projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
btrfs: drop null testing before destroy functions
[deliverable/linux.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
f28a4928
FM		 69struct btrfs_dio_data {
70 u64 outstanding_extents;
71 u64 reserve;
72 u64 unsubmitted_oe_range_start;
73 u64 unsubmitted_oe_range_end;
74};
75
6e1d5dcc
AD		 76static const struct inode_operations btrfs_dir_inode_operations;
77static const struct inode_operations btrfs_symlink_inode_operations;
78static const struct inode_operations btrfs_dir_ro_inode_operations;
79static const struct inode_operations btrfs_special_inode_operations;
80static const struct inode_operations btrfs_file_inode_operations;
7f09410b
AD		 81static const struct address_space_operations btrfs_aops;
82static const struct address_space_operations btrfs_symlink_aops;
828c0950 83static const struct file_operations btrfs_dir_file_operations;
20e5506b 84static const struct extent_io_ops btrfs_extent_io_ops;
39279cc3
CM		 85
86static struct kmem_cache *btrfs_inode_cachep;
87struct kmem_cache *btrfs_trans_handle_cachep;
88struct kmem_cache *btrfs_transaction_cachep;
39279cc3 89struct kmem_cache *btrfs_path_cachep;
dc89e982 90struct kmem_cache *btrfs_free_space_cachep;
39279cc3
CM		 91
92#define S_SHIFT 12
4d4ab6d6 93static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
39279cc3
CM		 94 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
95 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
96 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
97 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
98 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
99 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
100 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
101};
102
3972f260 103static int btrfs_setsize(struct inode *inode, struct iattr *attr);
a41ad394 104static int btrfs_truncate(struct inode *inode);
5fd02043 105static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
771ed689
CM		 106static noinline int cow_file_range(struct inode *inode,
107 struct page *locked_page,
108 u64 start, u64 end, int *page_started,
109 unsigned long *nr_written, int unlock);
70c8a91c
JB		 110static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
111 u64 len, u64 orig_start,
112 u64 block_start, u64 block_len,
cc95bef6
JB		 113 u64 orig_block_len, u64 ram_bytes,
114 int type);
7b128766 115
48a3b636 116static int btrfs_dirty_inode(struct inode *inode);
7b128766 117
6a3891c5
JB		 118#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
119void btrfs_test_inode_set_ops(struct inode *inode)
120{
121 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
122}
123#endif
124
f34f57a3 125static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
2a7dba39
EP		 126 struct inode *inode, struct inode *dir,
127 const struct qstr *qstr)
0279b4cd
JO		 128{
129 int err;
130
f34f57a3 131 err = btrfs_init_acl(trans, inode, dir);
0279b4cd 132 if (!err)
2a7dba39 133 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
0279b4cd
JO		 134 return err;
135}
136
c8b97818
CM		 137/*
138 * this does all the hard work for inserting an inline extent into
139 * the btree. The caller should have done a btrfs_drop_extents so that
140 * no overlapping inline items exist in the btree
141 */
40f76580 142static int insert_inline_extent(struct btrfs_trans_handle *trans,
1acae57b 143 struct btrfs_path *path, int extent_inserted,
c8b97818
CM		 144 struct btrfs_root *root, struct inode *inode,
145 u64 start, size_t size, size_t compressed_size,
fe3f566c 146 int compress_type,
c8b97818
CM		 147 struct page **compressed_pages)
148{
c8b97818
CM		 149 struct extent_buffer *leaf;
150 struct page *page = NULL;
151 char *kaddr;
152 unsigned long ptr;
153 struct btrfs_file_extent_item *ei;
154 int err = 0;
155 int ret;
156 size_t cur_size = size;
c8b97818 157 unsigned long offset;
c8b97818 158
fe3f566c 159 if (compressed_size && compressed_pages)
c8b97818 160 cur_size = compressed_size;
c8b97818 161
1acae57b 162 inode_add_bytes(inode, size);
c8b97818 163
1acae57b
FDBM		 164 if (!extent_inserted) {
165 struct btrfs_key key;
166 size_t datasize;
c8b97818 167
1acae57b
FDBM		 168 key.objectid = btrfs_ino(inode);
169 key.offset = start;
962a298f 170 key.type = BTRFS_EXTENT_DATA_KEY;
c8b97818 171
1acae57b
FDBM		 172 datasize = btrfs_file_extent_calc_inline_size(cur_size);
173 path->leave_spinning = 1;
174 ret = btrfs_insert_empty_item(trans, root, path, &key,
175 datasize);
176 if (ret) {
177 err = ret;
178 goto fail;
179 }
c8b97818
CM		 180 }
181 leaf = path->nodes[0];
182 ei = btrfs_item_ptr(leaf, path->slots[0],
183 struct btrfs_file_extent_item);
184 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
185 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
186 btrfs_set_file_extent_encryption(leaf, ei, 0);
187 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
188 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
189 ptr = btrfs_file_extent_inline_start(ei);
190
261507a0 191 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818
CM		 192 struct page *cpage;
193 int i = 0;
d397712b 194 while (compressed_size > 0) {
c8b97818 195 cpage = compressed_pages[i];
5b050f04 196 cur_size = min_t(unsigned long, compressed_size,
c8b97818
CM		 197 PAGE_CACHE_SIZE);
198
7ac687d9 199 kaddr = kmap_atomic(cpage);
c8b97818 200 write_extent_buffer(leaf, kaddr, ptr, cur_size);
7ac687d9 201 kunmap_atomic(kaddr);
c8b97818
CM		 202
203 i++;
204 ptr += cur_size;
205 compressed_size -= cur_size;
206 }
207 btrfs_set_file_extent_compression(leaf, ei,
261507a0 208 compress_type);
c8b97818
CM		 209 } else {
210 page = find_get_page(inode->i_mapping,
211 start >> PAGE_CACHE_SHIFT);
212 btrfs_set_file_extent_compression(leaf, ei, 0);
7ac687d9 213 kaddr = kmap_atomic(page);
c8b97818
CM		 214 offset = start & (PAGE_CACHE_SIZE - 1);
215 write_extent_buffer(leaf, kaddr + offset, ptr, size);
7ac687d9 216 kunmap_atomic(kaddr);
c8b97818
CM		 217 page_cache_release(page);
218 }
219 btrfs_mark_buffer_dirty(leaf);
1acae57b 220 btrfs_release_path(path);
c8b97818 221
c2167754
YZ		 222 /*
223 * we're an inline extent, so nobody can
224 * extend the file past i_size without locking
225 * a page we already have locked.
226 *
227 * We must do any isize and inode updates
228 * before we unlock the pages. Otherwise we
229 * could end up racing with unlink.
230 */
c8b97818 231 BTRFS_I(inode)->disk_i_size = inode->i_size;
79787eaa 232 ret = btrfs_update_inode(trans, root, inode);
c2167754 233
79787eaa 234 return ret;
c8b97818 235fail:
c8b97818
CM		 236 return err;
237}
238
239
240/*
241 * conditionally insert an inline extent into the file. This
242 * does the checks required to make sure the data is small enough
243 * to fit as an inline extent.
244 */
00361589
JB		 245static noinline int cow_file_range_inline(struct btrfs_root *root,
246 struct inode *inode, u64 start,
247 u64 end, size_t compressed_size,
248 int compress_type,
249 struct page **compressed_pages)
c8b97818 250{
00361589 251 struct btrfs_trans_handle *trans;
c8b97818
CM		 252 u64 isize = i_size_read(inode);
253 u64 actual_end = min(end + 1, isize);
254 u64 inline_len = actual_end - start;
fda2832f 255 u64 aligned_end = ALIGN(end, root->sectorsize);
c8b97818
CM		 256 u64 data_len = inline_len;
257 int ret;
1acae57b
FDBM		 258 struct btrfs_path *path;
259 int extent_inserted = 0;
260 u32 extent_item_size;
c8b97818
CM		 261
262 if (compressed_size)
263 data_len = compressed_size;
264
265 if (start > 0 ||
354877be
WS		 266 actual_end > PAGE_CACHE_SIZE ||
267 data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
c8b97818
CM		 268 (!compressed_size &&
269 (actual_end & (root->sectorsize - 1)) == 0) ||
270 end + 1 < isize ||
271 data_len > root->fs_info->max_inline) {
272 return 1;
273 }
274
1acae57b
FDBM		 275 path = btrfs_alloc_path();
276 if (!path)
277 return -ENOMEM;
278
00361589 279 trans = btrfs_join_transaction(root);
1acae57b
FDBM		 280 if (IS_ERR(trans)) {
281 btrfs_free_path(path);
00361589 282 return PTR_ERR(trans);
1acae57b 283 }
00361589
JB		 284 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
285
1acae57b
FDBM		 286 if (compressed_size && compressed_pages)
287 extent_item_size = btrfs_file_extent_calc_inline_size(
288 compressed_size);
289 else
290 extent_item_size = btrfs_file_extent_calc_inline_size(
291 inline_len);
292
293 ret = __btrfs_drop_extents(trans, root, inode, path,
294 start, aligned_end, NULL,
295 1, 1, extent_item_size, &extent_inserted);
00361589
JB		 296 if (ret) {
297 btrfs_abort_transaction(trans, root, ret);
298 goto out;
299 }
c8b97818
CM		 300
301 if (isize > actual_end)
302 inline_len = min_t(u64, isize, actual_end);
1acae57b
FDBM		 303 ret = insert_inline_extent(trans, path, extent_inserted,
304 root, inode, start,
c8b97818 305 inline_len, compressed_size,
fe3f566c 306 compress_type, compressed_pages);
2adcac1a 307 if (ret && ret != -ENOSPC) {
79787eaa 308 btrfs_abort_transaction(trans, root, ret);
00361589 309 goto out;
2adcac1a 310 } else if (ret == -ENOSPC) {
00361589
JB		 311 ret = 1;
312 goto out;
79787eaa 313 }
2adcac1a 314
bdc20e67 315 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
0ca1f7ce 316 btrfs_delalloc_release_metadata(inode, end + 1 - start);
a1ed835e 317 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
00361589 318out:
94ed938a
QW		 319 /*
320 * Don't forget to free the reserved space, as for inlined extent
321 * it won't count as data extent, free them directly here.
322 * And at reserve time, it's always aligned to page size, so
323 * just free one page here.
324 */
325 btrfs_qgroup_free_data(inode, 0, PAGE_CACHE_SIZE);
1acae57b 326 btrfs_free_path(path);
00361589
JB		 327 btrfs_end_transaction(trans, root);
328 return ret;
c8b97818
CM		 329}
330
771ed689
CM		 331struct async_extent {
332 u64 start;
333 u64 ram_size;
334 u64 compressed_size;
335 struct page **pages;
336 unsigned long nr_pages;
261507a0 337 int compress_type;
771ed689
CM		 338 struct list_head list;
339};
340
341struct async_cow {
342 struct inode *inode;
343 struct btrfs_root *root;
344 struct page *locked_page;
345 u64 start;
346 u64 end;
347 struct list_head extents;
348 struct btrfs_work work;
349};
350
351static noinline int add_async_extent(struct async_cow *cow,
352 u64 start, u64 ram_size,
353 u64 compressed_size,
354 struct page **pages,
261507a0
LZ		 355 unsigned long nr_pages,
356 int compress_type)
771ed689
CM		 357{
358 struct async_extent *async_extent;
359
360 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
79787eaa 361 BUG_ON(!async_extent); /* -ENOMEM */
771ed689
CM		 362 async_extent->start = start;
363 async_extent->ram_size = ram_size;
364 async_extent->compressed_size = compressed_size;
365 async_extent->pages = pages;
366 async_extent->nr_pages = nr_pages;
261507a0 367 async_extent->compress_type = compress_type;
771ed689
CM		 368 list_add_tail(&async_extent->list, &cow->extents);
369 return 0;
370}
371
f79707b0
WS		 372static inline int inode_need_compress(struct inode *inode)
373{
374 struct btrfs_root *root = BTRFS_I(inode)->root;
375
376 /* force compress */
377 if (btrfs_test_opt(root, FORCE_COMPRESS))
378 return 1;
379 /* bad compression ratios */
380 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
381 return 0;
382 if (btrfs_test_opt(root, COMPRESS) ||
383 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
384 BTRFS_I(inode)->force_compress)
385 return 1;
386 return 0;
387}
388
d352ac68 389/*
771ed689
CM		 390 * we create compressed extents in two phases. The first
391 * phase compresses a range of pages that have already been
392 * locked (both pages and state bits are locked).
c8b97818 393 *
771ed689
CM		 394 * This is done inside an ordered work queue, and the compression
395 * is spread across many cpus. The actual IO submission is step
396 * two, and the ordered work queue takes care of making sure that
397 * happens in the same order things were put onto the queue by
398 * writepages and friends.
c8b97818 399 *
771ed689
CM		 400 * If this code finds it can't get good compression, it puts an
401 * entry onto the work queue to write the uncompressed bytes. This
402 * makes sure that both compressed inodes and uncompressed inodes
b2570314
AB		 403 * are written in the same order that the flusher thread sent them
404 * down.
d352ac68 405 */
c44f649e 406static noinline void compress_file_range(struct inode *inode,
771ed689
CM		 407 struct page *locked_page,
408 u64 start, u64 end,
409 struct async_cow *async_cow,
410 int *num_added)
b888db2b
CM		 411{
412 struct btrfs_root *root = BTRFS_I(inode)->root;
db94535d 413 u64 num_bytes;
db94535d 414 u64 blocksize = root->sectorsize;
c8b97818 415 u64 actual_end;
42dc7bab 416 u64 isize = i_size_read(inode);
e6dcd2dc 417 int ret = 0;
c8b97818
CM		 418 struct page **pages = NULL;
419 unsigned long nr_pages;
420 unsigned long nr_pages_ret = 0;
421 unsigned long total_compressed = 0;
422 unsigned long total_in = 0;
ee22184b
BL		 423 unsigned long max_compressed = SZ_128K;
424 unsigned long max_uncompressed = SZ_128K;
c8b97818
CM		 425 int i;
426 int will_compress;
261507a0 427 int compress_type = root->fs_info->compress_type;
4adaa611 428 int redirty = 0;
b888db2b 429
4cb13e5d 430 /* if this is a small write inside eof, kick off a defrag */
ee22184b 431 if ((end - start + 1) < SZ_16K &&
4cb13e5d 432 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
4cb5300b
CM		 433 btrfs_add_inode_defrag(NULL, inode);
434
42dc7bab 435 actual_end = min_t(u64, isize, end + 1);
c8b97818
CM		 436again:
437 will_compress = 0;
438 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
ee22184b 439 nr_pages = min_t(unsigned long, nr_pages, SZ_128K / PAGE_CACHE_SIZE);
be20aa9d 440
f03d9301
CM		 441 /*
442 * we don't want to send crud past the end of i_size through
443 * compression, that's just a waste of CPU time. So, if the
444 * end of the file is before the start of our current
445 * requested range of bytes, we bail out to the uncompressed
446 * cleanup code that can deal with all of this.
447 *
448 * It isn't really the fastest way to fix things, but this is a
449 * very uncommon corner.
450 */
451 if (actual_end <= start)
452 goto cleanup_and_bail_uncompressed;
453
c8b97818
CM		 454 total_compressed = actual_end - start;
455
4bcbb332
SW		 456 /*
457 * skip compression for a small file range(<=blocksize) that
458 * isn't an inline extent, since it dosen't save disk space at all.
459 */
460 if (total_compressed <= blocksize &&
461 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
462 goto cleanup_and_bail_uncompressed;
463
c8b97818
CM		 464 /* we want to make sure that amount of ram required to uncompress
465 * an extent is reasonable, so we limit the total size in ram
771ed689
CM		 466 * of a compressed extent to 128k. This is a crucial number
467 * because it also controls how easily we can spread reads across
468 * cpus for decompression.
469 *
470 * We also want to make sure the amount of IO required to do
471 * a random read is reasonably small, so we limit the size of
472 * a compressed extent to 128k.
c8b97818
CM		 473 */
474 total_compressed = min(total_compressed, max_uncompressed);
fda2832f 475 num_bytes = ALIGN(end - start + 1, blocksize);
be20aa9d 476 num_bytes = max(blocksize, num_bytes);
c8b97818
CM		 477 total_in = 0;
478 ret = 0;
db94535d 479
771ed689
CM		 480 /*
481 * we do compression for mount -o compress and when the
482 * inode has not been flagged as nocompress. This flag can
483 * change at any time if we discover bad compression ratios.
c8b97818 484 */
f79707b0 485 if (inode_need_compress(inode)) {
c8b97818 486 WARN_ON(pages);
31e818fe 487 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
560f7d75
LZ		 488 if (!pages) {
489 /* just bail out to the uncompressed code */
490 goto cont;
491 }
c8b97818 492
261507a0
LZ		 493 if (BTRFS_I(inode)->force_compress)
494 compress_type = BTRFS_I(inode)->force_compress;
495
4adaa611
CM		 496 /*
497 * we need to call clear_page_dirty_for_io on each
498 * page in the range. Otherwise applications with the file
499 * mmap'd can wander in and change the page contents while
500 * we are compressing them.
501 *
502 * If the compression fails for any reason, we set the pages
503 * dirty again later on.
504 */
505 extent_range_clear_dirty_for_io(inode, start, end);
506 redirty = 1;
261507a0
LZ		 507 ret = btrfs_compress_pages(compress_type,
508 inode->i_mapping, start,
509 total_compressed, pages,
510 nr_pages, &nr_pages_ret,
511 &total_in,
512 &total_compressed,
513 max_compressed);
c8b97818
CM		 514
515 if (!ret) {
516 unsigned long offset = total_compressed &
517 (PAGE_CACHE_SIZE - 1);
518 struct page *page = pages[nr_pages_ret - 1];
519 char *kaddr;
520
521 /* zero the tail end of the last page, we might be
522 * sending it down to disk
523 */
524 if (offset) {
7ac687d9 525 kaddr = kmap_atomic(page);
c8b97818
CM		 526 memset(kaddr + offset, 0,
527 PAGE_CACHE_SIZE - offset);
7ac687d9 528 kunmap_atomic(kaddr);
c8b97818
CM		 529 }
530 will_compress = 1;
531 }
532 }
560f7d75 533cont:
c8b97818
CM		 534 if (start == 0) {
535 /* lets try to make an inline extent */
771ed689 536 if (ret || total_in < (actual_end - start)) {
c8b97818 537 /* we didn't compress the entire range, try
771ed689 538 * to make an uncompressed inline extent.
c8b97818 539 */
00361589
JB		 540 ret = cow_file_range_inline(root, inode, start, end,
541 0, 0, NULL);
c8b97818 542 } else {
771ed689 543 /* try making a compressed inline extent */
00361589 544 ret = cow_file_range_inline(root, inode, start, end,
fe3f566c
LZ		 545 total_compressed,
546 compress_type, pages);
c8b97818 547 }
79787eaa 548 if (ret <= 0) {
151a41bc
JB		 549 unsigned long clear_flags = EXTENT_DELALLOC |
550 EXTENT_DEFRAG;
e6eb4314
FM		 551 unsigned long page_error_op;
552
151a41bc 553 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
e6eb4314 554 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
151a41bc 555
771ed689 556 /*
79787eaa
JM		 557 * inline extent creation worked or returned error,
558 * we don't need to create any more async work items.
559 * Unlock and free up our temp pages.
771ed689 560 */
c2790a2e 561 extent_clear_unlock_delalloc(inode, start, end, NULL,
151a41bc 562 clear_flags, PAGE_UNLOCK |
c2790a2e
JB		 563 PAGE_CLEAR_DIRTY |
564 PAGE_SET_WRITEBACK |
e6eb4314 565 page_error_op |
c2790a2e 566 PAGE_END_WRITEBACK);
c8b97818
CM		 567 goto free_pages_out;
568 }
569 }
570
571 if (will_compress) {
572 /*
573 * we aren't doing an inline extent round the compressed size
574 * up to a block size boundary so the allocator does sane
575 * things
576 */
fda2832f 577 total_compressed = ALIGN(total_compressed, blocksize);
c8b97818
CM		 578
579 /*
580 * one last check to make sure the compression is really a
581 * win, compare the page count read with the blocks on disk
582 */
fda2832f 583 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
c8b97818
CM		 584 if (total_compressed >= total_in) {
585 will_compress = 0;
586 } else {
c8b97818
CM		 587 num_bytes = total_in;
588 }
589 }
590 if (!will_compress && pages) {
591 /*
592 * the compression code ran but failed to make things smaller,
593 * free any pages it allocated and our page pointer array
594 */
595 for (i = 0; i < nr_pages_ret; i++) {
70b99e69 596 WARN_ON(pages[i]->mapping);
c8b97818
CM		 597 page_cache_release(pages[i]);
598 }
599 kfree(pages);
600 pages = NULL;
601 total_compressed = 0;
602 nr_pages_ret = 0;
603
604 /* flag the file so we don't compress in the future */
1e701a32
CM		 605 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
606 !(BTRFS_I(inode)->force_compress)) {
a555f810 607 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
1e701a32 608 }
c8b97818 609 }
771ed689
CM		 610 if (will_compress) {
611 *num_added += 1;
c8b97818 612
771ed689
CM		 613 /* the async work queues will take care of doing actual
614 * allocation on disk for these compressed pages,
615 * and will submit them to the elevator.
616 */
617 add_async_extent(async_cow, start, num_bytes,
261507a0
LZ		 618 total_compressed, pages, nr_pages_ret,
619 compress_type);
179e29e4 620
24ae6365 621 if (start + num_bytes < end) {
771ed689
CM		 622 start += num_bytes;
623 pages = NULL;
624 cond_resched();
625 goto again;
626 }
627 } else {
f03d9301 628cleanup_and_bail_uncompressed:
771ed689
CM		 629 /*
630 * No compression, but we still need to write the pages in
631 * the file we've been given so far. redirty the locked
632 * page if it corresponds to our extent and set things up
633 * for the async work queue to run cow_file_range to do
634 * the normal delalloc dance
635 */
636 if (page_offset(locked_page) >= start &&
637 page_offset(locked_page) <= end) {
638 __set_page_dirty_nobuffers(locked_page);
639 /* unlocked later on in the async handlers */
640 }
4adaa611
CM		 641 if (redirty)
642 extent_range_redirty_for_io(inode, start, end);
261507a0
LZ		 643 add_async_extent(async_cow, start, end - start + 1,
644 0, NULL, 0, BTRFS_COMPRESS_NONE);
771ed689
CM		 645 *num_added += 1;
646 }
3b951516 647
c44f649e 648 return;
771ed689
CM		 649
650free_pages_out:
651 for (i = 0; i < nr_pages_ret; i++) {
652 WARN_ON(pages[i]->mapping);
653 page_cache_release(pages[i]);
654 }
d397712b 655 kfree(pages);
771ed689 656}
771ed689 657
40ae837b
FM		 658static void free_async_extent_pages(struct async_extent *async_extent)
659{
660 int i;
661
662 if (!async_extent->pages)
663 return;
664
665 for (i = 0; i < async_extent->nr_pages; i++) {
666 WARN_ON(async_extent->pages[i]->mapping);
667 page_cache_release(async_extent->pages[i]);
668 }
669 kfree(async_extent->pages);
670 async_extent->nr_pages = 0;
671 async_extent->pages = NULL;
771ed689
CM		 672}
673
674/*
675 * phase two of compressed writeback. This is the ordered portion
676 * of the code, which only gets called in the order the work was
677 * queued. We walk all the async extents created by compress_file_range
678 * and send them down to the disk.
679 */
dec8f175 680static noinline void submit_compressed_extents(struct inode *inode,
771ed689
CM		 681 struct async_cow *async_cow)
682{
683 struct async_extent *async_extent;
684 u64 alloc_hint = 0;
771ed689
CM		 685 struct btrfs_key ins;
686 struct extent_map *em;
687 struct btrfs_root *root = BTRFS_I(inode)->root;
688 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
689 struct extent_io_tree *io_tree;
f5a84ee3 690 int ret = 0;
771ed689 691
3e04e7f1 692again:
d397712b 693 while (!list_empty(&async_cow->extents)) {
771ed689
CM		 694 async_extent = list_entry(async_cow->extents.next,
695 struct async_extent, list);
696 list_del(&async_extent->list);
c8b97818 697
771ed689
CM		 698 io_tree = &BTRFS_I(inode)->io_tree;
699
f5a84ee3 700retry:
771ed689
CM		 701 /* did the compression code fall back to uncompressed IO? */
702 if (!async_extent->pages) {
703 int page_started = 0;
704 unsigned long nr_written = 0;
705
706 lock_extent(io_tree, async_extent->start,
2ac55d41 707 async_extent->start +
d0082371 708 async_extent->ram_size - 1);
771ed689
CM		 709
710 /* allocate blocks */
f5a84ee3
JB		 711 ret = cow_file_range(inode, async_cow->locked_page,
712 async_extent->start,
713 async_extent->start +
714 async_extent->ram_size - 1,
715 &page_started, &nr_written, 0);
771ed689 716
79787eaa
JM		 717 /* JDM XXX */
718
771ed689
CM		 719 /*
720 * if page_started, cow_file_range inserted an
721 * inline extent and took care of all the unlocking
722 * and IO for us. Otherwise, we need to submit
723 * all those pages down to the drive.
724 */
f5a84ee3 725 if (!page_started && !ret)
771ed689
CM		 726 extent_write_locked_range(io_tree,
727 inode, async_extent->start,
d397712b 728 async_extent->start +
771ed689
CM		 729 async_extent->ram_size - 1,
730 btrfs_get_extent,
731 WB_SYNC_ALL);
3e04e7f1
JB		 732 else if (ret)
733 unlock_page(async_cow->locked_page);
771ed689
CM		 734 kfree(async_extent);
735 cond_resched();
736 continue;
737 }
738
739 lock_extent(io_tree, async_extent->start,
d0082371 740 async_extent->start + async_extent->ram_size - 1);
771ed689 741
00361589 742 ret = btrfs_reserve_extent(root,
771ed689
CM		 743 async_extent->compressed_size,
744 async_extent->compressed_size,
e570fd27 745 0, alloc_hint, &ins, 1, 1);
f5a84ee3 746 if (ret) {
40ae837b 747 free_async_extent_pages(async_extent);
3e04e7f1 748
fdf8e2ea
JB		 749 if (ret == -ENOSPC) {
750 unlock_extent(io_tree, async_extent->start,
751 async_extent->start +
752 async_extent->ram_size - 1);
ce62003f
LB		 753
754 /*
755 * we need to redirty the pages if we decide to
756 * fallback to uncompressed IO, otherwise we
757 * will not submit these pages down to lower
758 * layers.
759 */
760 extent_range_redirty_for_io(inode,
761 async_extent->start,
762 async_extent->start +
763 async_extent->ram_size - 1);
764
79787eaa 765 goto retry;
fdf8e2ea 766 }
3e04e7f1 767 goto out_free;
f5a84ee3 768 }
c2167754
YZ		 769 /*
770 * here we're doing allocation and writeback of the
771 * compressed pages
772 */
773 btrfs_drop_extent_cache(inode, async_extent->start,
774 async_extent->start +
775 async_extent->ram_size - 1, 0);
776
172ddd60 777 em = alloc_extent_map();
b9aa55be
LB		 778 if (!em) {
779 ret = -ENOMEM;
3e04e7f1 780 goto out_free_reserve;
b9aa55be 781 }
771ed689
CM		 782 em->start = async_extent->start;
783 em->len = async_extent->ram_size;
445a6944 784 em->orig_start = em->start;
2ab28f32
JB		 785 em->mod_start = em->start;
786 em->mod_len = em->len;
c8b97818 787
771ed689
CM		 788 em->block_start = ins.objectid;
789 em->block_len = ins.offset;
b4939680 790 em->orig_block_len = ins.offset;
cc95bef6 791 em->ram_bytes = async_extent->ram_size;
771ed689 792 em->bdev = root->fs_info->fs_devices->latest_bdev;
261507a0 793 em->compress_type = async_extent->compress_type;
771ed689
CM		 794 set_bit(EXTENT_FLAG_PINNED, &em->flags);
795 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
70c8a91c 796 em->generation = -1;
771ed689 797
d397712b 798 while (1) {
890871be 799 write_lock(&em_tree->lock);
09a2a8f9 800 ret = add_extent_mapping(em_tree, em, 1);
890871be 801 write_unlock(&em_tree->lock);
771ed689
CM		 802 if (ret != -EEXIST) {
803 free_extent_map(em);
804 break;
805 }
806 btrfs_drop_extent_cache(inode, async_extent->start,
807 async_extent->start +
808 async_extent->ram_size - 1, 0);
809 }
810
3e04e7f1
JB		 811 if (ret)
812 goto out_free_reserve;
813
261507a0
LZ		 814 ret = btrfs_add_ordered_extent_compress(inode,
815 async_extent->start,
816 ins.objectid,
817 async_extent->ram_size,
818 ins.offset,
819 BTRFS_ORDERED_COMPRESSED,
820 async_extent->compress_type);
d9f85963
FM		 821 if (ret) {
822 btrfs_drop_extent_cache(inode, async_extent->start,
823 async_extent->start +
824 async_extent->ram_size - 1, 0);
3e04e7f1 825 goto out_free_reserve;
d9f85963 826 }
771ed689 827
771ed689
CM		 828 /*
829 * clear dirty, set writeback and unlock the pages.
830 */
c2790a2e 831 extent_clear_unlock_delalloc(inode, async_extent->start,
a791e35e
CM		 832 async_extent->start +
833 async_extent->ram_size - 1,
151a41bc
JB		 834 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
835 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
c2790a2e 836 PAGE_SET_WRITEBACK);
771ed689 837 ret = btrfs_submit_compressed_write(inode,
d397712b
CM		 838 async_extent->start,
839 async_extent->ram_size,
840 ins.objectid,
841 ins.offset, async_extent->pages,
842 async_extent->nr_pages);
fce2a4e6
FM		 843 if (ret) {
844 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
845 struct page *p = async_extent->pages[0];
846 const u64 start = async_extent->start;
847 const u64 end = start + async_extent->ram_size - 1;
848
849 p->mapping = inode->i_mapping;
850 tree->ops->writepage_end_io_hook(p, start, end,
851 NULL, 0);
852 p->mapping = NULL;
853 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
854 PAGE_END_WRITEBACK |
855 PAGE_SET_ERROR);
40ae837b 856 free_async_extent_pages(async_extent);
fce2a4e6 857 }
771ed689
CM		 858 alloc_hint = ins.objectid + ins.offset;
859 kfree(async_extent);
860 cond_resched();
861 }
dec8f175 862 return;
3e04e7f1 863out_free_reserve:
e570fd27 864 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
79787eaa 865out_free:
c2790a2e 866 extent_clear_unlock_delalloc(inode, async_extent->start,
3e04e7f1
JB		 867 async_extent->start +
868 async_extent->ram_size - 1,
c2790a2e 869 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc
JB		 870 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
871 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
704de49d
FM		 872 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
873 PAGE_SET_ERROR);
40ae837b 874 free_async_extent_pages(async_extent);
79787eaa 875 kfree(async_extent);
3e04e7f1 876 goto again;
771ed689
CM		 877}
878
4b46fce2
JB		 879static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
880 u64 num_bytes)
881{
882 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
883 struct extent_map *em;
884 u64 alloc_hint = 0;
885
886 read_lock(&em_tree->lock);
887 em = search_extent_mapping(em_tree, start, num_bytes);
888 if (em) {
889 /*
890 * if block start isn't an actual block number then find the
891 * first block in this inode and use that as a hint. If that
892 * block is also bogus then just don't worry about it.
893 */
894 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
895 free_extent_map(em);
896 em = search_extent_mapping(em_tree, 0, 0);
897 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
898 alloc_hint = em->block_start;
899 if (em)
900 free_extent_map(em);
901 } else {
902 alloc_hint = em->block_start;
903 free_extent_map(em);
904 }
905 }
906 read_unlock(&em_tree->lock);
907
908 return alloc_hint;
909}
910
771ed689
CM		 911/*
912 * when extent_io.c finds a delayed allocation range in the file,
913 * the call backs end up in this code. The basic idea is to
914 * allocate extents on disk for the range, and create ordered data structs
915 * in ram to track those extents.
916 *
917 * locked_page is the page that writepage had locked already. We use
918 * it to make sure we don't do extra locks or unlocks.
919 *
920 * *page_started is set to one if we unlock locked_page and do everything
921 * required to start IO on it. It may be clean and already done with
922 * IO when we return.
923 */
00361589
JB		 924static noinline int cow_file_range(struct inode *inode,
925 struct page *locked_page,
926 u64 start, u64 end, int *page_started,
927 unsigned long *nr_written,
928 int unlock)
771ed689 929{
00361589 930 struct btrfs_root *root = BTRFS_I(inode)->root;
771ed689
CM		 931 u64 alloc_hint = 0;
932 u64 num_bytes;
933 unsigned long ram_size;
934 u64 disk_num_bytes;
935 u64 cur_alloc_size;
936 u64 blocksize = root->sectorsize;
771ed689
CM		 937 struct btrfs_key ins;
938 struct extent_map *em;
939 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
940 int ret = 0;
941
02ecd2c2
JB		 942 if (btrfs_is_free_space_inode(inode)) {
943 WARN_ON_ONCE(1);
29bce2f3
JB		 944 ret = -EINVAL;
945 goto out_unlock;
02ecd2c2 946 }
771ed689 947
fda2832f 948 num_bytes = ALIGN(end - start + 1, blocksize);
771ed689
CM		 949 num_bytes = max(blocksize, num_bytes);
950 disk_num_bytes = num_bytes;
771ed689 951
4cb5300b 952 /* if this is a small write inside eof, kick off defrag */
ee22184b 953 if (num_bytes < SZ_64K &&
4cb13e5d 954 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
00361589 955 btrfs_add_inode_defrag(NULL, inode);
4cb5300b 956
771ed689
CM		 957 if (start == 0) {
958 /* lets try to make an inline extent */
00361589
JB		 959 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
960 NULL);
771ed689 961 if (ret == 0) {
c2790a2e
JB		 962 extent_clear_unlock_delalloc(inode, start, end, NULL,
963 EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc 964 EXTENT_DEFRAG, PAGE_UNLOCK |
c2790a2e
JB		 965 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
966 PAGE_END_WRITEBACK);
c2167754 967
771ed689
CM		 968 *nr_written = *nr_written +
969 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
970 *page_started = 1;
771ed689 971 goto out;
79787eaa 972 } else if (ret < 0) {
79787eaa 973 goto out_unlock;
771ed689
CM		 974 }
975 }
976
977 BUG_ON(disk_num_bytes >
6c41761f 978 btrfs_super_total_bytes(root->fs_info->super_copy));
771ed689 979
4b46fce2 980 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
771ed689
CM		 981 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
982
d397712b 983 while (disk_num_bytes > 0) {
a791e35e
CM		 984 unsigned long op;
985
287a0ab9 986 cur_alloc_size = disk_num_bytes;
00361589 987 ret = btrfs_reserve_extent(root, cur_alloc_size,
771ed689 988 root->sectorsize, 0, alloc_hint,
e570fd27 989 &ins, 1, 1);
00361589 990 if (ret < 0)
79787eaa 991 goto out_unlock;
d397712b 992
172ddd60 993 em = alloc_extent_map();
b9aa55be
LB		 994 if (!em) {
995 ret = -ENOMEM;
ace68bac 996 goto out_reserve;
b9aa55be 997 }
e6dcd2dc 998 em->start = start;
445a6944 999 em->orig_start = em->start;
771ed689
CM		 1000 ram_size = ins.offset;
1001 em->len = ins.offset;
2ab28f32
JB		 1002 em->mod_start = em->start;
1003 em->mod_len = em->len;
c8b97818 1004
e6dcd2dc 1005 em->block_start = ins.objectid;
c8b97818 1006 em->block_len = ins.offset;
b4939680 1007 em->orig_block_len = ins.offset;
cc95bef6 1008 em->ram_bytes = ram_size;
e6dcd2dc 1009 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 1010 set_bit(EXTENT_FLAG_PINNED, &em->flags);
70c8a91c 1011 em->generation = -1;
c8b97818 1012
d397712b 1013 while (1) {
890871be 1014 write_lock(&em_tree->lock);
09a2a8f9 1015 ret = add_extent_mapping(em_tree, em, 1);
890871be 1016 write_unlock(&em_tree->lock);
e6dcd2dc
CM		 1017 if (ret != -EEXIST) {
1018 free_extent_map(em);
1019 break;
1020 }
1021 btrfs_drop_extent_cache(inode, start,
c8b97818 1022 start + ram_size - 1, 0);
e6dcd2dc 1023 }
ace68bac
LB		 1024 if (ret)
1025 goto out_reserve;
e6dcd2dc 1026
98d20f67 1027 cur_alloc_size = ins.offset;
e6dcd2dc 1028 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771ed689 1029 ram_size, cur_alloc_size, 0);
ace68bac 1030 if (ret)
d9f85963 1031 goto out_drop_extent_cache;
c8b97818 1032
17d217fe
YZ		 1033 if (root->root_key.objectid ==
1034 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1035 ret = btrfs_reloc_clone_csums(inode, start,
1036 cur_alloc_size);
00361589 1037 if (ret)
d9f85963 1038 goto out_drop_extent_cache;
17d217fe
YZ		 1039 }
1040
d397712b 1041 if (disk_num_bytes < cur_alloc_size)
3b951516 1042 break;
d397712b 1043
c8b97818
CM		 1044 /* we're not doing compressed IO, don't unlock the first
1045 * page (which the caller expects to stay locked), don't
1046 * clear any dirty bits and don't set any writeback bits
8b62b72b
CM		 1047 *
1048 * Do set the Private2 bit so we know this page was properly
1049 * setup for writepage
c8b97818 1050 */
c2790a2e
JB		 1051 op = unlock ? PAGE_UNLOCK : 0;
1052 op |= PAGE_SET_PRIVATE2;
a791e35e 1053
c2790a2e
JB		 1054 extent_clear_unlock_delalloc(inode, start,
1055 start + ram_size - 1, locked_page,
1056 EXTENT_LOCKED | EXTENT_DELALLOC,
1057 op);
c8b97818 1058 disk_num_bytes -= cur_alloc_size;
c59f8951
CM		 1059 num_bytes -= cur_alloc_size;
1060 alloc_hint = ins.objectid + ins.offset;
1061 start += cur_alloc_size;
b888db2b 1062 }
79787eaa 1063out:
be20aa9d 1064 return ret;
b7d5b0a8 1065
d9f85963
FM		 1066out_drop_extent_cache:
1067 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
ace68bac 1068out_reserve:
e570fd27 1069 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
79787eaa 1070out_unlock:
c2790a2e 1071 extent_clear_unlock_delalloc(inode, start, end, locked_page,
151a41bc
JB		 1072 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1073 EXTENT_DELALLOC | EXTENT_DEFRAG,
1074 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1075 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
79787eaa 1076 goto out;
771ed689 1077}
c8b97818 1078
771ed689
CM		 1079/*
1080 * work queue call back to started compression on a file and pages
1081 */
1082static noinline void async_cow_start(struct btrfs_work *work)
1083{
1084 struct async_cow *async_cow;
1085 int num_added = 0;
1086 async_cow = container_of(work, struct async_cow, work);
1087
1088 compress_file_range(async_cow->inode, async_cow->locked_page,
1089 async_cow->start, async_cow->end, async_cow,
1090 &num_added);
8180ef88 1091 if (num_added == 0) {
cb77fcd8 1092 btrfs_add_delayed_iput(async_cow->inode);
771ed689 1093 async_cow->inode = NULL;
8180ef88 1094 }
771ed689
CM		 1095}
1096
1097/*
1098 * work queue call back to submit previously compressed pages
1099 */
1100static noinline void async_cow_submit(struct btrfs_work *work)
1101{
1102 struct async_cow *async_cow;
1103 struct btrfs_root *root;
1104 unsigned long nr_pages;
1105
1106 async_cow = container_of(work, struct async_cow, work);
1107
1108 root = async_cow->root;
1109 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1110 PAGE_CACHE_SHIFT;
1111
ee863954
DS		 1112 /*
1113 * atomic_sub_return implies a barrier for waitqueue_active
1114 */
66657b31 1115 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
ee22184b 1116 5 * SZ_1M &&
771ed689
CM		 1117 waitqueue_active(&root->fs_info->async_submit_wait))
1118 wake_up(&root->fs_info->async_submit_wait);
1119
d397712b 1120 if (async_cow->inode)
771ed689 1121 submit_compressed_extents(async_cow->inode, async_cow);
771ed689 1122}
c8b97818 1123
771ed689
CM		 1124static noinline void async_cow_free(struct btrfs_work *work)
1125{
1126 struct async_cow *async_cow;
1127 async_cow = container_of(work, struct async_cow, work);
8180ef88 1128 if (async_cow->inode)
cb77fcd8 1129 btrfs_add_delayed_iput(async_cow->inode);
771ed689
CM		 1130 kfree(async_cow);
1131}
1132
1133static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1134 u64 start, u64 end, int *page_started,
1135 unsigned long *nr_written)
1136{
1137 struct async_cow *async_cow;
1138 struct btrfs_root *root = BTRFS_I(inode)->root;
1139 unsigned long nr_pages;
1140 u64 cur_end;
ee22184b 1141 int limit = 10 * SZ_1M;
771ed689 1142
a3429ab7
CM		 1143 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1144 1, 0, NULL, GFP_NOFS);
d397712b 1145 while (start < end) {
771ed689 1146 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
79787eaa 1147 BUG_ON(!async_cow); /* -ENOMEM */
8180ef88 1148 async_cow->inode = igrab(inode);
771ed689
CM		 1149 async_cow->root = root;
1150 async_cow->locked_page = locked_page;
1151 async_cow->start = start;
1152
f79707b0
WS		 1153 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1154 !btrfs_test_opt(root, FORCE_COMPRESS))
771ed689
CM		 1155 cur_end = end;
1156 else
ee22184b 1157 cur_end = min(end, start + SZ_512K - 1);
771ed689
CM		 1158
1159 async_cow->end = cur_end;
1160 INIT_LIST_HEAD(&async_cow->extents);
1161
9e0af237
LB		 1162 btrfs_init_work(&async_cow->work,
1163 btrfs_delalloc_helper,
1164 async_cow_start, async_cow_submit,
1165 async_cow_free);
771ed689 1166
771ed689
CM		 1167 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1168 PAGE_CACHE_SHIFT;
1169 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1170
afe3d242
QW		 1171 btrfs_queue_work(root->fs_info->delalloc_workers,
1172 &async_cow->work);
771ed689
CM		 1173
1174 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1175 wait_event(root->fs_info->async_submit_wait,
1176 (atomic_read(&root->fs_info->async_delalloc_pages) <
1177 limit));
1178 }
1179
d397712b 1180 while (atomic_read(&root->fs_info->async_submit_draining) &&
771ed689
CM		 1181 atomic_read(&root->fs_info->async_delalloc_pages)) {
1182 wait_event(root->fs_info->async_submit_wait,
1183 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1184 0));
1185 }
1186
1187 *nr_written += nr_pages;
1188 start = cur_end + 1;
1189 }
1190 *page_started = 1;
1191 return 0;
be20aa9d
CM		 1192}
1193
d397712b 1194static noinline int csum_exist_in_range(struct btrfs_root *root,
17d217fe
YZ		 1195 u64 bytenr, u64 num_bytes)
1196{
1197 int ret;
1198 struct btrfs_ordered_sum *sums;
1199 LIST_HEAD(list);
1200
07d400a6 1201 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
a2de733c 1202 bytenr + num_bytes - 1, &list, 0);
17d217fe
YZ		 1203 if (ret == 0 && list_empty(&list))
1204 return 0;
1205
1206 while (!list_empty(&list)) {
1207 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1208 list_del(&sums->list);
1209 kfree(sums);
1210 }
1211 return 1;
1212}
1213
d352ac68
CM		 1214/*
1215 * when nowcow writeback call back. This checks for snapshots or COW copies
1216 * of the extents that exist in the file, and COWs the file as required.
1217 *
1218 * If no cow copies or snapshots exist, we write directly to the existing
1219 * blocks on disk
1220 */
7f366cfe
CM		 1221static noinline int run_delalloc_nocow(struct inode *inode,
1222 struct page *locked_page,
771ed689
CM		 1223 u64 start, u64 end, int *page_started, int force,
1224 unsigned long *nr_written)
be20aa9d 1225{
be20aa9d 1226 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 1227 struct btrfs_trans_handle *trans;
be20aa9d 1228 struct extent_buffer *leaf;
be20aa9d 1229 struct btrfs_path *path;
80ff3856 1230 struct btrfs_file_extent_item *fi;
be20aa9d 1231 struct btrfs_key found_key;
80ff3856
YZ		 1232 u64 cow_start;
1233 u64 cur_offset;
1234 u64 extent_end;
5d4f98a2 1235 u64 extent_offset;
80ff3856
YZ		 1236 u64 disk_bytenr;
1237 u64 num_bytes;
b4939680 1238 u64 disk_num_bytes;
cc95bef6 1239 u64 ram_bytes;
80ff3856 1240 int extent_type;
79787eaa 1241 int ret, err;
d899e052 1242 int type;
80ff3856
YZ		 1243 int nocow;
1244 int check_prev = 1;
82d5902d 1245 bool nolock;
33345d01 1246 u64 ino = btrfs_ino(inode);
be20aa9d
CM		 1247
1248 path = btrfs_alloc_path();
17ca04af 1249 if (!path) {
c2790a2e
JB		 1250 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1251 EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc
JB		 1252 EXTENT_DO_ACCOUNTING |
1253 EXTENT_DEFRAG, PAGE_UNLOCK |
c2790a2e
JB		 1254 PAGE_CLEAR_DIRTY |
1255 PAGE_SET_WRITEBACK |
1256 PAGE_END_WRITEBACK);
d8926bb3 1257 return -ENOMEM;
17ca04af 1258 }
82d5902d 1259
83eea1f1 1260 nolock = btrfs_is_free_space_inode(inode);
82d5902d
LZ		 1261
1262 if (nolock)
7a7eaa40 1263 trans = btrfs_join_transaction_nolock(root);
82d5902d 1264 else
7a7eaa40 1265 trans = btrfs_join_transaction(root);
ff5714cc 1266
79787eaa 1267 if (IS_ERR(trans)) {
c2790a2e
JB		 1268 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1269 EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc
JB		 1270 EXTENT_DO_ACCOUNTING |
1271 EXTENT_DEFRAG, PAGE_UNLOCK |
c2790a2e
JB		 1272 PAGE_CLEAR_DIRTY |
1273 PAGE_SET_WRITEBACK |
1274 PAGE_END_WRITEBACK);
79787eaa
JM		 1275 btrfs_free_path(path);
1276 return PTR_ERR(trans);
1277 }
1278
74b21075 1279 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
be20aa9d 1280
80ff3856
YZ		 1281 cow_start = (u64)-1;
1282 cur_offset = start;
1283 while (1) {
33345d01 1284 ret = btrfs_lookup_file_extent(trans, root, path, ino,
80ff3856 1285 cur_offset, 0);
d788a349 1286 if (ret < 0)
79787eaa 1287 goto error;
80ff3856
YZ		 1288 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1289 leaf = path->nodes[0];
1290 btrfs_item_key_to_cpu(leaf, &found_key,
1291 path->slots[0] - 1);
33345d01 1292 if (found_key.objectid == ino &&
80ff3856
YZ		 1293 found_key.type == BTRFS_EXTENT_DATA_KEY)
1294 path->slots[0]--;
1295 }
1296 check_prev = 0;
1297next_slot:
1298 leaf = path->nodes[0];
1299 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1300 ret = btrfs_next_leaf(root, path);
d788a349 1301 if (ret < 0)
79787eaa 1302 goto error;
80ff3856
YZ		 1303 if (ret > 0)
1304 break;
1305 leaf = path->nodes[0];
1306 }
be20aa9d 1307
80ff3856
YZ		 1308 nocow = 0;
1309 disk_bytenr = 0;
17d217fe 1310 num_bytes = 0;
80ff3856
YZ		 1311 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1312
1d512cb7
FM		 1313 if (found_key.objectid > ino)
1314 break;
1315 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1316 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1317 path->slots[0]++;
1318 goto next_slot;
1319 }
1320 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
80ff3856
YZ		 1321 found_key.offset > end)
1322 break;
1323
1324 if (found_key.offset > cur_offset) {
1325 extent_end = found_key.offset;
e9061e21 1326 extent_type = 0;
80ff3856
YZ		 1327 goto out_check;
1328 }
1329
1330 fi = btrfs_item_ptr(leaf, path->slots[0],
1331 struct btrfs_file_extent_item);
1332 extent_type = btrfs_file_extent_type(leaf, fi);
1333
cc95bef6 1334 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
d899e052
YZ		 1335 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1336 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856 1337 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5d4f98a2 1338 extent_offset = btrfs_file_extent_offset(leaf, fi);
80ff3856
YZ		 1339 extent_end = found_key.offset +
1340 btrfs_file_extent_num_bytes(leaf, fi);
b4939680
JB		 1341 disk_num_bytes =
1342 btrfs_file_extent_disk_num_bytes(leaf, fi);
80ff3856
YZ		 1343 if (extent_end <= start) {
1344 path->slots[0]++;
1345 goto next_slot;
1346 }
17d217fe
YZ		 1347 if (disk_bytenr == 0)
1348 goto out_check;
80ff3856
YZ		 1349 if (btrfs_file_extent_compression(leaf, fi) ||
1350 btrfs_file_extent_encryption(leaf, fi) ||
1351 btrfs_file_extent_other_encoding(leaf, fi))
1352 goto out_check;
d899e052
YZ		 1353 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1354 goto out_check;
d2fb3437 1355 if (btrfs_extent_readonly(root, disk_bytenr))
80ff3856 1356 goto out_check;
33345d01 1357 if (btrfs_cross_ref_exist(trans, root, ino,
5d4f98a2
YZ		 1358 found_key.offset -
1359 extent_offset, disk_bytenr))
17d217fe 1360 goto out_check;
5d4f98a2 1361 disk_bytenr += extent_offset;
17d217fe
YZ		 1362 disk_bytenr += cur_offset - found_key.offset;
1363 num_bytes = min(end + 1, extent_end) - cur_offset;
e9894fd3
WS		 1364 /*
1365 * if there are pending snapshots for this root,
1366 * we fall into common COW way.
1367 */
1368 if (!nolock) {
9ea24bbe 1369 err = btrfs_start_write_no_snapshoting(root);
e9894fd3
WS		 1370 if (!err)
1371 goto out_check;
1372 }
17d217fe
YZ		 1373 /*
1374 * force cow if csum exists in the range.
1375 * this ensure that csum for a given extent are
1376 * either valid or do not exist.
1377 */
1378 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1379 goto out_check;
80ff3856
YZ		 1380 nocow = 1;
1381 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1382 extent_end = found_key.offset +
514ac8ad
CM		 1383 btrfs_file_extent_inline_len(leaf,
1384 path->slots[0], fi);
80ff3856
YZ		 1385 extent_end = ALIGN(extent_end, root->sectorsize);
1386 } else {
1387 BUG_ON(1);
1388 }
1389out_check:
1390 if (extent_end <= start) {
1391 path->slots[0]++;
e9894fd3 1392 if (!nolock && nocow)
9ea24bbe 1393 btrfs_end_write_no_snapshoting(root);
80ff3856
YZ		 1394 goto next_slot;
1395 }
1396 if (!nocow) {
1397 if (cow_start == (u64)-1)
1398 cow_start = cur_offset;
1399 cur_offset = extent_end;
1400 if (cur_offset > end)
1401 break;
1402 path->slots[0]++;
1403 goto next_slot;
7ea394f1
YZ		 1404 }
1405
b3b4aa74 1406 btrfs_release_path(path);
80ff3856 1407 if (cow_start != (u64)-1) {
00361589
JB		 1408 ret = cow_file_range(inode, locked_page,
1409 cow_start, found_key.offset - 1,
1410 page_started, nr_written, 1);
e9894fd3
WS		 1411 if (ret) {
1412 if (!nolock && nocow)
9ea24bbe 1413 btrfs_end_write_no_snapshoting(root);
79787eaa 1414 goto error;
e9894fd3 1415 }
80ff3856 1416 cow_start = (u64)-1;
7ea394f1 1417 }
80ff3856 1418
d899e052
YZ		 1419 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1420 struct extent_map *em;
1421 struct extent_map_tree *em_tree;
1422 em_tree = &BTRFS_I(inode)->extent_tree;
172ddd60 1423 em = alloc_extent_map();
79787eaa 1424 BUG_ON(!em); /* -ENOMEM */
d899e052 1425 em->start = cur_offset;
70c8a91c 1426 em->orig_start = found_key.offset - extent_offset;
d899e052
YZ		 1427 em->len = num_bytes;
1428 em->block_len = num_bytes;
1429 em->block_start = disk_bytenr;
b4939680 1430 em->orig_block_len = disk_num_bytes;
cc95bef6 1431 em->ram_bytes = ram_bytes;
d899e052 1432 em->bdev = root->fs_info->fs_devices->latest_bdev;
2ab28f32
JB		 1433 em->mod_start = em->start;
1434 em->mod_len = em->len;
d899e052 1435 set_bit(EXTENT_FLAG_PINNED, &em->flags);
b11e234d 1436 set_bit(EXTENT_FLAG_FILLING, &em->flags);
70c8a91c 1437 em->generation = -1;
d899e052 1438 while (1) {
890871be 1439 write_lock(&em_tree->lock);
09a2a8f9 1440 ret = add_extent_mapping(em_tree, em, 1);
890871be 1441 write_unlock(&em_tree->lock);
d899e052
YZ		 1442 if (ret != -EEXIST) {
1443 free_extent_map(em);
1444 break;
1445 }
1446 btrfs_drop_extent_cache(inode, em->start,
1447 em->start + em->len - 1, 0);
1448 }
1449 type = BTRFS_ORDERED_PREALLOC;
1450 } else {
1451 type = BTRFS_ORDERED_NOCOW;
1452 }
80ff3856
YZ		 1453
1454 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052 1455 num_bytes, num_bytes, type);
79787eaa 1456 BUG_ON(ret); /* -ENOMEM */
771ed689 1457
efa56464
YZ		 1458 if (root->root_key.objectid ==
1459 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1460 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1461 num_bytes);
e9894fd3
WS		 1462 if (ret) {
1463 if (!nolock && nocow)
9ea24bbe 1464 btrfs_end_write_no_snapshoting(root);
79787eaa 1465 goto error;
e9894fd3 1466 }
efa56464
YZ		 1467 }
1468
c2790a2e
JB		 1469 extent_clear_unlock_delalloc(inode, cur_offset,
1470 cur_offset + num_bytes - 1,
1471 locked_page, EXTENT_LOCKED |
1472 EXTENT_DELALLOC, PAGE_UNLOCK |
1473 PAGE_SET_PRIVATE2);
e9894fd3 1474 if (!nolock && nocow)
9ea24bbe 1475 btrfs_end_write_no_snapshoting(root);
80ff3856
YZ		 1476 cur_offset = extent_end;
1477 if (cur_offset > end)
1478 break;
be20aa9d 1479 }
b3b4aa74 1480 btrfs_release_path(path);
80ff3856 1481
17ca04af 1482 if (cur_offset <= end && cow_start == (u64)-1) {
80ff3856 1483 cow_start = cur_offset;
17ca04af
JB		 1484 cur_offset = end;
1485 }
1486
80ff3856 1487 if (cow_start != (u64)-1) {
00361589
JB		 1488 ret = cow_file_range(inode, locked_page, cow_start, end,
1489 page_started, nr_written, 1);
d788a349 1490 if (ret)
79787eaa 1491 goto error;
80ff3856
YZ		 1492 }
1493
79787eaa 1494error:
a698d075 1495 err = btrfs_end_transaction(trans, root);
79787eaa
JM		 1496 if (!ret)
1497 ret = err;
1498
17ca04af 1499 if (ret && cur_offset < end)
c2790a2e
JB		 1500 extent_clear_unlock_delalloc(inode, cur_offset, end,
1501 locked_page, EXTENT_LOCKED |
151a41bc
JB		 1502 EXTENT_DELALLOC | EXTENT_DEFRAG |
1503 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1504 PAGE_CLEAR_DIRTY |
c2790a2e
JB		 1505 PAGE_SET_WRITEBACK |
1506 PAGE_END_WRITEBACK);
7ea394f1 1507 btrfs_free_path(path);
79787eaa 1508 return ret;
be20aa9d
CM		 1509}
1510
47059d93
WS		 1511static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1512{
1513
1514 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1515 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1516 return 0;
1517
1518 /*
1519 * @defrag_bytes is a hint value, no spinlock held here,
1520 * if is not zero, it means the file is defragging.
1521 * Force cow if given extent needs to be defragged.
1522 */
1523 if (BTRFS_I(inode)->defrag_bytes &&
1524 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1525 EXTENT_DEFRAG, 0, NULL))
1526 return 1;
1527
1528 return 0;
1529}
1530
d352ac68
CM		 1531/*
1532 * extent_io.c call back to do delayed allocation processing
1533 */
c8b97818 1534static int run_delalloc_range(struct inode *inode, struct page *locked_page,
771ed689
CM		 1535 u64 start, u64 end, int *page_started,
1536 unsigned long *nr_written)
be20aa9d 1537{
be20aa9d 1538 int ret;
47059d93 1539 int force_cow = need_force_cow(inode, start, end);
a2135011 1540
47059d93 1541 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
c8b97818 1542 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1543 page_started, 1, nr_written);
47059d93 1544 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
d899e052 1545 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1546 page_started, 0, nr_written);
7816030e 1547 } else if (!inode_need_compress(inode)) {
7f366cfe
CM		 1548 ret = cow_file_range(inode, locked_page, start, end,
1549 page_started, nr_written, 1);
7ddf5a42
JB		 1550 } else {
1551 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1552 &BTRFS_I(inode)->runtime_flags);
771ed689 1553 ret = cow_file_range_async(inode, locked_page, start, end,
d397712b 1554 page_started, nr_written);
7ddf5a42 1555 }
b888db2b
CM		 1556 return ret;
1557}
1558
1bf85046
JM		 1559static void btrfs_split_extent_hook(struct inode *inode,
1560 struct extent_state *orig, u64 split)
9ed74f2d 1561{
dcab6a3b
JB		 1562 u64 size;
1563
0ca1f7ce 1564 /* not delalloc, ignore it */
9ed74f2d 1565 if (!(orig->state & EXTENT_DELALLOC))
1bf85046 1566 return;
9ed74f2d 1567
dcab6a3b
JB		 1568 size = orig->end - orig->start + 1;
1569 if (size > BTRFS_MAX_EXTENT_SIZE) {
1570 u64 num_extents;
1571 u64 new_size;
1572
1573 /*
ba117213
JB		 1574 * See the explanation in btrfs_merge_extent_hook, the same
1575 * applies here, just in reverse.
dcab6a3b
JB		 1576 */
1577 new_size = orig->end - split + 1;
ba117213 1578 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
dcab6a3b 1579 BTRFS_MAX_EXTENT_SIZE);
ba117213
JB		 1580 new_size = split - orig->start;
1581 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1582 BTRFS_MAX_EXTENT_SIZE);
1583 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1584 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
dcab6a3b
JB		 1585 return;
1586 }
1587
9e0baf60
JB		 1588 spin_lock(&BTRFS_I(inode)->lock);
1589 BTRFS_I(inode)->outstanding_extents++;
1590 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1591}
1592
1593/*
1594 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1595 * extents so we can keep track of new extents that are just merged onto old
1596 * extents, such as when we are doing sequential writes, so we can properly
1597 * account for the metadata space we'll need.
1598 */
1bf85046
JM		 1599static void btrfs_merge_extent_hook(struct inode *inode,
1600 struct extent_state *new,
1601 struct extent_state *other)
9ed74f2d 1602{
dcab6a3b
JB		 1603 u64 new_size, old_size;
1604 u64 num_extents;
1605
9ed74f2d
JB		 1606 /* not delalloc, ignore it */
1607 if (!(other->state & EXTENT_DELALLOC))
1bf85046 1608 return;
9ed74f2d 1609
8461a3de
JB		 1610 if (new->start > other->start)
1611 new_size = new->end - other->start + 1;
1612 else
1613 new_size = other->end - new->start + 1;
dcab6a3b
JB		 1614
1615 /* we're not bigger than the max, unreserve the space and go */
1616 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1617 spin_lock(&BTRFS_I(inode)->lock);
1618 BTRFS_I(inode)->outstanding_extents--;
1619 spin_unlock(&BTRFS_I(inode)->lock);
1620 return;
1621 }
1622
1623 /*
ba117213
JB		 1624 * We have to add up either side to figure out how many extents were
1625 * accounted for before we merged into one big extent. If the number of
1626 * extents we accounted for is <= the amount we need for the new range
1627 * then we can return, otherwise drop. Think of it like this
1628 *
1629 * [ 4k][MAX_SIZE]
1630 *
1631 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1632 * need 2 outstanding extents, on one side we have 1 and the other side
1633 * we have 1 so they are == and we can return. But in this case
1634 *
1635 * [MAX_SIZE+4k][MAX_SIZE+4k]
1636 *
1637 * Each range on their own accounts for 2 extents, but merged together
1638 * they are only 3 extents worth of accounting, so we need to drop in
1639 * this case.
dcab6a3b 1640 */
ba117213 1641 old_size = other->end - other->start + 1;
dcab6a3b
JB		 1642 num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1643 BTRFS_MAX_EXTENT_SIZE);
ba117213
JB		 1644 old_size = new->end - new->start + 1;
1645 num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1646 BTRFS_MAX_EXTENT_SIZE);
1647
dcab6a3b 1648 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
ba117213 1649 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
dcab6a3b
JB		 1650 return;
1651
9e0baf60
JB		 1652 spin_lock(&BTRFS_I(inode)->lock);
1653 BTRFS_I(inode)->outstanding_extents--;
1654 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1655}
1656
eb73c1b7
MX		 1657static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1658 struct inode *inode)
1659{
1660 spin_lock(&root->delalloc_lock);
1661 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1662 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1663 &root->delalloc_inodes);
1664 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1665 &BTRFS_I(inode)->runtime_flags);
1666 root->nr_delalloc_inodes++;
1667 if (root->nr_delalloc_inodes == 1) {
1668 spin_lock(&root->fs_info->delalloc_root_lock);
1669 BUG_ON(!list_empty(&root->delalloc_root));
1670 list_add_tail(&root->delalloc_root,
1671 &root->fs_info->delalloc_roots);
1672 spin_unlock(&root->fs_info->delalloc_root_lock);
1673 }
1674 }
1675 spin_unlock(&root->delalloc_lock);
1676}
1677
1678static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1679 struct inode *inode)
1680{
1681 spin_lock(&root->delalloc_lock);
1682 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1683 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1684 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1685 &BTRFS_I(inode)->runtime_flags);
1686 root->nr_delalloc_inodes--;
1687 if (!root->nr_delalloc_inodes) {
1688 spin_lock(&root->fs_info->delalloc_root_lock);
1689 BUG_ON(list_empty(&root->delalloc_root));
1690 list_del_init(&root->delalloc_root);
1691 spin_unlock(&root->fs_info->delalloc_root_lock);
1692 }
1693 }
1694 spin_unlock(&root->delalloc_lock);
1695}
1696
d352ac68
CM		 1697/*
1698 * extent_io.c set_bit_hook, used to track delayed allocation
1699 * bytes in this file, and to maintain the list of inodes that
1700 * have pending delalloc work to be done.
1701 */
1bf85046 1702static void btrfs_set_bit_hook(struct inode *inode,
9ee49a04 1703 struct extent_state *state, unsigned *bits)
291d673e 1704{
9ed74f2d 1705
47059d93
WS		 1706 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1707 WARN_ON(1);
75eff68e
CM		 1708 /*
1709 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1710 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1711 * bit, which is only set or cleared with irqs on
1712 */
0ca1f7ce 1713 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1714 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1715 u64 len = state->end + 1 - state->start;
83eea1f1 1716 bool do_list = !btrfs_is_free_space_inode(inode);
9ed74f2d 1717
9e0baf60 1718 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1719 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1720 } else {
1721 spin_lock(&BTRFS_I(inode)->lock);
1722 BTRFS_I(inode)->outstanding_extents++;
1723 spin_unlock(&BTRFS_I(inode)->lock);
1724 }
287a0ab9 1725
6a3891c5
JB		 1726 /* For sanity tests */
1727 if (btrfs_test_is_dummy_root(root))
1728 return;
1729
963d678b
MX		 1730 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1731 root->fs_info->delalloc_batch);
df0af1a5 1732 spin_lock(&BTRFS_I(inode)->lock);
0ca1f7ce 1733 BTRFS_I(inode)->delalloc_bytes += len;
47059d93
WS		 1734 if (*bits & EXTENT_DEFRAG)
1735 BTRFS_I(inode)->defrag_bytes += len;
df0af1a5 1736 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
eb73c1b7
MX		 1737 &BTRFS_I(inode)->runtime_flags))
1738 btrfs_add_delalloc_inodes(root, inode);
df0af1a5 1739 spin_unlock(&BTRFS_I(inode)->lock);
291d673e 1740 }
291d673e
CM		 1741}
1742
d352ac68
CM		 1743/*
1744 * extent_io.c clear_bit_hook, see set_bit_hook for why
1745 */
1bf85046 1746static void btrfs_clear_bit_hook(struct inode *inode,
41074888 1747 struct extent_state *state,
9ee49a04 1748 unsigned *bits)
291d673e 1749{
47059d93 1750 u64 len = state->end + 1 - state->start;
dcab6a3b
JB		 1751 u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1752 BTRFS_MAX_EXTENT_SIZE);
47059d93
WS		 1753
1754 spin_lock(&BTRFS_I(inode)->lock);
1755 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1756 BTRFS_I(inode)->defrag_bytes -= len;
1757 spin_unlock(&BTRFS_I(inode)->lock);
1758
75eff68e
CM		 1759 /*
1760 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1761 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1762 * bit, which is only set or cleared with irqs on
1763 */
0ca1f7ce 1764 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1765 struct btrfs_root *root = BTRFS_I(inode)->root;
83eea1f1 1766 bool do_list = !btrfs_is_free_space_inode(inode);
bcbfce8a 1767
9e0baf60 1768 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1769 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1770 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1771 spin_lock(&BTRFS_I(inode)->lock);
dcab6a3b 1772 BTRFS_I(inode)->outstanding_extents -= num_extents;
9e0baf60
JB		 1773 spin_unlock(&BTRFS_I(inode)->lock);
1774 }
0ca1f7ce 1775
b6d08f06
JB		 1776 /*
1777 * We don't reserve metadata space for space cache inodes so we
1778 * don't need to call dellalloc_release_metadata if there is an
1779 * error.
1780 */
1781 if (*bits & EXTENT_DO_ACCOUNTING &&
1782 root != root->fs_info->tree_root)
0ca1f7ce
YZ		 1783 btrfs_delalloc_release_metadata(inode, len);
1784
6a3891c5
JB		 1785 /* For sanity tests. */
1786 if (btrfs_test_is_dummy_root(root))
1787 return;
1788
0cb59c99 1789 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
7ee9e440 1790 && do_list && !(state->state & EXTENT_NORESERVE))
51773bec
QW		 1791 btrfs_free_reserved_data_space_noquota(inode,
1792 state->start, len);
9ed74f2d 1793
963d678b
MX		 1794 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1795 root->fs_info->delalloc_batch);
df0af1a5 1796 spin_lock(&BTRFS_I(inode)->lock);
0ca1f7ce 1797 BTRFS_I(inode)->delalloc_bytes -= len;
0cb59c99 1798 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
df0af1a5 1799 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
eb73c1b7
MX		 1800 &BTRFS_I(inode)->runtime_flags))
1801 btrfs_del_delalloc_inode(root, inode);
df0af1a5 1802 spin_unlock(&BTRFS_I(inode)->lock);
291d673e 1803 }
291d673e
CM		 1804}
1805
d352ac68
CM		 1806/*
1807 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1808 * we don't create bios that span stripes or chunks
1809 */
64a16701 1810int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
c8b97818
CM		 1811 size_t size, struct bio *bio,
1812 unsigned long bio_flags)
239b14b3
CM		 1813{
1814 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
4f024f37 1815 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
239b14b3
CM		 1816 u64 length = 0;
1817 u64 map_length;
239b14b3
CM		 1818 int ret;
1819
771ed689
CM		 1820 if (bio_flags & EXTENT_BIO_COMPRESSED)
1821 return 0;
1822
4f024f37 1823 length = bio->bi_iter.bi_size;
239b14b3 1824 map_length = length;
64a16701 1825 ret = btrfs_map_block(root->fs_info, rw, logical,
f188591e 1826 &map_length, NULL, 0);
3ec706c8 1827 /* Will always return 0 with map_multi == NULL */
3444a972 1828 BUG_ON(ret < 0);
d397712b 1829 if (map_length < length + size)
239b14b3 1830 return 1;
3444a972 1831 return 0;
239b14b3
CM		 1832}
1833
d352ac68
CM		 1834/*
1835 * in order to insert checksums into the metadata in large chunks,
1836 * we wait until bio submission time. All the pages in the bio are
1837 * checksummed and sums are attached onto the ordered extent record.
1838 *
1839 * At IO completion time the cums attached on the ordered extent record
1840 * are inserted into the btree
1841 */
d397712b
CM		 1842static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1843 struct bio *bio, int mirror_num,
eaf25d93
CM		 1844 unsigned long bio_flags,
1845 u64 bio_offset)
065631f6 1846{
065631f6 1847 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 1848 int ret = 0;
e015640f 1849
d20f7043 1850 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
79787eaa 1851 BUG_ON(ret); /* -ENOMEM */
4a69a410
CM		 1852 return 0;
1853}
e015640f 1854
4a69a410
CM		 1855/*
1856 * in order to insert checksums into the metadata in large chunks,
1857 * we wait until bio submission time. All the pages in the bio are
1858 * checksummed and sums are attached onto the ordered extent record.
1859 *
1860 * At IO completion time the cums attached on the ordered extent record
1861 * are inserted into the btree
1862 */
b2950863 1863static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1864 int mirror_num, unsigned long bio_flags,
1865 u64 bio_offset)
4a69a410
CM		 1866{
1867 struct btrfs_root *root = BTRFS_I(inode)->root;
61891923
SB		 1868 int ret;
1869
1870 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
4246a0b6
CH		 1871 if (ret) {
1872 bio->bi_error = ret;
1873 bio_endio(bio);
1874 }
61891923 1875 return ret;
44b8bd7e
CM		 1876}
1877
d352ac68 1878/*
cad321ad
CM		 1879 * extent_io.c submission hook. This does the right thing for csum calculation
1880 * on write, or reading the csums from the tree before a read
d352ac68 1881 */
b2950863 1882static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1883 int mirror_num, unsigned long bio_flags,
1884 u64 bio_offset)
44b8bd7e
CM		 1885{
1886 struct btrfs_root *root = BTRFS_I(inode)->root;
0d51e28a 1887 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
44b8bd7e 1888 int ret = 0;
19b9bdb0 1889 int skip_sum;
b812ce28 1890 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
44b8bd7e 1891
6cbff00f 1892 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
cad321ad 1893
83eea1f1 1894 if (btrfs_is_free_space_inode(inode))
0d51e28a 1895 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
0417341e 1896
7b6d91da 1897 if (!(rw & REQ_WRITE)) {
5fd02043
JB		 1898 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1899 if (ret)
61891923 1900 goto out;
5fd02043 1901
d20f7043 1902 if (bio_flags & EXTENT_BIO_COMPRESSED) {
61891923
SB		 1903 ret = btrfs_submit_compressed_read(inode, bio,
1904 mirror_num,
1905 bio_flags);
1906 goto out;
c2db1073
TI		 1907 } else if (!skip_sum) {
1908 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1909 if (ret)
61891923 1910 goto out;
c2db1073 1911 }
4d1b5fb4 1912 goto mapit;
b812ce28 1913 } else if (async && !skip_sum) {
17d217fe
YZ		 1914 /* csum items have already been cloned */
1915 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1916 goto mapit;
19b9bdb0 1917 /* we're doing a write, do the async checksumming */
61891923 1918 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 1919 inode, rw, bio, mirror_num,
eaf25d93
CM		 1920 bio_flags, bio_offset,
1921 __btrfs_submit_bio_start,
4a69a410 1922 __btrfs_submit_bio_done);
61891923 1923 goto out;
b812ce28
JB		 1924 } else if (!skip_sum) {
1925 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1926 if (ret)
1927 goto out;
19b9bdb0
CM		 1928 }
1929
0b86a832 1930mapit:
61891923
SB		 1931 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1932
1933out:
4246a0b6
CH		 1934 if (ret < 0) {
1935 bio->bi_error = ret;
1936 bio_endio(bio);
1937 }
61891923 1938 return ret;
065631f6 1939}
6885f308 1940
d352ac68
CM		 1941/*
1942 * given a list of ordered sums record them in the inode. This happens
1943 * at IO completion time based on sums calculated at bio submission time.
1944 */
ba1da2f4 1945static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM		 1946 struct inode *inode, u64 file_offset,
1947 struct list_head *list)
1948{
e6dcd2dc
CM		 1949 struct btrfs_ordered_sum *sum;
1950
c6e30871 1951 list_for_each_entry(sum, list, list) {
39847c4d 1952 trans->adding_csums = 1;
d20f7043
CM		 1953 btrfs_csum_file_blocks(trans,
1954 BTRFS_I(inode)->root->fs_info->csum_root, sum);
39847c4d 1955 trans->adding_csums = 0;
e6dcd2dc
CM		 1956 }
1957 return 0;
1958}
1959
2ac55d41
JB		 1960int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1961 struct extent_state **cached_state)
ea8c2819 1962{
6c1500f2 1963 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
ea8c2819 1964 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2ac55d41 1965 cached_state, GFP_NOFS);
ea8c2819
CM		 1966}
1967
d352ac68 1968/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM		 1969struct btrfs_writepage_fixup {
1970 struct page *page;
1971 struct btrfs_work work;
1972};
1973
b2950863 1974static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
247e743c
CM		 1975{
1976 struct btrfs_writepage_fixup *fixup;
1977 struct btrfs_ordered_extent *ordered;
2ac55d41 1978 struct extent_state *cached_state = NULL;
247e743c
CM		 1979 struct page *page;
1980 struct inode *inode;
1981 u64 page_start;
1982 u64 page_end;
87826df0 1983 int ret;
247e743c
CM		 1984
1985 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1986 page = fixup->page;
4a096752 1987again:
247e743c
CM		 1988 lock_page(page);
1989 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1990 ClearPageChecked(page);
1991 goto out_page;
1992 }
1993
1994 inode = page->mapping->host;
1995 page_start = page_offset(page);
1996 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1997
ff13db41 1998 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
d0082371 1999 &cached_state);
4a096752
CM		 2000
2001 /* already ordered? We're done */
8b62b72b 2002 if (PagePrivate2(page))
247e743c 2003 goto out;
4a096752
CM		 2004
2005 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2006 if (ordered) {
2ac55d41
JB		 2007 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2008 page_end, &cached_state, GFP_NOFS);
4a096752
CM		 2009 unlock_page(page);
2010 btrfs_start_ordered_extent(inode, ordered, 1);
87826df0 2011 btrfs_put_ordered_extent(ordered);
4a096752
CM		 2012 goto again;
2013 }
247e743c 2014
7cf5b976
QW		 2015 ret = btrfs_delalloc_reserve_space(inode, page_start,
2016 PAGE_CACHE_SIZE);
87826df0
JM		 2017 if (ret) {
2018 mapping_set_error(page->mapping, ret);
2019 end_extent_writepage(page, ret, page_start, page_end);
2020 ClearPageChecked(page);
2021 goto out;
2022 }
2023
2ac55d41 2024 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
247e743c 2025 ClearPageChecked(page);
87826df0 2026 set_page_dirty(page);
247e743c 2027out:
2ac55d41
JB		 2028 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2029 &cached_state, GFP_NOFS);
247e743c
CM		 2030out_page:
2031 unlock_page(page);
2032 page_cache_release(page);
b897abec 2033 kfree(fixup);
247e743c
CM		 2034}
2035
2036/*
2037 * There are a few paths in the higher layers of the kernel that directly
2038 * set the page dirty bit without asking the filesystem if it is a
2039 * good idea. This causes problems because we want to make sure COW
2040 * properly happens and the data=ordered rules are followed.
2041 *
c8b97818 2042 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 2043 * hasn't been properly setup for IO. We kick off an async process
2044 * to fix it up. The async helper will wait for ordered extents, set
2045 * the delalloc bit and make it safe to write the page.
2046 */
b2950863 2047static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
247e743c
CM		 2048{
2049 struct inode *inode = page->mapping->host;
2050 struct btrfs_writepage_fixup *fixup;
2051 struct btrfs_root *root = BTRFS_I(inode)->root;
247e743c 2052
8b62b72b
CM		 2053 /* this page is properly in the ordered list */
2054 if (TestClearPagePrivate2(page))
247e743c
CM		 2055 return 0;
2056
2057 if (PageChecked(page))
2058 return -EAGAIN;
2059
2060 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2061 if (!fixup)
2062 return -EAGAIN;
f421950f 2063
247e743c
CM		 2064 SetPageChecked(page);
2065 page_cache_get(page);
9e0af237
LB		 2066 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2067 btrfs_writepage_fixup_worker, NULL, NULL);
247e743c 2068 fixup->page = page;
dc6e3209 2069 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
87826df0 2070 return -EBUSY;
247e743c
CM		 2071}
2072
d899e052
YZ		 2073static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2074 struct inode *inode, u64 file_pos,
2075 u64 disk_bytenr, u64 disk_num_bytes,
2076 u64 num_bytes, u64 ram_bytes,
2077 u8 compression, u8 encryption,
2078 u16 other_encoding, int extent_type)
2079{
2080 struct btrfs_root *root = BTRFS_I(inode)->root;
2081 struct btrfs_file_extent_item *fi;
2082 struct btrfs_path *path;
2083 struct extent_buffer *leaf;
2084 struct btrfs_key ins;
1acae57b 2085 int extent_inserted = 0;
d899e052
YZ		 2086 int ret;
2087
2088 path = btrfs_alloc_path();
d8926bb3
MF		 2089 if (!path)
2090 return -ENOMEM;
d899e052 2091
a1ed835e
CM		 2092 /*
2093 * we may be replacing one extent in the tree with another.
2094 * The new extent is pinned in the extent map, and we don't want
2095 * to drop it from the cache until it is completely in the btree.
2096 *
2097 * So, tell btrfs_drop_extents to leave this extent in the cache.
2098 * the caller is expected to unpin it and allow it to be merged
2099 * with the others.
2100 */
1acae57b
FDBM		 2101 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2102 file_pos + num_bytes, NULL, 0,
2103 1, sizeof(*fi), &extent_inserted);
79787eaa
JM		 2104 if (ret)
2105 goto out;
d899e052 2106
1acae57b
FDBM		 2107 if (!extent_inserted) {
2108 ins.objectid = btrfs_ino(inode);
2109 ins.offset = file_pos;
2110 ins.type = BTRFS_EXTENT_DATA_KEY;
2111
2112 path->leave_spinning = 1;
2113 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2114 sizeof(*fi));
2115 if (ret)
2116 goto out;
2117 }
d899e052
YZ		 2118 leaf = path->nodes[0];
2119 fi = btrfs_item_ptr(leaf, path->slots[0],
2120 struct btrfs_file_extent_item);
2121 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2122 btrfs_set_file_extent_type(leaf, fi, extent_type);
2123 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2124 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2125 btrfs_set_file_extent_offset(leaf, fi, 0);
2126 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2127 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2128 btrfs_set_file_extent_compression(leaf, fi, compression);
2129 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2130 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
b9473439 2131
d899e052 2132 btrfs_mark_buffer_dirty(leaf);
ce195332 2133 btrfs_release_path(path);
d899e052
YZ		 2134
2135 inode_add_bytes(inode, num_bytes);
d899e052
YZ		 2136
2137 ins.objectid = disk_bytenr;
2138 ins.offset = disk_num_bytes;
2139 ins.type = BTRFS_EXTENT_ITEM_KEY;
5d4f98a2
YZ		 2140 ret = btrfs_alloc_reserved_file_extent(trans, root,
2141 root->root_key.objectid,
5846a3c2
QW		 2142 btrfs_ino(inode), file_pos,
2143 ram_bytes, &ins);
297d750b 2144 /*
5846a3c2
QW		 2145 * Release the reserved range from inode dirty range map, as it is
2146 * already moved into delayed_ref_head
297d750b
QW		 2147 */
2148 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
79787eaa 2149out:
d899e052 2150 btrfs_free_path(path);
b9473439 2151
79787eaa 2152 return ret;
d899e052
YZ		 2153}
2154
38c227d8
LB		 2155/* snapshot-aware defrag */
2156struct sa_defrag_extent_backref {
2157 struct rb_node node;
2158 struct old_sa_defrag_extent *old;
2159 u64 root_id;
2160 u64 inum;
2161 u64 file_pos;
2162 u64 extent_offset;
2163 u64 num_bytes;
2164 u64 generation;
2165};
2166
2167struct old_sa_defrag_extent {
2168 struct list_head list;
2169 struct new_sa_defrag_extent *new;
2170
2171 u64 extent_offset;
2172 u64 bytenr;
2173 u64 offset;
2174 u64 len;
2175 int count;
2176};
2177
2178struct new_sa_defrag_extent {
2179 struct rb_root root;
2180 struct list_head head;
2181 struct btrfs_path *path;
2182 struct inode *inode;
2183 u64 file_pos;
2184 u64 len;
2185 u64 bytenr;
2186 u64 disk_len;
2187 u8 compress_type;
2188};
2189
2190static int backref_comp(struct sa_defrag_extent_backref *b1,
2191 struct sa_defrag_extent_backref *b2)
2192{
2193 if (b1->root_id < b2->root_id)
2194 return -1;
2195 else if (b1->root_id > b2->root_id)
2196 return 1;
2197
2198 if (b1->inum < b2->inum)
2199 return -1;
2200 else if (b1->inum > b2->inum)
2201 return 1;
2202
2203 if (b1->file_pos < b2->file_pos)
2204 return -1;
2205 else if (b1->file_pos > b2->file_pos)
2206 return 1;
2207
2208 /*
2209 * [------------------------------] ===> (a range of space)
2210 * |<--->| |<---->| =============> (fs/file tree A)
2211 * |<---------------------------->| ===> (fs/file tree B)
2212 *
2213 * A range of space can refer to two file extents in one tree while
2214 * refer to only one file extent in another tree.
2215 *
2216 * So we may process a disk offset more than one time(two extents in A)
2217 * and locate at the same extent(one extent in B), then insert two same
2218 * backrefs(both refer to the extent in B).
2219 */
2220 return 0;
2221}
2222
2223static void backref_insert(struct rb_root *root,
2224 struct sa_defrag_extent_backref *backref)
2225{
2226 struct rb_node **p = &root->rb_node;
2227 struct rb_node *parent = NULL;
2228 struct sa_defrag_extent_backref *entry;
2229 int ret;
2230
2231 while (*p) {
2232 parent = *p;
2233 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2234
2235 ret = backref_comp(backref, entry);
2236 if (ret < 0)
2237 p = &(*p)->rb_left;
2238 else
2239 p = &(*p)->rb_right;
2240 }
2241
2242 rb_link_node(&backref->node, parent, p);
2243 rb_insert_color(&backref->node, root);
2244}
2245
2246/*
2247 * Note the backref might has changed, and in this case we just return 0.
2248 */
2249static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2250 void *ctx)
2251{
2252 struct btrfs_file_extent_item *extent;
2253 struct btrfs_fs_info *fs_info;
2254 struct old_sa_defrag_extent *old = ctx;
2255 struct new_sa_defrag_extent *new = old->new;
2256 struct btrfs_path *path = new->path;
2257 struct btrfs_key key;
2258 struct btrfs_root *root;
2259 struct sa_defrag_extent_backref *backref;
2260 struct extent_buffer *leaf;
2261 struct inode *inode = new->inode;
2262 int slot;
2263 int ret;
2264 u64 extent_offset;
2265 u64 num_bytes;
2266
2267 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2268 inum == btrfs_ino(inode))
2269 return 0;
2270
2271 key.objectid = root_id;
2272 key.type = BTRFS_ROOT_ITEM_KEY;
2273 key.offset = (u64)-1;
2274
2275 fs_info = BTRFS_I(inode)->root->fs_info;
2276 root = btrfs_read_fs_root_no_name(fs_info, &key);
2277 if (IS_ERR(root)) {
2278 if (PTR_ERR(root) == -ENOENT)
2279 return 0;
2280 WARN_ON(1);
2281 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2282 inum, offset, root_id);
2283 return PTR_ERR(root);
2284 }
2285
2286 key.objectid = inum;
2287 key.type = BTRFS_EXTENT_DATA_KEY;
2288 if (offset > (u64)-1 << 32)
2289 key.offset = 0;
2290 else
2291 key.offset = offset;
2292
2293 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
fae7f21c 2294 if (WARN_ON(ret < 0))
38c227d8 2295 return ret;
50f1319c 2296 ret = 0;
38c227d8
LB		 2297
2298 while (1) {
2299 cond_resched();
2300
2301 leaf = path->nodes[0];
2302 slot = path->slots[0];
2303
2304 if (slot >= btrfs_header_nritems(leaf)) {
2305 ret = btrfs_next_leaf(root, path);
2306 if (ret < 0) {
2307 goto out;
2308 } else if (ret > 0) {
2309 ret = 0;
2310 goto out;
2311 }
2312 continue;
2313 }
2314
2315 path->slots[0]++;
2316
2317 btrfs_item_key_to_cpu(leaf, &key, slot);
2318
2319 if (key.objectid > inum)
2320 goto out;
2321
2322 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2323 continue;
2324
2325 extent = btrfs_item_ptr(leaf, slot,
2326 struct btrfs_file_extent_item);
2327
2328 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2329 continue;
2330
e68afa49
LB		 2331 /*
2332 * 'offset' refers to the exact key.offset,
2333 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2334 * (key.offset - extent_offset).
2335 */
2336 if (key.offset != offset)
38c227d8
LB		 2337 continue;
2338
e68afa49 2339 extent_offset = btrfs_file_extent_offset(leaf, extent);
38c227d8 2340 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
e68afa49 2341
38c227d8
LB		 2342 if (extent_offset >= old->extent_offset + old->offset +
2343 old->len || extent_offset + num_bytes <=
2344 old->extent_offset + old->offset)
2345 continue;
38c227d8
LB		 2346 break;
2347 }
2348
2349 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2350 if (!backref) {
2351 ret = -ENOENT;
2352 goto out;
2353 }
2354
2355 backref->root_id = root_id;
2356 backref->inum = inum;
e68afa49 2357 backref->file_pos = offset;
38c227d8
LB		 2358 backref->num_bytes = num_bytes;
2359 backref->extent_offset = extent_offset;
2360 backref->generation = btrfs_file_extent_generation(leaf, extent);
2361 backref->old = old;
2362 backref_insert(&new->root, backref);
2363 old->count++;
2364out:
2365 btrfs_release_path(path);
2366 WARN_ON(ret);
2367 return ret;
2368}
2369
2370static noinline bool record_extent_backrefs(struct btrfs_path *path,
2371 struct new_sa_defrag_extent *new)
2372{
2373 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2374 struct old_sa_defrag_extent *old, *tmp;
2375 int ret;
2376
2377 new->path = path;
2378
2379 list_for_each_entry_safe(old, tmp, &new->head, list) {
e68afa49
LB		 2380 ret = iterate_inodes_from_logical(old->bytenr +
2381 old->extent_offset, fs_info,
38c227d8
LB		 2382 path, record_one_backref,
2383 old);
4724b106
JB		 2384 if (ret < 0 && ret != -ENOENT)
2385 return false;
38c227d8
LB		 2386
2387 /* no backref to be processed for this extent */
2388 if (!old->count) {
2389 list_del(&old->list);
2390 kfree(old);
2391 }
2392 }
2393
2394 if (list_empty(&new->head))
2395 return false;
2396
2397 return true;
2398}
2399
2400static int relink_is_mergable(struct extent_buffer *leaf,
2401 struct btrfs_file_extent_item *fi,
116e0024 2402 struct new_sa_defrag_extent *new)
38c227d8 2403{
116e0024 2404 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
38c227d8
LB		 2405 return 0;
2406
2407 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2408 return 0;
2409
116e0024
LB		 2410 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2411 return 0;
2412
2413 if (btrfs_file_extent_encryption(leaf, fi) ||
38c227d8
LB		 2414 btrfs_file_extent_other_encoding(leaf, fi))
2415 return 0;
2416
2417 return 1;
2418}
2419
2420/*
2421 * Note the backref might has changed, and in this case we just return 0.
2422 */
2423static noinline int relink_extent_backref(struct btrfs_path *path,
2424 struct sa_defrag_extent_backref *prev,
2425 struct sa_defrag_extent_backref *backref)
2426{
2427 struct btrfs_file_extent_item *extent;
2428 struct btrfs_file_extent_item *item;
2429 struct btrfs_ordered_extent *ordered;
2430 struct btrfs_trans_handle *trans;
2431 struct btrfs_fs_info *fs_info;
2432 struct btrfs_root *root;
2433 struct btrfs_key key;
2434 struct extent_buffer *leaf;
2435 struct old_sa_defrag_extent *old = backref->old;
2436 struct new_sa_defrag_extent *new = old->new;
2437 struct inode *src_inode = new->inode;
2438 struct inode *inode;
2439 struct extent_state *cached = NULL;
2440 int ret = 0;
2441 u64 start;
2442 u64 len;
2443 u64 lock_start;
2444 u64 lock_end;
2445 bool merge = false;
2446 int index;
2447
2448 if (prev && prev->root_id == backref->root_id &&
2449 prev->inum == backref->inum &&
2450 prev->file_pos + prev->num_bytes == backref->file_pos)
2451 merge = true;
2452
2453 /* step 1: get root */
2454 key.objectid = backref->root_id;
2455 key.type = BTRFS_ROOT_ITEM_KEY;
2456 key.offset = (u64)-1;
2457
2458 fs_info = BTRFS_I(src_inode)->root->fs_info;
2459 index = srcu_read_lock(&fs_info->subvol_srcu);
2460
2461 root = btrfs_read_fs_root_no_name(fs_info, &key);
2462 if (IS_ERR(root)) {
2463 srcu_read_unlock(&fs_info->subvol_srcu, index);
2464 if (PTR_ERR(root) == -ENOENT)
2465 return 0;
2466 return PTR_ERR(root);
2467 }
38c227d8 2468
bcbba5e6
WS		 2469 if (btrfs_root_readonly(root)) {
2470 srcu_read_unlock(&fs_info->subvol_srcu, index);
2471 return 0;
2472 }
2473
38c227d8
LB		 2474 /* step 2: get inode */
2475 key.objectid = backref->inum;
2476 key.type = BTRFS_INODE_ITEM_KEY;
2477 key.offset = 0;
2478
2479 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2480 if (IS_ERR(inode)) {
2481 srcu_read_unlock(&fs_info->subvol_srcu, index);
2482 return 0;
2483 }
2484
2485 srcu_read_unlock(&fs_info->subvol_srcu, index);
2486
2487 /* step 3: relink backref */
2488 lock_start = backref->file_pos;
2489 lock_end = backref->file_pos + backref->num_bytes - 1;
2490 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
ff13db41 2491 &cached);
38c227d8
LB		 2492
2493 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2494 if (ordered) {
2495 btrfs_put_ordered_extent(ordered);
2496 goto out_unlock;
2497 }
2498
2499 trans = btrfs_join_transaction(root);
2500 if (IS_ERR(trans)) {
2501 ret = PTR_ERR(trans);
2502 goto out_unlock;
2503 }
2504
2505 key.objectid = backref->inum;
2506 key.type = BTRFS_EXTENT_DATA_KEY;
2507 key.offset = backref->file_pos;
2508
2509 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2510 if (ret < 0) {
2511 goto out_free_path;
2512 } else if (ret > 0) {
2513 ret = 0;
2514 goto out_free_path;
2515 }
2516
2517 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2518 struct btrfs_file_extent_item);
2519
2520 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2521 backref->generation)
2522 goto out_free_path;
2523
2524 btrfs_release_path(path);
2525
2526 start = backref->file_pos;
2527 if (backref->extent_offset < old->extent_offset + old->offset)
2528 start += old->extent_offset + old->offset -
2529 backref->extent_offset;
2530
2531 len = min(backref->extent_offset + backref->num_bytes,
2532 old->extent_offset + old->offset + old->len);
2533 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2534
2535 ret = btrfs_drop_extents(trans, root, inode, start,
2536 start + len, 1);
2537 if (ret)
2538 goto out_free_path;
2539again:
2540 key.objectid = btrfs_ino(inode);
2541 key.type = BTRFS_EXTENT_DATA_KEY;
2542 key.offset = start;
2543
a09a0a70 2544 path->leave_spinning = 1;
38c227d8
LB		 2545 if (merge) {
2546 struct btrfs_file_extent_item *fi;
2547 u64 extent_len;
2548 struct btrfs_key found_key;
2549
3c9665df 2550 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
38c227d8
LB		 2551 if (ret < 0)
2552 goto out_free_path;
2553
2554 path->slots[0]--;
2555 leaf = path->nodes[0];
2556 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2557
2558 fi = btrfs_item_ptr(leaf, path->slots[0],
2559 struct btrfs_file_extent_item);
2560 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2561
116e0024
LB		 2562 if (extent_len + found_key.offset == start &&
2563 relink_is_mergable(leaf, fi, new)) {
38c227d8
LB		 2564 btrfs_set_file_extent_num_bytes(leaf, fi,
2565 extent_len + len);
2566 btrfs_mark_buffer_dirty(leaf);
2567 inode_add_bytes(inode, len);
2568
2569 ret = 1;
2570 goto out_free_path;
2571 } else {
2572 merge = false;
2573 btrfs_release_path(path);
2574 goto again;
2575 }
2576 }
2577
2578 ret = btrfs_insert_empty_item(trans, root, path, &key,
2579 sizeof(*extent));
2580 if (ret) {
2581 btrfs_abort_transaction(trans, root, ret);
2582 goto out_free_path;
2583 }
2584
2585 leaf = path->nodes[0];
2586 item = btrfs_item_ptr(leaf, path->slots[0],
2587 struct btrfs_file_extent_item);
2588 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2589 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2590 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2591 btrfs_set_file_extent_num_bytes(leaf, item, len);
2592 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2593 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2594 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2595 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2596 btrfs_set_file_extent_encryption(leaf, item, 0);
2597 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2598
2599 btrfs_mark_buffer_dirty(leaf);
2600 inode_add_bytes(inode, len);
a09a0a70 2601 btrfs_release_path(path);
38c227d8
LB		 2602
2603 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2604 new->disk_len, 0,
2605 backref->root_id, backref->inum,
b06c4bf5 2606 new->file_pos); /* start - extent_offset */
38c227d8
LB		 2607 if (ret) {
2608 btrfs_abort_transaction(trans, root, ret);
2609 goto out_free_path;
2610 }
2611
2612 ret = 1;
2613out_free_path:
2614 btrfs_release_path(path);
a09a0a70 2615 path->leave_spinning = 0;
38c227d8
LB		 2616 btrfs_end_transaction(trans, root);
2617out_unlock:
2618 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2619 &cached, GFP_NOFS);
2620 iput(inode);
2621 return ret;
2622}
2623
6f519564
LB		 2624static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2625{
2626 struct old_sa_defrag_extent *old, *tmp;
2627
2628 if (!new)
2629 return;
2630
2631 list_for_each_entry_safe(old, tmp, &new->head, list) {
6f519564
LB		 2632 kfree(old);
2633 }
2634 kfree(new);
2635}
2636
38c227d8
LB		 2637static void relink_file_extents(struct new_sa_defrag_extent *new)
2638{
2639 struct btrfs_path *path;
38c227d8
LB		 2640 struct sa_defrag_extent_backref *backref;
2641 struct sa_defrag_extent_backref *prev = NULL;
2642 struct inode *inode;
2643 struct btrfs_root *root;
2644 struct rb_node *node;
2645 int ret;
2646
2647 inode = new->inode;
2648 root = BTRFS_I(inode)->root;
2649
2650 path = btrfs_alloc_path();
2651 if (!path)
2652 return;
2653
2654 if (!record_extent_backrefs(path, new)) {
2655 btrfs_free_path(path);
2656 goto out;
2657 }
2658 btrfs_release_path(path);
2659
2660 while (1) {
2661 node = rb_first(&new->root);
2662 if (!node)
2663 break;
2664 rb_erase(node, &new->root);
2665
2666 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2667
2668 ret = relink_extent_backref(path, prev, backref);
2669 WARN_ON(ret < 0);
2670
2671 kfree(prev);
2672
2673 if (ret == 1)
2674 prev = backref;
2675 else
2676 prev = NULL;
2677 cond_resched();
2678 }
2679 kfree(prev);
2680
2681 btrfs_free_path(path);
38c227d8 2682out:
6f519564
LB		 2683 free_sa_defrag_extent(new);
2684
38c227d8
LB		 2685 atomic_dec(&root->fs_info->defrag_running);
2686 wake_up(&root->fs_info->transaction_wait);
38c227d8
LB		 2687}
2688
2689static struct new_sa_defrag_extent *
2690record_old_file_extents(struct inode *inode,
2691 struct btrfs_ordered_extent *ordered)
2692{
2693 struct btrfs_root *root = BTRFS_I(inode)->root;
2694 struct btrfs_path *path;
2695 struct btrfs_key key;
6f519564 2696 struct old_sa_defrag_extent *old;
38c227d8
LB		 2697 struct new_sa_defrag_extent *new;
2698 int ret;
2699
2700 new = kmalloc(sizeof(*new), GFP_NOFS);
2701 if (!new)
2702 return NULL;
2703
2704 new->inode = inode;
2705 new->file_pos = ordered->file_offset;
2706 new->len = ordered->len;
2707 new->bytenr = ordered->start;
2708 new->disk_len = ordered->disk_len;
2709 new->compress_type = ordered->compress_type;
2710 new->root = RB_ROOT;
2711 INIT_LIST_HEAD(&new->head);
2712
2713 path = btrfs_alloc_path();
2714 if (!path)
2715 goto out_kfree;
2716
2717 key.objectid = btrfs_ino(inode);
2718 key.type = BTRFS_EXTENT_DATA_KEY;
2719 key.offset = new->file_pos;
2720
2721 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2722 if (ret < 0)
2723 goto out_free_path;
2724 if (ret > 0 && path->slots[0] > 0)
2725 path->slots[0]--;
2726
2727 /* find out all the old extents for the file range */
2728 while (1) {
2729 struct btrfs_file_extent_item *extent;
2730 struct extent_buffer *l;
2731 int slot;
2732 u64 num_bytes;
2733 u64 offset;
2734 u64 end;
2735 u64 disk_bytenr;
2736 u64 extent_offset;
2737
2738 l = path->nodes[0];
2739 slot = path->slots[0];
2740
2741 if (slot >= btrfs_header_nritems(l)) {
2742 ret = btrfs_next_leaf(root, path);
2743 if (ret < 0)
6f519564 2744 goto out_free_path;
38c227d8
LB		 2745 else if (ret > 0)
2746 break;
2747 continue;
2748 }
2749
2750 btrfs_item_key_to_cpu(l, &key, slot);
2751
2752 if (key.objectid != btrfs_ino(inode))
2753 break;
2754 if (key.type != BTRFS_EXTENT_DATA_KEY)
2755 break;
2756 if (key.offset >= new->file_pos + new->len)
2757 break;
2758
2759 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2760
2761 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2762 if (key.offset + num_bytes < new->file_pos)
2763 goto next;
2764
2765 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2766 if (!disk_bytenr)
2767 goto next;
2768
2769 extent_offset = btrfs_file_extent_offset(l, extent);
2770
2771 old = kmalloc(sizeof(*old), GFP_NOFS);
2772 if (!old)
6f519564 2773 goto out_free_path;
38c227d8
LB		 2774
2775 offset = max(new->file_pos, key.offset);
2776 end = min(new->file_pos + new->len, key.offset + num_bytes);
2777
2778 old->bytenr = disk_bytenr;
2779 old->extent_offset = extent_offset;
2780 old->offset = offset - key.offset;
2781 old->len = end - offset;
2782 old->new = new;
2783 old->count = 0;
2784 list_add_tail(&old->list, &new->head);
2785next:
2786 path->slots[0]++;
2787 cond_resched();
2788 }
2789
2790 btrfs_free_path(path);
2791 atomic_inc(&root->fs_info->defrag_running);
2792
2793 return new;
2794
38c227d8
LB		 2795out_free_path:
2796 btrfs_free_path(path);
2797out_kfree:
6f519564 2798 free_sa_defrag_extent(new);
38c227d8
LB		 2799 return NULL;
2800}
2801
e570fd27
MX		 2802static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2803 u64 start, u64 len)
2804{
2805 struct btrfs_block_group_cache *cache;
2806
2807 cache = btrfs_lookup_block_group(root->fs_info, start);
2808 ASSERT(cache);
2809
2810 spin_lock(&cache->lock);
2811 cache->delalloc_bytes -= len;
2812 spin_unlock(&cache->lock);
2813
2814 btrfs_put_block_group(cache);
2815}
2816
d352ac68
CM		 2817/* as ordered data IO finishes, this gets called so we can finish
2818 * an ordered extent if the range of bytes in the file it covers are
2819 * fully written.
2820 */
5fd02043 2821static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
e6dcd2dc 2822{
5fd02043 2823 struct inode *inode = ordered_extent->inode;
e6dcd2dc 2824 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 2825 struct btrfs_trans_handle *trans = NULL;
e6dcd2dc 2826 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2ac55d41 2827 struct extent_state *cached_state = NULL;
38c227d8 2828 struct new_sa_defrag_extent *new = NULL;
261507a0 2829 int compress_type = 0;
77cef2ec
JB		 2830 int ret = 0;
2831 u64 logical_len = ordered_extent->len;
82d5902d 2832 bool nolock;
77cef2ec 2833 bool truncated = false;
e6dcd2dc 2834
83eea1f1 2835 nolock = btrfs_is_free_space_inode(inode);
0cb59c99 2836
5fd02043
JB		 2837 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2838 ret = -EIO;
2839 goto out;
2840 }
2841
f612496b
MX		 2842 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2843 ordered_extent->file_offset +
2844 ordered_extent->len - 1);
2845
77cef2ec
JB		 2846 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2847 truncated = true;
2848 logical_len = ordered_extent->truncated_len;
2849 /* Truncated the entire extent, don't bother adding */
2850 if (!logical_len)
2851 goto out;
2852 }
2853
c2167754 2854 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
79787eaa 2855 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
94ed938a
QW		 2856
2857 /*
2858 * For mwrite(mmap + memset to write) case, we still reserve
2859 * space for NOCOW range.
2860 * As NOCOW won't cause a new delayed ref, just free the space
2861 */
2862 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2863 ordered_extent->len);
6c760c07
JB		 2864 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2865 if (nolock)
2866 trans = btrfs_join_transaction_nolock(root);
2867 else
2868 trans = btrfs_join_transaction(root);
2869 if (IS_ERR(trans)) {
2870 ret = PTR_ERR(trans);
2871 trans = NULL;
2872 goto out;
c2167754 2873 }
6c760c07
JB		 2874 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2875 ret = btrfs_update_inode_fallback(trans, root, inode);
2876 if (ret) /* -ENOMEM or corruption */
2877 btrfs_abort_transaction(trans, root, ret);
c2167754
YZ		 2878 goto out;
2879 }
e6dcd2dc 2880
2ac55d41
JB		 2881 lock_extent_bits(io_tree, ordered_extent->file_offset,
2882 ordered_extent->file_offset + ordered_extent->len - 1,
ff13db41 2883 &cached_state);
e6dcd2dc 2884
38c227d8
LB		 2885 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2886 ordered_extent->file_offset + ordered_extent->len - 1,
2887 EXTENT_DEFRAG, 1, cached_state);
2888 if (ret) {
2889 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
8101c8db 2890 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
38c227d8
LB		 2891 /* the inode is shared */
2892 new = record_old_file_extents(inode, ordered_extent);
2893
2894 clear_extent_bit(io_tree, ordered_extent->file_offset,
2895 ordered_extent->file_offset + ordered_extent->len - 1,
2896 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2897 }
2898
0cb59c99 2899 if (nolock)
7a7eaa40 2900 trans = btrfs_join_transaction_nolock(root);
0cb59c99 2901 else
7a7eaa40 2902 trans = btrfs_join_transaction(root);
79787eaa
JM		 2903 if (IS_ERR(trans)) {
2904 ret = PTR_ERR(trans);
2905 trans = NULL;
2906 goto out_unlock;
2907 }
a79b7d4b 2908
0ca1f7ce 2909 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
c2167754 2910
c8b97818 2911 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
261507a0 2912 compress_type = ordered_extent->compress_type;
d899e052 2913 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
261507a0 2914 BUG_ON(compress_type);
920bbbfb 2915 ret = btrfs_mark_extent_written(trans, inode,
d899e052
YZ		 2916 ordered_extent->file_offset,
2917 ordered_extent->file_offset +
77cef2ec 2918 logical_len);
d899e052 2919 } else {
0af3d00b 2920 BUG_ON(root == root->fs_info->tree_root);
d899e052
YZ		 2921 ret = insert_reserved_file_extent(trans, inode,
2922 ordered_extent->file_offset,
2923 ordered_extent->start,
2924 ordered_extent->disk_len,
77cef2ec 2925 logical_len, logical_len,
261507a0 2926 compress_type, 0, 0,
d899e052 2927 BTRFS_FILE_EXTENT_REG);
e570fd27
MX		 2928 if (!ret)
2929 btrfs_release_delalloc_bytes(root,
2930 ordered_extent->start,
2931 ordered_extent->disk_len);
d899e052 2932 }
5dc562c5
JB		 2933 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2934 ordered_extent->file_offset, ordered_extent->len,
2935 trans->transid);
79787eaa
JM		 2936 if (ret < 0) {
2937 btrfs_abort_transaction(trans, root, ret);
5fd02043 2938 goto out_unlock;
79787eaa 2939 }
2ac55d41 2940
e6dcd2dc
CM		 2941 add_pending_csums(trans, inode, ordered_extent->file_offset,
2942 &ordered_extent->list);
2943
6c760c07
JB		 2944 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2945 ret = btrfs_update_inode_fallback(trans, root, inode);
2946 if (ret) { /* -ENOMEM or corruption */
2947 btrfs_abort_transaction(trans, root, ret);
2948 goto out_unlock;
1ef30be1
JB		 2949 }
2950 ret = 0;
5fd02043
JB		 2951out_unlock:
2952 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2953 ordered_extent->file_offset +
2954 ordered_extent->len - 1, &cached_state, GFP_NOFS);
c2167754 2955out:
5b0e95bf 2956 if (root != root->fs_info->tree_root)
0cb59c99 2957 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
a698d075
MX		 2958 if (trans)
2959 btrfs_end_transaction(trans, root);
0cb59c99 2960
77cef2ec
JB		 2961 if (ret || truncated) {
2962 u64 start, end;
2963
2964 if (truncated)
2965 start = ordered_extent->file_offset + logical_len;
2966 else
2967 start = ordered_extent->file_offset;
2968 end = ordered_extent->file_offset + ordered_extent->len - 1;
2969 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2970
2971 /* Drop the cache for the part of the extent we didn't write. */
2972 btrfs_drop_extent_cache(inode, start, end, 0);
5fd02043 2973
0bec9ef5
JB		 2974 /*
2975 * If the ordered extent had an IOERR or something else went
2976 * wrong we need to return the space for this ordered extent
77cef2ec
JB		 2977 * back to the allocator. We only free the extent in the
2978 * truncated case if we didn't write out the extent at all.
0bec9ef5 2979 */
77cef2ec
JB		 2980 if ((ret || !logical_len) &&
2981 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
0bec9ef5
JB		 2982 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2983 btrfs_free_reserved_extent(root, ordered_extent->start,
e570fd27 2984 ordered_extent->disk_len, 1);
0bec9ef5
JB		 2985 }
2986
2987
5fd02043 2988 /*
8bad3c02
LB		 2989 * This needs to be done to make sure anybody waiting knows we are done
2990 * updating everything for this ordered extent.
5fd02043
JB		 2991 */
2992 btrfs_remove_ordered_extent(inode, ordered_extent);
2993
38c227d8 2994 /* for snapshot-aware defrag */
6f519564
LB		 2995 if (new) {
2996 if (ret) {
2997 free_sa_defrag_extent(new);
2998 atomic_dec(&root->fs_info->defrag_running);
2999 } else {
3000 relink_file_extents(new);
3001 }
3002 }
38c227d8 3003
e6dcd2dc
CM		 3004 /* once for us */
3005 btrfs_put_ordered_extent(ordered_extent);
3006 /* once for the tree */
3007 btrfs_put_ordered_extent(ordered_extent);
3008
5fd02043
JB		 3009 return ret;
3010}
3011
3012static void finish_ordered_fn(struct btrfs_work *work)
3013{
3014 struct btrfs_ordered_extent *ordered_extent;
3015 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3016 btrfs_finish_ordered_io(ordered_extent);
e6dcd2dc
CM		 3017}
3018
b2950863 3019static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM		 3020 struct extent_state *state, int uptodate)
3021{
5fd02043
JB		 3022 struct inode *inode = page->mapping->host;
3023 struct btrfs_root *root = BTRFS_I(inode)->root;
3024 struct btrfs_ordered_extent *ordered_extent = NULL;
9e0af237
LB		 3025 struct btrfs_workqueue *wq;
3026 btrfs_work_func_t func;
5fd02043 3027
1abe9b8a 3028 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3029
8b62b72b 3030 ClearPagePrivate2(page);
5fd02043
JB		 3031 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3032 end - start + 1, uptodate))
3033 return 0;
3034
9e0af237
LB		 3035 if (btrfs_is_free_space_inode(inode)) {
3036 wq = root->fs_info->endio_freespace_worker;
3037 func = btrfs_freespace_write_helper;
3038 } else {
3039 wq = root->fs_info->endio_write_workers;
3040 func = btrfs_endio_write_helper;
3041 }
5fd02043 3042
9e0af237
LB		 3043 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3044 NULL);
3045 btrfs_queue_work(wq, &ordered_extent->work);
5fd02043
JB		 3046
3047 return 0;
211f90e6
CM		 3048}
3049
dc380aea
MX		 3050static int __readpage_endio_check(struct inode *inode,
3051 struct btrfs_io_bio *io_bio,
3052 int icsum, struct page *page,
3053 int pgoff, u64 start, size_t len)
3054{
3055 char *kaddr;
3056 u32 csum_expected;
3057 u32 csum = ~(u32)0;
dc380aea
MX		 3058
3059 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3060
3061 kaddr = kmap_atomic(page);
3062 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3063 btrfs_csum_final(csum, (char *)&csum);
3064 if (csum != csum_expected)
3065 goto zeroit;
3066
3067 kunmap_atomic(kaddr);
3068 return 0;
3069zeroit:
94647322
DS		 3070 btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3071 "csum failed ino %llu off %llu csum %u expected csum %u",
dc380aea
MX		 3072 btrfs_ino(inode), start, csum, csum_expected);
3073 memset(kaddr + pgoff, 1, len);
3074 flush_dcache_page(page);
3075 kunmap_atomic(kaddr);
3076 if (csum_expected == 0)
3077 return 0;
3078 return -EIO;
3079}
3080
d352ac68
CM		 3081/*
3082 * when reads are done, we need to check csums to verify the data is correct
4a54c8c1
JS		 3083 * if there's a match, we allow the bio to finish. If not, the code in
3084 * extent_io.c will try to find good copies for us.
d352ac68 3085 */
facc8a22
MX		 3086static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3087 u64 phy_offset, struct page *page,
3088 u64 start, u64 end, int mirror)
07157aac 3089{
4eee4fa4 3090 size_t offset = start - page_offset(page);
07157aac 3091 struct inode *inode = page->mapping->host;
d1310b2e 3092 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
ff79f819 3093 struct btrfs_root *root = BTRFS_I(inode)->root;
d1310b2e 3094
d20f7043
CM		 3095 if (PageChecked(page)) {
3096 ClearPageChecked(page);
dc380aea 3097 return 0;
d20f7043 3098 }
6cbff00f
CH		 3099
3100 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
dc380aea 3101 return 0;
17d217fe
YZ		 3102
3103 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
9655d298 3104 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
17d217fe
YZ		 3105 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
3106 GFP_NOFS);
b6cda9bc 3107 return 0;
17d217fe 3108 }
d20f7043 3109
facc8a22 3110 phy_offset >>= inode->i_sb->s_blocksize_bits;
dc380aea
MX		 3111 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3112 start, (size_t)(end - start + 1));
07157aac 3113}
b888db2b 3114
24bbcf04
YZ		 3115void btrfs_add_delayed_iput(struct inode *inode)
3116{
3117 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
8089fe62 3118 struct btrfs_inode *binode = BTRFS_I(inode);
24bbcf04
YZ		 3119
3120 if (atomic_add_unless(&inode->i_count, -1, 1))
3121 return;
3122
24bbcf04 3123 spin_lock(&fs_info->delayed_iput_lock);
8089fe62
DS		 3124 if (binode->delayed_iput_count == 0) {
3125 ASSERT(list_empty(&binode->delayed_iput));
3126 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3127 } else {
3128 binode->delayed_iput_count++;
3129 }
24bbcf04
YZ		 3130 spin_unlock(&fs_info->delayed_iput_lock);
3131}
3132
3133void btrfs_run_delayed_iputs(struct btrfs_root *root)
3134{
24bbcf04 3135 struct btrfs_fs_info *fs_info = root->fs_info;
24bbcf04 3136
24bbcf04 3137 spin_lock(&fs_info->delayed_iput_lock);
8089fe62
DS		 3138 while (!list_empty(&fs_info->delayed_iputs)) {
3139 struct btrfs_inode *inode;
3140
3141 inode = list_first_entry(&fs_info->delayed_iputs,
3142 struct btrfs_inode, delayed_iput);
3143 if (inode->delayed_iput_count) {
3144 inode->delayed_iput_count--;
3145 list_move_tail(&inode->delayed_iput,
3146 &fs_info->delayed_iputs);
3147 } else {
3148 list_del_init(&inode->delayed_iput);
3149 }
3150 spin_unlock(&fs_info->delayed_iput_lock);
3151 iput(&inode->vfs_inode);
3152 spin_lock(&fs_info->delayed_iput_lock);
24bbcf04 3153 }
8089fe62 3154 spin_unlock(&fs_info->delayed_iput_lock);
24bbcf04
YZ		 3155}
3156
d68fc57b 3157/*
42b2aa86 3158 * This is called in transaction commit time. If there are no orphan
d68fc57b
YZ		 3159 * files in the subvolume, it removes orphan item and frees block_rsv
3160 * structure.
3161 */
3162void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3163 struct btrfs_root *root)
3164{
90290e19 3165 struct btrfs_block_rsv *block_rsv;
d68fc57b
YZ		 3166 int ret;
3167
8a35d95f 3168 if (atomic_read(&root->orphan_inodes) ||
d68fc57b
YZ		 3169 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3170 return;
3171
90290e19 3172 spin_lock(&root->orphan_lock);
8a35d95f 3173 if (atomic_read(&root->orphan_inodes)) {
90290e19
JB		 3174 spin_unlock(&root->orphan_lock);
3175 return;
3176 }
3177
3178 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3179 spin_unlock(&root->orphan_lock);
3180 return;
3181 }
3182
3183 block_rsv = root->orphan_block_rsv;
3184 root->orphan_block_rsv = NULL;
3185 spin_unlock(&root->orphan_lock);
3186
27cdeb70 3187 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
d68fc57b
YZ		 3188 btrfs_root_refs(&root->root_item) > 0) {
3189 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3190 root->root_key.objectid);
4ef31a45
JB		 3191 if (ret)
3192 btrfs_abort_transaction(trans, root, ret);
3193 else
27cdeb70
MX		 3194 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3195 &root->state);
d68fc57b
YZ		 3196 }
3197
90290e19
JB		 3198 if (block_rsv) {
3199 WARN_ON(block_rsv->size > 0);
3200 btrfs_free_block_rsv(root, block_rsv);
d68fc57b
YZ		 3201 }
3202}
3203
7b128766
JB		 3204/*
3205 * This creates an orphan entry for the given inode in case something goes
3206 * wrong in the middle of an unlink/truncate.
d68fc57b
YZ		 3207 *
3208 * NOTE: caller of this function should reserve 5 units of metadata for
3209 * this function.
7b128766
JB		 3210 */
3211int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3212{
3213 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 3214 struct btrfs_block_rsv *block_rsv = NULL;
3215 int reserve = 0;
3216 int insert = 0;
3217 int ret;
7b128766 3218
d68fc57b 3219 if (!root->orphan_block_rsv) {
66d8f3dd 3220 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
b532402e
TI		 3221 if (!block_rsv)
3222 return -ENOMEM;
d68fc57b 3223 }
7b128766 3224
d68fc57b
YZ		 3225 spin_lock(&root->orphan_lock);
3226 if (!root->orphan_block_rsv) {
3227 root->orphan_block_rsv = block_rsv;
3228 } else if (block_rsv) {
3229 btrfs_free_block_rsv(root, block_rsv);
3230 block_rsv = NULL;
7b128766 3231 }
7b128766 3232
8a35d95f
JB		 3233 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3234 &BTRFS_I(inode)->runtime_flags)) {
d68fc57b
YZ		 3235#if 0
3236 /*
3237 * For proper ENOSPC handling, we should do orphan
3238 * cleanup when mounting. But this introduces backward
3239 * compatibility issue.
3240 */
3241 if (!xchg(&root->orphan_item_inserted, 1))
3242 insert = 2;
3243 else
3244 insert = 1;
3245#endif
3246 insert = 1;
321f0e70 3247 atomic_inc(&root->orphan_inodes);
7b128766
JB		 3248 }
3249
72ac3c0d
JB		 3250 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3251 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3252 reserve = 1;
d68fc57b 3253 spin_unlock(&root->orphan_lock);
7b128766 3254
d68fc57b
YZ		 3255 /* grab metadata reservation from transaction handle */
3256 if (reserve) {
3257 ret = btrfs_orphan_reserve_metadata(trans, inode);
79787eaa 3258 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
d68fc57b 3259 }
7b128766 3260
d68fc57b
YZ		 3261 /* insert an orphan item to track this unlinked/truncated file */
3262 if (insert >= 1) {
33345d01 3263 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
4ef31a45 3264 if (ret) {
703c88e0 3265 atomic_dec(&root->orphan_inodes);
4ef31a45
JB		 3266 if (reserve) {
3267 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3268 &BTRFS_I(inode)->runtime_flags);
3269 btrfs_orphan_release_metadata(inode);
3270 }
3271 if (ret != -EEXIST) {
e8e7cff6
JB		 3272 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3273 &BTRFS_I(inode)->runtime_flags);
4ef31a45
JB		 3274 btrfs_abort_transaction(trans, root, ret);
3275 return ret;
3276 }
79787eaa
JM		 3277 }
3278 ret = 0;
d68fc57b
YZ		 3279 }
3280
3281 /* insert an orphan item to track subvolume contains orphan files */
3282 if (insert >= 2) {
3283 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3284 root->root_key.objectid);
79787eaa
JM		 3285 if (ret && ret != -EEXIST) {
3286 btrfs_abort_transaction(trans, root, ret);
3287 return ret;
3288 }
d68fc57b
YZ		 3289 }
3290 return 0;
7b128766
JB		 3291}
3292
3293/*
3294 * We have done the truncate/delete so we can go ahead and remove the orphan
3295 * item for this particular inode.
3296 */
48a3b636
ES		 3297static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3298 struct inode *inode)
7b128766
JB		 3299{
3300 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 3301 int delete_item = 0;
3302 int release_rsv = 0;
7b128766
JB		 3303 int ret = 0;
3304
d68fc57b 3305 spin_lock(&root->orphan_lock);
8a35d95f
JB		 3306 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3307 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3308 delete_item = 1;
7b128766 3309
72ac3c0d
JB		 3310 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3311 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3312 release_rsv = 1;
d68fc57b 3313 spin_unlock(&root->orphan_lock);
7b128766 3314
703c88e0 3315 if (delete_item) {
8a35d95f 3316 atomic_dec(&root->orphan_inodes);
703c88e0
FDBM		 3317 if (trans)
3318 ret = btrfs_del_orphan_item(trans, root,
3319 btrfs_ino(inode));
8a35d95f 3320 }
7b128766 3321
703c88e0
FDBM		 3322 if (release_rsv)
3323 btrfs_orphan_release_metadata(inode);
3324
4ef31a45 3325 return ret;
7b128766
JB		 3326}
3327
3328/*
3329 * this cleans up any orphans that may be left on the list from the last use
3330 * of this root.
3331 */
66b4ffd1 3332int btrfs_orphan_cleanup(struct btrfs_root *root)
7b128766
JB		 3333{
3334 struct btrfs_path *path;
3335 struct extent_buffer *leaf;
7b128766
JB		 3336 struct btrfs_key key, found_key;
3337 struct btrfs_trans_handle *trans;
3338 struct inode *inode;
8f6d7f4f 3339 u64 last_objectid = 0;
7b128766
JB		 3340 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3341
d68fc57b 3342 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
66b4ffd1 3343 return 0;
c71bf099
YZ		 3344
3345 path = btrfs_alloc_path();
66b4ffd1
JB		 3346 if (!path) {
3347 ret = -ENOMEM;
3348 goto out;
3349 }
e4058b54 3350 path->reada = READA_BACK;
7b128766
JB		 3351
3352 key.objectid = BTRFS_ORPHAN_OBJECTID;
962a298f 3353 key.type = BTRFS_ORPHAN_ITEM_KEY;
7b128766
JB		 3354 key.offset = (u64)-1;
3355
7b128766
JB		 3356 while (1) {
3357 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
66b4ffd1
JB		 3358 if (ret < 0)
3359 goto out;
7b128766
JB		 3360
3361 /*
3362 * if ret == 0 means we found what we were searching for, which
25985edc 3363 * is weird, but possible, so only screw with path if we didn't
7b128766
JB		 3364 * find the key and see if we have stuff that matches
3365 */
3366 if (ret > 0) {
66b4ffd1 3367 ret = 0;
7b128766
JB		 3368 if (path->slots[0] == 0)
3369 break;
3370 path->slots[0]--;
3371 }
3372
3373 /* pull out the item */
3374 leaf = path->nodes[0];
7b128766
JB		 3375 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3376
3377 /* make sure the item matches what we want */
3378 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3379 break;
962a298f 3380 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
7b128766
JB		 3381 break;
3382
3383 /* release the path since we're done with it */
b3b4aa74 3384 btrfs_release_path(path);
7b128766
JB		 3385
3386 /*
3387 * this is where we are basically btrfs_lookup, without the
3388 * crossing root thing. we store the inode number in the
3389 * offset of the orphan item.
3390 */
8f6d7f4f
JB		 3391
3392 if (found_key.offset == last_objectid) {
c2cf52eb
SK		 3393 btrfs_err(root->fs_info,
3394 "Error removing orphan entry, stopping orphan cleanup");
8f6d7f4f
JB		 3395 ret = -EINVAL;
3396 goto out;
3397 }
3398
3399 last_objectid = found_key.offset;
3400
5d4f98a2
YZ		 3401 found_key.objectid = found_key.offset;
3402 found_key.type = BTRFS_INODE_ITEM_KEY;
3403 found_key.offset = 0;
73f73415 3404 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
8c6ffba0 3405 ret = PTR_ERR_OR_ZERO(inode);
a8c9e576 3406 if (ret && ret != -ESTALE)
66b4ffd1 3407 goto out;
7b128766 3408
f8e9e0b0
AJ		 3409 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3410 struct btrfs_root *dead_root;
3411 struct btrfs_fs_info *fs_info = root->fs_info;
3412 int is_dead_root = 0;
3413
3414 /*
3415 * this is an orphan in the tree root. Currently these
3416 * could come from 2 sources:
3417 * a) a snapshot deletion in progress
3418 * b) a free space cache inode
3419 * We need to distinguish those two, as the snapshot
3420 * orphan must not get deleted.
3421 * find_dead_roots already ran before us, so if this
3422 * is a snapshot deletion, we should find the root
3423 * in the dead_roots list
3424 */
3425 spin_lock(&fs_info->trans_lock);
3426 list_for_each_entry(dead_root, &fs_info->dead_roots,
3427 root_list) {
3428 if (dead_root->root_key.objectid ==
3429 found_key.objectid) {
3430 is_dead_root = 1;
3431 break;
3432 }
3433 }
3434 spin_unlock(&fs_info->trans_lock);
3435 if (is_dead_root) {
3436 /* prevent this orphan from being found again */
3437 key.offset = found_key.objectid - 1;
3438 continue;
3439 }
3440 }
7b128766 3441 /*
a8c9e576
JB		 3442 * Inode is already gone but the orphan item is still there,
3443 * kill the orphan item.
7b128766 3444 */
a8c9e576
JB		 3445 if (ret == -ESTALE) {
3446 trans = btrfs_start_transaction(root, 1);
66b4ffd1
JB		 3447 if (IS_ERR(trans)) {
3448 ret = PTR_ERR(trans);
3449 goto out;
3450 }
c2cf52eb
SK		 3451 btrfs_debug(root->fs_info, "auto deleting %Lu",
3452 found_key.objectid);
a8c9e576
JB		 3453 ret = btrfs_del_orphan_item(trans, root,
3454 found_key.objectid);
5b21f2ed 3455 btrfs_end_transaction(trans, root);
4ef31a45
JB		 3456 if (ret)
3457 goto out;
7b128766
JB		 3458 continue;
3459 }
3460
a8c9e576
JB		 3461 /*
3462 * add this inode to the orphan list so btrfs_orphan_del does
3463 * the proper thing when we hit it
3464 */
8a35d95f
JB		 3465 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3466 &BTRFS_I(inode)->runtime_flags);
925396ec 3467 atomic_inc(&root->orphan_inodes);
a8c9e576 3468
7b128766
JB		 3469 /* if we have links, this was a truncate, lets do that */
3470 if (inode->i_nlink) {
fae7f21c 3471 if (WARN_ON(!S_ISREG(inode->i_mode))) {
a41ad394
JB		 3472 iput(inode);
3473 continue;
3474 }
7b128766 3475 nr_truncate++;
f3fe820c
JB		 3476
3477 /* 1 for the orphan item deletion. */
3478 trans = btrfs_start_transaction(root, 1);
3479 if (IS_ERR(trans)) {
c69b26b0 3480 iput(inode);
f3fe820c
JB		 3481 ret = PTR_ERR(trans);
3482 goto out;
3483 }
3484 ret = btrfs_orphan_add(trans, inode);
3485 btrfs_end_transaction(trans, root);
c69b26b0
JB		 3486 if (ret) {
3487 iput(inode);
f3fe820c 3488 goto out;
c69b26b0 3489 }
f3fe820c 3490
66b4ffd1 3491 ret = btrfs_truncate(inode);
4a7d0f68
JB		 3492 if (ret)
3493 btrfs_orphan_del(NULL, inode);
7b128766
JB		 3494 } else {
3495 nr_unlink++;
3496 }
3497
3498 /* this will do delete_inode and everything for us */
3499 iput(inode);
66b4ffd1
JB		 3500 if (ret)
3501 goto out;
7b128766 3502 }
3254c876
MX		 3503 /* release the path since we're done with it */
3504 btrfs_release_path(path);
3505
d68fc57b
YZ		 3506 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3507
3508 if (root->orphan_block_rsv)
3509 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3510 (u64)-1);
3511
27cdeb70
MX		 3512 if (root->orphan_block_rsv ||
3513 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
7a7eaa40 3514 trans = btrfs_join_transaction(root);
66b4ffd1
JB		 3515 if (!IS_ERR(trans))
3516 btrfs_end_transaction(trans, root);
d68fc57b 3517 }
7b128766
JB		 3518
3519 if (nr_unlink)
4884b476 3520 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
7b128766 3521 if (nr_truncate)
4884b476 3522 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
66b4ffd1
JB		 3523
3524out:
3525 if (ret)
68b663d1 3526 btrfs_err(root->fs_info,
c2cf52eb 3527 "could not do orphan cleanup %d", ret);
66b4ffd1
JB		 3528 btrfs_free_path(path);
3529 return ret;
7b128766
JB		 3530}
3531
46a53cca
CM		 3532/*
3533 * very simple check to peek ahead in the leaf looking for xattrs. If we
3534 * don't find any xattrs, we know there can't be any acls.
3535 *
3536 * slot is the slot the inode is in, objectid is the objectid of the inode
3537 */
3538static noinline int acls_after_inode_item(struct extent_buffer *leaf,
63541927
FDBM		 3539 int slot, u64 objectid,
3540 int *first_xattr_slot)
46a53cca
CM		 3541{
3542 u32 nritems = btrfs_header_nritems(leaf);
3543 struct btrfs_key found_key;
f23b5a59
JB		 3544 static u64 xattr_access = 0;
3545 static u64 xattr_default = 0;
46a53cca
CM		 3546 int scanned = 0;
3547
f23b5a59 3548 if (!xattr_access) {
97d79299
AG		 3549 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3550 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3551 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3552 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
f23b5a59
JB		 3553 }
3554
46a53cca 3555 slot++;
63541927 3556 *first_xattr_slot = -1;
46a53cca
CM		 3557 while (slot < nritems) {
3558 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3559
3560 /* we found a different objectid, there must not be acls */
3561 if (found_key.objectid != objectid)
3562 return 0;
3563
3564 /* we found an xattr, assume we've got an acl */
f23b5a59 3565 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
63541927
FDBM		 3566 if (*first_xattr_slot == -1)
3567 *first_xattr_slot = slot;
f23b5a59
JB		 3568 if (found_key.offset == xattr_access ||
3569 found_key.offset == xattr_default)
3570 return 1;
3571 }
46a53cca
CM		 3572
3573 /*
3574 * we found a key greater than an xattr key, there can't
3575 * be any acls later on
3576 */
3577 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3578 return 0;
3579
3580 slot++;
3581 scanned++;
3582
3583 /*
3584 * it goes inode, inode backrefs, xattrs, extents,
3585 * so if there are a ton of hard links to an inode there can
3586 * be a lot of backrefs. Don't waste time searching too hard,
3587 * this is just an optimization
3588 */
3589 if (scanned >= 8)
3590 break;
3591 }
3592 /* we hit the end of the leaf before we found an xattr or
3593 * something larger than an xattr. We have to assume the inode
3594 * has acls
3595 */
63541927
FDBM		 3596 if (*first_xattr_slot == -1)
3597 *first_xattr_slot = slot;
46a53cca
CM		 3598 return 1;
3599}
3600
d352ac68
CM		 3601/*
3602 * read an inode from the btree into the in-memory inode
3603 */
5d4f98a2 3604static void btrfs_read_locked_inode(struct inode *inode)
39279cc3
CM		 3605{
3606 struct btrfs_path *path;
5f39d397 3607 struct extent_buffer *leaf;
39279cc3
CM		 3608 struct btrfs_inode_item *inode_item;
3609 struct btrfs_root *root = BTRFS_I(inode)->root;
3610 struct btrfs_key location;
67de1176 3611 unsigned long ptr;
46a53cca 3612 int maybe_acls;
618e21d5 3613 u32 rdev;
39279cc3 3614 int ret;
2f7e33d4 3615 bool filled = false;
63541927 3616 int first_xattr_slot;
2f7e33d4
MX		 3617
3618 ret = btrfs_fill_inode(inode, &rdev);
3619 if (!ret)
3620 filled = true;
39279cc3
CM		 3621
3622 path = btrfs_alloc_path();
1748f843
MF		 3623 if (!path)
3624 goto make_bad;
3625
39279cc3 3626 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 3627
39279cc3 3628 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 3629 if (ret)
39279cc3 3630 goto make_bad;
39279cc3 3631
5f39d397 3632 leaf = path->nodes[0];
2f7e33d4
MX		 3633
3634 if (filled)
67de1176 3635 goto cache_index;
2f7e33d4 3636
5f39d397
CM		 3637 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3638 struct btrfs_inode_item);
5f39d397 3639 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
bfe86848 3640 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2f2f43d3
EB		 3641 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3642 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
dbe674a9 3643 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397 3644
a937b979
DS		 3645 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3646 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
5f39d397 3647
a937b979
DS		 3648 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3649 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
5f39d397 3650
a937b979
DS		 3651 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3652 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
5f39d397 3653
9cc97d64 3654 BTRFS_I(inode)->i_otime.tv_sec =
3655 btrfs_timespec_sec(leaf, &inode_item->otime);
3656 BTRFS_I(inode)->i_otime.tv_nsec =
3657 btrfs_timespec_nsec(leaf, &inode_item->otime);
5f39d397 3658
a76a3cd4 3659 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 3660 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
5dc562c5
JB		 3661 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3662
6e17d30b
YD		 3663 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3664 inode->i_generation = BTRFS_I(inode)->generation;
3665 inode->i_rdev = 0;
3666 rdev = btrfs_inode_rdev(leaf, inode_item);
3667
3668 BTRFS_I(inode)->index_cnt = (u64)-1;
3669 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3670
3671cache_index:
5dc562c5
JB		 3672 /*
3673 * If we were modified in the current generation and evicted from memory
3674 * and then re-read we need to do a full sync since we don't have any
3675 * idea about which extents were modified before we were evicted from
3676 * cache.
6e17d30b
YD		 3677 *
3678 * This is required for both inode re-read from disk and delayed inode
3679 * in delayed_nodes_tree.
5dc562c5
JB		 3680 */
3681 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3682 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3683 &BTRFS_I(inode)->runtime_flags);
3684
bde6c242
FM		 3685 /*
3686 * We don't persist the id of the transaction where an unlink operation
3687 * against the inode was last made. So here we assume the inode might
3688 * have been evicted, and therefore the exact value of last_unlink_trans
3689 * lost, and set it to last_trans to avoid metadata inconsistencies
3690 * between the inode and its parent if the inode is fsync'ed and the log
3691 * replayed. For example, in the scenario:
3692 *
3693 * touch mydir/foo
3694 * ln mydir/foo mydir/bar
3695 * sync
3696 * unlink mydir/bar
3697 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3698 * xfs_io -c fsync mydir/foo
3699 * <power failure>
3700 * mount fs, triggers fsync log replay
3701 *
3702 * We must make sure that when we fsync our inode foo we also log its
3703 * parent inode, otherwise after log replay the parent still has the
3704 * dentry with the "bar" name but our inode foo has a link count of 1
3705 * and doesn't have an inode ref with the name "bar" anymore.
3706 *
3707 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3708 * but it guarantees correctness at the expense of ocassional full
3709 * transaction commits on fsync if our inode is a directory, or if our
3710 * inode is not a directory, logging its parent unnecessarily.
3711 */
3712 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3713
67de1176
MX		 3714 path->slots[0]++;
3715 if (inode->i_nlink != 1 ||
3716 path->slots[0] >= btrfs_header_nritems(leaf))
3717 goto cache_acl;
3718
3719 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3720 if (location.objectid != btrfs_ino(inode))
3721 goto cache_acl;
3722
3723 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3724 if (location.type == BTRFS_INODE_REF_KEY) {
3725 struct btrfs_inode_ref *ref;
3726
3727 ref = (struct btrfs_inode_ref *)ptr;
3728 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3729 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3730 struct btrfs_inode_extref *extref;
3731
3732 extref = (struct btrfs_inode_extref *)ptr;
3733 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3734 extref);
3735 }
2f7e33d4 3736cache_acl:
46a53cca
CM		 3737 /*
3738 * try to precache a NULL acl entry for files that don't have
3739 * any xattrs or acls
3740 */
33345d01 3741 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
63541927
FDBM		 3742 btrfs_ino(inode), &first_xattr_slot);
3743 if (first_xattr_slot != -1) {
3744 path->slots[0] = first_xattr_slot;
3745 ret = btrfs_load_inode_props(inode, path);
3746 if (ret)
3747 btrfs_err(root->fs_info,
351fd353 3748 "error loading props for ino %llu (root %llu): %d",
63541927
FDBM		 3749 btrfs_ino(inode),
3750 root->root_key.objectid, ret);
3751 }
3752 btrfs_free_path(path);
3753
72c04902
AV		 3754 if (!maybe_acls)
3755 cache_no_acl(inode);
46a53cca 3756
39279cc3 3757 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 3758 case S_IFREG:
3759 inode->i_mapping->a_ops = &btrfs_aops;
d1310b2e 3760 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 3761 inode->i_fop = &btrfs_file_operations;
3762 inode->i_op = &btrfs_file_inode_operations;
3763 break;
3764 case S_IFDIR:
3765 inode->i_fop = &btrfs_dir_file_operations;
3766 if (root == root->fs_info->tree_root)
3767 inode->i_op = &btrfs_dir_ro_inode_operations;
3768 else
3769 inode->i_op = &btrfs_dir_inode_operations;
3770 break;
3771 case S_IFLNK:
3772 inode->i_op = &btrfs_symlink_inode_operations;
21fc61c7 3773 inode_nohighmem(inode);
39279cc3
CM		 3774 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3775 break;
618e21d5 3776 default:
0279b4cd 3777 inode->i_op = &btrfs_special_inode_operations;
618e21d5
JB		 3778 init_special_inode(inode, inode->i_mode, rdev);
3779 break;
39279cc3 3780 }
6cbff00f
CH		 3781
3782 btrfs_update_iflags(inode);
39279cc3
CM		 3783 return;
3784
3785make_bad:
39279cc3 3786 btrfs_free_path(path);
39279cc3
CM		 3787 make_bad_inode(inode);
3788}
3789
d352ac68
CM		 3790/*
3791 * given a leaf and an inode, copy the inode fields into the leaf
3792 */
e02119d5
CM		 3793static void fill_inode_item(struct btrfs_trans_handle *trans,
3794 struct extent_buffer *leaf,
5f39d397 3795 struct btrfs_inode_item *item,
39279cc3
CM		 3796 struct inode *inode)
3797{
51fab693
LB		 3798 struct btrfs_map_token token;
3799
3800 btrfs_init_map_token(&token);
5f39d397 3801
51fab693
LB		 3802 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3803 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3804 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3805 &token);
3806 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3807 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
5f39d397 3808
a937b979 3809 btrfs_set_token_timespec_sec(leaf, &item->atime,
51fab693 3810 inode->i_atime.tv_sec, &token);
a937b979 3811 btrfs_set_token_timespec_nsec(leaf, &item->atime,
51fab693 3812 inode->i_atime.tv_nsec, &token);
5f39d397 3813
a937b979 3814 btrfs_set_token_timespec_sec(leaf, &item->mtime,
51fab693 3815 inode->i_mtime.tv_sec, &token);
a937b979 3816 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
51fab693 3817 inode->i_mtime.tv_nsec, &token);
5f39d397 3818
a937b979 3819 btrfs_set_token_timespec_sec(leaf, &item->ctime,
51fab693 3820 inode->i_ctime.tv_sec, &token);
a937b979 3821 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
51fab693 3822 inode->i_ctime.tv_nsec, &token);
5f39d397 3823
9cc97d64 3824 btrfs_set_token_timespec_sec(leaf, &item->otime,
3825 BTRFS_I(inode)->i_otime.tv_sec, &token);
3826 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3827 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3828
51fab693
LB		 3829 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3830 &token);
3831 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3832 &token);
3833 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3834 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3835 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3836 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3837 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
39279cc3
CM		 3838}
3839
d352ac68
CM		 3840/*
3841 * copy everything in the in-memory inode into the btree.
3842 */
2115133f 3843static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
d397712b 3844 struct btrfs_root *root, struct inode *inode)
39279cc3
CM		 3845{
3846 struct btrfs_inode_item *inode_item;
3847 struct btrfs_path *path;
5f39d397 3848 struct extent_buffer *leaf;
39279cc3
CM		 3849 int ret;
3850
3851 path = btrfs_alloc_path();
16cdcec7
MX		 3852 if (!path)
3853 return -ENOMEM;
3854
b9473439 3855 path->leave_spinning = 1;
16cdcec7
MX		 3856 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3857 1);
39279cc3
CM		 3858 if (ret) {
3859 if (ret > 0)
3860 ret = -ENOENT;
3861 goto failed;
3862 }
3863
5f39d397
CM		 3864 leaf = path->nodes[0];
3865 inode_item = btrfs_item_ptr(leaf, path->slots[0],
16cdcec7 3866 struct btrfs_inode_item);
39279cc3 3867
e02119d5 3868 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 3869 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 3870 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 3871 ret = 0;
3872failed:
39279cc3
CM		 3873 btrfs_free_path(path);
3874 return ret;
3875}
3876
2115133f
CM		 3877/*
3878 * copy everything in the in-memory inode into the btree.
3879 */
3880noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3881 struct btrfs_root *root, struct inode *inode)
3882{
3883 int ret;
3884
3885 /*
3886 * If the inode is a free space inode, we can deadlock during commit
3887 * if we put it into the delayed code.
3888 *
3889 * The data relocation inode should also be directly updated
3890 * without delay
3891 */
83eea1f1 3892 if (!btrfs_is_free_space_inode(inode)
1d52c78a
JB		 3893 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3894 && !root->fs_info->log_root_recovering) {
8ea05e3a
AB		 3895 btrfs_update_root_times(trans, root);
3896
2115133f
CM		 3897 ret = btrfs_delayed_update_inode(trans, root, inode);
3898 if (!ret)
3899 btrfs_set_inode_last_trans(trans, inode);
3900 return ret;
3901 }
3902
3903 return btrfs_update_inode_item(trans, root, inode);
3904}
3905
be6aef60
JB		 3906noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3907 struct btrfs_root *root,
3908 struct inode *inode)
2115133f
CM		 3909{
3910 int ret;
3911
3912 ret = btrfs_update_inode(trans, root, inode);
3913 if (ret == -ENOSPC)
3914 return btrfs_update_inode_item(trans, root, inode);
3915 return ret;
3916}
3917
d352ac68
CM		 3918/*
3919 * unlink helper that gets used here in inode.c and in the tree logging
3920 * recovery code. It remove a link in a directory with a given name, and
3921 * also drops the back refs in the inode to the directory
3922 */
92986796
AV		 3923static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3924 struct btrfs_root *root,
3925 struct inode *dir, struct inode *inode,
3926 const char *name, int name_len)
39279cc3
CM		 3927{
3928 struct btrfs_path *path;
39279cc3 3929 int ret = 0;
5f39d397 3930 struct extent_buffer *leaf;
39279cc3 3931 struct btrfs_dir_item *di;
5f39d397 3932 struct btrfs_key key;
aec7477b 3933 u64 index;
33345d01
LZ		 3934 u64 ino = btrfs_ino(inode);
3935 u64 dir_ino = btrfs_ino(dir);
39279cc3
CM		 3936
3937 path = btrfs_alloc_path();
54aa1f4d
CM		 3938 if (!path) {
3939 ret = -ENOMEM;
554233a6 3940 goto out;
54aa1f4d
CM		 3941 }
3942
b9473439 3943 path->leave_spinning = 1;
33345d01 3944 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
39279cc3
CM		 3945 name, name_len, -1);
3946 if (IS_ERR(di)) {
3947 ret = PTR_ERR(di);
3948 goto err;
3949 }
3950 if (!di) {
3951 ret = -ENOENT;
3952 goto err;
3953 }
5f39d397
CM		 3954 leaf = path->nodes[0];
3955 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 3956 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 3957 if (ret)
3958 goto err;
b3b4aa74 3959 btrfs_release_path(path);
39279cc3 3960
67de1176
MX		 3961 /*
3962 * If we don't have dir index, we have to get it by looking up
3963 * the inode ref, since we get the inode ref, remove it directly,
3964 * it is unnecessary to do delayed deletion.
3965 *
3966 * But if we have dir index, needn't search inode ref to get it.
3967 * Since the inode ref is close to the inode item, it is better
3968 * that we delay to delete it, and just do this deletion when
3969 * we update the inode item.
3970 */
3971 if (BTRFS_I(inode)->dir_index) {
3972 ret = btrfs_delayed_delete_inode_ref(inode);
3973 if (!ret) {
3974 index = BTRFS_I(inode)->dir_index;
3975 goto skip_backref;
3976 }
3977 }
3978
33345d01
LZ		 3979 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3980 dir_ino, &index);
aec7477b 3981 if (ret) {
c2cf52eb
SK		 3982 btrfs_info(root->fs_info,
3983 "failed to delete reference to %.*s, inode %llu parent %llu",
c1c9ff7c 3984 name_len, name, ino, dir_ino);
79787eaa 3985 btrfs_abort_transaction(trans, root, ret);
aec7477b
JB		 3986 goto err;
3987 }
67de1176 3988skip_backref:
16cdcec7 3989 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 3990 if (ret) {
3991 btrfs_abort_transaction(trans, root, ret);
39279cc3 3992 goto err;
79787eaa 3993 }
39279cc3 3994
e02119d5 3995 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
33345d01 3996 inode, dir_ino);
79787eaa
JM		 3997 if (ret != 0 && ret != -ENOENT) {
3998 btrfs_abort_transaction(trans, root, ret);
3999 goto err;
4000 }
e02119d5
CM		 4001
4002 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4003 dir, index);
6418c961
CM		 4004 if (ret == -ENOENT)
4005 ret = 0;
d4e3991b
ZB		 4006 else if (ret)
4007 btrfs_abort_transaction(trans, root, ret);
39279cc3
CM		 4008err:
4009 btrfs_free_path(path);
e02119d5
CM		 4010 if (ret)
4011 goto out;
4012
4013 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95
JB		 4014 inode_inc_iversion(inode);
4015 inode_inc_iversion(dir);
04b285f3
DD		 4016 inode->i_ctime = dir->i_mtime =
4017 dir->i_ctime = current_fs_time(inode->i_sb);
b9959295 4018 ret = btrfs_update_inode(trans, root, dir);
e02119d5 4019out:
39279cc3
CM		 4020 return ret;
4021}
4022
92986796
AV		 4023int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4024 struct btrfs_root *root,
4025 struct inode *dir, struct inode *inode,
4026 const char *name, int name_len)
4027{
4028 int ret;
4029 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4030 if (!ret) {
8b558c5f 4031 drop_nlink(inode);
92986796
AV		 4032 ret = btrfs_update_inode(trans, root, inode);
4033 }
4034 return ret;
4035}
39279cc3 4036
a22285a6
YZ		 4037/*
4038 * helper to start transaction for unlink and rmdir.
4039 *
d52be818
JB		 4040 * unlink and rmdir are special in btrfs, they do not always free space, so
4041 * if we cannot make our reservations the normal way try and see if there is
4042 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4043 * allow the unlink to occur.
a22285a6 4044 */
d52be818 4045static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4df27c4d 4046{
a22285a6 4047 struct btrfs_root *root = BTRFS_I(dir)->root;
4df27c4d 4048
e70bea5f
JB		 4049 /*
4050 * 1 for the possible orphan item
4051 * 1 for the dir item
4052 * 1 for the dir index
4053 * 1 for the inode ref
e70bea5f
JB		 4054 * 1 for the inode
4055 */
8eab77ff 4056 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
a22285a6
YZ		 4057}
4058
4059static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4060{
4061 struct btrfs_root *root = BTRFS_I(dir)->root;
4062 struct btrfs_trans_handle *trans;
2b0143b5 4063 struct inode *inode = d_inode(dentry);
a22285a6 4064 int ret;
a22285a6 4065
d52be818 4066 trans = __unlink_start_trans(dir);
a22285a6
YZ		 4067 if (IS_ERR(trans))
4068 return PTR_ERR(trans);
5f39d397 4069
2b0143b5 4070 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
12fcfd22 4071
2b0143b5 4072 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
e02119d5 4073 dentry->d_name.name, dentry->d_name.len);
b532402e
TI		 4074 if (ret)
4075 goto out;
7b128766 4076
a22285a6 4077 if (inode->i_nlink == 0) {
7b128766 4078 ret = btrfs_orphan_add(trans, inode);
b532402e
TI		 4079 if (ret)
4080 goto out;
a22285a6 4081 }
7b128766 4082
b532402e 4083out:
d52be818 4084 btrfs_end_transaction(trans, root);
b53d3f5d 4085 btrfs_btree_balance_dirty(root);
39279cc3
CM		 4086 return ret;
4087}
4088
4df27c4d
YZ		 4089int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4090 struct btrfs_root *root,
4091 struct inode *dir, u64 objectid,
4092 const char *name, int name_len)
4093{
4094 struct btrfs_path *path;
4095 struct extent_buffer *leaf;
4096 struct btrfs_dir_item *di;
4097 struct btrfs_key key;
4098 u64 index;
4099 int ret;
33345d01 4100 u64 dir_ino = btrfs_ino(dir);
4df27c4d
YZ		 4101
4102 path = btrfs_alloc_path();
4103 if (!path)
4104 return -ENOMEM;
4105
33345d01 4106 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4df27c4d 4107 name, name_len, -1);
79787eaa
JM		 4108 if (IS_ERR_OR_NULL(di)) {
4109 if (!di)
4110 ret = -ENOENT;
4111 else
4112 ret = PTR_ERR(di);
4113 goto out;
4114 }
4df27c4d
YZ		 4115
4116 leaf = path->nodes[0];
4117 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4118 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4119 ret = btrfs_delete_one_dir_name(trans, root, path, di);
79787eaa
JM		 4120 if (ret) {
4121 btrfs_abort_transaction(trans, root, ret);
4122 goto out;
4123 }
b3b4aa74 4124 btrfs_release_path(path);
4df27c4d
YZ		 4125
4126 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4127 objectid, root->root_key.objectid,
33345d01 4128 dir_ino, &index, name, name_len);
4df27c4d 4129 if (ret < 0) {
79787eaa
JM		 4130 if (ret != -ENOENT) {
4131 btrfs_abort_transaction(trans, root, ret);
4132 goto out;
4133 }
33345d01 4134 di = btrfs_search_dir_index_item(root, path, dir_ino,
4df27c4d 4135 name, name_len);
79787eaa
JM		 4136 if (IS_ERR_OR_NULL(di)) {
4137 if (!di)
4138 ret = -ENOENT;
4139 else
4140 ret = PTR_ERR(di);
4141 btrfs_abort_transaction(trans, root, ret);
4142 goto out;
4143 }
4df27c4d
YZ		 4144
4145 leaf = path->nodes[0];
4146 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
b3b4aa74 4147 btrfs_release_path(path);
4df27c4d
YZ		 4148 index = key.offset;
4149 }
945d8962 4150 btrfs_release_path(path);
4df27c4d 4151
16cdcec7 4152 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 4153 if (ret) {
4154 btrfs_abort_transaction(trans, root, ret);
4155 goto out;
4156 }
4df27c4d
YZ		 4157
4158 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95 4159 inode_inc_iversion(dir);
04b285f3 4160 dir->i_mtime = dir->i_ctime = current_fs_time(dir->i_sb);
5a24e84c 4161 ret = btrfs_update_inode_fallback(trans, root, dir);
79787eaa
JM		 4162 if (ret)
4163 btrfs_abort_transaction(trans, root, ret);
4164out:
71d7aed0 4165 btrfs_free_path(path);
79787eaa 4166 return ret;
4df27c4d
YZ		 4167}
4168
39279cc3
CM		 4169static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4170{
2b0143b5 4171 struct inode *inode = d_inode(dentry);
1832a6d5 4172 int err = 0;
39279cc3 4173 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 4174 struct btrfs_trans_handle *trans;
39279cc3 4175
b3ae244e 4176 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
134d4512 4177 return -ENOTEMPTY;
b3ae244e
DS		 4178 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4179 return -EPERM;
134d4512 4180
d52be818 4181 trans = __unlink_start_trans(dir);
a22285a6 4182 if (IS_ERR(trans))
5df6a9f6 4183 return PTR_ERR(trans);
5df6a9f6 4184
33345d01 4185 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4df27c4d
YZ		 4186 err = btrfs_unlink_subvol(trans, root, dir,
4187 BTRFS_I(inode)->location.objectid,
4188 dentry->d_name.name,
4189 dentry->d_name.len);
4190 goto out;
4191 }
4192
7b128766
JB		 4193 err = btrfs_orphan_add(trans, inode);
4194 if (err)
4df27c4d 4195 goto out;
7b128766 4196
39279cc3 4197 /* now the directory is empty */
2b0143b5 4198 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
e02119d5 4199 dentry->d_name.name, dentry->d_name.len);
d397712b 4200 if (!err)
dbe674a9 4201 btrfs_i_size_write(inode, 0);
4df27c4d 4202out:
d52be818 4203 btrfs_end_transaction(trans, root);
b53d3f5d 4204 btrfs_btree_balance_dirty(root);
3954401f 4205
39279cc3
CM		 4206 return err;
4207}
4208
28f75a0e
CM		 4209static int truncate_space_check(struct btrfs_trans_handle *trans,
4210 struct btrfs_root *root,
4211 u64 bytes_deleted)
4212{
4213 int ret;
4214
4215 bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4216 ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4217 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4218 if (!ret)
4219 trans->bytes_reserved += bytes_deleted;
4220 return ret;
4221
4222}
4223
0305cd5f
FM		 4224static int truncate_inline_extent(struct inode *inode,
4225 struct btrfs_path *path,
4226 struct btrfs_key *found_key,
4227 const u64 item_end,
4228 const u64 new_size)
4229{
4230 struct extent_buffer *leaf = path->nodes[0];
4231 int slot = path->slots[0];
4232 struct btrfs_file_extent_item *fi;
4233 u32 size = (u32)(new_size - found_key->offset);
4234 struct btrfs_root *root = BTRFS_I(inode)->root;
4235
4236 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4237
4238 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4239 loff_t offset = new_size;
4240 loff_t page_end = ALIGN(offset, PAGE_CACHE_SIZE);
4241
4242 /*
4243 * Zero out the remaining of the last page of our inline extent,
4244 * instead of directly truncating our inline extent here - that
4245 * would be much more complex (decompressing all the data, then
4246 * compressing the truncated data, which might be bigger than
4247 * the size of the inline extent, resize the extent, etc).
4248 * We release the path because to get the page we might need to
4249 * read the extent item from disk (data not in the page cache).
4250 */
4251 btrfs_release_path(path);
4252 return btrfs_truncate_page(inode, offset, page_end - offset, 0);
4253 }
4254
4255 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4256 size = btrfs_file_extent_calc_inline_size(size);
4257 btrfs_truncate_item(root, path, size, 1);
4258
4259 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4260 inode_sub_bytes(inode, item_end + 1 - new_size);
4261
4262 return 0;
4263}
4264
39279cc3
CM		 4265/*
4266 * this can truncate away extent items, csum items and directory items.
4267 * It starts at a high offset and removes keys until it can't find
d352ac68 4268 * any higher than new_size
39279cc3
CM		 4269 *
4270 * csum items that cross the new i_size are truncated to the new size
4271 * as well.
7b128766
JB		 4272 *
4273 * min_type is the minimum key type to truncate down to. If set to 0, this
4274 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 4275 */
8082510e
YZ		 4276int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4277 struct btrfs_root *root,
4278 struct inode *inode,
4279 u64 new_size, u32 min_type)
39279cc3 4280{
39279cc3 4281 struct btrfs_path *path;
5f39d397 4282 struct extent_buffer *leaf;
39279cc3 4283 struct btrfs_file_extent_item *fi;
8082510e
YZ		 4284 struct btrfs_key key;
4285 struct btrfs_key found_key;
39279cc3 4286 u64 extent_start = 0;
db94535d 4287 u64 extent_num_bytes = 0;
5d4f98a2 4288 u64 extent_offset = 0;
39279cc3 4289 u64 item_end = 0;
c1aa4575 4290 u64 last_size = new_size;
8082510e 4291 u32 found_type = (u8)-1;
39279cc3
CM		 4292 int found_extent;
4293 int del_item;
85e21bac
CM		 4294 int pending_del_nr = 0;
4295 int pending_del_slot = 0;
179e29e4 4296 int extent_type = -1;
8082510e
YZ		 4297 int ret;
4298 int err = 0;
33345d01 4299 u64 ino = btrfs_ino(inode);
28ed1345 4300 u64 bytes_deleted = 0;
1262133b
JB		 4301 bool be_nice = 0;
4302 bool should_throttle = 0;
28f75a0e 4303 bool should_end = 0;
8082510e
YZ		 4304
4305 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
39279cc3 4306
28ed1345
CM		 4307 /*
4308 * for non-free space inodes and ref cows, we want to back off from
4309 * time to time
4310 */
4311 if (!btrfs_is_free_space_inode(inode) &&
4312 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4313 be_nice = 1;
4314
0eb0e19c
MF		 4315 path = btrfs_alloc_path();
4316 if (!path)
4317 return -ENOMEM;
e4058b54 4318 path->reada = READA_BACK;
0eb0e19c 4319
5dc562c5
JB		 4320 /*
4321 * We want to drop from the next block forward in case this new size is
4322 * not block aligned since we will be keeping the last block of the
4323 * extent just the way it is.
4324 */
27cdeb70
MX		 4325 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4326 root == root->fs_info->tree_root)
fda2832f
QW		 4327 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4328 root->sectorsize), (u64)-1, 0);
8082510e 4329
16cdcec7
MX		 4330 /*
4331 * This function is also used to drop the items in the log tree before
4332 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4333 * it is used to drop the loged items. So we shouldn't kill the delayed
4334 * items.
4335 */
4336 if (min_type == 0 && root == BTRFS_I(inode)->root)
4337 btrfs_kill_delayed_inode_items(inode);
4338
33345d01 4339 key.objectid = ino;
39279cc3 4340 key.offset = (u64)-1;
5f39d397
CM		 4341 key.type = (u8)-1;
4342
85e21bac 4343search_again:
28ed1345
CM		 4344 /*
4345 * with a 16K leaf size and 128MB extents, you can actually queue
4346 * up a huge file in a single leaf. Most of the time that
4347 * bytes_deleted is > 0, it will be huge by the time we get here
4348 */
ee22184b 4349 if (be_nice && bytes_deleted > SZ_32M) {
28ed1345
CM		 4350 if (btrfs_should_end_transaction(trans, root)) {
4351 err = -EAGAIN;
4352 goto error;
4353 }
4354 }
4355
4356
b9473439 4357 path->leave_spinning = 1;
85e21bac 4358 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8082510e
YZ		 4359 if (ret < 0) {
4360 err = ret;
4361 goto out;
4362 }
d397712b 4363
85e21bac 4364 if (ret > 0) {
e02119d5
CM		 4365 /* there are no items in the tree for us to truncate, we're
4366 * done
4367 */
8082510e
YZ		 4368 if (path->slots[0] == 0)
4369 goto out;
85e21bac
CM		 4370 path->slots[0]--;
4371 }
4372
d397712b 4373 while (1) {
39279cc3 4374 fi = NULL;
5f39d397
CM		 4375 leaf = path->nodes[0];
4376 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
962a298f 4377 found_type = found_key.type;
39279cc3 4378
33345d01 4379 if (found_key.objectid != ino)
39279cc3 4380 break;
5f39d397 4381
85e21bac 4382 if (found_type < min_type)
39279cc3
CM		 4383 break;
4384
5f39d397 4385 item_end = found_key.offset;
39279cc3 4386 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 4387 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 4388 struct btrfs_file_extent_item);
179e29e4
CM		 4389 extent_type = btrfs_file_extent_type(leaf, fi);
4390 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 4391 item_end +=
db94535d 4392 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 4393 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 4394 item_end += btrfs_file_extent_inline_len(leaf,
514ac8ad 4395 path->slots[0], fi);
39279cc3 4396 }
008630c1 4397 item_end--;
39279cc3 4398 }
8082510e
YZ		 4399 if (found_type > min_type) {
4400 del_item = 1;
4401 } else {
4402 if (item_end < new_size)
b888db2b 4403 break;
8082510e
YZ		 4404 if (found_key.offset >= new_size)
4405 del_item = 1;
4406 else
4407 del_item = 0;
39279cc3 4408 }
39279cc3 4409 found_extent = 0;
39279cc3 4410 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 4411 if (found_type != BTRFS_EXTENT_DATA_KEY)
4412 goto delete;
4413
7f4f6e0a
JB		 4414 if (del_item)
4415 last_size = found_key.offset;
4416 else
4417 last_size = new_size;
4418
179e29e4 4419 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 4420 u64 num_dec;
db94535d 4421 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
f70a9a6b 4422 if (!del_item) {
db94535d
CM		 4423 u64 orig_num_bytes =
4424 btrfs_file_extent_num_bytes(leaf, fi);
fda2832f
QW		 4425 extent_num_bytes = ALIGN(new_size -
4426 found_key.offset,
4427 root->sectorsize);
db94535d
CM		 4428 btrfs_set_file_extent_num_bytes(leaf, fi,
4429 extent_num_bytes);
4430 num_dec = (orig_num_bytes -
9069218d 4431 extent_num_bytes);
27cdeb70
MX		 4432 if (test_bit(BTRFS_ROOT_REF_COWS,
4433 &root->state) &&
4434 extent_start != 0)
a76a3cd4 4435 inode_sub_bytes(inode, num_dec);
5f39d397 4436 btrfs_mark_buffer_dirty(leaf);
39279cc3 4437 } else {
db94535d
CM		 4438 extent_num_bytes =
4439 btrfs_file_extent_disk_num_bytes(leaf,
4440 fi);
5d4f98a2
YZ		 4441 extent_offset = found_key.offset -
4442 btrfs_file_extent_offset(leaf, fi);
4443
39279cc3 4444 /* FIXME blocksize != 4096 */
9069218d 4445 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 4446 if (extent_start != 0) {
4447 found_extent = 1;
27cdeb70
MX		 4448 if (test_bit(BTRFS_ROOT_REF_COWS,
4449 &root->state))
a76a3cd4 4450 inode_sub_bytes(inode, num_dec);
e02119d5 4451 }
39279cc3 4452 }
9069218d 4453 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 4454 /*
4455 * we can't truncate inline items that have had
4456 * special encodings
4457 */
4458 if (!del_item &&
c8b97818
CM		 4459 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4460 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
514ac8ad
CM		 4461
4462 /*
0305cd5f
FM		 4463 * Need to release path in order to truncate a
4464 * compressed extent. So delete any accumulated
4465 * extent items so far.
514ac8ad 4466 */
0305cd5f
FM		 4467 if (btrfs_file_extent_compression(leaf, fi) !=
4468 BTRFS_COMPRESS_NONE && pending_del_nr) {
4469 err = btrfs_del_items(trans, root, path,
4470 pending_del_slot,
4471 pending_del_nr);
4472 if (err) {
4473 btrfs_abort_transaction(trans,
4474 root,
4475 err);
4476 goto error;
4477 }
4478 pending_del_nr = 0;
4479 }
4480
4481 err = truncate_inline_extent(inode, path,
4482 &found_key,
4483 item_end,
4484 new_size);
4485 if (err) {
4486 btrfs_abort_transaction(trans,
4487 root, err);
4488 goto error;
4489 }
27cdeb70
MX		 4490 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4491 &root->state)) {
0305cd5f 4492 inode_sub_bytes(inode, item_end + 1 - new_size);
9069218d 4493 }
39279cc3 4494 }
179e29e4 4495delete:
39279cc3 4496 if (del_item) {
85e21bac
CM		 4497 if (!pending_del_nr) {
4498 /* no pending yet, add ourselves */
4499 pending_del_slot = path->slots[0];
4500 pending_del_nr = 1;
4501 } else if (pending_del_nr &&
4502 path->slots[0] + 1 == pending_del_slot) {
4503 /* hop on the pending chunk */
4504 pending_del_nr++;
4505 pending_del_slot = path->slots[0];
4506 } else {
d397712b 4507 BUG();
85e21bac 4508 }
39279cc3
CM		 4509 } else {
4510 break;
4511 }
28f75a0e
CM		 4512 should_throttle = 0;
4513
27cdeb70
MX		 4514 if (found_extent &&
4515 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4516 root == root->fs_info->tree_root)) {
b9473439 4517 btrfs_set_path_blocking(path);
28ed1345 4518 bytes_deleted += extent_num_bytes;
39279cc3 4519 ret = btrfs_free_extent(trans, root, extent_start,
5d4f98a2
YZ		 4520 extent_num_bytes, 0,
4521 btrfs_header_owner(leaf),
b06c4bf5 4522 ino, extent_offset);
39279cc3 4523 BUG_ON(ret);
1262133b 4524 if (btrfs_should_throttle_delayed_refs(trans, root))
28ed1345
CM		 4525 btrfs_async_run_delayed_refs(root,
4526 trans->delayed_ref_updates * 2, 0);
28f75a0e
CM		 4527 if (be_nice) {
4528 if (truncate_space_check(trans, root,
4529 extent_num_bytes)) {
4530 should_end = 1;
4531 }
4532 if (btrfs_should_throttle_delayed_refs(trans,
4533 root)) {
4534 should_throttle = 1;
4535 }
4536 }
39279cc3 4537 }
85e21bac 4538
8082510e
YZ		 4539 if (found_type == BTRFS_INODE_ITEM_KEY)
4540 break;
4541
4542 if (path->slots[0] == 0 ||
1262133b 4543 path->slots[0] != pending_del_slot ||
28f75a0e 4544 should_throttle || should_end) {
8082510e
YZ		 4545 if (pending_del_nr) {
4546 ret = btrfs_del_items(trans, root, path,
4547 pending_del_slot,
4548 pending_del_nr);
79787eaa
JM		 4549 if (ret) {
4550 btrfs_abort_transaction(trans,
4551 root, ret);
4552 goto error;
4553 }
8082510e
YZ		 4554 pending_del_nr = 0;
4555 }
b3b4aa74 4556 btrfs_release_path(path);
28f75a0e 4557 if (should_throttle) {
1262133b
JB		 4558 unsigned long updates = trans->delayed_ref_updates;
4559 if (updates) {
4560 trans->delayed_ref_updates = 0;
4561 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4562 if (ret && !err)
4563 err = ret;
4564 }
4565 }
28f75a0e
CM		 4566 /*
4567 * if we failed to refill our space rsv, bail out
4568 * and let the transaction restart
4569 */
4570 if (should_end) {
4571 err = -EAGAIN;
4572 goto error;
4573 }
85e21bac 4574 goto search_again;
8082510e
YZ		 4575 } else {
4576 path->slots[0]--;
85e21bac 4577 }
39279cc3 4578 }
8082510e 4579out:
85e21bac
CM		 4580 if (pending_del_nr) {
4581 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4582 pending_del_nr);
79787eaa
JM		 4583 if (ret)
4584 btrfs_abort_transaction(trans, root, ret);
85e21bac 4585 }
79787eaa 4586error:
c1aa4575 4587 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
7f4f6e0a 4588 btrfs_ordered_update_i_size(inode, last_size, NULL);
28ed1345 4589
39279cc3 4590 btrfs_free_path(path);
28ed1345 4591
ee22184b 4592 if (be_nice && bytes_deleted > SZ_32M) {
28ed1345
CM		 4593 unsigned long updates = trans->delayed_ref_updates;
4594 if (updates) {
4595 trans->delayed_ref_updates = 0;
4596 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4597 if (ret && !err)
4598 err = ret;
4599 }
4600 }
8082510e 4601 return err;
39279cc3
CM		 4602}
4603
4604/*
2aaa6655
JB		 4605 * btrfs_truncate_page - read, zero a chunk and write a page
4606 * @inode - inode that we're zeroing
4607 * @from - the offset to start zeroing
4608 * @len - the length to zero, 0 to zero the entire range respective to the
4609 * offset
4610 * @front - zero up to the offset instead of from the offset on
4611 *
4612 * This will find the page for the "from" offset and cow the page and zero the
4613 * part we want to zero. This is used with truncate and hole punching.
39279cc3 4614 */
2aaa6655
JB		 4615int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4616 int front)
39279cc3 4617{
2aaa6655 4618 struct address_space *mapping = inode->i_mapping;
db94535d 4619 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 4620 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4621 struct btrfs_ordered_extent *ordered;
2ac55d41 4622 struct extent_state *cached_state = NULL;
e6dcd2dc 4623 char *kaddr;
db94535d 4624 u32 blocksize = root->sectorsize;
39279cc3
CM		 4625 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4626 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4627 struct page *page;
3b16a4e3 4628 gfp_t mask = btrfs_alloc_write_mask(mapping);
39279cc3 4629 int ret = 0;
a52d9a80 4630 u64 page_start;
e6dcd2dc 4631 u64 page_end;
39279cc3 4632
2aaa6655
JB		 4633 if ((offset & (blocksize - 1)) == 0 &&
4634 (!len || ((len & (blocksize - 1)) == 0)))
39279cc3 4635 goto out;
7cf5b976 4636 ret = btrfs_delalloc_reserve_space(inode,
df480633 4637 round_down(from, PAGE_CACHE_SIZE), PAGE_CACHE_SIZE);
5d5e103a
JB		 4638 if (ret)
4639 goto out;
39279cc3 4640
211c17f5 4641again:
3b16a4e3 4642 page = find_or_create_page(mapping, index, mask);
5d5e103a 4643 if (!page) {
7cf5b976 4644 btrfs_delalloc_release_space(inode,
df480633
QW		 4645 round_down(from, PAGE_CACHE_SIZE),
4646 PAGE_CACHE_SIZE);
ac6a2b36 4647 ret = -ENOMEM;
39279cc3 4648 goto out;
5d5e103a 4649 }
e6dcd2dc
CM		 4650
4651 page_start = page_offset(page);
4652 page_end = page_start + PAGE_CACHE_SIZE - 1;
4653
39279cc3 4654 if (!PageUptodate(page)) {
9ebefb18 4655 ret = btrfs_readpage(NULL, page);
39279cc3 4656 lock_page(page);
211c17f5
CM		 4657 if (page->mapping != mapping) {
4658 unlock_page(page);
4659 page_cache_release(page);
4660 goto again;
4661 }
39279cc3
CM		 4662 if (!PageUptodate(page)) {
4663 ret = -EIO;
89642229 4664 goto out_unlock;
39279cc3
CM		 4665 }
4666 }
211c17f5 4667 wait_on_page_writeback(page);
e6dcd2dc 4668
ff13db41 4669 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
e6dcd2dc
CM		 4670 set_page_extent_mapped(page);
4671
4672 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4673 if (ordered) {
2ac55d41
JB		 4674 unlock_extent_cached(io_tree, page_start, page_end,
4675 &cached_state, GFP_NOFS);
e6dcd2dc
CM		 4676 unlock_page(page);
4677 page_cache_release(page);
eb84ae03 4678 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 4679 btrfs_put_ordered_extent(ordered);
4680 goto again;
4681 }
4682
2ac55d41 4683 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 4684 EXTENT_DIRTY | EXTENT_DELALLOC |
4685 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 4686 0, 0, &cached_state, GFP_NOFS);
5d5e103a 4687
2ac55d41
JB		 4688 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4689 &cached_state);
9ed74f2d 4690 if (ret) {
2ac55d41
JB		 4691 unlock_extent_cached(io_tree, page_start, page_end,
4692 &cached_state, GFP_NOFS);
9ed74f2d
JB		 4693 goto out_unlock;
4694 }
4695
e6dcd2dc 4696 if (offset != PAGE_CACHE_SIZE) {
2aaa6655
JB		 4697 if (!len)
4698 len = PAGE_CACHE_SIZE - offset;
e6dcd2dc 4699 kaddr = kmap(page);
2aaa6655
JB		 4700 if (front)
4701 memset(kaddr, 0, offset);
4702 else
4703 memset(kaddr + offset, 0, len);
e6dcd2dc
CM		 4704 flush_dcache_page(page);
4705 kunmap(page);
4706 }
247e743c 4707 ClearPageChecked(page);
e6dcd2dc 4708 set_page_dirty(page);
2ac55d41
JB		 4709 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4710 GFP_NOFS);
39279cc3 4711
89642229 4712out_unlock:
5d5e103a 4713 if (ret)
7cf5b976
QW		 4714 btrfs_delalloc_release_space(inode, page_start,
4715 PAGE_CACHE_SIZE);
39279cc3
CM		 4716 unlock_page(page);
4717 page_cache_release(page);
4718out:
4719 return ret;
4720}
4721
16e7549f
JB		 4722static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4723 u64 offset, u64 len)
4724{
4725 struct btrfs_trans_handle *trans;
4726 int ret;
4727
4728 /*
4729 * Still need to make sure the inode looks like it's been updated so
4730 * that any holes get logged if we fsync.
4731 */
4732 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4733 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4734 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4735 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4736 return 0;
4737 }
4738
4739 /*
4740 * 1 - for the one we're dropping
4741 * 1 - for the one we're adding
4742 * 1 - for updating the inode.
4743 */
4744 trans = btrfs_start_transaction(root, 3);
4745 if (IS_ERR(trans))
4746 return PTR_ERR(trans);
4747
4748 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4749 if (ret) {
4750 btrfs_abort_transaction(trans, root, ret);
4751 btrfs_end_transaction(trans, root);
4752 return ret;
4753 }
4754
4755 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4756 0, 0, len, 0, len, 0, 0, 0);
4757 if (ret)
4758 btrfs_abort_transaction(trans, root, ret);
4759 else
4760 btrfs_update_inode(trans, root, inode);
4761 btrfs_end_transaction(trans, root);
4762 return ret;
4763}
4764
695a0d0d
JB		 4765/*
4766 * This function puts in dummy file extents for the area we're creating a hole
4767 * for. So if we are truncating this file to a larger size we need to insert
4768 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4769 * the range between oldsize and size
4770 */
a41ad394 4771int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
39279cc3 4772{
9036c102
YZ		 4773 struct btrfs_root *root = BTRFS_I(inode)->root;
4774 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a22285a6 4775 struct extent_map *em = NULL;
2ac55d41 4776 struct extent_state *cached_state = NULL;
5dc562c5 4777 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
fda2832f
QW		 4778 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4779 u64 block_end = ALIGN(size, root->sectorsize);
9036c102
YZ		 4780 u64 last_byte;
4781 u64 cur_offset;
4782 u64 hole_size;
9ed74f2d 4783 int err = 0;
39279cc3 4784
a71754fc
JB		 4785 /*
4786 * If our size started in the middle of a page we need to zero out the
4787 * rest of the page before we expand the i_size, otherwise we could
4788 * expose stale data.
4789 */
4790 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4791 if (err)
4792 return err;
4793
9036c102
YZ		 4794 if (size <= hole_start)
4795 return 0;
4796
9036c102
YZ		 4797 while (1) {
4798 struct btrfs_ordered_extent *ordered;
fa7c1494 4799
ff13db41 4800 lock_extent_bits(io_tree, hole_start, block_end - 1,
d0082371 4801 &cached_state);
fa7c1494
MX		 4802 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4803 block_end - hole_start);
9036c102
YZ		 4804 if (!ordered)
4805 break;
2ac55d41
JB		 4806 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4807 &cached_state, GFP_NOFS);
fa7c1494 4808 btrfs_start_ordered_extent(inode, ordered, 1);
9036c102
YZ		 4809 btrfs_put_ordered_extent(ordered);
4810 }
39279cc3 4811
9036c102
YZ		 4812 cur_offset = hole_start;
4813 while (1) {
4814 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4815 block_end - cur_offset, 0);
79787eaa
JM		 4816 if (IS_ERR(em)) {
4817 err = PTR_ERR(em);
f2767956 4818 em = NULL;
79787eaa
JM		 4819 break;
4820 }
9036c102 4821 last_byte = min(extent_map_end(em), block_end);
fda2832f 4822 last_byte = ALIGN(last_byte , root->sectorsize);
8082510e 4823 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5dc562c5 4824 struct extent_map *hole_em;
9036c102 4825 hole_size = last_byte - cur_offset;
9ed74f2d 4826
16e7549f
JB		 4827 err = maybe_insert_hole(root, inode, cur_offset,
4828 hole_size);
4829 if (err)
3893e33b 4830 break;
5dc562c5
JB		 4831 btrfs_drop_extent_cache(inode, cur_offset,
4832 cur_offset + hole_size - 1, 0);
4833 hole_em = alloc_extent_map();
4834 if (!hole_em) {
4835 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4836 &BTRFS_I(inode)->runtime_flags);
4837 goto next;
4838 }
4839 hole_em->start = cur_offset;
4840 hole_em->len = hole_size;
4841 hole_em->orig_start = cur_offset;
8082510e 4842
5dc562c5
JB		 4843 hole_em->block_start = EXTENT_MAP_HOLE;
4844 hole_em->block_len = 0;
b4939680 4845 hole_em->orig_block_len = 0;
cc95bef6 4846 hole_em->ram_bytes = hole_size;
5dc562c5
JB		 4847 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4848 hole_em->compress_type = BTRFS_COMPRESS_NONE;
16e7549f 4849 hole_em->generation = root->fs_info->generation;
8082510e 4850
5dc562c5
JB		 4851 while (1) {
4852 write_lock(&em_tree->lock);
09a2a8f9 4853 err = add_extent_mapping(em_tree, hole_em, 1);
5dc562c5
JB		 4854 write_unlock(&em_tree->lock);
4855 if (err != -EEXIST)
4856 break;
4857 btrfs_drop_extent_cache(inode, cur_offset,
4858 cur_offset +
4859 hole_size - 1, 0);
4860 }
4861 free_extent_map(hole_em);
9036c102 4862 }
16e7549f 4863next:
9036c102 4864 free_extent_map(em);
a22285a6 4865 em = NULL;
9036c102 4866 cur_offset = last_byte;
8082510e 4867 if (cur_offset >= block_end)
9036c102
YZ		 4868 break;
4869 }
a22285a6 4870 free_extent_map(em);
2ac55d41
JB		 4871 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4872 GFP_NOFS);
9036c102
YZ		 4873 return err;
4874}
39279cc3 4875
3972f260 4876static int btrfs_setsize(struct inode *inode, struct iattr *attr)
8082510e 4877{
f4a2f4c5
MX		 4878 struct btrfs_root *root = BTRFS_I(inode)->root;
4879 struct btrfs_trans_handle *trans;
a41ad394 4880 loff_t oldsize = i_size_read(inode);
3972f260
ES		 4881 loff_t newsize = attr->ia_size;
4882 int mask = attr->ia_valid;
8082510e
YZ		 4883 int ret;
4884
3972f260
ES		 4885 /*
4886 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4887 * special case where we need to update the times despite not having
4888 * these flags set. For all other operations the VFS set these flags
4889 * explicitly if it wants a timestamp update.
4890 */
dff6efc3
CH		 4891 if (newsize != oldsize) {
4892 inode_inc_iversion(inode);
4893 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4894 inode->i_ctime = inode->i_mtime =
4895 current_fs_time(inode->i_sb);
4896 }
3972f260 4897
a41ad394 4898 if (newsize > oldsize) {
7caef267 4899 truncate_pagecache(inode, newsize);
9ea24bbe
FM		 4900 /*
4901 * Don't do an expanding truncate while snapshoting is ongoing.
4902 * This is to ensure the snapshot captures a fully consistent
4903 * state of this file - if the snapshot captures this expanding
4904 * truncation, it must capture all writes that happened before
4905 * this truncation.
4906 */
0bc19f90 4907 btrfs_wait_for_snapshot_creation(root);
a41ad394 4908 ret = btrfs_cont_expand(inode, oldsize, newsize);
9ea24bbe
FM		 4909 if (ret) {
4910 btrfs_end_write_no_snapshoting(root);
8082510e 4911 return ret;
9ea24bbe 4912 }
8082510e 4913
f4a2f4c5 4914 trans = btrfs_start_transaction(root, 1);
9ea24bbe
FM		 4915 if (IS_ERR(trans)) {
4916 btrfs_end_write_no_snapshoting(root);
f4a2f4c5 4917 return PTR_ERR(trans);
9ea24bbe 4918 }
f4a2f4c5
MX		 4919
4920 i_size_write(inode, newsize);
4921 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4922 ret = btrfs_update_inode(trans, root, inode);
9ea24bbe 4923 btrfs_end_write_no_snapshoting(root);
7ad85bb7 4924 btrfs_end_transaction(trans, root);
a41ad394 4925 } else {
8082510e 4926
a41ad394
JB		 4927 /*
4928 * We're truncating a file that used to have good data down to
4929 * zero. Make sure it gets into the ordered flush list so that
4930 * any new writes get down to disk quickly.
4931 */
4932 if (newsize == 0)
72ac3c0d
JB		 4933 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4934 &BTRFS_I(inode)->runtime_flags);
8082510e 4935
f3fe820c
JB		 4936 /*
4937 * 1 for the orphan item we're going to add
4938 * 1 for the orphan item deletion.
4939 */
4940 trans = btrfs_start_transaction(root, 2);
4941 if (IS_ERR(trans))
4942 return PTR_ERR(trans);
4943
4944 /*
4945 * We need to do this in case we fail at _any_ point during the
4946 * actual truncate. Once we do the truncate_setsize we could
4947 * invalidate pages which forces any outstanding ordered io to
4948 * be instantly completed which will give us extents that need
4949 * to be truncated. If we fail to get an orphan inode down we
4950 * could have left over extents that were never meant to live,
4951 * so we need to garuntee from this point on that everything
4952 * will be consistent.
4953 */
4954 ret = btrfs_orphan_add(trans, inode);
4955 btrfs_end_transaction(trans, root);
4956 if (ret)
4957 return ret;
4958
a41ad394
JB		 4959 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4960 truncate_setsize(inode, newsize);
2e60a51e
MX		 4961
4962 /* Disable nonlocked read DIO to avoid the end less truncate */
4963 btrfs_inode_block_unlocked_dio(inode);
4964 inode_dio_wait(inode);
4965 btrfs_inode_resume_unlocked_dio(inode);
4966
a41ad394 4967 ret = btrfs_truncate(inode);
7f4f6e0a
JB		 4968 if (ret && inode->i_nlink) {
4969 int err;
4970
4971 /*
4972 * failed to truncate, disk_i_size is only adjusted down
4973 * as we remove extents, so it should represent the true
4974 * size of the inode, so reset the in memory size and
4975 * delete our orphan entry.
4976 */
4977 trans = btrfs_join_transaction(root);
4978 if (IS_ERR(trans)) {
4979 btrfs_orphan_del(NULL, inode);
4980 return ret;
4981 }
4982 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4983 err = btrfs_orphan_del(trans, inode);
4984 if (err)
4985 btrfs_abort_transaction(trans, root, err);
4986 btrfs_end_transaction(trans, root);
4987 }
8082510e
YZ		 4988 }
4989
a41ad394 4990 return ret;
8082510e
YZ		 4991}
4992
9036c102
YZ		 4993static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4994{
2b0143b5 4995 struct inode *inode = d_inode(dentry);
b83cc969 4996 struct btrfs_root *root = BTRFS_I(inode)->root;
9036c102 4997 int err;
39279cc3 4998
b83cc969
LZ		 4999 if (btrfs_root_readonly(root))
5000 return -EROFS;
5001
9036c102
YZ		 5002 err = inode_change_ok(inode, attr);
5003 if (err)
5004 return err;
2bf5a725 5005
5a3f23d5 5006 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3972f260 5007 err = btrfs_setsize(inode, attr);
8082510e
YZ		 5008 if (err)
5009 return err;
39279cc3 5010 }
9036c102 5011
1025774c
CH		 5012 if (attr->ia_valid) {
5013 setattr_copy(inode, attr);
0c4d2d95 5014 inode_inc_iversion(inode);
22c44fe6 5015 err = btrfs_dirty_inode(inode);
1025774c 5016
22c44fe6 5017 if (!err && attr->ia_valid & ATTR_MODE)
996a710d 5018 err = posix_acl_chmod(inode, inode->i_mode);
1025774c 5019 }
33268eaf 5020
39279cc3
CM		 5021 return err;
5022}
61295eb8 5023
131e404a
FDBM		 5024/*
5025 * While truncating the inode pages during eviction, we get the VFS calling
5026 * btrfs_invalidatepage() against each page of the inode. This is slow because
5027 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5028 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5029 * extent_state structures over and over, wasting lots of time.
5030 *
5031 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5032 * those expensive operations on a per page basis and do only the ordered io
5033 * finishing, while we release here the extent_map and extent_state structures,
5034 * without the excessive merging and splitting.
5035 */
5036static void evict_inode_truncate_pages(struct inode *inode)
5037{
5038 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5039 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5040 struct rb_node *node;
5041
5042 ASSERT(inode->i_state & I_FREEING);
91b0abe3 5043 truncate_inode_pages_final(&inode->i_data);
131e404a
FDBM		 5044
5045 write_lock(&map_tree->lock);
5046 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5047 struct extent_map *em;
5048
5049 node = rb_first(&map_tree->map);
5050 em = rb_entry(node, struct extent_map, rb_node);
180589ef
WS		 5051 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5052 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
131e404a
FDBM		 5053 remove_extent_mapping(map_tree, em);
5054 free_extent_map(em);
7064dd5c
FM		 5055 if (need_resched()) {
5056 write_unlock(&map_tree->lock);
5057 cond_resched();
5058 write_lock(&map_tree->lock);
5059 }
131e404a
FDBM		 5060 }
5061 write_unlock(&map_tree->lock);
5062
6ca07097
FM		 5063 /*
5064 * Keep looping until we have no more ranges in the io tree.
5065 * We can have ongoing bios started by readpages (called from readahead)
9c6429d9
FM		 5066 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5067 * still in progress (unlocked the pages in the bio but did not yet
5068 * unlocked the ranges in the io tree). Therefore this means some
6ca07097
FM		 5069 * ranges can still be locked and eviction started because before
5070 * submitting those bios, which are executed by a separate task (work
5071 * queue kthread), inode references (inode->i_count) were not taken
5072 * (which would be dropped in the end io callback of each bio).
5073 * Therefore here we effectively end up waiting for those bios and
5074 * anyone else holding locked ranges without having bumped the inode's
5075 * reference count - if we don't do it, when they access the inode's
5076 * io_tree to unlock a range it may be too late, leading to an
5077 * use-after-free issue.
5078 */
131e404a
FDBM		 5079 spin_lock(&io_tree->lock);
5080 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5081 struct extent_state *state;
5082 struct extent_state *cached_state = NULL;
6ca07097
FM		 5083 u64 start;
5084 u64 end;
131e404a
FDBM		 5085
5086 node = rb_first(&io_tree->state);
5087 state = rb_entry(node, struct extent_state, rb_node);
6ca07097
FM		 5088 start = state->start;
5089 end = state->end;
131e404a
FDBM		 5090 spin_unlock(&io_tree->lock);
5091
ff13db41 5092 lock_extent_bits(io_tree, start, end, &cached_state);
b9d0b389
QW		 5093
5094 /*
5095 * If still has DELALLOC flag, the extent didn't reach disk,
5096 * and its reserved space won't be freed by delayed_ref.
5097 * So we need to free its reserved space here.
5098 * (Refer to comment in btrfs_invalidatepage, case 2)
5099 *
5100 * Note, end is the bytenr of last byte, so we need + 1 here.
5101 */
5102 if (state->state & EXTENT_DELALLOC)
5103 btrfs_qgroup_free_data(inode, start, end - start + 1);
5104
6ca07097 5105 clear_extent_bit(io_tree, start, end,
131e404a
FDBM		 5106 EXTENT_LOCKED | EXTENT_DIRTY |
5107 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5108 EXTENT_DEFRAG, 1, 1,
5109 &cached_state, GFP_NOFS);
131e404a 5110
7064dd5c 5111 cond_resched();
131e404a
FDBM		 5112 spin_lock(&io_tree->lock);
5113 }
5114 spin_unlock(&io_tree->lock);
5115}
5116
bd555975 5117void btrfs_evict_inode(struct inode *inode)
39279cc3
CM		 5118{
5119 struct btrfs_trans_handle *trans;
5120 struct btrfs_root *root = BTRFS_I(inode)->root;
726c35fa 5121 struct btrfs_block_rsv *rsv, *global_rsv;
3bce876f 5122 int steal_from_global = 0;
07127184 5123 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3
CM		 5124 int ret;
5125
1abe9b8a 5126 trace_btrfs_inode_evict(inode);
5127
131e404a
FDBM		 5128 evict_inode_truncate_pages(inode);
5129
69e9c6c6
SB		 5130 if (inode->i_nlink &&
5131 ((btrfs_root_refs(&root->root_item) != 0 &&
5132 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5133 btrfs_is_free_space_inode(inode)))
bd555975
AV		 5134 goto no_delete;
5135
39279cc3 5136 if (is_bad_inode(inode)) {
7b128766 5137 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 5138 goto no_delete;
5139 }
bd555975 5140 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
a30e577c
JM		 5141 if (!special_file(inode->i_mode))
5142 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 5143
f612496b
MX		 5144 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5145
c71bf099 5146 if (root->fs_info->log_root_recovering) {
6bf02314 5147 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8a35d95f 5148 &BTRFS_I(inode)->runtime_flags));
c71bf099
YZ		 5149 goto no_delete;
5150 }
5151
76dda93c 5152 if (inode->i_nlink > 0) {
69e9c6c6
SB		 5153 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5154 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
76dda93c
YZ		 5155 goto no_delete;
5156 }
5157
0e8c36a9
MX		 5158 ret = btrfs_commit_inode_delayed_inode(inode);
5159 if (ret) {
5160 btrfs_orphan_del(NULL, inode);
5161 goto no_delete;
5162 }
5163
66d8f3dd 5164 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4289a667
JB		 5165 if (!rsv) {
5166 btrfs_orphan_del(NULL, inode);
5167 goto no_delete;
5168 }
4a338542 5169 rsv->size = min_size;
ca7e70f5 5170 rsv->failfast = 1;
726c35fa 5171 global_rsv = &root->fs_info->global_block_rsv;
4289a667 5172
dbe674a9 5173 btrfs_i_size_write(inode, 0);
5f39d397 5174
4289a667 5175 /*
8407aa46
MX		 5176 * This is a bit simpler than btrfs_truncate since we've already
5177 * reserved our space for our orphan item in the unlink, so we just
5178 * need to reserve some slack space in case we add bytes and update
5179 * inode item when doing the truncate.
4289a667 5180 */
8082510e 5181 while (1) {
08e007d2
MX		 5182 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5183 BTRFS_RESERVE_FLUSH_LIMIT);
726c35fa
JB		 5184
5185 /*
5186 * Try and steal from the global reserve since we will
5187 * likely not use this space anyway, we want to try as
5188 * hard as possible to get this to work.
5189 */
5190 if (ret)
3bce876f
JB		 5191 steal_from_global++;
5192 else
5193 steal_from_global = 0;
5194 ret = 0;
d68fc57b 5195
3bce876f
JB		 5196 /*
5197 * steal_from_global == 0: we reserved stuff, hooray!
5198 * steal_from_global == 1: we didn't reserve stuff, boo!
5199 * steal_from_global == 2: we've committed, still not a lot of
5200 * room but maybe we'll have room in the global reserve this
5201 * time.
5202 * steal_from_global == 3: abandon all hope!
5203 */
5204 if (steal_from_global > 2) {
c2cf52eb
SK		 5205 btrfs_warn(root->fs_info,
5206 "Could not get space for a delete, will truncate on mount %d",
5207 ret);
4289a667
JB		 5208 btrfs_orphan_del(NULL, inode);
5209 btrfs_free_block_rsv(root, rsv);
5210 goto no_delete;
d68fc57b 5211 }
7b128766 5212
0e8c36a9 5213 trans = btrfs_join_transaction(root);
4289a667
JB		 5214 if (IS_ERR(trans)) {
5215 btrfs_orphan_del(NULL, inode);
5216 btrfs_free_block_rsv(root, rsv);
5217 goto no_delete;
d68fc57b 5218 }
7b128766 5219
3bce876f
JB		 5220 /*
5221 * We can't just steal from the global reserve, we need tomake
5222 * sure there is room to do it, if not we need to commit and try
5223 * again.
5224 */
5225 if (steal_from_global) {
5226 if (!btrfs_check_space_for_delayed_refs(trans, root))
5227 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5228 min_size);
5229 else
5230 ret = -ENOSPC;
5231 }
5232
5233 /*
5234 * Couldn't steal from the global reserve, we have too much
5235 * pending stuff built up, commit the transaction and try it
5236 * again.
5237 */
5238 if (ret) {
5239 ret = btrfs_commit_transaction(trans, root);
5240 if (ret) {
5241 btrfs_orphan_del(NULL, inode);
5242 btrfs_free_block_rsv(root, rsv);
5243 goto no_delete;
5244 }
5245 continue;
5246 } else {
5247 steal_from_global = 0;
5248 }
5249
4289a667
JB		 5250 trans->block_rsv = rsv;
5251
d68fc57b 5252 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
28ed1345 5253 if (ret != -ENOSPC && ret != -EAGAIN)
8082510e 5254 break;
85e21bac 5255
8407aa46 5256 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e
YZ		 5257 btrfs_end_transaction(trans, root);
5258 trans = NULL;
b53d3f5d 5259 btrfs_btree_balance_dirty(root);
8082510e 5260 }
5f39d397 5261
4289a667
JB		 5262 btrfs_free_block_rsv(root, rsv);
5263
4ef31a45
JB		 5264 /*
5265 * Errors here aren't a big deal, it just means we leave orphan items
5266 * in the tree. They will be cleaned up on the next mount.
5267 */
8082510e 5268 if (ret == 0) {
4289a667 5269 trans->block_rsv = root->orphan_block_rsv;
4ef31a45
JB		 5270 btrfs_orphan_del(trans, inode);
5271 } else {
5272 btrfs_orphan_del(NULL, inode);
8082510e 5273 }
54aa1f4d 5274
4289a667 5275 trans->block_rsv = &root->fs_info->trans_block_rsv;
581bb050
LZ		 5276 if (!(root == root->fs_info->tree_root ||
5277 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
33345d01 5278 btrfs_return_ino(root, btrfs_ino(inode));
581bb050 5279
54aa1f4d 5280 btrfs_end_transaction(trans, root);
b53d3f5d 5281 btrfs_btree_balance_dirty(root);
39279cc3 5282no_delete:
89042e5a 5283 btrfs_remove_delayed_node(inode);
dbd5768f 5284 clear_inode(inode);
39279cc3
CM		 5285}
5286
5287/*
5288 * this returns the key found in the dir entry in the location pointer.
5289 * If no dir entries were found, location->objectid is 0.
5290 */
5291static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5292 struct btrfs_key *location)
5293{
5294 const char *name = dentry->d_name.name;
5295 int namelen = dentry->d_name.len;
5296 struct btrfs_dir_item *di;
5297 struct btrfs_path *path;
5298 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 5299 int ret = 0;
39279cc3
CM		 5300
5301 path = btrfs_alloc_path();
d8926bb3
MF		 5302 if (!path)
5303 return -ENOMEM;
3954401f 5304
33345d01 5305 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
39279cc3 5306 namelen, 0);
0d9f7f3e
Y		 5307 if (IS_ERR(di))
5308 ret = PTR_ERR(di);
d397712b 5309
c704005d 5310 if (IS_ERR_OR_NULL(di))
3954401f 5311 goto out_err;
d397712b 5312
5f39d397 5313 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 5314out:
39279cc3
CM		 5315 btrfs_free_path(path);
5316 return ret;
3954401f
CM		 5317out_err:
5318 location->objectid = 0;
5319 goto out;
39279cc3
CM		 5320}
5321
5322/*
5323 * when we hit a tree root in a directory, the btrfs part of the inode
5324 * needs to be changed to reflect the root directory of the tree root. This
5325 * is kind of like crossing a mount point.
5326 */
5327static int fixup_tree_root_location(struct btrfs_root *root,
4df27c4d
YZ		 5328 struct inode *dir,
5329 struct dentry *dentry,
5330 struct btrfs_key *location,
5331 struct btrfs_root **sub_root)
39279cc3 5332{
4df27c4d
YZ		 5333 struct btrfs_path *path;
5334 struct btrfs_root *new_root;
5335 struct btrfs_root_ref *ref;
5336 struct extent_buffer *leaf;
1d4c08e0 5337 struct btrfs_key key;
4df27c4d
YZ		 5338 int ret;
5339 int err = 0;
39279cc3 5340
4df27c4d
YZ		 5341 path = btrfs_alloc_path();
5342 if (!path) {
5343 err = -ENOMEM;
5344 goto out;
5345 }
39279cc3 5346
4df27c4d 5347 err = -ENOENT;
1d4c08e0
DS		 5348 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5349 key.type = BTRFS_ROOT_REF_KEY;
5350 key.offset = location->objectid;
5351
5352 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5353 0, 0);
4df27c4d
YZ		 5354 if (ret) {
5355 if (ret < 0)
5356 err = ret;
5357 goto out;
5358 }
39279cc3 5359
4df27c4d
YZ		 5360 leaf = path->nodes[0];
5361 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
33345d01 5362 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4df27c4d
YZ		 5363 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5364 goto out;
39279cc3 5365
4df27c4d
YZ		 5366 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5367 (unsigned long)(ref + 1),
5368 dentry->d_name.len);
5369 if (ret)
5370 goto out;
5371
b3b4aa74 5372 btrfs_release_path(path);
4df27c4d
YZ		 5373
5374 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5375 if (IS_ERR(new_root)) {
5376 err = PTR_ERR(new_root);
5377 goto out;
5378 }
5379
4df27c4d
YZ		 5380 *sub_root = new_root;
5381 location->objectid = btrfs_root_dirid(&new_root->root_item);
5382 location->type = BTRFS_INODE_ITEM_KEY;
5383 location->offset = 0;
5384 err = 0;
5385out:
5386 btrfs_free_path(path);
5387 return err;
39279cc3
CM		 5388}
5389
5d4f98a2
YZ		 5390static void inode_tree_add(struct inode *inode)
5391{
5392 struct btrfs_root *root = BTRFS_I(inode)->root;
5393 struct btrfs_inode *entry;
03e860bd
FNP		 5394 struct rb_node **p;
5395 struct rb_node *parent;
cef21937 5396 struct rb_node *new = &BTRFS_I(inode)->rb_node;
33345d01 5397 u64 ino = btrfs_ino(inode);
5d4f98a2 5398
1d3382cb 5399 if (inode_unhashed(inode))
76dda93c 5400 return;
e1409cef 5401 parent = NULL;
5d4f98a2 5402 spin_lock(&root->inode_lock);
e1409cef 5403 p = &root->inode_tree.rb_node;
5d4f98a2
YZ		 5404 while (*p) {
5405 parent = *p;
5406 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5407
33345d01 5408 if (ino < btrfs_ino(&entry->vfs_inode))
03e860bd 5409 p = &parent->rb_left;
33345d01 5410 else if (ino > btrfs_ino(&entry->vfs_inode))
03e860bd 5411 p = &parent->rb_right;
5d4f98a2
YZ		 5412 else {
5413 WARN_ON(!(entry->vfs_inode.i_state &
a4ffdde6 5414 (I_WILL_FREE | I_FREEING)));
cef21937 5415 rb_replace_node(parent, new, &root->inode_tree);
03e860bd
FNP		 5416 RB_CLEAR_NODE(parent);
5417 spin_unlock(&root->inode_lock);
cef21937 5418 return;
5d4f98a2
YZ		 5419 }
5420 }
cef21937
FDBM		 5421 rb_link_node(new, parent, p);
5422 rb_insert_color(new, &root->inode_tree);
5d4f98a2
YZ		 5423 spin_unlock(&root->inode_lock);
5424}
5425
5426static void inode_tree_del(struct inode *inode)
5427{
5428 struct btrfs_root *root = BTRFS_I(inode)->root;
76dda93c 5429 int empty = 0;
5d4f98a2 5430
03e860bd 5431 spin_lock(&root->inode_lock);
5d4f98a2 5432 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5d4f98a2 5433 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5d4f98a2 5434 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
76dda93c 5435 empty = RB_EMPTY_ROOT(&root->inode_tree);
5d4f98a2 5436 }
03e860bd 5437 spin_unlock(&root->inode_lock);
76dda93c 5438
69e9c6c6 5439 if (empty && btrfs_root_refs(&root->root_item) == 0) {
76dda93c
YZ		 5440 synchronize_srcu(&root->fs_info->subvol_srcu);
5441 spin_lock(&root->inode_lock);
5442 empty = RB_EMPTY_ROOT(&root->inode_tree);
5443 spin_unlock(&root->inode_lock);
5444 if (empty)
5445 btrfs_add_dead_root(root);
5446 }
5447}
5448
143bede5 5449void btrfs_invalidate_inodes(struct btrfs_root *root)
76dda93c
YZ		 5450{
5451 struct rb_node *node;
5452 struct rb_node *prev;
5453 struct btrfs_inode *entry;
5454 struct inode *inode;
5455 u64 objectid = 0;
5456
7813b3db
LB		 5457 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5458 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
76dda93c
YZ		 5459
5460 spin_lock(&root->inode_lock);
5461again:
5462 node = root->inode_tree.rb_node;
5463 prev = NULL;
5464 while (node) {
5465 prev = node;
5466 entry = rb_entry(node, struct btrfs_inode, rb_node);
5467
33345d01 5468 if (objectid < btrfs_ino(&entry->vfs_inode))
76dda93c 5469 node = node->rb_left;
33345d01 5470 else if (objectid > btrfs_ino(&entry->vfs_inode))
76dda93c
YZ		 5471 node = node->rb_right;
5472 else
5473 break;
5474 }
5475 if (!node) {
5476 while (prev) {
5477 entry = rb_entry(prev, struct btrfs_inode, rb_node);
33345d01 5478 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
76dda93c
YZ		 5479 node = prev;
5480 break;
5481 }
5482 prev = rb_next(prev);
5483 }
5484 }
5485 while (node) {
5486 entry = rb_entry(node, struct btrfs_inode, rb_node);
33345d01 5487 objectid = btrfs_ino(&entry->vfs_inode) + 1;
76dda93c
YZ		 5488 inode = igrab(&entry->vfs_inode);
5489 if (inode) {
5490 spin_unlock(&root->inode_lock);
5491 if (atomic_read(&inode->i_count) > 1)
5492 d_prune_aliases(inode);
5493 /*
45321ac5 5494 * btrfs_drop_inode will have it removed from
76dda93c
YZ		 5495 * the inode cache when its usage count
5496 * hits zero.
5497 */
5498 iput(inode);
5499 cond_resched();
5500 spin_lock(&root->inode_lock);
5501 goto again;
5502 }
5503
5504 if (cond_resched_lock(&root->inode_lock))
5505 goto again;
5506
5507 node = rb_next(node);
5508 }
5509 spin_unlock(&root->inode_lock);
5d4f98a2
YZ		 5510}
5511
e02119d5
CM		 5512static int btrfs_init_locked_inode(struct inode *inode, void *p)
5513{
5514 struct btrfs_iget_args *args = p;
90d3e592
CM		 5515 inode->i_ino = args->location->objectid;
5516 memcpy(&BTRFS_I(inode)->location, args->location,
5517 sizeof(*args->location));
e02119d5 5518 BTRFS_I(inode)->root = args->root;
39279cc3
CM		 5519 return 0;
5520}
5521
5522static int btrfs_find_actor(struct inode *inode, void *opaque)
5523{
5524 struct btrfs_iget_args *args = opaque;
90d3e592 5525 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
d397712b 5526 args->root == BTRFS_I(inode)->root;
39279cc3
CM		 5527}
5528
5d4f98a2 5529static struct inode *btrfs_iget_locked(struct super_block *s,
90d3e592 5530 struct btrfs_key *location,
5d4f98a2 5531 struct btrfs_root *root)
39279cc3
CM		 5532{
5533 struct inode *inode;
5534 struct btrfs_iget_args args;
90d3e592 5535 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
778ba82b 5536
90d3e592 5537 args.location = location;
39279cc3
CM		 5538 args.root = root;
5539
778ba82b 5540 inode = iget5_locked(s, hashval, btrfs_find_actor,
39279cc3
CM		 5541 btrfs_init_locked_inode,
5542 (void *)&args);
5543 return inode;
5544}
5545
1a54ef8c
BR		 5546/* Get an inode object given its location and corresponding root.
5547 * Returns in *is_new if the inode was read from disk
5548 */
5549struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
73f73415 5550 struct btrfs_root *root, int *new)
1a54ef8c
BR		 5551{
5552 struct inode *inode;
5553
90d3e592 5554 inode = btrfs_iget_locked(s, location, root);
1a54ef8c 5555 if (!inode)
5d4f98a2 5556 return ERR_PTR(-ENOMEM);
1a54ef8c
BR		 5557
5558 if (inode->i_state & I_NEW) {
1a54ef8c 5559 btrfs_read_locked_inode(inode);
1748f843
MF		 5560 if (!is_bad_inode(inode)) {
5561 inode_tree_add(inode);
5562 unlock_new_inode(inode);
5563 if (new)
5564 *new = 1;
5565 } else {
e0b6d65b
ST		 5566 unlock_new_inode(inode);
5567 iput(inode);
5568 inode = ERR_PTR(-ESTALE);
1748f843
MF		 5569 }
5570 }
5571
1a54ef8c
BR		 5572 return inode;
5573}
5574
4df27c4d
YZ		 5575static struct inode *new_simple_dir(struct super_block *s,
5576 struct btrfs_key *key,
5577 struct btrfs_root *root)
5578{
5579 struct inode *inode = new_inode(s);
5580
5581 if (!inode)
5582 return ERR_PTR(-ENOMEM);
5583
4df27c4d
YZ		 5584 BTRFS_I(inode)->root = root;
5585 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
72ac3c0d 5586 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4df27c4d
YZ		 5587
5588 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
848cce0d 5589 inode->i_op = &btrfs_dir_ro_inode_operations;
4df27c4d
YZ		 5590 inode->i_fop = &simple_dir_operations;
5591 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
04b285f3 5592 inode->i_mtime = current_fs_time(inode->i_sb);
9cc97d64 5593 inode->i_atime = inode->i_mtime;
5594 inode->i_ctime = inode->i_mtime;
5595 BTRFS_I(inode)->i_otime = inode->i_mtime;
4df27c4d
YZ		 5596
5597 return inode;
5598}
5599
3de4586c 5600struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3 5601{
d397712b 5602 struct inode *inode;
4df27c4d 5603 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3
CM		 5604 struct btrfs_root *sub_root = root;
5605 struct btrfs_key location;
76dda93c 5606 int index;
b4aff1f8 5607 int ret = 0;
39279cc3
CM		 5608
5609 if (dentry->d_name.len > BTRFS_NAME_LEN)
5610 return ERR_PTR(-ENAMETOOLONG);
5f39d397 5611
39e3c955 5612 ret = btrfs_inode_by_name(dir, dentry, &location);
39279cc3
CM		 5613 if (ret < 0)
5614 return ERR_PTR(ret);
5f39d397 5615
4df27c4d 5616 if (location.objectid == 0)
5662344b 5617 return ERR_PTR(-ENOENT);
4df27c4d
YZ		 5618
5619 if (location.type == BTRFS_INODE_ITEM_KEY) {
73f73415 5620 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4df27c4d
YZ		 5621 return inode;
5622 }
5623
5624 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5625
76dda93c 5626 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4df27c4d
YZ		 5627 ret = fixup_tree_root_location(root, dir, dentry,
5628 &location, &sub_root);
5629 if (ret < 0) {
5630 if (ret != -ENOENT)
5631 inode = ERR_PTR(ret);
5632 else
5633 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5634 } else {
73f73415 5635 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
39279cc3 5636 }
76dda93c
YZ		 5637 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5638
34d19bad 5639 if (!IS_ERR(inode) && root != sub_root) {
c71bf099
YZ		 5640 down_read(&root->fs_info->cleanup_work_sem);
5641 if (!(inode->i_sb->s_flags & MS_RDONLY))
66b4ffd1 5642 ret = btrfs_orphan_cleanup(sub_root);
c71bf099 5643 up_read(&root->fs_info->cleanup_work_sem);
01cd3367
JB		 5644 if (ret) {
5645 iput(inode);
66b4ffd1 5646 inode = ERR_PTR(ret);
01cd3367 5647 }
c71bf099
YZ		 5648 }
5649
3de4586c
CM		 5650 return inode;
5651}
5652
fe15ce44 5653static int btrfs_dentry_delete(const struct dentry *dentry)
76dda93c
YZ		 5654{
5655 struct btrfs_root *root;
2b0143b5 5656 struct inode *inode = d_inode(dentry);
76dda93c 5657
848cce0d 5658 if (!inode && !IS_ROOT(dentry))
2b0143b5 5659 inode = d_inode(dentry->d_parent);
76dda93c 5660
848cce0d
LZ		 5661 if (inode) {
5662 root = BTRFS_I(inode)->root;
efefb143
YZ		 5663 if (btrfs_root_refs(&root->root_item) == 0)
5664 return 1;
848cce0d
LZ		 5665
5666 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5667 return 1;
efefb143 5668 }
76dda93c
YZ		 5669 return 0;
5670}
5671
b4aff1f8
JB		 5672static void btrfs_dentry_release(struct dentry *dentry)
5673{
944a4515 5674 kfree(dentry->d_fsdata);
b4aff1f8
JB		 5675}
5676
3de4586c 5677static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
00cd8dd3 5678 unsigned int flags)
3de4586c 5679{
5662344b 5680 struct inode *inode;
a66e7cc6 5681
5662344b
TI		 5682 inode = btrfs_lookup_dentry(dir, dentry);
5683 if (IS_ERR(inode)) {
5684 if (PTR_ERR(inode) == -ENOENT)
5685 inode = NULL;
5686 else
5687 return ERR_CAST(inode);
5688 }
5689
41d28bca 5690 return d_splice_alias(inode, dentry);
39279cc3
CM		 5691}
5692
16cdcec7 5693unsigned char btrfs_filetype_table[] = {
39279cc3
CM		 5694 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5695};
5696
9cdda8d3 5697static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
39279cc3 5698{
9cdda8d3 5699 struct inode *inode = file_inode(file);
39279cc3
CM		 5700 struct btrfs_root *root = BTRFS_I(inode)->root;
5701 struct btrfs_item *item;
5702 struct btrfs_dir_item *di;
5703 struct btrfs_key key;
5f39d397 5704 struct btrfs_key found_key;
39279cc3 5705 struct btrfs_path *path;
16cdcec7
MX		 5706 struct list_head ins_list;
5707 struct list_head del_list;
39279cc3 5708 int ret;
5f39d397 5709 struct extent_buffer *leaf;
39279cc3 5710 int slot;
39279cc3
CM		 5711 unsigned char d_type;
5712 int over = 0;
5713 u32 di_cur;
5714 u32 di_total;
5715 u32 di_len;
5716 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 5717 char tmp_name[32];
5718 char *name_ptr;
5719 int name_len;
9cdda8d3 5720 int is_curr = 0; /* ctx->pos points to the current index? */
39279cc3
CM		 5721
5722 /* FIXME, use a real flag for deciding about the key type */
5723 if (root->fs_info->tree_root == root)
5724 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 5725
9cdda8d3
AV		 5726 if (!dir_emit_dots(file, ctx))
5727 return 0;
5728
49593bfa 5729 path = btrfs_alloc_path();
16cdcec7
MX		 5730 if (!path)
5731 return -ENOMEM;
ff5714cc 5732
e4058b54 5733 path->reada = READA_FORWARD;
49593bfa 5734
16cdcec7
MX		 5735 if (key_type == BTRFS_DIR_INDEX_KEY) {
5736 INIT_LIST_HEAD(&ins_list);
5737 INIT_LIST_HEAD(&del_list);
5738 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5739 }
5740
962a298f 5741 key.type = key_type;
9cdda8d3 5742 key.offset = ctx->pos;
33345d01 5743 key.objectid = btrfs_ino(inode);
5f39d397 5744
39279cc3
CM		 5745 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5746 if (ret < 0)
5747 goto err;
49593bfa
DW		 5748
5749 while (1) {
5f39d397 5750 leaf = path->nodes[0];
39279cc3 5751 slot = path->slots[0];
b9e03af0
LZ		 5752 if (slot >= btrfs_header_nritems(leaf)) {
5753 ret = btrfs_next_leaf(root, path);
5754 if (ret < 0)
5755 goto err;
5756 else if (ret > 0)
5757 break;
5758 continue;
39279cc3 5759 }
3de4586c 5760
dd3cc16b 5761 item = btrfs_item_nr(slot);
5f39d397
CM		 5762 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5763
5764 if (found_key.objectid != key.objectid)
39279cc3 5765 break;
962a298f 5766 if (found_key.type != key_type)
39279cc3 5767 break;
9cdda8d3 5768 if (found_key.offset < ctx->pos)
b9e03af0 5769 goto next;
16cdcec7
MX		 5770 if (key_type == BTRFS_DIR_INDEX_KEY &&
5771 btrfs_should_delete_dir_index(&del_list,
5772 found_key.offset))
5773 goto next;
5f39d397 5774
9cdda8d3 5775 ctx->pos = found_key.offset;
16cdcec7 5776 is_curr = 1;
49593bfa 5777
39279cc3
CM		 5778 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5779 di_cur = 0;
5f39d397 5780 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 5781
5782 while (di_cur < di_total) {
5f39d397
CM		 5783 struct btrfs_key location;
5784
22a94d44
JB		 5785 if (verify_dir_item(root, leaf, di))
5786 break;
5787
5f39d397 5788 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 5789 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 5790 name_ptr = tmp_name;
5791 } else {
5792 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 5793 if (!name_ptr) {
5794 ret = -ENOMEM;
5795 goto err;
5796 }
5f39d397
CM		 5797 }
5798 read_extent_buffer(leaf, name_ptr,
5799 (unsigned long)(di + 1), name_len);
5800
5801 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5802 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c 5803
fede766f 5804
3de4586c 5805 /* is this a reference to our own snapshot? If so
8c9c2bf7
AJ		 5806 * skip it.
5807 *
5808 * In contrast to old kernels, we insert the snapshot's
5809 * dir item and dir index after it has been created, so
5810 * we won't find a reference to our own snapshot. We
5811 * still keep the following code for backward
5812 * compatibility.
3de4586c
CM		 5813 */
5814 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5815 location.objectid == root->root_key.objectid) {
5816 over = 0;
5817 goto skip;
5818 }
9cdda8d3
AV		 5819 over = !dir_emit(ctx, name_ptr, name_len,
5820 location.objectid, d_type);
5f39d397 5821
3de4586c 5822skip:
5f39d397
CM		 5823 if (name_ptr != tmp_name)
5824 kfree(name_ptr);
5825
39279cc3
CM		 5826 if (over)
5827 goto nopos;
5103e947 5828 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 5829 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 5830 di_cur += di_len;
5831 di = (struct btrfs_dir_item *)((char *)di + di_len);
5832 }
b9e03af0
LZ		 5833next:
5834 path->slots[0]++;
39279cc3 5835 }
49593bfa 5836
16cdcec7
MX		 5837 if (key_type == BTRFS_DIR_INDEX_KEY) {
5838 if (is_curr)
9cdda8d3
AV		 5839 ctx->pos++;
5840 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
16cdcec7
MX		 5841 if (ret)
5842 goto nopos;
5843 }
5844
49593bfa 5845 /* Reached end of directory/root. Bump pos past the last item. */
db62efbb
ZB		 5846 ctx->pos++;
5847
5848 /*
5849 * Stop new entries from being returned after we return the last
5850 * entry.
5851 *
5852 * New directory entries are assigned a strictly increasing
5853 * offset. This means that new entries created during readdir
5854 * are *guaranteed* to be seen in the future by that readdir.
5855 * This has broken buggy programs which operate on names as
5856 * they're returned by readdir. Until we re-use freed offsets
5857 * we have this hack to stop new entries from being returned
5858 * under the assumption that they'll never reach this huge
5859 * offset.
5860 *
5861 * This is being careful not to overflow 32bit loff_t unless the
5862 * last entry requires it because doing so has broken 32bit apps
5863 * in the past.
5864 */
5865 if (key_type == BTRFS_DIR_INDEX_KEY) {
5866 if (ctx->pos >= INT_MAX)
5867 ctx->pos = LLONG_MAX;
5868 else
5869 ctx->pos = INT_MAX;
5870 }
39279cc3
CM		 5871nopos:
5872 ret = 0;
5873err:
16cdcec7
MX		 5874 if (key_type == BTRFS_DIR_INDEX_KEY)
5875 btrfs_put_delayed_items(&ins_list, &del_list);
39279cc3 5876 btrfs_free_path(path);
39279cc3
CM		 5877 return ret;
5878}
5879
a9185b41 5880int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
39279cc3
CM		 5881{
5882 struct btrfs_root *root = BTRFS_I(inode)->root;
5883 struct btrfs_trans_handle *trans;
5884 int ret = 0;
0af3d00b 5885 bool nolock = false;
39279cc3 5886
72ac3c0d 5887 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4ca8b41e
CM		 5888 return 0;
5889
83eea1f1 5890 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
82d5902d 5891 nolock = true;
0af3d00b 5892
a9185b41 5893 if (wbc->sync_mode == WB_SYNC_ALL) {
0af3d00b 5894 if (nolock)
7a7eaa40 5895 trans = btrfs_join_transaction_nolock(root);
0af3d00b 5896 else
7a7eaa40 5897 trans = btrfs_join_transaction(root);
3612b495
TI		 5898 if (IS_ERR(trans))
5899 return PTR_ERR(trans);
a698d075 5900 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 5901 }
5902 return ret;
5903}
5904
5905/*
54aa1f4d 5906 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 5907 * inode changes. But, it is most likely to find the inode in cache.
5908 * FIXME, needs more benchmarking...there are no reasons other than performance
5909 * to keep or drop this code.
5910 */
48a3b636 5911static int btrfs_dirty_inode(struct inode *inode)
39279cc3
CM		 5912{
5913 struct btrfs_root *root = BTRFS_I(inode)->root;
5914 struct btrfs_trans_handle *trans;
8929ecfa
YZ		 5915 int ret;
5916
72ac3c0d 5917 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
22c44fe6 5918 return 0;
39279cc3 5919
7a7eaa40 5920 trans = btrfs_join_transaction(root);
22c44fe6
JB		 5921 if (IS_ERR(trans))
5922 return PTR_ERR(trans);
8929ecfa
YZ		 5923
5924 ret = btrfs_update_inode(trans, root, inode);
94b60442
CM		 5925 if (ret && ret == -ENOSPC) {
5926 /* whoops, lets try again with the full transaction */
5927 btrfs_end_transaction(trans, root);
5928 trans = btrfs_start_transaction(root, 1);
22c44fe6
JB		 5929 if (IS_ERR(trans))
5930 return PTR_ERR(trans);
8929ecfa 5931
94b60442 5932 ret = btrfs_update_inode(trans, root, inode);
94b60442 5933 }
39279cc3 5934 btrfs_end_transaction(trans, root);
16cdcec7
MX		 5935 if (BTRFS_I(inode)->delayed_node)
5936 btrfs_balance_delayed_items(root);
22c44fe6
JB		 5937
5938 return ret;
5939}
5940
5941/*
5942 * This is a copy of file_update_time. We need this so we can return error on
5943 * ENOSPC for updating the inode in the case of file write and mmap writes.
5944 */
e41f941a
JB		 5945static int btrfs_update_time(struct inode *inode, struct timespec *now,
5946 int flags)
22c44fe6 5947{
2bc55652
AB		 5948 struct btrfs_root *root = BTRFS_I(inode)->root;
5949
5950 if (btrfs_root_readonly(root))
5951 return -EROFS;
5952
e41f941a 5953 if (flags & S_VERSION)
22c44fe6 5954 inode_inc_iversion(inode);
e41f941a
JB		 5955 if (flags & S_CTIME)
5956 inode->i_ctime = *now;
5957 if (flags & S_MTIME)
5958 inode->i_mtime = *now;
5959 if (flags & S_ATIME)
5960 inode->i_atime = *now;
5961 return btrfs_dirty_inode(inode);
39279cc3
CM		 5962}
5963
d352ac68
CM		 5964/*
5965 * find the highest existing sequence number in a directory
5966 * and then set the in-memory index_cnt variable to reflect
5967 * free sequence numbers
5968 */
aec7477b
JB		 5969static int btrfs_set_inode_index_count(struct inode *inode)
5970{
5971 struct btrfs_root *root = BTRFS_I(inode)->root;
5972 struct btrfs_key key, found_key;
5973 struct btrfs_path *path;
5974 struct extent_buffer *leaf;
5975 int ret;
5976
33345d01 5977 key.objectid = btrfs_ino(inode);
962a298f 5978 key.type = BTRFS_DIR_INDEX_KEY;
aec7477b
JB		 5979 key.offset = (u64)-1;
5980
5981 path = btrfs_alloc_path();
5982 if (!path)
5983 return -ENOMEM;
5984
5985 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5986 if (ret < 0)
5987 goto out;
5988 /* FIXME: we should be able to handle this */
5989 if (ret == 0)
5990 goto out;
5991 ret = 0;
5992
5993 /*
5994 * MAGIC NUMBER EXPLANATION:
5995 * since we search a directory based on f_pos we have to start at 2
5996 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5997 * else has to start at 2
5998 */
5999 if (path->slots[0] == 0) {
6000 BTRFS_I(inode)->index_cnt = 2;
6001 goto out;
6002 }
6003
6004 path->slots[0]--;
6005
6006 leaf = path->nodes[0];
6007 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6008
33345d01 6009 if (found_key.objectid != btrfs_ino(inode) ||
962a298f 6010 found_key.type != BTRFS_DIR_INDEX_KEY) {
aec7477b
JB		 6011 BTRFS_I(inode)->index_cnt = 2;
6012 goto out;
6013 }
6014
6015 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
6016out:
6017 btrfs_free_path(path);
6018 return ret;
6019}
6020
d352ac68
CM		 6021/*
6022 * helper to find a free sequence number in a given directory. This current
6023 * code is very simple, later versions will do smarter things in the btree
6024 */
3de4586c 6025int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 6026{
6027 int ret = 0;
6028
6029 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
16cdcec7
MX		 6030 ret = btrfs_inode_delayed_dir_index_count(dir);
6031 if (ret) {
6032 ret = btrfs_set_inode_index_count(dir);
6033 if (ret)
6034 return ret;
6035 }
aec7477b
JB		 6036 }
6037
00e4e6b3 6038 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 6039 BTRFS_I(dir)->index_cnt++;
6040
6041 return ret;
6042}
6043
b0d5d10f
CM		 6044static int btrfs_insert_inode_locked(struct inode *inode)
6045{
6046 struct btrfs_iget_args args;
6047 args.location = &BTRFS_I(inode)->location;
6048 args.root = BTRFS_I(inode)->root;
6049
6050 return insert_inode_locked4(inode,
6051 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6052 btrfs_find_actor, &args);
6053}
6054
39279cc3
CM		 6055static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6056 struct btrfs_root *root,
aec7477b 6057 struct inode *dir,
9c58309d 6058 const char *name, int name_len,
175a4eb7
AV		 6059 u64 ref_objectid, u64 objectid,
6060 umode_t mode, u64 *index)
39279cc3
CM		 6061{
6062 struct inode *inode;
5f39d397 6063 struct btrfs_inode_item *inode_item;
39279cc3 6064 struct btrfs_key *location;
5f39d397 6065 struct btrfs_path *path;
9c58309d
CM		 6066 struct btrfs_inode_ref *ref;
6067 struct btrfs_key key[2];
6068 u32 sizes[2];
ef3b9af5 6069 int nitems = name ? 2 : 1;
9c58309d 6070 unsigned long ptr;
39279cc3 6071 int ret;
39279cc3 6072
5f39d397 6073 path = btrfs_alloc_path();
d8926bb3
MF		 6074 if (!path)
6075 return ERR_PTR(-ENOMEM);
5f39d397 6076
39279cc3 6077 inode = new_inode(root->fs_info->sb);
8fb27640
YS		 6078 if (!inode) {
6079 btrfs_free_path(path);
39279cc3 6080 return ERR_PTR(-ENOMEM);
8fb27640 6081 }
39279cc3 6082
5762b5c9
FM		 6083 /*
6084 * O_TMPFILE, set link count to 0, so that after this point,
6085 * we fill in an inode item with the correct link count.
6086 */
6087 if (!name)
6088 set_nlink(inode, 0);
6089
581bb050
LZ		 6090 /*
6091 * we have to initialize this early, so we can reclaim the inode
6092 * number if we fail afterwards in this function.
6093 */
6094 inode->i_ino = objectid;
6095
ef3b9af5 6096 if (dir && name) {
1abe9b8a 6097 trace_btrfs_inode_request(dir);
6098
3de4586c 6099 ret = btrfs_set_inode_index(dir, index);
09771430 6100 if (ret) {
8fb27640 6101 btrfs_free_path(path);
09771430 6102 iput(inode);
aec7477b 6103 return ERR_PTR(ret);
09771430 6104 }
ef3b9af5
FM		 6105 } else if (dir) {
6106 *index = 0;
aec7477b
JB		 6107 }
6108 /*
6109 * index_cnt is ignored for everything but a dir,
6110 * btrfs_get_inode_index_count has an explanation for the magic
6111 * number
6112 */
6113 BTRFS_I(inode)->index_cnt = 2;
67de1176 6114 BTRFS_I(inode)->dir_index = *index;
39279cc3 6115 BTRFS_I(inode)->root = root;
e02119d5 6116 BTRFS_I(inode)->generation = trans->transid;
76195853 6117 inode->i_generation = BTRFS_I(inode)->generation;
b888db2b 6118
5dc562c5
JB		 6119 /*
6120 * We could have gotten an inode number from somebody who was fsynced
6121 * and then removed in this same transaction, so let's just set full
6122 * sync since it will be a full sync anyway and this will blow away the
6123 * old info in the log.
6124 */
6125 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6126
9c58309d 6127 key[0].objectid = objectid;
962a298f 6128 key[0].type = BTRFS_INODE_ITEM_KEY;
9c58309d
CM		 6129 key[0].offset = 0;
6130
9c58309d 6131 sizes[0] = sizeof(struct btrfs_inode_item);
ef3b9af5
FM		 6132
6133 if (name) {
6134 /*
6135 * Start new inodes with an inode_ref. This is slightly more
6136 * efficient for small numbers of hard links since they will
6137 * be packed into one item. Extended refs will kick in if we
6138 * add more hard links than can fit in the ref item.
6139 */
6140 key[1].objectid = objectid;
962a298f 6141 key[1].type = BTRFS_INODE_REF_KEY;
ef3b9af5
FM		 6142 key[1].offset = ref_objectid;
6143
6144 sizes[1] = name_len + sizeof(*ref);
6145 }
9c58309d 6146
b0d5d10f
CM		 6147 location = &BTRFS_I(inode)->location;
6148 location->objectid = objectid;
6149 location->offset = 0;
962a298f 6150 location->type = BTRFS_INODE_ITEM_KEY;
b0d5d10f
CM		 6151
6152 ret = btrfs_insert_inode_locked(inode);
6153 if (ret < 0)
6154 goto fail;
6155
b9473439 6156 path->leave_spinning = 1;
ef3b9af5 6157 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
9c58309d 6158 if (ret != 0)
b0d5d10f 6159 goto fail_unlock;
5f39d397 6160
ecc11fab 6161 inode_init_owner(inode, dir, mode);
a76a3cd4 6162 inode_set_bytes(inode, 0);
9cc97d64 6163
04b285f3 6164 inode->i_mtime = current_fs_time(inode->i_sb);
9cc97d64 6165 inode->i_atime = inode->i_mtime;
6166 inode->i_ctime = inode->i_mtime;
6167 BTRFS_I(inode)->i_otime = inode->i_mtime;
6168
5f39d397
CM		 6169 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6170 struct btrfs_inode_item);
293f7e07
LZ		 6171 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6172 sizeof(*inode_item));
e02119d5 6173 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d 6174
ef3b9af5
FM		 6175 if (name) {
6176 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6177 struct btrfs_inode_ref);
6178 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6179 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6180 ptr = (unsigned long)(ref + 1);
6181 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6182 }
9c58309d 6183
5f39d397
CM		 6184 btrfs_mark_buffer_dirty(path->nodes[0]);
6185 btrfs_free_path(path);
6186
6cbff00f
CH		 6187 btrfs_inherit_iflags(inode, dir);
6188
569254b0 6189 if (S_ISREG(mode)) {
94272164
CM		 6190 if (btrfs_test_opt(root, NODATASUM))
6191 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
213490b3 6192 if (btrfs_test_opt(root, NODATACOW))
f2bdf9a8
JB		 6193 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6194 BTRFS_INODE_NODATASUM;
94272164
CM		 6195 }
6196
5d4f98a2 6197 inode_tree_add(inode);
1abe9b8a 6198
6199 trace_btrfs_inode_new(inode);
1973f0fa 6200 btrfs_set_inode_last_trans(trans, inode);
1abe9b8a 6201
8ea05e3a
AB		 6202 btrfs_update_root_times(trans, root);
6203
63541927
FDBM		 6204 ret = btrfs_inode_inherit_props(trans, inode, dir);
6205 if (ret)
6206 btrfs_err(root->fs_info,
6207 "error inheriting props for ino %llu (root %llu): %d",
6208 btrfs_ino(inode), root->root_key.objectid, ret);
6209
39279cc3 6210 return inode;
b0d5d10f
CM		 6211
6212fail_unlock:
6213 unlock_new_inode(inode);
5f39d397 6214fail:
ef3b9af5 6215 if (dir && name)
aec7477b 6216 BTRFS_I(dir)->index_cnt--;
5f39d397 6217 btrfs_free_path(path);
09771430 6218 iput(inode);
5f39d397 6219 return ERR_PTR(ret);
39279cc3
CM		 6220}
6221
6222static inline u8 btrfs_inode_type(struct inode *inode)
6223{
6224 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6225}
6226
d352ac68
CM		 6227/*
6228 * utility function to add 'inode' into 'parent_inode' with
6229 * a give name and a given sequence number.
6230 * if 'add_backref' is true, also insert a backref from the
6231 * inode to the parent directory.
6232 */
e02119d5
CM		 6233int btrfs_add_link(struct btrfs_trans_handle *trans,
6234 struct inode *parent_inode, struct inode *inode,
6235 const char *name, int name_len, int add_backref, u64 index)
39279cc3 6236{
4df27c4d 6237 int ret = 0;
39279cc3 6238 struct btrfs_key key;
e02119d5 6239 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
33345d01
LZ		 6240 u64 ino = btrfs_ino(inode);
6241 u64 parent_ino = btrfs_ino(parent_inode);
5f39d397 6242
33345d01 6243 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 6244 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6245 } else {
33345d01 6246 key.objectid = ino;
962a298f 6247 key.type = BTRFS_INODE_ITEM_KEY;
4df27c4d
YZ		 6248 key.offset = 0;
6249 }
6250
33345d01 6251 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 6252 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6253 key.objectid, root->root_key.objectid,
33345d01 6254 parent_ino, index, name, name_len);
4df27c4d 6255 } else if (add_backref) {
33345d01
LZ		 6256 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6257 parent_ino, index);
4df27c4d 6258 }
39279cc3 6259
79787eaa
JM		 6260 /* Nothing to clean up yet */
6261 if (ret)
6262 return ret;
4df27c4d 6263
79787eaa
JM		 6264 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6265 parent_inode, &key,
6266 btrfs_inode_type(inode), index);
9c52057c 6267 if (ret == -EEXIST || ret == -EOVERFLOW)
79787eaa
JM		 6268 goto fail_dir_item;
6269 else if (ret) {
6270 btrfs_abort_transaction(trans, root, ret);
6271 return ret;
39279cc3 6272 }
79787eaa
JM		 6273
6274 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6275 name_len * 2);
0c4d2d95 6276 inode_inc_iversion(parent_inode);
04b285f3
DD		 6277 parent_inode->i_mtime = parent_inode->i_ctime =
6278 current_fs_time(parent_inode->i_sb);
79787eaa
JM		 6279 ret = btrfs_update_inode(trans, root, parent_inode);
6280 if (ret)
6281 btrfs_abort_transaction(trans, root, ret);
39279cc3 6282 return ret;
fe66a05a
CM		 6283
6284fail_dir_item:
6285 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6286 u64 local_index;
6287 int err;
6288 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6289 key.objectid, root->root_key.objectid,
6290 parent_ino, &local_index, name, name_len);
6291
6292 } else if (add_backref) {
6293 u64 local_index;
6294 int err;
6295
6296 err = btrfs_del_inode_ref(trans, root, name, name_len,
6297 ino, parent_ino, &local_index);
6298 }
6299 return ret;
39279cc3
CM		 6300}
6301
6302static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
a1b075d2
JB		 6303 struct inode *dir, struct dentry *dentry,
6304 struct inode *inode, int backref, u64 index)
39279cc3 6305{
a1b075d2
JB		 6306 int err = btrfs_add_link(trans, dir, inode,
6307 dentry->d_name.name, dentry->d_name.len,
6308 backref, index);
39279cc3
CM		 6309 if (err > 0)
6310 err = -EEXIST;
6311 return err;
6312}
6313
618e21d5 6314static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1a67aafb 6315 umode_t mode, dev_t rdev)
618e21d5
JB		 6316{
6317 struct btrfs_trans_handle *trans;
6318 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 6319 struct inode *inode = NULL;
618e21d5
JB		 6320 int err;
6321 int drop_inode = 0;
6322 u64 objectid;
00e4e6b3 6323 u64 index = 0;
618e21d5 6324
9ed74f2d
JB		 6325 /*
6326 * 2 for inode item and ref
6327 * 2 for dir items
6328 * 1 for xattr if selinux is on
6329 */
a22285a6
YZ		 6330 trans = btrfs_start_transaction(root, 5);
6331 if (IS_ERR(trans))
6332 return PTR_ERR(trans);
1832a6d5 6333
581bb050
LZ		 6334 err = btrfs_find_free_ino(root, &objectid);
6335 if (err)
6336 goto out_unlock;
6337
aec7477b 6338 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 6339 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 6340 mode, &index);
7cf96da3
TI		 6341 if (IS_ERR(inode)) {
6342 err = PTR_ERR(inode);
618e21d5 6343 goto out_unlock;
7cf96da3 6344 }
618e21d5 6345
ad19db71
CS		 6346 /*
6347 * If the active LSM wants to access the inode during
6348 * d_instantiate it needs these. Smack checks to see
6349 * if the filesystem supports xattrs by looking at the
6350 * ops vector.
6351 */
ad19db71 6352 inode->i_op = &btrfs_special_inode_operations;
b0d5d10f
CM		 6353 init_special_inode(inode, inode->i_mode, rdev);
6354
6355 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
618e21d5 6356 if (err)
b0d5d10f
CM		 6357 goto out_unlock_inode;
6358
6359 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6360 if (err) {
6361 goto out_unlock_inode;
6362 } else {
1b4ab1bb 6363 btrfs_update_inode(trans, root, inode);
b0d5d10f 6364 unlock_new_inode(inode);
08c422c2 6365 d_instantiate(dentry, inode);
618e21d5 6366 }
b0d5d10f 6367
618e21d5 6368out_unlock:
7ad85bb7 6369 btrfs_end_transaction(trans, root);
c581afc8 6370 btrfs_balance_delayed_items(root);
b53d3f5d 6371 btrfs_btree_balance_dirty(root);
618e21d5
JB		 6372 if (drop_inode) {
6373 inode_dec_link_count(inode);
6374 iput(inode);
6375 }
618e21d5 6376 return err;
b0d5d10f
CM		 6377
6378out_unlock_inode:
6379 drop_inode = 1;
6380 unlock_new_inode(inode);
6381 goto out_unlock;
6382
618e21d5
JB		 6383}
6384
39279cc3 6385static int btrfs_create(struct inode *dir, struct dentry *dentry,
ebfc3b49 6386 umode_t mode, bool excl)
39279cc3
CM		 6387{
6388 struct btrfs_trans_handle *trans;
6389 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 6390 struct inode *inode = NULL;
43baa579 6391 int drop_inode_on_err = 0;
a22285a6 6392 int err;
39279cc3 6393 u64 objectid;
00e4e6b3 6394 u64 index = 0;
39279cc3 6395
9ed74f2d
JB		 6396 /*
6397 * 2 for inode item and ref
6398 * 2 for dir items
6399 * 1 for xattr if selinux is on
6400 */
a22285a6
YZ		 6401 trans = btrfs_start_transaction(root, 5);
6402 if (IS_ERR(trans))
6403 return PTR_ERR(trans);
9ed74f2d 6404
581bb050
LZ		 6405 err = btrfs_find_free_ino(root, &objectid);
6406 if (err)
6407 goto out_unlock;
6408
aec7477b 6409 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 6410 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 6411 mode, &index);
7cf96da3
TI		 6412 if (IS_ERR(inode)) {
6413 err = PTR_ERR(inode);
39279cc3 6414 goto out_unlock;
7cf96da3 6415 }
43baa579 6416 drop_inode_on_err = 1;
ad19db71
CS		 6417 /*
6418 * If the active LSM wants to access the inode during
6419 * d_instantiate it needs these. Smack checks to see
6420 * if the filesystem supports xattrs by looking at the
6421 * ops vector.
6422 */
6423 inode->i_fop = &btrfs_file_operations;
6424 inode->i_op = &btrfs_file_inode_operations;
b0d5d10f 6425 inode->i_mapping->a_ops = &btrfs_aops;
b0d5d10f
CM		 6426
6427 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6428 if (err)
6429 goto out_unlock_inode;
6430
6431 err = btrfs_update_inode(trans, root, inode);
6432 if (err)
6433 goto out_unlock_inode;
ad19db71 6434
a1b075d2 6435 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3 6436 if (err)
b0d5d10f 6437 goto out_unlock_inode;
43baa579 6438
43baa579 6439 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
b0d5d10f 6440 unlock_new_inode(inode);
43baa579
FB		 6441 d_instantiate(dentry, inode);
6442
39279cc3 6443out_unlock:
7ad85bb7 6444 btrfs_end_transaction(trans, root);
43baa579 6445 if (err && drop_inode_on_err) {
39279cc3
CM		 6446 inode_dec_link_count(inode);
6447 iput(inode);
6448 }
c581afc8 6449 btrfs_balance_delayed_items(root);
b53d3f5d 6450 btrfs_btree_balance_dirty(root);
39279cc3 6451 return err;
b0d5d10f
CM		 6452
6453out_unlock_inode:
6454 unlock_new_inode(inode);
6455 goto out_unlock;
6456
39279cc3
CM		 6457}
6458
6459static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6460 struct dentry *dentry)
6461{
271dba45 6462 struct btrfs_trans_handle *trans = NULL;
39279cc3 6463 struct btrfs_root *root = BTRFS_I(dir)->root;
2b0143b5 6464 struct inode *inode = d_inode(old_dentry);
00e4e6b3 6465 u64 index;
39279cc3
CM		 6466 int err;
6467 int drop_inode = 0;
6468
4a8be425
TH		 6469 /* do not allow sys_link's with other subvols of the same device */
6470 if (root->objectid != BTRFS_I(inode)->root->objectid)
3ab3564f 6471 return -EXDEV;
4a8be425 6472
f186373f 6473 if (inode->i_nlink >= BTRFS_LINK_MAX)
c055e99e 6474 return -EMLINK;
4a8be425 6475
3de4586c 6476 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 6477 if (err)
6478 goto fail;
6479
a22285a6 6480 /*
7e6b6465 6481 * 2 items for inode and inode ref
a22285a6 6482 * 2 items for dir items
7e6b6465 6483 * 1 item for parent inode
a22285a6 6484 */
7e6b6465 6485 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 6486 if (IS_ERR(trans)) {
6487 err = PTR_ERR(trans);
271dba45 6488 trans = NULL;
a22285a6
YZ		 6489 goto fail;
6490 }
5f39d397 6491
67de1176
MX		 6492 /* There are several dir indexes for this inode, clear the cache. */
6493 BTRFS_I(inode)->dir_index = 0ULL;
8b558c5f 6494 inc_nlink(inode);
0c4d2d95 6495 inode_inc_iversion(inode);
04b285f3 6496 inode->i_ctime = current_fs_time(inode->i_sb);
7de9c6ee 6497 ihold(inode);
e9976151 6498 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
aec7477b 6499
a1b075d2 6500 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5f39d397 6501
a5719521 6502 if (err) {
54aa1f4d 6503 drop_inode = 1;
a5719521 6504 } else {
10d9f309 6505 struct dentry *parent = dentry->d_parent;
a5719521 6506 err = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 6507 if (err)
6508 goto fail;
ef3b9af5
FM		 6509 if (inode->i_nlink == 1) {
6510 /*
6511 * If new hard link count is 1, it's a file created
6512 * with open(2) O_TMPFILE flag.
6513 */
6514 err = btrfs_orphan_del(trans, inode);
6515 if (err)
6516 goto fail;
6517 }
08c422c2 6518 d_instantiate(dentry, inode);
6a912213 6519 btrfs_log_new_name(trans, inode, NULL, parent);
a5719521 6520 }
39279cc3 6521
c581afc8 6522 btrfs_balance_delayed_items(root);
1832a6d5 6523fail:
271dba45
FM		 6524 if (trans)
6525 btrfs_end_transaction(trans, root);
39279cc3
CM		 6526 if (drop_inode) {
6527 inode_dec_link_count(inode);
6528 iput(inode);
6529 }
b53d3f5d 6530 btrfs_btree_balance_dirty(root);
39279cc3
CM		 6531 return err;
6532}
6533
18bb1db3 6534static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
39279cc3 6535{
b9d86667 6536 struct inode *inode = NULL;
39279cc3
CM		 6537 struct btrfs_trans_handle *trans;
6538 struct btrfs_root *root = BTRFS_I(dir)->root;
6539 int err = 0;
6540 int drop_on_err = 0;
b9d86667 6541 u64 objectid = 0;
00e4e6b3 6542 u64 index = 0;
39279cc3 6543
9ed74f2d
JB		 6544 /*
6545 * 2 items for inode and ref
6546 * 2 items for dir items
6547 * 1 for xattr if selinux is on
6548 */
a22285a6
YZ		 6549 trans = btrfs_start_transaction(root, 5);
6550 if (IS_ERR(trans))
6551 return PTR_ERR(trans);
39279cc3 6552
581bb050
LZ		 6553 err = btrfs_find_free_ino(root, &objectid);
6554 if (err)
6555 goto out_fail;
6556
aec7477b 6557 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 6558 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 6559 S_IFDIR | mode, &index);
39279cc3
CM		 6560 if (IS_ERR(inode)) {
6561 err = PTR_ERR(inode);
6562 goto out_fail;
6563 }
5f39d397 6564
39279cc3 6565 drop_on_err = 1;
b0d5d10f
CM		 6566 /* these must be set before we unlock the inode */
6567 inode->i_op = &btrfs_dir_inode_operations;
6568 inode->i_fop = &btrfs_dir_file_operations;
33268eaf 6569
2a7dba39 6570 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf 6571 if (err)
b0d5d10f 6572 goto out_fail_inode;
39279cc3 6573
dbe674a9 6574 btrfs_i_size_write(inode, 0);
39279cc3
CM		 6575 err = btrfs_update_inode(trans, root, inode);
6576 if (err)
b0d5d10f 6577 goto out_fail_inode;
5f39d397 6578
a1b075d2
JB		 6579 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6580 dentry->d_name.len, 0, index);
39279cc3 6581 if (err)
b0d5d10f 6582 goto out_fail_inode;
5f39d397 6583
39279cc3 6584 d_instantiate(dentry, inode);
b0d5d10f
CM		 6585 /*
6586 * mkdir is special. We're unlocking after we call d_instantiate
6587 * to avoid a race with nfsd calling d_instantiate.
6588 */
6589 unlock_new_inode(inode);
39279cc3 6590 drop_on_err = 0;
39279cc3
CM		 6591
6592out_fail:
7ad85bb7 6593 btrfs_end_transaction(trans, root);
c7cfb8a5
WS		 6594 if (drop_on_err) {
6595 inode_dec_link_count(inode);
39279cc3 6596 iput(inode);
c7cfb8a5 6597 }
c581afc8 6598 btrfs_balance_delayed_items(root);
b53d3f5d 6599 btrfs_btree_balance_dirty(root);
39279cc3 6600 return err;
b0d5d10f
CM		 6601
6602out_fail_inode:
6603 unlock_new_inode(inode);
6604 goto out_fail;
39279cc3
CM		 6605}
6606
e6c4efd8
QW		 6607/* Find next extent map of a given extent map, caller needs to ensure locks */
6608static struct extent_map *next_extent_map(struct extent_map *em)
6609{
6610 struct rb_node *next;
6611
6612 next = rb_next(&em->rb_node);
6613 if (!next)
6614 return NULL;
6615 return container_of(next, struct extent_map, rb_node);
6616}
6617
6618static struct extent_map *prev_extent_map(struct extent_map *em)
6619{
6620 struct rb_node *prev;
6621
6622 prev = rb_prev(&em->rb_node);
6623 if (!prev)
6624 return NULL;
6625 return container_of(prev, struct extent_map, rb_node);
6626}
6627
d352ac68 6628/* helper for btfs_get_extent. Given an existing extent in the tree,
e6c4efd8 6629 * the existing extent is the nearest extent to map_start,
d352ac68 6630 * and an extent that you want to insert, deal with overlap and insert
e6c4efd8 6631 * the best fitted new extent into the tree.
d352ac68 6632 */
3b951516
CM		 6633static int merge_extent_mapping(struct extent_map_tree *em_tree,
6634 struct extent_map *existing,
e6dcd2dc 6635 struct extent_map *em,
51f395ad 6636 u64 map_start)
3b951516 6637{
e6c4efd8
QW		 6638 struct extent_map *prev;
6639 struct extent_map *next;
6640 u64 start;
6641 u64 end;
3b951516 6642 u64 start_diff;
3b951516 6643
e6dcd2dc 6644 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
e6c4efd8
QW		 6645
6646 if (existing->start > map_start) {
6647 next = existing;
6648 prev = prev_extent_map(next);
6649 } else {
6650 prev = existing;
6651 next = next_extent_map(prev);
6652 }
6653
6654 start = prev ? extent_map_end(prev) : em->start;
6655 start = max_t(u64, start, em->start);
6656 end = next ? next->start : extent_map_end(em);
6657 end = min_t(u64, end, extent_map_end(em));
6658 start_diff = start - em->start;
6659 em->start = start;
6660 em->len = end - start;
c8b97818
CM		 6661 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6662 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 6663 em->block_start += start_diff;
c8b97818
CM		 6664 em->block_len -= start_diff;
6665 }
09a2a8f9 6666 return add_extent_mapping(em_tree, em, 0);
3b951516
CM		 6667}
6668
c8b97818 6669static noinline int uncompress_inline(struct btrfs_path *path,
e40da0e5 6670 struct page *page,
c8b97818
CM		 6671 size_t pg_offset, u64 extent_offset,
6672 struct btrfs_file_extent_item *item)
6673{
6674 int ret;
6675 struct extent_buffer *leaf = path->nodes[0];
6676 char *tmp;
6677 size_t max_size;
6678 unsigned long inline_size;
6679 unsigned long ptr;
261507a0 6680 int compress_type;
c8b97818
CM		 6681
6682 WARN_ON(pg_offset != 0);
261507a0 6683 compress_type = btrfs_file_extent_compression(leaf, item);
c8b97818
CM		 6684 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6685 inline_size = btrfs_file_extent_inline_item_len(leaf,
dd3cc16b 6686 btrfs_item_nr(path->slots[0]));
c8b97818 6687 tmp = kmalloc(inline_size, GFP_NOFS);
8d413713
TI		 6688 if (!tmp)
6689 return -ENOMEM;
c8b97818
CM		 6690 ptr = btrfs_file_extent_inline_start(item);
6691
6692 read_extent_buffer(leaf, tmp, ptr, inline_size);
6693
5b050f04 6694 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
261507a0
LZ		 6695 ret = btrfs_decompress(compress_type, tmp, page,
6696 extent_offset, inline_size, max_size);
c8b97818 6697 kfree(tmp);
166ae5a4 6698 return ret;
c8b97818
CM		 6699}
6700
d352ac68
CM		 6701/*
6702 * a bit scary, this does extent mapping from logical file offset to the disk.
d397712b
CM		 6703 * the ugly parts come from merging extents from the disk with the in-ram
6704 * representation. This gets more complex because of the data=ordered code,
d352ac68
CM		 6705 * where the in-ram extents might be locked pending data=ordered completion.
6706 *
6707 * This also copies inline extents directly into the page.
6708 */
d397712b 6709
a52d9a80 6710struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 6711 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 6712 int create)
6713{
6714 int ret;
6715 int err = 0;
a52d9a80
CM		 6716 u64 extent_start = 0;
6717 u64 extent_end = 0;
33345d01 6718 u64 objectid = btrfs_ino(inode);
a52d9a80 6719 u32 found_type;
f421950f 6720 struct btrfs_path *path = NULL;
a52d9a80
CM		 6721 struct btrfs_root *root = BTRFS_I(inode)->root;
6722 struct btrfs_file_extent_item *item;
5f39d397
CM		 6723 struct extent_buffer *leaf;
6724 struct btrfs_key found_key;
a52d9a80
CM		 6725 struct extent_map *em = NULL;
6726 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 6727 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 6728 struct btrfs_trans_handle *trans = NULL;
7ffbb598 6729 const bool new_inline = !page || create;
a52d9a80 6730
a52d9a80 6731again:
890871be 6732 read_lock(&em_tree->lock);
d1310b2e 6733 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 6734 if (em)
6735 em->bdev = root->fs_info->fs_devices->latest_bdev;
890871be 6736 read_unlock(&em_tree->lock);
d1310b2e 6737
a52d9a80 6738 if (em) {
e1c4b745
CM		 6739 if (em->start > start || em->start + em->len <= start)
6740 free_extent_map(em);
6741 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 6742 free_extent_map(em);
6743 else
6744 goto out;
a52d9a80 6745 }
172ddd60 6746 em = alloc_extent_map();
a52d9a80 6747 if (!em) {
d1310b2e
CM		 6748 err = -ENOMEM;
6749 goto out;
a52d9a80 6750 }
e6dcd2dc 6751 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 6752 em->start = EXTENT_MAP_HOLE;
445a6944 6753 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 6754 em->len = (u64)-1;
c8b97818 6755 em->block_len = (u64)-1;
f421950f
CM		 6756
6757 if (!path) {
6758 path = btrfs_alloc_path();
026fd317
JB		 6759 if (!path) {
6760 err = -ENOMEM;
6761 goto out;
6762 }
6763 /*
6764 * Chances are we'll be called again, so go ahead and do
6765 * readahead
6766 */
e4058b54 6767 path->reada = READA_FORWARD;
f421950f
CM		 6768 }
6769
179e29e4
CM		 6770 ret = btrfs_lookup_file_extent(trans, root, path,
6771 objectid, start, trans != NULL);
a52d9a80
CM		 6772 if (ret < 0) {
6773 err = ret;
6774 goto out;
6775 }
6776
6777 if (ret != 0) {
6778 if (path->slots[0] == 0)
6779 goto not_found;
6780 path->slots[0]--;
6781 }
6782
5f39d397
CM		 6783 leaf = path->nodes[0];
6784 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 6785 struct btrfs_file_extent_item);
a52d9a80 6786 /* are we inside the extent that was found? */
5f39d397 6787 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
962a298f 6788 found_type = found_key.type;
5f39d397 6789 if (found_key.objectid != objectid ||
a52d9a80 6790 found_type != BTRFS_EXTENT_DATA_KEY) {
25a50341
JB		 6791 /*
6792 * If we backup past the first extent we want to move forward
6793 * and see if there is an extent in front of us, otherwise we'll
6794 * say there is a hole for our whole search range which can
6795 * cause problems.
6796 */
6797 extent_end = start;
6798 goto next;
a52d9a80
CM		 6799 }
6800
5f39d397
CM		 6801 found_type = btrfs_file_extent_type(leaf, item);
6802 extent_start = found_key.offset;
d899e052
YZ		 6803 if (found_type == BTRFS_FILE_EXTENT_REG ||
6804 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 6805 extent_end = extent_start +
db94535d 6806 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 6807 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6808 size_t size;
514ac8ad 6809 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
fda2832f 6810 extent_end = ALIGN(extent_start + size, root->sectorsize);
9036c102 6811 }
25a50341 6812next:
9036c102
YZ		 6813 if (start >= extent_end) {
6814 path->slots[0]++;
6815 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6816 ret = btrfs_next_leaf(root, path);
6817 if (ret < 0) {
6818 err = ret;
6819 goto out;
a52d9a80 6820 }
9036c102
YZ		 6821 if (ret > 0)
6822 goto not_found;
6823 leaf = path->nodes[0];
a52d9a80 6824 }
9036c102
YZ		 6825 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6826 if (found_key.objectid != objectid ||
6827 found_key.type != BTRFS_EXTENT_DATA_KEY)
6828 goto not_found;
6829 if (start + len <= found_key.offset)
6830 goto not_found;
e2eca69d
WS		 6831 if (start > found_key.offset)
6832 goto next;
9036c102 6833 em->start = start;
70c8a91c 6834 em->orig_start = start;
9036c102
YZ		 6835 em->len = found_key.offset - start;
6836 goto not_found_em;
6837 }
6838
7ffbb598
FM		 6839 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6840
d899e052
YZ		 6841 if (found_type == BTRFS_FILE_EXTENT_REG ||
6842 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80
CM		 6843 goto insert;
6844 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 6845 unsigned long ptr;
a52d9a80 6846 char *map;
3326d1b0
CM		 6847 size_t size;
6848 size_t extent_offset;
6849 size_t copy_size;
a52d9a80 6850
7ffbb598 6851 if (new_inline)
689f9346 6852 goto out;
5f39d397 6853
514ac8ad 6854 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
9036c102 6855 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 6856 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 6857 size - extent_offset);
3326d1b0 6858 em->start = extent_start + extent_offset;
fda2832f 6859 em->len = ALIGN(copy_size, root->sectorsize);
b4939680 6860 em->orig_block_len = em->len;
70c8a91c 6861 em->orig_start = em->start;
689f9346 6862 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 6863 if (create == 0 && !PageUptodate(page)) {
261507a0
LZ		 6864 if (btrfs_file_extent_compression(leaf, item) !=
6865 BTRFS_COMPRESS_NONE) {
e40da0e5 6866 ret = uncompress_inline(path, page, pg_offset,
c8b97818 6867 extent_offset, item);
166ae5a4
ZB		 6868 if (ret) {
6869 err = ret;
6870 goto out;
6871 }
c8b97818
CM		 6872 } else {
6873 map = kmap(page);
6874 read_extent_buffer(leaf, map + pg_offset, ptr,
6875 copy_size);
93c82d57
CM		 6876 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6877 memset(map + pg_offset + copy_size, 0,
6878 PAGE_CACHE_SIZE - pg_offset -
6879 copy_size);
6880 }
c8b97818
CM		 6881 kunmap(page);
6882 }
179e29e4
CM		 6883 flush_dcache_page(page);
6884 } else if (create && PageUptodate(page)) {
6bf7e080 6885 BUG();
179e29e4
CM		 6886 if (!trans) {
6887 kunmap(page);
6888 free_extent_map(em);
6889 em = NULL;
ff5714cc 6890
b3b4aa74 6891 btrfs_release_path(path);
7a7eaa40 6892 trans = btrfs_join_transaction(root);
ff5714cc 6893
3612b495
TI		 6894 if (IS_ERR(trans))
6895 return ERR_CAST(trans);
179e29e4
CM		 6896 goto again;
6897 }
c8b97818 6898 map = kmap(page);
70dec807 6899 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 6900 copy_size);
c8b97818 6901 kunmap(page);
179e29e4 6902 btrfs_mark_buffer_dirty(leaf);
a52d9a80 6903 }
d1310b2e 6904 set_extent_uptodate(io_tree, em->start,
507903b8 6905 extent_map_end(em) - 1, NULL, GFP_NOFS);
a52d9a80 6906 goto insert;
a52d9a80
CM		 6907 }
6908not_found:
6909 em->start = start;
70c8a91c 6910 em->orig_start = start;
d1310b2e 6911 em->len = len;
a52d9a80 6912not_found_em:
5f39d397 6913 em->block_start = EXTENT_MAP_HOLE;
9036c102 6914 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80 6915insert:
b3b4aa74 6916 btrfs_release_path(path);
d1310b2e 6917 if (em->start > start || extent_map_end(em) <= start) {
c2cf52eb 6918 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
c1c9ff7c 6919 em->start, em->len, start, len);
a52d9a80
CM		 6920 err = -EIO;
6921 goto out;
6922 }
d1310b2e
CM		 6923
6924 err = 0;
890871be 6925 write_lock(&em_tree->lock);
09a2a8f9 6926 ret = add_extent_mapping(em_tree, em, 0);
3b951516
CM		 6927 /* it is possible that someone inserted the extent into the tree
6928 * while we had the lock dropped. It is also possible that
6929 * an overlapping map exists in the tree
6930 */
a52d9a80 6931 if (ret == -EEXIST) {
3b951516 6932 struct extent_map *existing;
e6dcd2dc
CM		 6933
6934 ret = 0;
6935
e6c4efd8
QW		 6936 existing = search_extent_mapping(em_tree, start, len);
6937 /*
6938 * existing will always be non-NULL, since there must be
6939 * extent causing the -EEXIST.
6940 */
6941 if (start >= extent_map_end(existing) ||
32be3a1a 6942 start <= existing->start) {
e6c4efd8
QW		 6943 /*
6944 * The existing extent map is the one nearest to
6945 * the [start, start + len) range which overlaps
6946 */
6947 err = merge_extent_mapping(em_tree, existing,
6948 em, start);
e1c4b745 6949 free_extent_map(existing);
e6c4efd8 6950 if (err) {
3b951516
CM		 6951 free_extent_map(em);
6952 em = NULL;
6953 }
6954 } else {
6955 free_extent_map(em);
6956 em = existing;
e6dcd2dc 6957 err = 0;
a52d9a80 6958 }
a52d9a80 6959 }
890871be 6960 write_unlock(&em_tree->lock);
a52d9a80 6961out:
1abe9b8a 6962
4cd8587c 6963 trace_btrfs_get_extent(root, em);
1abe9b8a 6964
527afb44 6965 btrfs_free_path(path);
a52d9a80
CM		 6966 if (trans) {
6967 ret = btrfs_end_transaction(trans, root);
d397712b 6968 if (!err)
a52d9a80
CM		 6969 err = ret;
6970 }
a52d9a80
CM		 6971 if (err) {
6972 free_extent_map(em);
a52d9a80
CM		 6973 return ERR_PTR(err);
6974 }
79787eaa 6975 BUG_ON(!em); /* Error is always set */
a52d9a80
CM		 6976 return em;
6977}
6978
ec29ed5b
CM		 6979struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6980 size_t pg_offset, u64 start, u64 len,
6981 int create)
6982{
6983 struct extent_map *em;
6984 struct extent_map *hole_em = NULL;
6985 u64 range_start = start;
6986 u64 end;
6987 u64 found;
6988 u64 found_end;
6989 int err = 0;
6990
6991 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6992 if (IS_ERR(em))
6993 return em;
6994 if (em) {
6995 /*
f9e4fb53
LB		 6996 * if our em maps to
6997 * - a hole or
6998 * - a pre-alloc extent,
6999 * there might actually be delalloc bytes behind it.
ec29ed5b 7000 */
f9e4fb53
LB		 7001 if (em->block_start != EXTENT_MAP_HOLE &&
7002 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
ec29ed5b
CM		 7003 return em;
7004 else
7005 hole_em = em;
7006 }
7007
7008 /* check to see if we've wrapped (len == -1 or similar) */
7009 end = start + len;
7010 if (end < start)
7011 end = (u64)-1;
7012 else
7013 end -= 1;
7014
7015 em = NULL;
7016
7017 /* ok, we didn't find anything, lets look for delalloc */
7018 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
7019 end, len, EXTENT_DELALLOC, 1);
7020 found_end = range_start + found;
7021 if (found_end < range_start)
7022 found_end = (u64)-1;
7023
7024 /*
7025 * we didn't find anything useful, return
7026 * the original results from get_extent()
7027 */
7028 if (range_start > end || found_end <= start) {
7029 em = hole_em;
7030 hole_em = NULL;
7031 goto out;
7032 }
7033
7034 /* adjust the range_start to make sure it doesn't
7035 * go backwards from the start they passed in
7036 */
67871254 7037 range_start = max(start, range_start);
ec29ed5b
CM		 7038 found = found_end - range_start;
7039
7040 if (found > 0) {
7041 u64 hole_start = start;
7042 u64 hole_len = len;
7043
172ddd60 7044 em = alloc_extent_map();
ec29ed5b
CM		 7045 if (!em) {
7046 err = -ENOMEM;
7047 goto out;
7048 }
7049 /*
7050 * when btrfs_get_extent can't find anything it
7051 * returns one huge hole
7052 *
7053 * make sure what it found really fits our range, and
7054 * adjust to make sure it is based on the start from
7055 * the caller
7056 */
7057 if (hole_em) {
7058 u64 calc_end = extent_map_end(hole_em);
7059
7060 if (calc_end <= start || (hole_em->start > end)) {
7061 free_extent_map(hole_em);
7062 hole_em = NULL;
7063 } else {
7064 hole_start = max(hole_em->start, start);
7065 hole_len = calc_end - hole_start;
7066 }
7067 }
7068 em->bdev = NULL;
7069 if (hole_em && range_start > hole_start) {
7070 /* our hole starts before our delalloc, so we
7071 * have to return just the parts of the hole
7072 * that go until the delalloc starts
7073 */
7074 em->len = min(hole_len,
7075 range_start - hole_start);
7076 em->start = hole_start;
7077 em->orig_start = hole_start;
7078 /*
7079 * don't adjust block start at all,
7080 * it is fixed at EXTENT_MAP_HOLE
7081 */
7082 em->block_start = hole_em->block_start;
7083 em->block_len = hole_len;
f9e4fb53
LB		 7084 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7085 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
ec29ed5b
CM		 7086 } else {
7087 em->start = range_start;
7088 em->len = found;
7089 em->orig_start = range_start;
7090 em->block_start = EXTENT_MAP_DELALLOC;
7091 em->block_len = found;
7092 }
7093 } else if (hole_em) {
7094 return hole_em;
7095 }
7096out:
7097
7098 free_extent_map(hole_em);
7099 if (err) {
7100 free_extent_map(em);
7101 return ERR_PTR(err);
7102 }
7103 return em;
7104}
7105
4b46fce2
JB		 7106static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7107 u64 start, u64 len)
7108{
7109 struct btrfs_root *root = BTRFS_I(inode)->root;
70c8a91c 7110 struct extent_map *em;
4b46fce2
JB		 7111 struct btrfs_key ins;
7112 u64 alloc_hint;
7113 int ret;
4b46fce2 7114
4b46fce2 7115 alloc_hint = get_extent_allocation_hint(inode, start, len);
00361589 7116 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
e570fd27 7117 alloc_hint, &ins, 1, 1);
00361589
JB		 7118 if (ret)
7119 return ERR_PTR(ret);
4b46fce2 7120
de0ee0ed
FM		 7121 /*
7122 * Create the ordered extent before the extent map. This is to avoid
7123 * races with the fast fsync path that would lead to it logging file
7124 * extent items that point to disk extents that were not yet written to.
7125 * The fast fsync path collects ordered extents into a local list and
7126 * then collects all the new extent maps, so we must create the ordered
7127 * extent first and make sure the fast fsync path collects any new
7128 * ordered extents after collecting new extent maps as well.
7129 * The fsync path simply can not rely on inode_dio_wait() because it
7130 * causes deadlock with AIO.
7131 */
4b46fce2
JB		 7132 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
7133 ins.offset, ins.offset, 0);
7134 if (ret) {
e570fd27 7135 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
00361589 7136 return ERR_PTR(ret);
4b46fce2 7137 }
00361589 7138
de0ee0ed
FM		 7139 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
7140 ins.offset, ins.offset, ins.offset, 0);
7141 if (IS_ERR(em)) {
7142 struct btrfs_ordered_extent *oe;
7143
7144 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7145 oe = btrfs_lookup_ordered_extent(inode, start);
7146 ASSERT(oe);
7147 if (WARN_ON(!oe))
7148 return em;
7149 set_bit(BTRFS_ORDERED_IOERR, &oe->flags);
7150 set_bit(BTRFS_ORDERED_IO_DONE, &oe->flags);
7151 btrfs_remove_ordered_extent(inode, oe);
7152 /* Once for our lookup and once for the ordered extents tree. */
7153 btrfs_put_ordered_extent(oe);
7154 btrfs_put_ordered_extent(oe);
7155 }
4b46fce2
JB		 7156 return em;
7157}
7158
46bfbb5c
CM		 7159/*
7160 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7161 * block must be cow'd
7162 */
00361589 7163noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7ee9e440
JB		 7164 u64 *orig_start, u64 *orig_block_len,
7165 u64 *ram_bytes)
46bfbb5c 7166{
00361589 7167 struct btrfs_trans_handle *trans;
46bfbb5c
CM		 7168 struct btrfs_path *path;
7169 int ret;
7170 struct extent_buffer *leaf;
7171 struct btrfs_root *root = BTRFS_I(inode)->root;
7b2b7085 7172 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
46bfbb5c
CM		 7173 struct btrfs_file_extent_item *fi;
7174 struct btrfs_key key;
7175 u64 disk_bytenr;
7176 u64 backref_offset;
7177 u64 extent_end;
7178 u64 num_bytes;
7179 int slot;
7180 int found_type;
7ee9e440 7181 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
e77751aa 7182
46bfbb5c
CM		 7183 path = btrfs_alloc_path();
7184 if (!path)
7185 return -ENOMEM;
7186
00361589 7187 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
46bfbb5c
CM		 7188 offset, 0);
7189 if (ret < 0)
7190 goto out;
7191
7192 slot = path->slots[0];
7193 if (ret == 1) {
7194 if (slot == 0) {
7195 /* can't find the item, must cow */
7196 ret = 0;
7197 goto out;
7198 }
7199 slot--;
7200 }
7201 ret = 0;
7202 leaf = path->nodes[0];
7203 btrfs_item_key_to_cpu(leaf, &key, slot);
33345d01 7204 if (key.objectid != btrfs_ino(inode) ||
46bfbb5c
CM		 7205 key.type != BTRFS_EXTENT_DATA_KEY) {
7206 /* not our file or wrong item type, must cow */
7207 goto out;
7208 }
7209
7210 if (key.offset > offset) {
7211 /* Wrong offset, must cow */
7212 goto out;
7213 }
7214
7215 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7216 found_type = btrfs_file_extent_type(leaf, fi);
7217 if (found_type != BTRFS_FILE_EXTENT_REG &&
7218 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7219 /* not a regular extent, must cow */
7220 goto out;
7221 }
7ee9e440
JB		 7222
7223 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7224 goto out;
7225
e77751aa
MX		 7226 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7227 if (extent_end <= offset)
7228 goto out;
7229
46bfbb5c 7230 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7ee9e440
JB		 7231 if (disk_bytenr == 0)
7232 goto out;
7233
7234 if (btrfs_file_extent_compression(leaf, fi) ||
7235 btrfs_file_extent_encryption(leaf, fi) ||
7236 btrfs_file_extent_other_encoding(leaf, fi))
7237 goto out;
7238
46bfbb5c
CM		 7239 backref_offset = btrfs_file_extent_offset(leaf, fi);
7240
7ee9e440
JB		 7241 if (orig_start) {
7242 *orig_start = key.offset - backref_offset;
7243 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7244 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7245 }
eb384b55 7246
46bfbb5c
CM		 7247 if (btrfs_extent_readonly(root, disk_bytenr))
7248 goto out;
7b2b7085
MX		 7249
7250 num_bytes = min(offset + *len, extent_end) - offset;
7251 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7252 u64 range_end;
7253
7254 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7255 ret = test_range_bit(io_tree, offset, range_end,
7256 EXTENT_DELALLOC, 0, NULL);
7257 if (ret) {
7258 ret = -EAGAIN;
7259 goto out;
7260 }
7261 }
7262
1bda19eb 7263 btrfs_release_path(path);
46bfbb5c
CM		 7264
7265 /*
7266 * look for other files referencing this extent, if we
7267 * find any we must cow
7268 */
00361589
JB		 7269 trans = btrfs_join_transaction(root);
7270 if (IS_ERR(trans)) {
7271 ret = 0;
46bfbb5c 7272 goto out;
00361589
JB		 7273 }
7274
7275 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7276 key.offset - backref_offset, disk_bytenr);
7277 btrfs_end_transaction(trans, root);
7278 if (ret) {
7279 ret = 0;
7280 goto out;
7281 }
46bfbb5c
CM		 7282
7283 /*
7284 * adjust disk_bytenr and num_bytes to cover just the bytes
7285 * in this extent we are about to write. If there
7286 * are any csums in that range we have to cow in order
7287 * to keep the csums correct
7288 */
7289 disk_bytenr += backref_offset;
7290 disk_bytenr += offset - key.offset;
46bfbb5c
CM		 7291 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7292 goto out;
7293 /*
7294 * all of the above have passed, it is safe to overwrite this extent
7295 * without cow
7296 */
eb384b55 7297 *len = num_bytes;
46bfbb5c
CM		 7298 ret = 1;
7299out:
7300 btrfs_free_path(path);
7301 return ret;
7302}
7303
fc4adbff
AG		 7304bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7305{
7306 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7307 int found = false;
7308 void **pagep = NULL;
7309 struct page *page = NULL;
7310 int start_idx;
7311 int end_idx;
7312
7313 start_idx = start >> PAGE_CACHE_SHIFT;
7314
7315 /*
7316 * end is the last byte in the last page. end == start is legal
7317 */
7318 end_idx = end >> PAGE_CACHE_SHIFT;
7319
7320 rcu_read_lock();
7321
7322 /* Most of the code in this while loop is lifted from
7323 * find_get_page. It's been modified to begin searching from a
7324 * page and return just the first page found in that range. If the
7325 * found idx is less than or equal to the end idx then we know that
7326 * a page exists. If no pages are found or if those pages are
7327 * outside of the range then we're fine (yay!) */
7328 while (page == NULL &&
7329 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7330 page = radix_tree_deref_slot(pagep);
7331 if (unlikely(!page))
7332 break;
7333
7334 if (radix_tree_exception(page)) {
809f9016
FM		 7335 if (radix_tree_deref_retry(page)) {
7336 page = NULL;
fc4adbff 7337 continue;
809f9016 7338 }
fc4adbff
AG		 7339 /*
7340 * Otherwise, shmem/tmpfs must be storing a swap entry
7341 * here as an exceptional entry: so return it without
7342 * attempting to raise page count.
7343 */
6fdef6d4 7344 page = NULL;
fc4adbff
AG		 7345 break; /* TODO: Is this relevant for this use case? */
7346 }
7347
91405151
FM		 7348 if (!page_cache_get_speculative(page)) {
7349 page = NULL;
fc4adbff 7350 continue;
91405151 7351 }
fc4adbff
AG		 7352
7353 /*
7354 * Has the page moved?
7355 * This is part of the lockless pagecache protocol. See
7356 * include/linux/pagemap.h for details.
7357 */
7358 if (unlikely(page != *pagep)) {
7359 page_cache_release(page);
7360 page = NULL;
7361 }
7362 }
7363
7364 if (page) {
7365 if (page->index <= end_idx)
7366 found = true;
7367 page_cache_release(page);
7368 }
7369
7370 rcu_read_unlock();
7371 return found;
7372}
7373
eb838e73
JB		 7374static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7375 struct extent_state **cached_state, int writing)
7376{
7377 struct btrfs_ordered_extent *ordered;
7378 int ret = 0;
7379
7380 while (1) {
7381 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
ff13db41 7382 cached_state);
eb838e73
JB		 7383 /*
7384 * We're concerned with the entire range that we're going to be
7385 * doing DIO to, so we need to make sure theres no ordered
7386 * extents in this range.
7387 */
7388 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7389 lockend - lockstart + 1);
7390
7391 /*
7392 * We need to make sure there are no buffered pages in this
7393 * range either, we could have raced between the invalidate in
7394 * generic_file_direct_write and locking the extent. The
7395 * invalidate needs to happen so that reads after a write do not
7396 * get stale data.
7397 */
fc4adbff
AG		 7398 if (!ordered &&
7399 (!writing ||
7400 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
eb838e73
JB		 7401 break;
7402
7403 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7404 cached_state, GFP_NOFS);
7405
7406 if (ordered) {
7407 btrfs_start_ordered_extent(inode, ordered, 1);
7408 btrfs_put_ordered_extent(ordered);
7409 } else {
eb838e73 7410 /*
b850ae14
FM		 7411 * We could trigger writeback for this range (and wait
7412 * for it to complete) and then invalidate the pages for
7413 * this range (through invalidate_inode_pages2_range()),
7414 * but that can lead us to a deadlock with a concurrent
7415 * call to readpages() (a buffered read or a defrag call
7416 * triggered a readahead) on a page lock due to an
7417 * ordered dio extent we created before but did not have
7418 * yet a corresponding bio submitted (whence it can not
7419 * complete), which makes readpages() wait for that
7420 * ordered extent to complete while holding a lock on
7421 * that page.
eb838e73 7422 */
b850ae14
FM		 7423 ret = -ENOTBLK;
7424 break;
eb838e73
JB		 7425 }
7426
7427 cond_resched();
7428 }
7429
7430 return ret;
7431}
7432
69ffb543
JB		 7433static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7434 u64 len, u64 orig_start,
7435 u64 block_start, u64 block_len,
cc95bef6
JB		 7436 u64 orig_block_len, u64 ram_bytes,
7437 int type)
69ffb543
JB		 7438{
7439 struct extent_map_tree *em_tree;
7440 struct extent_map *em;
7441 struct btrfs_root *root = BTRFS_I(inode)->root;
7442 int ret;
7443
7444 em_tree = &BTRFS_I(inode)->extent_tree;
7445 em = alloc_extent_map();
7446 if (!em)
7447 return ERR_PTR(-ENOMEM);
7448
7449 em->start = start;
7450 em->orig_start = orig_start;
2ab28f32
JB		 7451 em->mod_start = start;
7452 em->mod_len = len;
69ffb543
JB		 7453 em->len = len;
7454 em->block_len = block_len;
7455 em->block_start = block_start;
7456 em->bdev = root->fs_info->fs_devices->latest_bdev;
b4939680 7457 em->orig_block_len = orig_block_len;
cc95bef6 7458 em->ram_bytes = ram_bytes;
70c8a91c 7459 em->generation = -1;
69ffb543
JB		 7460 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7461 if (type == BTRFS_ORDERED_PREALLOC)
b11e234d 7462 set_bit(EXTENT_FLAG_FILLING, &em->flags);
69ffb543
JB		 7463
7464 do {
7465 btrfs_drop_extent_cache(inode, em->start,
7466 em->start + em->len - 1, 0);
7467 write_lock(&em_tree->lock);
09a2a8f9 7468 ret = add_extent_mapping(em_tree, em, 1);
69ffb543
JB		 7469 write_unlock(&em_tree->lock);
7470 } while (ret == -EEXIST);
7471
7472 if (ret) {
7473 free_extent_map(em);
7474 return ERR_PTR(ret);
7475 }
7476
7477 return em;
7478}
7479
9c9464cc
FM		 7480static void adjust_dio_outstanding_extents(struct inode *inode,
7481 struct btrfs_dio_data *dio_data,
7482 const u64 len)
7483{
7484 unsigned num_extents;
7485
7486 num_extents = (unsigned) div64_u64(len + BTRFS_MAX_EXTENT_SIZE - 1,
7487 BTRFS_MAX_EXTENT_SIZE);
7488 /*
7489 * If we have an outstanding_extents count still set then we're
7490 * within our reservation, otherwise we need to adjust our inode
7491 * counter appropriately.
7492 */
7493 if (dio_data->outstanding_extents) {
7494 dio_data->outstanding_extents -= num_extents;
7495 } else {
7496 spin_lock(&BTRFS_I(inode)->lock);
7497 BTRFS_I(inode)->outstanding_extents += num_extents;
7498 spin_unlock(&BTRFS_I(inode)->lock);
7499 }
7500}
7501
4b46fce2
JB		 7502static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7503 struct buffer_head *bh_result, int create)
7504{
7505 struct extent_map *em;
7506 struct btrfs_root *root = BTRFS_I(inode)->root;
eb838e73 7507 struct extent_state *cached_state = NULL;
50745b0a 7508 struct btrfs_dio_data *dio_data = NULL;
4b46fce2 7509 u64 start = iblock << inode->i_blkbits;
eb838e73 7510 u64 lockstart, lockend;
4b46fce2 7511 u64 len = bh_result->b_size;
eb838e73 7512 int unlock_bits = EXTENT_LOCKED;
0934856d 7513 int ret = 0;
eb838e73 7514
172a5049 7515 if (create)
3266789f 7516 unlock_bits |= EXTENT_DIRTY;
172a5049 7517 else
c329861d 7518 len = min_t(u64, len, root->sectorsize);
eb838e73 7519
c329861d
JB		 7520 lockstart = start;
7521 lockend = start + len - 1;
7522
e1cbbfa5
JB		 7523 if (current->journal_info) {
7524 /*
7525 * Need to pull our outstanding extents and set journal_info to NULL so
7526 * that anything that needs to check if there's a transction doesn't get
7527 * confused.
7528 */
50745b0a 7529 dio_data = current->journal_info;
e1cbbfa5
JB		 7530 current->journal_info = NULL;
7531 }
7532
eb838e73
JB		 7533 /*
7534 * If this errors out it's because we couldn't invalidate pagecache for
7535 * this range and we need to fallback to buffered.
7536 */
9c9464cc
FM		 7537 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7538 create)) {
7539 ret = -ENOTBLK;
7540 goto err;
7541 }
eb838e73 7542
4b46fce2 7543 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
eb838e73
JB		 7544 if (IS_ERR(em)) {
7545 ret = PTR_ERR(em);
7546 goto unlock_err;
7547 }
4b46fce2
JB		 7548
7549 /*
7550 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7551 * io. INLINE is special, and we could probably kludge it in here, but
7552 * it's still buffered so for safety lets just fall back to the generic
7553 * buffered path.
7554 *
7555 * For COMPRESSED we _have_ to read the entire extent in so we can
7556 * decompress it, so there will be buffering required no matter what we
7557 * do, so go ahead and fallback to buffered.
7558 *
7559 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7560 * to buffered IO. Don't blame me, this is the price we pay for using
7561 * the generic code.
7562 */
7563 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7564 em->block_start == EXTENT_MAP_INLINE) {
7565 free_extent_map(em);
eb838e73
JB		 7566 ret = -ENOTBLK;
7567 goto unlock_err;
4b46fce2
JB		 7568 }
7569
7570 /* Just a good old fashioned hole, return */
7571 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7572 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7573 free_extent_map(em);
eb838e73 7574 goto unlock_err;
4b46fce2
JB		 7575 }
7576
7577 /*
7578 * We don't allocate a new extent in the following cases
7579 *
7580 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7581 * existing extent.
7582 * 2) The extent is marked as PREALLOC. We're good to go here and can
7583 * just use the extent.
7584 *
7585 */
46bfbb5c 7586 if (!create) {
eb838e73
JB		 7587 len = min(len, em->len - (start - em->start));
7588 lockstart = start + len;
7589 goto unlock;
46bfbb5c 7590 }
4b46fce2
JB		 7591
7592 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7593 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7594 em->block_start != EXTENT_MAP_HOLE)) {
4b46fce2 7595 int type;
eb384b55 7596 u64 block_start, orig_start, orig_block_len, ram_bytes;
4b46fce2
JB		 7597
7598 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7599 type = BTRFS_ORDERED_PREALLOC;
7600 else
7601 type = BTRFS_ORDERED_NOCOW;
46bfbb5c 7602 len = min(len, em->len - (start - em->start));
4b46fce2 7603 block_start = em->block_start + (start - em->start);
46bfbb5c 7604
00361589 7605 if (can_nocow_extent(inode, start, &len, &orig_start,
7ee9e440 7606 &orig_block_len, &ram_bytes) == 1) {
69ffb543
JB		 7607 if (type == BTRFS_ORDERED_PREALLOC) {
7608 free_extent_map(em);
7609 em = create_pinned_em(inode, start, len,
7610 orig_start,
b4939680 7611 block_start, len,
cc95bef6
JB		 7612 orig_block_len,
7613 ram_bytes, type);
555e1286
FM		 7614 if (IS_ERR(em)) {
7615 ret = PTR_ERR(em);
69ffb543 7616 goto unlock_err;
555e1286 7617 }
69ffb543
JB		 7618 }
7619
46bfbb5c
CM		 7620 ret = btrfs_add_ordered_extent_dio(inode, start,
7621 block_start, len, len, type);
46bfbb5c
CM		 7622 if (ret) {
7623 free_extent_map(em);
eb838e73 7624 goto unlock_err;
46bfbb5c
CM		 7625 }
7626 goto unlock;
4b46fce2 7627 }
4b46fce2 7628 }
00361589 7629
46bfbb5c
CM		 7630 /*
7631 * this will cow the extent, reset the len in case we changed
7632 * it above
7633 */
7634 len = bh_result->b_size;
70c8a91c
JB		 7635 free_extent_map(em);
7636 em = btrfs_new_extent_direct(inode, start, len);
eb838e73
JB		 7637 if (IS_ERR(em)) {
7638 ret = PTR_ERR(em);
7639 goto unlock_err;
7640 }
46bfbb5c
CM		 7641 len = min(len, em->len - (start - em->start));
7642unlock:
4b46fce2
JB		 7643 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7644 inode->i_blkbits;
46bfbb5c 7645 bh_result->b_size = len;
4b46fce2
JB		 7646 bh_result->b_bdev = em->bdev;
7647 set_buffer_mapped(bh_result);
c3473e83
JB		 7648 if (create) {
7649 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7650 set_buffer_new(bh_result);
7651
7652 /*
7653 * Need to update the i_size under the extent lock so buffered
7654 * readers will get the updated i_size when we unlock.
7655 */
7656 if (start + len > i_size_read(inode))
7657 i_size_write(inode, start + len);
0934856d 7658
9c9464cc 7659 adjust_dio_outstanding_extents(inode, dio_data, len);
7cf5b976 7660 btrfs_free_reserved_data_space(inode, start, len);
50745b0a 7661 WARN_ON(dio_data->reserve < len);
7662 dio_data->reserve -= len;
f28a4928 7663 dio_data->unsubmitted_oe_range_end = start + len;
50745b0a 7664 current->journal_info = dio_data;
c3473e83 7665 }
4b46fce2 7666
eb838e73
JB		 7667 /*
7668 * In the case of write we need to clear and unlock the entire range,
7669 * in the case of read we need to unlock only the end area that we
7670 * aren't using if there is any left over space.
7671 */
24c03fa5 7672 if (lockstart < lockend) {
0934856d
MX		 7673 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7674 lockend, unlock_bits, 1, 0,
7675 &cached_state, GFP_NOFS);
24c03fa5 7676 } else {
eb838e73 7677 free_extent_state(cached_state);
24c03fa5 7678 }
eb838e73 7679
4b46fce2
JB		 7680 free_extent_map(em);
7681
7682 return 0;
eb838e73
JB		 7683
7684unlock_err:
eb838e73
JB		 7685 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7686 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
9c9464cc 7687err:
50745b0a 7688 if (dio_data)
7689 current->journal_info = dio_data;
9c9464cc
FM		 7690 /*
7691 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7692 * write less data then expected, so that we don't underflow our inode's
7693 * outstanding extents counter.
7694 */
7695 if (create && dio_data)
7696 adjust_dio_outstanding_extents(inode, dio_data, len);
7697
eb838e73 7698 return ret;
4b46fce2
JB		 7699}
7700
8b110e39
MX		 7701static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7702 int rw, int mirror_num)
7703{
7704 struct btrfs_root *root = BTRFS_I(inode)->root;
7705 int ret;
7706
7707 BUG_ON(rw & REQ_WRITE);
7708
7709 bio_get(bio);
7710
7711 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7712 BTRFS_WQ_ENDIO_DIO_REPAIR);
7713 if (ret)
7714 goto err;
7715
7716 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7717err:
7718 bio_put(bio);
7719 return ret;
7720}
7721
7722static int btrfs_check_dio_repairable(struct inode *inode,
7723 struct bio *failed_bio,
7724 struct io_failure_record *failrec,
7725 int failed_mirror)
7726{
7727 int num_copies;
7728
7729 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7730 failrec->logical, failrec->len);
7731 if (num_copies == 1) {
7732 /*
7733 * we only have a single copy of the data, so don't bother with
7734 * all the retry and error correction code that follows. no
7735 * matter what the error is, it is very likely to persist.
7736 */
7737 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7738 num_copies, failrec->this_mirror, failed_mirror);
7739 return 0;
7740 }
7741
7742 failrec->failed_mirror = failed_mirror;
7743 failrec->this_mirror++;
7744 if (failrec->this_mirror == failed_mirror)
7745 failrec->this_mirror++;
7746
7747 if (failrec->this_mirror > num_copies) {
7748 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7749 num_copies, failrec->this_mirror, failed_mirror);
7750 return 0;
7751 }
7752
7753 return 1;
7754}
7755
7756static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7757 struct page *page, u64 start, u64 end,
7758 int failed_mirror, bio_end_io_t *repair_endio,
7759 void *repair_arg)
7760{
7761 struct io_failure_record *failrec;
7762 struct bio *bio;
7763 int isector;
7764 int read_mode;
7765 int ret;
7766
7767 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7768
7769 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7770 if (ret)
7771 return ret;
7772
7773 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7774 failed_mirror);
7775 if (!ret) {
7776 free_io_failure(inode, failrec);
7777 return -EIO;
7778 }
7779
7780 if (failed_bio->bi_vcnt > 1)
7781 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7782 else
7783 read_mode = READ_SYNC;
7784
7785 isector = start - btrfs_io_bio(failed_bio)->logical;
7786 isector >>= inode->i_sb->s_blocksize_bits;
7787 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7788 0, isector, repair_endio, repair_arg);
7789 if (!bio) {
7790 free_io_failure(inode, failrec);
7791 return -EIO;
7792 }
7793
7794 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7795 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7796 read_mode, failrec->this_mirror, failrec->in_validation);
7797
7798 ret = submit_dio_repair_bio(inode, bio, read_mode,
7799 failrec->this_mirror);
7800 if (ret) {
7801 free_io_failure(inode, failrec);
7802 bio_put(bio);
7803 }
7804
7805 return ret;
7806}
7807
7808struct btrfs_retry_complete {
7809 struct completion done;
7810 struct inode *inode;
7811 u64 start;
7812 int uptodate;
7813};
7814
4246a0b6 7815static void btrfs_retry_endio_nocsum(struct bio *bio)
8b110e39
MX		 7816{
7817 struct btrfs_retry_complete *done = bio->bi_private;
7818 struct bio_vec *bvec;
7819 int i;
7820
4246a0b6 7821 if (bio->bi_error)
8b110e39
MX		 7822 goto end;
7823
7824 done->uptodate = 1;
7825 bio_for_each_segment_all(bvec, bio, i)
7826 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7827end:
7828 complete(&done->done);
7829 bio_put(bio);
7830}
7831
7832static int __btrfs_correct_data_nocsum(struct inode *inode,
7833 struct btrfs_io_bio *io_bio)
4b46fce2 7834{
2c30c71b 7835 struct bio_vec *bvec;
8b110e39 7836 struct btrfs_retry_complete done;
4b46fce2 7837 u64 start;
2c30c71b 7838 int i;
c1dc0896 7839 int ret;
4b46fce2 7840
8b110e39
MX		 7841 start = io_bio->logical;
7842 done.inode = inode;
7843
7844 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7845try_again:
7846 done.uptodate = 0;
7847 done.start = start;
7848 init_completion(&done.done);
7849
7850 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7851 start + bvec->bv_len - 1,
7852 io_bio->mirror_num,
7853 btrfs_retry_endio_nocsum, &done);
7854 if (ret)
7855 return ret;
7856
7857 wait_for_completion(&done.done);
7858
7859 if (!done.uptodate) {
7860 /* We might have another mirror, so try again */
7861 goto try_again;
7862 }
7863
7864 start += bvec->bv_len;
7865 }
7866
7867 return 0;
7868}
7869
4246a0b6 7870static void btrfs_retry_endio(struct bio *bio)
8b110e39
MX		 7871{
7872 struct btrfs_retry_complete *done = bio->bi_private;
7873 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7874 struct bio_vec *bvec;
7875 int uptodate;
7876 int ret;
7877 int i;
7878
4246a0b6 7879 if (bio->bi_error)
8b110e39
MX		 7880 goto end;
7881
7882 uptodate = 1;
7883 bio_for_each_segment_all(bvec, bio, i) {
7884 ret = __readpage_endio_check(done->inode, io_bio, i,
7885 bvec->bv_page, 0,
7886 done->start, bvec->bv_len);
7887 if (!ret)
7888 clean_io_failure(done->inode, done->start,
7889 bvec->bv_page, 0);
7890 else
7891 uptodate = 0;
7892 }
7893
7894 done->uptodate = uptodate;
7895end:
7896 complete(&done->done);
7897 bio_put(bio);
7898}
7899
7900static int __btrfs_subio_endio_read(struct inode *inode,
7901 struct btrfs_io_bio *io_bio, int err)
7902{
7903 struct bio_vec *bvec;
7904 struct btrfs_retry_complete done;
7905 u64 start;
7906 u64 offset = 0;
7907 int i;
7908 int ret;
dc380aea 7909
8b110e39 7910 err = 0;
c1dc0896 7911 start = io_bio->logical;
8b110e39
MX		 7912 done.inode = inode;
7913
c1dc0896 7914 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
dc380aea
MX		 7915 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7916 0, start, bvec->bv_len);
8b110e39
MX		 7917 if (likely(!ret))
7918 goto next;
7919try_again:
7920 done.uptodate = 0;
7921 done.start = start;
7922 init_completion(&done.done);
7923
7924 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7925 start + bvec->bv_len - 1,
7926 io_bio->mirror_num,
7927 btrfs_retry_endio, &done);
7928 if (ret) {
7929 err = ret;
7930 goto next;
7931 }
7932
7933 wait_for_completion(&done.done);
7934
7935 if (!done.uptodate) {
7936 /* We might have another mirror, so try again */
7937 goto try_again;
7938 }
7939next:
7940 offset += bvec->bv_len;
4b46fce2 7941 start += bvec->bv_len;
2c30c71b 7942 }
c1dc0896
MX		 7943
7944 return err;
7945}
7946
8b110e39
MX		 7947static int btrfs_subio_endio_read(struct inode *inode,
7948 struct btrfs_io_bio *io_bio, int err)
7949{
7950 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7951
7952 if (skip_csum) {
7953 if (unlikely(err))
7954 return __btrfs_correct_data_nocsum(inode, io_bio);
7955 else
7956 return 0;
7957 } else {
7958 return __btrfs_subio_endio_read(inode, io_bio, err);
7959 }
7960}
7961
4246a0b6 7962static void btrfs_endio_direct_read(struct bio *bio)
c1dc0896
MX		 7963{
7964 struct btrfs_dio_private *dip = bio->bi_private;
7965 struct inode *inode = dip->inode;
7966 struct bio *dio_bio;
7967 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
4246a0b6 7968 int err = bio->bi_error;
c1dc0896 7969
8b110e39
MX		 7970 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
7971 err = btrfs_subio_endio_read(inode, io_bio, err);
c1dc0896 7972
4b46fce2 7973 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
d0082371 7974 dip->logical_offset + dip->bytes - 1);
9be3395b 7975 dio_bio = dip->dio_bio;
4b46fce2 7976
4b46fce2 7977 kfree(dip);
c0da7aa1 7978
4246a0b6 7979 dio_end_io(dio_bio, bio->bi_error);
23ea8e5a
MX		 7980
7981 if (io_bio->end_io)
7982 io_bio->end_io(io_bio, err);
9be3395b 7983 bio_put(bio);
4b46fce2
JB		 7984}
7985
14543774
FM		 7986static void btrfs_endio_direct_write_update_ordered(struct inode *inode,
7987 const u64 offset,
7988 const u64 bytes,
7989 const int uptodate)
4b46fce2 7990{
4b46fce2 7991 struct btrfs_root *root = BTRFS_I(inode)->root;
4b46fce2 7992 struct btrfs_ordered_extent *ordered = NULL;
14543774
FM		 7993 u64 ordered_offset = offset;
7994 u64 ordered_bytes = bytes;
4b46fce2
JB		 7995 int ret;
7996
163cf09c
CM		 7997again:
7998 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7999 &ordered_offset,
4246a0b6 8000 ordered_bytes,
14543774 8001 uptodate);
4b46fce2 8002 if (!ret)
163cf09c 8003 goto out_test;
4b46fce2 8004
9e0af237
LB		 8005 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8006 finish_ordered_fn, NULL, NULL);
fccb5d86
QW		 8007 btrfs_queue_work(root->fs_info->endio_write_workers,
8008 &ordered->work);
163cf09c
CM		 8009out_test:
8010 /*
8011 * our bio might span multiple ordered extents. If we haven't
8012 * completed the accounting for the whole dio, go back and try again
8013 */
14543774
FM		 8014 if (ordered_offset < offset + bytes) {
8015 ordered_bytes = offset + bytes - ordered_offset;
5fd02043 8016 ordered = NULL;
163cf09c
CM		 8017 goto again;
8018 }
14543774
FM		 8019}
8020
8021static void btrfs_endio_direct_write(struct bio *bio)
8022{
8023 struct btrfs_dio_private *dip = bio->bi_private;
8024 struct bio *dio_bio = dip->dio_bio;
8025
8026 btrfs_endio_direct_write_update_ordered(dip->inode,
8027 dip->logical_offset,
8028 dip->bytes,
8029 !bio->bi_error);
4b46fce2 8030
4b46fce2 8031 kfree(dip);
c0da7aa1 8032
4246a0b6 8033 dio_end_io(dio_bio, bio->bi_error);
9be3395b 8034 bio_put(bio);
4b46fce2
JB		 8035}
8036
eaf25d93
CM		 8037static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
8038 struct bio *bio, int mirror_num,
8039 unsigned long bio_flags, u64 offset)
8040{
8041 int ret;
8042 struct btrfs_root *root = BTRFS_I(inode)->root;
8043 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
79787eaa 8044 BUG_ON(ret); /* -ENOMEM */
eaf25d93
CM		 8045 return 0;
8046}
8047
4246a0b6 8048static void btrfs_end_dio_bio(struct bio *bio)
e65e1535
MX		 8049{
8050 struct btrfs_dio_private *dip = bio->bi_private;
4246a0b6 8051 int err = bio->bi_error;
e65e1535 8052
8b110e39
MX		 8053 if (err)
8054 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8055 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
8056 btrfs_ino(dip->inode), bio->bi_rw,
8057 (unsigned long long)bio->bi_iter.bi_sector,
8058 bio->bi_iter.bi_size, err);
8059
8060 if (dip->subio_endio)
8061 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
c1dc0896
MX		 8062
8063 if (err) {
e65e1535
MX		 8064 dip->errors = 1;
8065
8066 /*
8067 * before atomic variable goto zero, we must make sure
8068 * dip->errors is perceived to be set.
8069 */
4e857c58 8070 smp_mb__before_atomic();
e65e1535
MX		 8071 }
8072
8073 /* if there are more bios still pending for this dio, just exit */
8074 if (!atomic_dec_and_test(&dip->pending_bios))
8075 goto out;
8076
9be3395b 8077 if (dip->errors) {
e65e1535 8078 bio_io_error(dip->orig_bio);
9be3395b 8079 } else {
4246a0b6
CH		 8080 dip->dio_bio->bi_error = 0;
8081 bio_endio(dip->orig_bio);
e65e1535
MX		 8082 }
8083out:
8084 bio_put(bio);
8085}
8086
8087static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8088 u64 first_sector, gfp_t gfp_flags)
8089{
da2f0f74 8090 struct bio *bio;
22365979 8091 bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
da2f0f74
CM		 8092 if (bio)
8093 bio_associate_current(bio);
8094 return bio;
e65e1535
MX		 8095}
8096
c1dc0896
MX		 8097static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
8098 struct inode *inode,
8099 struct btrfs_dio_private *dip,
8100 struct bio *bio,
8101 u64 file_offset)
8102{
8103 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8104 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8105 int ret;
8106
8107 /*
8108 * We load all the csum data we need when we submit
8109 * the first bio to reduce the csum tree search and
8110 * contention.
8111 */
8112 if (dip->logical_offset == file_offset) {
8113 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
8114 file_offset);
8115 if (ret)
8116 return ret;
8117 }
8118
8119 if (bio == dip->orig_bio)
8120 return 0;
8121
8122 file_offset -= dip->logical_offset;
8123 file_offset >>= inode->i_sb->s_blocksize_bits;
8124 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8125
8126 return 0;
8127}
8128
e65e1535
MX		 8129static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8130 int rw, u64 file_offset, int skip_sum,
c329861d 8131 int async_submit)
e65e1535 8132{
facc8a22 8133 struct btrfs_dio_private *dip = bio->bi_private;
e65e1535
MX		 8134 int write = rw & REQ_WRITE;
8135 struct btrfs_root *root = BTRFS_I(inode)->root;
8136 int ret;
8137
b812ce28
JB		 8138 if (async_submit)
8139 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8140
e65e1535 8141 bio_get(bio);
5fd02043
JB		 8142
8143 if (!write) {
bfebd8b5
DS		 8144 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8145 BTRFS_WQ_ENDIO_DATA);
5fd02043
JB		 8146 if (ret)
8147 goto err;
8148 }
e65e1535 8149
1ae39938
JB		 8150 if (skip_sum)
8151 goto map;
8152
8153 if (write && async_submit) {
e65e1535
MX		 8154 ret = btrfs_wq_submit_bio(root->fs_info,
8155 inode, rw, bio, 0, 0,
8156 file_offset,
8157 __btrfs_submit_bio_start_direct_io,
8158 __btrfs_submit_bio_done);
8159 goto err;
1ae39938
JB		 8160 } else if (write) {
8161 /*
8162 * If we aren't doing async submit, calculate the csum of the
8163 * bio now.
8164 */
8165 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8166 if (ret)
8167 goto err;
23ea8e5a 8168 } else {
c1dc0896
MX		 8169 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8170 file_offset);
c2db1073
TI		 8171 if (ret)
8172 goto err;
8173 }
1ae39938
JB		 8174map:
8175 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
e65e1535
MX		 8176err:
8177 bio_put(bio);
8178 return ret;
8179}
8180
8181static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
8182 int skip_sum)
8183{
8184 struct inode *inode = dip->inode;
8185 struct btrfs_root *root = BTRFS_I(inode)->root;
e65e1535
MX		 8186 struct bio *bio;
8187 struct bio *orig_bio = dip->orig_bio;
8188 struct bio_vec *bvec = orig_bio->bi_io_vec;
4f024f37 8189 u64 start_sector = orig_bio->bi_iter.bi_sector;
e65e1535
MX		 8190 u64 file_offset = dip->logical_offset;
8191 u64 submit_len = 0;
8192 u64 map_length;
8193 int nr_pages = 0;
23ea8e5a 8194 int ret;
1ae39938 8195 int async_submit = 0;
e65e1535 8196
4f024f37 8197 map_length = orig_bio->bi_iter.bi_size;
53b381b3 8198 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
e65e1535 8199 &map_length, NULL, 0);
7a5c3c9b 8200 if (ret)
e65e1535 8201 return -EIO;
facc8a22 8202
4f024f37 8203 if (map_length >= orig_bio->bi_iter.bi_size) {
02f57c7a 8204 bio = orig_bio;
c1dc0896 8205 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
02f57c7a
JB		 8206 goto submit;
8207 }
8208
53b381b3 8209 /* async crcs make it difficult to collect full stripe writes. */
ffe2d203 8210 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
53b381b3
DW		 8211 async_submit = 0;
8212 else
8213 async_submit = 1;
8214
02f57c7a
JB		 8215 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8216 if (!bio)
8217 return -ENOMEM;
7a5c3c9b 8218
02f57c7a
JB		 8219 bio->bi_private = dip;
8220 bio->bi_end_io = btrfs_end_dio_bio;
c1dc0896 8221 btrfs_io_bio(bio)->logical = file_offset;
02f57c7a
JB		 8222 atomic_inc(&dip->pending_bios);
8223
e65e1535 8224 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
ee39b432 8225 if (map_length < submit_len + bvec->bv_len ||
e65e1535 8226 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
ee39b432 8227 bvec->bv_offset) < bvec->bv_len) {
e65e1535
MX		 8228 /*
8229 * inc the count before we submit the bio so
8230 * we know the end IO handler won't happen before
8231 * we inc the count. Otherwise, the dip might get freed
8232 * before we're done setting it up
8233 */
8234 atomic_inc(&dip->pending_bios);
8235 ret = __btrfs_submit_dio_bio(bio, inode, rw,
8236 file_offset, skip_sum,
c329861d 8237 async_submit);
e65e1535
MX		 8238 if (ret) {
8239 bio_put(bio);
8240 atomic_dec(&dip->pending_bios);
8241 goto out_err;
8242 }
8243
e65e1535
MX		 8244 start_sector += submit_len >> 9;
8245 file_offset += submit_len;
8246
8247 submit_len = 0;
8248 nr_pages = 0;
8249
8250 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8251 start_sector, GFP_NOFS);
8252 if (!bio)
8253 goto out_err;
8254 bio->bi_private = dip;
8255 bio->bi_end_io = btrfs_end_dio_bio;
c1dc0896 8256 btrfs_io_bio(bio)->logical = file_offset;
e65e1535 8257
4f024f37 8258 map_length = orig_bio->bi_iter.bi_size;
53b381b3 8259 ret = btrfs_map_block(root->fs_info, rw,
3ec706c8 8260 start_sector << 9,
e65e1535
MX		 8261 &map_length, NULL, 0);
8262 if (ret) {
8263 bio_put(bio);
8264 goto out_err;
8265 }
8266 } else {
8267 submit_len += bvec->bv_len;
67871254 8268 nr_pages++;
e65e1535
MX		 8269 bvec++;
8270 }
8271 }
8272
02f57c7a 8273submit:
e65e1535 8274 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
c329861d 8275 async_submit);
e65e1535
MX		 8276 if (!ret)
8277 return 0;
8278
8279 bio_put(bio);
8280out_err:
8281 dip->errors = 1;
8282 /*
8283 * before atomic variable goto zero, we must
8284 * make sure dip->errors is perceived to be set.
8285 */
4e857c58 8286 smp_mb__before_atomic();
e65e1535
MX		 8287 if (atomic_dec_and_test(&dip->pending_bios))
8288 bio_io_error(dip->orig_bio);
8289
8290 /* bio_end_io() will handle error, so we needn't return it */
8291 return 0;
8292}
8293
9be3395b
CM		 8294static void btrfs_submit_direct(int rw, struct bio *dio_bio,
8295 struct inode *inode, loff_t file_offset)
4b46fce2 8296{
61de718f
FM		 8297 struct btrfs_dio_private *dip = NULL;
8298 struct bio *io_bio = NULL;
23ea8e5a 8299 struct btrfs_io_bio *btrfs_bio;
4b46fce2 8300 int skip_sum;
7b6d91da 8301 int write = rw & REQ_WRITE;
4b46fce2
JB		 8302 int ret = 0;
8303
8304 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8305
9be3395b 8306 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
9be3395b
CM		 8307 if (!io_bio) {
8308 ret = -ENOMEM;
8309 goto free_ordered;
8310 }
8311
c1dc0896 8312 dip = kzalloc(sizeof(*dip), GFP_NOFS);
4b46fce2
JB		 8313 if (!dip) {
8314 ret = -ENOMEM;
61de718f 8315 goto free_ordered;
4b46fce2 8316 }
4b46fce2 8317
9be3395b 8318 dip->private = dio_bio->bi_private;
4b46fce2
JB		 8319 dip->inode = inode;
8320 dip->logical_offset = file_offset;
4f024f37
KO		 8321 dip->bytes = dio_bio->bi_iter.bi_size;
8322 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
9be3395b 8323 io_bio->bi_private = dip;
9be3395b
CM		 8324 dip->orig_bio = io_bio;
8325 dip->dio_bio = dio_bio;
e65e1535 8326 atomic_set(&dip->pending_bios, 0);
c1dc0896
MX		 8327 btrfs_bio = btrfs_io_bio(io_bio);
8328 btrfs_bio->logical = file_offset;
4b46fce2 8329
c1dc0896 8330 if (write) {
9be3395b 8331 io_bio->bi_end_io = btrfs_endio_direct_write;
c1dc0896 8332 } else {
9be3395b 8333 io_bio->bi_end_io = btrfs_endio_direct_read;
c1dc0896
MX		 8334 dip->subio_endio = btrfs_subio_endio_read;
8335 }
4b46fce2 8336
f28a4928
FM		 8337 /*
8338 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8339 * even if we fail to submit a bio, because in such case we do the
8340 * corresponding error handling below and it must not be done a second
8341 * time by btrfs_direct_IO().
8342 */
8343 if (write) {
8344 struct btrfs_dio_data *dio_data = current->journal_info;
8345
8346 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8347 dip->bytes;
8348 dio_data->unsubmitted_oe_range_start =
8349 dio_data->unsubmitted_oe_range_end;
8350 }
8351
e65e1535
MX		 8352 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8353 if (!ret)
eaf25d93 8354 return;
9be3395b 8355
23ea8e5a
MX		 8356 if (btrfs_bio->end_io)
8357 btrfs_bio->end_io(btrfs_bio, ret);
9be3395b 8358
4b46fce2
JB		 8359free_ordered:
8360 /*
61de718f
FM		 8361 * If we arrived here it means either we failed to submit the dip
8362 * or we either failed to clone the dio_bio or failed to allocate the
8363 * dip. If we cloned the dio_bio and allocated the dip, we can just
8364 * call bio_endio against our io_bio so that we get proper resource
8365 * cleanup if we fail to submit the dip, otherwise, we must do the
8366 * same as btrfs_endio_direct_[write|read] because we can't call these
8367 * callbacks - they require an allocated dip and a clone of dio_bio.
4b46fce2 8368 */
61de718f 8369 if (io_bio && dip) {
4246a0b6
CH		 8370 io_bio->bi_error = -EIO;
8371 bio_endio(io_bio);
61de718f
FM		 8372 /*
8373 * The end io callbacks free our dip, do the final put on io_bio
8374 * and all the cleanup and final put for dio_bio (through
8375 * dio_end_io()).
8376 */
8377 dip = NULL;
8378 io_bio = NULL;
8379 } else {
14543774
FM		 8380 if (write)
8381 btrfs_endio_direct_write_update_ordered(inode,
8382 file_offset,
8383 dio_bio->bi_iter.bi_size,
8384 0);
8385 else
61de718f
FM		 8386 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8387 file_offset + dio_bio->bi_iter.bi_size - 1);
14543774 8388
4246a0b6 8389 dio_bio->bi_error = -EIO;
61de718f
FM		 8390 /*
8391 * Releases and cleans up our dio_bio, no need to bio_put()
8392 * nor bio_endio()/bio_io_error() against dio_bio.
8393 */
8394 dio_end_io(dio_bio, ret);
4b46fce2 8395 }
61de718f
FM		 8396 if (io_bio)
8397 bio_put(io_bio);
8398 kfree(dip);
4b46fce2
JB		 8399}
8400
6f673763 8401static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
28060d5d 8402 const struct iov_iter *iter, loff_t offset)
5a5f79b5
CM		 8403{
8404 int seg;
a1b75f7d 8405 int i;
5a5f79b5
CM		 8406 unsigned blocksize_mask = root->sectorsize - 1;
8407 ssize_t retval = -EINVAL;
5a5f79b5
CM		 8408
8409 if (offset & blocksize_mask)
8410 goto out;
8411
28060d5d
AV		 8412 if (iov_iter_alignment(iter) & blocksize_mask)
8413 goto out;
a1b75f7d 8414
28060d5d 8415 /* If this is a write we don't need to check anymore */
6f673763 8416 if (iov_iter_rw(iter) == WRITE)
28060d5d
AV		 8417 return 0;
8418 /*
8419 * Check to make sure we don't have duplicate iov_base's in this
8420 * iovec, if so return EINVAL, otherwise we'll get csum errors
8421 * when reading back.
8422 */
8423 for (seg = 0; seg < iter->nr_segs; seg++) {
8424 for (i = seg + 1; i < iter->nr_segs; i++) {
8425 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
a1b75f7d
JB		 8426 goto out;
8427 }
5a5f79b5
CM		 8428 }
8429 retval = 0;
8430out:
8431 return retval;
8432}
eb838e73 8433
22c6186e
OS		 8434static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
8435 loff_t offset)
16432985 8436{
4b46fce2
JB		 8437 struct file *file = iocb->ki_filp;
8438 struct inode *inode = file->f_mapping->host;
50745b0a 8439 struct btrfs_root *root = BTRFS_I(inode)->root;
8440 struct btrfs_dio_data dio_data = { 0 };
0934856d 8441 size_t count = 0;
2e60a51e 8442 int flags = 0;
38851cc1
MX		 8443 bool wakeup = true;
8444 bool relock = false;
0934856d 8445 ssize_t ret;
4b46fce2 8446
6f673763 8447 if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
5a5f79b5 8448 return 0;
3f7c579c 8449
fe0f07d0 8450 inode_dio_begin(inode);
4e857c58 8451 smp_mb__after_atomic();
38851cc1 8452
0e267c44 8453 /*
41bd9ca4
MX		 8454 * The generic stuff only does filemap_write_and_wait_range, which
8455 * isn't enough if we've written compressed pages to this area, so
8456 * we need to flush the dirty pages again to make absolutely sure
8457 * that any outstanding dirty pages are on disk.
0e267c44 8458 */
a6cbcd4a 8459 count = iov_iter_count(iter);
41bd9ca4
MX		 8460 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8461 &BTRFS_I(inode)->runtime_flags))
9a025a08
WS		 8462 filemap_fdatawrite_range(inode->i_mapping, offset,
8463 offset + count - 1);
0e267c44 8464
6f673763 8465 if (iov_iter_rw(iter) == WRITE) {
38851cc1
MX		 8466 /*
8467 * If the write DIO is beyond the EOF, we need update
8468 * the isize, but it is protected by i_mutex. So we can
8469 * not unlock the i_mutex at this case.
8470 */
8471 if (offset + count <= inode->i_size) {
5955102c 8472 inode_unlock(inode);
38851cc1
MX		 8473 relock = true;
8474 }
7cf5b976 8475 ret = btrfs_delalloc_reserve_space(inode, offset, count);
0934856d 8476 if (ret)
38851cc1 8477 goto out;
50745b0a 8478 dio_data.outstanding_extents = div64_u64(count +
e1cbbfa5
JB		 8479 BTRFS_MAX_EXTENT_SIZE - 1,
8480 BTRFS_MAX_EXTENT_SIZE);
8481
8482 /*
8483 * We need to know how many extents we reserved so that we can
8484 * do the accounting properly if we go over the number we
8485 * originally calculated. Abuse current->journal_info for this.
8486 */
50745b0a 8487 dio_data.reserve = round_up(count, root->sectorsize);
f28a4928
FM		 8488 dio_data.unsubmitted_oe_range_start = (u64)offset;
8489 dio_data.unsubmitted_oe_range_end = (u64)offset;
50745b0a 8490 current->journal_info = &dio_data;
ee39b432
DS		 8491 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8492 &BTRFS_I(inode)->runtime_flags)) {
fe0f07d0 8493 inode_dio_end(inode);
38851cc1
MX		 8494 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8495 wakeup = false;
0934856d
MX		 8496 }
8497
17f8c842
OS		 8498 ret = __blockdev_direct_IO(iocb, inode,
8499 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8500 iter, offset, btrfs_get_blocks_direct, NULL,
8501 btrfs_submit_direct, flags);
6f673763 8502 if (iov_iter_rw(iter) == WRITE) {
e1cbbfa5 8503 current->journal_info = NULL;
ddba1bfc 8504 if (ret < 0 && ret != -EIOCBQUEUED) {
50745b0a 8505 if (dio_data.reserve)
7cf5b976
QW		 8506 btrfs_delalloc_release_space(inode, offset,
8507 dio_data.reserve);
f28a4928
FM		 8508 /*
8509 * On error we might have left some ordered extents
8510 * without submitting corresponding bios for them, so
8511 * cleanup them up to avoid other tasks getting them
8512 * and waiting for them to complete forever.
8513 */
8514 if (dio_data.unsubmitted_oe_range_start <
8515 dio_data.unsubmitted_oe_range_end)
8516 btrfs_endio_direct_write_update_ordered(inode,
8517 dio_data.unsubmitted_oe_range_start,
8518 dio_data.unsubmitted_oe_range_end -
8519 dio_data.unsubmitted_oe_range_start,
8520 0);
ddba1bfc 8521 } else if (ret >= 0 && (size_t)ret < count)
7cf5b976
QW		 8522 btrfs_delalloc_release_space(inode, offset,
8523 count - (size_t)ret);
0934856d 8524 }
38851cc1 8525out:
2e60a51e 8526 if (wakeup)
fe0f07d0 8527 inode_dio_end(inode);
38851cc1 8528 if (relock)
5955102c 8529 inode_lock(inode);
0934856d
MX		 8530
8531 return ret;
16432985
CM		 8532}
8533
05dadc09
TI		 8534#define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8535
1506fcc8
YS		 8536static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8537 __u64 start, __u64 len)
8538{
05dadc09
TI		 8539 int ret;
8540
8541 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8542 if (ret)
8543 return ret;
8544
ec29ed5b 8545 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
1506fcc8
YS		 8546}
8547
a52d9a80 8548int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 8549{
d1310b2e
CM		 8550 struct extent_io_tree *tree;
8551 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 8552 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
9ebefb18 8553}
1832a6d5 8554
a52d9a80 8555static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 8556{
d1310b2e 8557 struct extent_io_tree *tree;
be7bd730
JB		 8558 struct inode *inode = page->mapping->host;
8559 int ret;
b888db2b
CM		 8560
8561 if (current->flags & PF_MEMALLOC) {
8562 redirty_page_for_writepage(wbc, page);
8563 unlock_page(page);
8564 return 0;
8565 }
be7bd730
JB		 8566
8567 /*
8568 * If we are under memory pressure we will call this directly from the
8569 * VM, we need to make sure we have the inode referenced for the ordered
8570 * extent. If not just return like we didn't do anything.
8571 */
8572 if (!igrab(inode)) {
8573 redirty_page_for_writepage(wbc, page);
8574 return AOP_WRITEPAGE_ACTIVATE;
8575 }
d1310b2e 8576 tree = &BTRFS_I(page->mapping->host)->io_tree;
be7bd730
JB		 8577 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8578 btrfs_add_delayed_iput(inode);
8579 return ret;
9ebefb18
CM		 8580}
8581
48a3b636
ES		 8582static int btrfs_writepages(struct address_space *mapping,
8583 struct writeback_control *wbc)
b293f02e 8584{
d1310b2e 8585 struct extent_io_tree *tree;
771ed689 8586
d1310b2e 8587 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 8588 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8589}
8590
3ab2fb5a
CM		 8591static int
8592btrfs_readpages(struct file *file, struct address_space *mapping,
8593 struct list_head *pages, unsigned nr_pages)
8594{
d1310b2e
CM		 8595 struct extent_io_tree *tree;
8596 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 8597 return extent_readpages(tree, mapping, pages, nr_pages,
8598 btrfs_get_extent);
8599}
e6dcd2dc 8600static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 8601{
d1310b2e
CM		 8602 struct extent_io_tree *tree;
8603 struct extent_map_tree *map;
a52d9a80 8604 int ret;
8c2383c3 8605
d1310b2e
CM		 8606 tree = &BTRFS_I(page->mapping->host)->io_tree;
8607 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 8608 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 8609 if (ret == 1) {
8610 ClearPagePrivate(page);
8611 set_page_private(page, 0);
8612 page_cache_release(page);
39279cc3 8613 }
a52d9a80 8614 return ret;
39279cc3
CM		 8615}
8616
e6dcd2dc
CM		 8617static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8618{
98509cfc
CM		 8619 if (PageWriteback(page) || PageDirty(page))
8620 return 0;
b335b003 8621 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
e6dcd2dc
CM		 8622}
8623
d47992f8
LC		 8624static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8625 unsigned int length)
39279cc3 8626{
5fd02043 8627 struct inode *inode = page->mapping->host;
d1310b2e 8628 struct extent_io_tree *tree;
e6dcd2dc 8629 struct btrfs_ordered_extent *ordered;
2ac55d41 8630 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 8631 u64 page_start = page_offset(page);
8632 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
131e404a 8633 int inode_evicting = inode->i_state & I_FREEING;
39279cc3 8634
8b62b72b
CM		 8635 /*
8636 * we have the page locked, so new writeback can't start,
8637 * and the dirty bit won't be cleared while we are here.
8638 *
8639 * Wait for IO on this page so that we can safely clear
8640 * the PagePrivate2 bit and do ordered accounting
8641 */
e6dcd2dc 8642 wait_on_page_writeback(page);
8b62b72b 8643
5fd02043 8644 tree = &BTRFS_I(inode)->io_tree;
e6dcd2dc
CM		 8645 if (offset) {
8646 btrfs_releasepage(page, GFP_NOFS);
8647 return;
8648 }
131e404a
FDBM		 8649
8650 if (!inode_evicting)
ff13db41 8651 lock_extent_bits(tree, page_start, page_end, &cached_state);
131e404a 8652 ordered = btrfs_lookup_ordered_extent(inode, page_start);
e6dcd2dc 8653 if (ordered) {
eb84ae03
CM		 8654 /*
8655 * IO on this page will never be started, so we need
8656 * to account for any ordered extents now
8657 */
131e404a
FDBM		 8658 if (!inode_evicting)
8659 clear_extent_bit(tree, page_start, page_end,
8660 EXTENT_DIRTY | EXTENT_DELALLOC |
8661 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8662 EXTENT_DEFRAG, 1, 0, &cached_state,
8663 GFP_NOFS);
8b62b72b
CM		 8664 /*
8665 * whoever cleared the private bit is responsible
8666 * for the finish_ordered_io
8667 */
77cef2ec
JB		 8668 if (TestClearPagePrivate2(page)) {
8669 struct btrfs_ordered_inode_tree *tree;
8670 u64 new_len;
8671
8672 tree = &BTRFS_I(inode)->ordered_tree;
8673
8674 spin_lock_irq(&tree->lock);
8675 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8676 new_len = page_start - ordered->file_offset;
8677 if (new_len < ordered->truncated_len)
8678 ordered->truncated_len = new_len;
8679 spin_unlock_irq(&tree->lock);
8680
8681 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8682 page_start,
8683 PAGE_CACHE_SIZE, 1))
8684 btrfs_finish_ordered_io(ordered);
8b62b72b 8685 }
e6dcd2dc 8686 btrfs_put_ordered_extent(ordered);
131e404a
FDBM		 8687 if (!inode_evicting) {
8688 cached_state = NULL;
ff13db41 8689 lock_extent_bits(tree, page_start, page_end,
131e404a
FDBM		 8690 &cached_state);
8691 }
8692 }
8693
b9d0b389
QW		 8694 /*
8695 * Qgroup reserved space handler
8696 * Page here will be either
8697 * 1) Already written to disk
8698 * In this case, its reserved space is released from data rsv map
8699 * and will be freed by delayed_ref handler finally.
8700 * So even we call qgroup_free_data(), it won't decrease reserved
8701 * space.
8702 * 2) Not written to disk
8703 * This means the reserved space should be freed here.
8704 */
8705 btrfs_qgroup_free_data(inode, page_start, PAGE_CACHE_SIZE);
131e404a
FDBM		 8706 if (!inode_evicting) {
8707 clear_extent_bit(tree, page_start, page_end,
8708 EXTENT_LOCKED | EXTENT_DIRTY |
8709 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8710 EXTENT_DEFRAG, 1, 1,
8711 &cached_state, GFP_NOFS);
8712
8713 __btrfs_releasepage(page, GFP_NOFS);
e6dcd2dc 8714 }
e6dcd2dc 8715
4a096752 8716 ClearPageChecked(page);
9ad6b7bc 8717 if (PagePrivate(page)) {
9ad6b7bc
CM		 8718 ClearPagePrivate(page);
8719 set_page_private(page, 0);
8720 page_cache_release(page);
8721 }
39279cc3
CM		 8722}
8723
9ebefb18
CM		 8724/*
8725 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8726 * called from a page fault handler when a page is first dirtied. Hence we must
8727 * be careful to check for EOF conditions here. We set the page up correctly
8728 * for a written page which means we get ENOSPC checking when writing into
8729 * holes and correct delalloc and unwritten extent mapping on filesystems that
8730 * support these features.
8731 *
8732 * We are not allowed to take the i_mutex here so we have to play games to
8733 * protect against truncate races as the page could now be beyond EOF. Because
8734 * vmtruncate() writes the inode size before removing pages, once we have the
8735 * page lock we can determine safely if the page is beyond EOF. If it is not
8736 * beyond EOF, then the page is guaranteed safe against truncation until we
8737 * unlock the page.
8738 */
c2ec175c 8739int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
9ebefb18 8740{
c2ec175c 8741 struct page *page = vmf->page;
496ad9aa 8742 struct inode *inode = file_inode(vma->vm_file);
1832a6d5 8743 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 8744 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8745 struct btrfs_ordered_extent *ordered;
2ac55d41 8746 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 8747 char *kaddr;
8748 unsigned long zero_start;
9ebefb18 8749 loff_t size;
1832a6d5 8750 int ret;
9998eb70 8751 int reserved = 0;
a52d9a80 8752 u64 page_start;
e6dcd2dc 8753 u64 page_end;
9ebefb18 8754
b2b5ef5c 8755 sb_start_pagefault(inode->i_sb);
df480633
QW		 8756 page_start = page_offset(page);
8757 page_end = page_start + PAGE_CACHE_SIZE - 1;
8758
7cf5b976
QW		 8759 ret = btrfs_delalloc_reserve_space(inode, page_start,
8760 PAGE_CACHE_SIZE);
9998eb70 8761 if (!ret) {
e41f941a 8762 ret = file_update_time(vma->vm_file);
9998eb70
CM		 8763 reserved = 1;
8764 }
56a76f82
NP		 8765 if (ret) {
8766 if (ret == -ENOMEM)
8767 ret = VM_FAULT_OOM;
8768 else /* -ENOSPC, -EIO, etc */
8769 ret = VM_FAULT_SIGBUS;
9998eb70
CM		 8770 if (reserved)
8771 goto out;
8772 goto out_noreserve;
56a76f82 8773 }
1832a6d5 8774
56a76f82 8775 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
e6dcd2dc 8776again:
9ebefb18 8777 lock_page(page);
9ebefb18 8778 size = i_size_read(inode);
a52d9a80 8779
9ebefb18 8780 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 8781 (page_start >= size)) {
9ebefb18
CM		 8782 /* page got truncated out from underneath us */
8783 goto out_unlock;
8784 }
e6dcd2dc
CM		 8785 wait_on_page_writeback(page);
8786
ff13db41 8787 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
e6dcd2dc
CM		 8788 set_page_extent_mapped(page);
8789
eb84ae03
CM		 8790 /*
8791 * we can't set the delalloc bits if there are pending ordered
8792 * extents. Drop our locks and wait for them to finish
8793 */
e6dcd2dc
CM		 8794 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8795 if (ordered) {
2ac55d41
JB		 8796 unlock_extent_cached(io_tree, page_start, page_end,
8797 &cached_state, GFP_NOFS);
e6dcd2dc 8798 unlock_page(page);
eb84ae03 8799 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 8800 btrfs_put_ordered_extent(ordered);
8801 goto again;
8802 }
8803
fbf19087
JB		 8804 /*
8805 * XXX - page_mkwrite gets called every time the page is dirtied, even
8806 * if it was already dirty, so for space accounting reasons we need to
8807 * clear any delalloc bits for the range we are fixing to save. There
8808 * is probably a better way to do this, but for now keep consistent with
8809 * prepare_pages in the normal write path.
8810 */
2ac55d41 8811 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 8812 EXTENT_DIRTY | EXTENT_DELALLOC |
8813 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 8814 0, 0, &cached_state, GFP_NOFS);
fbf19087 8815
2ac55d41
JB		 8816 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
8817 &cached_state);
9ed74f2d 8818 if (ret) {
2ac55d41
JB		 8819 unlock_extent_cached(io_tree, page_start, page_end,
8820 &cached_state, GFP_NOFS);
9ed74f2d
JB		 8821 ret = VM_FAULT_SIGBUS;
8822 goto out_unlock;
8823 }
e6dcd2dc 8824 ret = 0;
9ebefb18
CM		 8825
8826 /* page is wholly or partially inside EOF */
a52d9a80 8827 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 8828 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 8829 else
e6dcd2dc 8830 zero_start = PAGE_CACHE_SIZE;
9ebefb18 8831
e6dcd2dc
CM		 8832 if (zero_start != PAGE_CACHE_SIZE) {
8833 kaddr = kmap(page);
8834 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
8835 flush_dcache_page(page);
8836 kunmap(page);
8837 }
247e743c 8838 ClearPageChecked(page);
e6dcd2dc 8839 set_page_dirty(page);
50a9b214 8840 SetPageUptodate(page);
5a3f23d5 8841
257c62e1
CM		 8842 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8843 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
46d8bc34 8844 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
257c62e1 8845
2ac55d41 8846 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9ebefb18
CM		 8847
8848out_unlock:
b2b5ef5c
JK		 8849 if (!ret) {
8850 sb_end_pagefault(inode->i_sb);
50a9b214 8851 return VM_FAULT_LOCKED;
b2b5ef5c 8852 }
9ebefb18 8853 unlock_page(page);
1832a6d5 8854out:
7cf5b976 8855 btrfs_delalloc_release_space(inode, page_start, PAGE_CACHE_SIZE);
9998eb70 8856out_noreserve:
b2b5ef5c 8857 sb_end_pagefault(inode->i_sb);
9ebefb18
CM		 8858 return ret;
8859}
8860
a41ad394 8861static int btrfs_truncate(struct inode *inode)
39279cc3
CM		 8862{
8863 struct btrfs_root *root = BTRFS_I(inode)->root;
fcb80c2a 8864 struct btrfs_block_rsv *rsv;
a71754fc 8865 int ret = 0;
3893e33b 8866 int err = 0;
39279cc3 8867 struct btrfs_trans_handle *trans;
dbe674a9 8868 u64 mask = root->sectorsize - 1;
07127184 8869 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3 8870
0ef8b726
JB		 8871 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8872 (u64)-1);
8873 if (ret)
8874 return ret;
39279cc3 8875
fcb80c2a
JB		 8876 /*
8877 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8878 * 3 things going on here
8879 *
8880 * 1) We need to reserve space for our orphan item and the space to
8881 * delete our orphan item. Lord knows we don't want to have a dangling
8882 * orphan item because we didn't reserve space to remove it.
8883 *
8884 * 2) We need to reserve space to update our inode.
8885 *
8886 * 3) We need to have something to cache all the space that is going to
8887 * be free'd up by the truncate operation, but also have some slack
8888 * space reserved in case it uses space during the truncate (thank you
8889 * very much snapshotting).
8890 *
8891 * And we need these to all be seperate. The fact is we can use alot of
8892 * space doing the truncate, and we have no earthly idea how much space
8893 * we will use, so we need the truncate reservation to be seperate so it
8894 * doesn't end up using space reserved for updating the inode or
8895 * removing the orphan item. We also need to be able to stop the
8896 * transaction and start a new one, which means we need to be able to
8897 * update the inode several times, and we have no idea of knowing how
8898 * many times that will be, so we can't just reserve 1 item for the
8899 * entirety of the opration, so that has to be done seperately as well.
8900 * Then there is the orphan item, which does indeed need to be held on
8901 * to for the whole operation, and we need nobody to touch this reserved
8902 * space except the orphan code.
8903 *
8904 * So that leaves us with
8905 *
8906 * 1) root->orphan_block_rsv - for the orphan deletion.
8907 * 2) rsv - for the truncate reservation, which we will steal from the
8908 * transaction reservation.
8909 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8910 * updating the inode.
8911 */
66d8f3dd 8912 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
fcb80c2a
JB		 8913 if (!rsv)
8914 return -ENOMEM;
4a338542 8915 rsv->size = min_size;
ca7e70f5 8916 rsv->failfast = 1;
f0cd846e 8917
907cbceb 8918 /*
07127184 8919 * 1 for the truncate slack space
907cbceb
JB		 8920 * 1 for updating the inode.
8921 */
f3fe820c 8922 trans = btrfs_start_transaction(root, 2);
fcb80c2a
JB		 8923 if (IS_ERR(trans)) {
8924 err = PTR_ERR(trans);
8925 goto out;
8926 }
f0cd846e 8927
907cbceb
JB		 8928 /* Migrate the slack space for the truncate to our reserve */
8929 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8930 min_size);
fcb80c2a 8931 BUG_ON(ret);
f0cd846e 8932
5dc562c5
JB		 8933 /*
8934 * So if we truncate and then write and fsync we normally would just
8935 * write the extents that changed, which is a problem if we need to
8936 * first truncate that entire inode. So set this flag so we write out
8937 * all of the extents in the inode to the sync log so we're completely
8938 * safe.
8939 */
8940 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
ca7e70f5 8941 trans->block_rsv = rsv;
907cbceb 8942
8082510e
YZ		 8943 while (1) {
8944 ret = btrfs_truncate_inode_items(trans, root, inode,
8945 inode->i_size,
8946 BTRFS_EXTENT_DATA_KEY);
28ed1345 8947 if (ret != -ENOSPC && ret != -EAGAIN) {
3893e33b 8948 err = ret;
8082510e 8949 break;
3893e33b 8950 }
39279cc3 8951
fcb80c2a 8952 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e 8953 ret = btrfs_update_inode(trans, root, inode);
3893e33b
JB		 8954 if (ret) {
8955 err = ret;
8956 break;
8957 }
ca7e70f5 8958
8082510e 8959 btrfs_end_transaction(trans, root);
b53d3f5d 8960 btrfs_btree_balance_dirty(root);
ca7e70f5
JB		 8961
8962 trans = btrfs_start_transaction(root, 2);
8963 if (IS_ERR(trans)) {
8964 ret = err = PTR_ERR(trans);
8965 trans = NULL;
8966 break;
8967 }
8968
8969 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
8970 rsv, min_size);
8971 BUG_ON(ret); /* shouldn't happen */
8972 trans->block_rsv = rsv;
8082510e
YZ		 8973 }
8974
8975 if (ret == 0 && inode->i_nlink > 0) {
fcb80c2a 8976 trans->block_rsv = root->orphan_block_rsv;
8082510e 8977 ret = btrfs_orphan_del(trans, inode);
3893e33b
JB		 8978 if (ret)
8979 err = ret;
8082510e
YZ		 8980 }
8981
917c16b2
CM		 8982 if (trans) {
8983 trans->block_rsv = &root->fs_info->trans_block_rsv;
8984 ret = btrfs_update_inode(trans, root, inode);
8985 if (ret && !err)
8986 err = ret;
7b128766 8987
7ad85bb7 8988 ret = btrfs_end_transaction(trans, root);
b53d3f5d 8989 btrfs_btree_balance_dirty(root);
917c16b2 8990 }
fcb80c2a
JB		 8991
8992out:
8993 btrfs_free_block_rsv(root, rsv);
8994
3893e33b
JB		 8995 if (ret && !err)
8996 err = ret;
a41ad394 8997
3893e33b 8998 return err;
39279cc3
CM		 8999}
9000
d352ac68
CM		 9001/*
9002 * create a new subvolume directory/inode (helper for the ioctl).
9003 */
d2fb3437 9004int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
63541927
FDBM		 9005 struct btrfs_root *new_root,
9006 struct btrfs_root *parent_root,
9007 u64 new_dirid)
39279cc3 9008{
39279cc3 9009 struct inode *inode;
76dda93c 9010 int err;
00e4e6b3 9011 u64 index = 0;
39279cc3 9012
12fc9d09
FA		 9013 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9014 new_dirid, new_dirid,
9015 S_IFDIR | (~current_umask() & S_IRWXUGO),
9016 &index);
54aa1f4d 9017 if (IS_ERR(inode))
f46b5a66 9018 return PTR_ERR(inode);
39279cc3
CM		 9019 inode->i_op = &btrfs_dir_inode_operations;
9020 inode->i_fop = &btrfs_dir_file_operations;
9021
bfe86848 9022 set_nlink(inode, 1);
dbe674a9 9023 btrfs_i_size_write(inode, 0);
b0d5d10f 9024 unlock_new_inode(inode);
3b96362c 9025
63541927
FDBM		 9026 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9027 if (err)
9028 btrfs_err(new_root->fs_info,
351fd353 9029 "error inheriting subvolume %llu properties: %d",
63541927
FDBM		 9030 new_root->root_key.objectid, err);
9031
76dda93c 9032 err = btrfs_update_inode(trans, new_root, inode);
cb8e7090 9033
76dda93c 9034 iput(inode);
ce598979 9035 return err;
39279cc3
CM		 9036}
9037
39279cc3
CM		 9038struct inode *btrfs_alloc_inode(struct super_block *sb)
9039{
9040 struct btrfs_inode *ei;
2ead6ae7 9041 struct inode *inode;
39279cc3
CM		 9042
9043 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9044 if (!ei)
9045 return NULL;
2ead6ae7
YZ		 9046
9047 ei->root = NULL;
2ead6ae7 9048 ei->generation = 0;
15ee9bc7 9049 ei->last_trans = 0;
257c62e1 9050 ei->last_sub_trans = 0;
e02119d5 9051 ei->logged_trans = 0;
2ead6ae7 9052 ei->delalloc_bytes = 0;
47059d93 9053 ei->defrag_bytes = 0;
2ead6ae7
YZ		 9054 ei->disk_i_size = 0;
9055 ei->flags = 0;
7709cde3 9056 ei->csum_bytes = 0;
2ead6ae7 9057 ei->index_cnt = (u64)-1;
67de1176 9058 ei->dir_index = 0;
2ead6ae7 9059 ei->last_unlink_trans = 0;
46d8bc34 9060 ei->last_log_commit = 0;
8089fe62 9061 ei->delayed_iput_count = 0;
2ead6ae7 9062
9e0baf60
JB		 9063 spin_lock_init(&ei->lock);
9064 ei->outstanding_extents = 0;
9065 ei->reserved_extents = 0;
2ead6ae7 9066
72ac3c0d 9067 ei->runtime_flags = 0;
261507a0 9068 ei->force_compress = BTRFS_COMPRESS_NONE;
2ead6ae7 9069
16cdcec7
MX		 9070 ei->delayed_node = NULL;
9071
9cc97d64 9072 ei->i_otime.tv_sec = 0;
9073 ei->i_otime.tv_nsec = 0;
9074
2ead6ae7 9075 inode = &ei->vfs_inode;
a8067e02 9076 extent_map_tree_init(&ei->extent_tree);
f993c883
DS		 9077 extent_io_tree_init(&ei->io_tree, &inode->i_data);
9078 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
0b32f4bb
JB		 9079 ei->io_tree.track_uptodate = 1;
9080 ei->io_failure_tree.track_uptodate = 1;
b812ce28 9081 atomic_set(&ei->sync_writers, 0);
2ead6ae7 9082 mutex_init(&ei->log_mutex);
f248679e 9083 mutex_init(&ei->delalloc_mutex);
e6dcd2dc 9084 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
2ead6ae7 9085 INIT_LIST_HEAD(&ei->delalloc_inodes);
8089fe62 9086 INIT_LIST_HEAD(&ei->delayed_iput);
2ead6ae7
YZ		 9087 RB_CLEAR_NODE(&ei->rb_node);
9088
9089 return inode;
39279cc3
CM		 9090}
9091
aaedb55b
JB		 9092#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9093void btrfs_test_destroy_inode(struct inode *inode)
9094{
9095 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9096 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9097}
9098#endif
9099
fa0d7e3d
NP		 9100static void btrfs_i_callback(struct rcu_head *head)
9101{
9102 struct inode *inode = container_of(head, struct inode, i_rcu);
fa0d7e3d
NP		 9103 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9104}
9105
39279cc3
CM		 9106void btrfs_destroy_inode(struct inode *inode)
9107{
e6dcd2dc 9108 struct btrfs_ordered_extent *ordered;
5a3f23d5
CM		 9109 struct btrfs_root *root = BTRFS_I(inode)->root;
9110
b3d9b7a3 9111 WARN_ON(!hlist_empty(&inode->i_dentry));
39279cc3 9112 WARN_ON(inode->i_data.nrpages);
9e0baf60
JB		 9113 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9114 WARN_ON(BTRFS_I(inode)->reserved_extents);
7709cde3
JB		 9115 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9116 WARN_ON(BTRFS_I(inode)->csum_bytes);
47059d93 9117 WARN_ON(BTRFS_I(inode)->defrag_bytes);
39279cc3 9118
a6dbd429
JB		 9119 /*
9120 * This can happen where we create an inode, but somebody else also
9121 * created the same inode and we need to destroy the one we already
9122 * created.
9123 */
9124 if (!root)
9125 goto free;
9126
8a35d95f
JB		 9127 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9128 &BTRFS_I(inode)->runtime_flags)) {
c2cf52eb 9129 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
c1c9ff7c 9130 btrfs_ino(inode));
8a35d95f 9131 atomic_dec(&root->orphan_inodes);
7b128766 9132 }
7b128766 9133
d397712b 9134 while (1) {
e6dcd2dc
CM		 9135 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9136 if (!ordered)
9137 break;
9138 else {
c2cf52eb 9139 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
c1c9ff7c 9140 ordered->file_offset, ordered->len);
e6dcd2dc
CM		 9141 btrfs_remove_ordered_extent(inode, ordered);
9142 btrfs_put_ordered_extent(ordered);
9143 btrfs_put_ordered_extent(ordered);
9144 }
9145 }
56fa9d07 9146 btrfs_qgroup_check_reserved_leak(inode);
5d4f98a2 9147 inode_tree_del(inode);
5b21f2ed 9148 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
a6dbd429 9149free:
fa0d7e3d 9150 call_rcu(&inode->i_rcu, btrfs_i_callback);
39279cc3
CM		 9151}
9152
45321ac5 9153int btrfs_drop_inode(struct inode *inode)
76dda93c
YZ		 9154{
9155 struct btrfs_root *root = BTRFS_I(inode)->root;
45321ac5 9156
6379ef9f
NA		 9157 if (root == NULL)
9158 return 1;
9159
fa6ac876 9160 /* the snap/subvol tree is on deleting */
69e9c6c6 9161 if (btrfs_root_refs(&root->root_item) == 0)
45321ac5 9162 return 1;
76dda93c 9163 else
45321ac5 9164 return generic_drop_inode(inode);
76dda93c
YZ		 9165}
9166
0ee0fda0 9167static void init_once(void *foo)
39279cc3
CM		 9168{
9169 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9170
9171 inode_init_once(&ei->vfs_inode);
9172}
9173
9174void btrfs_destroy_cachep(void)
9175{
8c0a8537
KS		 9176 /*
9177 * Make sure all delayed rcu free inodes are flushed before we
9178 * destroy cache.
9179 */
9180 rcu_barrier();
5598e900
KM		 9181 kmem_cache_destroy(btrfs_inode_cachep);
9182 kmem_cache_destroy(btrfs_trans_handle_cachep);
9183 kmem_cache_destroy(btrfs_transaction_cachep);
9184 kmem_cache_destroy(btrfs_path_cachep);
9185 kmem_cache_destroy(btrfs_free_space_cachep);
39279cc3
CM		 9186}
9187
9188int btrfs_init_cachep(void)
9189{
837e1972 9190 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9601e3f6 9191 sizeof(struct btrfs_inode), 0,
5d097056
VD		 9192 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9193 init_once);
39279cc3
CM		 9194 if (!btrfs_inode_cachep)
9195 goto fail;
9601e3f6 9196
837e1972 9197 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9601e3f6
CH		 9198 sizeof(struct btrfs_trans_handle), 0,
9199 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 9200 if (!btrfs_trans_handle_cachep)
9201 goto fail;
9601e3f6 9202
837e1972 9203 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9601e3f6
CH		 9204 sizeof(struct btrfs_transaction), 0,
9205 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 9206 if (!btrfs_transaction_cachep)
9207 goto fail;
9601e3f6 9208
837e1972 9209 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9601e3f6
CH		 9210 sizeof(struct btrfs_path), 0,
9211 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 9212 if (!btrfs_path_cachep)
9213 goto fail;
9601e3f6 9214
837e1972 9215 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
dc89e982
JB		 9216 sizeof(struct btrfs_free_space), 0,
9217 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9218 if (!btrfs_free_space_cachep)
9219 goto fail;
9220
39279cc3
CM		 9221 return 0;
9222fail:
9223 btrfs_destroy_cachep();
9224 return -ENOMEM;
9225}
9226
9227static int btrfs_getattr(struct vfsmount *mnt,
9228 struct dentry *dentry, struct kstat *stat)
9229{
df0af1a5 9230 u64 delalloc_bytes;
2b0143b5 9231 struct inode *inode = d_inode(dentry);
fadc0d8b
DS		 9232 u32 blocksize = inode->i_sb->s_blocksize;
9233
39279cc3 9234 generic_fillattr(inode, stat);
0ee5dc67 9235 stat->dev = BTRFS_I(inode)->root->anon_dev;
d6667462 9236 stat->blksize = PAGE_CACHE_SIZE;
df0af1a5
MX		 9237
9238 spin_lock(&BTRFS_I(inode)->lock);
9239 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9240 spin_unlock(&BTRFS_I(inode)->lock);
fadc0d8b 9241 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
df0af1a5 9242 ALIGN(delalloc_bytes, blocksize)) >> 9;
39279cc3
CM		 9243 return 0;
9244}
9245
d397712b
CM		 9246static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9247 struct inode *new_dir, struct dentry *new_dentry)
39279cc3
CM		 9248{
9249 struct btrfs_trans_handle *trans;
9250 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4df27c4d 9251 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
2b0143b5
DH		 9252 struct inode *new_inode = d_inode(new_dentry);
9253 struct inode *old_inode = d_inode(old_dentry);
00e4e6b3 9254 u64 index = 0;
4df27c4d 9255 u64 root_objectid;
39279cc3 9256 int ret;
33345d01 9257 u64 old_ino = btrfs_ino(old_inode);
39279cc3 9258
33345d01 9259 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
f679a840
YZ		 9260 return -EPERM;
9261
4df27c4d 9262 /* we only allow rename subvolume link between subvolumes */
33345d01 9263 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
3394e160
CM		 9264 return -EXDEV;
9265
33345d01
LZ		 9266 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9267 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
39279cc3 9268 return -ENOTEMPTY;
5f39d397 9269
4df27c4d
YZ		 9270 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9271 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9272 return -ENOTEMPTY;
9c52057c
CM		 9273
9274
9275 /* check for collisions, even if the name isn't there */
4871c158 9276 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9c52057c
CM		 9277 new_dentry->d_name.name,
9278 new_dentry->d_name.len);
9279
9280 if (ret) {
9281 if (ret == -EEXIST) {
9282 /* we shouldn't get
9283 * eexist without a new_inode */
fae7f21c 9284 if (WARN_ON(!new_inode)) {
9c52057c
CM		 9285 return ret;
9286 }
9287 } else {
9288 /* maybe -EOVERFLOW */
9289 return ret;
9290 }
9291 }
9292 ret = 0;
9293
5a3f23d5 9294 /*
8d875f95
CM		 9295 * we're using rename to replace one file with another. Start IO on it
9296 * now so we don't add too much work to the end of the transaction
5a3f23d5 9297 */
8d875f95 9298 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
5a3f23d5
CM		 9299 filemap_flush(old_inode->i_mapping);
9300
76dda93c 9301 /* close the racy window with snapshot create/destroy ioctl */
33345d01 9302 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 9303 down_read(&root->fs_info->subvol_sem);
a22285a6
YZ		 9304 /*
9305 * We want to reserve the absolute worst case amount of items. So if
9306 * both inodes are subvols and we need to unlink them then that would
9307 * require 4 item modifications, but if they are both normal inodes it
9308 * would require 5 item modifications, so we'll assume their normal
9309 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9310 * should cover the worst case number of items we'll modify.
9311 */
6e137ed3 9312 trans = btrfs_start_transaction(root, 11);
b44c59a8
JL		 9313 if (IS_ERR(trans)) {
9314 ret = PTR_ERR(trans);
9315 goto out_notrans;
9316 }
76dda93c 9317
4df27c4d
YZ		 9318 if (dest != root)
9319 btrfs_record_root_in_trans(trans, dest);
5f39d397 9320
a5719521
YZ		 9321 ret = btrfs_set_inode_index(new_dir, &index);
9322 if (ret)
9323 goto out_fail;
5a3f23d5 9324
67de1176 9325 BTRFS_I(old_inode)->dir_index = 0ULL;
33345d01 9326 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d 9327 /* force full log commit if subvolume involved. */
995946dd 9328 btrfs_set_log_full_commit(root->fs_info, trans);
4df27c4d 9329 } else {
a5719521
YZ		 9330 ret = btrfs_insert_inode_ref(trans, dest,
9331 new_dentry->d_name.name,
9332 new_dentry->d_name.len,
33345d01
LZ		 9333 old_ino,
9334 btrfs_ino(new_dir), index);
a5719521
YZ		 9335 if (ret)
9336 goto out_fail;
4df27c4d
YZ		 9337 /*
9338 * this is an ugly little race, but the rename is required
9339 * to make sure that if we crash, the inode is either at the
9340 * old name or the new one. pinning the log transaction lets
9341 * us make sure we don't allow a log commit to come in after
9342 * we unlink the name but before we add the new name back in.
9343 */
9344 btrfs_pin_log_trans(root);
9345 }
5a3f23d5 9346
0c4d2d95
JB		 9347 inode_inc_iversion(old_dir);
9348 inode_inc_iversion(new_dir);
9349 inode_inc_iversion(old_inode);
04b285f3
DD		 9350 old_dir->i_ctime = old_dir->i_mtime =
9351 new_dir->i_ctime = new_dir->i_mtime =
9352 old_inode->i_ctime = current_fs_time(old_dir->i_sb);
5f39d397 9353
12fcfd22
CM		 9354 if (old_dentry->d_parent != new_dentry->d_parent)
9355 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9356
33345d01 9357 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 9358 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9359 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9360 old_dentry->d_name.name,
9361 old_dentry->d_name.len);
9362 } else {
92986796 9363 ret = __btrfs_unlink_inode(trans, root, old_dir,
2b0143b5 9364 d_inode(old_dentry),
92986796
AV		 9365 old_dentry->d_name.name,
9366 old_dentry->d_name.len);
9367 if (!ret)
9368 ret = btrfs_update_inode(trans, root, old_inode);
4df27c4d 9369 }
79787eaa
JM		 9370 if (ret) {
9371 btrfs_abort_transaction(trans, root, ret);
9372 goto out_fail;
9373 }
39279cc3
CM		 9374
9375 if (new_inode) {
0c4d2d95 9376 inode_inc_iversion(new_inode);
04b285f3 9377 new_inode->i_ctime = current_fs_time(new_inode->i_sb);
33345d01 9378 if (unlikely(btrfs_ino(new_inode) ==
4df27c4d
YZ		 9379 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9380 root_objectid = BTRFS_I(new_inode)->location.objectid;
9381 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9382 root_objectid,
9383 new_dentry->d_name.name,
9384 new_dentry->d_name.len);
9385 BUG_ON(new_inode->i_nlink == 0);
9386 } else {
9387 ret = btrfs_unlink_inode(trans, dest, new_dir,
2b0143b5 9388 d_inode(new_dentry),
4df27c4d
YZ		 9389 new_dentry->d_name.name,
9390 new_dentry->d_name.len);
9391 }
4ef31a45 9392 if (!ret && new_inode->i_nlink == 0)
2b0143b5 9393 ret = btrfs_orphan_add(trans, d_inode(new_dentry));
79787eaa
JM		 9394 if (ret) {
9395 btrfs_abort_transaction(trans, root, ret);
9396 goto out_fail;
9397 }
39279cc3 9398 }
aec7477b 9399
4df27c4d
YZ		 9400 ret = btrfs_add_link(trans, new_dir, old_inode,
9401 new_dentry->d_name.name,
a5719521 9402 new_dentry->d_name.len, 0, index);
79787eaa
JM		 9403 if (ret) {
9404 btrfs_abort_transaction(trans, root, ret);
9405 goto out_fail;
9406 }
39279cc3 9407
67de1176
MX		 9408 if (old_inode->i_nlink == 1)
9409 BTRFS_I(old_inode)->dir_index = index;
9410
33345d01 9411 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
10d9f309 9412 struct dentry *parent = new_dentry->d_parent;
6a912213 9413 btrfs_log_new_name(trans, old_inode, old_dir, parent);
4df27c4d
YZ		 9414 btrfs_end_log_trans(root);
9415 }
39279cc3 9416out_fail:
7ad85bb7 9417 btrfs_end_transaction(trans, root);
b44c59a8 9418out_notrans:
33345d01 9419 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 9420 up_read(&root->fs_info->subvol_sem);
9ed74f2d 9421
39279cc3
CM		 9422 return ret;
9423}
9424
80ace85c
MS		 9425static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9426 struct inode *new_dir, struct dentry *new_dentry,
9427 unsigned int flags)
9428{
9429 if (flags & ~RENAME_NOREPLACE)
9430 return -EINVAL;
9431
9432 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
9433}
9434
8ccf6f19
MX		 9435static void btrfs_run_delalloc_work(struct btrfs_work *work)
9436{
9437 struct btrfs_delalloc_work *delalloc_work;
9f23e289 9438 struct inode *inode;
8ccf6f19
MX		 9439
9440 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9441 work);
9f23e289 9442 inode = delalloc_work->inode;
30424601
DS		 9443 filemap_flush(inode->i_mapping);
9444 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9445 &BTRFS_I(inode)->runtime_flags))
9f23e289 9446 filemap_flush(inode->i_mapping);
8ccf6f19
MX		 9447
9448 if (delalloc_work->delay_iput)
9f23e289 9449 btrfs_add_delayed_iput(inode);
8ccf6f19 9450 else
9f23e289 9451 iput(inode);
8ccf6f19
MX		 9452 complete(&delalloc_work->completion);
9453}
9454
9455struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
651d494a 9456 int delay_iput)
8ccf6f19
MX		 9457{
9458 struct btrfs_delalloc_work *work;
9459
100d5702 9460 work = kmalloc(sizeof(*work), GFP_NOFS);
8ccf6f19
MX		 9461 if (!work)
9462 return NULL;
9463
9464 init_completion(&work->completion);
9465 INIT_LIST_HEAD(&work->list);
9466 work->inode = inode;
8ccf6f19 9467 work->delay_iput = delay_iput;
9e0af237
LB		 9468 WARN_ON_ONCE(!inode);
9469 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9470 btrfs_run_delalloc_work, NULL, NULL);
8ccf6f19
MX		 9471
9472 return work;
9473}
9474
9475void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9476{
9477 wait_for_completion(&work->completion);
100d5702 9478 kfree(work);
8ccf6f19
MX		 9479}
9480
d352ac68
CM		 9481/*
9482 * some fairly slow code that needs optimization. This walks the list
9483 * of all the inodes with pending delalloc and forces them to disk.
9484 */
6c255e67
MX		 9485static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9486 int nr)
ea8c2819 9487{
ea8c2819 9488 struct btrfs_inode *binode;
5b21f2ed 9489 struct inode *inode;
8ccf6f19
MX		 9490 struct btrfs_delalloc_work *work, *next;
9491 struct list_head works;
1eafa6c7 9492 struct list_head splice;
8ccf6f19 9493 int ret = 0;
ea8c2819 9494
8ccf6f19 9495 INIT_LIST_HEAD(&works);
1eafa6c7 9496 INIT_LIST_HEAD(&splice);
63607cc8 9497
573bfb72 9498 mutex_lock(&root->delalloc_mutex);
eb73c1b7
MX		 9499 spin_lock(&root->delalloc_lock);
9500 list_splice_init(&root->delalloc_inodes, &splice);
1eafa6c7
MX		 9501 while (!list_empty(&splice)) {
9502 binode = list_entry(splice.next, struct btrfs_inode,
ea8c2819 9503 delalloc_inodes);
1eafa6c7 9504
eb73c1b7
MX		 9505 list_move_tail(&binode->delalloc_inodes,
9506 &root->delalloc_inodes);
5b21f2ed 9507 inode = igrab(&binode->vfs_inode);
df0af1a5 9508 if (!inode) {
eb73c1b7 9509 cond_resched_lock(&root->delalloc_lock);
1eafa6c7 9510 continue;
df0af1a5 9511 }
eb73c1b7 9512 spin_unlock(&root->delalloc_lock);
1eafa6c7 9513
651d494a 9514 work = btrfs_alloc_delalloc_work(inode, delay_iput);
5d99a998 9515 if (!work) {
f4ab9ea7
JB		 9516 if (delay_iput)
9517 btrfs_add_delayed_iput(inode);
9518 else
9519 iput(inode);
1eafa6c7 9520 ret = -ENOMEM;
a1ecaabb 9521 goto out;
5b21f2ed 9522 }
1eafa6c7 9523 list_add_tail(&work->list, &works);
a44903ab
QW		 9524 btrfs_queue_work(root->fs_info->flush_workers,
9525 &work->work);
6c255e67
MX		 9526 ret++;
9527 if (nr != -1 && ret >= nr)
a1ecaabb 9528 goto out;
5b21f2ed 9529 cond_resched();
eb73c1b7 9530 spin_lock(&root->delalloc_lock);
ea8c2819 9531 }
eb73c1b7 9532 spin_unlock(&root->delalloc_lock);
8c8bee1d 9533
a1ecaabb 9534out:
eb73c1b7
MX		 9535 list_for_each_entry_safe(work, next, &works, list) {
9536 list_del_init(&work->list);
9537 btrfs_wait_and_free_delalloc_work(work);
9538 }
9539
9540 if (!list_empty_careful(&splice)) {
9541 spin_lock(&root->delalloc_lock);
9542 list_splice_tail(&splice, &root->delalloc_inodes);
9543 spin_unlock(&root->delalloc_lock);
9544 }
573bfb72 9545 mutex_unlock(&root->delalloc_mutex);
eb73c1b7
MX		 9546 return ret;
9547}
1eafa6c7 9548
eb73c1b7
MX		 9549int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9550{
9551 int ret;
1eafa6c7 9552
2c21b4d7 9553 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
eb73c1b7
MX		 9554 return -EROFS;
9555
6c255e67
MX		 9556 ret = __start_delalloc_inodes(root, delay_iput, -1);
9557 if (ret > 0)
9558 ret = 0;
eb73c1b7
MX		 9559 /*
9560 * the filemap_flush will queue IO into the worker threads, but
8c8bee1d
CM		 9561 * we have to make sure the IO is actually started and that
9562 * ordered extents get created before we return
9563 */
9564 atomic_inc(&root->fs_info->async_submit_draining);
d397712b 9565 while (atomic_read(&root->fs_info->nr_async_submits) ||
771ed689 9566 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 9567 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 9568 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9569 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 9570 }
9571 atomic_dec(&root->fs_info->async_submit_draining);
eb73c1b7
MX		 9572 return ret;
9573}
9574
6c255e67
MX		 9575int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9576 int nr)
eb73c1b7
MX		 9577{
9578 struct btrfs_root *root;
9579 struct list_head splice;
9580 int ret;
9581
2c21b4d7 9582 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
eb73c1b7
MX		 9583 return -EROFS;
9584
9585 INIT_LIST_HEAD(&splice);
9586
573bfb72 9587 mutex_lock(&fs_info->delalloc_root_mutex);
eb73c1b7
MX		 9588 spin_lock(&fs_info->delalloc_root_lock);
9589 list_splice_init(&fs_info->delalloc_roots, &splice);
6c255e67 9590 while (!list_empty(&splice) && nr) {
eb73c1b7
MX		 9591 root = list_first_entry(&splice, struct btrfs_root,
9592 delalloc_root);
9593 root = btrfs_grab_fs_root(root);
9594 BUG_ON(!root);
9595 list_move_tail(&root->delalloc_root,
9596 &fs_info->delalloc_roots);
9597 spin_unlock(&fs_info->delalloc_root_lock);
9598
6c255e67 9599 ret = __start_delalloc_inodes(root, delay_iput, nr);
eb73c1b7 9600 btrfs_put_fs_root(root);
6c255e67 9601 if (ret < 0)
eb73c1b7
MX		 9602 goto out;
9603
6c255e67
MX		 9604 if (nr != -1) {
9605 nr -= ret;
9606 WARN_ON(nr < 0);
9607 }
eb73c1b7 9608 spin_lock(&fs_info->delalloc_root_lock);
8ccf6f19 9609 }
eb73c1b7 9610 spin_unlock(&fs_info->delalloc_root_lock);
1eafa6c7 9611
6c255e67 9612 ret = 0;
eb73c1b7
MX		 9613 atomic_inc(&fs_info->async_submit_draining);
9614 while (atomic_read(&fs_info->nr_async_submits) ||
9615 atomic_read(&fs_info->async_delalloc_pages)) {
9616 wait_event(fs_info->async_submit_wait,
9617 (atomic_read(&fs_info->nr_async_submits) == 0 &&
9618 atomic_read(&fs_info->async_delalloc_pages) == 0));
9619 }
9620 atomic_dec(&fs_info->async_submit_draining);
eb73c1b7 9621out:
1eafa6c7 9622 if (!list_empty_careful(&splice)) {
eb73c1b7
MX		 9623 spin_lock(&fs_info->delalloc_root_lock);
9624 list_splice_tail(&splice, &fs_info->delalloc_roots);
9625 spin_unlock(&fs_info->delalloc_root_lock);
1eafa6c7 9626 }
573bfb72 9627 mutex_unlock(&fs_info->delalloc_root_mutex);
8ccf6f19 9628 return ret;
ea8c2819
CM		 9629}
9630
39279cc3
CM		 9631static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9632 const char *symname)
9633{
9634 struct btrfs_trans_handle *trans;
9635 struct btrfs_root *root = BTRFS_I(dir)->root;
9636 struct btrfs_path *path;
9637 struct btrfs_key key;
1832a6d5 9638 struct inode *inode = NULL;
39279cc3
CM		 9639 int err;
9640 int drop_inode = 0;
9641 u64 objectid;
67871254 9642 u64 index = 0;
39279cc3
CM		 9643 int name_len;
9644 int datasize;
5f39d397 9645 unsigned long ptr;
39279cc3 9646 struct btrfs_file_extent_item *ei;
5f39d397 9647 struct extent_buffer *leaf;
39279cc3 9648
f06becc4 9649 name_len = strlen(symname);
39279cc3
CM		 9650 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9651 return -ENAMETOOLONG;
1832a6d5 9652
9ed74f2d
JB		 9653 /*
9654 * 2 items for inode item and ref
9655 * 2 items for dir items
9269d12b
FM		 9656 * 1 item for updating parent inode item
9657 * 1 item for the inline extent item
9ed74f2d
JB		 9658 * 1 item for xattr if selinux is on
9659 */
9269d12b 9660 trans = btrfs_start_transaction(root, 7);
a22285a6
YZ		 9661 if (IS_ERR(trans))
9662 return PTR_ERR(trans);
1832a6d5 9663
581bb050
LZ		 9664 err = btrfs_find_free_ino(root, &objectid);
9665 if (err)
9666 goto out_unlock;
9667
aec7477b 9668 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 9669 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 9670 S_IFLNK|S_IRWXUGO, &index);
7cf96da3
TI		 9671 if (IS_ERR(inode)) {
9672 err = PTR_ERR(inode);
39279cc3 9673 goto out_unlock;
7cf96da3 9674 }
39279cc3 9675
ad19db71
CS		 9676 /*
9677 * If the active LSM wants to access the inode during
9678 * d_instantiate it needs these. Smack checks to see
9679 * if the filesystem supports xattrs by looking at the
9680 * ops vector.
9681 */
9682 inode->i_fop = &btrfs_file_operations;
9683 inode->i_op = &btrfs_file_inode_operations;
b0d5d10f 9684 inode->i_mapping->a_ops = &btrfs_aops;
b0d5d10f
CM		 9685 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9686
9687 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9688 if (err)
9689 goto out_unlock_inode;
ad19db71 9690
39279cc3 9691 path = btrfs_alloc_path();
d8926bb3
MF		 9692 if (!path) {
9693 err = -ENOMEM;
b0d5d10f 9694 goto out_unlock_inode;
d8926bb3 9695 }
33345d01 9696 key.objectid = btrfs_ino(inode);
39279cc3 9697 key.offset = 0;
962a298f 9698 key.type = BTRFS_EXTENT_DATA_KEY;
39279cc3
CM		 9699 datasize = btrfs_file_extent_calc_inline_size(name_len);
9700 err = btrfs_insert_empty_item(trans, root, path, &key,
9701 datasize);
54aa1f4d 9702 if (err) {
b0839166 9703 btrfs_free_path(path);
b0d5d10f 9704 goto out_unlock_inode;
54aa1f4d 9705 }
5f39d397
CM		 9706 leaf = path->nodes[0];
9707 ei = btrfs_item_ptr(leaf, path->slots[0],
9708 struct btrfs_file_extent_item);
9709 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9710 btrfs_set_file_extent_type(leaf, ei,
39279cc3 9711 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 9712 btrfs_set_file_extent_encryption(leaf, ei, 0);
9713 btrfs_set_file_extent_compression(leaf, ei, 0);
9714 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9715 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9716
39279cc3 9717 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 9718 write_extent_buffer(leaf, symname, ptr, name_len);
9719 btrfs_mark_buffer_dirty(leaf);
39279cc3 9720 btrfs_free_path(path);
5f39d397 9721
39279cc3 9722 inode->i_op = &btrfs_symlink_inode_operations;
21fc61c7 9723 inode_nohighmem(inode);
39279cc3 9724 inode->i_mapping->a_ops = &btrfs_symlink_aops;
d899e052 9725 inode_set_bytes(inode, name_len);
f06becc4 9726 btrfs_i_size_write(inode, name_len);
54aa1f4d 9727 err = btrfs_update_inode(trans, root, inode);
d50866d0
FM		 9728 /*
9729 * Last step, add directory indexes for our symlink inode. This is the
9730 * last step to avoid extra cleanup of these indexes if an error happens
9731 * elsewhere above.
9732 */
9733 if (!err)
9734 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
b0d5d10f 9735 if (err) {
54aa1f4d 9736 drop_inode = 1;
b0d5d10f
CM		 9737 goto out_unlock_inode;
9738 }
9739
9740 unlock_new_inode(inode);
9741 d_instantiate(dentry, inode);
39279cc3
CM		 9742
9743out_unlock:
7ad85bb7 9744 btrfs_end_transaction(trans, root);
39279cc3
CM		 9745 if (drop_inode) {
9746 inode_dec_link_count(inode);
9747 iput(inode);
9748 }
b53d3f5d 9749 btrfs_btree_balance_dirty(root);
39279cc3 9750 return err;
b0d5d10f
CM		 9751
9752out_unlock_inode:
9753 drop_inode = 1;
9754 unlock_new_inode(inode);
9755 goto out_unlock;
39279cc3 9756}
16432985 9757
0af3d00b
JB		 9758static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9759 u64 start, u64 num_bytes, u64 min_size,
9760 loff_t actual_len, u64 *alloc_hint,
9761 struct btrfs_trans_handle *trans)
d899e052 9762{
5dc562c5
JB		 9763 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9764 struct extent_map *em;
d899e052
YZ		 9765 struct btrfs_root *root = BTRFS_I(inode)->root;
9766 struct btrfs_key ins;
d899e052 9767 u64 cur_offset = start;
55a61d1d 9768 u64 i_size;
154ea289 9769 u64 cur_bytes;
0b670dc4 9770 u64 last_alloc = (u64)-1;
d899e052 9771 int ret = 0;
0af3d00b 9772 bool own_trans = true;
d899e052 9773
0af3d00b
JB		 9774 if (trans)
9775 own_trans = false;
d899e052 9776 while (num_bytes > 0) {
0af3d00b
JB		 9777 if (own_trans) {
9778 trans = btrfs_start_transaction(root, 3);
9779 if (IS_ERR(trans)) {
9780 ret = PTR_ERR(trans);
9781 break;
9782 }
5a303d5d
YZ		 9783 }
9784
ee22184b 9785 cur_bytes = min_t(u64, num_bytes, SZ_256M);
154ea289 9786 cur_bytes = max(cur_bytes, min_size);
0b670dc4
JB		 9787 /*
9788 * If we are severely fragmented we could end up with really
9789 * small allocations, so if the allocator is returning small
9790 * chunks lets make its job easier by only searching for those
9791 * sized chunks.
9792 */
9793 cur_bytes = min(cur_bytes, last_alloc);
00361589 9794 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
e570fd27 9795 *alloc_hint, &ins, 1, 0);
5a303d5d 9796 if (ret) {
0af3d00b
JB		 9797 if (own_trans)
9798 btrfs_end_transaction(trans, root);
a22285a6 9799 break;
d899e052 9800 }
5a303d5d 9801
0b670dc4 9802 last_alloc = ins.offset;
d899e052
YZ		 9803 ret = insert_reserved_file_extent(trans, inode,
9804 cur_offset, ins.objectid,
9805 ins.offset, ins.offset,
920bbbfb 9806 ins.offset, 0, 0, 0,
d899e052 9807 BTRFS_FILE_EXTENT_PREALLOC);
79787eaa 9808 if (ret) {
857cc2fc 9809 btrfs_free_reserved_extent(root, ins.objectid,
e570fd27 9810 ins.offset, 0);
79787eaa
JM		 9811 btrfs_abort_transaction(trans, root, ret);
9812 if (own_trans)
9813 btrfs_end_transaction(trans, root);
9814 break;
9815 }
31193213 9816
a1ed835e
CM		 9817 btrfs_drop_extent_cache(inode, cur_offset,
9818 cur_offset + ins.offset -1, 0);
5a303d5d 9819
5dc562c5
JB		 9820 em = alloc_extent_map();
9821 if (!em) {
9822 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9823 &BTRFS_I(inode)->runtime_flags);
9824 goto next;
9825 }
9826
9827 em->start = cur_offset;
9828 em->orig_start = cur_offset;
9829 em->len = ins.offset;
9830 em->block_start = ins.objectid;
9831 em->block_len = ins.offset;
b4939680 9832 em->orig_block_len = ins.offset;
cc95bef6 9833 em->ram_bytes = ins.offset;
5dc562c5
JB		 9834 em->bdev = root->fs_info->fs_devices->latest_bdev;
9835 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9836 em->generation = trans->transid;
9837
9838 while (1) {
9839 write_lock(&em_tree->lock);
09a2a8f9 9840 ret = add_extent_mapping(em_tree, em, 1);
5dc562c5
JB		 9841 write_unlock(&em_tree->lock);
9842 if (ret != -EEXIST)
9843 break;
9844 btrfs_drop_extent_cache(inode, cur_offset,
9845 cur_offset + ins.offset - 1,
9846 0);
9847 }
9848 free_extent_map(em);
9849next:
d899e052
YZ		 9850 num_bytes -= ins.offset;
9851 cur_offset += ins.offset;
efa56464 9852 *alloc_hint = ins.objectid + ins.offset;
5a303d5d 9853
0c4d2d95 9854 inode_inc_iversion(inode);
04b285f3 9855 inode->i_ctime = current_fs_time(inode->i_sb);
6cbff00f 9856 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
d899e052 9857 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
efa56464
YZ		 9858 (actual_len > inode->i_size) &&
9859 (cur_offset > inode->i_size)) {
d1ea6a61 9860 if (cur_offset > actual_len)
55a61d1d 9861 i_size = actual_len;
d1ea6a61 9862 else
55a61d1d
JB		 9863 i_size = cur_offset;
9864 i_size_write(inode, i_size);
9865 btrfs_ordered_update_i_size(inode, i_size, NULL);
5a303d5d
YZ		 9866 }
9867
d899e052 9868 ret = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 9869
9870 if (ret) {
9871 btrfs_abort_transaction(trans, root, ret);
9872 if (own_trans)
9873 btrfs_end_transaction(trans, root);
9874 break;
9875 }
d899e052 9876
0af3d00b
JB		 9877 if (own_trans)
9878 btrfs_end_transaction(trans, root);
5a303d5d 9879 }
d899e052
YZ		 9880 return ret;
9881}
9882
0af3d00b
JB		 9883int btrfs_prealloc_file_range(struct inode *inode, int mode,
9884 u64 start, u64 num_bytes, u64 min_size,
9885 loff_t actual_len, u64 *alloc_hint)
9886{
9887 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9888 min_size, actual_len, alloc_hint,
9889 NULL);
9890}
9891
9892int btrfs_prealloc_file_range_trans(struct inode *inode,
9893 struct btrfs_trans_handle *trans, int mode,
9894 u64 start, u64 num_bytes, u64 min_size,
9895 loff_t actual_len, u64 *alloc_hint)
9896{
9897 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9898 min_size, actual_len, alloc_hint, trans);
9899}
9900
e6dcd2dc
CM		 9901static int btrfs_set_page_dirty(struct page *page)
9902{
e6dcd2dc
CM		 9903 return __set_page_dirty_nobuffers(page);
9904}
9905
10556cb2 9906static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd 9907{
b83cc969 9908 struct btrfs_root *root = BTRFS_I(inode)->root;
cb6db4e5 9909 umode_t mode = inode->i_mode;
b83cc969 9910
cb6db4e5
JM		 9911 if (mask & MAY_WRITE &&
9912 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9913 if (btrfs_root_readonly(root))
9914 return -EROFS;
9915 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9916 return -EACCES;
9917 }
2830ba7f 9918 return generic_permission(inode, mask);
fdebe2bd 9919}
39279cc3 9920
ef3b9af5
FM		 9921static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9922{
9923 struct btrfs_trans_handle *trans;
9924 struct btrfs_root *root = BTRFS_I(dir)->root;
9925 struct inode *inode = NULL;
9926 u64 objectid;
9927 u64 index;
9928 int ret = 0;
9929
9930 /*
9931 * 5 units required for adding orphan entry
9932 */
9933 trans = btrfs_start_transaction(root, 5);
9934 if (IS_ERR(trans))
9935 return PTR_ERR(trans);
9936
9937 ret = btrfs_find_free_ino(root, &objectid);
9938 if (ret)
9939 goto out;
9940
9941 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9942 btrfs_ino(dir), objectid, mode, &index);
9943 if (IS_ERR(inode)) {
9944 ret = PTR_ERR(inode);
9945 inode = NULL;
9946 goto out;
9947 }
9948
ef3b9af5
FM		 9949 inode->i_fop = &btrfs_file_operations;
9950 inode->i_op = &btrfs_file_inode_operations;
9951
9952 inode->i_mapping->a_ops = &btrfs_aops;
ef3b9af5
FM		 9953 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9954
b0d5d10f
CM		 9955 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9956 if (ret)
9957 goto out_inode;
9958
9959 ret = btrfs_update_inode(trans, root, inode);
9960 if (ret)
9961 goto out_inode;
ef3b9af5
FM		 9962 ret = btrfs_orphan_add(trans, inode);
9963 if (ret)
b0d5d10f 9964 goto out_inode;
ef3b9af5 9965
5762b5c9
FM		 9966 /*
9967 * We set number of links to 0 in btrfs_new_inode(), and here we set
9968 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9969 * through:
9970 *
9971 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9972 */
9973 set_nlink(inode, 1);
b0d5d10f 9974 unlock_new_inode(inode);
ef3b9af5
FM		 9975 d_tmpfile(dentry, inode);
9976 mark_inode_dirty(inode);
9977
9978out:
9979 btrfs_end_transaction(trans, root);
9980 if (ret)
9981 iput(inode);
9982 btrfs_balance_delayed_items(root);
9983 btrfs_btree_balance_dirty(root);
ef3b9af5 9984 return ret;
b0d5d10f
CM		 9985
9986out_inode:
9987 unlock_new_inode(inode);
9988 goto out;
9989
ef3b9af5
FM		 9990}
9991
b38ef71c
FM		 9992/* Inspired by filemap_check_errors() */
9993int btrfs_inode_check_errors(struct inode *inode)
9994{
9995 int ret = 0;
9996
9997 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
9998 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
9999 ret = -ENOSPC;
10000 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
10001 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
10002 ret = -EIO;
10003
10004 return ret;
10005}
10006
6e1d5dcc 10007static const struct inode_operations btrfs_dir_inode_operations = {
3394e160 10008 .getattr = btrfs_getattr,
39279cc3
CM		 10009 .lookup = btrfs_lookup,
10010 .create = btrfs_create,
10011 .unlink = btrfs_unlink,
10012 .link = btrfs_link,
10013 .mkdir = btrfs_mkdir,
10014 .rmdir = btrfs_rmdir,
80ace85c 10015 .rename2 = btrfs_rename2,
39279cc3
CM		 10016 .symlink = btrfs_symlink,
10017 .setattr = btrfs_setattr,
618e21d5 10018 .mknod = btrfs_mknod,
95819c05 10019 .setxattr = btrfs_setxattr,
9172abbc 10020 .getxattr = generic_getxattr,
5103e947 10021 .listxattr = btrfs_listxattr,
95819c05 10022 .removexattr = btrfs_removexattr,
fdebe2bd 10023 .permission = btrfs_permission,
4e34e719 10024 .get_acl = btrfs_get_acl,
996a710d 10025 .set_acl = btrfs_set_acl,
93fd63c2 10026 .update_time = btrfs_update_time,
ef3b9af5 10027 .tmpfile = btrfs_tmpfile,
39279cc3 10028};
6e1d5dcc 10029static const struct inode_operations btrfs_dir_ro_inode_operations = {
39279cc3 10030 .lookup = btrfs_lookup,
fdebe2bd 10031 .permission = btrfs_permission,
4e34e719 10032 .get_acl = btrfs_get_acl,
996a710d 10033 .set_acl = btrfs_set_acl,
93fd63c2 10034 .update_time = btrfs_update_time,
39279cc3 10035};
76dda93c 10036
828c0950 10037static const struct file_operations btrfs_dir_file_operations = {
39279cc3
CM		 10038 .llseek = generic_file_llseek,
10039 .read = generic_read_dir,
9cdda8d3 10040 .iterate = btrfs_real_readdir,
34287aa3 10041 .unlocked_ioctl = btrfs_ioctl,
39279cc3 10042#ifdef CONFIG_COMPAT
34287aa3 10043 .compat_ioctl = btrfs_ioctl,
39279cc3 10044#endif
6bf13c0c 10045 .release = btrfs_release_file,
e02119d5 10046 .fsync = btrfs_sync_file,
39279cc3
CM		 10047};
10048
20e5506b 10049static const struct extent_io_ops btrfs_extent_io_ops = {
07157aac 10050 .fill_delalloc = run_delalloc_range,
065631f6 10051 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 10052 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 10053 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 10054 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 10055 .writepage_start_hook = btrfs_writepage_start_hook,
b0c68f8b
CM		 10056 .set_bit_hook = btrfs_set_bit_hook,
10057 .clear_bit_hook = btrfs_clear_bit_hook,
9ed74f2d
JB		 10058 .merge_extent_hook = btrfs_merge_extent_hook,
10059 .split_extent_hook = btrfs_split_extent_hook,
07157aac
CM		 10060};
10061
35054394
CM		 10062/*
10063 * btrfs doesn't support the bmap operation because swapfiles
10064 * use bmap to make a mapping of extents in the file. They assume
10065 * these extents won't change over the life of the file and they
10066 * use the bmap result to do IO directly to the drive.
10067 *
10068 * the btrfs bmap call would return logical addresses that aren't
10069 * suitable for IO and they also will change frequently as COW
10070 * operations happen. So, swapfile + btrfs == corruption.
10071 *
10072 * For now we're avoiding this by dropping bmap.
10073 */
7f09410b 10074static const struct address_space_operations btrfs_aops = {
39279cc3
CM		 10075 .readpage = btrfs_readpage,
10076 .writepage = btrfs_writepage,
b293f02e 10077 .writepages = btrfs_writepages,
3ab2fb5a 10078 .readpages = btrfs_readpages,
16432985 10079 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 10080 .invalidatepage = btrfs_invalidatepage,
10081 .releasepage = btrfs_releasepage,
e6dcd2dc 10082 .set_page_dirty = btrfs_set_page_dirty,
465fdd97 10083 .error_remove_page = generic_error_remove_page,
39279cc3
CM		 10084};
10085
7f09410b 10086static const struct address_space_operations btrfs_symlink_aops = {
39279cc3
CM		 10087 .readpage = btrfs_readpage,
10088 .writepage = btrfs_writepage,
2bf5a725
CM		 10089 .invalidatepage = btrfs_invalidatepage,
10090 .releasepage = btrfs_releasepage,
39279cc3
CM		 10091};
10092
6e1d5dcc 10093static const struct inode_operations btrfs_file_inode_operations = {
39279cc3
CM		 10094 .getattr = btrfs_getattr,
10095 .setattr = btrfs_setattr,
95819c05 10096 .setxattr = btrfs_setxattr,
9172abbc 10097 .getxattr = generic_getxattr,
5103e947 10098 .listxattr = btrfs_listxattr,
95819c05 10099 .removexattr = btrfs_removexattr,
fdebe2bd 10100 .permission = btrfs_permission,
1506fcc8 10101 .fiemap = btrfs_fiemap,
4e34e719 10102 .get_acl = btrfs_get_acl,
996a710d 10103 .set_acl = btrfs_set_acl,
e41f941a 10104 .update_time = btrfs_update_time,
39279cc3 10105};
6e1d5dcc 10106static const struct inode_operations btrfs_special_inode_operations = {
618e21d5
JB		 10107 .getattr = btrfs_getattr,
10108 .setattr = btrfs_setattr,
fdebe2bd 10109 .permission = btrfs_permission,
95819c05 10110 .setxattr = btrfs_setxattr,
9172abbc 10111 .getxattr = generic_getxattr,
33268eaf 10112 .listxattr = btrfs_listxattr,
95819c05 10113 .removexattr = btrfs_removexattr,
4e34e719 10114 .get_acl = btrfs_get_acl,
996a710d 10115 .set_acl = btrfs_set_acl,
e41f941a 10116 .update_time = btrfs_update_time,
618e21d5 10117};
6e1d5dcc 10118static const struct inode_operations btrfs_symlink_inode_operations = {
39279cc3 10119 .readlink = generic_readlink,
6b255391 10120 .get_link = page_get_link,
f209561a 10121 .getattr = btrfs_getattr,
22c44fe6 10122 .setattr = btrfs_setattr,
fdebe2bd 10123 .permission = btrfs_permission,
0279b4cd 10124 .setxattr = btrfs_setxattr,
9172abbc 10125 .getxattr = generic_getxattr,
0279b4cd
JO		 10126 .listxattr = btrfs_listxattr,
10127 .removexattr = btrfs_removexattr,
e41f941a 10128 .update_time = btrfs_update_time,
39279cc3 10129};
76dda93c 10130
82d339d9 10131const struct dentry_operations btrfs_dentry_operations = {
76dda93c 10132 .d_delete = btrfs_dentry_delete,
b4aff1f8 10133 .d_release = btrfs_dentry_release,
76dda93c 10134};
Linux kernel - Deliverables
This page took 2.094335 seconds and 5 git commands to generate.