);
4246 return btrfs_truncate_page(inode, offset, page_end - offset, 0);
4247 }
4248
4249 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4250 size = btrfs_file_extent_calc_inline_size(size);
4251 btrfs_truncate_item(root, path, size, 1);
4252
4253 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4254 inode_sub_bytes(inode, item_end + 1 - new_size);
4255
4256 return 0;
4257}
4258
39279cc3
CM		 4259/*
4260 * this can truncate away extent items, csum items and directory items.
4261 * It starts at a high offset and removes keys until it can't find
d352ac68 4262 * any higher than new_size
39279cc3
CM		 4263 *
4264 * csum items that cross the new i_size are truncated to the new size
4265 * as well.
7b128766
JB		 4266 *
4267 * min_type is the minimum key type to truncate down to. If set to 0, this
4268 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 4269 */
8082510e
YZ		 4270int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4271 struct btrfs_root *root,
4272 struct inode *inode,
4273 u64 new_size, u32 min_type)
39279cc3 4274{
39279cc3 4275 struct btrfs_path *path;
5f39d397 4276 struct extent_buffer *leaf;
39279cc3 4277 struct btrfs_file_extent_item *fi;
8082510e
YZ		 4278 struct btrfs_key key;
4279 struct btrfs_key found_key;
39279cc3 4280 u64 extent_start = 0;
db94535d 4281 u64 extent_num_bytes = 0;
5d4f98a2 4282 u64 extent_offset = 0;
39279cc3 4283 u64 item_end = 0;
c1aa4575 4284 u64 last_size = new_size;
8082510e 4285 u32 found_type = (u8)-1;
39279cc3
CM		 4286 int found_extent;
4287 int del_item;
85e21bac
CM		 4288 int pending_del_nr = 0;
4289 int pending_del_slot = 0;
179e29e4 4290 int extent_type = -1;
8082510e
YZ		 4291 int ret;
4292 int err = 0;
33345d01 4293 u64 ino = btrfs_ino(inode);
28ed1345 4294 u64 bytes_deleted = 0;
1262133b
JB		 4295 bool be_nice = 0;
4296 bool should_throttle = 0;
28f75a0e 4297 bool should_end = 0;
8082510e
YZ		 4298
4299 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
39279cc3 4300
28ed1345
CM		 4301 /*
4302 * for non-free space inodes and ref cows, we want to back off from
4303 * time to time
4304 */
4305 if (!btrfs_is_free_space_inode(inode) &&
4306 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4307 be_nice = 1;
4308
0eb0e19c
MF		 4309 path = btrfs_alloc_path();
4310 if (!path)
4311 return -ENOMEM;
4312 path->reada = -1;
4313
5dc562c5
JB		 4314 /*
4315 * We want to drop from the next block forward in case this new size is
4316 * not block aligned since we will be keeping the last block of the
4317 * extent just the way it is.
4318 */
27cdeb70
MX		 4319 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4320 root == root->fs_info->tree_root)
fda2832f
QW		 4321 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4322 root->sectorsize), (u64)-1, 0);
8082510e 4323
16cdcec7
MX		 4324 /*
4325 * This function is also used to drop the items in the log tree before
4326 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4327 * it is used to drop the loged items. So we shouldn't kill the delayed
4328 * items.
4329 */
4330 if (min_type == 0 && root == BTRFS_I(inode)->root)
4331 btrfs_kill_delayed_inode_items(inode);
4332
33345d01 4333 key.objectid = ino;
39279cc3 4334 key.offset = (u64)-1;
5f39d397
CM		 4335 key.type = (u8)-1;
4336
85e21bac 4337search_again:
28ed1345
CM		 4338 /*
4339 * with a 16K leaf size and 128MB extents, you can actually queue
4340 * up a huge file in a single leaf. Most of the time that
4341 * bytes_deleted is > 0, it will be huge by the time we get here
4342 */
4343 if (be_nice && bytes_deleted > 32 * 1024 * 1024) {
4344 if (btrfs_should_end_transaction(trans, root)) {
4345 err = -EAGAIN;
4346 goto error;
4347 }
4348 }
4349
4350
b9473439 4351 path->leave_spinning = 1;
85e21bac 4352 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8082510e
YZ		 4353 if (ret < 0) {
4354 err = ret;
4355 goto out;
4356 }
d397712b 4357
85e21bac 4358 if (ret > 0) {
e02119d5
CM		 4359 /* there are no items in the tree for us to truncate, we're
4360 * done
4361 */
8082510e
YZ		 4362 if (path->slots[0] == 0)
4363 goto out;
85e21bac
CM		 4364 path->slots[0]--;
4365 }
4366
d397712b 4367 while (1) {
39279cc3 4368 fi = NULL;
5f39d397
CM		 4369 leaf = path->nodes[0];
4370 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
962a298f 4371 found_type = found_key.type;
39279cc3 4372
33345d01 4373 if (found_key.objectid != ino)
39279cc3 4374 break;
5f39d397 4375
85e21bac 4376 if (found_type < min_type)
39279cc3
CM		 4377 break;
4378
5f39d397 4379 item_end = found_key.offset;
39279cc3 4380 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 4381 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 4382 struct btrfs_file_extent_item);
179e29e4
CM		 4383 extent_type = btrfs_file_extent_type(leaf, fi);
4384 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 4385 item_end +=
db94535d 4386 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 4387 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 4388 item_end += btrfs_file_extent_inline_len(leaf,
514ac8ad 4389 path->slots[0], fi);
39279cc3 4390 }
008630c1 4391 item_end--;
39279cc3 4392 }
8082510e
YZ		 4393 if (found_type > min_type) {
4394 del_item = 1;
4395 } else {
4396 if (item_end < new_size)
b888db2b 4397 break;
8082510e
YZ		 4398 if (found_key.offset >= new_size)
4399 del_item = 1;
4400 else
4401 del_item = 0;
39279cc3 4402 }
39279cc3 4403 found_extent = 0;
39279cc3 4404 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 4405 if (found_type != BTRFS_EXTENT_DATA_KEY)
4406 goto delete;
4407
7f4f6e0a
JB		 4408 if (del_item)
4409 last_size = found_key.offset;
4410 else
4411 last_size = new_size;
4412
179e29e4 4413 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 4414 u64 num_dec;
db94535d 4415 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
f70a9a6b 4416 if (!del_item) {
db94535d
CM		 4417 u64 orig_num_bytes =
4418 btrfs_file_extent_num_bytes(leaf, fi);
fda2832f
QW		 4419 extent_num_bytes = ALIGN(new_size -
4420 found_key.offset,
4421 root->sectorsize);
db94535d
CM		 4422 btrfs_set_file_extent_num_bytes(leaf, fi,
4423 extent_num_bytes);
4424 num_dec = (orig_num_bytes -
9069218d 4425 extent_num_bytes);
27cdeb70
MX		 4426 if (test_bit(BTRFS_ROOT_REF_COWS,
4427 &root->state) &&
4428 extent_start != 0)
a76a3cd4 4429 inode_sub_bytes(inode, num_dec);
5f39d397 4430 btrfs_mark_buffer_dirty(leaf);
39279cc3 4431 } else {
db94535d
CM		 4432 extent_num_bytes =
4433 btrfs_file_extent_disk_num_bytes(leaf,
4434 fi);
5d4f98a2
YZ		 4435 extent_offset = found_key.offset -
4436 btrfs_file_extent_offset(leaf, fi);
4437
39279cc3 4438 /* FIXME blocksize != 4096 */
9069218d 4439 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 4440 if (extent_start != 0) {
4441 found_extent = 1;
27cdeb70
MX		 4442 if (test_bit(BTRFS_ROOT_REF_COWS,
4443 &root->state))
a76a3cd4 4444 inode_sub_bytes(inode, num_dec);
e02119d5 4445 }
39279cc3 4446 }
9069218d 4447 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 4448 /*
4449 * we can't truncate inline items that have had
4450 * special encodings
4451 */
4452 if (!del_item &&
c8b97818
CM		 4453 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4454 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
514ac8ad
CM		 4455
4456 /*
0305cd5f
FM		 4457 * Need to release path in order to truncate a
4458 * compressed extent. So delete any accumulated
4459 * extent items so far.
514ac8ad 4460 */
0305cd5f
FM		 4461 if (btrfs_file_extent_compression(leaf, fi) !=
4462 BTRFS_COMPRESS_NONE && pending_del_nr) {
4463 err = btrfs_del_items(trans, root, path,
4464 pending_del_slot,
4465 pending_del_nr);
4466 if (err) {
4467 btrfs_abort_transaction(trans,
4468 root,
4469 err);
4470 goto error;
4471 }
4472 pending_del_nr = 0;
4473 }
4474
4475 err = truncate_inline_extent(inode, path,
4476 &found_key,
4477 item_end,
4478 new_size);
4479 if (err) {
4480 btrfs_abort_transaction(trans,
4481 root, err);
4482 goto error;
4483 }
27cdeb70
MX		 4484 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4485 &root->state)) {
0305cd5f 4486 inode_sub_bytes(inode, item_end + 1 - new_size);
9069218d 4487 }
39279cc3 4488 }
179e29e4 4489delete:
39279cc3 4490 if (del_item) {
85e21bac
CM		 4491 if (!pending_del_nr) {
4492 /* no pending yet, add ourselves */
4493 pending_del_slot = path->slots[0];
4494 pending_del_nr = 1;
4495 } else if (pending_del_nr &&
4496 path->slots[0] + 1 == pending_del_slot) {
4497 /* hop on the pending chunk */
4498 pending_del_nr++;
4499 pending_del_slot = path->slots[0];
4500 } else {
d397712b 4501 BUG();
85e21bac 4502 }
39279cc3
CM		 4503 } else {
4504 break;
4505 }
28f75a0e
CM		 4506 should_throttle = 0;
4507
27cdeb70
MX		 4508 if (found_extent &&
4509 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4510 root == root->fs_info->tree_root)) {
b9473439 4511 btrfs_set_path_blocking(path);
28ed1345 4512 bytes_deleted += extent_num_bytes;
39279cc3 4513 ret = btrfs_free_extent(trans, root, extent_start,
5d4f98a2
YZ		 4514 extent_num_bytes, 0,
4515 btrfs_header_owner(leaf),
b06c4bf5 4516 ino, extent_offset);
39279cc3 4517 BUG_ON(ret);
1262133b 4518 if (btrfs_should_throttle_delayed_refs(trans, root))
28ed1345
CM		 4519 btrfs_async_run_delayed_refs(root,
4520 trans->delayed_ref_updates * 2, 0);
28f75a0e
CM		 4521 if (be_nice) {
4522 if (truncate_space_check(trans, root,
4523 extent_num_bytes)) {
4524 should_end = 1;
4525 }
4526 if (btrfs_should_throttle_delayed_refs(trans,
4527 root)) {
4528 should_throttle = 1;
4529 }
4530 }
39279cc3 4531 }
85e21bac 4532
8082510e
YZ		 4533 if (found_type == BTRFS_INODE_ITEM_KEY)
4534 break;
4535
4536 if (path->slots[0] == 0 ||
1262133b 4537 path->slots[0] != pending_del_slot ||
28f75a0e 4538 should_throttle || should_end) {
8082510e
YZ		 4539 if (pending_del_nr) {
4540 ret = btrfs_del_items(trans, root, path,
4541 pending_del_slot,
4542 pending_del_nr);
79787eaa
JM		 4543 if (ret) {
4544 btrfs_abort_transaction(trans,
4545 root, ret);
4546 goto error;
4547 }
8082510e
YZ		 4548 pending_del_nr = 0;
4549 }
b3b4aa74 4550 btrfs_release_path(path);
28f75a0e 4551 if (should_throttle) {
1262133b
JB		 4552 unsigned long updates = trans->delayed_ref_updates;
4553 if (updates) {
4554 trans->delayed_ref_updates = 0;
4555 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4556 if (ret && !err)
4557 err = ret;
4558 }
4559 }
28f75a0e
CM		 4560 /*
4561 * if we failed to refill our space rsv, bail out
4562 * and let the transaction restart
4563 */
4564 if (should_end) {
4565 err = -EAGAIN;
4566 goto error;
4567 }
85e21bac 4568 goto search_again;
8082510e
YZ		 4569 } else {
4570 path->slots[0]--;
85e21bac 4571 }
39279cc3 4572 }
8082510e 4573out:
85e21bac
CM		 4574 if (pending_del_nr) {
4575 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4576 pending_del_nr);
79787eaa
JM		 4577 if (ret)
4578 btrfs_abort_transaction(trans, root, ret);
85e21bac 4579 }
79787eaa 4580error:
c1aa4575 4581 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
7f4f6e0a 4582 btrfs_ordered_update_i_size(inode, last_size, NULL);
28ed1345 4583
39279cc3 4584 btrfs_free_path(path);
28ed1345 4585
28f75a0e 4586 if (be_nice && bytes_deleted > 32 * 1024 * 1024) {
28ed1345
CM		 4587 unsigned long updates = trans->delayed_ref_updates;
4588 if (updates) {
4589 trans->delayed_ref_updates = 0;
4590 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4591 if (ret && !err)
4592 err = ret;
4593 }
4594 }
8082510e 4595 return err;
39279cc3
CM		 4596}
4597
4598/*
2aaa6655
JB		 4599 * btrfs_truncate_page - read, zero a chunk and write a page
4600 * @inode - inode that we're zeroing
4601 * @from - the offset to start zeroing
4602 * @len - the length to zero, 0 to zero the entire range respective to the
4603 * offset
4604 * @front - zero up to the offset instead of from the offset on
4605 *
4606 * This will find the page for the "from" offset and cow the page and zero the
4607 * part we want to zero. This is used with truncate and hole punching.
39279cc3 4608 */
2aaa6655
JB		 4609int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4610 int front)
39279cc3 4611{
2aaa6655 4612 struct address_space *mapping = inode->i_mapping;
db94535d 4613 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 4614 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4615 struct btrfs_ordered_extent *ordered;
2ac55d41 4616 struct extent_state *cached_state = NULL;
e6dcd2dc 4617 char *kaddr;
db94535d 4618 u32 blocksize = root->sectorsize;
39279cc3
CM		 4619 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4620 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4621 struct page *page;
3b16a4e3 4622 gfp_t mask = btrfs_alloc_write_mask(mapping);
39279cc3 4623 int ret = 0;
a52d9a80 4624 u64 page_start;
e6dcd2dc 4625 u64 page_end;
39279cc3 4626
2aaa6655
JB		 4627 if ((offset & (blocksize - 1)) == 0 &&
4628 (!len || ((len & (blocksize - 1)) == 0)))
39279cc3 4629 goto out;
7cf5b976 4630 ret = btrfs_delalloc_reserve_space(inode,
df480633 4631 round_down(from, PAGE_CACHE_SIZE), PAGE_CACHE_SIZE);
5d5e103a
JB		 4632 if (ret)
4633 goto out;
39279cc3 4634
211c17f5 4635again:
3b16a4e3 4636 page = find_or_create_page(mapping, index, mask);
5d5e103a 4637 if (!page) {
7cf5b976 4638 btrfs_delalloc_release_space(inode,
df480633
QW		 4639 round_down(from, PAGE_CACHE_SIZE),
4640 PAGE_CACHE_SIZE);
ac6a2b36 4641 ret = -ENOMEM;
39279cc3 4642 goto out;
5d5e103a 4643 }
e6dcd2dc
CM		 4644
4645 page_start = page_offset(page);
4646 page_end = page_start + PAGE_CACHE_SIZE - 1;
4647
39279cc3 4648 if (!PageUptodate(page)) {
9ebefb18 4649 ret = btrfs_readpage(NULL, page);
39279cc3 4650 lock_page(page);
211c17f5
CM		 4651 if (page->mapping != mapping) {
4652 unlock_page(page);
4653 page_cache_release(page);
4654 goto again;
4655 }
39279cc3
CM		 4656 if (!PageUptodate(page)) {
4657 ret = -EIO;
89642229 4658 goto out_unlock;
39279cc3
CM		 4659 }
4660 }
211c17f5 4661 wait_on_page_writeback(page);
e6dcd2dc 4662
d0082371 4663 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 4664 set_page_extent_mapped(page);
4665
4666 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4667 if (ordered) {
2ac55d41
JB		 4668 unlock_extent_cached(io_tree, page_start, page_end,
4669 &cached_state, GFP_NOFS);
e6dcd2dc
CM		 4670 unlock_page(page);
4671 page_cache_release(page);
eb84ae03 4672 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 4673 btrfs_put_ordered_extent(ordered);
4674 goto again;
4675 }
4676
2ac55d41 4677 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 4678 EXTENT_DIRTY | EXTENT_DELALLOC |
4679 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 4680 0, 0, &cached_state, GFP_NOFS);
5d5e103a 4681
2ac55d41
JB		 4682 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4683 &cached_state);
9ed74f2d 4684 if (ret) {
2ac55d41
JB		 4685 unlock_extent_cached(io_tree, page_start, page_end,
4686 &cached_state, GFP_NOFS);
9ed74f2d
JB		 4687 goto out_unlock;
4688 }
4689
e6dcd2dc 4690 if (offset != PAGE_CACHE_SIZE) {
2aaa6655
JB		 4691 if (!len)
4692 len = PAGE_CACHE_SIZE - offset;
e6dcd2dc 4693 kaddr = kmap(page);
2aaa6655
JB		 4694 if (front)
4695 memset(kaddr, 0, offset);
4696 else
4697 memset(kaddr + offset, 0, len);
e6dcd2dc
CM		 4698 flush_dcache_page(page);
4699 kunmap(page);
4700 }
247e743c 4701 ClearPageChecked(page);
e6dcd2dc 4702 set_page_dirty(page);
2ac55d41
JB		 4703 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4704 GFP_NOFS);
39279cc3 4705
89642229 4706out_unlock:
5d5e103a 4707 if (ret)
7cf5b976
QW		 4708 btrfs_delalloc_release_space(inode, page_start,
4709 PAGE_CACHE_SIZE);
39279cc3
CM		 4710 unlock_page(page);
4711 page_cache_release(page);
4712out:
4713 return ret;
4714}
4715
16e7549f
JB		 4716static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4717 u64 offset, u64 len)
4718{
4719 struct btrfs_trans_handle *trans;
4720 int ret;
4721
4722 /*
4723 * Still need to make sure the inode looks like it's been updated so
4724 * that any holes get logged if we fsync.
4725 */
4726 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4727 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4728 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4729 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4730 return 0;
4731 }
4732
4733 /*
4734 * 1 - for the one we're dropping
4735 * 1 - for the one we're adding
4736 * 1 - for updating the inode.
4737 */
4738 trans = btrfs_start_transaction(root, 3);
4739 if (IS_ERR(trans))
4740 return PTR_ERR(trans);
4741
4742 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4743 if (ret) {
4744 btrfs_abort_transaction(trans, root, ret);
4745 btrfs_end_transaction(trans, root);
4746 return ret;
4747 }
4748
4749 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4750 0, 0, len, 0, len, 0, 0, 0);
4751 if (ret)
4752 btrfs_abort_transaction(trans, root, ret);
4753 else
4754 btrfs_update_inode(trans, root, inode);
4755 btrfs_end_transaction(trans, root);
4756 return ret;
4757}
4758
695a0d0d
JB		 4759/*
4760 * This function puts in dummy file extents for the area we're creating a hole
4761 * for. So if we are truncating this file to a larger size we need to insert
4762 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4763 * the range between oldsize and size
4764 */
a41ad394 4765int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
39279cc3 4766{
9036c102
YZ		 4767 struct btrfs_root *root = BTRFS_I(inode)->root;
4768 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a22285a6 4769 struct extent_map *em = NULL;
2ac55d41 4770 struct extent_state *cached_state = NULL;
5dc562c5 4771 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
fda2832f
QW		 4772 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4773 u64 block_end = ALIGN(size, root->sectorsize);
9036c102
YZ		 4774 u64 last_byte;
4775 u64 cur_offset;
4776 u64 hole_size;
9ed74f2d 4777 int err = 0;
39279cc3 4778
a71754fc
JB		 4779 /*
4780 * If our size started in the middle of a page we need to zero out the
4781 * rest of the page before we expand the i_size, otherwise we could
4782 * expose stale data.
4783 */
4784 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4785 if (err)
4786 return err;
4787
9036c102
YZ		 4788 if (size <= hole_start)
4789 return 0;
4790
9036c102
YZ		 4791 while (1) {
4792 struct btrfs_ordered_extent *ordered;
fa7c1494 4793
2ac55d41 4794 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
d0082371 4795 &cached_state);
fa7c1494
MX		 4796 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4797 block_end - hole_start);
9036c102
YZ		 4798 if (!ordered)
4799 break;
2ac55d41
JB		 4800 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4801 &cached_state, GFP_NOFS);
fa7c1494 4802 btrfs_start_ordered_extent(inode, ordered, 1);
9036c102
YZ		 4803 btrfs_put_ordered_extent(ordered);
4804 }
39279cc3 4805
9036c102
YZ		 4806 cur_offset = hole_start;
4807 while (1) {
4808 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4809 block_end - cur_offset, 0);
79787eaa
JM		 4810 if (IS_ERR(em)) {
4811 err = PTR_ERR(em);
f2767956 4812 em = NULL;
79787eaa
JM		 4813 break;
4814 }
9036c102 4815 last_byte = min(extent_map_end(em), block_end);
fda2832f 4816 last_byte = ALIGN(last_byte , root->sectorsize);
8082510e 4817 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5dc562c5 4818 struct extent_map *hole_em;
9036c102 4819 hole_size = last_byte - cur_offset;
9ed74f2d 4820
16e7549f
JB		 4821 err = maybe_insert_hole(root, inode, cur_offset,
4822 hole_size);
4823 if (err)
3893e33b 4824 break;
5dc562c5
JB		 4825 btrfs_drop_extent_cache(inode, cur_offset,
4826 cur_offset + hole_size - 1, 0);
4827 hole_em = alloc_extent_map();
4828 if (!hole_em) {
4829 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4830 &BTRFS_I(inode)->runtime_flags);
4831 goto next;
4832 }
4833 hole_em->start = cur_offset;
4834 hole_em->len = hole_size;
4835 hole_em->orig_start = cur_offset;
8082510e 4836
5dc562c5
JB		 4837 hole_em->block_start = EXTENT_MAP_HOLE;
4838 hole_em->block_len = 0;
b4939680 4839 hole_em->orig_block_len = 0;
cc95bef6 4840 hole_em->ram_bytes = hole_size;
5dc562c5
JB		 4841 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4842 hole_em->compress_type = BTRFS_COMPRESS_NONE;
16e7549f 4843 hole_em->generation = root->fs_info->generation;
8082510e 4844
5dc562c5
JB		 4845 while (1) {
4846 write_lock(&em_tree->lock);
09a2a8f9 4847 err = add_extent_mapping(em_tree, hole_em, 1);
5dc562c5
JB		 4848 write_unlock(&em_tree->lock);
4849 if (err != -EEXIST)
4850 break;
4851 btrfs_drop_extent_cache(inode, cur_offset,
4852 cur_offset +
4853 hole_size - 1, 0);
4854 }
4855 free_extent_map(hole_em);
9036c102 4856 }
16e7549f 4857next:
9036c102 4858 free_extent_map(em);
a22285a6 4859 em = NULL;
9036c102 4860 cur_offset = last_byte;
8082510e 4861 if (cur_offset >= block_end)
9036c102
YZ		 4862 break;
4863 }
a22285a6 4864 free_extent_map(em);
2ac55d41
JB		 4865 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4866 GFP_NOFS);
9036c102
YZ		 4867 return err;
4868}
39279cc3 4869
9ea24bbe
FM		 4870static int wait_snapshoting_atomic_t(atomic_t *a)
4871{
4872 schedule();
4873 return 0;
4874}
4875
4876static void wait_for_snapshot_creation(struct btrfs_root *root)
4877{
4878 while (true) {
4879 int ret;
4880
4881 ret = btrfs_start_write_no_snapshoting(root);
4882 if (ret)
4883 break;
4884 wait_on_atomic_t(&root->will_be_snapshoted,
4885 wait_snapshoting_atomic_t,
4886 TASK_UNINTERRUPTIBLE);
4887 }
4888}
4889
3972f260 4890static int btrfs_setsize(struct inode *inode, struct iattr *attr)
8082510e 4891{
f4a2f4c5
MX		 4892 struct btrfs_root *root = BTRFS_I(inode)->root;
4893 struct btrfs_trans_handle *trans;
a41ad394 4894 loff_t oldsize = i_size_read(inode);
3972f260
ES		 4895 loff_t newsize = attr->ia_size;
4896 int mask = attr->ia_valid;
8082510e
YZ		 4897 int ret;
4898
3972f260
ES		 4899 /*
4900 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4901 * special case where we need to update the times despite not having
4902 * these flags set. For all other operations the VFS set these flags
4903 * explicitly if it wants a timestamp update.
4904 */
dff6efc3
CH		 4905 if (newsize != oldsize) {
4906 inode_inc_iversion(inode);
4907 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4908 inode->i_ctime = inode->i_mtime =
4909 current_fs_time(inode->i_sb);
4910 }
3972f260 4911
a41ad394 4912 if (newsize > oldsize) {
7caef267 4913 truncate_pagecache(inode, newsize);
9ea24bbe
FM		 4914 /*
4915 * Don't do an expanding truncate while snapshoting is ongoing.
4916 * This is to ensure the snapshot captures a fully consistent
4917 * state of this file - if the snapshot captures this expanding
4918 * truncation, it must capture all writes that happened before
4919 * this truncation.
4920 */
4921 wait_for_snapshot_creation(root);
a41ad394 4922 ret = btrfs_cont_expand(inode, oldsize, newsize);
9ea24bbe
FM		 4923 if (ret) {
4924 btrfs_end_write_no_snapshoting(root);
8082510e 4925 return ret;
9ea24bbe 4926 }
8082510e 4927
f4a2f4c5 4928 trans = btrfs_start_transaction(root, 1);
9ea24bbe
FM		 4929 if (IS_ERR(trans)) {
4930 btrfs_end_write_no_snapshoting(root);
f4a2f4c5 4931 return PTR_ERR(trans);
9ea24bbe 4932 }
f4a2f4c5
MX		 4933
4934 i_size_write(inode, newsize);
4935 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4936 ret = btrfs_update_inode(trans, root, inode);
9ea24bbe 4937 btrfs_end_write_no_snapshoting(root);
7ad85bb7 4938 btrfs_end_transaction(trans, root);
a41ad394 4939 } else {
8082510e 4940
a41ad394
JB		 4941 /*
4942 * We're truncating a file that used to have good data down to
4943 * zero. Make sure it gets into the ordered flush list so that
4944 * any new writes get down to disk quickly.
4945 */
4946 if (newsize == 0)
72ac3c0d
JB		 4947 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4948 &BTRFS_I(inode)->runtime_flags);
8082510e 4949
f3fe820c
JB		 4950 /*
4951 * 1 for the orphan item we're going to add
4952 * 1 for the orphan item deletion.
4953 */
4954 trans = btrfs_start_transaction(root, 2);
4955 if (IS_ERR(trans))
4956 return PTR_ERR(trans);
4957
4958 /*
4959 * We need to do this in case we fail at _any_ point during the
4960 * actual truncate. Once we do the truncate_setsize we could
4961 * invalidate pages which forces any outstanding ordered io to
4962 * be instantly completed which will give us extents that need
4963 * to be truncated. If we fail to get an orphan inode down we
4964 * could have left over extents that were never meant to live,
4965 * so we need to garuntee from this point on that everything
4966 * will be consistent.
4967 */
4968 ret = btrfs_orphan_add(trans, inode);
4969 btrfs_end_transaction(trans, root);
4970 if (ret)
4971 return ret;
4972
a41ad394
JB		 4973 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4974 truncate_setsize(inode, newsize);
2e60a51e
MX		 4975
4976 /* Disable nonlocked read DIO to avoid the end less truncate */
4977 btrfs_inode_block_unlocked_dio(inode);
4978 inode_dio_wait(inode);
4979 btrfs_inode_resume_unlocked_dio(inode);
4980
a41ad394 4981 ret = btrfs_truncate(inode);
7f4f6e0a
JB		 4982 if (ret && inode->i_nlink) {
4983 int err;
4984
4985 /*
4986 * failed to truncate, disk_i_size is only adjusted down
4987 * as we remove extents, so it should represent the true
4988 * size of the inode, so reset the in memory size and
4989 * delete our orphan entry.
4990 */
4991 trans = btrfs_join_transaction(root);
4992 if (IS_ERR(trans)) {
4993 btrfs_orphan_del(NULL, inode);
4994 return ret;
4995 }
4996 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4997 err = btrfs_orphan_del(trans, inode);
4998 if (err)
4999 btrfs_abort_transaction(trans, root, err);
5000 btrfs_end_transaction(trans, root);
5001 }
8082510e
YZ		 5002 }
5003
a41ad394 5004 return ret;
8082510e
YZ		 5005}
5006
9036c102
YZ		 5007static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5008{
2b0143b5 5009 struct inode *inode = d_inode(dentry);
b83cc969 5010 struct btrfs_root *root = BTRFS_I(inode)->root;
9036c102 5011 int err;
39279cc3 5012
b83cc969
LZ		 5013 if (btrfs_root_readonly(root))
5014 return -EROFS;
5015
9036c102
YZ		 5016 err = inode_change_ok(inode, attr);
5017 if (err)
5018 return err;
2bf5a725 5019
5a3f23d5 5020 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3972f260 5021 err = btrfs_setsize(inode, attr);
8082510e
YZ		 5022 if (err)
5023 return err;
39279cc3 5024 }
9036c102 5025
1025774c
CH		 5026 if (attr->ia_valid) {
5027 setattr_copy(inode, attr);
0c4d2d95 5028 inode_inc_iversion(inode);
22c44fe6 5029 err = btrfs_dirty_inode(inode);
1025774c 5030
22c44fe6 5031 if (!err && attr->ia_valid & ATTR_MODE)
996a710d 5032 err = posix_acl_chmod(inode, inode->i_mode);
1025774c 5033 }
33268eaf 5034
39279cc3
CM		 5035 return err;
5036}
61295eb8 5037
131e404a
FDBM		 5038/*
5039 * While truncating the inode pages during eviction, we get the VFS calling
5040 * btrfs_invalidatepage() against each page of the inode. This is slow because
5041 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5042 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5043 * extent_state structures over and over, wasting lots of time.
5044 *
5045 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5046 * those expensive operations on a per page basis and do only the ordered io
5047 * finishing, while we release here the extent_map and extent_state structures,
5048 * without the excessive merging and splitting.
5049 */
5050static void evict_inode_truncate_pages(struct inode *inode)
5051{
5052 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5053 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5054 struct rb_node *node;
5055
5056 ASSERT(inode->i_state & I_FREEING);
91b0abe3 5057 truncate_inode_pages_final(&inode->i_data);
131e404a
FDBM		 5058
5059 write_lock(&map_tree->lock);
5060 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5061 struct extent_map *em;
5062
5063 node = rb_first(&map_tree->map);
5064 em = rb_entry(node, struct extent_map, rb_node);
180589ef
WS		 5065 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5066 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
131e404a
FDBM		 5067 remove_extent_mapping(map_tree, em);
5068 free_extent_map(em);
7064dd5c
FM		 5069 if (need_resched()) {
5070 write_unlock(&map_tree->lock);
5071 cond_resched();
5072 write_lock(&map_tree->lock);
5073 }
131e404a
FDBM		 5074 }
5075 write_unlock(&map_tree->lock);
5076
6ca07097
FM		 5077 /*
5078 * Keep looping until we have no more ranges in the io tree.
5079 * We can have ongoing bios started by readpages (called from readahead)
9c6429d9
FM		 5080 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5081 * still in progress (unlocked the pages in the bio but did not yet
5082 * unlocked the ranges in the io tree). Therefore this means some
6ca07097
FM		 5083 * ranges can still be locked and eviction started because before
5084 * submitting those bios, which are executed by a separate task (work
5085 * queue kthread), inode references (inode->i_count) were not taken
5086 * (which would be dropped in the end io callback of each bio).
5087 * Therefore here we effectively end up waiting for those bios and
5088 * anyone else holding locked ranges without having bumped the inode's
5089 * reference count - if we don't do it, when they access the inode's
5090 * io_tree to unlock a range it may be too late, leading to an
5091 * use-after-free issue.
5092 */
131e404a
FDBM		 5093 spin_lock(&io_tree->lock);
5094 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5095 struct extent_state *state;
5096 struct extent_state *cached_state = NULL;
6ca07097
FM		 5097 u64 start;
5098 u64 end;
131e404a
FDBM		 5099
5100 node = rb_first(&io_tree->state);
5101 state = rb_entry(node, struct extent_state, rb_node);
6ca07097
FM		 5102 start = state->start;
5103 end = state->end;
131e404a
FDBM		 5104 spin_unlock(&io_tree->lock);
5105
6ca07097 5106 lock_extent_bits(io_tree, start, end, 0, &cached_state);
b9d0b389
QW		 5107
5108 /*
5109 * If still has DELALLOC flag, the extent didn't reach disk,
5110 * and its reserved space won't be freed by delayed_ref.
5111 * So we need to free its reserved space here.
5112 * (Refer to comment in btrfs_invalidatepage, case 2)
5113 *
5114 * Note, end is the bytenr of last byte, so we need + 1 here.
5115 */
5116 if (state->state & EXTENT_DELALLOC)
5117 btrfs_qgroup_free_data(inode, start, end - start + 1);
5118
6ca07097 5119 clear_extent_bit(io_tree, start, end,
131e404a
FDBM		 5120 EXTENT_LOCKED | EXTENT_DIRTY |
5121 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5122 EXTENT_DEFRAG, 1, 1,
5123 &cached_state, GFP_NOFS);
131e404a 5124
7064dd5c 5125 cond_resched();
131e404a
FDBM		 5126 spin_lock(&io_tree->lock);
5127 }
5128 spin_unlock(&io_tree->lock);
5129}
5130
bd555975 5131void btrfs_evict_inode(struct inode *inode)
39279cc3
CM		 5132{
5133 struct btrfs_trans_handle *trans;
5134 struct btrfs_root *root = BTRFS_I(inode)->root;
726c35fa 5135 struct btrfs_block_rsv *rsv, *global_rsv;
3bce876f 5136 int steal_from_global = 0;
07127184 5137 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3
CM		 5138 int ret;
5139
1abe9b8a 5140 trace_btrfs_inode_evict(inode);
5141
131e404a
FDBM		 5142 evict_inode_truncate_pages(inode);
5143
69e9c6c6
SB		 5144 if (inode->i_nlink &&
5145 ((btrfs_root_refs(&root->root_item) != 0 &&
5146 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5147 btrfs_is_free_space_inode(inode)))
bd555975
AV		 5148 goto no_delete;
5149
39279cc3 5150 if (is_bad_inode(inode)) {
7b128766 5151 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 5152 goto no_delete;
5153 }
bd555975 5154 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
a30e577c
JM		 5155 if (!special_file(inode->i_mode))
5156 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 5157
f612496b
MX		 5158 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5159
c71bf099 5160 if (root->fs_info->log_root_recovering) {
6bf02314 5161 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8a35d95f 5162 &BTRFS_I(inode)->runtime_flags));
c71bf099
YZ		 5163 goto no_delete;
5164 }
5165
76dda93c 5166 if (inode->i_nlink > 0) {
69e9c6c6
SB		 5167 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5168 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
76dda93c
YZ		 5169 goto no_delete;
5170 }
5171
0e8c36a9
MX		 5172 ret = btrfs_commit_inode_delayed_inode(inode);
5173 if (ret) {
5174 btrfs_orphan_del(NULL, inode);
5175 goto no_delete;
5176 }
5177
66d8f3dd 5178 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4289a667
JB		 5179 if (!rsv) {
5180 btrfs_orphan_del(NULL, inode);
5181 goto no_delete;
5182 }
4a338542 5183 rsv->size = min_size;
ca7e70f5 5184 rsv->failfast = 1;
726c35fa 5185 global_rsv = &root->fs_info->global_block_rsv;
4289a667 5186
dbe674a9 5187 btrfs_i_size_write(inode, 0);
5f39d397 5188
4289a667 5189 /*
8407aa46
MX		 5190 * This is a bit simpler than btrfs_truncate since we've already
5191 * reserved our space for our orphan item in the unlink, so we just
5192 * need to reserve some slack space in case we add bytes and update
5193 * inode item when doing the truncate.
4289a667 5194 */
8082510e 5195 while (1) {
08e007d2
MX		 5196 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5197 BTRFS_RESERVE_FLUSH_LIMIT);
726c35fa
JB		 5198
5199 /*
5200 * Try and steal from the global reserve since we will
5201 * likely not use this space anyway, we want to try as
5202 * hard as possible to get this to work.
5203 */
5204 if (ret)
3bce876f
JB		 5205 steal_from_global++;
5206 else
5207 steal_from_global = 0;
5208 ret = 0;
d68fc57b 5209
3bce876f
JB		 5210 /*
5211 * steal_from_global == 0: we reserved stuff, hooray!
5212 * steal_from_global == 1: we didn't reserve stuff, boo!
5213 * steal_from_global == 2: we've committed, still not a lot of
5214 * room but maybe we'll have room in the global reserve this
5215 * time.
5216 * steal_from_global == 3: abandon all hope!
5217 */
5218 if (steal_from_global > 2) {
c2cf52eb
SK		 5219 btrfs_warn(root->fs_info,
5220 "Could not get space for a delete, will truncate on mount %d",
5221 ret);
4289a667
JB		 5222 btrfs_orphan_del(NULL, inode);
5223 btrfs_free_block_rsv(root, rsv);
5224 goto no_delete;
d68fc57b 5225 }
7b128766 5226
0e8c36a9 5227 trans = btrfs_join_transaction(root);
4289a667
JB		 5228 if (IS_ERR(trans)) {
5229 btrfs_orphan_del(NULL, inode);
5230 btrfs_free_block_rsv(root, rsv);
5231 goto no_delete;
d68fc57b 5232 }
7b128766 5233
3bce876f
JB		 5234 /*
5235 * We can't just steal from the global reserve, we need tomake
5236 * sure there is room to do it, if not we need to commit and try
5237 * again.
5238 */
5239 if (steal_from_global) {
5240 if (!btrfs_check_space_for_delayed_refs(trans, root))
5241 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5242 min_size);
5243 else
5244 ret = -ENOSPC;
5245 }
5246
5247 /*
5248 * Couldn't steal from the global reserve, we have too much
5249 * pending stuff built up, commit the transaction and try it
5250 * again.
5251 */
5252 if (ret) {
5253 ret = btrfs_commit_transaction(trans, root);
5254 if (ret) {
5255 btrfs_orphan_del(NULL, inode);
5256 btrfs_free_block_rsv(root, rsv);
5257 goto no_delete;
5258 }
5259 continue;
5260 } else {
5261 steal_from_global = 0;
5262 }
5263
4289a667
JB		 5264 trans->block_rsv = rsv;
5265
d68fc57b 5266 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
28ed1345 5267 if (ret != -ENOSPC && ret != -EAGAIN)
8082510e 5268 break;
85e21bac 5269
8407aa46 5270 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e
YZ		 5271 btrfs_end_transaction(trans, root);
5272 trans = NULL;
b53d3f5d 5273 btrfs_btree_balance_dirty(root);
8082510e 5274 }
5f39d397 5275
4289a667
JB		 5276 btrfs_free_block_rsv(root, rsv);
5277
4ef31a45
JB		 5278 /*
5279 * Errors here aren't a big deal, it just means we leave orphan items
5280 * in the tree. They will be cleaned up on the next mount.
5281 */
8082510e 5282 if (ret == 0) {
4289a667 5283 trans->block_rsv = root->orphan_block_rsv;
4ef31a45
JB		 5284 btrfs_orphan_del(trans, inode);
5285 } else {
5286 btrfs_orphan_del(NULL, inode);
8082510e 5287 }
54aa1f4d 5288
4289a667 5289 trans->block_rsv = &root->fs_info->trans_block_rsv;
581bb050
LZ		 5290 if (!(root == root->fs_info->tree_root ||
5291 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
33345d01 5292 btrfs_return_ino(root, btrfs_ino(inode));
581bb050 5293
54aa1f4d 5294 btrfs_end_transaction(trans, root);
b53d3f5d 5295 btrfs_btree_balance_dirty(root);
39279cc3 5296no_delete:
89042e5a 5297 btrfs_remove_delayed_node(inode);
dbd5768f 5298 clear_inode(inode);
8082510e 5299 return;
39279cc3
CM		 5300}
5301
5302/*
5303 * this returns the key found in the dir entry in the location pointer.
5304 * If no dir entries were found, location->objectid is 0.
5305 */
5306static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5307 struct btrfs_key *location)
5308{
5309 const char *name = dentry->d_name.name;
5310 int namelen = dentry->d_name.len;
5311 struct btrfs_dir_item *di;
5312 struct btrfs_path *path;
5313 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 5314 int ret = 0;
39279cc3
CM		 5315
5316 path = btrfs_alloc_path();
d8926bb3
MF		 5317 if (!path)
5318 return -ENOMEM;
3954401f 5319
33345d01 5320 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
39279cc3 5321 namelen, 0);
0d9f7f3e
Y		 5322 if (IS_ERR(di))
5323 ret = PTR_ERR(di);
d397712b 5324
c704005d 5325 if (IS_ERR_OR_NULL(di))
3954401f 5326 goto out_err;
d397712b 5327
5f39d397 5328 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 5329out:
39279cc3
CM		 5330 btrfs_free_path(path);
5331 return ret;
3954401f
CM		 5332out_err:
5333 location->objectid = 0;
5334 goto out;
39279cc3
CM		 5335}
5336
5337/*
5338 * when we hit a tree root in a directory, the btrfs part of the inode
5339 * needs to be changed to reflect the root directory of the tree root. This
5340 * is kind of like crossing a mount point.
5341 */
5342static int fixup_tree_root_location(struct btrfs_root *root,
4df27c4d
YZ		 5343 struct inode *dir,
5344 struct dentry *dentry,
5345 struct btrfs_key *location,
5346 struct btrfs_root **sub_root)
39279cc3 5347{
4df27c4d
YZ		 5348 struct btrfs_path *path;
5349 struct btrfs_root *new_root;
5350 struct btrfs_root_ref *ref;
5351 struct extent_buffer *leaf;
1d4c08e0 5352 struct btrfs_key key;
4df27c4d
YZ		 5353 int ret;
5354 int err = 0;
39279cc3 5355
4df27c4d
YZ		 5356 path = btrfs_alloc_path();
5357 if (!path) {
5358 err = -ENOMEM;
5359 goto out;
5360 }
39279cc3 5361
4df27c4d 5362 err = -ENOENT;
1d4c08e0
DS		 5363 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5364 key.type = BTRFS_ROOT_REF_KEY;
5365 key.offset = location->objectid;
5366
5367 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5368 0, 0);
4df27c4d
YZ		 5369 if (ret) {
5370 if (ret < 0)
5371 err = ret;
5372 goto out;
5373 }
39279cc3 5374
4df27c4d
YZ		 5375 leaf = path->nodes[0];
5376 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
33345d01 5377 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4df27c4d
YZ		 5378 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5379 goto out;
39279cc3 5380
4df27c4d
YZ		 5381 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5382 (unsigned long)(ref + 1),
5383 dentry->d_name.len);
5384 if (ret)
5385 goto out;
5386
b3b4aa74 5387 btrfs_release_path(path);
4df27c4d
YZ		 5388
5389 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5390 if (IS_ERR(new_root)) {
5391 err = PTR_ERR(new_root);
5392 goto out;
5393 }
5394
4df27c4d
YZ		 5395 *sub_root = new_root;
5396 location->objectid = btrfs_root_dirid(&new_root->root_item);
5397 location->type = BTRFS_INODE_ITEM_KEY;
5398 location->offset = 0;
5399 err = 0;
5400out:
5401 btrfs_free_path(path);
5402 return err;
39279cc3
CM		 5403}
5404
5d4f98a2
YZ		 5405static void inode_tree_add(struct inode *inode)
5406{
5407 struct btrfs_root *root = BTRFS_I(inode)->root;
5408 struct btrfs_inode *entry;
03e860bd
FNP		 5409 struct rb_node **p;
5410 struct rb_node *parent;
cef21937 5411 struct rb_node *new = &BTRFS_I(inode)->rb_node;
33345d01 5412 u64 ino = btrfs_ino(inode);
5d4f98a2 5413
1d3382cb 5414 if (inode_unhashed(inode))
76dda93c 5415 return;
e1409cef 5416 parent = NULL;
5d4f98a2 5417 spin_lock(&root->inode_lock);
e1409cef 5418 p = &root->inode_tree.rb_node;
5d4f98a2
YZ		 5419 while (*p) {
5420 parent = *p;
5421 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5422
33345d01 5423 if (ino < btrfs_ino(&entry->vfs_inode))
03e860bd 5424 p = &parent->rb_left;
33345d01 5425 else if (ino > btrfs_ino(&entry->vfs_inode))
03e860bd 5426 p = &parent->rb_right;
5d4f98a2
YZ		 5427 else {
5428 WARN_ON(!(entry->vfs_inode.i_state &
a4ffdde6 5429 (I_WILL_FREE | I_FREEING)));
cef21937 5430 rb_replace_node(parent, new, &root->inode_tree);
03e860bd
FNP		 5431 RB_CLEAR_NODE(parent);
5432 spin_unlock(&root->inode_lock);
cef21937 5433 return;
5d4f98a2
YZ		 5434 }
5435 }
cef21937
FDBM		 5436 rb_link_node(new, parent, p);
5437 rb_insert_color(new, &root->inode_tree);
5d4f98a2
YZ		 5438 spin_unlock(&root->inode_lock);
5439}
5440
5441static void inode_tree_del(struct inode *inode)
5442{
5443 struct btrfs_root *root = BTRFS_I(inode)->root;
76dda93c 5444 int empty = 0;
5d4f98a2 5445
03e860bd 5446 spin_lock(&root->inode_lock);
5d4f98a2 5447 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5d4f98a2 5448 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5d4f98a2 5449 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
76dda93c 5450 empty = RB_EMPTY_ROOT(&root->inode_tree);
5d4f98a2 5451 }
03e860bd 5452 spin_unlock(&root->inode_lock);
76dda93c 5453
69e9c6c6 5454 if (empty && btrfs_root_refs(&root->root_item) == 0) {
76dda93c
YZ		 5455 synchronize_srcu(&root->fs_info->subvol_srcu);
5456 spin_lock(&root->inode_lock);
5457 empty = RB_EMPTY_ROOT(&root->inode_tree);
5458 spin_unlock(&root->inode_lock);
5459 if (empty)
5460 btrfs_add_dead_root(root);
5461 }
5462}
5463
143bede5 5464void btrfs_invalidate_inodes(struct btrfs_root *root)
76dda93c
YZ		 5465{
5466 struct rb_node *node;
5467 struct rb_node *prev;
5468 struct btrfs_inode *entry;
5469 struct inode *inode;
5470 u64 objectid = 0;
5471
7813b3db
LB		 5472 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5473 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
76dda93c
YZ		 5474
5475 spin_lock(&root->inode_lock);
5476again:
5477 node = root->inode_tree.rb_node;
5478 prev = NULL;
5479 while (node) {
5480 prev = node;
5481 entry = rb_entry(node, struct btrfs_inode, rb_node);
5482
33345d01 5483 if (objectid < btrfs_ino(&entry->vfs_inode))
76dda93c 5484 node = node->rb_left;
33345d01 5485 else if (objectid > btrfs_ino(&entry->vfs_inode))
76dda93c
YZ		 5486 node = node->rb_right;
5487 else
5488 break;
5489 }
5490 if (!node) {
5491 while (prev) {
5492 entry = rb_entry(prev, struct btrfs_inode, rb_node);
33345d01 5493 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
76dda93c
YZ		 5494 node = prev;
5495 break;
5496 }
5497 prev = rb_next(prev);
5498 }
5499 }
5500 while (node) {
5501 entry = rb_entry(node, struct btrfs_inode, rb_node);
33345d01 5502 objectid = btrfs_ino(&entry->vfs_inode) + 1;
76dda93c
YZ		 5503 inode = igrab(&entry->vfs_inode);
5504 if (inode) {
5505 spin_unlock(&root->inode_lock);
5506 if (atomic_read(&inode->i_count) > 1)
5507 d_prune_aliases(inode);
5508 /*
45321ac5 5509 * btrfs_drop_inode will have it removed from
76dda93c
YZ		 5510 * the inode cache when its usage count
5511 * hits zero.
5512 */
5513 iput(inode);
5514 cond_resched();
5515 spin_lock(&root->inode_lock);
5516 goto again;
5517 }
5518
5519 if (cond_resched_lock(&root->inode_lock))
5520 goto again;
5521
5522 node = rb_next(node);
5523 }
5524 spin_unlock(&root->inode_lock);
5d4f98a2
YZ		 5525}
5526
e02119d5
CM		 5527static int btrfs_init_locked_inode(struct inode *inode, void *p)
5528{
5529 struct btrfs_iget_args *args = p;
90d3e592
CM		 5530 inode->i_ino = args->location->objectid;
5531 memcpy(&BTRFS_I(inode)->location, args->location,
5532 sizeof(*args->location));
e02119d5 5533 BTRFS_I(inode)->root = args->root;
39279cc3
CM		 5534 return 0;
5535}
5536
5537static int btrfs_find_actor(struct inode *inode, void *opaque)
5538{
5539 struct btrfs_iget_args *args = opaque;
90d3e592 5540 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
d397712b 5541 args->root == BTRFS_I(inode)->root;
39279cc3
CM		 5542}
5543
5d4f98a2 5544static struct inode *btrfs_iget_locked(struct super_block *s,
90d3e592 5545 struct btrfs_key *location,
5d4f98a2 5546 struct btrfs_root *root)
39279cc3
CM		 5547{
5548 struct inode *inode;
5549 struct btrfs_iget_args args;
90d3e592 5550 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
778ba82b 5551
90d3e592 5552 args.location = location;
39279cc3
CM		 5553 args.root = root;
5554
778ba82b 5555 inode = iget5_locked(s, hashval, btrfs_find_actor,
39279cc3
CM		 5556 btrfs_init_locked_inode,
5557 (void *)&args);
5558 return inode;
5559}
5560
1a54ef8c
BR		 5561/* Get an inode object given its location and corresponding root.
5562 * Returns in *is_new if the inode was read from disk
5563 */
5564struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
73f73415 5565 struct btrfs_root *root, int *new)
1a54ef8c
BR		 5566{
5567 struct inode *inode;
5568
90d3e592 5569 inode = btrfs_iget_locked(s, location, root);
1a54ef8c 5570 if (!inode)
5d4f98a2 5571 return ERR_PTR(-ENOMEM);
1a54ef8c
BR		 5572
5573 if (inode->i_state & I_NEW) {
1a54ef8c 5574 btrfs_read_locked_inode(inode);
1748f843
MF		 5575 if (!is_bad_inode(inode)) {
5576 inode_tree_add(inode);
5577 unlock_new_inode(inode);
5578 if (new)
5579 *new = 1;
5580 } else {
e0b6d65b
ST		 5581 unlock_new_inode(inode);
5582 iput(inode);
5583 inode = ERR_PTR(-ESTALE);
1748f843
MF		 5584 }
5585 }
5586
1a54ef8c
BR		 5587 return inode;
5588}
5589
4df27c4d
YZ		 5590static struct inode *new_simple_dir(struct super_block *s,
5591 struct btrfs_key *key,
5592 struct btrfs_root *root)
5593{
5594 struct inode *inode = new_inode(s);
5595
5596 if (!inode)
5597 return ERR_PTR(-ENOMEM);
5598
4df27c4d
YZ		 5599 BTRFS_I(inode)->root = root;
5600 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
72ac3c0d 5601 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4df27c4d
YZ		 5602
5603 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
848cce0d 5604 inode->i_op = &btrfs_dir_ro_inode_operations;
4df27c4d
YZ		 5605 inode->i_fop = &simple_dir_operations;
5606 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
9cc97d64 5607 inode->i_mtime = CURRENT_TIME;
5608 inode->i_atime = inode->i_mtime;
5609 inode->i_ctime = inode->i_mtime;
5610 BTRFS_I(inode)->i_otime = inode->i_mtime;
4df27c4d
YZ		 5611
5612 return inode;
5613}
5614
3de4586c 5615struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3 5616{
d397712b 5617 struct inode *inode;
4df27c4d 5618 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3
CM		 5619 struct btrfs_root *sub_root = root;
5620 struct btrfs_key location;
76dda93c 5621 int index;
b4aff1f8 5622 int ret = 0;
39279cc3
CM		 5623
5624 if (dentry->d_name.len > BTRFS_NAME_LEN)
5625 return ERR_PTR(-ENAMETOOLONG);
5f39d397 5626
39e3c955 5627 ret = btrfs_inode_by_name(dir, dentry, &location);
39279cc3
CM		 5628 if (ret < 0)
5629 return ERR_PTR(ret);
5f39d397 5630
4df27c4d 5631 if (location.objectid == 0)
5662344b 5632 return ERR_PTR(-ENOENT);
4df27c4d
YZ		 5633
5634 if (location.type == BTRFS_INODE_ITEM_KEY) {
73f73415 5635 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4df27c4d
YZ		 5636 return inode;
5637 }
5638
5639 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5640
76dda93c 5641 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4df27c4d
YZ		 5642 ret = fixup_tree_root_location(root, dir, dentry,
5643 &location, &sub_root);
5644 if (ret < 0) {
5645 if (ret != -ENOENT)
5646 inode = ERR_PTR(ret);
5647 else
5648 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5649 } else {
73f73415 5650 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
39279cc3 5651 }
76dda93c
YZ		 5652 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5653
34d19bad 5654 if (!IS_ERR(inode) && root != sub_root) {
c71bf099
YZ		 5655 down_read(&root->fs_info->cleanup_work_sem);
5656 if (!(inode->i_sb->s_flags & MS_RDONLY))
66b4ffd1 5657 ret = btrfs_orphan_cleanup(sub_root);
c71bf099 5658 up_read(&root->fs_info->cleanup_work_sem);
01cd3367
JB		 5659 if (ret) {
5660 iput(inode);
66b4ffd1 5661 inode = ERR_PTR(ret);
01cd3367 5662 }
c71bf099
YZ		 5663 }
5664
3de4586c
CM		 5665 return inode;
5666}
5667
fe15ce44 5668static int btrfs_dentry_delete(const struct dentry *dentry)
76dda93c
YZ		 5669{
5670 struct btrfs_root *root;
2b0143b5 5671 struct inode *inode = d_inode(dentry);
76dda93c 5672
848cce0d 5673 if (!inode && !IS_ROOT(dentry))
2b0143b5 5674 inode = d_inode(dentry->d_parent);
76dda93c 5675
848cce0d
LZ		 5676 if (inode) {
5677 root = BTRFS_I(inode)->root;
efefb143
YZ		 5678 if (btrfs_root_refs(&root->root_item) == 0)
5679 return 1;
848cce0d
LZ		 5680
5681 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5682 return 1;
efefb143 5683 }
76dda93c
YZ		 5684 return 0;
5685}
5686
b4aff1f8
JB		 5687static void btrfs_dentry_release(struct dentry *dentry)
5688{
944a4515 5689 kfree(dentry->d_fsdata);
b4aff1f8
JB		 5690}
5691
3de4586c 5692static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
00cd8dd3 5693 unsigned int flags)
3de4586c 5694{
5662344b 5695 struct inode *inode;
a66e7cc6 5696
5662344b
TI		 5697 inode = btrfs_lookup_dentry(dir, dentry);
5698 if (IS_ERR(inode)) {
5699 if (PTR_ERR(inode) == -ENOENT)
5700 inode = NULL;
5701 else
5702 return ERR_CAST(inode);
5703 }
5704
41d28bca 5705 return d_splice_alias(inode, dentry);
39279cc3
CM		 5706}
5707
16cdcec7 5708unsigned char btrfs_filetype_table[] = {
39279cc3
CM		 5709 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5710};
5711
9cdda8d3 5712static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
39279cc3 5713{
9cdda8d3 5714 struct inode *inode = file_inode(file);
39279cc3
CM		 5715 struct btrfs_root *root = BTRFS_I(inode)->root;
5716 struct btrfs_item *item;
5717 struct btrfs_dir_item *di;
5718 struct btrfs_key key;
5f39d397 5719 struct btrfs_key found_key;
39279cc3 5720 struct btrfs_path *path;
16cdcec7
MX		 5721 struct list_head ins_list;
5722 struct list_head del_list;
39279cc3 5723 int ret;
5f39d397 5724 struct extent_buffer *leaf;
39279cc3 5725 int slot;
39279cc3
CM		 5726 unsigned char d_type;
5727 int over = 0;
5728 u32 di_cur;
5729 u32 di_total;
5730 u32 di_len;
5731 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 5732 char tmp_name[32];
5733 char *name_ptr;
5734 int name_len;
9cdda8d3 5735 int is_curr = 0; /* ctx->pos points to the current index? */
39279cc3
CM		 5736
5737 /* FIXME, use a real flag for deciding about the key type */
5738 if (root->fs_info->tree_root == root)
5739 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 5740
9cdda8d3
AV		 5741 if (!dir_emit_dots(file, ctx))
5742 return 0;
5743
49593bfa 5744 path = btrfs_alloc_path();
16cdcec7
MX		 5745 if (!path)
5746 return -ENOMEM;
ff5714cc 5747
026fd317 5748 path->reada = 1;
49593bfa 5749
16cdcec7
MX		 5750 if (key_type == BTRFS_DIR_INDEX_KEY) {
5751 INIT_LIST_HEAD(&ins_list);
5752 INIT_LIST_HEAD(&del_list);
5753 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5754 }
5755
962a298f 5756 key.type = key_type;
9cdda8d3 5757 key.offset = ctx->pos;
33345d01 5758 key.objectid = btrfs_ino(inode);
5f39d397 5759
39279cc3
CM		 5760 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5761 if (ret < 0)
5762 goto err;
49593bfa
DW		 5763
5764 while (1) {
5f39d397 5765 leaf = path->nodes[0];
39279cc3 5766 slot = path->slots[0];
b9e03af0
LZ		 5767 if (slot >= btrfs_header_nritems(leaf)) {
5768 ret = btrfs_next_leaf(root, path);
5769 if (ret < 0)
5770 goto err;
5771 else if (ret > 0)
5772 break;
5773 continue;
39279cc3 5774 }
3de4586c 5775
dd3cc16b 5776 item = btrfs_item_nr(slot);
5f39d397
CM		 5777 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5778
5779 if (found_key.objectid != key.objectid)
39279cc3 5780 break;
962a298f 5781 if (found_key.type != key_type)
39279cc3 5782 break;
9cdda8d3 5783 if (found_key.offset < ctx->pos)
b9e03af0 5784 goto next;
16cdcec7
MX		 5785 if (key_type == BTRFS_DIR_INDEX_KEY &&
5786 btrfs_should_delete_dir_index(&del_list,
5787 found_key.offset))
5788 goto next;
5f39d397 5789
9cdda8d3 5790 ctx->pos = found_key.offset;
16cdcec7 5791 is_curr = 1;
49593bfa 5792
39279cc3
CM		 5793 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5794 di_cur = 0;
5f39d397 5795 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 5796
5797 while (di_cur < di_total) {
5f39d397
CM		 5798 struct btrfs_key location;
5799
22a94d44
JB		 5800 if (verify_dir_item(root, leaf, di))
5801 break;
5802
5f39d397 5803 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 5804 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 5805 name_ptr = tmp_name;
5806 } else {
5807 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 5808 if (!name_ptr) {
5809 ret = -ENOMEM;
5810 goto err;
5811 }
5f39d397
CM		 5812 }
5813 read_extent_buffer(leaf, name_ptr,
5814 (unsigned long)(di + 1), name_len);
5815
5816 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5817 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c 5818
fede766f 5819
3de4586c 5820 /* is this a reference to our own snapshot? If so
8c9c2bf7
AJ		 5821 * skip it.
5822 *
5823 * In contrast to old kernels, we insert the snapshot's
5824 * dir item and dir index after it has been created, so
5825 * we won't find a reference to our own snapshot. We
5826 * still keep the following code for backward
5827 * compatibility.
3de4586c
CM		 5828 */
5829 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5830 location.objectid == root->root_key.objectid) {
5831 over = 0;
5832 goto skip;
5833 }
9cdda8d3
AV		 5834 over = !dir_emit(ctx, name_ptr, name_len,
5835 location.objectid, d_type);
5f39d397 5836
3de4586c 5837skip:
5f39d397
CM		 5838 if (name_ptr != tmp_name)
5839 kfree(name_ptr);
5840
39279cc3
CM		 5841 if (over)
5842 goto nopos;
5103e947 5843 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 5844 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 5845 di_cur += di_len;
5846 di = (struct btrfs_dir_item *)((char *)di + di_len);
5847 }
b9e03af0
LZ		 5848next:
5849 path->slots[0]++;
39279cc3 5850 }
49593bfa 5851
16cdcec7
MX		 5852 if (key_type == BTRFS_DIR_INDEX_KEY) {
5853 if (is_curr)
9cdda8d3
AV		 5854 ctx->pos++;
5855 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
16cdcec7
MX		 5856 if (ret)
5857 goto nopos;
5858 }
5859
49593bfa 5860 /* Reached end of directory/root. Bump pos past the last item. */
db62efbb
ZB		 5861 ctx->pos++;
5862
5863 /*
5864 * Stop new entries from being returned after we return the last
5865 * entry.
5866 *
5867 * New directory entries are assigned a strictly increasing
5868 * offset. This means that new entries created during readdir
5869 * are *guaranteed* to be seen in the future by that readdir.
5870 * This has broken buggy programs which operate on names as
5871 * they're returned by readdir. Until we re-use freed offsets
5872 * we have this hack to stop new entries from being returned
5873 * under the assumption that they'll never reach this huge
5874 * offset.
5875 *
5876 * This is being careful not to overflow 32bit loff_t unless the
5877 * last entry requires it because doing so has broken 32bit apps
5878 * in the past.
5879 */
5880 if (key_type == BTRFS_DIR_INDEX_KEY) {
5881 if (ctx->pos >= INT_MAX)
5882 ctx->pos = LLONG_MAX;
5883 else
5884 ctx->pos = INT_MAX;
5885 }
39279cc3
CM		 5886nopos:
5887 ret = 0;
5888err:
16cdcec7
MX		 5889 if (key_type == BTRFS_DIR_INDEX_KEY)
5890 btrfs_put_delayed_items(&ins_list, &del_list);
39279cc3 5891 btrfs_free_path(path);
39279cc3
CM		 5892 return ret;
5893}
5894
a9185b41 5895int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
39279cc3
CM		 5896{
5897 struct btrfs_root *root = BTRFS_I(inode)->root;
5898 struct btrfs_trans_handle *trans;
5899 int ret = 0;
0af3d00b 5900 bool nolock = false;
39279cc3 5901
72ac3c0d 5902 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4ca8b41e
CM		 5903 return 0;
5904
83eea1f1 5905 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
82d5902d 5906 nolock = true;
0af3d00b 5907
a9185b41 5908 if (wbc->sync_mode == WB_SYNC_ALL) {
0af3d00b 5909 if (nolock)
7a7eaa40 5910 trans = btrfs_join_transaction_nolock(root);
0af3d00b 5911 else
7a7eaa40 5912 trans = btrfs_join_transaction(root);
3612b495
TI		 5913 if (IS_ERR(trans))
5914 return PTR_ERR(trans);
a698d075 5915 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 5916 }
5917 return ret;
5918}
5919
5920/*
54aa1f4d 5921 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 5922 * inode changes. But, it is most likely to find the inode in cache.
5923 * FIXME, needs more benchmarking...there are no reasons other than performance
5924 * to keep or drop this code.
5925 */
48a3b636 5926static int btrfs_dirty_inode(struct inode *inode)
39279cc3
CM		 5927{
5928 struct btrfs_root *root = BTRFS_I(inode)->root;
5929 struct btrfs_trans_handle *trans;
8929ecfa
YZ		 5930 int ret;
5931
72ac3c0d 5932 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
22c44fe6 5933 return 0;
39279cc3 5934
7a7eaa40 5935 trans = btrfs_join_transaction(root);
22c44fe6
JB		 5936 if (IS_ERR(trans))
5937 return PTR_ERR(trans);
8929ecfa
YZ		 5938
5939 ret = btrfs_update_inode(trans, root, inode);
94b60442
CM		 5940 if (ret && ret == -ENOSPC) {
5941 /* whoops, lets try again with the full transaction */
5942 btrfs_end_transaction(trans, root);
5943 trans = btrfs_start_transaction(root, 1);
22c44fe6
JB		 5944 if (IS_ERR(trans))
5945 return PTR_ERR(trans);
8929ecfa 5946
94b60442 5947 ret = btrfs_update_inode(trans, root, inode);
94b60442 5948 }
39279cc3 5949 btrfs_end_transaction(trans, root);
16cdcec7
MX		 5950 if (BTRFS_I(inode)->delayed_node)
5951 btrfs_balance_delayed_items(root);
22c44fe6
JB		 5952
5953 return ret;
5954}
5955
5956/*
5957 * This is a copy of file_update_time. We need this so we can return error on
5958 * ENOSPC for updating the inode in the case of file write and mmap writes.
5959 */
e41f941a
JB		 5960static int btrfs_update_time(struct inode *inode, struct timespec *now,
5961 int flags)
22c44fe6 5962{
2bc55652
AB		 5963 struct btrfs_root *root = BTRFS_I(inode)->root;
5964
5965 if (btrfs_root_readonly(root))
5966 return -EROFS;
5967
e41f941a 5968 if (flags & S_VERSION)
22c44fe6 5969 inode_inc_iversion(inode);
e41f941a
JB		 5970 if (flags & S_CTIME)
5971 inode->i_ctime = *now;
5972 if (flags & S_MTIME)
5973 inode->i_mtime = *now;
5974 if (flags & S_ATIME)
5975 inode->i_atime = *now;
5976 return btrfs_dirty_inode(inode);
39279cc3
CM		 5977}
5978
d352ac68
CM		 5979/*
5980 * find the highest existing sequence number in a directory
5981 * and then set the in-memory index_cnt variable to reflect
5982 * free sequence numbers
5983 */
aec7477b
JB		 5984static int btrfs_set_inode_index_count(struct inode *inode)
5985{
5986 struct btrfs_root *root = BTRFS_I(inode)->root;
5987 struct btrfs_key key, found_key;
5988 struct btrfs_path *path;
5989 struct extent_buffer *leaf;
5990 int ret;
5991
33345d01 5992 key.objectid = btrfs_ino(inode);
962a298f 5993 key.type = BTRFS_DIR_INDEX_KEY;
aec7477b
JB		 5994 key.offset = (u64)-1;
5995
5996 path = btrfs_alloc_path();
5997 if (!path)
5998 return -ENOMEM;
5999
6000 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6001 if (ret < 0)
6002 goto out;
6003 /* FIXME: we should be able to handle this */
6004 if (ret == 0)
6005 goto out;
6006 ret = 0;
6007
6008 /*
6009 * MAGIC NUMBER EXPLANATION:
6010 * since we search a directory based on f_pos we have to start at 2
6011 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6012 * else has to start at 2
6013 */
6014 if (path->slots[0] == 0) {
6015 BTRFS_I(inode)->index_cnt = 2;
6016 goto out;
6017 }
6018
6019 path->slots[0]--;
6020
6021 leaf = path->nodes[0];
6022 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6023
33345d01 6024 if (found_key.objectid != btrfs_ino(inode) ||
962a298f 6025 found_key.type != BTRFS_DIR_INDEX_KEY) {
aec7477b
JB		 6026 BTRFS_I(inode)->index_cnt = 2;
6027 goto out;
6028 }
6029
6030 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
6031out:
6032 btrfs_free_path(path);
6033 return ret;
6034}
6035
d352ac68
CM		 6036/*
6037 * helper to find a free sequence number in a given directory. This current
6038 * code is very simple, later versions will do smarter things in the btree
6039 */
3de4586c 6040int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 6041{
6042 int ret = 0;
6043
6044 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
16cdcec7
MX		 6045 ret = btrfs_inode_delayed_dir_index_count(dir);
6046 if (ret) {
6047 ret = btrfs_set_inode_index_count(dir);
6048 if (ret)
6049 return ret;
6050 }
aec7477b
JB		 6051 }
6052
00e4e6b3 6053 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 6054 BTRFS_I(dir)->index_cnt++;
6055
6056 return ret;
6057}
6058
b0d5d10f
CM		 6059static int btrfs_insert_inode_locked(struct inode *inode)
6060{
6061 struct btrfs_iget_args args;
6062 args.location = &BTRFS_I(inode)->location;
6063 args.root = BTRFS_I(inode)->root;
6064
6065 return insert_inode_locked4(inode,
6066 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6067 btrfs_find_actor, &args);
6068}
6069
39279cc3
CM		 6070static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6071 struct btrfs_root *root,
aec7477b 6072 struct inode *dir,
9c58309d 6073 const char *name, int name_len,
175a4eb7
AV		 6074 u64 ref_objectid, u64 objectid,
6075 umode_t mode, u64 *index)
39279cc3
CM		 6076{
6077 struct inode *inode;
5f39d397 6078 struct btrfs_inode_item *inode_item;
39279cc3 6079 struct btrfs_key *location;
5f39d397 6080 struct btrfs_path *path;
9c58309d
CM		 6081 struct btrfs_inode_ref *ref;
6082 struct btrfs_key key[2];
6083 u32 sizes[2];
ef3b9af5 6084 int nitems = name ? 2 : 1;
9c58309d 6085 unsigned long ptr;
39279cc3 6086 int ret;
39279cc3 6087
5f39d397 6088 path = btrfs_alloc_path();
d8926bb3
MF		 6089 if (!path)
6090 return ERR_PTR(-ENOMEM);
5f39d397 6091
39279cc3 6092 inode = new_inode(root->fs_info->sb);
8fb27640
YS		 6093 if (!inode) {
6094 btrfs_free_path(path);
39279cc3 6095 return ERR_PTR(-ENOMEM);
8fb27640 6096 }
39279cc3 6097
5762b5c9
FM		 6098 /*
6099 * O_TMPFILE, set link count to 0, so that after this point,
6100 * we fill in an inode item with the correct link count.
6101 */
6102 if (!name)
6103 set_nlink(inode, 0);
6104
581bb050
LZ		 6105 /*
6106 * we have to initialize this early, so we can reclaim the inode
6107 * number if we fail afterwards in this function.
6108 */
6109 inode->i_ino = objectid;
6110
ef3b9af5 6111 if (dir && name) {
1abe9b8a 6112 trace_btrfs_inode_request(dir);
6113
3de4586c 6114 ret = btrfs_set_inode_index(dir, index);
09771430 6115 if (ret) {
8fb27640 6116 btrfs_free_path(path);
09771430 6117 iput(inode);
aec7477b 6118 return ERR_PTR(ret);
09771430 6119 }
ef3b9af5
FM		 6120 } else if (dir) {
6121 *index = 0;
aec7477b
JB		 6122 }
6123 /*
6124 * index_cnt is ignored for everything but a dir,
6125 * btrfs_get_inode_index_count has an explanation for the magic
6126 * number
6127 */
6128 BTRFS_I(inode)->index_cnt = 2;
67de1176 6129 BTRFS_I(inode)->dir_index = *index;
39279cc3 6130 BTRFS_I(inode)->root = root;
e02119d5 6131 BTRFS_I(inode)->generation = trans->transid;
76195853 6132 inode->i_generation = BTRFS_I(inode)->generation;
b888db2b 6133
5dc562c5
JB		 6134 /*
6135 * We could have gotten an inode number from somebody who was fsynced
6136 * and then removed in this same transaction, so let's just set full
6137 * sync since it will be a full sync anyway and this will blow away the
6138 * old info in the log.
6139 */
6140 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6141
9c58309d 6142 key[0].objectid = objectid;
962a298f 6143 key[0].type = BTRFS_INODE_ITEM_KEY;
9c58309d
CM		 6144 key[0].offset = 0;
6145
9c58309d 6146 sizes[0] = sizeof(struct btrfs_inode_item);
ef3b9af5
FM		 6147
6148 if (name) {
6149 /*
6150 * Start new inodes with an inode_ref. This is slightly more
6151 * efficient for small numbers of hard links since they will
6152 * be packed into one item. Extended refs will kick in if we
6153 * add more hard links than can fit in the ref item.
6154 */
6155 key[1].objectid = objectid;
962a298f 6156 key[1].type = BTRFS_INODE_REF_KEY;
ef3b9af5
FM		 6157 key[1].offset = ref_objectid;
6158
6159 sizes[1] = name_len + sizeof(*ref);
6160 }
9c58309d 6161
b0d5d10f
CM		 6162 location = &BTRFS_I(inode)->location;
6163 location->objectid = objectid;
6164 location->offset = 0;
962a298f 6165 location->type = BTRFS_INODE_ITEM_KEY;
b0d5d10f
CM		 6166
6167 ret = btrfs_insert_inode_locked(inode);
6168 if (ret < 0)
6169 goto fail;
6170
b9473439 6171 path->leave_spinning = 1;
ef3b9af5 6172 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
9c58309d 6173 if (ret != 0)
b0d5d10f 6174 goto fail_unlock;
5f39d397 6175
ecc11fab 6176 inode_init_owner(inode, dir, mode);
a76a3cd4 6177 inode_set_bytes(inode, 0);
9cc97d64 6178
6179 inode->i_mtime = CURRENT_TIME;
6180 inode->i_atime = inode->i_mtime;
6181 inode->i_ctime = inode->i_mtime;
6182 BTRFS_I(inode)->i_otime = inode->i_mtime;
6183
5f39d397
CM		 6184 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6185 struct btrfs_inode_item);
293f7e07
LZ		 6186 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6187 sizeof(*inode_item));
e02119d5 6188 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d 6189
ef3b9af5
FM		 6190 if (name) {
6191 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6192 struct btrfs_inode_ref);
6193 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6194 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6195 ptr = (unsigned long)(ref + 1);
6196 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6197 }
9c58309d 6198
5f39d397
CM		 6199 btrfs_mark_buffer_dirty(path->nodes[0]);
6200 btrfs_free_path(path);
6201
6cbff00f
CH		 6202 btrfs_inherit_iflags(inode, dir);
6203
569254b0 6204 if (S_ISREG(mode)) {
94272164
CM		 6205 if (btrfs_test_opt(root, NODATASUM))
6206 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
213490b3 6207 if (btrfs_test_opt(root, NODATACOW))
f2bdf9a8
JB		 6208 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6209 BTRFS_INODE_NODATASUM;
94272164
CM		 6210 }
6211
5d4f98a2 6212 inode_tree_add(inode);
1abe9b8a 6213
6214 trace_btrfs_inode_new(inode);
1973f0fa 6215 btrfs_set_inode_last_trans(trans, inode);
1abe9b8a 6216
8ea05e3a
AB		 6217 btrfs_update_root_times(trans, root);
6218
63541927
FDBM		 6219 ret = btrfs_inode_inherit_props(trans, inode, dir);
6220 if (ret)
6221 btrfs_err(root->fs_info,
6222 "error inheriting props for ino %llu (root %llu): %d",
6223 btrfs_ino(inode), root->root_key.objectid, ret);
6224
39279cc3 6225 return inode;
b0d5d10f
CM		 6226
6227fail_unlock:
6228 unlock_new_inode(inode);
5f39d397 6229fail:
ef3b9af5 6230 if (dir && name)
aec7477b 6231 BTRFS_I(dir)->index_cnt--;
5f39d397 6232 btrfs_free_path(path);
09771430 6233 iput(inode);
5f39d397 6234 return ERR_PTR(ret);
39279cc3
CM		 6235}
6236
6237static inline u8 btrfs_inode_type(struct inode *inode)
6238{
6239 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6240}
6241
d352ac68
CM		 6242/*
6243 * utility function to add 'inode' into 'parent_inode' with
6244 * a give name and a given sequence number.
6245 * if 'add_backref' is true, also insert a backref from the
6246 * inode to the parent directory.
6247 */
e02119d5
CM		 6248int btrfs_add_link(struct btrfs_trans_handle *trans,
6249 struct inode *parent_inode, struct inode *inode,
6250 const char *name, int name_len, int add_backref, u64 index)
39279cc3 6251{
4df27c4d 6252 int ret = 0;
39279cc3 6253 struct btrfs_key key;
e02119d5 6254 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
33345d01
LZ		 6255 u64 ino = btrfs_ino(inode);
6256 u64 parent_ino = btrfs_ino(parent_inode);
5f39d397 6257
33345d01 6258 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 6259 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6260 } else {
33345d01 6261 key.objectid = ino;
962a298f 6262 key.type = BTRFS_INODE_ITEM_KEY;
4df27c4d
YZ		 6263 key.offset = 0;
6264 }
6265
33345d01 6266 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 6267 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6268 key.objectid, root->root_key.objectid,
33345d01 6269 parent_ino, index, name, name_len);
4df27c4d 6270 } else if (add_backref) {
33345d01
LZ		 6271 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6272 parent_ino, index);
4df27c4d 6273 }
39279cc3 6274
79787eaa
JM		 6275 /* Nothing to clean up yet */
6276 if (ret)
6277 return ret;
4df27c4d 6278
79787eaa
JM		 6279 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6280 parent_inode, &key,
6281 btrfs_inode_type(inode), index);
9c52057c 6282 if (ret == -EEXIST || ret == -EOVERFLOW)
79787eaa
JM		 6283 goto fail_dir_item;
6284 else if (ret) {
6285 btrfs_abort_transaction(trans, root, ret);
6286 return ret;
39279cc3 6287 }
79787eaa
JM		 6288
6289 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6290 name_len * 2);
0c4d2d95 6291 inode_inc_iversion(parent_inode);
79787eaa
JM		 6292 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
6293 ret = btrfs_update_inode(trans, root, parent_inode);
6294 if (ret)
6295 btrfs_abort_transaction(trans, root, ret);
39279cc3 6296 return ret;
fe66a05a
CM		 6297
6298fail_dir_item:
6299 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6300 u64 local_index;
6301 int err;
6302 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6303 key.objectid, root->root_key.objectid,
6304 parent_ino, &local_index, name, name_len);
6305
6306 } else if (add_backref) {
6307 u64 local_index;
6308 int err;
6309
6310 err = btrfs_del_inode_ref(trans, root, name, name_len,
6311 ino, parent_ino, &local_index);
6312 }
6313 return ret;
39279cc3
CM		 6314}
6315
6316static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
a1b075d2
JB		 6317 struct inode *dir, struct dentry *dentry,
6318 struct inode *inode, int backref, u64 index)
39279cc3 6319{
a1b075d2
JB		 6320 int err = btrfs_add_link(trans, dir, inode,
6321 dentry->d_name.name, dentry->d_name.len,
6322 backref, index);
39279cc3
CM		 6323 if (err > 0)
6324 err = -EEXIST;
6325 return err;
6326}
6327
618e21d5 6328static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1a67aafb 6329 umode_t mode, dev_t rdev)
618e21d5
JB		 6330{
6331 struct btrfs_trans_handle *trans;
6332 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 6333 struct inode *inode = NULL;
618e21d5
JB		 6334 int err;
6335 int drop_inode = 0;
6336 u64 objectid;
00e4e6b3 6337 u64 index = 0;
618e21d5 6338
9ed74f2d
JB		 6339 /*
6340 * 2 for inode item and ref
6341 * 2 for dir items
6342 * 1 for xattr if selinux is on
6343 */
a22285a6
YZ		 6344 trans = btrfs_start_transaction(root, 5);
6345 if (IS_ERR(trans))
6346 return PTR_ERR(trans);
1832a6d5 6347
581bb050
LZ		 6348 err = btrfs_find_free_ino(root, &objectid);
6349 if (err)
6350 goto out_unlock;
6351
aec7477b 6352 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 6353 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 6354 mode, &index);
7cf96da3
TI		 6355 if (IS_ERR(inode)) {
6356 err = PTR_ERR(inode);
618e21d5 6357 goto out_unlock;
7cf96da3 6358 }
618e21d5 6359
ad19db71
CS		 6360 /*
6361 * If the active LSM wants to access the inode during
6362 * d_instantiate it needs these. Smack checks to see
6363 * if the filesystem supports xattrs by looking at the
6364 * ops vector.
6365 */
ad19db71 6366 inode->i_op = &btrfs_special_inode_operations;
b0d5d10f
CM		 6367 init_special_inode(inode, inode->i_mode, rdev);
6368
6369 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
618e21d5 6370 if (err)
b0d5d10f
CM		 6371 goto out_unlock_inode;
6372
6373 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6374 if (err) {
6375 goto out_unlock_inode;
6376 } else {
1b4ab1bb 6377 btrfs_update_inode(trans, root, inode);
b0d5d10f 6378 unlock_new_inode(inode);
08c422c2 6379 d_instantiate(dentry, inode);
618e21d5 6380 }
b0d5d10f 6381
618e21d5 6382out_unlock:
7ad85bb7 6383 btrfs_end_transaction(trans, root);
c581afc8 6384 btrfs_balance_delayed_items(root);
b53d3f5d 6385 btrfs_btree_balance_dirty(root);
618e21d5
JB		 6386 if (drop_inode) {
6387 inode_dec_link_count(inode);
6388 iput(inode);
6389 }
618e21d5 6390 return err;
b0d5d10f
CM		 6391
6392out_unlock_inode:
6393 drop_inode = 1;
6394 unlock_new_inode(inode);
6395 goto out_unlock;
6396
618e21d5
JB		 6397}
6398
39279cc3 6399static int btrfs_create(struct inode *dir, struct dentry *dentry,
ebfc3b49 6400 umode_t mode, bool excl)
39279cc3
CM		 6401{
6402 struct btrfs_trans_handle *trans;
6403 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 6404 struct inode *inode = NULL;
43baa579 6405 int drop_inode_on_err = 0;
a22285a6 6406 int err;
39279cc3 6407 u64 objectid;
00e4e6b3 6408 u64 index = 0;
39279cc3 6409
9ed74f2d
JB		 6410 /*
6411 * 2 for inode item and ref
6412 * 2 for dir items
6413 * 1 for xattr if selinux is on
6414 */
a22285a6
YZ		 6415 trans = btrfs_start_transaction(root, 5);
6416 if (IS_ERR(trans))
6417 return PTR_ERR(trans);
9ed74f2d 6418
581bb050
LZ		 6419 err = btrfs_find_free_ino(root, &objectid);
6420 if (err)
6421 goto out_unlock;
6422
aec7477b 6423 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 6424 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 6425 mode, &index);
7cf96da3
TI		 6426 if (IS_ERR(inode)) {
6427 err = PTR_ERR(inode);
39279cc3 6428 goto out_unlock;
7cf96da3 6429 }
43baa579 6430 drop_inode_on_err = 1;
ad19db71
CS		 6431 /*
6432 * If the active LSM wants to access the inode during
6433 * d_instantiate it needs these. Smack checks to see
6434 * if the filesystem supports xattrs by looking at the
6435 * ops vector.
6436 */
6437 inode->i_fop = &btrfs_file_operations;
6438 inode->i_op = &btrfs_file_inode_operations;
b0d5d10f 6439 inode->i_mapping->a_ops = &btrfs_aops;
b0d5d10f
CM		 6440
6441 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6442 if (err)
6443 goto out_unlock_inode;
6444
6445 err = btrfs_update_inode(trans, root, inode);
6446 if (err)
6447 goto out_unlock_inode;
ad19db71 6448
a1b075d2 6449 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3 6450 if (err)
b0d5d10f 6451 goto out_unlock_inode;
43baa579 6452
43baa579 6453 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
b0d5d10f 6454 unlock_new_inode(inode);
43baa579
FB		 6455 d_instantiate(dentry, inode);
6456
39279cc3 6457out_unlock:
7ad85bb7 6458 btrfs_end_transaction(trans, root);
43baa579 6459 if (err && drop_inode_on_err) {
39279cc3
CM		 6460 inode_dec_link_count(inode);
6461 iput(inode);
6462 }
c581afc8 6463 btrfs_balance_delayed_items(root);
b53d3f5d 6464 btrfs_btree_balance_dirty(root);
39279cc3 6465 return err;
b0d5d10f
CM		 6466
6467out_unlock_inode:
6468 unlock_new_inode(inode);
6469 goto out_unlock;
6470
39279cc3
CM		 6471}
6472
6473static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6474 struct dentry *dentry)
6475{
6476 struct btrfs_trans_handle *trans;
6477 struct btrfs_root *root = BTRFS_I(dir)->root;
2b0143b5 6478 struct inode *inode = d_inode(old_dentry);
00e4e6b3 6479 u64 index;
39279cc3
CM		 6480 int err;
6481 int drop_inode = 0;
6482
4a8be425
TH		 6483 /* do not allow sys_link's with other subvols of the same device */
6484 if (root->objectid != BTRFS_I(inode)->root->objectid)
3ab3564f 6485 return -EXDEV;
4a8be425 6486
f186373f 6487 if (inode->i_nlink >= BTRFS_LINK_MAX)
c055e99e 6488 return -EMLINK;
4a8be425 6489
3de4586c 6490 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 6491 if (err)
6492 goto fail;
6493
a22285a6 6494 /*
7e6b6465 6495 * 2 items for inode and inode ref
a22285a6 6496 * 2 items for dir items
7e6b6465 6497 * 1 item for parent inode
a22285a6 6498 */
7e6b6465 6499 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 6500 if (IS_ERR(trans)) {
6501 err = PTR_ERR(trans);
6502 goto fail;
6503 }
5f39d397 6504
67de1176
MX		 6505 /* There are several dir indexes for this inode, clear the cache. */
6506 BTRFS_I(inode)->dir_index = 0ULL;
8b558c5f 6507 inc_nlink(inode);
0c4d2d95 6508 inode_inc_iversion(inode);
3153495d 6509 inode->i_ctime = CURRENT_TIME;
7de9c6ee 6510 ihold(inode);
e9976151 6511 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
aec7477b 6512
a1b075d2 6513 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5f39d397 6514
a5719521 6515 if (err) {
54aa1f4d 6516 drop_inode = 1;
a5719521 6517 } else {
10d9f309 6518 struct dentry *parent = dentry->d_parent;
a5719521 6519 err = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 6520 if (err)
6521 goto fail;
ef3b9af5
FM		 6522 if (inode->i_nlink == 1) {
6523 /*
6524 * If new hard link count is 1, it's a file created
6525 * with open(2) O_TMPFILE flag.
6526 */
6527 err = btrfs_orphan_del(trans, inode);
6528 if (err)
6529 goto fail;
6530 }
08c422c2 6531 d_instantiate(dentry, inode);
6a912213 6532 btrfs_log_new_name(trans, inode, NULL, parent);
a5719521 6533 }
39279cc3 6534
7ad85bb7 6535 btrfs_end_transaction(trans, root);
c581afc8 6536 btrfs_balance_delayed_items(root);
1832a6d5 6537fail:
39279cc3
CM		 6538 if (drop_inode) {
6539 inode_dec_link_count(inode);
6540 iput(inode);
6541 }
b53d3f5d 6542 btrfs_btree_balance_dirty(root);
39279cc3
CM		 6543 return err;
6544}
6545
18bb1db3 6546static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
39279cc3 6547{
b9d86667 6548 struct inode *inode = NULL;
39279cc3
CM		 6549 struct btrfs_trans_handle *trans;
6550 struct btrfs_root *root = BTRFS_I(dir)->root;
6551 int err = 0;
6552 int drop_on_err = 0;
b9d86667 6553 u64 objectid = 0;
00e4e6b3 6554 u64 index = 0;
39279cc3 6555
9ed74f2d
JB		 6556 /*
6557 * 2 items for inode and ref
6558 * 2 items for dir items
6559 * 1 for xattr if selinux is on
6560 */
a22285a6
YZ		 6561 trans = btrfs_start_transaction(root, 5);
6562 if (IS_ERR(trans))
6563 return PTR_ERR(trans);
39279cc3 6564
581bb050
LZ		 6565 err = btrfs_find_free_ino(root, &objectid);
6566 if (err)
6567 goto out_fail;
6568
aec7477b 6569 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 6570 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 6571 S_IFDIR | mode, &index);
39279cc3
CM		 6572 if (IS_ERR(inode)) {
6573 err = PTR_ERR(inode);
6574 goto out_fail;
6575 }
5f39d397 6576
39279cc3 6577 drop_on_err = 1;
b0d5d10f
CM		 6578 /* these must be set before we unlock the inode */
6579 inode->i_op = &btrfs_dir_inode_operations;
6580 inode->i_fop = &btrfs_dir_file_operations;
33268eaf 6581
2a7dba39 6582 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf 6583 if (err)
b0d5d10f 6584 goto out_fail_inode;
39279cc3 6585
dbe674a9 6586 btrfs_i_size_write(inode, 0);
39279cc3
CM		 6587 err = btrfs_update_inode(trans, root, inode);
6588 if (err)
b0d5d10f 6589 goto out_fail_inode;
5f39d397 6590
a1b075d2
JB		 6591 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6592 dentry->d_name.len, 0, index);
39279cc3 6593 if (err)
b0d5d10f 6594 goto out_fail_inode;
5f39d397 6595
39279cc3 6596 d_instantiate(dentry, inode);
b0d5d10f
CM		 6597 /*
6598 * mkdir is special. We're unlocking after we call d_instantiate
6599 * to avoid a race with nfsd calling d_instantiate.
6600 */
6601 unlock_new_inode(inode);
39279cc3 6602 drop_on_err = 0;
39279cc3
CM		 6603
6604out_fail:
7ad85bb7 6605 btrfs_end_transaction(trans, root);
c7cfb8a5
WS		 6606 if (drop_on_err) {
6607 inode_dec_link_count(inode);
39279cc3 6608 iput(inode);
c7cfb8a5 6609 }
c581afc8 6610 btrfs_balance_delayed_items(root);
b53d3f5d 6611 btrfs_btree_balance_dirty(root);
39279cc3 6612 return err;
b0d5d10f
CM		 6613
6614out_fail_inode:
6615 unlock_new_inode(inode);
6616 goto out_fail;
39279cc3
CM		 6617}
6618
e6c4efd8
QW		 6619/* Find next extent map of a given extent map, caller needs to ensure locks */
6620static struct extent_map *next_extent_map(struct extent_map *em)
6621{
6622 struct rb_node *next;
6623
6624 next = rb_next(&em->rb_node);
6625 if (!next)
6626 return NULL;
6627 return container_of(next, struct extent_map, rb_node);
6628}
6629
6630static struct extent_map *prev_extent_map(struct extent_map *em)
6631{
6632 struct rb_node *prev;
6633
6634 prev = rb_prev(&em->rb_node);
6635 if (!prev)
6636 return NULL;
6637 return container_of(prev, struct extent_map, rb_node);
6638}
6639
d352ac68 6640/* helper for btfs_get_extent. Given an existing extent in the tree,
e6c4efd8 6641 * the existing extent is the nearest extent to map_start,
d352ac68 6642 * and an extent that you want to insert, deal with overlap and insert
e6c4efd8 6643 * the best fitted new extent into the tree.
d352ac68 6644 */
3b951516
CM		 6645static int merge_extent_mapping(struct extent_map_tree *em_tree,
6646 struct extent_map *existing,
e6dcd2dc 6647 struct extent_map *em,
51f395ad 6648 u64 map_start)
3b951516 6649{
e6c4efd8
QW		 6650 struct extent_map *prev;
6651 struct extent_map *next;
6652 u64 start;
6653 u64 end;
3b951516 6654 u64 start_diff;
3b951516 6655
e6dcd2dc 6656 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
e6c4efd8
QW		 6657
6658 if (existing->start > map_start) {
6659 next = existing;
6660 prev = prev_extent_map(next);
6661 } else {
6662 prev = existing;
6663 next = next_extent_map(prev);
6664 }
6665
6666 start = prev ? extent_map_end(prev) : em->start;
6667 start = max_t(u64, start, em->start);
6668 end = next ? next->start : extent_map_end(em);
6669 end = min_t(u64, end, extent_map_end(em));
6670 start_diff = start - em->start;
6671 em->start = start;
6672 em->len = end - start;
c8b97818
CM		 6673 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6674 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 6675 em->block_start += start_diff;
c8b97818
CM		 6676 em->block_len -= start_diff;
6677 }
09a2a8f9 6678 return add_extent_mapping(em_tree, em, 0);
3b951516
CM		 6679}
6680
c8b97818
CM		 6681static noinline int uncompress_inline(struct btrfs_path *path,
6682 struct inode *inode, struct page *page,
6683 size_t pg_offset, u64 extent_offset,
6684 struct btrfs_file_extent_item *item)
6685{
6686 int ret;
6687 struct extent_buffer *leaf = path->nodes[0];
6688 char *tmp;
6689 size_t max_size;
6690 unsigned long inline_size;
6691 unsigned long ptr;
261507a0 6692 int compress_type;
c8b97818
CM		 6693
6694 WARN_ON(pg_offset != 0);
261507a0 6695 compress_type = btrfs_file_extent_compression(leaf, item);
c8b97818
CM		 6696 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6697 inline_size = btrfs_file_extent_inline_item_len(leaf,
dd3cc16b 6698 btrfs_item_nr(path->slots[0]));
c8b97818 6699 tmp = kmalloc(inline_size, GFP_NOFS);
8d413713
TI		 6700 if (!tmp)
6701 return -ENOMEM;
c8b97818
CM		 6702 ptr = btrfs_file_extent_inline_start(item);
6703
6704 read_extent_buffer(leaf, tmp, ptr, inline_size);
6705
5b050f04 6706 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
261507a0
LZ		 6707 ret = btrfs_decompress(compress_type, tmp, page,
6708 extent_offset, inline_size, max_size);
c8b97818 6709 kfree(tmp);
166ae5a4 6710 return ret;
c8b97818
CM		 6711}
6712
d352ac68
CM		 6713/*
6714 * a bit scary, this does extent mapping from logical file offset to the disk.
d397712b
CM		 6715 * the ugly parts come from merging extents from the disk with the in-ram
6716 * representation. This gets more complex because of the data=ordered code,
d352ac68
CM		 6717 * where the in-ram extents might be locked pending data=ordered completion.
6718 *
6719 * This also copies inline extents directly into the page.
6720 */
d397712b 6721
a52d9a80 6722struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 6723 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 6724 int create)
6725{
6726 int ret;
6727 int err = 0;
a52d9a80
CM		 6728 u64 extent_start = 0;
6729 u64 extent_end = 0;
33345d01 6730 u64 objectid = btrfs_ino(inode);
a52d9a80 6731 u32 found_type;
f421950f 6732 struct btrfs_path *path = NULL;
a52d9a80
CM		 6733 struct btrfs_root *root = BTRFS_I(inode)->root;
6734 struct btrfs_file_extent_item *item;
5f39d397
CM		 6735 struct extent_buffer *leaf;
6736 struct btrfs_key found_key;
a52d9a80
CM		 6737 struct extent_map *em = NULL;
6738 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 6739 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 6740 struct btrfs_trans_handle *trans = NULL;
7ffbb598 6741 const bool new_inline = !page || create;
a52d9a80 6742
a52d9a80 6743again:
890871be 6744 read_lock(&em_tree->lock);
d1310b2e 6745 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 6746 if (em)
6747 em->bdev = root->fs_info->fs_devices->latest_bdev;
890871be 6748 read_unlock(&em_tree->lock);
d1310b2e 6749
a52d9a80 6750 if (em) {
e1c4b745
CM		 6751 if (em->start > start || em->start + em->len <= start)
6752 free_extent_map(em);
6753 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 6754 free_extent_map(em);
6755 else
6756 goto out;
a52d9a80 6757 }
172ddd60 6758 em = alloc_extent_map();
a52d9a80 6759 if (!em) {
d1310b2e
CM		 6760 err = -ENOMEM;
6761 goto out;
a52d9a80 6762 }
e6dcd2dc 6763 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 6764 em->start = EXTENT_MAP_HOLE;
445a6944 6765 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 6766 em->len = (u64)-1;
c8b97818 6767 em->block_len = (u64)-1;
f421950f
CM		 6768
6769 if (!path) {
6770 path = btrfs_alloc_path();
026fd317
JB		 6771 if (!path) {
6772 err = -ENOMEM;
6773 goto out;
6774 }
6775 /*
6776 * Chances are we'll be called again, so go ahead and do
6777 * readahead
6778 */
6779 path->reada = 1;
f421950f
CM		 6780 }
6781
179e29e4
CM		 6782 ret = btrfs_lookup_file_extent(trans, root, path,
6783 objectid, start, trans != NULL);
a52d9a80
CM		 6784 if (ret < 0) {
6785 err = ret;
6786 goto out;
6787 }
6788
6789 if (ret != 0) {
6790 if (path->slots[0] == 0)
6791 goto not_found;
6792 path->slots[0]--;
6793 }
6794
5f39d397
CM		 6795 leaf = path->nodes[0];
6796 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 6797 struct btrfs_file_extent_item);
a52d9a80 6798 /* are we inside the extent that was found? */
5f39d397 6799 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
962a298f 6800 found_type = found_key.type;
5f39d397 6801 if (found_key.objectid != objectid ||
a52d9a80 6802 found_type != BTRFS_EXTENT_DATA_KEY) {
25a50341
JB		 6803 /*
6804 * If we backup past the first extent we want to move forward
6805 * and see if there is an extent in front of us, otherwise we'll
6806 * say there is a hole for our whole search range which can
6807 * cause problems.
6808 */
6809 extent_end = start;
6810 goto next;
a52d9a80
CM		 6811 }
6812
5f39d397
CM		 6813 found_type = btrfs_file_extent_type(leaf, item);
6814 extent_start = found_key.offset;
d899e052
YZ		 6815 if (found_type == BTRFS_FILE_EXTENT_REG ||
6816 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 6817 extent_end = extent_start +
db94535d 6818 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 6819 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6820 size_t size;
514ac8ad 6821 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
fda2832f 6822 extent_end = ALIGN(extent_start + size, root->sectorsize);
9036c102 6823 }
25a50341 6824next:
9036c102
YZ		 6825 if (start >= extent_end) {
6826 path->slots[0]++;
6827 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6828 ret = btrfs_next_leaf(root, path);
6829 if (ret < 0) {
6830 err = ret;
6831 goto out;
a52d9a80 6832 }
9036c102
YZ		 6833 if (ret > 0)
6834 goto not_found;
6835 leaf = path->nodes[0];
a52d9a80 6836 }
9036c102
YZ		 6837 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6838 if (found_key.objectid != objectid ||
6839 found_key.type != BTRFS_EXTENT_DATA_KEY)
6840 goto not_found;
6841 if (start + len <= found_key.offset)
6842 goto not_found;
e2eca69d
WS		 6843 if (start > found_key.offset)
6844 goto next;
9036c102 6845 em->start = start;
70c8a91c 6846 em->orig_start = start;
9036c102
YZ		 6847 em->len = found_key.offset - start;
6848 goto not_found_em;
6849 }
6850
7ffbb598
FM		 6851 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6852
d899e052
YZ		 6853 if (found_type == BTRFS_FILE_EXTENT_REG ||
6854 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80
CM		 6855 goto insert;
6856 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 6857 unsigned long ptr;
a52d9a80 6858 char *map;
3326d1b0
CM		 6859 size_t size;
6860 size_t extent_offset;
6861 size_t copy_size;
a52d9a80 6862
7ffbb598 6863 if (new_inline)
689f9346 6864 goto out;
5f39d397 6865
514ac8ad 6866 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
9036c102 6867 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 6868 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 6869 size - extent_offset);
3326d1b0 6870 em->start = extent_start + extent_offset;
fda2832f 6871 em->len = ALIGN(copy_size, root->sectorsize);
b4939680 6872 em->orig_block_len = em->len;
70c8a91c 6873 em->orig_start = em->start;
689f9346 6874 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 6875 if (create == 0 && !PageUptodate(page)) {
261507a0
LZ		 6876 if (btrfs_file_extent_compression(leaf, item) !=
6877 BTRFS_COMPRESS_NONE) {
c8b97818
CM		 6878 ret = uncompress_inline(path, inode, page,
6879 pg_offset,
6880 extent_offset, item);
166ae5a4
ZB		 6881 if (ret) {
6882 err = ret;
6883 goto out;
6884 }
c8b97818
CM		 6885 } else {
6886 map = kmap(page);
6887 read_extent_buffer(leaf, map + pg_offset, ptr,
6888 copy_size);
93c82d57
CM		 6889 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6890 memset(map + pg_offset + copy_size, 0,
6891 PAGE_CACHE_SIZE - pg_offset -
6892 copy_size);
6893 }
c8b97818
CM		 6894 kunmap(page);
6895 }
179e29e4
CM		 6896 flush_dcache_page(page);
6897 } else if (create && PageUptodate(page)) {
6bf7e080 6898 BUG();
179e29e4
CM		 6899 if (!trans) {
6900 kunmap(page);
6901 free_extent_map(em);
6902 em = NULL;
ff5714cc 6903
b3b4aa74 6904 btrfs_release_path(path);
7a7eaa40 6905 trans = btrfs_join_transaction(root);
ff5714cc 6906
3612b495
TI		 6907 if (IS_ERR(trans))
6908 return ERR_CAST(trans);
179e29e4
CM		 6909 goto again;
6910 }
c8b97818 6911 map = kmap(page);
70dec807 6912 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 6913 copy_size);
c8b97818 6914 kunmap(page);
179e29e4 6915 btrfs_mark_buffer_dirty(leaf);
a52d9a80 6916 }
d1310b2e 6917 set_extent_uptodate(io_tree, em->start,
507903b8 6918 extent_map_end(em) - 1, NULL, GFP_NOFS);
a52d9a80 6919 goto insert;
a52d9a80
CM		 6920 }
6921not_found:
6922 em->start = start;
70c8a91c 6923 em->orig_start = start;
d1310b2e 6924 em->len = len;
a52d9a80 6925not_found_em:
5f39d397 6926 em->block_start = EXTENT_MAP_HOLE;
9036c102 6927 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80 6928insert:
b3b4aa74 6929 btrfs_release_path(path);
d1310b2e 6930 if (em->start > start || extent_map_end(em) <= start) {
c2cf52eb 6931 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
c1c9ff7c 6932 em->start, em->len, start, len);
a52d9a80
CM		 6933 err = -EIO;
6934 goto out;
6935 }
d1310b2e
CM		 6936
6937 err = 0;
890871be 6938 write_lock(&em_tree->lock);
09a2a8f9 6939 ret = add_extent_mapping(em_tree, em, 0);
3b951516
CM		 6940 /* it is possible that someone inserted the extent into the tree
6941 * while we had the lock dropped. It is also possible that
6942 * an overlapping map exists in the tree
6943 */
a52d9a80 6944 if (ret == -EEXIST) {
3b951516 6945 struct extent_map *existing;
e6dcd2dc
CM		 6946
6947 ret = 0;
6948
e6c4efd8
QW		 6949 existing = search_extent_mapping(em_tree, start, len);
6950 /*
6951 * existing will always be non-NULL, since there must be
6952 * extent causing the -EEXIST.
6953 */
6954 if (start >= extent_map_end(existing) ||
32be3a1a 6955 start <= existing->start) {
e6c4efd8
QW		 6956 /*
6957 * The existing extent map is the one nearest to
6958 * the [start, start + len) range which overlaps
6959 */
6960 err = merge_extent_mapping(em_tree, existing,
6961 em, start);
e1c4b745 6962 free_extent_map(existing);
e6c4efd8 6963 if (err) {
3b951516
CM		 6964 free_extent_map(em);
6965 em = NULL;
6966 }
6967 } else {
6968 free_extent_map(em);
6969 em = existing;
e6dcd2dc 6970 err = 0;
a52d9a80 6971 }
a52d9a80 6972 }
890871be 6973 write_unlock(&em_tree->lock);
a52d9a80 6974out:
1abe9b8a 6975
4cd8587c 6976 trace_btrfs_get_extent(root, em);
1abe9b8a 6977
527afb44 6978 btrfs_free_path(path);
a52d9a80
CM		 6979 if (trans) {
6980 ret = btrfs_end_transaction(trans, root);
d397712b 6981 if (!err)
a52d9a80
CM		 6982 err = ret;
6983 }
a52d9a80
CM		 6984 if (err) {
6985 free_extent_map(em);
a52d9a80
CM		 6986 return ERR_PTR(err);
6987 }
79787eaa 6988 BUG_ON(!em); /* Error is always set */
a52d9a80
CM		 6989 return em;
6990}
6991
ec29ed5b
CM		 6992struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6993 size_t pg_offset, u64 start, u64 len,
6994 int create)
6995{
6996 struct extent_map *em;
6997 struct extent_map *hole_em = NULL;
6998 u64 range_start = start;
6999 u64 end;
7000 u64 found;
7001 u64 found_end;
7002 int err = 0;
7003
7004 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7005 if (IS_ERR(em))
7006 return em;
7007 if (em) {
7008 /*
f9e4fb53
LB		 7009 * if our em maps to
7010 * - a hole or
7011 * - a pre-alloc extent,
7012 * there might actually be delalloc bytes behind it.
ec29ed5b 7013 */
f9e4fb53
LB		 7014 if (em->block_start != EXTENT_MAP_HOLE &&
7015 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
ec29ed5b
CM		 7016 return em;
7017 else
7018 hole_em = em;
7019 }
7020
7021 /* check to see if we've wrapped (len == -1 or similar) */
7022 end = start + len;
7023 if (end < start)
7024 end = (u64)-1;
7025 else
7026 end -= 1;
7027
7028 em = NULL;
7029
7030 /* ok, we didn't find anything, lets look for delalloc */
7031 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
7032 end, len, EXTENT_DELALLOC, 1);
7033 found_end = range_start + found;
7034 if (found_end < range_start)
7035 found_end = (u64)-1;
7036
7037 /*
7038 * we didn't find anything useful, return
7039 * the original results from get_extent()
7040 */
7041 if (range_start > end || found_end <= start) {
7042 em = hole_em;
7043 hole_em = NULL;
7044 goto out;
7045 }
7046
7047 /* adjust the range_start to make sure it doesn't
7048 * go backwards from the start they passed in
7049 */
67871254 7050 range_start = max(start, range_start);
ec29ed5b
CM		 7051 found = found_end - range_start;
7052
7053 if (found > 0) {
7054 u64 hole_start = start;
7055 u64 hole_len = len;
7056
172ddd60 7057 em = alloc_extent_map();
ec29ed5b
CM		 7058 if (!em) {
7059 err = -ENOMEM;
7060 goto out;
7061 }
7062 /*
7063 * when btrfs_get_extent can't find anything it
7064 * returns one huge hole
7065 *
7066 * make sure what it found really fits our range, and
7067 * adjust to make sure it is based on the start from
7068 * the caller
7069 */
7070 if (hole_em) {
7071 u64 calc_end = extent_map_end(hole_em);
7072
7073 if (calc_end <= start || (hole_em->start > end)) {
7074 free_extent_map(hole_em);
7075 hole_em = NULL;
7076 } else {
7077 hole_start = max(hole_em->start, start);
7078 hole_len = calc_end - hole_start;
7079 }
7080 }
7081 em->bdev = NULL;
7082 if (hole_em && range_start > hole_start) {
7083 /* our hole starts before our delalloc, so we
7084 * have to return just the parts of the hole
7085 * that go until the delalloc starts
7086 */
7087 em->len = min(hole_len,
7088 range_start - hole_start);
7089 em->start = hole_start;
7090 em->orig_start = hole_start;
7091 /*
7092 * don't adjust block start at all,
7093 * it is fixed at EXTENT_MAP_HOLE
7094 */
7095 em->block_start = hole_em->block_start;
7096 em->block_len = hole_len;
f9e4fb53
LB		 7097 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7098 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
ec29ed5b
CM		 7099 } else {
7100 em->start = range_start;
7101 em->len = found;
7102 em->orig_start = range_start;
7103 em->block_start = EXTENT_MAP_DELALLOC;
7104 em->block_len = found;
7105 }
7106 } else if (hole_em) {
7107 return hole_em;
7108 }
7109out:
7110
7111 free_extent_map(hole_em);
7112 if (err) {
7113 free_extent_map(em);
7114 return ERR_PTR(err);
7115 }
7116 return em;
7117}
7118
4b46fce2
JB		 7119static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7120 u64 start, u64 len)
7121{
7122 struct btrfs_root *root = BTRFS_I(inode)->root;
70c8a91c 7123 struct extent_map *em;
4b46fce2
JB		 7124 struct btrfs_key ins;
7125 u64 alloc_hint;
7126 int ret;
4b46fce2 7127
4b46fce2 7128 alloc_hint = get_extent_allocation_hint(inode, start, len);
00361589 7129 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
e570fd27 7130 alloc_hint, &ins, 1, 1);
00361589
JB		 7131 if (ret)
7132 return ERR_PTR(ret);
4b46fce2 7133
70c8a91c 7134 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
cc95bef6 7135 ins.offset, ins.offset, ins.offset, 0);
00361589 7136 if (IS_ERR(em)) {
e570fd27 7137 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
00361589
JB		 7138 return em;
7139 }
4b46fce2
JB		 7140
7141 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
7142 ins.offset, ins.offset, 0);
7143 if (ret) {
e570fd27 7144 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
00361589
JB		 7145 free_extent_map(em);
7146 return ERR_PTR(ret);
4b46fce2 7147 }
00361589 7148
4b46fce2
JB		 7149 return em;
7150}
7151
46bfbb5c
CM		 7152/*
7153 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7154 * block must be cow'd
7155 */
00361589 7156noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7ee9e440
JB		 7157 u64 *orig_start, u64 *orig_block_len,
7158 u64 *ram_bytes)
46bfbb5c 7159{
00361589 7160 struct btrfs_trans_handle *trans;
46bfbb5c
CM		 7161 struct btrfs_path *path;
7162 int ret;
7163 struct extent_buffer *leaf;
7164 struct btrfs_root *root = BTRFS_I(inode)->root;
7b2b7085 7165 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
46bfbb5c
CM		 7166 struct btrfs_file_extent_item *fi;
7167 struct btrfs_key key;
7168 u64 disk_bytenr;
7169 u64 backref_offset;
7170 u64 extent_end;
7171 u64 num_bytes;
7172 int slot;
7173 int found_type;
7ee9e440 7174 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
e77751aa 7175
46bfbb5c
CM		 7176 path = btrfs_alloc_path();
7177 if (!path)
7178 return -ENOMEM;
7179
00361589 7180 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
46bfbb5c
CM		 7181 offset, 0);
7182 if (ret < 0)
7183 goto out;
7184
7185 slot = path->slots[0];
7186 if (ret == 1) {
7187 if (slot == 0) {
7188 /* can't find the item, must cow */
7189 ret = 0;
7190 goto out;
7191 }
7192 slot--;
7193 }
7194 ret = 0;
7195 leaf = path->nodes[0];
7196 btrfs_item_key_to_cpu(leaf, &key, slot);
33345d01 7197 if (key.objectid != btrfs_ino(inode) ||
46bfbb5c
CM		 7198 key.type != BTRFS_EXTENT_DATA_KEY) {
7199 /* not our file or wrong item type, must cow */
7200 goto out;
7201 }
7202
7203 if (key.offset > offset) {
7204 /* Wrong offset, must cow */
7205 goto out;
7206 }
7207
7208 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7209 found_type = btrfs_file_extent_type(leaf, fi);
7210 if (found_type != BTRFS_FILE_EXTENT_REG &&
7211 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7212 /* not a regular extent, must cow */
7213 goto out;
7214 }
7ee9e440
JB		 7215
7216 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7217 goto out;
7218
e77751aa
MX		 7219 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7220 if (extent_end <= offset)
7221 goto out;
7222
46bfbb5c 7223 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7ee9e440
JB		 7224 if (disk_bytenr == 0)
7225 goto out;
7226
7227 if (btrfs_file_extent_compression(leaf, fi) ||
7228 btrfs_file_extent_encryption(leaf, fi) ||
7229 btrfs_file_extent_other_encoding(leaf, fi))
7230 goto out;
7231
46bfbb5c
CM		 7232 backref_offset = btrfs_file_extent_offset(leaf, fi);
7233
7ee9e440
JB		 7234 if (orig_start) {
7235 *orig_start = key.offset - backref_offset;
7236 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7237 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7238 }
eb384b55 7239
46bfbb5c
CM		 7240 if (btrfs_extent_readonly(root, disk_bytenr))
7241 goto out;
7b2b7085
MX		 7242
7243 num_bytes = min(offset + *len, extent_end) - offset;
7244 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7245 u64 range_end;
7246
7247 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7248 ret = test_range_bit(io_tree, offset, range_end,
7249 EXTENT_DELALLOC, 0, NULL);
7250 if (ret) {
7251 ret = -EAGAIN;
7252 goto out;
7253 }
7254 }
7255
1bda19eb 7256 btrfs_release_path(path);
46bfbb5c
CM		 7257
7258 /*
7259 * look for other files referencing this extent, if we
7260 * find any we must cow
7261 */
00361589
JB		 7262 trans = btrfs_join_transaction(root);
7263 if (IS_ERR(trans)) {
7264 ret = 0;
46bfbb5c 7265 goto out;
00361589
JB		 7266 }
7267
7268 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7269 key.offset - backref_offset, disk_bytenr);
7270 btrfs_end_transaction(trans, root);
7271 if (ret) {
7272 ret = 0;
7273 goto out;
7274 }
46bfbb5c
CM		 7275
7276 /*
7277 * adjust disk_bytenr and num_bytes to cover just the bytes
7278 * in this extent we are about to write. If there
7279 * are any csums in that range we have to cow in order
7280 * to keep the csums correct
7281 */
7282 disk_bytenr += backref_offset;
7283 disk_bytenr += offset - key.offset;
46bfbb5c
CM		 7284 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7285 goto out;
7286 /*
7287 * all of the above have passed, it is safe to overwrite this extent
7288 * without cow
7289 */
eb384b55 7290 *len = num_bytes;
46bfbb5c
CM		 7291 ret = 1;
7292out:
7293 btrfs_free_path(path);
7294 return ret;
7295}
7296
fc4adbff
AG		 7297bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7298{
7299 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7300 int found = false;
7301 void **pagep = NULL;
7302 struct page *page = NULL;
7303 int start_idx;
7304 int end_idx;
7305
7306 start_idx = start >> PAGE_CACHE_SHIFT;
7307
7308 /*
7309 * end is the last byte in the last page. end == start is legal
7310 */
7311 end_idx = end >> PAGE_CACHE_SHIFT;
7312
7313 rcu_read_lock();
7314
7315 /* Most of the code in this while loop is lifted from
7316 * find_get_page. It's been modified to begin searching from a
7317 * page and return just the first page found in that range. If the
7318 * found idx is less than or equal to the end idx then we know that
7319 * a page exists. If no pages are found or if those pages are
7320 * outside of the range then we're fine (yay!) */
7321 while (page == NULL &&
7322 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7323 page = radix_tree_deref_slot(pagep);
7324 if (unlikely(!page))
7325 break;
7326
7327 if (radix_tree_exception(page)) {
809f9016
FM		 7328 if (radix_tree_deref_retry(page)) {
7329 page = NULL;
fc4adbff 7330 continue;
809f9016 7331 }
fc4adbff
AG		 7332 /*
7333 * Otherwise, shmem/tmpfs must be storing a swap entry
7334 * here as an exceptional entry: so return it without
7335 * attempting to raise page count.
7336 */
6fdef6d4 7337 page = NULL;
fc4adbff
AG		 7338 break; /* TODO: Is this relevant for this use case? */
7339 }
7340
91405151
FM		 7341 if (!page_cache_get_speculative(page)) {
7342 page = NULL;
fc4adbff 7343 continue;
91405151 7344 }
fc4adbff
AG		 7345
7346 /*
7347 * Has the page moved?
7348 * This is part of the lockless pagecache protocol. See
7349 * include/linux/pagemap.h for details.
7350 */
7351 if (unlikely(page != *pagep)) {
7352 page_cache_release(page);
7353 page = NULL;
7354 }
7355 }
7356
7357 if (page) {
7358 if (page->index <= end_idx)
7359 found = true;
7360 page_cache_release(page);
7361 }
7362
7363 rcu_read_unlock();
7364 return found;
7365}
7366
eb838e73
JB		 7367static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7368 struct extent_state **cached_state, int writing)
7369{
7370 struct btrfs_ordered_extent *ordered;
7371 int ret = 0;
7372
7373 while (1) {
7374 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7375 0, cached_state);
7376 /*
7377 * We're concerned with the entire range that we're going to be
7378 * doing DIO to, so we need to make sure theres no ordered
7379 * extents in this range.
7380 */
7381 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7382 lockend - lockstart + 1);
7383
7384 /*
7385 * We need to make sure there are no buffered pages in this
7386 * range either, we could have raced between the invalidate in
7387 * generic_file_direct_write and locking the extent. The
7388 * invalidate needs to happen so that reads after a write do not
7389 * get stale data.
7390 */
fc4adbff
AG		 7391 if (!ordered &&
7392 (!writing ||
7393 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
eb838e73
JB		 7394 break;
7395
7396 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7397 cached_state, GFP_NOFS);
7398
7399 if (ordered) {
7400 btrfs_start_ordered_extent(inode, ordered, 1);
7401 btrfs_put_ordered_extent(ordered);
7402 } else {
7403 /* Screw you mmap */
728404da 7404 ret = btrfs_fdatawrite_range(inode, lockstart, lockend);
075bdbdb
FM		 7405 if (ret)
7406 break;
7407 ret = filemap_fdatawait_range(inode->i_mapping,
7408 lockstart,
7409 lockend);
eb838e73
JB		 7410 if (ret)
7411 break;
7412
7413 /*
7414 * If we found a page that couldn't be invalidated just
7415 * fall back to buffered.
7416 */
7417 ret = invalidate_inode_pages2_range(inode->i_mapping,
7418 lockstart >> PAGE_CACHE_SHIFT,
7419 lockend >> PAGE_CACHE_SHIFT);
7420 if (ret)
7421 break;
7422 }
7423
7424 cond_resched();
7425 }
7426
7427 return ret;
7428}
7429
69ffb543
JB		 7430static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7431 u64 len, u64 orig_start,
7432 u64 block_start, u64 block_len,
cc95bef6
JB		 7433 u64 orig_block_len, u64 ram_bytes,
7434 int type)
69ffb543
JB		 7435{
7436 struct extent_map_tree *em_tree;
7437 struct extent_map *em;
7438 struct btrfs_root *root = BTRFS_I(inode)->root;
7439 int ret;
7440
7441 em_tree = &BTRFS_I(inode)->extent_tree;
7442 em = alloc_extent_map();
7443 if (!em)
7444 return ERR_PTR(-ENOMEM);
7445
7446 em->start = start;
7447 em->orig_start = orig_start;
2ab28f32
JB		 7448 em->mod_start = start;
7449 em->mod_len = len;
69ffb543
JB		 7450 em->len = len;
7451 em->block_len = block_len;
7452 em->block_start = block_start;
7453 em->bdev = root->fs_info->fs_devices->latest_bdev;
b4939680 7454 em->orig_block_len = orig_block_len;
cc95bef6 7455 em->ram_bytes = ram_bytes;
70c8a91c 7456 em->generation = -1;
69ffb543
JB		 7457 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7458 if (type == BTRFS_ORDERED_PREALLOC)
b11e234d 7459 set_bit(EXTENT_FLAG_FILLING, &em->flags);
69ffb543
JB		 7460
7461 do {
7462 btrfs_drop_extent_cache(inode, em->start,
7463 em->start + em->len - 1, 0);
7464 write_lock(&em_tree->lock);
09a2a8f9 7465 ret = add_extent_mapping(em_tree, em, 1);
69ffb543
JB		 7466 write_unlock(&em_tree->lock);
7467 } while (ret == -EEXIST);
7468
7469 if (ret) {
7470 free_extent_map(em);
7471 return ERR_PTR(ret);
7472 }
7473
7474 return em;
7475}
7476
50745b0a 7477struct btrfs_dio_data {
7478 u64 outstanding_extents;
7479 u64 reserve;
7480};
69ffb543 7481
9c9464cc
FM		 7482static void adjust_dio_outstanding_extents(struct inode *inode,
7483 struct btrfs_dio_data *dio_data,
7484 const u64 len)
7485{
7486 unsigned num_extents;
7487
7488 num_extents = (unsigned) div64_u64(len + BTRFS_MAX_EXTENT_SIZE - 1,
7489 BTRFS_MAX_EXTENT_SIZE);
7490 /*
7491 * If we have an outstanding_extents count still set then we're
7492 * within our reservation, otherwise we need to adjust our inode
7493 * counter appropriately.
7494 */
7495 if (dio_data->outstanding_extents) {
7496 dio_data->outstanding_extents -= num_extents;
7497 } else {
7498 spin_lock(&BTRFS_I(inode)->lock);
7499 BTRFS_I(inode)->outstanding_extents += num_extents;
7500 spin_unlock(&BTRFS_I(inode)->lock);
7501 }
7502}
7503
4b46fce2
JB		 7504static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7505 struct buffer_head *bh_result, int create)
7506{
7507 struct extent_map *em;
7508 struct btrfs_root *root = BTRFS_I(inode)->root;
eb838e73 7509 struct extent_state *cached_state = NULL;
50745b0a 7510 struct btrfs_dio_data *dio_data = NULL;
4b46fce2 7511 u64 start = iblock << inode->i_blkbits;
eb838e73 7512 u64 lockstart, lockend;
4b46fce2 7513 u64 len = bh_result->b_size;
eb838e73 7514 int unlock_bits = EXTENT_LOCKED;
0934856d 7515 int ret = 0;
eb838e73 7516
172a5049 7517 if (create)
3266789f 7518 unlock_bits |= EXTENT_DIRTY;
172a5049 7519 else
c329861d 7520 len = min_t(u64, len, root->sectorsize);
eb838e73 7521
c329861d
JB		 7522 lockstart = start;
7523 lockend = start + len - 1;
7524
e1cbbfa5
JB		 7525 if (current->journal_info) {
7526 /*
7527 * Need to pull our outstanding extents and set journal_info to NULL so
7528 * that anything that needs to check if there's a transction doesn't get
7529 * confused.
7530 */
50745b0a 7531 dio_data = current->journal_info;
e1cbbfa5
JB		 7532 current->journal_info = NULL;
7533 }
7534
eb838e73
JB		 7535 /*
7536 * If this errors out it's because we couldn't invalidate pagecache for
7537 * this range and we need to fallback to buffered.
7538 */
9c9464cc
FM		 7539 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7540 create)) {
7541 ret = -ENOTBLK;
7542 goto err;
7543 }
eb838e73 7544
4b46fce2 7545 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
eb838e73
JB		 7546 if (IS_ERR(em)) {
7547 ret = PTR_ERR(em);
7548 goto unlock_err;
7549 }
4b46fce2
JB		 7550
7551 /*
7552 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7553 * io. INLINE is special, and we could probably kludge it in here, but
7554 * it's still buffered so for safety lets just fall back to the generic
7555 * buffered path.
7556 *
7557 * For COMPRESSED we _have_ to read the entire extent in so we can
7558 * decompress it, so there will be buffering required no matter what we
7559 * do, so go ahead and fallback to buffered.
7560 *
7561 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7562 * to buffered IO. Don't blame me, this is the price we pay for using
7563 * the generic code.
7564 */
7565 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7566 em->block_start == EXTENT_MAP_INLINE) {
7567 free_extent_map(em);
eb838e73
JB		 7568 ret = -ENOTBLK;
7569 goto unlock_err;
4b46fce2
JB		 7570 }
7571
7572 /* Just a good old fashioned hole, return */
7573 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7574 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7575 free_extent_map(em);
eb838e73 7576 goto unlock_err;
4b46fce2
JB		 7577 }
7578
7579 /*
7580 * We don't allocate a new extent in the following cases
7581 *
7582 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7583 * existing extent.
7584 * 2) The extent is marked as PREALLOC. We're good to go here and can
7585 * just use the extent.
7586 *
7587 */
46bfbb5c 7588 if (!create) {
eb838e73
JB		 7589 len = min(len, em->len - (start - em->start));
7590 lockstart = start + len;
7591 goto unlock;
46bfbb5c 7592 }
4b46fce2
JB		 7593
7594 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7595 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7596 em->block_start != EXTENT_MAP_HOLE)) {
4b46fce2 7597 int type;
eb384b55 7598 u64 block_start, orig_start, orig_block_len, ram_bytes;
4b46fce2
JB		 7599
7600 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7601 type = BTRFS_ORDERED_PREALLOC;
7602 else
7603 type = BTRFS_ORDERED_NOCOW;
46bfbb5c 7604 len = min(len, em->len - (start - em->start));
4b46fce2 7605 block_start = em->block_start + (start - em->start);
46bfbb5c 7606
00361589 7607 if (can_nocow_extent(inode, start, &len, &orig_start,
7ee9e440 7608 &orig_block_len, &ram_bytes) == 1) {
69ffb543
JB		 7609 if (type == BTRFS_ORDERED_PREALLOC) {
7610 free_extent_map(em);
7611 em = create_pinned_em(inode, start, len,
7612 orig_start,
b4939680 7613 block_start, len,
cc95bef6
JB		 7614 orig_block_len,
7615 ram_bytes, type);
555e1286
FM		 7616 if (IS_ERR(em)) {
7617 ret = PTR_ERR(em);
69ffb543 7618 goto unlock_err;
555e1286 7619 }
69ffb543
JB		 7620 }
7621
46bfbb5c
CM		 7622 ret = btrfs_add_ordered_extent_dio(inode, start,
7623 block_start, len, len, type);
46bfbb5c
CM		 7624 if (ret) {
7625 free_extent_map(em);
eb838e73 7626 goto unlock_err;
46bfbb5c
CM		 7627 }
7628 goto unlock;
4b46fce2 7629 }
4b46fce2 7630 }
00361589 7631
46bfbb5c
CM		 7632 /*
7633 * this will cow the extent, reset the len in case we changed
7634 * it above
7635 */
7636 len = bh_result->b_size;
70c8a91c
JB		 7637 free_extent_map(em);
7638 em = btrfs_new_extent_direct(inode, start, len);
eb838e73
JB		 7639 if (IS_ERR(em)) {
7640 ret = PTR_ERR(em);
7641 goto unlock_err;
7642 }
46bfbb5c
CM		 7643 len = min(len, em->len - (start - em->start));
7644unlock:
4b46fce2
JB		 7645 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7646 inode->i_blkbits;
46bfbb5c 7647 bh_result->b_size = len;
4b46fce2
JB		 7648 bh_result->b_bdev = em->bdev;
7649 set_buffer_mapped(bh_result);
c3473e83
JB		 7650 if (create) {
7651 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7652 set_buffer_new(bh_result);
7653
7654 /*
7655 * Need to update the i_size under the extent lock so buffered
7656 * readers will get the updated i_size when we unlock.
7657 */
7658 if (start + len > i_size_read(inode))
7659 i_size_write(inode, start + len);
0934856d 7660
9c9464cc 7661 adjust_dio_outstanding_extents(inode, dio_data, len);
7cf5b976 7662 btrfs_free_reserved_data_space(inode, start, len);
50745b0a 7663 WARN_ON(dio_data->reserve < len);
7664 dio_data->reserve -= len;
7665 current->journal_info = dio_data;
c3473e83 7666 }
4b46fce2 7667
eb838e73
JB		 7668 /*
7669 * In the case of write we need to clear and unlock the entire range,
7670 * in the case of read we need to unlock only the end area that we
7671 * aren't using if there is any left over space.
7672 */
24c03fa5 7673 if (lockstart < lockend) {
0934856d
MX		 7674 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7675 lockend, unlock_bits, 1, 0,
7676 &cached_state, GFP_NOFS);
24c03fa5 7677 } else {
eb838e73 7678 free_extent_state(cached_state);
24c03fa5 7679 }
eb838e73 7680
4b46fce2
JB		 7681 free_extent_map(em);
7682
7683 return 0;
eb838e73
JB		 7684
7685unlock_err:
eb838e73
JB		 7686 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7687 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
9c9464cc 7688err:
50745b0a 7689 if (dio_data)
7690 current->journal_info = dio_data;
9c9464cc
FM		 7691 /*
7692 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7693 * write less data then expected, so that we don't underflow our inode's
7694 * outstanding extents counter.
7695 */
7696 if (create && dio_data)
7697 adjust_dio_outstanding_extents(inode, dio_data, len);
7698
eb838e73 7699 return ret;
4b46fce2
JB		 7700}
7701
8b110e39
MX		 7702static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7703 int rw, int mirror_num)
7704{
7705 struct btrfs_root *root = BTRFS_I(inode)->root;
7706 int ret;
7707
7708 BUG_ON(rw & REQ_WRITE);
7709
7710 bio_get(bio);
7711
7712 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7713 BTRFS_WQ_ENDIO_DIO_REPAIR);
7714 if (ret)
7715 goto err;
7716
7717 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7718err:
7719 bio_put(bio);
7720 return ret;
7721}
7722
7723static int btrfs_check_dio_repairable(struct inode *inode,
7724 struct bio *failed_bio,
7725 struct io_failure_record *failrec,
7726 int failed_mirror)
7727{
7728 int num_copies;
7729
7730 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7731 failrec->logical, failrec->len);
7732 if (num_copies == 1) {
7733 /*
7734 * we only have a single copy of the data, so don't bother with
7735 * all the retry and error correction code that follows. no
7736 * matter what the error is, it is very likely to persist.
7737 */
7738 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7739 num_copies, failrec->this_mirror, failed_mirror);
7740 return 0;
7741 }
7742
7743 failrec->failed_mirror = failed_mirror;
7744 failrec->this_mirror++;
7745 if (failrec->this_mirror == failed_mirror)
7746 failrec->this_mirror++;
7747
7748 if (failrec->this_mirror > num_copies) {
7749 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7750 num_copies, failrec->this_mirror, failed_mirror);
7751 return 0;
7752 }
7753
7754 return 1;
7755}
7756
7757static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7758 struct page *page, u64 start, u64 end,
7759 int failed_mirror, bio_end_io_t *repair_endio,
7760 void *repair_arg)
7761{
7762 struct io_failure_record *failrec;
7763 struct bio *bio;
7764 int isector;
7765 int read_mode;
7766 int ret;
7767
7768 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7769
7770 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7771 if (ret)
7772 return ret;
7773
7774 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7775 failed_mirror);
7776 if (!ret) {
7777 free_io_failure(inode, failrec);
7778 return -EIO;
7779 }
7780
7781 if (failed_bio->bi_vcnt > 1)
7782 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7783 else
7784 read_mode = READ_SYNC;
7785
7786 isector = start - btrfs_io_bio(failed_bio)->logical;
7787 isector >>= inode->i_sb->s_blocksize_bits;
7788 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7789 0, isector, repair_endio, repair_arg);
7790 if (!bio) {
7791 free_io_failure(inode, failrec);
7792 return -EIO;
7793 }
7794
7795 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7796 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7797 read_mode, failrec->this_mirror, failrec->in_validation);
7798
7799 ret = submit_dio_repair_bio(inode, bio, read_mode,
7800 failrec->this_mirror);
7801 if (ret) {
7802 free_io_failure(inode, failrec);
7803 bio_put(bio);
7804 }
7805
7806 return ret;
7807}
7808
7809struct btrfs_retry_complete {
7810 struct completion done;
7811 struct inode *inode;
7812 u64 start;
7813 int uptodate;
7814};
7815
4246a0b6 7816static void btrfs_retry_endio_nocsum(struct bio *bio)
8b110e39
MX		 7817{
7818 struct btrfs_retry_complete *done = bio->bi_private;
7819 struct bio_vec *bvec;
7820 int i;
7821
4246a0b6 7822 if (bio->bi_error)
8b110e39
MX		 7823 goto end;
7824
7825 done->uptodate = 1;
7826 bio_for_each_segment_all(bvec, bio, i)
7827 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7828end:
7829 complete(&done->done);
7830 bio_put(bio);
7831}
7832
7833static int __btrfs_correct_data_nocsum(struct inode *inode,
7834 struct btrfs_io_bio *io_bio)
4b46fce2 7835{
2c30c71b 7836 struct bio_vec *bvec;
8b110e39 7837 struct btrfs_retry_complete done;
4b46fce2 7838 u64 start;
2c30c71b 7839 int i;
c1dc0896 7840 int ret;
4b46fce2 7841
8b110e39
MX		 7842 start = io_bio->logical;
7843 done.inode = inode;
7844
7845 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7846try_again:
7847 done.uptodate = 0;
7848 done.start = start;
7849 init_completion(&done.done);
7850
7851 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7852 start + bvec->bv_len - 1,
7853 io_bio->mirror_num,
7854 btrfs_retry_endio_nocsum, &done);
7855 if (ret)
7856 return ret;
7857
7858 wait_for_completion(&done.done);
7859
7860 if (!done.uptodate) {
7861 /* We might have another mirror, so try again */
7862 goto try_again;
7863 }
7864
7865 start += bvec->bv_len;
7866 }
7867
7868 return 0;
7869}
7870
4246a0b6 7871static void btrfs_retry_endio(struct bio *bio)
8b110e39
MX		 7872{
7873 struct btrfs_retry_complete *done = bio->bi_private;
7874 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7875 struct bio_vec *bvec;
7876 int uptodate;
7877 int ret;
7878 int i;
7879
4246a0b6 7880 if (bio->bi_error)
8b110e39
MX		 7881 goto end;
7882
7883 uptodate = 1;
7884 bio_for_each_segment_all(bvec, bio, i) {
7885 ret = __readpage_endio_check(done->inode, io_bio, i,
7886 bvec->bv_page, 0,
7887 done->start, bvec->bv_len);
7888 if (!ret)
7889 clean_io_failure(done->inode, done->start,
7890 bvec->bv_page, 0);
7891 else
7892 uptodate = 0;
7893 }
7894
7895 done->uptodate = uptodate;
7896end:
7897 complete(&done->done);
7898 bio_put(bio);
7899}
7900
7901static int __btrfs_subio_endio_read(struct inode *inode,
7902 struct btrfs_io_bio *io_bio, int err)
7903{
7904 struct bio_vec *bvec;
7905 struct btrfs_retry_complete done;
7906 u64 start;
7907 u64 offset = 0;
7908 int i;
7909 int ret;
dc380aea 7910
8b110e39 7911 err = 0;
c1dc0896 7912 start = io_bio->logical;
8b110e39
MX		 7913 done.inode = inode;
7914
c1dc0896 7915 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
dc380aea
MX		 7916 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7917 0, start, bvec->bv_len);
8b110e39
MX		 7918 if (likely(!ret))
7919 goto next;
7920try_again:
7921 done.uptodate = 0;
7922 done.start = start;
7923 init_completion(&done.done);
7924
7925 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7926 start + bvec->bv_len - 1,
7927 io_bio->mirror_num,
7928 btrfs_retry_endio, &done);
7929 if (ret) {
7930 err = ret;
7931 goto next;
7932 }
7933
7934 wait_for_completion(&done.done);
7935
7936 if (!done.uptodate) {
7937 /* We might have another mirror, so try again */
7938 goto try_again;
7939 }
7940next:
7941 offset += bvec->bv_len;
4b46fce2 7942 start += bvec->bv_len;
2c30c71b 7943 }
c1dc0896
MX		 7944
7945 return err;
7946}
7947
8b110e39
MX		 7948static int btrfs_subio_endio_read(struct inode *inode,
7949 struct btrfs_io_bio *io_bio, int err)
7950{
7951 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7952
7953 if (skip_csum) {
7954 if (unlikely(err))
7955 return __btrfs_correct_data_nocsum(inode, io_bio);
7956 else
7957 return 0;
7958 } else {
7959 return __btrfs_subio_endio_read(inode, io_bio, err);
7960 }
7961}
7962
4246a0b6 7963static void btrfs_endio_direct_read(struct bio *bio)
c1dc0896
MX		 7964{
7965 struct btrfs_dio_private *dip = bio->bi_private;
7966 struct inode *inode = dip->inode;
7967 struct bio *dio_bio;
7968 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
4246a0b6 7969 int err = bio->bi_error;
c1dc0896 7970
8b110e39
MX		 7971 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
7972 err = btrfs_subio_endio_read(inode, io_bio, err);
c1dc0896 7973
4b46fce2 7974 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
d0082371 7975 dip->logical_offset + dip->bytes - 1);
9be3395b 7976 dio_bio = dip->dio_bio;
4b46fce2 7977
4b46fce2 7978 kfree(dip);
c0da7aa1 7979
4246a0b6 7980 dio_end_io(dio_bio, bio->bi_error);
23ea8e5a
MX		 7981
7982 if (io_bio->end_io)
7983 io_bio->end_io(io_bio, err);
9be3395b 7984 bio_put(bio);
4b46fce2
JB		 7985}
7986
4246a0b6 7987static void btrfs_endio_direct_write(struct bio *bio)
4b46fce2
JB		 7988{
7989 struct btrfs_dio_private *dip = bio->bi_private;
7990 struct inode *inode = dip->inode;
7991 struct btrfs_root *root = BTRFS_I(inode)->root;
4b46fce2 7992 struct btrfs_ordered_extent *ordered = NULL;
163cf09c
CM		 7993 u64 ordered_offset = dip->logical_offset;
7994 u64 ordered_bytes = dip->bytes;
9be3395b 7995 struct bio *dio_bio;
4b46fce2
JB		 7996 int ret;
7997
163cf09c
CM		 7998again:
7999 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8000 &ordered_offset,
4246a0b6
CH		 8001 ordered_bytes,
8002 !bio->bi_error);
4b46fce2 8003 if (!ret)
163cf09c 8004 goto out_test;
4b46fce2 8005
9e0af237
LB		 8006 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8007 finish_ordered_fn, NULL, NULL);
fccb5d86
QW		 8008 btrfs_queue_work(root->fs_info->endio_write_workers,
8009 &ordered->work);
163cf09c
CM		 8010out_test:
8011 /*
8012 * our bio might span multiple ordered extents. If we haven't
8013 * completed the accounting for the whole dio, go back and try again
8014 */
8015 if (ordered_offset < dip->logical_offset + dip->bytes) {
8016 ordered_bytes = dip->logical_offset + dip->bytes -
8017 ordered_offset;
5fd02043 8018 ordered = NULL;
163cf09c
CM		 8019 goto again;
8020 }
9be3395b 8021 dio_bio = dip->dio_bio;
4b46fce2 8022
4b46fce2 8023 kfree(dip);
c0da7aa1 8024
4246a0b6 8025 dio_end_io(dio_bio, bio->bi_error);
9be3395b 8026 bio_put(bio);
4b46fce2
JB		 8027}
8028
eaf25d93
CM		 8029static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
8030 struct bio *bio, int mirror_num,
8031 unsigned long bio_flags, u64 offset)
8032{
8033 int ret;
8034 struct btrfs_root *root = BTRFS_I(inode)->root;
8035 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
79787eaa 8036 BUG_ON(ret); /* -ENOMEM */
eaf25d93
CM		 8037 return 0;
8038}
8039
4246a0b6 8040static void btrfs_end_dio_bio(struct bio *bio)
e65e1535
MX		 8041{
8042 struct btrfs_dio_private *dip = bio->bi_private;
4246a0b6 8043 int err = bio->bi_error;
e65e1535 8044
8b110e39
MX		 8045 if (err)
8046 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8047 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
8048 btrfs_ino(dip->inode), bio->bi_rw,
8049 (unsigned long long)bio->bi_iter.bi_sector,
8050 bio->bi_iter.bi_size, err);
8051
8052 if (dip->subio_endio)
8053 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
c1dc0896
MX		 8054
8055 if (err) {
e65e1535
MX		 8056 dip->errors = 1;
8057
8058 /*
8059 * before atomic variable goto zero, we must make sure
8060 * dip->errors is perceived to be set.
8061 */
4e857c58 8062 smp_mb__before_atomic();
e65e1535
MX		 8063 }
8064
8065 /* if there are more bios still pending for this dio, just exit */
8066 if (!atomic_dec_and_test(&dip->pending_bios))
8067 goto out;
8068
9be3395b 8069 if (dip->errors) {
e65e1535 8070 bio_io_error(dip->orig_bio);
9be3395b 8071 } else {
4246a0b6
CH		 8072 dip->dio_bio->bi_error = 0;
8073 bio_endio(dip->orig_bio);
e65e1535
MX		 8074 }
8075out:
8076 bio_put(bio);
8077}
8078
8079static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8080 u64 first_sector, gfp_t gfp_flags)
8081{
da2f0f74 8082 struct bio *bio;
22365979 8083 bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
da2f0f74
CM		 8084 if (bio)
8085 bio_associate_current(bio);
8086 return bio;
e65e1535
MX		 8087}
8088
c1dc0896
MX		 8089static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
8090 struct inode *inode,
8091 struct btrfs_dio_private *dip,
8092 struct bio *bio,
8093 u64 file_offset)
8094{
8095 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8096 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8097 int ret;
8098
8099 /*
8100 * We load all the csum data we need when we submit
8101 * the first bio to reduce the csum tree search and
8102 * contention.
8103 */
8104 if (dip->logical_offset == file_offset) {
8105 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
8106 file_offset);
8107 if (ret)
8108 return ret;
8109 }
8110
8111 if (bio == dip->orig_bio)
8112 return 0;
8113
8114 file_offset -= dip->logical_offset;
8115 file_offset >>= inode->i_sb->s_blocksize_bits;
8116 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8117
8118 return 0;
8119}
8120
e65e1535
MX		 8121static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8122 int rw, u64 file_offset, int skip_sum,
c329861d 8123 int async_submit)
e65e1535 8124{
facc8a22 8125 struct btrfs_dio_private *dip = bio->bi_private;
e65e1535
MX		 8126 int write = rw & REQ_WRITE;
8127 struct btrfs_root *root = BTRFS_I(inode)->root;
8128 int ret;
8129
b812ce28
JB		 8130 if (async_submit)
8131 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8132
e65e1535 8133 bio_get(bio);
5fd02043
JB		 8134
8135 if (!write) {
bfebd8b5
DS		 8136 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8137 BTRFS_WQ_ENDIO_DATA);
5fd02043
JB		 8138 if (ret)
8139 goto err;
8140 }
e65e1535 8141
1ae39938
JB		 8142 if (skip_sum)
8143 goto map;
8144
8145 if (write && async_submit) {
e65e1535
MX		 8146 ret = btrfs_wq_submit_bio(root->fs_info,
8147 inode, rw, bio, 0, 0,
8148 file_offset,
8149 __btrfs_submit_bio_start_direct_io,
8150 __btrfs_submit_bio_done);
8151 goto err;
1ae39938
JB		 8152 } else if (write) {
8153 /*
8154 * If we aren't doing async submit, calculate the csum of the
8155 * bio now.
8156 */
8157 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8158 if (ret)
8159 goto err;
23ea8e5a 8160 } else {
c1dc0896
MX		 8161 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8162 file_offset);
c2db1073
TI		 8163 if (ret)
8164 goto err;
8165 }
1ae39938
JB		 8166map:
8167 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
e65e1535
MX		 8168err:
8169 bio_put(bio);
8170 return ret;
8171}
8172
8173static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
8174 int skip_sum)
8175{
8176 struct inode *inode = dip->inode;
8177 struct btrfs_root *root = BTRFS_I(inode)->root;
e65e1535
MX		 8178 struct bio *bio;
8179 struct bio *orig_bio = dip->orig_bio;
8180 struct bio_vec *bvec = orig_bio->bi_io_vec;
4f024f37 8181 u64 start_sector = orig_bio->bi_iter.bi_sector;
e65e1535
MX		 8182 u64 file_offset = dip->logical_offset;
8183 u64 submit_len = 0;
8184 u64 map_length;
8185 int nr_pages = 0;
23ea8e5a 8186 int ret;
1ae39938 8187 int async_submit = 0;
e65e1535 8188
4f024f37 8189 map_length = orig_bio->bi_iter.bi_size;
53b381b3 8190 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
e65e1535 8191 &map_length, NULL, 0);
7a5c3c9b 8192 if (ret)
e65e1535 8193 return -EIO;
facc8a22 8194
4f024f37 8195 if (map_length >= orig_bio->bi_iter.bi_size) {
02f57c7a 8196 bio = orig_bio;
c1dc0896 8197 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
02f57c7a
JB		 8198 goto submit;
8199 }
8200
53b381b3 8201 /* async crcs make it difficult to collect full stripe writes. */
ffe2d203 8202 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
53b381b3
DW		 8203 async_submit = 0;
8204 else
8205 async_submit = 1;
8206
02f57c7a
JB		 8207 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8208 if (!bio)
8209 return -ENOMEM;
7a5c3c9b 8210
02f57c7a
JB		 8211 bio->bi_private = dip;
8212 bio->bi_end_io = btrfs_end_dio_bio;
c1dc0896 8213 btrfs_io_bio(bio)->logical = file_offset;
02f57c7a
JB		 8214 atomic_inc(&dip->pending_bios);
8215
e65e1535 8216 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
ee39b432 8217 if (map_length < submit_len + bvec->bv_len ||
e65e1535 8218 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
ee39b432 8219 bvec->bv_offset) < bvec->bv_len) {
e65e1535
MX		 8220 /*
8221 * inc the count before we submit the bio so
8222 * we know the end IO handler won't happen before
8223 * we inc the count. Otherwise, the dip might get freed
8224 * before we're done setting it up
8225 */
8226 atomic_inc(&dip->pending_bios);
8227 ret = __btrfs_submit_dio_bio(bio, inode, rw,
8228 file_offset, skip_sum,
c329861d 8229 async_submit);
e65e1535
MX		 8230 if (ret) {
8231 bio_put(bio);
8232 atomic_dec(&dip->pending_bios);
8233 goto out_err;
8234 }
8235
e65e1535
MX		 8236 start_sector += submit_len >> 9;
8237 file_offset += submit_len;
8238
8239 submit_len = 0;
8240 nr_pages = 0;
8241
8242 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8243 start_sector, GFP_NOFS);
8244 if (!bio)
8245 goto out_err;
8246 bio->bi_private = dip;
8247 bio->bi_end_io = btrfs_end_dio_bio;
c1dc0896 8248 btrfs_io_bio(bio)->logical = file_offset;
e65e1535 8249
4f024f37 8250 map_length = orig_bio->bi_iter.bi_size;
53b381b3 8251 ret = btrfs_map_block(root->fs_info, rw,
3ec706c8 8252 start_sector << 9,
e65e1535
MX		 8253 &map_length, NULL, 0);
8254 if (ret) {
8255 bio_put(bio);
8256 goto out_err;
8257 }
8258 } else {
8259 submit_len += bvec->bv_len;
67871254 8260 nr_pages++;
e65e1535
MX		 8261 bvec++;
8262 }
8263 }
8264
02f57c7a 8265submit:
e65e1535 8266 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
c329861d 8267 async_submit);
e65e1535
MX		 8268 if (!ret)
8269 return 0;
8270
8271 bio_put(bio);
8272out_err:
8273 dip->errors = 1;
8274 /*
8275 * before atomic variable goto zero, we must
8276 * make sure dip->errors is perceived to be set.
8277 */
4e857c58 8278 smp_mb__before_atomic();
e65e1535
MX		 8279 if (atomic_dec_and_test(&dip->pending_bios))
8280 bio_io_error(dip->orig_bio);
8281
8282 /* bio_end_io() will handle error, so we needn't return it */
8283 return 0;
8284}
8285
9be3395b
CM		 8286static void btrfs_submit_direct(int rw, struct bio *dio_bio,
8287 struct inode *inode, loff_t file_offset)
4b46fce2 8288{
61de718f
FM		 8289 struct btrfs_dio_private *dip = NULL;
8290 struct bio *io_bio = NULL;
23ea8e5a 8291 struct btrfs_io_bio *btrfs_bio;
4b46fce2 8292 int skip_sum;
7b6d91da 8293 int write = rw & REQ_WRITE;
4b46fce2
JB		 8294 int ret = 0;
8295
8296 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8297
9be3395b 8298 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
9be3395b
CM		 8299 if (!io_bio) {
8300 ret = -ENOMEM;
8301 goto free_ordered;
8302 }
8303
c1dc0896 8304 dip = kzalloc(sizeof(*dip), GFP_NOFS);
4b46fce2
JB		 8305 if (!dip) {
8306 ret = -ENOMEM;
61de718f 8307 goto free_ordered;
4b46fce2 8308 }
4b46fce2 8309
9be3395b 8310 dip->private = dio_bio->bi_private;
4b46fce2
JB		 8311 dip->inode = inode;
8312 dip->logical_offset = file_offset;
4f024f37
KO		 8313 dip->bytes = dio_bio->bi_iter.bi_size;
8314 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
9be3395b 8315 io_bio->bi_private = dip;
9be3395b
CM		 8316 dip->orig_bio = io_bio;
8317 dip->dio_bio = dio_bio;
e65e1535 8318 atomic_set(&dip->pending_bios, 0);
c1dc0896
MX		 8319 btrfs_bio = btrfs_io_bio(io_bio);
8320 btrfs_bio->logical = file_offset;
4b46fce2 8321
c1dc0896 8322 if (write) {
9be3395b 8323 io_bio->bi_end_io = btrfs_endio_direct_write;
c1dc0896 8324 } else {
9be3395b 8325 io_bio->bi_end_io = btrfs_endio_direct_read;
c1dc0896
MX		 8326 dip->subio_endio = btrfs_subio_endio_read;
8327 }
4b46fce2 8328
e65e1535
MX		 8329 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8330 if (!ret)
eaf25d93 8331 return;
9be3395b 8332
23ea8e5a
MX		 8333 if (btrfs_bio->end_io)
8334 btrfs_bio->end_io(btrfs_bio, ret);
9be3395b 8335
4b46fce2
JB		 8336free_ordered:
8337 /*
61de718f
FM		 8338 * If we arrived here it means either we failed to submit the dip
8339 * or we either failed to clone the dio_bio or failed to allocate the
8340 * dip. If we cloned the dio_bio and allocated the dip, we can just
8341 * call bio_endio against our io_bio so that we get proper resource
8342 * cleanup if we fail to submit the dip, otherwise, we must do the
8343 * same as btrfs_endio_direct_[write|read] because we can't call these
8344 * callbacks - they require an allocated dip and a clone of dio_bio.
4b46fce2 8345 */
61de718f 8346 if (io_bio && dip) {
4246a0b6
CH		 8347 io_bio->bi_error = -EIO;
8348 bio_endio(io_bio);
61de718f
FM		 8349 /*
8350 * The end io callbacks free our dip, do the final put on io_bio
8351 * and all the cleanup and final put for dio_bio (through
8352 * dio_end_io()).
8353 */
8354 dip = NULL;
8355 io_bio = NULL;
8356 } else {
8357 if (write) {
8358 struct btrfs_ordered_extent *ordered;
8359
8360 ordered = btrfs_lookup_ordered_extent(inode,
8361 file_offset);
8362 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
8363 /*
8364 * Decrements our ref on the ordered extent and removes
8365 * the ordered extent from the inode's ordered tree,
8366 * doing all the proper resource cleanup such as for the
8367 * reserved space and waking up any waiters for this
8368 * ordered extent (through btrfs_remove_ordered_extent).
8369 */
8370 btrfs_finish_ordered_io(ordered);
8371 } else {
8372 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8373 file_offset + dio_bio->bi_iter.bi_size - 1);
8374 }
4246a0b6 8375 dio_bio->bi_error = -EIO;
61de718f
FM		 8376 /*
8377 * Releases and cleans up our dio_bio, no need to bio_put()
8378 * nor bio_endio()/bio_io_error() against dio_bio.
8379 */
8380 dio_end_io(dio_bio, ret);
4b46fce2 8381 }
61de718f
FM		 8382 if (io_bio)
8383 bio_put(io_bio);
8384 kfree(dip);
4b46fce2
JB		 8385}
8386
6f673763 8387static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
28060d5d 8388 const struct iov_iter *iter, loff_t offset)
5a5f79b5
CM		 8389{
8390 int seg;
a1b75f7d 8391 int i;
5a5f79b5
CM		 8392 unsigned blocksize_mask = root->sectorsize - 1;
8393 ssize_t retval = -EINVAL;
5a5f79b5
CM		 8394
8395 if (offset & blocksize_mask)
8396 goto out;
8397
28060d5d
AV		 8398 if (iov_iter_alignment(iter) & blocksize_mask)
8399 goto out;
a1b75f7d 8400
28060d5d 8401 /* If this is a write we don't need to check anymore */
6f673763 8402 if (iov_iter_rw(iter) == WRITE)
28060d5d
AV		 8403 return 0;
8404 /*
8405 * Check to make sure we don't have duplicate iov_base's in this
8406 * iovec, if so return EINVAL, otherwise we'll get csum errors
8407 * when reading back.
8408 */
8409 for (seg = 0; seg < iter->nr_segs; seg++) {
8410 for (i = seg + 1; i < iter->nr_segs; i++) {
8411 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
a1b75f7d
JB		 8412 goto out;
8413 }
5a5f79b5
CM		 8414 }
8415 retval = 0;
8416out:
8417 return retval;
8418}
eb838e73 8419
22c6186e
OS		 8420static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
8421 loff_t offset)
16432985 8422{
4b46fce2
JB		 8423 struct file *file = iocb->ki_filp;
8424 struct inode *inode = file->f_mapping->host;
50745b0a 8425 struct btrfs_root *root = BTRFS_I(inode)->root;
8426 struct btrfs_dio_data dio_data = { 0 };
0934856d 8427 size_t count = 0;
2e60a51e 8428 int flags = 0;
38851cc1
MX		 8429 bool wakeup = true;
8430 bool relock = false;
0934856d 8431 ssize_t ret;
4b46fce2 8432
6f673763 8433 if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
5a5f79b5 8434 return 0;
3f7c579c 8435
fe0f07d0 8436 inode_dio_begin(inode);
4e857c58 8437 smp_mb__after_atomic();
38851cc1 8438
0e267c44 8439 /*
41bd9ca4
MX		 8440 * The generic stuff only does filemap_write_and_wait_range, which
8441 * isn't enough if we've written compressed pages to this area, so
8442 * we need to flush the dirty pages again to make absolutely sure
8443 * that any outstanding dirty pages are on disk.
0e267c44 8444 */
a6cbcd4a 8445 count = iov_iter_count(iter);
41bd9ca4
MX		 8446 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8447 &BTRFS_I(inode)->runtime_flags))
9a025a08
WS		 8448 filemap_fdatawrite_range(inode->i_mapping, offset,
8449 offset + count - 1);
0e267c44 8450
6f673763 8451 if (iov_iter_rw(iter) == WRITE) {
38851cc1
MX		 8452 /*
8453 * If the write DIO is beyond the EOF, we need update
8454 * the isize, but it is protected by i_mutex. So we can
8455 * not unlock the i_mutex at this case.
8456 */
8457 if (offset + count <= inode->i_size) {
8458 mutex_unlock(&inode->i_mutex);
8459 relock = true;
8460 }
7cf5b976 8461 ret = btrfs_delalloc_reserve_space(inode, offset, count);
0934856d 8462 if (ret)
38851cc1 8463 goto out;
50745b0a 8464 dio_data.outstanding_extents = div64_u64(count +
e1cbbfa5
JB		 8465 BTRFS_MAX_EXTENT_SIZE - 1,
8466 BTRFS_MAX_EXTENT_SIZE);
8467
8468 /*
8469 * We need to know how many extents we reserved so that we can
8470 * do the accounting properly if we go over the number we
8471 * originally calculated. Abuse current->journal_info for this.
8472 */
50745b0a 8473 dio_data.reserve = round_up(count, root->sectorsize);
8474 current->journal_info = &dio_data;
ee39b432
DS		 8475 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8476 &BTRFS_I(inode)->runtime_flags)) {
fe0f07d0 8477 inode_dio_end(inode);
38851cc1
MX		 8478 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8479 wakeup = false;
0934856d
MX		 8480 }
8481
17f8c842
OS		 8482 ret = __blockdev_direct_IO(iocb, inode,
8483 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8484 iter, offset, btrfs_get_blocks_direct, NULL,
8485 btrfs_submit_direct, flags);
6f673763 8486 if (iov_iter_rw(iter) == WRITE) {
e1cbbfa5 8487 current->journal_info = NULL;
ddba1bfc 8488 if (ret < 0 && ret != -EIOCBQUEUED) {
50745b0a 8489 if (dio_data.reserve)
7cf5b976
QW		 8490 btrfs_delalloc_release_space(inode, offset,
8491 dio_data.reserve);
ddba1bfc 8492 } else if (ret >= 0 && (size_t)ret < count)
7cf5b976
QW		 8493 btrfs_delalloc_release_space(inode, offset,
8494 count - (size_t)ret);
0934856d 8495 }
38851cc1 8496out:
2e60a51e 8497 if (wakeup)
fe0f07d0 8498 inode_dio_end(inode);
38851cc1
MX		 8499 if (relock)
8500 mutex_lock(&inode->i_mutex);
0934856d
MX		 8501
8502 return ret;
16432985
CM		 8503}
8504
05dadc09
TI		 8505#define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8506
1506fcc8
YS		 8507static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8508 __u64 start, __u64 len)
8509{
05dadc09
TI		 8510 int ret;
8511
8512 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8513 if (ret)
8514 return ret;
8515
ec29ed5b 8516 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
1506fcc8
YS		 8517}
8518
a52d9a80 8519int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 8520{
d1310b2e
CM		 8521 struct extent_io_tree *tree;
8522 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 8523 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
9ebefb18 8524}
1832a6d5 8525
a52d9a80 8526static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 8527{
d1310b2e 8528 struct extent_io_tree *tree;
be7bd730
JB		 8529 struct inode *inode = page->mapping->host;
8530 int ret;
b888db2b
CM		 8531
8532 if (current->flags & PF_MEMALLOC) {
8533 redirty_page_for_writepage(wbc, page);
8534 unlock_page(page);
8535 return 0;
8536 }
be7bd730
JB		 8537
8538 /*
8539 * If we are under memory pressure we will call this directly from the
8540 * VM, we need to make sure we have the inode referenced for the ordered
8541 * extent. If not just return like we didn't do anything.
8542 */
8543 if (!igrab(inode)) {
8544 redirty_page_for_writepage(wbc, page);
8545 return AOP_WRITEPAGE_ACTIVATE;
8546 }
d1310b2e 8547 tree = &BTRFS_I(page->mapping->host)->io_tree;
be7bd730
JB		 8548 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8549 btrfs_add_delayed_iput(inode);
8550 return ret;
9ebefb18
CM		 8551}
8552
48a3b636
ES		 8553static int btrfs_writepages(struct address_space *mapping,
8554 struct writeback_control *wbc)
b293f02e 8555{
d1310b2e 8556 struct extent_io_tree *tree;
771ed689 8557
d1310b2e 8558 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 8559 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8560}
8561
3ab2fb5a
CM		 8562static int
8563btrfs_readpages(struct file *file, struct address_space *mapping,
8564 struct list_head *pages, unsigned nr_pages)
8565{
d1310b2e
CM		 8566 struct extent_io_tree *tree;
8567 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 8568 return extent_readpages(tree, mapping, pages, nr_pages,
8569 btrfs_get_extent);
8570}
e6dcd2dc 8571static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 8572{
d1310b2e
CM		 8573 struct extent_io_tree *tree;
8574 struct extent_map_tree *map;
a52d9a80 8575 int ret;
8c2383c3 8576
d1310b2e
CM		 8577 tree = &BTRFS_I(page->mapping->host)->io_tree;
8578 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 8579 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 8580 if (ret == 1) {
8581 ClearPagePrivate(page);
8582 set_page_private(page, 0);
8583 page_cache_release(page);
39279cc3 8584 }
a52d9a80 8585 return ret;
39279cc3
CM		 8586}
8587
e6dcd2dc
CM		 8588static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8589{
98509cfc
CM		 8590 if (PageWriteback(page) || PageDirty(page))
8591 return 0;
b335b003 8592 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
e6dcd2dc
CM		 8593}
8594
d47992f8
LC		 8595static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8596 unsigned int length)
39279cc3 8597{
5fd02043 8598 struct inode *inode = page->mapping->host;
d1310b2e 8599 struct extent_io_tree *tree;
e6dcd2dc 8600 struct btrfs_ordered_extent *ordered;
2ac55d41 8601 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 8602 u64 page_start = page_offset(page);
8603 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
131e404a 8604 int inode_evicting = inode->i_state & I_FREEING;
39279cc3 8605
8b62b72b
CM		 8606 /*
8607 * we have the page locked, so new writeback can't start,
8608 * and the dirty bit won't be cleared while we are here.
8609 *
8610 * Wait for IO on this page so that we can safely clear
8611 * the PagePrivate2 bit and do ordered accounting
8612 */
e6dcd2dc 8613 wait_on_page_writeback(page);
8b62b72b 8614
5fd02043 8615 tree = &BTRFS_I(inode)->io_tree;
e6dcd2dc
CM		 8616 if (offset) {
8617 btrfs_releasepage(page, GFP_NOFS);
8618 return;
8619 }
131e404a
FDBM		 8620
8621 if (!inode_evicting)
8622 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
8623 ordered = btrfs_lookup_ordered_extent(inode, page_start);
e6dcd2dc 8624 if (ordered) {
eb84ae03
CM		 8625 /*
8626 * IO on this page will never be started, so we need
8627 * to account for any ordered extents now
8628 */
131e404a
FDBM		 8629 if (!inode_evicting)
8630 clear_extent_bit(tree, page_start, page_end,
8631 EXTENT_DIRTY | EXTENT_DELALLOC |
8632 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8633 EXTENT_DEFRAG, 1, 0, &cached_state,
8634 GFP_NOFS);
8b62b72b
CM		 8635 /*
8636 * whoever cleared the private bit is responsible
8637 * for the finish_ordered_io
8638 */
77cef2ec
JB		 8639 if (TestClearPagePrivate2(page)) {
8640 struct btrfs_ordered_inode_tree *tree;
8641 u64 new_len;
8642
8643 tree = &BTRFS_I(inode)->ordered_tree;
8644
8645 spin_lock_irq(&tree->lock);
8646 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8647 new_len = page_start - ordered->file_offset;
8648 if (new_len < ordered->truncated_len)
8649 ordered->truncated_len = new_len;
8650 spin_unlock_irq(&tree->lock);
8651
8652 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8653 page_start,
8654 PAGE_CACHE_SIZE, 1))
8655 btrfs_finish_ordered_io(ordered);
8b62b72b 8656 }
e6dcd2dc 8657 btrfs_put_ordered_extent(ordered);
131e404a
FDBM		 8658 if (!inode_evicting) {
8659 cached_state = NULL;
8660 lock_extent_bits(tree, page_start, page_end, 0,
8661 &cached_state);
8662 }
8663 }
8664
b9d0b389
QW		 8665 /*
8666 * Qgroup reserved space handler
8667 * Page here will be either
8668 * 1) Already written to disk
8669 * In this case, its reserved space is released from data rsv map
8670 * and will be freed by delayed_ref handler finally.
8671 * So even we call qgroup_free_data(), it won't decrease reserved
8672 * space.
8673 * 2) Not written to disk
8674 * This means the reserved space should be freed here.
8675 */
8676 btrfs_qgroup_free_data(inode, page_start, PAGE_CACHE_SIZE);
131e404a
FDBM		 8677 if (!inode_evicting) {
8678 clear_extent_bit(tree, page_start, page_end,
8679 EXTENT_LOCKED | EXTENT_DIRTY |
8680 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8681 EXTENT_DEFRAG, 1, 1,
8682 &cached_state, GFP_NOFS);
8683
8684 __btrfs_releasepage(page, GFP_NOFS);
e6dcd2dc 8685 }
e6dcd2dc 8686
4a096752 8687 ClearPageChecked(page);
9ad6b7bc 8688 if (PagePrivate(page)) {
9ad6b7bc
CM		 8689 ClearPagePrivate(page);
8690 set_page_private(page, 0);
8691 page_cache_release(page);
8692 }
39279cc3
CM		 8693}
8694
9ebefb18
CM		 8695/*
8696 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8697 * called from a page fault handler when a page is first dirtied. Hence we must
8698 * be careful to check for EOF conditions here. We set the page up correctly
8699 * for a written page which means we get ENOSPC checking when writing into
8700 * holes and correct delalloc and unwritten extent mapping on filesystems that
8701 * support these features.
8702 *
8703 * We are not allowed to take the i_mutex here so we have to play games to
8704 * protect against truncate races as the page could now be beyond EOF. Because
8705 * vmtruncate() writes the inode size before removing pages, once we have the
8706 * page lock we can determine safely if the page is beyond EOF. If it is not
8707 * beyond EOF, then the page is guaranteed safe against truncation until we
8708 * unlock the page.
8709 */
c2ec175c 8710int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
9ebefb18 8711{
c2ec175c 8712 struct page *page = vmf->page;
496ad9aa 8713 struct inode *inode = file_inode(vma->vm_file);
1832a6d5 8714 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 8715 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8716 struct btrfs_ordered_extent *ordered;
2ac55d41 8717 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 8718 char *kaddr;
8719 unsigned long zero_start;
9ebefb18 8720 loff_t size;
1832a6d5 8721 int ret;
9998eb70 8722 int reserved = 0;
a52d9a80 8723 u64 page_start;
e6dcd2dc 8724 u64 page_end;
9ebefb18 8725
b2b5ef5c 8726 sb_start_pagefault(inode->i_sb);
df480633
QW		 8727 page_start = page_offset(page);
8728 page_end = page_start + PAGE_CACHE_SIZE - 1;
8729
7cf5b976
QW		 8730 ret = btrfs_delalloc_reserve_space(inode, page_start,
8731 PAGE_CACHE_SIZE);
9998eb70 8732 if (!ret) {
e41f941a 8733 ret = file_update_time(vma->vm_file);
9998eb70
CM		 8734 reserved = 1;
8735 }
56a76f82
NP		 8736 if (ret) {
8737 if (ret == -ENOMEM)
8738 ret = VM_FAULT_OOM;
8739 else /* -ENOSPC, -EIO, etc */
8740 ret = VM_FAULT_SIGBUS;
9998eb70
CM		 8741 if (reserved)
8742 goto out;
8743 goto out_noreserve;
56a76f82 8744 }
1832a6d5 8745
56a76f82 8746 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
e6dcd2dc 8747again:
9ebefb18 8748 lock_page(page);
9ebefb18 8749 size = i_size_read(inode);
a52d9a80 8750
9ebefb18 8751 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 8752 (page_start >= size)) {
9ebefb18
CM		 8753 /* page got truncated out from underneath us */
8754 goto out_unlock;
8755 }
e6dcd2dc
CM		 8756 wait_on_page_writeback(page);
8757
d0082371 8758 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 8759 set_page_extent_mapped(page);
8760
eb84ae03
CM		 8761 /*
8762 * we can't set the delalloc bits if there are pending ordered
8763 * extents. Drop our locks and wait for them to finish
8764 */
e6dcd2dc
CM		 8765 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8766 if (ordered) {
2ac55d41
JB		 8767 unlock_extent_cached(io_tree, page_start, page_end,
8768 &cached_state, GFP_NOFS);
e6dcd2dc 8769 unlock_page(page);
eb84ae03 8770 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 8771 btrfs_put_ordered_extent(ordered);
8772 goto again;
8773 }
8774
fbf19087
JB		 8775 /*
8776 * XXX - page_mkwrite gets called every time the page is dirtied, even
8777 * if it was already dirty, so for space accounting reasons we need to
8778 * clear any delalloc bits for the range we are fixing to save. There
8779 * is probably a better way to do this, but for now keep consistent with
8780 * prepare_pages in the normal write path.
8781 */
2ac55d41 8782 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 8783 EXTENT_DIRTY | EXTENT_DELALLOC |
8784 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 8785 0, 0, &cached_state, GFP_NOFS);
fbf19087 8786
2ac55d41
JB		 8787 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
8788 &cached_state);
9ed74f2d 8789 if (ret) {
2ac55d41
JB		 8790 unlock_extent_cached(io_tree, page_start, page_end,
8791 &cached_state, GFP_NOFS);
9ed74f2d
JB		 8792 ret = VM_FAULT_SIGBUS;
8793 goto out_unlock;
8794 }
e6dcd2dc 8795 ret = 0;
9ebefb18
CM		 8796
8797 /* page is wholly or partially inside EOF */
a52d9a80 8798 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 8799 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 8800 else
e6dcd2dc 8801 zero_start = PAGE_CACHE_SIZE;
9ebefb18 8802
e6dcd2dc
CM		 8803 if (zero_start != PAGE_CACHE_SIZE) {
8804 kaddr = kmap(page);
8805 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
8806 flush_dcache_page(page);
8807 kunmap(page);
8808 }
247e743c 8809 ClearPageChecked(page);
e6dcd2dc 8810 set_page_dirty(page);
50a9b214 8811 SetPageUptodate(page);
5a3f23d5 8812
257c62e1
CM		 8813 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8814 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
46d8bc34 8815 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
257c62e1 8816
2ac55d41 8817 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9ebefb18
CM		 8818
8819out_unlock:
b2b5ef5c
JK		 8820 if (!ret) {
8821 sb_end_pagefault(inode->i_sb);
50a9b214 8822 return VM_FAULT_LOCKED;
b2b5ef5c 8823 }
9ebefb18 8824 unlock_page(page);
1832a6d5 8825out:
7cf5b976 8826 btrfs_delalloc_release_space(inode, page_start, PAGE_CACHE_SIZE);
9998eb70 8827out_noreserve:
b2b5ef5c 8828 sb_end_pagefault(inode->i_sb);
9ebefb18
CM		 8829 return ret;
8830}
8831
a41ad394 8832static int btrfs_truncate(struct inode *inode)
39279cc3
CM		 8833{
8834 struct btrfs_root *root = BTRFS_I(inode)->root;
fcb80c2a 8835 struct btrfs_block_rsv *rsv;
a71754fc 8836 int ret = 0;
3893e33b 8837 int err = 0;
39279cc3 8838 struct btrfs_trans_handle *trans;
dbe674a9 8839 u64 mask = root->sectorsize - 1;
07127184 8840 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3 8841
0ef8b726
JB		 8842 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8843 (u64)-1);
8844 if (ret)
8845 return ret;
39279cc3 8846
fcb80c2a
JB		 8847 /*
8848 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8849 * 3 things going on here
8850 *
8851 * 1) We need to reserve space for our orphan item and the space to
8852 * delete our orphan item. Lord knows we don't want to have a dangling
8853 * orphan item because we didn't reserve space to remove it.
8854 *
8855 * 2) We need to reserve space to update our inode.
8856 *
8857 * 3) We need to have something to cache all the space that is going to
8858 * be free'd up by the truncate operation, but also have some slack
8859 * space reserved in case it uses space during the truncate (thank you
8860 * very much snapshotting).
8861 *
8862 * And we need these to all be seperate. The fact is we can use alot of
8863 * space doing the truncate, and we have no earthly idea how much space
8864 * we will use, so we need the truncate reservation to be seperate so it
8865 * doesn't end up using space reserved for updating the inode or
8866 * removing the orphan item. We also need to be able to stop the
8867 * transaction and start a new one, which means we need to be able to
8868 * update the inode several times, and we have no idea of knowing how
8869 * many times that will be, so we can't just reserve 1 item for the
8870 * entirety of the opration, so that has to be done seperately as well.
8871 * Then there is the orphan item, which does indeed need to be held on
8872 * to for the whole operation, and we need nobody to touch this reserved
8873 * space except the orphan code.
8874 *
8875 * So that leaves us with
8876 *
8877 * 1) root->orphan_block_rsv - for the orphan deletion.
8878 * 2) rsv - for the truncate reservation, which we will steal from the
8879 * transaction reservation.
8880 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8881 * updating the inode.
8882 */
66d8f3dd 8883 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
fcb80c2a
JB		 8884 if (!rsv)
8885 return -ENOMEM;
4a338542 8886 rsv->size = min_size;
ca7e70f5 8887 rsv->failfast = 1;
f0cd846e 8888
907cbceb 8889 /*
07127184 8890 * 1 for the truncate slack space
907cbceb
JB		 8891 * 1 for updating the inode.
8892 */
f3fe820c 8893 trans = btrfs_start_transaction(root, 2);
fcb80c2a
JB		 8894 if (IS_ERR(trans)) {
8895 err = PTR_ERR(trans);
8896 goto out;
8897 }
f0cd846e 8898
907cbceb
JB		 8899 /* Migrate the slack space for the truncate to our reserve */
8900 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8901 min_size);
fcb80c2a 8902 BUG_ON(ret);
f0cd846e 8903
5dc562c5
JB		 8904 /*
8905 * So if we truncate and then write and fsync we normally would just
8906 * write the extents that changed, which is a problem if we need to
8907 * first truncate that entire inode. So set this flag so we write out
8908 * all of the extents in the inode to the sync log so we're completely
8909 * safe.
8910 */
8911 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
ca7e70f5 8912 trans->block_rsv = rsv;
907cbceb 8913
8082510e
YZ		 8914 while (1) {
8915 ret = btrfs_truncate_inode_items(trans, root, inode,
8916 inode->i_size,
8917 BTRFS_EXTENT_DATA_KEY);
28ed1345 8918 if (ret != -ENOSPC && ret != -EAGAIN) {
3893e33b 8919 err = ret;
8082510e 8920 break;
3893e33b 8921 }
39279cc3 8922
fcb80c2a 8923 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e 8924 ret = btrfs_update_inode(trans, root, inode);
3893e33b
JB		 8925 if (ret) {
8926 err = ret;
8927 break;
8928 }
ca7e70f5 8929
8082510e 8930 btrfs_end_transaction(trans, root);
b53d3f5d 8931 btrfs_btree_balance_dirty(root);
ca7e70f5
JB		 8932
8933 trans = btrfs_start_transaction(root, 2);
8934 if (IS_ERR(trans)) {
8935 ret = err = PTR_ERR(trans);
8936 trans = NULL;
8937 break;
8938 }
8939
8940 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
8941 rsv, min_size);
8942 BUG_ON(ret); /* shouldn't happen */
8943 trans->block_rsv = rsv;
8082510e
YZ		 8944 }
8945
8946 if (ret == 0 && inode->i_nlink > 0) {
fcb80c2a 8947 trans->block_rsv = root->orphan_block_rsv;
8082510e 8948 ret = btrfs_orphan_del(trans, inode);
3893e33b
JB		 8949 if (ret)
8950 err = ret;
8082510e
YZ		 8951 }
8952
917c16b2
CM		 8953 if (trans) {
8954 trans->block_rsv = &root->fs_info->trans_block_rsv;
8955 ret = btrfs_update_inode(trans, root, inode);
8956 if (ret && !err)
8957 err = ret;
7b128766 8958
7ad85bb7 8959 ret = btrfs_end_transaction(trans, root);
b53d3f5d 8960 btrfs_btree_balance_dirty(root);
917c16b2 8961 }
fcb80c2a
JB		 8962
8963out:
8964 btrfs_free_block_rsv(root, rsv);
8965
3893e33b
JB		 8966 if (ret && !err)
8967 err = ret;
a41ad394 8968
3893e33b 8969 return err;
39279cc3
CM		 8970}
8971
d352ac68
CM		 8972/*
8973 * create a new subvolume directory/inode (helper for the ioctl).
8974 */
d2fb3437 8975int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
63541927
FDBM		 8976 struct btrfs_root *new_root,
8977 struct btrfs_root *parent_root,
8978 u64 new_dirid)
39279cc3 8979{
39279cc3 8980 struct inode *inode;
76dda93c 8981 int err;
00e4e6b3 8982 u64 index = 0;
39279cc3 8983
12fc9d09
FA		 8984 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8985 new_dirid, new_dirid,
8986 S_IFDIR | (~current_umask() & S_IRWXUGO),
8987 &index);
54aa1f4d 8988 if (IS_ERR(inode))
f46b5a66 8989 return PTR_ERR(inode);
39279cc3
CM		 8990 inode->i_op = &btrfs_dir_inode_operations;
8991 inode->i_fop = &btrfs_dir_file_operations;
8992
bfe86848 8993 set_nlink(inode, 1);
dbe674a9 8994 btrfs_i_size_write(inode, 0);
b0d5d10f 8995 unlock_new_inode(inode);
3b96362c 8996
63541927
FDBM		 8997 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8998 if (err)
8999 btrfs_err(new_root->fs_info,
351fd353 9000 "error inheriting subvolume %llu properties: %d",
63541927
FDBM		 9001 new_root->root_key.objectid, err);
9002
76dda93c 9003 err = btrfs_update_inode(trans, new_root, inode);
cb8e7090 9004
76dda93c 9005 iput(inode);
ce598979 9006 return err;
39279cc3
CM		 9007}
9008
39279cc3
CM		 9009struct inode *btrfs_alloc_inode(struct super_block *sb)
9010{
9011 struct btrfs_inode *ei;
2ead6ae7 9012 struct inode *inode;
39279cc3
CM		 9013
9014 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9015 if (!ei)
9016 return NULL;
2ead6ae7
YZ		 9017
9018 ei->root = NULL;
2ead6ae7 9019 ei->generation = 0;
15ee9bc7 9020 ei->last_trans = 0;
257c62e1 9021 ei->last_sub_trans = 0;
e02119d5 9022 ei->logged_trans = 0;
2ead6ae7 9023 ei->delalloc_bytes = 0;
47059d93 9024 ei->defrag_bytes = 0;
2ead6ae7
YZ		 9025 ei->disk_i_size = 0;
9026 ei->flags = 0;
7709cde3 9027 ei->csum_bytes = 0;
2ead6ae7 9028 ei->index_cnt = (u64)-1;
67de1176 9029 ei->dir_index = 0;
2ead6ae7 9030 ei->last_unlink_trans = 0;
46d8bc34 9031 ei->last_log_commit = 0;
8089fe62 9032 ei->delayed_iput_count = 0;
2ead6ae7 9033
9e0baf60
JB		 9034 spin_lock_init(&ei->lock);
9035 ei->outstanding_extents = 0;
9036 ei->reserved_extents = 0;
2ead6ae7 9037
72ac3c0d 9038 ei->runtime_flags = 0;
261507a0 9039 ei->force_compress = BTRFS_COMPRESS_NONE;
2ead6ae7 9040
16cdcec7
MX		 9041 ei->delayed_node = NULL;
9042
9cc97d64 9043 ei->i_otime.tv_sec = 0;
9044 ei->i_otime.tv_nsec = 0;
9045
2ead6ae7 9046 inode = &ei->vfs_inode;
a8067e02 9047 extent_map_tree_init(&ei->extent_tree);
f993c883
DS		 9048 extent_io_tree_init(&ei->io_tree, &inode->i_data);
9049 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
0b32f4bb
JB		 9050 ei->io_tree.track_uptodate = 1;
9051 ei->io_failure_tree.track_uptodate = 1;
b812ce28 9052 atomic_set(&ei->sync_writers, 0);
2ead6ae7 9053 mutex_init(&ei->log_mutex);
f248679e 9054 mutex_init(&ei->delalloc_mutex);
e6dcd2dc 9055 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
2ead6ae7 9056 INIT_LIST_HEAD(&ei->delalloc_inodes);
8089fe62 9057 INIT_LIST_HEAD(&ei->delayed_iput);
2ead6ae7
YZ		 9058 RB_CLEAR_NODE(&ei->rb_node);
9059
9060 return inode;
39279cc3
CM		 9061}
9062
aaedb55b
JB		 9063#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9064void btrfs_test_destroy_inode(struct inode *inode)
9065{
9066 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9067 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9068}
9069#endif
9070
fa0d7e3d
NP		 9071static void btrfs_i_callback(struct rcu_head *head)
9072{
9073 struct inode *inode = container_of(head, struct inode, i_rcu);
fa0d7e3d
NP		 9074 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9075}
9076
39279cc3
CM		 9077void btrfs_destroy_inode(struct inode *inode)
9078{
e6dcd2dc 9079 struct btrfs_ordered_extent *ordered;
5a3f23d5
CM		 9080 struct btrfs_root *root = BTRFS_I(inode)->root;
9081
b3d9b7a3 9082 WARN_ON(!hlist_empty(&inode->i_dentry));
39279cc3 9083 WARN_ON(inode->i_data.nrpages);
9e0baf60
JB		 9084 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9085 WARN_ON(BTRFS_I(inode)->reserved_extents);
7709cde3
JB		 9086 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9087 WARN_ON(BTRFS_I(inode)->csum_bytes);
47059d93 9088 WARN_ON(BTRFS_I(inode)->defrag_bytes);
39279cc3 9089
a6dbd429
JB		 9090 /*
9091 * This can happen where we create an inode, but somebody else also
9092 * created the same inode and we need to destroy the one we already
9093 * created.
9094 */
9095 if (!root)
9096 goto free;
9097
8a35d95f
JB		 9098 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9099 &BTRFS_I(inode)->runtime_flags)) {
c2cf52eb 9100 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
c1c9ff7c 9101 btrfs_ino(inode));
8a35d95f 9102 atomic_dec(&root->orphan_inodes);
7b128766 9103 }
7b128766 9104
d397712b 9105 while (1) {
e6dcd2dc
CM		 9106 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9107 if (!ordered)
9108 break;
9109 else {
c2cf52eb 9110 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
c1c9ff7c 9111 ordered->file_offset, ordered->len);
e6dcd2dc
CM		 9112 btrfs_remove_ordered_extent(inode, ordered);
9113 btrfs_put_ordered_extent(ordered);
9114 btrfs_put_ordered_extent(ordered);
9115 }
9116 }
56fa9d07 9117 btrfs_qgroup_check_reserved_leak(inode);
5d4f98a2 9118 inode_tree_del(inode);
5b21f2ed 9119 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
a6dbd429 9120free:
fa0d7e3d 9121 call_rcu(&inode->i_rcu, btrfs_i_callback);
39279cc3
CM		 9122}
9123
45321ac5 9124int btrfs_drop_inode(struct inode *inode)
76dda93c
YZ		 9125{
9126 struct btrfs_root *root = BTRFS_I(inode)->root;
45321ac5 9127
6379ef9f
NA		 9128 if (root == NULL)
9129 return 1;
9130
fa6ac876 9131 /* the snap/subvol tree is on deleting */
69e9c6c6 9132 if (btrfs_root_refs(&root->root_item) == 0)
45321ac5 9133 return 1;
76dda93c 9134 else
45321ac5 9135 return generic_drop_inode(inode);
76dda93c
YZ		 9136}
9137
0ee0fda0 9138static void init_once(void *foo)
39279cc3
CM		 9139{
9140 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9141
9142 inode_init_once(&ei->vfs_inode);
9143}
9144
9145void btrfs_destroy_cachep(void)
9146{
8c0a8537
KS		 9147 /*
9148 * Make sure all delayed rcu free inodes are flushed before we
9149 * destroy cache.
9150 */
9151 rcu_barrier();
39279cc3
CM		 9152 if (btrfs_inode_cachep)
9153 kmem_cache_destroy(btrfs_inode_cachep);
9154 if (btrfs_trans_handle_cachep)
9155 kmem_cache_destroy(btrfs_trans_handle_cachep);
9156 if (btrfs_transaction_cachep)
9157 kmem_cache_destroy(btrfs_transaction_cachep);
39279cc3
CM		 9158 if (btrfs_path_cachep)
9159 kmem_cache_destroy(btrfs_path_cachep);
dc89e982
JB		 9160 if (btrfs_free_space_cachep)
9161 kmem_cache_destroy(btrfs_free_space_cachep);
8ccf6f19
MX		 9162 if (btrfs_delalloc_work_cachep)
9163 kmem_cache_destroy(btrfs_delalloc_work_cachep);
39279cc3
CM		 9164}
9165
9166int btrfs_init_cachep(void)
9167{
837e1972 9168 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9601e3f6
CH		 9169 sizeof(struct btrfs_inode), 0,
9170 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
39279cc3
CM		 9171 if (!btrfs_inode_cachep)
9172 goto fail;
9601e3f6 9173
837e1972 9174 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9601e3f6
CH		 9175 sizeof(struct btrfs_trans_handle), 0,
9176 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 9177 if (!btrfs_trans_handle_cachep)
9178 goto fail;
9601e3f6 9179
837e1972 9180 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9601e3f6
CH		 9181 sizeof(struct btrfs_transaction), 0,
9182 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 9183 if (!btrfs_transaction_cachep)
9184 goto fail;
9601e3f6 9185
837e1972 9186 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9601e3f6
CH		 9187 sizeof(struct btrfs_path), 0,
9188 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 9189 if (!btrfs_path_cachep)
9190 goto fail;
9601e3f6 9191
837e1972 9192 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
dc89e982
JB		 9193 sizeof(struct btrfs_free_space), 0,
9194 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9195 if (!btrfs_free_space_cachep)
9196 goto fail;
9197
8ccf6f19
MX		 9198 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
9199 sizeof(struct btrfs_delalloc_work), 0,
9200 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
9201 NULL);
9202 if (!btrfs_delalloc_work_cachep)
9203 goto fail;
9204
39279cc3
CM		 9205 return 0;
9206fail:
9207 btrfs_destroy_cachep();
9208 return -ENOMEM;
9209}
9210
9211static int btrfs_getattr(struct vfsmount *mnt,
9212 struct dentry *dentry, struct kstat *stat)
9213{
df0af1a5 9214 u64 delalloc_bytes;
2b0143b5 9215 struct inode *inode = d_inode(dentry);
fadc0d8b
DS		 9216 u32 blocksize = inode->i_sb->s_blocksize;
9217
39279cc3 9218 generic_fillattr(inode, stat);
0ee5dc67 9219 stat->dev = BTRFS_I(inode)->root->anon_dev;
d6667462 9220 stat->blksize = PAGE_CACHE_SIZE;
df0af1a5
MX		 9221
9222 spin_lock(&BTRFS_I(inode)->lock);
9223 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9224 spin_unlock(&BTRFS_I(inode)->lock);
fadc0d8b 9225 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
df0af1a5 9226 ALIGN(delalloc_bytes, blocksize)) >> 9;
39279cc3
CM		 9227 return 0;
9228}
9229
d397712b
CM		 9230static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9231 struct inode *new_dir, struct dentry *new_dentry)
39279cc3
CM		 9232{
9233 struct btrfs_trans_handle *trans;
9234 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4df27c4d 9235 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
2b0143b5
DH		 9236 struct inode *new_inode = d_inode(new_dentry);
9237 struct inode *old_inode = d_inode(old_dentry);
39279cc3 9238 struct timespec ctime = CURRENT_TIME;
00e4e6b3 9239 u64 index = 0;
4df27c4d 9240 u64 root_objectid;
39279cc3 9241 int ret;
33345d01 9242 u64 old_ino = btrfs_ino(old_inode);
39279cc3 9243
33345d01 9244 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
f679a840
YZ		 9245 return -EPERM;
9246
4df27c4d 9247 /* we only allow rename subvolume link between subvolumes */
33345d01 9248 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
3394e160
CM		 9249 return -EXDEV;
9250
33345d01
LZ		 9251 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9252 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
39279cc3 9253 return -ENOTEMPTY;
5f39d397 9254
4df27c4d
YZ		 9255 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9256 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9257 return -ENOTEMPTY;
9c52057c
CM		 9258
9259
9260 /* check for collisions, even if the name isn't there */
4871c158 9261 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9c52057c
CM		 9262 new_dentry->d_name.name,
9263 new_dentry->d_name.len);
9264
9265 if (ret) {
9266 if (ret == -EEXIST) {
9267 /* we shouldn't get
9268 * eexist without a new_inode */
fae7f21c 9269 if (WARN_ON(!new_inode)) {
9c52057c
CM		 9270 return ret;
9271 }
9272 } else {
9273 /* maybe -EOVERFLOW */
9274 return ret;
9275 }
9276 }
9277 ret = 0;
9278
5a3f23d5 9279 /*
8d875f95
CM		 9280 * we're using rename to replace one file with another. Start IO on it
9281 * now so we don't add too much work to the end of the transaction
5a3f23d5 9282 */
8d875f95 9283 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
5a3f23d5
CM		 9284 filemap_flush(old_inode->i_mapping);
9285
76dda93c 9286 /* close the racy window with snapshot create/destroy ioctl */
33345d01 9287 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 9288 down_read(&root->fs_info->subvol_sem);
a22285a6
YZ		 9289 /*
9290 * We want to reserve the absolute worst case amount of items. So if
9291 * both inodes are subvols and we need to unlink them then that would
9292 * require 4 item modifications, but if they are both normal inodes it
9293 * would require 5 item modifications, so we'll assume their normal
9294 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9295 * should cover the worst case number of items we'll modify.
9296 */
6e137ed3 9297 trans = btrfs_start_transaction(root, 11);
b44c59a8
JL		 9298 if (IS_ERR(trans)) {
9299 ret = PTR_ERR(trans);
9300 goto out_notrans;
9301 }
76dda93c 9302
4df27c4d
YZ		 9303 if (dest != root)
9304 btrfs_record_root_in_trans(trans, dest);
5f39d397 9305
a5719521
YZ		 9306 ret = btrfs_set_inode_index(new_dir, &index);
9307 if (ret)
9308 goto out_fail;
5a3f23d5 9309
67de1176 9310 BTRFS_I(old_inode)->dir_index = 0ULL;
33345d01 9311 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d 9312 /* force full log commit if subvolume involved. */
995946dd 9313 btrfs_set_log_full_commit(root->fs_info, trans);
4df27c4d 9314 } else {
a5719521
YZ		 9315 ret = btrfs_insert_inode_ref(trans, dest,
9316 new_dentry->d_name.name,
9317 new_dentry->d_name.len,
33345d01
LZ		 9318 old_ino,
9319 btrfs_ino(new_dir), index);
a5719521
YZ		 9320 if (ret)
9321 goto out_fail;
4df27c4d
YZ		 9322 /*
9323 * this is an ugly little race, but the rename is required
9324 * to make sure that if we crash, the inode is either at the
9325 * old name or the new one. pinning the log transaction lets
9326 * us make sure we don't allow a log commit to come in after
9327 * we unlink the name but before we add the new name back in.
9328 */
9329 btrfs_pin_log_trans(root);
9330 }
5a3f23d5 9331
0c4d2d95
JB		 9332 inode_inc_iversion(old_dir);
9333 inode_inc_iversion(new_dir);
9334 inode_inc_iversion(old_inode);
39279cc3
CM		 9335 old_dir->i_ctime = old_dir->i_mtime = ctime;
9336 new_dir->i_ctime = new_dir->i_mtime = ctime;
9337 old_inode->i_ctime = ctime;
5f39d397 9338
12fcfd22
CM		 9339 if (old_dentry->d_parent != new_dentry->d_parent)
9340 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9341
33345d01 9342 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 9343 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9344 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9345 old_dentry->d_name.name,
9346 old_dentry->d_name.len);
9347 } else {
92986796 9348 ret = __btrfs_unlink_inode(trans, root, old_dir,
2b0143b5 9349 d_inode(old_dentry),
92986796
AV		 9350 old_dentry->d_name.name,
9351 old_dentry->d_name.len);
9352 if (!ret)
9353 ret = btrfs_update_inode(trans, root, old_inode);
4df27c4d 9354 }
79787eaa
JM		 9355 if (ret) {
9356 btrfs_abort_transaction(trans, root, ret);
9357 goto out_fail;
9358 }
39279cc3
CM		 9359
9360 if (new_inode) {
0c4d2d95 9361 inode_inc_iversion(new_inode);
39279cc3 9362 new_inode->i_ctime = CURRENT_TIME;
33345d01 9363 if (unlikely(btrfs_ino(new_inode) ==
4df27c4d
YZ		 9364 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9365 root_objectid = BTRFS_I(new_inode)->location.objectid;
9366 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9367 root_objectid,
9368 new_dentry->d_name.name,
9369 new_dentry->d_name.len);
9370 BUG_ON(new_inode->i_nlink == 0);
9371 } else {
9372 ret = btrfs_unlink_inode(trans, dest, new_dir,
2b0143b5 9373 d_inode(new_dentry),
4df27c4d
YZ		 9374 new_dentry->d_name.name,
9375 new_dentry->d_name.len);
9376 }
4ef31a45 9377 if (!ret && new_inode->i_nlink == 0)
2b0143b5 9378 ret = btrfs_orphan_add(trans, d_inode(new_dentry));
79787eaa
JM		 9379 if (ret) {
9380 btrfs_abort_transaction(trans, root, ret);
9381 goto out_fail;
9382 }
39279cc3 9383 }
aec7477b 9384
4df27c4d
YZ		 9385 ret = btrfs_add_link(trans, new_dir, old_inode,
9386 new_dentry->d_name.name,
a5719521 9387 new_dentry->d_name.len, 0, index);
79787eaa
JM		 9388 if (ret) {
9389 btrfs_abort_transaction(trans, root, ret);
9390 goto out_fail;
9391 }
39279cc3 9392
67de1176
MX		 9393 if (old_inode->i_nlink == 1)
9394 BTRFS_I(old_inode)->dir_index = index;
9395
33345d01 9396 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
10d9f309 9397 struct dentry *parent = new_dentry->d_parent;
6a912213 9398 btrfs_log_new_name(trans, old_inode, old_dir, parent);
4df27c4d
YZ		 9399 btrfs_end_log_trans(root);
9400 }
39279cc3 9401out_fail:
7ad85bb7 9402 btrfs_end_transaction(trans, root);
b44c59a8 9403out_notrans:
33345d01 9404 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 9405 up_read(&root->fs_info->subvol_sem);
9ed74f2d 9406
39279cc3
CM		 9407 return ret;
9408}
9409
80ace85c
MS		 9410static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9411 struct inode *new_dir, struct dentry *new_dentry,
9412 unsigned int flags)
9413{
9414 if (flags & ~RENAME_NOREPLACE)
9415 return -EINVAL;
9416
9417 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
9418}
9419
8ccf6f19
MX		 9420static void btrfs_run_delalloc_work(struct btrfs_work *work)
9421{
9422 struct btrfs_delalloc_work *delalloc_work;
9f23e289 9423 struct inode *inode;
8ccf6f19
MX		 9424
9425 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9426 work);
9f23e289
JB		 9427 inode = delalloc_work->inode;
9428 if (delalloc_work->wait) {
9429 btrfs_wait_ordered_range(inode, 0, (u64)-1);
9430 } else {
9431 filemap_flush(inode->i_mapping);
9432 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9433 &BTRFS_I(inode)->runtime_flags))
9434 filemap_flush(inode->i_mapping);
9435 }
8ccf6f19
MX		 9436
9437 if (delalloc_work->delay_iput)
9f23e289 9438 btrfs_add_delayed_iput(inode);
8ccf6f19 9439 else
9f23e289 9440 iput(inode);
8ccf6f19
MX		 9441 complete(&delalloc_work->completion);
9442}
9443
9444struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9445 int wait, int delay_iput)
9446{
9447 struct btrfs_delalloc_work *work;
9448
9449 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
9450 if (!work)
9451 return NULL;
9452
9453 init_completion(&work->completion);
9454 INIT_LIST_HEAD(&work->list);
9455 work->inode = inode;
9456 work->wait = wait;
9457 work->delay_iput = delay_iput;
9e0af237
LB		 9458 WARN_ON_ONCE(!inode);
9459 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9460 btrfs_run_delalloc_work, NULL, NULL);
8ccf6f19
MX		 9461
9462 return work;
9463}
9464
9465void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9466{
9467 wait_for_completion(&work->completion);
9468 kmem_cache_free(btrfs_delalloc_work_cachep, work);
9469}
9470
d352ac68
CM		 9471/*
9472 * some fairly slow code that needs optimization. This walks the list
9473 * of all the inodes with pending delalloc and forces them to disk.
9474 */
6c255e67
MX		 9475static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9476 int nr)
ea8c2819 9477{
ea8c2819 9478 struct btrfs_inode *binode;
5b21f2ed 9479 struct inode *inode;
8ccf6f19
MX		 9480 struct btrfs_delalloc_work *work, *next;
9481 struct list_head works;
1eafa6c7 9482 struct list_head splice;
8ccf6f19 9483 int ret = 0;
ea8c2819 9484
8ccf6f19 9485 INIT_LIST_HEAD(&works);
1eafa6c7 9486 INIT_LIST_HEAD(&splice);
63607cc8 9487
573bfb72 9488 mutex_lock(&root->delalloc_mutex);
eb73c1b7
MX		 9489 spin_lock(&root->delalloc_lock);
9490 list_splice_init(&root->delalloc_inodes, &splice);
1eafa6c7
MX		 9491 while (!list_empty(&splice)) {
9492 binode = list_entry(splice.next, struct btrfs_inode,
ea8c2819 9493 delalloc_inodes);
1eafa6c7 9494
eb73c1b7
MX		 9495 list_move_tail(&binode->delalloc_inodes,
9496 &root->delalloc_inodes);
5b21f2ed 9497 inode = igrab(&binode->vfs_inode);
df0af1a5 9498 if (!inode) {
eb73c1b7 9499 cond_resched_lock(&root->delalloc_lock);
1eafa6c7 9500 continue;
df0af1a5 9501 }
eb73c1b7 9502 spin_unlock(&root->delalloc_lock);
1eafa6c7
MX		 9503
9504 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
5d99a998 9505 if (!work) {
f4ab9ea7
JB		 9506 if (delay_iput)
9507 btrfs_add_delayed_iput(inode);
9508 else
9509 iput(inode);
1eafa6c7 9510 ret = -ENOMEM;
a1ecaabb 9511 goto out;
5b21f2ed 9512 }
1eafa6c7 9513 list_add_tail(&work->list, &works);
a44903ab
QW		 9514 btrfs_queue_work(root->fs_info->flush_workers,
9515 &work->work);
6c255e67
MX		 9516 ret++;
9517 if (nr != -1 && ret >= nr)
a1ecaabb 9518 goto out;
5b21f2ed 9519 cond_resched();
eb73c1b7 9520 spin_lock(&root->delalloc_lock);
ea8c2819 9521 }
eb73c1b7 9522 spin_unlock(&root->delalloc_lock);
8c8bee1d 9523
a1ecaabb 9524out:
eb73c1b7
MX		 9525 list_for_each_entry_safe(work, next, &works, list) {
9526 list_del_init(&work->list);
9527 btrfs_wait_and_free_delalloc_work(work);
9528 }
9529
9530 if (!list_empty_careful(&splice)) {
9531 spin_lock(&root->delalloc_lock);
9532 list_splice_tail(&splice, &root->delalloc_inodes);
9533 spin_unlock(&root->delalloc_lock);
9534 }
573bfb72 9535 mutex_unlock(&root->delalloc_mutex);
eb73c1b7
MX		 9536 return ret;
9537}
1eafa6c7 9538
eb73c1b7
MX		 9539int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9540{
9541 int ret;
1eafa6c7 9542
2c21b4d7 9543 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
eb73c1b7
MX		 9544 return -EROFS;
9545
6c255e67
MX		 9546 ret = __start_delalloc_inodes(root, delay_iput, -1);
9547 if (ret > 0)
9548 ret = 0;
eb73c1b7
MX		 9549 /*
9550 * the filemap_flush will queue IO into the worker threads, but
8c8bee1d
CM		 9551 * we have to make sure the IO is actually started and that
9552 * ordered extents get created before we return
9553 */
9554 atomic_inc(&root->fs_info->async_submit_draining);
d397712b 9555 while (atomic_read(&root->fs_info->nr_async_submits) ||
771ed689 9556 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 9557 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 9558 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9559 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 9560 }
9561 atomic_dec(&root->fs_info->async_submit_draining);
eb73c1b7
MX		 9562 return ret;
9563}
9564
6c255e67
MX		 9565int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9566 int nr)
eb73c1b7
MX		 9567{
9568 struct btrfs_root *root;
9569 struct list_head splice;
9570 int ret;
9571
2c21b4d7 9572 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
eb73c1b7
MX		 9573 return -EROFS;
9574
9575 INIT_LIST_HEAD(&splice);
9576
573bfb72 9577 mutex_lock(&fs_info->delalloc_root_mutex);
eb73c1b7
MX		 9578 spin_lock(&fs_info->delalloc_root_lock);
9579 list_splice_init(&fs_info->delalloc_roots, &splice);
6c255e67 9580 while (!list_empty(&splice) && nr) {
eb73c1b7
MX		 9581 root = list_first_entry(&splice, struct btrfs_root,
9582 delalloc_root);
9583 root = btrfs_grab_fs_root(root);
9584 BUG_ON(!root);
9585 list_move_tail(&root->delalloc_root,
9586 &fs_info->delalloc_roots);
9587 spin_unlock(&fs_info->delalloc_root_lock);
9588
6c255e67 9589 ret = __start_delalloc_inodes(root, delay_iput, nr);
eb73c1b7 9590 btrfs_put_fs_root(root);
6c255e67 9591 if (ret < 0)
eb73c1b7
MX		 9592 goto out;
9593
6c255e67
MX		 9594 if (nr != -1) {
9595 nr -= ret;
9596 WARN_ON(nr < 0);
9597 }
eb73c1b7 9598 spin_lock(&fs_info->delalloc_root_lock);
8ccf6f19 9599 }
eb73c1b7 9600 spin_unlock(&fs_info->delalloc_root_lock);
1eafa6c7 9601
6c255e67 9602 ret = 0;
eb73c1b7
MX		 9603 atomic_inc(&fs_info->async_submit_draining);
9604 while (atomic_read(&fs_info->nr_async_submits) ||
9605 atomic_read(&fs_info->async_delalloc_pages)) {
9606 wait_event(fs_info->async_submit_wait,
9607 (atomic_read(&fs_info->nr_async_submits) == 0 &&
9608 atomic_read(&fs_info->async_delalloc_pages) == 0));
9609 }
9610 atomic_dec(&fs_info->async_submit_draining);
eb73c1b7 9611out:
1eafa6c7 9612 if (!list_empty_careful(&splice)) {
eb73c1b7
MX		 9613 spin_lock(&fs_info->delalloc_root_lock);
9614 list_splice_tail(&splice, &fs_info->delalloc_roots);
9615 spin_unlock(&fs_info->delalloc_root_lock);
1eafa6c7 9616 }
573bfb72 9617 mutex_unlock(&fs_info->delalloc_root_mutex);
8ccf6f19 9618 return ret;
ea8c2819
CM		 9619}
9620
39279cc3
CM		 9621static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9622 const char *symname)
9623{
9624 struct btrfs_trans_handle *trans;
9625 struct btrfs_root *root = BTRFS_I(dir)->root;
9626 struct btrfs_path *path;
9627 struct btrfs_key key;
1832a6d5 9628 struct inode *inode = NULL;
39279cc3
CM		 9629 int err;
9630 int drop_inode = 0;
9631 u64 objectid;
67871254 9632 u64 index = 0;
39279cc3
CM		 9633 int name_len;
9634 int datasize;
5f39d397 9635 unsigned long ptr;
39279cc3 9636 struct btrfs_file_extent_item *ei;
5f39d397 9637 struct extent_buffer *leaf;
39279cc3 9638
f06becc4 9639 name_len = strlen(symname);
39279cc3
CM		 9640 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9641 return -ENAMETOOLONG;
1832a6d5 9642
9ed74f2d
JB		 9643 /*
9644 * 2 items for inode item and ref
9645 * 2 items for dir items
9646 * 1 item for xattr if selinux is on
9647 */
a22285a6
YZ		 9648 trans = btrfs_start_transaction(root, 5);
9649 if (IS_ERR(trans))
9650 return PTR_ERR(trans);
1832a6d5 9651
581bb050
LZ		 9652 err = btrfs_find_free_ino(root, &objectid);
9653 if (err)
9654 goto out_unlock;
9655
aec7477b 9656 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 9657 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 9658 S_IFLNK|S_IRWXUGO, &index);
7cf96da3
TI		 9659 if (IS_ERR(inode)) {
9660 err = PTR_ERR(inode);
39279cc3 9661 goto out_unlock;
7cf96da3 9662 }
39279cc3 9663
ad19db71
CS		 9664 /*
9665 * If the active LSM wants to access the inode during
9666 * d_instantiate it needs these. Smack checks to see
9667 * if the filesystem supports xattrs by looking at the
9668 * ops vector.
9669 */
9670 inode->i_fop = &btrfs_file_operations;
9671 inode->i_op = &btrfs_file_inode_operations;
b0d5d10f 9672 inode->i_mapping->a_ops = &btrfs_aops;
b0d5d10f
CM		 9673 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9674
9675 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9676 if (err)
9677 goto out_unlock_inode;
ad19db71 9678
a1b075d2 9679 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3 9680 if (err)
b0d5d10f 9681 goto out_unlock_inode;
39279cc3
CM		 9682
9683 path = btrfs_alloc_path();
d8926bb3
MF		 9684 if (!path) {
9685 err = -ENOMEM;
b0d5d10f 9686 goto out_unlock_inode;
d8926bb3 9687 }
33345d01 9688 key.objectid = btrfs_ino(inode);
39279cc3 9689 key.offset = 0;
962a298f 9690 key.type = BTRFS_EXTENT_DATA_KEY;
39279cc3
CM		 9691 datasize = btrfs_file_extent_calc_inline_size(name_len);
9692 err = btrfs_insert_empty_item(trans, root, path, &key,
9693 datasize);
54aa1f4d 9694 if (err) {
b0839166 9695 btrfs_free_path(path);
b0d5d10f 9696 goto out_unlock_inode;
54aa1f4d 9697 }
5f39d397
CM		 9698 leaf = path->nodes[0];
9699 ei = btrfs_item_ptr(leaf, path->slots[0],
9700 struct btrfs_file_extent_item);
9701 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9702 btrfs_set_file_extent_type(leaf, ei,
39279cc3 9703 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 9704 btrfs_set_file_extent_encryption(leaf, ei, 0);
9705 btrfs_set_file_extent_compression(leaf, ei, 0);
9706 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9707 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9708
39279cc3 9709 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 9710 write_extent_buffer(leaf, symname, ptr, name_len);
9711 btrfs_mark_buffer_dirty(leaf);
39279cc3 9712 btrfs_free_path(path);
5f39d397 9713
39279cc3
CM		 9714 inode->i_op = &btrfs_symlink_inode_operations;
9715 inode->i_mapping->a_ops = &btrfs_symlink_aops;
d899e052 9716 inode_set_bytes(inode, name_len);
f06becc4 9717 btrfs_i_size_write(inode, name_len);
54aa1f4d 9718 err = btrfs_update_inode(trans, root, inode);
b0d5d10f 9719 if (err) {
54aa1f4d 9720 drop_inode = 1;
b0d5d10f
CM		 9721 goto out_unlock_inode;
9722 }
9723
9724 unlock_new_inode(inode);
9725 d_instantiate(dentry, inode);
39279cc3
CM		 9726
9727out_unlock:
7ad85bb7 9728 btrfs_end_transaction(trans, root);
39279cc3
CM		 9729 if (drop_inode) {
9730 inode_dec_link_count(inode);
9731 iput(inode);
9732 }
b53d3f5d 9733 btrfs_btree_balance_dirty(root);
39279cc3 9734 return err;
b0d5d10f
CM		 9735
9736out_unlock_inode:
9737 drop_inode = 1;
9738 unlock_new_inode(inode);
9739 goto out_unlock;
39279cc3 9740}
16432985 9741
0af3d00b
JB		 9742static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9743 u64 start, u64 num_bytes, u64 min_size,
9744 loff_t actual_len, u64 *alloc_hint,
9745 struct btrfs_trans_handle *trans)
d899e052 9746{
5dc562c5
JB		 9747 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9748 struct extent_map *em;
d899e052
YZ		 9749 struct btrfs_root *root = BTRFS_I(inode)->root;
9750 struct btrfs_key ins;
d899e052 9751 u64 cur_offset = start;
55a61d1d 9752 u64 i_size;
154ea289 9753 u64 cur_bytes;
0b670dc4 9754 u64 last_alloc = (u64)-1;
d899e052 9755 int ret = 0;
0af3d00b 9756 bool own_trans = true;
d899e052 9757
0af3d00b
JB		 9758 if (trans)
9759 own_trans = false;
d899e052 9760 while (num_bytes > 0) {
0af3d00b
JB		 9761 if (own_trans) {
9762 trans = btrfs_start_transaction(root, 3);
9763 if (IS_ERR(trans)) {
9764 ret = PTR_ERR(trans);
9765 break;
9766 }
5a303d5d
YZ		 9767 }
9768
154ea289
CM		 9769 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
9770 cur_bytes = max(cur_bytes, min_size);
0b670dc4
JB		 9771 /*
9772 * If we are severely fragmented we could end up with really
9773 * small allocations, so if the allocator is returning small
9774 * chunks lets make its job easier by only searching for those
9775 * sized chunks.
9776 */
9777 cur_bytes = min(cur_bytes, last_alloc);
00361589 9778 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
e570fd27 9779 *alloc_hint, &ins, 1, 0);
5a303d5d 9780 if (ret) {
0af3d00b
JB		 9781 if (own_trans)
9782 btrfs_end_transaction(trans, root);
a22285a6 9783 break;
d899e052 9784 }
5a303d5d 9785
0b670dc4 9786 last_alloc = ins.offset;
d899e052
YZ		 9787 ret = insert_reserved_file_extent(trans, inode,
9788 cur_offset, ins.objectid,
9789 ins.offset, ins.offset,
920bbbfb 9790 ins.offset, 0, 0, 0,
d899e052 9791 BTRFS_FILE_EXTENT_PREALLOC);
79787eaa 9792 if (ret) {
857cc2fc 9793 btrfs_free_reserved_extent(root, ins.objectid,
e570fd27 9794 ins.offset, 0);
79787eaa
JM		 9795 btrfs_abort_transaction(trans, root, ret);
9796 if (own_trans)
9797 btrfs_end_transaction(trans, root);
9798 break;
9799 }
31193213 9800
a1ed835e
CM		 9801 btrfs_drop_extent_cache(inode, cur_offset,
9802 cur_offset + ins.offset -1, 0);
5a303d5d 9803
5dc562c5
JB		 9804 em = alloc_extent_map();
9805 if (!em) {
9806 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9807 &BTRFS_I(inode)->runtime_flags);
9808 goto next;
9809 }
9810
9811 em->start = cur_offset;
9812 em->orig_start = cur_offset;
9813 em->len = ins.offset;
9814 em->block_start = ins.objectid;
9815 em->block_len = ins.offset;
b4939680 9816 em->orig_block_len = ins.offset;
cc95bef6 9817 em->ram_bytes = ins.offset;
5dc562c5
JB		 9818 em->bdev = root->fs_info->fs_devices->latest_bdev;
9819 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9820 em->generation = trans->transid;
9821
9822 while (1) {
9823 write_lock(&em_tree->lock);
09a2a8f9 9824 ret = add_extent_mapping(em_tree, em, 1);
5dc562c5
JB		 9825 write_unlock(&em_tree->lock);
9826 if (ret != -EEXIST)
9827 break;
9828 btrfs_drop_extent_cache(inode, cur_offset,
9829 cur_offset + ins.offset - 1,
9830 0);
9831 }
9832 free_extent_map(em);
9833next:
d899e052
YZ		 9834 num_bytes -= ins.offset;
9835 cur_offset += ins.offset;
efa56464 9836 *alloc_hint = ins.objectid + ins.offset;
5a303d5d 9837
0c4d2d95 9838 inode_inc_iversion(inode);
d899e052 9839 inode->i_ctime = CURRENT_TIME;
6cbff00f 9840 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
d899e052 9841 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
efa56464
YZ		 9842 (actual_len > inode->i_size) &&
9843 (cur_offset > inode->i_size)) {
d1ea6a61 9844 if (cur_offset > actual_len)
55a61d1d 9845 i_size = actual_len;
d1ea6a61 9846 else
55a61d1d
JB		 9847 i_size = cur_offset;
9848 i_size_write(inode, i_size);
9849 btrfs_ordered_update_i_size(inode, i_size, NULL);
5a303d5d
YZ		 9850 }
9851
d899e052 9852 ret = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 9853
9854 if (ret) {
9855 btrfs_abort_transaction(trans, root, ret);
9856 if (own_trans)
9857 btrfs_end_transaction(trans, root);
9858 break;
9859 }
d899e052 9860
0af3d00b
JB		 9861 if (own_trans)
9862 btrfs_end_transaction(trans, root);
5a303d5d 9863 }
d899e052
YZ		 9864 return ret;
9865}
9866
0af3d00b
JB		 9867int btrfs_prealloc_file_range(struct inode *inode, int mode,
9868 u64 start, u64 num_bytes, u64 min_size,
9869 loff_t actual_len, u64 *alloc_hint)
9870{
9871 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9872 min_size, actual_len, alloc_hint,
9873 NULL);
9874}
9875
9876int btrfs_prealloc_file_range_trans(struct inode *inode,
9877 struct btrfs_trans_handle *trans, int mode,
9878 u64 start, u64 num_bytes, u64 min_size,
9879 loff_t actual_len, u64 *alloc_hint)
9880{
9881 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9882 min_size, actual_len, alloc_hint, trans);
9883}
9884
e6dcd2dc
CM		 9885static int btrfs_set_page_dirty(struct page *page)
9886{
e6dcd2dc
CM		 9887 return __set_page_dirty_nobuffers(page);
9888}
9889
10556cb2 9890static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd 9891{
b83cc969 9892 struct btrfs_root *root = BTRFS_I(inode)->root;
cb6db4e5 9893 umode_t mode = inode->i_mode;
b83cc969 9894
cb6db4e5
JM		 9895 if (mask & MAY_WRITE &&
9896 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9897 if (btrfs_root_readonly(root))
9898 return -EROFS;
9899 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9900 return -EACCES;
9901 }
2830ba7f 9902 return generic_permission(inode, mask);
fdebe2bd 9903}
39279cc3 9904
ef3b9af5
FM		 9905static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9906{
9907 struct btrfs_trans_handle *trans;
9908 struct btrfs_root *root = BTRFS_I(dir)->root;
9909 struct inode *inode = NULL;
9910 u64 objectid;
9911 u64 index;
9912 int ret = 0;
9913
9914 /*
9915 * 5 units required for adding orphan entry
9916 */
9917 trans = btrfs_start_transaction(root, 5);
9918 if (IS_ERR(trans))
9919 return PTR_ERR(trans);
9920
9921 ret = btrfs_find_free_ino(root, &objectid);
9922 if (ret)
9923 goto out;
9924
9925 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9926 btrfs_ino(dir), objectid, mode, &index);
9927 if (IS_ERR(inode)) {
9928 ret = PTR_ERR(inode);
9929 inode = NULL;
9930 goto out;
9931 }
9932
ef3b9af5
FM		 9933 inode->i_fop = &btrfs_file_operations;
9934 inode->i_op = &btrfs_file_inode_operations;
9935
9936 inode->i_mapping->a_ops = &btrfs_aops;
ef3b9af5
FM		 9937 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9938
b0d5d10f
CM		 9939 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9940 if (ret)
9941 goto out_inode;
9942
9943 ret = btrfs_update_inode(trans, root, inode);
9944 if (ret)
9945 goto out_inode;
ef3b9af5
FM		 9946 ret = btrfs_orphan_add(trans, inode);
9947 if (ret)
b0d5d10f 9948 goto out_inode;
ef3b9af5 9949
5762b5c9
FM		 9950 /*
9951 * We set number of links to 0 in btrfs_new_inode(), and here we set
9952 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9953 * through:
9954 *
9955 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9956 */
9957 set_nlink(inode, 1);
b0d5d10f 9958 unlock_new_inode(inode);
ef3b9af5
FM		 9959 d_tmpfile(dentry, inode);
9960 mark_inode_dirty(inode);
9961
9962out:
9963 btrfs_end_transaction(trans, root);
9964 if (ret)
9965 iput(inode);
9966 btrfs_balance_delayed_items(root);
9967 btrfs_btree_balance_dirty(root);
ef3b9af5 9968 return ret;
b0d5d10f
CM		 9969
9970out_inode:
9971 unlock_new_inode(inode);
9972 goto out;
9973
ef3b9af5
FM		 9974}
9975
b38ef71c
FM		 9976/* Inspired by filemap_check_errors() */
9977int btrfs_inode_check_errors(struct inode *inode)
9978{
9979 int ret = 0;
9980
9981 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
9982 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
9983 ret = -ENOSPC;
9984 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
9985 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
9986 ret = -EIO;
9987
9988 return ret;
9989}
9990
6e1d5dcc 9991static const struct inode_operations btrfs_dir_inode_operations = {
3394e160 9992 .getattr = btrfs_getattr,
39279cc3
CM		 9993 .lookup = btrfs_lookup,
9994 .create = btrfs_create,
9995 .unlink = btrfs_unlink,
9996 .link = btrfs_link,
9997 .mkdir = btrfs_mkdir,
9998 .rmdir = btrfs_rmdir,
80ace85c 9999 .rename2 = btrfs_rename2,
39279cc3
CM		 10000 .symlink = btrfs_symlink,
10001 .setattr = btrfs_setattr,
618e21d5 10002 .mknod = btrfs_mknod,
95819c05
CH		 10003 .setxattr = btrfs_setxattr,
10004 .getxattr = btrfs_getxattr,
5103e947 10005 .listxattr = btrfs_listxattr,
95819c05 10006 .removexattr = btrfs_removexattr,
fdebe2bd 10007 .permission = btrfs_permission,
4e34e719 10008 .get_acl = btrfs_get_acl,
996a710d 10009 .set_acl = btrfs_set_acl,
93fd63c2 10010 .update_time = btrfs_update_time,
ef3b9af5 10011 .tmpfile = btrfs_tmpfile,
39279cc3 10012};
6e1d5dcc 10013static const struct inode_operations btrfs_dir_ro_inode_operations = {
39279cc3 10014 .lookup = btrfs_lookup,
fdebe2bd 10015 .permission = btrfs_permission,
4e34e719 10016 .get_acl = btrfs_get_acl,
996a710d 10017 .set_acl = btrfs_set_acl,
93fd63c2 10018 .update_time = btrfs_update_time,
39279cc3 10019};
76dda93c 10020
828c0950 10021static const struct file_operations btrfs_dir_file_operations = {
39279cc3
CM		 10022 .llseek = generic_file_llseek,
10023 .read = generic_read_dir,
9cdda8d3 10024 .iterate = btrfs_real_readdir,
34287aa3 10025 .unlocked_ioctl = btrfs_ioctl,
39279cc3 10026#ifdef CONFIG_COMPAT
34287aa3 10027 .compat_ioctl = btrfs_ioctl,
39279cc3 10028#endif
6bf13c0c 10029 .release = btrfs_release_file,
e02119d5 10030 .fsync = btrfs_sync_file,
39279cc3
CM		 10031};
10032
d1310b2e 10033static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 10034 .fill_delalloc = run_delalloc_range,
065631f6 10035 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 10036 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 10037 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 10038 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 10039 .writepage_start_hook = btrfs_writepage_start_hook,
b0c68f8b
CM		 10040 .set_bit_hook = btrfs_set_bit_hook,
10041 .clear_bit_hook = btrfs_clear_bit_hook,
9ed74f2d
JB		 10042 .merge_extent_hook = btrfs_merge_extent_hook,
10043 .split_extent_hook = btrfs_split_extent_hook,
07157aac
CM		 10044};
10045
35054394
CM		 10046/*
10047 * btrfs doesn't support the bmap operation because swapfiles
10048 * use bmap to make a mapping of extents in the file. They assume
10049 * these extents won't change over the life of the file and they
10050 * use the bmap result to do IO directly to the drive.
10051 *
10052 * the btrfs bmap call would return logical addresses that aren't
10053 * suitable for IO and they also will change frequently as COW
10054 * operations happen. So, swapfile + btrfs == corruption.
10055 *
10056 * For now we're avoiding this by dropping bmap.
10057 */
7f09410b 10058static const struct address_space_operations btrfs_aops = {
39279cc3
CM		 10059 .readpage = btrfs_readpage,
10060 .writepage = btrfs_writepage,
b293f02e 10061 .writepages = btrfs_writepages,
3ab2fb5a 10062 .readpages = btrfs_readpages,
16432985 10063 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 10064 .invalidatepage = btrfs_invalidatepage,
10065 .releasepage = btrfs_releasepage,
e6dcd2dc 10066 .set_page_dirty = btrfs_set_page_dirty,
465fdd97 10067 .error_remove_page = generic_error_remove_page,
39279cc3
CM		 10068};
10069
7f09410b 10070static const struct address_space_operations btrfs_symlink_aops = {
39279cc3
CM		 10071 .readpage = btrfs_readpage,
10072 .writepage = btrfs_writepage,
2bf5a725
CM		 10073 .invalidatepage = btrfs_invalidatepage,
10074 .releasepage = btrfs_releasepage,
39279cc3
CM		 10075};
10076
6e1d5dcc 10077static const struct inode_operations btrfs_file_inode_operations = {
39279cc3
CM		 10078 .getattr = btrfs_getattr,
10079 .setattr = btrfs_setattr,
95819c05
CH		 10080 .setxattr = btrfs_setxattr,
10081 .getxattr = btrfs_getxattr,
5103e947 10082 .listxattr = btrfs_listxattr,
95819c05 10083 .removexattr = btrfs_removexattr,
fdebe2bd 10084 .permission = btrfs_permission,
1506fcc8 10085 .fiemap = btrfs_fiemap,
4e34e719 10086 .get_acl = btrfs_get_acl,
996a710d 10087 .set_acl = btrfs_set_acl,
e41f941a 10088 .update_time = btrfs_update_time,
39279cc3 10089};
6e1d5dcc 10090static const struct inode_operations btrfs_special_inode_operations = {
618e21d5
JB		 10091 .getattr = btrfs_getattr,
10092 .setattr = btrfs_setattr,
fdebe2bd 10093 .permission = btrfs_permission,
95819c05
CH		 10094 .setxattr = btrfs_setxattr,
10095 .getxattr = btrfs_getxattr,
33268eaf 10096 .listxattr = btrfs_listxattr,
95819c05 10097 .removexattr = btrfs_removexattr,
4e34e719 10098 .get_acl = btrfs_get_acl,
996a710d 10099 .set_acl = btrfs_set_acl,
e41f941a 10100 .update_time = btrfs_update_time,
618e21d5 10101};
6e1d5dcc 10102static const struct inode_operations btrfs_symlink_inode_operations = {
39279cc3
CM		 10103 .readlink = generic_readlink,
10104 .follow_link = page_follow_link_light,
10105 .put_link = page_put_link,
f209561a 10106 .getattr = btrfs_getattr,
22c44fe6 10107 .setattr = btrfs_setattr,
fdebe2bd 10108 .permission = btrfs_permission,
0279b4cd
JO		 10109 .setxattr = btrfs_setxattr,
10110 .getxattr = btrfs_getxattr,
10111 .listxattr = btrfs_listxattr,
10112 .removexattr = btrfs_removexattr,
e41f941a 10113 .update_time = btrfs_update_time,
39279cc3 10114};
76dda93c 10115
82d339d9 10116const struct dentry_operations btrfs_dentry_operations = {
76dda93c 10117 .d_delete = btrfs_dentry_delete,
b4aff1f8 10118 .d_release = btrfs_dentry_release,
76dda93c 10119};
Linux 6.x block layer and io_uring tree(s)