projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: fallocate: Work with sectorsized blocks
[linux-2.6-block.git] / fs / btrfs / inode.c
CommitLineData
6cbd5570
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
8f18cf13 19#include <linux/kernel.h>
065631f6 20#include <linux/bio.h>
39279cc3 21#include <linux/buffer_head.h>
f2eb0a24 22#include <linux/file.h>
39279cc3
CM		 23#include <linux/fs.h>
24#include <linux/pagemap.h>
25#include <linux/highmem.h>
26#include <linux/time.h>
27#include <linux/init.h>
28#include <linux/string.h>
39279cc3
CM		 29#include <linux/backing-dev.h>
30#include <linux/mpage.h>
31#include <linux/swap.h>
32#include <linux/writeback.h>
33#include <linux/statfs.h>
34#include <linux/compat.h>
9ebefb18 35#include <linux/bit_spinlock.h>
5103e947 36#include <linux/xattr.h>
33268eaf 37#include <linux/posix_acl.h>
d899e052 38#include <linux/falloc.h>
5a0e3ad6 39#include <linux/slab.h>
7a36ddec 40#include <linux/ratelimit.h>
22c44fe6 41#include <linux/mount.h>
55e301fd 42#include <linux/btrfs.h>
53b381b3 43#include <linux/blkdev.h>
f23b5a59 44#include <linux/posix_acl_xattr.h>
e2e40f2c 45#include <linux/uio.h>
39279cc3
CM		 46#include "ctree.h"
47#include "disk-io.h"
48#include "transaction.h"
49#include "btrfs_inode.h"
39279cc3 50#include "print-tree.h"
e6dcd2dc 51#include "ordered-data.h"
95819c05 52#include "xattr.h"
e02119d5 53#include "tree-log.h"
4a54c8c1 54#include "volumes.h"
c8b97818 55#include "compression.h"
b4ce94de 56#include "locking.h"
dc89e982 57#include "free-space-cache.h"
581bb050 58#include "inode-map.h"
38c227d8 59#include "backref.h"
f23b5a59 60#include "hash.h"
63541927 61#include "props.h"
31193213 62#include "qgroup.h"
39279cc3
CM		 63
64struct btrfs_iget_args {
90d3e592 65 struct btrfs_key *location;
39279cc3
CM		 66 struct btrfs_root *root;
67};
68
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
JB		 3548 if (!xattr_access) {
3549 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3550 strlen(POSIX_ACL_XATTR_ACCESS));
3551 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3552 strlen(POSIX_ACL_XATTR_DEFAULT));
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;
3773 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3774 break;
618e21d5 3775 default:
0279b4cd 3776 inode->i_op = &btrfs_special_inode_operations;
618e21d5
JB		 3777 init_special_inode(inode, inode->i_mode, rdev);
3778 break;
39279cc3 3779 }
6cbff00f
CH		 3780
3781 btrfs_update_iflags(inode);
39279cc3
CM		 3782 return;
3783
3784make_bad:
39279cc3 3785 btrfs_free_path(path);
39279cc3
CM		 3786 make_bad_inode(inode);
3787}
3788
d352ac68
CM		 3789/*
3790 * given a leaf and an inode, copy the inode fields into the leaf
3791 */
e02119d5
CM		 3792static void fill_inode_item(struct btrfs_trans_handle *trans,
3793 struct extent_buffer *leaf,
5f39d397 3794 struct btrfs_inode_item *item,
39279cc3
CM		 3795 struct inode *inode)
3796{
51fab693
LB		 3797 struct btrfs_map_token token;
3798
3799 btrfs_init_map_token(&token);
5f39d397 3800
51fab693
LB		 3801 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3802 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3803 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3804 &token);
3805 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3806 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
5f39d397 3807
a937b979 3808 btrfs_set_token_timespec_sec(leaf, &item->atime,
51fab693 3809 inode->i_atime.tv_sec, &token);
a937b979 3810 btrfs_set_token_timespec_nsec(leaf, &item->atime,
51fab693 3811 inode->i_atime.tv_nsec, &token);
5f39d397 3812
a937b979 3813 btrfs_set_token_timespec_sec(leaf, &item->mtime,
51fab693 3814 inode->i_mtime.tv_sec, &token);
a937b979 3815 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
51fab693 3816 inode->i_mtime.tv_nsec, &token);
5f39d397 3817
a937b979 3818 btrfs_set_token_timespec_sec(leaf, &item->ctime,
51fab693 3819 inode->i_ctime.tv_sec, &token);
a937b979 3820 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
51fab693 3821 inode->i_ctime.tv_nsec, &token);
5f39d397 3822
9cc97d64 3823 btrfs_set_token_timespec_sec(leaf, &item->otime,
3824 BTRFS_I(inode)->i_otime.tv_sec, &token);
3825 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3826 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3827
51fab693
LB		 3828 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3829 &token);
3830 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3831 &token);
3832 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3833 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3834 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3835 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3836 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
39279cc3
CM		 3837}
3838
d352ac68
CM		 3839/*
3840 * copy everything in the in-memory inode into the btree.
3841 */
2115133f 3842static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
d397712b 3843 struct btrfs_root *root, struct inode *inode)
39279cc3
CM		 3844{
3845 struct btrfs_inode_item *inode_item;
3846 struct btrfs_path *path;
5f39d397 3847 struct extent_buffer *leaf;
39279cc3
CM		 3848 int ret;
3849
3850 path = btrfs_alloc_path();
16cdcec7
MX		 3851 if (!path)
3852 return -ENOMEM;
3853
b9473439 3854 path->leave_spinning = 1;
16cdcec7
MX		 3855 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3856 1);
39279cc3
CM		 3857 if (ret) {
3858 if (ret > 0)
3859 ret = -ENOENT;
3860 goto failed;
3861 }
3862
5f39d397
CM		 3863 leaf = path->nodes[0];
3864 inode_item = btrfs_item_ptr(leaf, path->slots[0],
16cdcec7 3865 struct btrfs_inode_item);
39279cc3 3866
e02119d5 3867 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 3868 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 3869 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 3870 ret = 0;
3871failed:
39279cc3
CM		 3872 btrfs_free_path(path);
3873 return ret;
3874}
3875
2115133f
CM		 3876/*
3877 * copy everything in the in-memory inode into the btree.
3878 */
3879noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3880 struct btrfs_root *root, struct inode *inode)
3881{
3882 int ret;
3883
3884 /*
3885 * If the inode is a free space inode, we can deadlock during commit
3886 * if we put it into the delayed code.
3887 *
3888 * The data relocation inode should also be directly updated
3889 * without delay
3890 */
83eea1f1 3891 if (!btrfs_is_free_space_inode(inode)
1d52c78a
JB		 3892 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3893 && !root->fs_info->log_root_recovering) {
8ea05e3a
AB		 3894 btrfs_update_root_times(trans, root);
3895
2115133f
CM		 3896 ret = btrfs_delayed_update_inode(trans, root, inode);
3897 if (!ret)
3898 btrfs_set_inode_last_trans(trans, inode);
3899 return ret;
3900 }
3901
3902 return btrfs_update_inode_item(trans, root, inode);
3903}
3904
be6aef60
JB		 3905noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3906 struct btrfs_root *root,
3907 struct inode *inode)
2115133f
CM		 3908{
3909 int ret;
3910
3911 ret = btrfs_update_inode(trans, root, inode);
3912 if (ret == -ENOSPC)
3913 return btrfs_update_inode_item(trans, root, inode);
3914 return ret;
3915}
3916
d352ac68
CM		 3917/*
3918 * unlink helper that gets used here in inode.c and in the tree logging
3919 * recovery code. It remove a link in a directory with a given name, and
3920 * also drops the back refs in the inode to the directory
3921 */
92986796
AV		 3922static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3923 struct btrfs_root *root,
3924 struct inode *dir, struct inode *inode,
3925 const char *name, int name_len)
39279cc3
CM		 3926{
3927 struct btrfs_path *path;
39279cc3 3928 int ret = 0;
5f39d397 3929 struct extent_buffer *leaf;
39279cc3 3930 struct btrfs_dir_item *di;
5f39d397 3931 struct btrfs_key key;
aec7477b 3932 u64 index;
33345d01
LZ		 3933 u64 ino = btrfs_ino(inode);
3934 u64 dir_ino = btrfs_ino(dir);
39279cc3
CM		 3935
3936 path = btrfs_alloc_path();
54aa1f4d
CM		 3937 if (!path) {
3938 ret = -ENOMEM;
554233a6 3939 goto out;
54aa1f4d
CM		 3940 }
3941
b9473439 3942 path->leave_spinning = 1;
33345d01 3943 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
39279cc3
CM		 3944 name, name_len, -1);
3945 if (IS_ERR(di)) {
3946 ret = PTR_ERR(di);
3947 goto err;
3948 }
3949 if (!di) {
3950 ret = -ENOENT;
3951 goto err;
3952 }
5f39d397
CM		 3953 leaf = path->nodes[0];
3954 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 3955 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 3956 if (ret)
3957 goto err;
b3b4aa74 3958 btrfs_release_path(path);
39279cc3 3959
67de1176
MX		 3960 /*
3961 * If we don't have dir index, we have to get it by looking up
3962 * the inode ref, since we get the inode ref, remove it directly,
3963 * it is unnecessary to do delayed deletion.
3964 *
3965 * But if we have dir index, needn't search inode ref to get it.
3966 * Since the inode ref is close to the inode item, it is better
3967 * that we delay to delete it, and just do this deletion when
3968 * we update the inode item.
3969 */
3970 if (BTRFS_I(inode)->dir_index) {
3971 ret = btrfs_delayed_delete_inode_ref(inode);
3972 if (!ret) {
3973 index = BTRFS_I(inode)->dir_index;
3974 goto skip_backref;
3975 }
3976 }
3977
33345d01
LZ		 3978 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3979 dir_ino, &index);
aec7477b 3980 if (ret) {
c2cf52eb
SK		 3981 btrfs_info(root->fs_info,
3982 "failed to delete reference to %.*s, inode %llu parent %llu",
c1c9ff7c 3983 name_len, name, ino, dir_ino);
79787eaa 3984 btrfs_abort_transaction(trans, root, ret);
aec7477b
JB		 3985 goto err;
3986 }
67de1176 3987skip_backref:
16cdcec7 3988 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 3989 if (ret) {
3990 btrfs_abort_transaction(trans, root, ret);
39279cc3 3991 goto err;
79787eaa 3992 }
39279cc3 3993
e02119d5 3994 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
33345d01 3995 inode, dir_ino);
79787eaa
JM		 3996 if (ret != 0 && ret != -ENOENT) {
3997 btrfs_abort_transaction(trans, root, ret);
3998 goto err;
3999 }
e02119d5
CM		 4000
4001 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4002 dir, index);
6418c961
CM		 4003 if (ret == -ENOENT)
4004 ret = 0;
d4e3991b
ZB		 4005 else if (ret)
4006 btrfs_abort_transaction(trans, root, ret);
39279cc3
CM		 4007err:
4008 btrfs_free_path(path);
e02119d5
CM		 4009 if (ret)
4010 goto out;
4011
4012 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95
JB		 4013 inode_inc_iversion(inode);
4014 inode_inc_iversion(dir);
e02119d5 4015 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
b9959295 4016 ret = btrfs_update_inode(trans, root, dir);
e02119d5 4017out:
39279cc3
CM		 4018 return ret;
4019}
4020
92986796
AV		 4021int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4022 struct btrfs_root *root,
4023 struct inode *dir, struct inode *inode,
4024 const char *name, int name_len)
4025{
4026 int ret;
4027 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4028 if (!ret) {
8b558c5f 4029 drop_nlink(inode);
92986796
AV		 4030 ret = btrfs_update_inode(trans, root, inode);
4031 }
4032 return ret;
4033}
39279cc3 4034
a22285a6
YZ		 4035/*
4036 * helper to start transaction for unlink and rmdir.
4037 *
d52be818
JB		 4038 * unlink and rmdir are special in btrfs, they do not always free space, so
4039 * if we cannot make our reservations the normal way try and see if there is
4040 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4041 * allow the unlink to occur.
a22285a6 4042 */
d52be818 4043static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4df27c4d 4044{
a22285a6 4045 struct btrfs_root *root = BTRFS_I(dir)->root;
4df27c4d 4046
e70bea5f
JB		 4047 /*
4048 * 1 for the possible orphan item
4049 * 1 for the dir item
4050 * 1 for the dir index
4051 * 1 for the inode ref
e70bea5f
JB		 4052 * 1 for the inode
4053 */
8eab77ff 4054 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
a22285a6
YZ		 4055}
4056
4057static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4058{
4059 struct btrfs_root *root = BTRFS_I(dir)->root;
4060 struct btrfs_trans_handle *trans;
2b0143b5 4061 struct inode *inode = d_inode(dentry);
a22285a6 4062 int ret;
a22285a6 4063
d52be818 4064 trans = __unlink_start_trans(dir);
a22285a6
YZ		 4065 if (IS_ERR(trans))
4066 return PTR_ERR(trans);
5f39d397 4067
2b0143b5 4068 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
12fcfd22 4069
2b0143b5 4070 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
e02119d5 4071 dentry->d_name.name, dentry->d_name.len);
b532402e
TI		 4072 if (ret)
4073 goto out;
7b128766 4074
a22285a6 4075 if (inode->i_nlink == 0) {
7b128766 4076 ret = btrfs_orphan_add(trans, inode);
b532402e
TI		 4077 if (ret)
4078 goto out;
a22285a6 4079 }
7b128766 4080
b532402e 4081out:
d52be818 4082 btrfs_end_transaction(trans, root);
b53d3f5d 4083 btrfs_btree_balance_dirty(root);
39279cc3
CM		 4084 return ret;
4085}
4086
4df27c4d
YZ		 4087int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4088 struct btrfs_root *root,
4089 struct inode *dir, u64 objectid,
4090 const char *name, int name_len)
4091{
4092 struct btrfs_path *path;
4093 struct extent_buffer *leaf;
4094 struct btrfs_dir_item *di;
4095 struct btrfs_key key;
4096 u64 index;
4097 int ret;
33345d01 4098 u64 dir_ino = btrfs_ino(dir);
4df27c4d
YZ		 4099
4100 path = btrfs_alloc_path();
4101 if (!path)
4102 return -ENOMEM;
4103
33345d01 4104 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4df27c4d 4105 name, name_len, -1);
79787eaa
JM		 4106 if (IS_ERR_OR_NULL(di)) {
4107 if (!di)
4108 ret = -ENOENT;
4109 else
4110 ret = PTR_ERR(di);
4111 goto out;
4112 }
4df27c4d
YZ		 4113
4114 leaf = path->nodes[0];
4115 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4116 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4117 ret = btrfs_delete_one_dir_name(trans, root, path, di);
79787eaa
JM		 4118 if (ret) {
4119 btrfs_abort_transaction(trans, root, ret);
4120 goto out;
4121 }
b3b4aa74 4122 btrfs_release_path(path);
4df27c4d
YZ		 4123
4124 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4125 objectid, root->root_key.objectid,
33345d01 4126 dir_ino, &index, name, name_len);
4df27c4d 4127 if (ret < 0) {
79787eaa
JM		 4128 if (ret != -ENOENT) {
4129 btrfs_abort_transaction(trans, root, ret);
4130 goto out;
4131 }
33345d01 4132 di = btrfs_search_dir_index_item(root, path, dir_ino,
4df27c4d 4133 name, name_len);
79787eaa
JM		 4134 if (IS_ERR_OR_NULL(di)) {
4135 if (!di)
4136 ret = -ENOENT;
4137 else
4138 ret = PTR_ERR(di);
4139 btrfs_abort_transaction(trans, root, ret);
4140 goto out;
4141 }
4df27c4d
YZ		 4142
4143 leaf = path->nodes[0];
4144 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
b3b4aa74 4145 btrfs_release_path(path);
4df27c4d
YZ		 4146 index = key.offset;
4147 }
945d8962 4148 btrfs_release_path(path);
4df27c4d 4149
16cdcec7 4150 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 4151 if (ret) {
4152 btrfs_abort_transaction(trans, root, ret);
4153 goto out;
4154 }
4df27c4d
YZ		 4155
4156 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95 4157 inode_inc_iversion(dir);
4df27c4d 4158 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
5a24e84c 4159 ret = btrfs_update_inode_fallback(trans, root, dir);
79787eaa
JM		 4160 if (ret)
4161 btrfs_abort_transaction(trans, root, ret);
4162out:
71d7aed0 4163 btrfs_free_path(path);
79787eaa 4164 return ret;
4df27c4d
YZ		 4165}
4166
39279cc3
CM		 4167static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4168{
2b0143b5 4169 struct inode *inode = d_inode(dentry);
1832a6d5 4170 int err = 0;
39279cc3 4171 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 4172 struct btrfs_trans_handle *trans;
39279cc3 4173
b3ae244e 4174 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
134d4512 4175 return -ENOTEMPTY;
b3ae244e
DS		 4176 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4177 return -EPERM;
134d4512 4178
d52be818 4179 trans = __unlink_start_trans(dir);
a22285a6 4180 if (IS_ERR(trans))
5df6a9f6 4181 return PTR_ERR(trans);
5df6a9f6 4182
33345d01 4183 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4df27c4d
YZ		 4184 err = btrfs_unlink_subvol(trans, root, dir,
4185 BTRFS_I(inode)->location.objectid,
4186 dentry->d_name.name,
4187 dentry->d_name.len);
4188 goto out;
4189 }
4190
7b128766
JB		 4191 err = btrfs_orphan_add(trans, inode);
4192 if (err)
4df27c4d 4193 goto out;
7b128766 4194
39279cc3 4195 /* now the directory is empty */
2b0143b5 4196 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
e02119d5 4197 dentry->d_name.name, dentry->d_name.len);
d397712b 4198 if (!err)
dbe674a9 4199 btrfs_i_size_write(inode, 0);
4df27c4d 4200out:
d52be818 4201 btrfs_end_transaction(trans, root);
b53d3f5d 4202 btrfs_btree_balance_dirty(root);
3954401f 4203
39279cc3
CM		 4204 return err;
4205}
4206
28f75a0e
CM		 4207static int truncate_space_check(struct btrfs_trans_handle *trans,
4208 struct btrfs_root *root,
4209 u64 bytes_deleted)
4210{
4211 int ret;
4212
4213 bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4214 ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4215 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4216 if (!ret)
4217 trans->bytes_reserved += bytes_deleted;
4218 return ret;
4219
4220}
4221
0305cd5f
FM		 4222static int truncate_inline_extent(struct inode *inode,
4223 struct btrfs_path *path,
4224 struct btrfs_key *found_key,
4225 const u64 item_end,
4226 const u64 new_size)
4227{
4228 struct extent_buffer *leaf = path->nodes[0];
4229 int slot = path->slots[0];
4230 struct btrfs_file_extent_item *fi;
4231 u32 size = (u32)(new_size - found_key->offset);
4232 struct btrfs_root *root = BTRFS_I(inode)->root;
4233
4234 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4235
4236 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4237 loff_t offset = new_size;
4238 loff_t page_end = ALIGN(offset, PAGE_CACHE_SIZE);
4239
4240 /*
4241 * Zero out the remaining of the last page of our inline extent,
4242 * instead of directly truncating our inline extent here - that
4243 * would be much more complex (decompressing all the data, then
4244 * compressing the truncated data, which might be bigger than
4245 * the size of the inline extent, resize the extent, etc).
4246 * We release the path because to get the page we might need to
4247 * read the extent item from disk (data not in the page cache).
4248 */
4249 btrfs_release_path(path);
9703fefe
CR		 4250 return btrfs_truncate_block(inode, offset, page_end - offset,
4251 0);
0305cd5f
FM		 4252 }
4253
4254 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4255 size = btrfs_file_extent_calc_inline_size(size);
4256 btrfs_truncate_item(root, path, size, 1);
4257
4258 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4259 inode_sub_bytes(inode, item_end + 1 - new_size);
4260
4261 return 0;
4262}
4263
39279cc3
CM		 4264/*
4265 * this can truncate away extent items, csum items and directory items.
4266 * It starts at a high offset and removes keys until it can't find
d352ac68 4267 * any higher than new_size
39279cc3
CM		 4268 *
4269 * csum items that cross the new i_size are truncated to the new size
4270 * as well.
7b128766
JB		 4271 *
4272 * min_type is the minimum key type to truncate down to. If set to 0, this
4273 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 4274 */
8082510e
YZ		 4275int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4276 struct btrfs_root *root,
4277 struct inode *inode,
4278 u64 new_size, u32 min_type)
39279cc3 4279{
39279cc3 4280 struct btrfs_path *path;
5f39d397 4281 struct extent_buffer *leaf;
39279cc3 4282 struct btrfs_file_extent_item *fi;
8082510e
YZ		 4283 struct btrfs_key key;
4284 struct btrfs_key found_key;
39279cc3 4285 u64 extent_start = 0;
db94535d 4286 u64 extent_num_bytes = 0;
5d4f98a2 4287 u64 extent_offset = 0;
39279cc3 4288 u64 item_end = 0;
c1aa4575 4289 u64 last_size = new_size;
8082510e 4290 u32 found_type = (u8)-1;
39279cc3
CM		 4291 int found_extent;
4292 int del_item;
85e21bac
CM		 4293 int pending_del_nr = 0;
4294 int pending_del_slot = 0;
179e29e4 4295 int extent_type = -1;
8082510e
YZ		 4296 int ret;
4297 int err = 0;
33345d01 4298 u64 ino = btrfs_ino(inode);
28ed1345 4299 u64 bytes_deleted = 0;
1262133b
JB		 4300 bool be_nice = 0;
4301 bool should_throttle = 0;
28f75a0e 4302 bool should_end = 0;
8082510e
YZ		 4303
4304 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
39279cc3 4305
28ed1345
CM		 4306 /*
4307 * for non-free space inodes and ref cows, we want to back off from
4308 * time to time
4309 */
4310 if (!btrfs_is_free_space_inode(inode) &&
4311 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4312 be_nice = 1;
4313
0eb0e19c
MF		 4314 path = btrfs_alloc_path();
4315 if (!path)
4316 return -ENOMEM;
e4058b54 4317 path->reada = READA_BACK;
0eb0e19c 4318
5dc562c5
JB		 4319 /*
4320 * We want to drop from the next block forward in case this new size is
4321 * not block aligned since we will be keeping the last block of the
4322 * extent just the way it is.
4323 */
27cdeb70
MX		 4324 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4325 root == root->fs_info->tree_root)
fda2832f
QW		 4326 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4327 root->sectorsize), (u64)-1, 0);
8082510e 4328
16cdcec7
MX		 4329 /*
4330 * This function is also used to drop the items in the log tree before
4331 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4332 * it is used to drop the loged items. So we shouldn't kill the delayed
4333 * items.
4334 */
4335 if (min_type == 0 && root == BTRFS_I(inode)->root)
4336 btrfs_kill_delayed_inode_items(inode);
4337
33345d01 4338 key.objectid = ino;
39279cc3 4339 key.offset = (u64)-1;
5f39d397
CM		 4340 key.type = (u8)-1;
4341
85e21bac 4342search_again:
28ed1345
CM		 4343 /*
4344 * with a 16K leaf size and 128MB extents, you can actually queue
4345 * up a huge file in a single leaf. Most of the time that
4346 * bytes_deleted is > 0, it will be huge by the time we get here
4347 */
ee22184b 4348 if (be_nice && bytes_deleted > SZ_32M) {
28ed1345
CM		 4349 if (btrfs_should_end_transaction(trans, root)) {
4350 err = -EAGAIN;
4351 goto error;
4352 }
4353 }
4354
4355
b9473439 4356 path->leave_spinning = 1;
85e21bac 4357 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8082510e
YZ		 4358 if (ret < 0) {
4359 err = ret;
4360 goto out;
4361 }
d397712b 4362
85e21bac 4363 if (ret > 0) {
e02119d5
CM		 4364 /* there are no items in the tree for us to truncate, we're
4365 * done
4366 */
8082510e
YZ		 4367 if (path->slots[0] == 0)
4368 goto out;
85e21bac
CM		 4369 path->slots[0]--;
4370 }
4371
d397712b 4372 while (1) {
39279cc3 4373 fi = NULL;
5f39d397
CM		 4374 leaf = path->nodes[0];
4375 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
962a298f 4376 found_type = found_key.type;
39279cc3 4377
33345d01 4378 if (found_key.objectid != ino)
39279cc3 4379 break;
5f39d397 4380
85e21bac 4381 if (found_type < min_type)
39279cc3
CM		 4382 break;
4383
5f39d397 4384 item_end = found_key.offset;
39279cc3 4385 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 4386 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 4387 struct btrfs_file_extent_item);
179e29e4
CM		 4388 extent_type = btrfs_file_extent_type(leaf, fi);
4389 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 4390 item_end +=
db94535d 4391 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 4392 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 4393 item_end += btrfs_file_extent_inline_len(leaf,
514ac8ad 4394 path->slots[0], fi);
39279cc3 4395 }
008630c1 4396 item_end--;
39279cc3 4397 }
8082510e
YZ		 4398 if (found_type > min_type) {
4399 del_item = 1;
4400 } else {
4401 if (item_end < new_size)
b888db2b 4402 break;
8082510e
YZ		 4403 if (found_key.offset >= new_size)
4404 del_item = 1;
4405 else
4406 del_item = 0;
39279cc3 4407 }
39279cc3 4408 found_extent = 0;
39279cc3 4409 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 4410 if (found_type != BTRFS_EXTENT_DATA_KEY)
4411 goto delete;
4412
7f4f6e0a
JB		 4413 if (del_item)
4414 last_size = found_key.offset;
4415 else
4416 last_size = new_size;
4417
179e29e4 4418 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 4419 u64 num_dec;
db94535d 4420 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
f70a9a6b 4421 if (!del_item) {
db94535d
CM		 4422 u64 orig_num_bytes =
4423 btrfs_file_extent_num_bytes(leaf, fi);
fda2832f
QW		 4424 extent_num_bytes = ALIGN(new_size -
4425 found_key.offset,
4426 root->sectorsize);
db94535d
CM		 4427 btrfs_set_file_extent_num_bytes(leaf, fi,
4428 extent_num_bytes);
4429 num_dec = (orig_num_bytes -
9069218d 4430 extent_num_bytes);
27cdeb70
MX		 4431 if (test_bit(BTRFS_ROOT_REF_COWS,
4432 &root->state) &&
4433 extent_start != 0)
a76a3cd4 4434 inode_sub_bytes(inode, num_dec);
5f39d397 4435 btrfs_mark_buffer_dirty(leaf);
39279cc3 4436 } else {
db94535d
CM		 4437 extent_num_bytes =
4438 btrfs_file_extent_disk_num_bytes(leaf,
4439 fi);
5d4f98a2
YZ		 4440 extent_offset = found_key.offset -
4441 btrfs_file_extent_offset(leaf, fi);
4442
39279cc3 4443 /* FIXME blocksize != 4096 */
9069218d 4444 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 4445 if (extent_start != 0) {
4446 found_extent = 1;
27cdeb70
MX		 4447 if (test_bit(BTRFS_ROOT_REF_COWS,
4448 &root->state))
a76a3cd4 4449 inode_sub_bytes(inode, num_dec);
e02119d5 4450 }
39279cc3 4451 }
9069218d 4452 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 4453 /*
4454 * we can't truncate inline items that have had
4455 * special encodings
4456 */
4457 if (!del_item &&
c8b97818
CM		 4458 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4459 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
514ac8ad
CM		 4460
4461 /*
0305cd5f
FM		 4462 * Need to release path in order to truncate a
4463 * compressed extent. So delete any accumulated
4464 * extent items so far.
514ac8ad 4465 */
0305cd5f
FM		 4466 if (btrfs_file_extent_compression(leaf, fi) !=
4467 BTRFS_COMPRESS_NONE && pending_del_nr) {
4468 err = btrfs_del_items(trans, root, path,
4469 pending_del_slot,
4470 pending_del_nr);
4471 if (err) {
4472 btrfs_abort_transaction(trans,
4473 root,
4474 err);
4475 goto error;
4476 }
4477 pending_del_nr = 0;
4478 }
4479
4480 err = truncate_inline_extent(inode, path,
4481 &found_key,
4482 item_end,
4483 new_size);
4484 if (err) {
4485 btrfs_abort_transaction(trans,
4486 root, err);
4487 goto error;
4488 }
27cdeb70
MX		 4489 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4490 &root->state)) {
0305cd5f 4491 inode_sub_bytes(inode, item_end + 1 - new_size);
9069218d 4492 }
39279cc3 4493 }
179e29e4 4494delete:
39279cc3 4495 if (del_item) {
85e21bac
CM		 4496 if (!pending_del_nr) {
4497 /* no pending yet, add ourselves */
4498 pending_del_slot = path->slots[0];
4499 pending_del_nr = 1;
4500 } else if (pending_del_nr &&
4501 path->slots[0] + 1 == pending_del_slot) {
4502 /* hop on the pending chunk */
4503 pending_del_nr++;
4504 pending_del_slot = path->slots[0];
4505 } else {
d397712b 4506 BUG();
85e21bac 4507 }
39279cc3
CM		 4508 } else {
4509 break;
4510 }
28f75a0e
CM		 4511 should_throttle = 0;
4512
27cdeb70
MX		 4513 if (found_extent &&
4514 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4515 root == root->fs_info->tree_root)) {
b9473439 4516 btrfs_set_path_blocking(path);
28ed1345 4517 bytes_deleted += extent_num_bytes;
39279cc3 4518 ret = btrfs_free_extent(trans, root, extent_start,
5d4f98a2
YZ		 4519 extent_num_bytes, 0,
4520 btrfs_header_owner(leaf),
b06c4bf5 4521 ino, extent_offset);
39279cc3 4522 BUG_ON(ret);
1262133b 4523 if (btrfs_should_throttle_delayed_refs(trans, root))
28ed1345
CM		 4524 btrfs_async_run_delayed_refs(root,
4525 trans->delayed_ref_updates * 2, 0);
28f75a0e
CM		 4526 if (be_nice) {
4527 if (truncate_space_check(trans, root,
4528 extent_num_bytes)) {
4529 should_end = 1;
4530 }
4531 if (btrfs_should_throttle_delayed_refs(trans,
4532 root)) {
4533 should_throttle = 1;
4534 }
4535 }
39279cc3 4536 }
85e21bac 4537
8082510e
YZ		 4538 if (found_type == BTRFS_INODE_ITEM_KEY)
4539 break;
4540
4541 if (path->slots[0] == 0 ||
1262133b 4542 path->slots[0] != pending_del_slot ||
28f75a0e 4543 should_throttle || should_end) {
8082510e
YZ		 4544 if (pending_del_nr) {
4545 ret = btrfs_del_items(trans, root, path,
4546 pending_del_slot,
4547 pending_del_nr);
79787eaa
JM		 4548 if (ret) {
4549 btrfs_abort_transaction(trans,
4550 root, ret);
4551 goto error;
4552 }
8082510e
YZ		 4553 pending_del_nr = 0;
4554 }
b3b4aa74 4555 btrfs_release_path(path);
28f75a0e 4556 if (should_throttle) {
1262133b
JB		 4557 unsigned long updates = trans->delayed_ref_updates;
4558 if (updates) {
4559 trans->delayed_ref_updates = 0;
4560 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4561 if (ret && !err)
4562 err = ret;
4563 }
4564 }
28f75a0e
CM		 4565 /*
4566 * if we failed to refill our space rsv, bail out
4567 * and let the transaction restart
4568 */
4569 if (should_end) {
4570 err = -EAGAIN;
4571 goto error;
4572 }
85e21bac 4573 goto search_again;
8082510e
YZ		 4574 } else {
4575 path->slots[0]--;
85e21bac 4576 }
39279cc3 4577 }
8082510e 4578out:
85e21bac
CM		 4579 if (pending_del_nr) {
4580 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4581 pending_del_nr);
79787eaa
JM		 4582 if (ret)
4583 btrfs_abort_transaction(trans, root, ret);
85e21bac 4584 }
79787eaa 4585error:
c1aa4575 4586 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
7f4f6e0a 4587 btrfs_ordered_update_i_size(inode, last_size, NULL);
28ed1345 4588
39279cc3 4589 btrfs_free_path(path);
28ed1345 4590
ee22184b 4591 if (be_nice && bytes_deleted > SZ_32M) {
28ed1345
CM		 4592 unsigned long updates = trans->delayed_ref_updates;
4593 if (updates) {
4594 trans->delayed_ref_updates = 0;
4595 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4596 if (ret && !err)
4597 err = ret;
4598 }
4599 }
8082510e 4600 return err;
39279cc3
CM		 4601}
4602
4603/*
9703fefe 4604 * btrfs_truncate_block - read, zero a chunk and write a block
2aaa6655
JB		 4605 * @inode - inode that we're zeroing
4606 * @from - the offset to start zeroing
4607 * @len - the length to zero, 0 to zero the entire range respective to the
4608 * offset
4609 * @front - zero up to the offset instead of from the offset on
4610 *
9703fefe 4611 * This will find the block for the "from" offset and cow the block and zero the
2aaa6655 4612 * part we want to zero. This is used with truncate and hole punching.
39279cc3 4613 */
9703fefe 4614int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
2aaa6655 4615 int front)
39279cc3 4616{
2aaa6655 4617 struct address_space *mapping = inode->i_mapping;
db94535d 4618 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 4619 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4620 struct btrfs_ordered_extent *ordered;
2ac55d41 4621 struct extent_state *cached_state = NULL;
e6dcd2dc 4622 char *kaddr;
db94535d 4623 u32 blocksize = root->sectorsize;
39279cc3 4624 pgoff_t index = from >> PAGE_CACHE_SHIFT;
9703fefe 4625 unsigned offset = from & (blocksize - 1);
39279cc3 4626 struct page *page;
3b16a4e3 4627 gfp_t mask = btrfs_alloc_write_mask(mapping);
39279cc3 4628 int ret = 0;
9703fefe
CR		 4629 u64 block_start;
4630 u64 block_end;
39279cc3 4631
2aaa6655
JB		 4632 if ((offset & (blocksize - 1)) == 0 &&
4633 (!len || ((len & (blocksize - 1)) == 0)))
39279cc3 4634 goto out;
9703fefe 4635
7cf5b976 4636 ret = btrfs_delalloc_reserve_space(inode,
9703fefe 4637 round_down(from, blocksize), blocksize);
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,
9703fefe
CR		 4645 round_down(from, blocksize),
4646 blocksize);
ac6a2b36 4647 ret = -ENOMEM;
39279cc3 4648 goto out;
5d5e103a 4649 }
e6dcd2dc 4650
9703fefe
CR		 4651 block_start = round_down(from, blocksize);
4652 block_end = block_start + blocksize - 1;
e6dcd2dc 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
9703fefe 4669 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
e6dcd2dc
CM		 4670 set_page_extent_mapped(page);
4671
9703fefe 4672 ordered = btrfs_lookup_ordered_extent(inode, block_start);
e6dcd2dc 4673 if (ordered) {
9703fefe 4674 unlock_extent_cached(io_tree, block_start, block_end,
2ac55d41 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
9703fefe 4683 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
9e8a4a8b
LB		 4684 EXTENT_DIRTY | EXTENT_DELALLOC |
4685 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 4686 0, 0, &cached_state, GFP_NOFS);
5d5e103a 4687
9703fefe 4688 ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
2ac55d41 4689 &cached_state);
9ed74f2d 4690 if (ret) {
9703fefe 4691 unlock_extent_cached(io_tree, block_start, block_end,
2ac55d41 4692 &cached_state, GFP_NOFS);
9ed74f2d
JB		 4693 goto out_unlock;
4694 }
4695
9703fefe 4696 if (offset != blocksize) {
2aaa6655 4697 if (!len)
9703fefe 4698 len = blocksize - offset;
e6dcd2dc 4699 kaddr = kmap(page);
2aaa6655 4700 if (front)
9703fefe
CR		 4701 memset(kaddr + (block_start - page_offset(page)),
4702 0, offset);
2aaa6655 4703 else
9703fefe
CR		 4704 memset(kaddr + (block_start - page_offset(page)) + offset,
4705 0, len);
e6dcd2dc
CM		 4706 flush_dcache_page(page);
4707 kunmap(page);
4708 }
247e743c 4709 ClearPageChecked(page);
e6dcd2dc 4710 set_page_dirty(page);
9703fefe 4711 unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
2ac55d41 4712 GFP_NOFS);
39279cc3 4713
89642229 4714out_unlock:
5d5e103a 4715 if (ret)
9703fefe
CR		 4716 btrfs_delalloc_release_space(inode, block_start,
4717 blocksize);
39279cc3
CM		 4718 unlock_page(page);
4719 page_cache_release(page);
4720out:
4721 return ret;
4722}
4723
16e7549f
JB		 4724static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4725 u64 offset, u64 len)
4726{
4727 struct btrfs_trans_handle *trans;
4728 int ret;
4729
4730 /*
4731 * Still need to make sure the inode looks like it's been updated so
4732 * that any holes get logged if we fsync.
4733 */
4734 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4735 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4736 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4737 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4738 return 0;
4739 }
4740
4741 /*
4742 * 1 - for the one we're dropping
4743 * 1 - for the one we're adding
4744 * 1 - for updating the inode.
4745 */
4746 trans = btrfs_start_transaction(root, 3);
4747 if (IS_ERR(trans))
4748 return PTR_ERR(trans);
4749
4750 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4751 if (ret) {
4752 btrfs_abort_transaction(trans, root, ret);
4753 btrfs_end_transaction(trans, root);
4754 return ret;
4755 }
4756
4757 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4758 0, 0, len, 0, len, 0, 0, 0);
4759 if (ret)
4760 btrfs_abort_transaction(trans, root, ret);
4761 else
4762 btrfs_update_inode(trans, root, inode);
4763 btrfs_end_transaction(trans, root);
4764 return ret;
4765}
4766
695a0d0d
JB		 4767/*
4768 * This function puts in dummy file extents for the area we're creating a hole
4769 * for. So if we are truncating this file to a larger size we need to insert
4770 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4771 * the range between oldsize and size
4772 */
a41ad394 4773int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
39279cc3 4774{
9036c102
YZ		 4775 struct btrfs_root *root = BTRFS_I(inode)->root;
4776 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a22285a6 4777 struct extent_map *em = NULL;
2ac55d41 4778 struct extent_state *cached_state = NULL;
5dc562c5 4779 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
fda2832f
QW		 4780 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4781 u64 block_end = ALIGN(size, root->sectorsize);
9036c102
YZ		 4782 u64 last_byte;
4783 u64 cur_offset;
4784 u64 hole_size;
9ed74f2d 4785 int err = 0;
39279cc3 4786
a71754fc 4787 /*
9703fefe
CR		 4788 * If our size started in the middle of a block we need to zero out the
4789 * rest of the block before we expand the i_size, otherwise we could
a71754fc
JB		 4790 * expose stale data.
4791 */
9703fefe 4792 err = btrfs_truncate_block(inode, oldsize, 0, 0);
a71754fc
JB		 4793 if (err)
4794 return err;
4795
9036c102
YZ		 4796 if (size <= hole_start)
4797 return 0;
4798
9036c102
YZ		 4799 while (1) {
4800 struct btrfs_ordered_extent *ordered;
fa7c1494 4801
ff13db41 4802 lock_extent_bits(io_tree, hole_start, block_end - 1,
d0082371 4803 &cached_state);
fa7c1494
MX		 4804 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4805 block_end - hole_start);
9036c102
YZ		 4806 if (!ordered)
4807 break;
2ac55d41
JB		 4808 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4809 &cached_state, GFP_NOFS);
fa7c1494 4810 btrfs_start_ordered_extent(inode, ordered, 1);
9036c102
YZ		 4811 btrfs_put_ordered_extent(ordered);
4812 }
39279cc3 4813
9036c102
YZ		 4814 cur_offset = hole_start;
4815 while (1) {
4816 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4817 block_end - cur_offset, 0);
79787eaa
JM		 4818 if (IS_ERR(em)) {
4819 err = PTR_ERR(em);
f2767956 4820 em = NULL;
79787eaa
JM		 4821 break;
4822 }
9036c102 4823 last_byte = min(extent_map_end(em), block_end);
fda2832f 4824 last_byte = ALIGN(last_byte , root->sectorsize);
8082510e 4825 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5dc562c5 4826 struct extent_map *hole_em;
9036c102 4827 hole_size = last_byte - cur_offset;
9ed74f2d 4828
16e7549f
JB		 4829 err = maybe_insert_hole(root, inode, cur_offset,
4830 hole_size);
4831 if (err)
3893e33b 4832 break;
5dc562c5
JB		 4833 btrfs_drop_extent_cache(inode, cur_offset,
4834 cur_offset + hole_size - 1, 0);
4835 hole_em = alloc_extent_map();
4836 if (!hole_em) {
4837 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4838 &BTRFS_I(inode)->runtime_flags);
4839 goto next;
4840 }
4841 hole_em->start = cur_offset;
4842 hole_em->len = hole_size;
4843 hole_em->orig_start = cur_offset;
8082510e 4844
5dc562c5
JB		 4845 hole_em->block_start = EXTENT_MAP_HOLE;
4846 hole_em->block_len = 0;
b4939680 4847 hole_em->orig_block_len = 0;
cc95bef6 4848 hole_em->ram_bytes = hole_size;
5dc562c5
JB		 4849 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4850 hole_em->compress_type = BTRFS_COMPRESS_NONE;
16e7549f 4851 hole_em->generation = root->fs_info->generation;
8082510e 4852
5dc562c5
JB		 4853 while (1) {
4854 write_lock(&em_tree->lock);
09a2a8f9 4855 err = add_extent_mapping(em_tree, hole_em, 1);
5dc562c5
JB		 4856 write_unlock(&em_tree->lock);
4857 if (err != -EEXIST)
4858 break;
4859 btrfs_drop_extent_cache(inode, cur_offset,
4860 cur_offset +
4861 hole_size - 1, 0);
4862 }
4863 free_extent_map(hole_em);
9036c102 4864 }
16e7549f 4865next:
9036c102 4866 free_extent_map(em);
a22285a6 4867 em = NULL;
9036c102 4868 cur_offset = last_byte;
8082510e 4869 if (cur_offset >= block_end)
9036c102
YZ		 4870 break;
4871 }
a22285a6 4872 free_extent_map(em);
2ac55d41
JB		 4873 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4874 GFP_NOFS);
9036c102
YZ		 4875 return err;
4876}
39279cc3 4877
3972f260 4878static int btrfs_setsize(struct inode *inode, struct iattr *attr)
8082510e 4879{
f4a2f4c5
MX		 4880 struct btrfs_root *root = BTRFS_I(inode)->root;
4881 struct btrfs_trans_handle *trans;
a41ad394 4882 loff_t oldsize = i_size_read(inode);
3972f260
ES		 4883 loff_t newsize = attr->ia_size;
4884 int mask = attr->ia_valid;
8082510e
YZ		 4885 int ret;
4886
3972f260
ES		 4887 /*
4888 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4889 * special case where we need to update the times despite not having
4890 * these flags set. For all other operations the VFS set these flags
4891 * explicitly if it wants a timestamp update.
4892 */
dff6efc3
CH		 4893 if (newsize != oldsize) {
4894 inode_inc_iversion(inode);
4895 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4896 inode->i_ctime = inode->i_mtime =
4897 current_fs_time(inode->i_sb);
4898 }
3972f260 4899
a41ad394 4900 if (newsize > oldsize) {
7caef267 4901 truncate_pagecache(inode, newsize);
9ea24bbe
FM		 4902 /*
4903 * Don't do an expanding truncate while snapshoting is ongoing.
4904 * This is to ensure the snapshot captures a fully consistent
4905 * state of this file - if the snapshot captures this expanding
4906 * truncation, it must capture all writes that happened before
4907 * this truncation.
4908 */
0bc19f90 4909 btrfs_wait_for_snapshot_creation(root);
a41ad394 4910 ret = btrfs_cont_expand(inode, oldsize, newsize);
9ea24bbe
FM		 4911 if (ret) {
4912 btrfs_end_write_no_snapshoting(root);
8082510e 4913 return ret;
9ea24bbe 4914 }
8082510e 4915
f4a2f4c5 4916 trans = btrfs_start_transaction(root, 1);
9ea24bbe
FM		 4917 if (IS_ERR(trans)) {
4918 btrfs_end_write_no_snapshoting(root);
f4a2f4c5 4919 return PTR_ERR(trans);
9ea24bbe 4920 }
f4a2f4c5
MX		 4921
4922 i_size_write(inode, newsize);
4923 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4924 ret = btrfs_update_inode(trans, root, inode);
9ea24bbe 4925 btrfs_end_write_no_snapshoting(root);
7ad85bb7 4926 btrfs_end_transaction(trans, root);
a41ad394 4927 } else {
8082510e 4928
a41ad394
JB		 4929 /*
4930 * We're truncating a file that used to have good data down to
4931 * zero. Make sure it gets into the ordered flush list so that
4932 * any new writes get down to disk quickly.
4933 */
4934 if (newsize == 0)
72ac3c0d
JB		 4935 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4936 &BTRFS_I(inode)->runtime_flags);
8082510e 4937
f3fe820c
JB		 4938 /*
4939 * 1 for the orphan item we're going to add
4940 * 1 for the orphan item deletion.
4941 */
4942 trans = btrfs_start_transaction(root, 2);
4943 if (IS_ERR(trans))
4944 return PTR_ERR(trans);
4945
4946 /*
4947 * We need to do this in case we fail at _any_ point during the
4948 * actual truncate. Once we do the truncate_setsize we could
4949 * invalidate pages which forces any outstanding ordered io to
4950 * be instantly completed which will give us extents that need
4951 * to be truncated. If we fail to get an orphan inode down we
4952 * could have left over extents that were never meant to live,
4953 * so we need to garuntee from this point on that everything
4954 * will be consistent.
4955 */
4956 ret = btrfs_orphan_add(trans, inode);
4957 btrfs_end_transaction(trans, root);
4958 if (ret)
4959 return ret;
4960
a41ad394
JB		 4961 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4962 truncate_setsize(inode, newsize);
2e60a51e
MX		 4963
4964 /* Disable nonlocked read DIO to avoid the end less truncate */
4965 btrfs_inode_block_unlocked_dio(inode);
4966 inode_dio_wait(inode);
4967 btrfs_inode_resume_unlocked_dio(inode);
4968
a41ad394 4969 ret = btrfs_truncate(inode);
7f4f6e0a
JB		 4970 if (ret && inode->i_nlink) {
4971 int err;
4972
4973 /*
4974 * failed to truncate, disk_i_size is only adjusted down
4975 * as we remove extents, so it should represent the true
4976 * size of the inode, so reset the in memory size and
4977 * delete our orphan entry.
4978 */
4979 trans = btrfs_join_transaction(root);
4980 if (IS_ERR(trans)) {
4981 btrfs_orphan_del(NULL, inode);
4982 return ret;
4983 }
4984 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4985 err = btrfs_orphan_del(trans, inode);
4986 if (err)
4987 btrfs_abort_transaction(trans, root, err);
4988 btrfs_end_transaction(trans, root);
4989 }
8082510e
YZ		 4990 }
4991
a41ad394 4992 return ret;
8082510e
YZ		 4993}
4994
9036c102
YZ		 4995static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4996{
2b0143b5 4997 struct inode *inode = d_inode(dentry);
b83cc969 4998 struct btrfs_root *root = BTRFS_I(inode)->root;
9036c102 4999 int err;
39279cc3 5000
b83cc969
LZ		 5001 if (btrfs_root_readonly(root))
5002 return -EROFS;
5003
9036c102
YZ		 5004 err = inode_change_ok(inode, attr);
5005 if (err)
5006 return err;
2bf5a725 5007
5a3f23d5 5008 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3972f260 5009 err = btrfs_setsize(inode, attr);
8082510e
YZ		 5010 if (err)
5011 return err;
39279cc3 5012 }
9036c102 5013
1025774c
CH		 5014 if (attr->ia_valid) {
5015 setattr_copy(inode, attr);
0c4d2d95 5016 inode_inc_iversion(inode);
22c44fe6 5017 err = btrfs_dirty_inode(inode);
1025774c 5018
22c44fe6 5019 if (!err && attr->ia_valid & ATTR_MODE)
996a710d 5020 err = posix_acl_chmod(inode, inode->i_mode);
1025774c 5021 }
33268eaf 5022
39279cc3
CM		 5023 return err;
5024}
61295eb8 5025
131e404a
FDBM		 5026/*
5027 * While truncating the inode pages during eviction, we get the VFS calling
5028 * btrfs_invalidatepage() against each page of the inode. This is slow because
5029 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5030 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5031 * extent_state structures over and over, wasting lots of time.
5032 *
5033 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5034 * those expensive operations on a per page basis and do only the ordered io
5035 * finishing, while we release here the extent_map and extent_state structures,
5036 * without the excessive merging and splitting.
5037 */
5038static void evict_inode_truncate_pages(struct inode *inode)
5039{
5040 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5041 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5042 struct rb_node *node;
5043
5044 ASSERT(inode->i_state & I_FREEING);
91b0abe3 5045 truncate_inode_pages_final(&inode->i_data);
131e404a
FDBM		 5046
5047 write_lock(&map_tree->lock);
5048 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5049 struct extent_map *em;
5050
5051 node = rb_first(&map_tree->map);
5052 em = rb_entry(node, struct extent_map, rb_node);
180589ef
WS		 5053 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5054 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
131e404a
FDBM		 5055 remove_extent_mapping(map_tree, em);
5056 free_extent_map(em);
7064dd5c
FM		 5057 if (need_resched()) {
5058 write_unlock(&map_tree->lock);
5059 cond_resched();
5060 write_lock(&map_tree->lock);
5061 }
131e404a
FDBM		 5062 }
5063 write_unlock(&map_tree->lock);
5064
6ca07097
FM		 5065 /*
5066 * Keep looping until we have no more ranges in the io tree.
5067 * We can have ongoing bios started by readpages (called from readahead)
9c6429d9
FM		 5068 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5069 * still in progress (unlocked the pages in the bio but did not yet
5070 * unlocked the ranges in the io tree). Therefore this means some
6ca07097
FM		 5071 * ranges can still be locked and eviction started because before
5072 * submitting those bios, which are executed by a separate task (work
5073 * queue kthread), inode references (inode->i_count) were not taken
5074 * (which would be dropped in the end io callback of each bio).
5075 * Therefore here we effectively end up waiting for those bios and
5076 * anyone else holding locked ranges without having bumped the inode's
5077 * reference count - if we don't do it, when they access the inode's
5078 * io_tree to unlock a range it may be too late, leading to an
5079 * use-after-free issue.
5080 */
131e404a
FDBM		 5081 spin_lock(&io_tree->lock);
5082 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5083 struct extent_state *state;
5084 struct extent_state *cached_state = NULL;
6ca07097
FM		 5085 u64 start;
5086 u64 end;
131e404a
FDBM		 5087
5088 node = rb_first(&io_tree->state);
5089 state = rb_entry(node, struct extent_state, rb_node);
6ca07097
FM		 5090 start = state->start;
5091 end = state->end;
131e404a
FDBM		 5092 spin_unlock(&io_tree->lock);
5093
ff13db41 5094 lock_extent_bits(io_tree, start, end, &cached_state);
b9d0b389
QW		 5095
5096 /*
5097 * If still has DELALLOC flag, the extent didn't reach disk,
5098 * and its reserved space won't be freed by delayed_ref.
5099 * So we need to free its reserved space here.
5100 * (Refer to comment in btrfs_invalidatepage, case 2)
5101 *
5102 * Note, end is the bytenr of last byte, so we need + 1 here.
5103 */
5104 if (state->state & EXTENT_DELALLOC)
5105 btrfs_qgroup_free_data(inode, start, end - start + 1);
5106
6ca07097 5107 clear_extent_bit(io_tree, start, end,
131e404a
FDBM		 5108 EXTENT_LOCKED | EXTENT_DIRTY |
5109 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5110 EXTENT_DEFRAG, 1, 1,
5111 &cached_state, GFP_NOFS);
131e404a 5112
7064dd5c 5113 cond_resched();
131e404a
FDBM		 5114 spin_lock(&io_tree->lock);
5115 }
5116 spin_unlock(&io_tree->lock);
5117}
5118
bd555975 5119void btrfs_evict_inode(struct inode *inode)
39279cc3
CM		 5120{
5121 struct btrfs_trans_handle *trans;
5122 struct btrfs_root *root = BTRFS_I(inode)->root;
726c35fa 5123 struct btrfs_block_rsv *rsv, *global_rsv;
3bce876f 5124 int steal_from_global = 0;
07127184 5125 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3
CM		 5126 int ret;
5127
1abe9b8a 5128 trace_btrfs_inode_evict(inode);
5129
131e404a
FDBM		 5130 evict_inode_truncate_pages(inode);
5131
69e9c6c6
SB		 5132 if (inode->i_nlink &&
5133 ((btrfs_root_refs(&root->root_item) != 0 &&
5134 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5135 btrfs_is_free_space_inode(inode)))
bd555975
AV		 5136 goto no_delete;
5137
39279cc3 5138 if (is_bad_inode(inode)) {
7b128766 5139 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 5140 goto no_delete;
5141 }
bd555975 5142 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
a30e577c
JM		 5143 if (!special_file(inode->i_mode))
5144 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 5145
f612496b
MX		 5146 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5147
c71bf099 5148 if (root->fs_info->log_root_recovering) {
6bf02314 5149 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8a35d95f 5150 &BTRFS_I(inode)->runtime_flags));
c71bf099
YZ		 5151 goto no_delete;
5152 }
5153
76dda93c 5154 if (inode->i_nlink > 0) {
69e9c6c6
SB		 5155 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5156 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
76dda93c
YZ		 5157 goto no_delete;
5158 }
5159
0e8c36a9
MX		 5160 ret = btrfs_commit_inode_delayed_inode(inode);
5161 if (ret) {
5162 btrfs_orphan_del(NULL, inode);
5163 goto no_delete;
5164 }
5165
66d8f3dd 5166 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4289a667
JB		 5167 if (!rsv) {
5168 btrfs_orphan_del(NULL, inode);
5169 goto no_delete;
5170 }
4a338542 5171 rsv->size = min_size;
ca7e70f5 5172 rsv->failfast = 1;
726c35fa 5173 global_rsv = &root->fs_info->global_block_rsv;
4289a667 5174
dbe674a9 5175 btrfs_i_size_write(inode, 0);
5f39d397 5176
4289a667 5177 /*
8407aa46
MX		 5178 * This is a bit simpler than btrfs_truncate since we've already
5179 * reserved our space for our orphan item in the unlink, so we just
5180 * need to reserve some slack space in case we add bytes and update
5181 * inode item when doing the truncate.
4289a667 5182 */
8082510e 5183 while (1) {
08e007d2
MX		 5184 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5185 BTRFS_RESERVE_FLUSH_LIMIT);
726c35fa
JB		 5186
5187 /*
5188 * Try and steal from the global reserve since we will
5189 * likely not use this space anyway, we want to try as
5190 * hard as possible to get this to work.
5191 */
5192 if (ret)
3bce876f
JB		 5193 steal_from_global++;
5194 else
5195 steal_from_global = 0;
5196 ret = 0;
d68fc57b 5197
3bce876f
JB		 5198 /*
5199 * steal_from_global == 0: we reserved stuff, hooray!
5200 * steal_from_global == 1: we didn't reserve stuff, boo!
5201 * steal_from_global == 2: we've committed, still not a lot of
5202 * room but maybe we'll have room in the global reserve this
5203 * time.
5204 * steal_from_global == 3: abandon all hope!
5205 */
5206 if (steal_from_global > 2) {
c2cf52eb
SK		 5207 btrfs_warn(root->fs_info,
5208 "Could not get space for a delete, will truncate on mount %d",
5209 ret);
4289a667
JB		 5210 btrfs_orphan_del(NULL, inode);
5211 btrfs_free_block_rsv(root, rsv);
5212 goto no_delete;
d68fc57b 5213 }
7b128766 5214
0e8c36a9 5215 trans = btrfs_join_transaction(root);
4289a667
JB		 5216 if (IS_ERR(trans)) {
5217 btrfs_orphan_del(NULL, inode);
5218 btrfs_free_block_rsv(root, rsv);
5219 goto no_delete;
d68fc57b 5220 }
7b128766 5221
3bce876f
JB		 5222 /*
5223 * We can't just steal from the global reserve, we need tomake
5224 * sure there is room to do it, if not we need to commit and try
5225 * again.
5226 */
5227 if (steal_from_global) {
5228 if (!btrfs_check_space_for_delayed_refs(trans, root))
5229 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5230 min_size);
5231 else
5232 ret = -ENOSPC;
5233 }
5234
5235 /*
5236 * Couldn't steal from the global reserve, we have too much
5237 * pending stuff built up, commit the transaction and try it
5238 * again.
5239 */
5240 if (ret) {
5241 ret = btrfs_commit_transaction(trans, root);
5242 if (ret) {
5243 btrfs_orphan_del(NULL, inode);
5244 btrfs_free_block_rsv(root, rsv);
5245 goto no_delete;
5246 }
5247 continue;
5248 } else {
5249 steal_from_global = 0;
5250 }
5251
4289a667
JB		 5252 trans->block_rsv = rsv;
5253
d68fc57b 5254 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
28ed1345 5255 if (ret != -ENOSPC && ret != -EAGAIN)
8082510e 5256 break;
85e21bac 5257
8407aa46 5258 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e
YZ		 5259 btrfs_end_transaction(trans, root);
5260 trans = NULL;
b53d3f5d 5261 btrfs_btree_balance_dirty(root);
8082510e 5262 }
5f39d397 5263
4289a667
JB		 5264 btrfs_free_block_rsv(root, rsv);
5265
4ef31a45
JB		 5266 /*
5267 * Errors here aren't a big deal, it just means we leave orphan items
5268 * in the tree. They will be cleaned up on the next mount.
5269 */
8082510e 5270 if (ret == 0) {
4289a667 5271 trans->block_rsv = root->orphan_block_rsv;
4ef31a45
JB		 5272 btrfs_orphan_del(trans, inode);
5273 } else {
5274 btrfs_orphan_del(NULL, inode);
8082510e 5275 }
54aa1f4d 5276
4289a667 5277 trans->block_rsv = &root->fs_info->trans_block_rsv;
581bb050
LZ		 5278 if (!(root == root->fs_info->tree_root ||
5279 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
33345d01 5280 btrfs_return_ino(root, btrfs_ino(inode));
581bb050 5281
54aa1f4d 5282 btrfs_end_transaction(trans, root);
b53d3f5d 5283 btrfs_btree_balance_dirty(root);
39279cc3 5284no_delete:
89042e5a 5285 btrfs_remove_delayed_node(inode);
dbd5768f 5286 clear_inode(inode);
39279cc3
CM		 5287}
5288
5289/*
5290 * this returns the key found in the dir entry in the location pointer.
5291 * If no dir entries were found, location->objectid is 0.
5292 */
5293static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5294 struct btrfs_key *location)
5295{
5296 const char *name = dentry->d_name.name;
5297 int namelen = dentry->d_name.len;
5298 struct btrfs_dir_item *di;
5299 struct btrfs_path *path;
5300 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 5301 int ret = 0;
39279cc3
CM		 5302
5303 path = btrfs_alloc_path();
d8926bb3
MF		 5304 if (!path)
5305 return -ENOMEM;
3954401f 5306
33345d01 5307 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
39279cc3 5308 namelen, 0);
0d9f7f3e
Y		 5309 if (IS_ERR(di))
5310 ret = PTR_ERR(di);
d397712b 5311
c704005d 5312 if (IS_ERR_OR_NULL(di))
3954401f 5313 goto out_err;
d397712b 5314
5f39d397 5315 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 5316out:
39279cc3
CM		 5317 btrfs_free_path(path);
5318 return ret;
3954401f
CM		 5319out_err:
5320 location->objectid = 0;
5321 goto out;
39279cc3
CM		 5322}
5323
5324/*
5325 * when we hit a tree root in a directory, the btrfs part of the inode
5326 * needs to be changed to reflect the root directory of the tree root. This
5327 * is kind of like crossing a mount point.
5328 */
5329static int fixup_tree_root_location(struct btrfs_root *root,
4df27c4d
YZ		 5330 struct inode *dir,
5331 struct dentry *dentry,
5332 struct btrfs_key *location,
5333 struct btrfs_root **sub_root)
39279cc3 5334{
4df27c4d
YZ		 5335 struct btrfs_path *path;
5336 struct btrfs_root *new_root;
5337 struct btrfs_root_ref *ref;
5338 struct extent_buffer *leaf;
1d4c08e0 5339 struct btrfs_key key;
4df27c4d
YZ		 5340 int ret;
5341 int err = 0;
39279cc3 5342
4df27c4d
YZ		 5343 path = btrfs_alloc_path();
5344 if (!path) {
5345 err = -ENOMEM;
5346 goto out;
5347 }
39279cc3 5348
4df27c4d 5349 err = -ENOENT;
1d4c08e0
DS		 5350 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5351 key.type = BTRFS_ROOT_REF_KEY;
5352 key.offset = location->objectid;
5353
5354 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5355 0, 0);
4df27c4d
YZ		 5356 if (ret) {
5357 if (ret < 0)
5358 err = ret;
5359 goto out;
5360 }
39279cc3 5361
4df27c4d
YZ		 5362 leaf = path->nodes[0];
5363 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
33345d01 5364 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4df27c4d
YZ		 5365 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5366 goto out;
39279cc3 5367
4df27c4d
YZ		 5368 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5369 (unsigned long)(ref + 1),
5370 dentry->d_name.len);
5371 if (ret)
5372 goto out;
5373
b3b4aa74 5374 btrfs_release_path(path);
4df27c4d
YZ		 5375
5376 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5377 if (IS_ERR(new_root)) {
5378 err = PTR_ERR(new_root);
5379 goto out;
5380 }
5381
4df27c4d
YZ		 5382 *sub_root = new_root;
5383 location->objectid = btrfs_root_dirid(&new_root->root_item);
5384 location->type = BTRFS_INODE_ITEM_KEY;
5385 location->offset = 0;
5386 err = 0;
5387out:
5388 btrfs_free_path(path);
5389 return err;
39279cc3
CM		 5390}
5391
5d4f98a2
YZ		 5392static void inode_tree_add(struct inode *inode)
5393{
5394 struct btrfs_root *root = BTRFS_I(inode)->root;
5395 struct btrfs_inode *entry;
03e860bd
FNP		 5396 struct rb_node **p;
5397 struct rb_node *parent;
cef21937 5398 struct rb_node *new = &BTRFS_I(inode)->rb_node;
33345d01 5399 u64 ino = btrfs_ino(inode);
5d4f98a2 5400
1d3382cb 5401 if (inode_unhashed(inode))
76dda93c 5402 return;
e1409cef 5403 parent = NULL;
5d4f98a2 5404 spin_lock(&root->inode_lock);
e1409cef 5405 p = &root->inode_tree.rb_node;
5d4f98a2
YZ		 5406 while (*p) {
5407 parent = *p;
5408 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5409
33345d01 5410 if (ino < btrfs_ino(&entry->vfs_inode))
03e860bd 5411 p = &parent->rb_left;
33345d01 5412 else if (ino > btrfs_ino(&entry->vfs_inode))
03e860bd 5413 p = &parent->rb_right;
5d4f98a2
YZ		 5414 else {
5415 WARN_ON(!(entry->vfs_inode.i_state &
a4ffdde6 5416 (I_WILL_FREE | I_FREEING)));
cef21937 5417 rb_replace_node(parent, new, &root->inode_tree);
03e860bd
FNP		 5418 RB_CLEAR_NODE(parent);
5419 spin_unlock(&root->inode_lock);
cef21937 5420 return;
5d4f98a2
YZ		 5421 }
5422 }
cef21937
FDBM		 5423 rb_link_node(new, parent, p);
5424 rb_insert_color(new, &root->inode_tree);
5d4f98a2
YZ		 5425 spin_unlock(&root->inode_lock);
5426}
5427
5428static void inode_tree_del(struct inode *inode)
5429{
5430 struct btrfs_root *root = BTRFS_I(inode)->root;
76dda93c 5431 int empty = 0;
5d4f98a2 5432
03e860bd 5433 spin_lock(&root->inode_lock);
5d4f98a2 5434 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5d4f98a2 5435 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5d4f98a2 5436 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
76dda93c 5437 empty = RB_EMPTY_ROOT(&root->inode_tree);
5d4f98a2 5438 }
03e860bd 5439 spin_unlock(&root->inode_lock);
76dda93c 5440
69e9c6c6 5441 if (empty && btrfs_root_refs(&root->root_item) == 0) {
76dda93c
YZ		 5442 synchronize_srcu(&root->fs_info->subvol_srcu);
5443 spin_lock(&root->inode_lock);
5444 empty = RB_EMPTY_ROOT(&root->inode_tree);
5445 spin_unlock(&root->inode_lock);
5446 if (empty)
5447 btrfs_add_dead_root(root);
5448 }
5449}
5450
143bede5 5451void btrfs_invalidate_inodes(struct btrfs_root *root)
76dda93c
YZ		 5452{
5453 struct rb_node *node;
5454 struct rb_node *prev;
5455 struct btrfs_inode *entry;
5456 struct inode *inode;
5457 u64 objectid = 0;
5458
7813b3db
LB		 5459 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5460 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
76dda93c
YZ		 5461
5462 spin_lock(&root->inode_lock);
5463again:
5464 node = root->inode_tree.rb_node;
5465 prev = NULL;
5466 while (node) {
5467 prev = node;
5468 entry = rb_entry(node, struct btrfs_inode, rb_node);
5469
33345d01 5470 if (objectid < btrfs_ino(&entry->vfs_inode))
76dda93c 5471 node = node->rb_left;
33345d01 5472 else if (objectid > btrfs_ino(&entry->vfs_inode))
76dda93c
YZ		 5473 node = node->rb_right;
5474 else
5475 break;
5476 }
5477 if (!node) {
5478 while (prev) {
5479 entry = rb_entry(prev, struct btrfs_inode, rb_node);
33345d01 5480 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
76dda93c
YZ		 5481 node = prev;
5482 break;
5483 }
5484 prev = rb_next(prev);
5485 }
5486 }
5487 while (node) {
5488 entry = rb_entry(node, struct btrfs_inode, rb_node);
33345d01 5489 objectid = btrfs_ino(&entry->vfs_inode) + 1;
76dda93c
YZ		 5490 inode = igrab(&entry->vfs_inode);
5491 if (inode) {
5492 spin_unlock(&root->inode_lock);
5493 if (atomic_read(&inode->i_count) > 1)
5494 d_prune_aliases(inode);
5495 /*
45321ac5 5496 * btrfs_drop_inode will have it removed from
76dda93c
YZ		 5497 * the inode cache when its usage count
5498 * hits zero.
5499 */
5500 iput(inode);
5501 cond_resched();
5502 spin_lock(&root->inode_lock);
5503 goto again;
5504 }
5505
5506 if (cond_resched_lock(&root->inode_lock))
5507 goto again;
5508
5509 node = rb_next(node);
5510 }
5511 spin_unlock(&root->inode_lock);
5d4f98a2
YZ		 5512}
5513
e02119d5
CM		 5514static int btrfs_init_locked_inode(struct inode *inode, void *p)
5515{
5516 struct btrfs_iget_args *args = p;
90d3e592
CM		 5517 inode->i_ino = args->location->objectid;
5518 memcpy(&BTRFS_I(inode)->location, args->location,
5519 sizeof(*args->location));
e02119d5 5520 BTRFS_I(inode)->root = args->root;
39279cc3
CM		 5521 return 0;
5522}
5523
5524static int btrfs_find_actor(struct inode *inode, void *opaque)
5525{
5526 struct btrfs_iget_args *args = opaque;
90d3e592 5527 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
d397712b 5528 args->root == BTRFS_I(inode)->root;
39279cc3
CM		 5529}
5530
5d4f98a2 5531static struct inode *btrfs_iget_locked(struct super_block *s,
90d3e592 5532 struct btrfs_key *location,
5d4f98a2 5533 struct btrfs_root *root)
39279cc3
CM		 5534{
5535 struct inode *inode;
5536 struct btrfs_iget_args args;
90d3e592 5537 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
778ba82b 5538
90d3e592 5539 args.location = location;
39279cc3
CM		 5540 args.root = root;
5541
778ba82b 5542 inode = iget5_locked(s, hashval, btrfs_find_actor,
39279cc3
CM		 5543 btrfs_init_locked_inode,
5544 (void *)&args);
5545 return inode;
5546}
5547
1a54ef8c
BR		 5548/* Get an inode object given its location and corresponding root.
5549 * Returns in *is_new if the inode was read from disk
5550 */
5551struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
73f73415 5552 struct btrfs_root *root, int *new)
1a54ef8c
BR		 5553{
5554 struct inode *inode;
5555
90d3e592 5556 inode = btrfs_iget_locked(s, location, root);
1a54ef8c 5557 if (!inode)
5d4f98a2 5558 return ERR_PTR(-ENOMEM);
1a54ef8c
BR		 5559
5560 if (inode->i_state & I_NEW) {
1a54ef8c 5561 btrfs_read_locked_inode(inode);
1748f843
MF		 5562 if (!is_bad_inode(inode)) {
5563 inode_tree_add(inode);
5564 unlock_new_inode(inode);
5565 if (new)
5566 *new = 1;
5567 } else {
e0b6d65b
ST		 5568 unlock_new_inode(inode);
5569 iput(inode);
5570 inode = ERR_PTR(-ESTALE);
1748f843
MF		 5571 }
5572 }
5573
1a54ef8c
BR		 5574 return inode;
5575}
5576
4df27c4d
YZ		 5577static struct inode *new_simple_dir(struct super_block *s,
5578 struct btrfs_key *key,
5579 struct btrfs_root *root)
5580{
5581 struct inode *inode = new_inode(s);
5582
5583 if (!inode)
5584 return ERR_PTR(-ENOMEM);
5585
4df27c4d
YZ		 5586 BTRFS_I(inode)->root = root;
5587 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
72ac3c0d 5588 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4df27c4d
YZ		 5589
5590 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
848cce0d 5591 inode->i_op = &btrfs_dir_ro_inode_operations;
4df27c4d
YZ		 5592 inode->i_fop = &simple_dir_operations;
5593 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
9cc97d64 5594 inode->i_mtime = CURRENT_TIME;
5595 inode->i_atime = inode->i_mtime;
5596 inode->i_ctime = inode->i_mtime;
5597 BTRFS_I(inode)->i_otime = inode->i_mtime;
4df27c4d
YZ		 5598
5599 return inode;
5600}
5601
3de4586c 5602struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3 5603{
d397712b 5604 struct inode *inode;
4df27c4d 5605 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3
CM		 5606 struct btrfs_root *sub_root = root;
5607 struct btrfs_key location;
76dda93c 5608 int index;
b4aff1f8 5609 int ret = 0;
39279cc3
CM		 5610
5611 if (dentry->d_name.len > BTRFS_NAME_LEN)
5612 return ERR_PTR(-ENAMETOOLONG);
5f39d397 5613
39e3c955 5614 ret = btrfs_inode_by_name(dir, dentry, &location);
39279cc3
CM		 5615 if (ret < 0)
5616 return ERR_PTR(ret);
5f39d397 5617
4df27c4d 5618 if (location.objectid == 0)
5662344b 5619 return ERR_PTR(-ENOENT);
4df27c4d
YZ		 5620
5621 if (location.type == BTRFS_INODE_ITEM_KEY) {
73f73415 5622 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4df27c4d
YZ		 5623 return inode;
5624 }
5625
5626 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5627
76dda93c 5628 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4df27c4d
YZ		 5629 ret = fixup_tree_root_location(root, dir, dentry,
5630 &location, &sub_root);
5631 if (ret < 0) {
5632 if (ret != -ENOENT)
5633 inode = ERR_PTR(ret);
5634 else
5635 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5636 } else {
73f73415 5637 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
39279cc3 5638 }
76dda93c
YZ		 5639 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5640
34d19bad 5641 if (!IS_ERR(inode) && root != sub_root) {
c71bf099
YZ		 5642 down_read(&root->fs_info->cleanup_work_sem);
5643 if (!(inode->i_sb->s_flags & MS_RDONLY))
66b4ffd1 5644 ret = btrfs_orphan_cleanup(sub_root);
c71bf099 5645 up_read(&root->fs_info->cleanup_work_sem);
01cd3367
JB		 5646 if (ret) {
5647 iput(inode);
66b4ffd1 5648 inode = ERR_PTR(ret);
01cd3367 5649 }
c71bf099
YZ		 5650 }
5651
3de4586c
CM		 5652 return inode;
5653}
5654
fe15ce44 5655static int btrfs_dentry_delete(const struct dentry *dentry)
76dda93c
YZ		 5656{
5657 struct btrfs_root *root;
2b0143b5 5658 struct inode *inode = d_inode(dentry);
76dda93c 5659
848cce0d 5660 if (!inode && !IS_ROOT(dentry))
2b0143b5 5661 inode = d_inode(dentry->d_parent);
76dda93c 5662
848cce0d
LZ		 5663 if (inode) {
5664 root = BTRFS_I(inode)->root;
efefb143
YZ		 5665 if (btrfs_root_refs(&root->root_item) == 0)
5666 return 1;
848cce0d
LZ		 5667
5668 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5669 return 1;
efefb143 5670 }
76dda93c
YZ		 5671 return 0;
5672}
5673
b4aff1f8
JB		 5674static void btrfs_dentry_release(struct dentry *dentry)
5675{
944a4515 5676 kfree(dentry->d_fsdata);
b4aff1f8
JB		 5677}
5678
3de4586c 5679static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
00cd8dd3 5680 unsigned int flags)
3de4586c 5681{
5662344b 5682 struct inode *inode;
a66e7cc6 5683
5662344b
TI		 5684 inode = btrfs_lookup_dentry(dir, dentry);
5685 if (IS_ERR(inode)) {
5686 if (PTR_ERR(inode) == -ENOENT)
5687 inode = NULL;
5688 else
5689 return ERR_CAST(inode);
5690 }
5691
41d28bca 5692 return d_splice_alias(inode, dentry);
39279cc3
CM		 5693}
5694
16cdcec7 5695unsigned char btrfs_filetype_table[] = {
39279cc3
CM		 5696 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5697};
5698
9cdda8d3 5699static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
39279cc3 5700{
9cdda8d3 5701 struct inode *inode = file_inode(file);
39279cc3
CM		 5702 struct btrfs_root *root = BTRFS_I(inode)->root;
5703 struct btrfs_item *item;
5704 struct btrfs_dir_item *di;
5705 struct btrfs_key key;
5f39d397 5706 struct btrfs_key found_key;
39279cc3 5707 struct btrfs_path *path;
16cdcec7
MX		 5708 struct list_head ins_list;
5709 struct list_head del_list;
39279cc3 5710 int ret;
5f39d397 5711 struct extent_buffer *leaf;
39279cc3 5712 int slot;
39279cc3
CM		 5713 unsigned char d_type;
5714 int over = 0;
5715 u32 di_cur;
5716 u32 di_total;
5717 u32 di_len;
5718 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 5719 char tmp_name[32];
5720 char *name_ptr;
5721 int name_len;
9cdda8d3 5722 int is_curr = 0; /* ctx->pos points to the current index? */
39279cc3
CM		 5723
5724 /* FIXME, use a real flag for deciding about the key type */
5725 if (root->fs_info->tree_root == root)
5726 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 5727
9cdda8d3
AV		 5728 if (!dir_emit_dots(file, ctx))
5729 return 0;
5730
49593bfa 5731 path = btrfs_alloc_path();
16cdcec7
MX		 5732 if (!path)
5733 return -ENOMEM;
ff5714cc 5734
e4058b54 5735 path->reada = READA_FORWARD;
49593bfa 5736
16cdcec7
MX		 5737 if (key_type == BTRFS_DIR_INDEX_KEY) {
5738 INIT_LIST_HEAD(&ins_list);
5739 INIT_LIST_HEAD(&del_list);
5740 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5741 }
5742
962a298f 5743 key.type = key_type;
9cdda8d3 5744 key.offset = ctx->pos;
33345d01 5745 key.objectid = btrfs_ino(inode);
5f39d397 5746
39279cc3
CM		 5747 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5748 if (ret < 0)
5749 goto err;
49593bfa
DW		 5750
5751 while (1) {
5f39d397 5752 leaf = path->nodes[0];
39279cc3 5753 slot = path->slots[0];
b9e03af0
LZ		 5754 if (slot >= btrfs_header_nritems(leaf)) {
5755 ret = btrfs_next_leaf(root, path);
5756 if (ret < 0)
5757 goto err;
5758 else if (ret > 0)
5759 break;
5760 continue;
39279cc3 5761 }
3de4586c 5762
dd3cc16b 5763 item = btrfs_item_nr(slot);
5f39d397
CM		 5764 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5765
5766 if (found_key.objectid != key.objectid)
39279cc3 5767 break;
962a298f 5768 if (found_key.type != key_type)
39279cc3 5769 break;
9cdda8d3 5770 if (found_key.offset < ctx->pos)
b9e03af0 5771 goto next;
16cdcec7
MX		 5772 if (key_type == BTRFS_DIR_INDEX_KEY &&
5773 btrfs_should_delete_dir_index(&del_list,
5774 found_key.offset))
5775 goto next;
5f39d397 5776
9cdda8d3 5777 ctx->pos = found_key.offset;
16cdcec7 5778 is_curr = 1;
49593bfa 5779
39279cc3
CM		 5780 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5781 di_cur = 0;
5f39d397 5782 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 5783
5784 while (di_cur < di_total) {
5f39d397
CM		 5785 struct btrfs_key location;
5786
22a94d44
JB		 5787 if (verify_dir_item(root, leaf, di))
5788 break;
5789
5f39d397 5790 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 5791 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 5792 name_ptr = tmp_name;
5793 } else {
5794 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 5795 if (!name_ptr) {
5796 ret = -ENOMEM;
5797 goto err;
5798 }
5f39d397
CM		 5799 }
5800 read_extent_buffer(leaf, name_ptr,
5801 (unsigned long)(di + 1), name_len);
5802
5803 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5804 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c 5805
fede766f 5806
3de4586c 5807 /* is this a reference to our own snapshot? If so
8c9c2bf7
AJ		 5808 * skip it.
5809 *
5810 * In contrast to old kernels, we insert the snapshot's
5811 * dir item and dir index after it has been created, so
5812 * we won't find a reference to our own snapshot. We
5813 * still keep the following code for backward
5814 * compatibility.
3de4586c
CM		 5815 */
5816 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5817 location.objectid == root->root_key.objectid) {
5818 over = 0;
5819 goto skip;
5820 }
9cdda8d3
AV		 5821 over = !dir_emit(ctx, name_ptr, name_len,
5822 location.objectid, d_type);
5f39d397 5823
3de4586c 5824skip:
5f39d397
CM		 5825 if (name_ptr != tmp_name)
5826 kfree(name_ptr);
5827
39279cc3
CM		 5828 if (over)
5829 goto nopos;
5103e947 5830 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 5831 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 5832 di_cur += di_len;
5833 di = (struct btrfs_dir_item *)((char *)di + di_len);
5834 }
b9e03af0
LZ		 5835next:
5836 path->slots[0]++;
39279cc3 5837 }
49593bfa 5838
16cdcec7
MX		 5839 if (key_type == BTRFS_DIR_INDEX_KEY) {
5840 if (is_curr)
9cdda8d3
AV		 5841 ctx->pos++;
5842 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
16cdcec7
MX		 5843 if (ret)
5844 goto nopos;
5845 }
5846
49593bfa 5847 /* Reached end of directory/root. Bump pos past the last item. */
db62efbb
ZB		 5848 ctx->pos++;
5849
5850 /*
5851 * Stop new entries from being returned after we return the last
5852 * entry.
5853 *
5854 * New directory entries are assigned a strictly increasing
5855 * offset. This means that new entries created during readdir
5856 * are *guaranteed* to be seen in the future by that readdir.
5857 * This has broken buggy programs which operate on names as
5858 * they're returned by readdir. Until we re-use freed offsets
5859 * we have this hack to stop new entries from being returned
5860 * under the assumption that they'll never reach this huge
5861 * offset.
5862 *
5863 * This is being careful not to overflow 32bit loff_t unless the
5864 * last entry requires it because doing so has broken 32bit apps
5865 * in the past.
5866 */
5867 if (key_type == BTRFS_DIR_INDEX_KEY) {
5868 if (ctx->pos >= INT_MAX)
5869 ctx->pos = LLONG_MAX;
5870 else
5871 ctx->pos = INT_MAX;
5872 }
39279cc3
CM		 5873nopos:
5874 ret = 0;
5875err:
16cdcec7
MX		 5876 if (key_type == BTRFS_DIR_INDEX_KEY)
5877 btrfs_put_delayed_items(&ins_list, &del_list);
39279cc3 5878 btrfs_free_path(path);
39279cc3
CM		 5879 return ret;
5880}
5881
a9185b41 5882int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
39279cc3
CM		 5883{
5884 struct btrfs_root *root = BTRFS_I(inode)->root;
5885 struct btrfs_trans_handle *trans;
5886 int ret = 0;
0af3d00b 5887 bool nolock = false;
39279cc3 5888
72ac3c0d 5889 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4ca8b41e
CM		 5890 return 0;
5891
83eea1f1 5892 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
82d5902d 5893 nolock = true;
0af3d00b 5894
a9185b41 5895 if (wbc->sync_mode == WB_SYNC_ALL) {
0af3d00b 5896 if (nolock)
7a7eaa40 5897 trans = btrfs_join_transaction_nolock(root);
0af3d00b 5898 else
7a7eaa40 5899 trans = btrfs_join_transaction(root);
3612b495
TI		 5900 if (IS_ERR(trans))
5901 return PTR_ERR(trans);
a698d075 5902 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 5903 }
5904 return ret;
5905}
5906
5907/*
54aa1f4d 5908 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 5909 * inode changes. But, it is most likely to find the inode in cache.
5910 * FIXME, needs more benchmarking...there are no reasons other than performance
5911 * to keep or drop this code.
5912 */
48a3b636 5913static int btrfs_dirty_inode(struct inode *inode)
39279cc3
CM		 5914{
5915 struct btrfs_root *root = BTRFS_I(inode)->root;
5916 struct btrfs_trans_handle *trans;
8929ecfa
YZ		 5917 int ret;
5918
72ac3c0d 5919 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
22c44fe6 5920 return 0;
39279cc3 5921
7a7eaa40 5922 trans = btrfs_join_transaction(root);
22c44fe6
JB		 5923 if (IS_ERR(trans))
5924 return PTR_ERR(trans);
8929ecfa
YZ		 5925
5926 ret = btrfs_update_inode(trans, root, inode);
94b60442
CM		 5927 if (ret && ret == -ENOSPC) {
5928 /* whoops, lets try again with the full transaction */
5929 btrfs_end_transaction(trans, root);
5930 trans = btrfs_start_transaction(root, 1);
22c44fe6
JB		 5931 if (IS_ERR(trans))
5932 return PTR_ERR(trans);
8929ecfa 5933
94b60442 5934 ret = btrfs_update_inode(trans, root, inode);
94b60442 5935 }
39279cc3 5936 btrfs_end_transaction(trans, root);
16cdcec7
MX		 5937 if (BTRFS_I(inode)->delayed_node)
5938 btrfs_balance_delayed_items(root);
22c44fe6
JB		 5939
5940 return ret;
5941}
5942
5943/*
5944 * This is a copy of file_update_time. We need this so we can return error on
5945 * ENOSPC for updating the inode in the case of file write and mmap writes.
5946 */
e41f941a
JB		 5947static int btrfs_update_time(struct inode *inode, struct timespec *now,
5948 int flags)
22c44fe6 5949{
2bc55652
AB		 5950 struct btrfs_root *root = BTRFS_I(inode)->root;
5951
5952 if (btrfs_root_readonly(root))
5953 return -EROFS;
5954
e41f941a 5955 if (flags & S_VERSION)
22c44fe6 5956 inode_inc_iversion(inode);
e41f941a
JB		 5957 if (flags & S_CTIME)
5958 inode->i_ctime = *now;
5959 if (flags & S_MTIME)
5960 inode->i_mtime = *now;
5961 if (flags & S_ATIME)
5962 inode->i_atime = *now;
5963 return btrfs_dirty_inode(inode);
39279cc3
CM		 5964}
5965
d352ac68
CM		 5966/*
5967 * find the highest existing sequence number in a directory
5968 * and then set the in-memory index_cnt variable to reflect
5969 * free sequence numbers
5970 */
aec7477b
JB		 5971static int btrfs_set_inode_index_count(struct inode *inode)
5972{
5973 struct btrfs_root *root = BTRFS_I(inode)->root;
5974 struct btrfs_key key, found_key;
5975 struct btrfs_path *path;
5976 struct extent_buffer *leaf;
5977 int ret;
5978
33345d01 5979 key.objectid = btrfs_ino(inode);
962a298f 5980 key.type = BTRFS_DIR_INDEX_KEY;
aec7477b
JB		 5981 key.offset = (u64)-1;
5982
5983 path = btrfs_alloc_path();
5984 if (!path)
5985 return -ENOMEM;
5986
5987 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5988 if (ret < 0)
5989 goto out;
5990 /* FIXME: we should be able to handle this */
5991 if (ret == 0)
5992 goto out;
5993 ret = 0;
5994
5995 /*
5996 * MAGIC NUMBER EXPLANATION:
5997 * since we search a directory based on f_pos we have to start at 2
5998 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5999 * else has to start at 2
6000 */
6001 if (path->slots[0] == 0) {
6002 BTRFS_I(inode)->index_cnt = 2;
6003 goto out;
6004 }
6005
6006 path->slots[0]--;
6007
6008 leaf = path->nodes[0];
6009 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6010
33345d01 6011 if (found_key.objectid != btrfs_ino(inode) ||
962a298f 6012 found_key.type != BTRFS_DIR_INDEX_KEY) {
aec7477b
JB		 6013 BTRFS_I(inode)->index_cnt = 2;
6014 goto out;
6015 }
6016
6017 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
6018out:
6019 btrfs_free_path(path);
6020 return ret;
6021}
6022
d352ac68
CM		 6023/*
6024 * helper to find a free sequence number in a given directory. This current
6025 * code is very simple, later versions will do smarter things in the btree
6026 */
3de4586c 6027int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 6028{
6029 int ret = 0;
6030
6031 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
16cdcec7
MX		 6032 ret = btrfs_inode_delayed_dir_index_count(dir);
6033 if (ret) {
6034 ret = btrfs_set_inode_index_count(dir);
6035 if (ret)
6036 return ret;
6037 }
aec7477b
JB		 6038 }
6039
00e4e6b3 6040 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 6041 BTRFS_I(dir)->index_cnt++;
6042
6043 return ret;
6044}
6045
b0d5d10f
CM		 6046static int btrfs_insert_inode_locked(struct inode *inode)
6047{
6048 struct btrfs_iget_args args;
6049 args.location = &BTRFS_I(inode)->location;
6050 args.root = BTRFS_I(inode)->root;
6051
6052 return insert_inode_locked4(inode,
6053 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6054 btrfs_find_actor, &args);
6055}
6056
39279cc3
CM		 6057static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6058 struct btrfs_root *root,
aec7477b 6059 struct inode *dir,
9c58309d 6060 const char *name, int name_len,
175a4eb7
AV		 6061 u64 ref_objectid, u64 objectid,
6062 umode_t mode, u64 *index)
39279cc3
CM		 6063{
6064 struct inode *inode;
5f39d397 6065 struct btrfs_inode_item *inode_item;
39279cc3 6066 struct btrfs_key *location;
5f39d397 6067 struct btrfs_path *path;
9c58309d
CM		 6068 struct btrfs_inode_ref *ref;
6069 struct btrfs_key key[2];
6070 u32 sizes[2];
ef3b9af5 6071 int nitems = name ? 2 : 1;
9c58309d 6072 unsigned long ptr;
39279cc3 6073 int ret;
39279cc3 6074
5f39d397 6075 path = btrfs_alloc_path();
d8926bb3
MF		 6076 if (!path)
6077 return ERR_PTR(-ENOMEM);
5f39d397 6078
39279cc3 6079 inode = new_inode(root->fs_info->sb);
8fb27640
YS		 6080 if (!inode) {
6081 btrfs_free_path(path);
39279cc3 6082 return ERR_PTR(-ENOMEM);
8fb27640 6083 }
39279cc3 6084
5762b5c9
FM		 6085 /*
6086 * O_TMPFILE, set link count to 0, so that after this point,
6087 * we fill in an inode item with the correct link count.
6088 */
6089 if (!name)
6090 set_nlink(inode, 0);
6091
581bb050
LZ		 6092 /*
6093 * we have to initialize this early, so we can reclaim the inode
6094 * number if we fail afterwards in this function.
6095 */
6096 inode->i_ino = objectid;
6097
ef3b9af5 6098 if (dir && name) {
1abe9b8a 6099 trace_btrfs_inode_request(dir);
6100
3de4586c 6101 ret = btrfs_set_inode_index(dir, index);
09771430 6102 if (ret) {
8fb27640 6103 btrfs_free_path(path);
09771430 6104 iput(inode);
aec7477b 6105 return ERR_PTR(ret);
09771430 6106 }
ef3b9af5
FM		 6107 } else if (dir) {
6108 *index = 0;
aec7477b
JB		 6109 }
6110 /*
6111 * index_cnt is ignored for everything but a dir,
6112 * btrfs_get_inode_index_count has an explanation for the magic
6113 * number
6114 */
6115 BTRFS_I(inode)->index_cnt = 2;
67de1176 6116 BTRFS_I(inode)->dir_index = *index;
39279cc3 6117 BTRFS_I(inode)->root = root;
e02119d5 6118 BTRFS_I(inode)->generation = trans->transid;
76195853 6119 inode->i_generation = BTRFS_I(inode)->generation;
b888db2b 6120
5dc562c5
JB		 6121 /*
6122 * We could have gotten an inode number from somebody who was fsynced
6123 * and then removed in this same transaction, so let's just set full
6124 * sync since it will be a full sync anyway and this will blow away the
6125 * old info in the log.
6126 */
6127 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6128
9c58309d 6129 key[0].objectid = objectid;
962a298f 6130 key[0].type = BTRFS_INODE_ITEM_KEY;
9c58309d
CM		 6131 key[0].offset = 0;
6132
9c58309d 6133 sizes[0] = sizeof(struct btrfs_inode_item);
ef3b9af5
FM		 6134
6135 if (name) {
6136 /*
6137 * Start new inodes with an inode_ref. This is slightly more
6138 * efficient for small numbers of hard links since they will
6139 * be packed into one item. Extended refs will kick in if we
6140 * add more hard links than can fit in the ref item.
6141 */
6142 key[1].objectid = objectid;
962a298f 6143 key[1].type = BTRFS_INODE_REF_KEY;
ef3b9af5
FM		 6144 key[1].offset = ref_objectid;
6145
6146 sizes[1] = name_len + sizeof(*ref);
6147 }
9c58309d 6148
b0d5d10f
CM		 6149 location = &BTRFS_I(inode)->location;
6150 location->objectid = objectid;
6151 location->offset = 0;
962a298f 6152 location->type = BTRFS_INODE_ITEM_KEY;
b0d5d10f
CM		 6153
6154 ret = btrfs_insert_inode_locked(inode);
6155 if (ret < 0)
6156 goto fail;
6157
b9473439 6158 path->leave_spinning = 1;
ef3b9af5 6159 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
9c58309d 6160 if (ret != 0)
b0d5d10f 6161 goto fail_unlock;
5f39d397 6162
ecc11fab 6163 inode_init_owner(inode, dir, mode);
a76a3cd4 6164 inode_set_bytes(inode, 0);
9cc97d64 6165
6166 inode->i_mtime = CURRENT_TIME;
6167 inode->i_atime = inode->i_mtime;
6168 inode->i_ctime = inode->i_mtime;
6169 BTRFS_I(inode)->i_otime = inode->i_mtime;
6170
5f39d397
CM		 6171 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6172 struct btrfs_inode_item);
293f7e07
LZ		 6173 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6174 sizeof(*inode_item));
e02119d5 6175 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d 6176
ef3b9af5
FM		 6177 if (name) {
6178 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6179 struct btrfs_inode_ref);
6180 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6181 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6182 ptr = (unsigned long)(ref + 1);
6183 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6184 }
9c58309d 6185
5f39d397
CM		 6186 btrfs_mark_buffer_dirty(path->nodes[0]);
6187 btrfs_free_path(path);
6188
6cbff00f
CH		 6189 btrfs_inherit_iflags(inode, dir);
6190
569254b0 6191 if (S_ISREG(mode)) {
94272164
CM		 6192 if (btrfs_test_opt(root, NODATASUM))
6193 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
213490b3 6194 if (btrfs_test_opt(root, NODATACOW))
f2bdf9a8
JB		 6195 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6196 BTRFS_INODE_NODATASUM;
94272164
CM		 6197 }
6198
5d4f98a2 6199 inode_tree_add(inode);
1abe9b8a 6200
6201 trace_btrfs_inode_new(inode);
1973f0fa 6202 btrfs_set_inode_last_trans(trans, inode);
1abe9b8a 6203
8ea05e3a
AB		 6204 btrfs_update_root_times(trans, root);
6205
63541927
FDBM		 6206 ret = btrfs_inode_inherit_props(trans, inode, dir);
6207 if (ret)
6208 btrfs_err(root->fs_info,
6209 "error inheriting props for ino %llu (root %llu): %d",
6210 btrfs_ino(inode), root->root_key.objectid, ret);
6211
39279cc3 6212 return inode;
b0d5d10f
CM		 6213
6214fail_unlock:
6215 unlock_new_inode(inode);
5f39d397 6216fail:
ef3b9af5 6217 if (dir && name)
aec7477b 6218 BTRFS_I(dir)->index_cnt--;
5f39d397 6219 btrfs_free_path(path);
09771430 6220 iput(inode);
5f39d397 6221 return ERR_PTR(ret);
39279cc3
CM		 6222}
6223
6224static inline u8 btrfs_inode_type(struct inode *inode)
6225{
6226 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6227}
6228
d352ac68
CM		 6229/*
6230 * utility function to add 'inode' into 'parent_inode' with
6231 * a give name and a given sequence number.
6232 * if 'add_backref' is true, also insert a backref from the
6233 * inode to the parent directory.
6234 */
e02119d5
CM		 6235int btrfs_add_link(struct btrfs_trans_handle *trans,
6236 struct inode *parent_inode, struct inode *inode,
6237 const char *name, int name_len, int add_backref, u64 index)
39279cc3 6238{
4df27c4d 6239 int ret = 0;
39279cc3 6240 struct btrfs_key key;
e02119d5 6241 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
33345d01
LZ		 6242 u64 ino = btrfs_ino(inode);
6243 u64 parent_ino = btrfs_ino(parent_inode);
5f39d397 6244
33345d01 6245 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 6246 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6247 } else {
33345d01 6248 key.objectid = ino;
962a298f 6249 key.type = BTRFS_INODE_ITEM_KEY;
4df27c4d
YZ		 6250 key.offset = 0;
6251 }
6252
33345d01 6253 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 6254 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6255 key.objectid, root->root_key.objectid,
33345d01 6256 parent_ino, index, name, name_len);
4df27c4d 6257 } else if (add_backref) {
33345d01
LZ		 6258 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6259 parent_ino, index);
4df27c4d 6260 }
39279cc3 6261
79787eaa
JM		 6262 /* Nothing to clean up yet */
6263 if (ret)
6264 return ret;
4df27c4d 6265
79787eaa
JM		 6266 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6267 parent_inode, &key,
6268 btrfs_inode_type(inode), index);
9c52057c 6269 if (ret == -EEXIST || ret == -EOVERFLOW)
79787eaa
JM		 6270 goto fail_dir_item;
6271 else if (ret) {
6272 btrfs_abort_transaction(trans, root, ret);
6273 return ret;
39279cc3 6274 }
79787eaa
JM		 6275
6276 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6277 name_len * 2);
0c4d2d95 6278 inode_inc_iversion(parent_inode);
79787eaa
JM		 6279 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
6280 ret = btrfs_update_inode(trans, root, parent_inode);
6281 if (ret)
6282 btrfs_abort_transaction(trans, root, ret);
39279cc3 6283 return ret;
fe66a05a
CM		 6284
6285fail_dir_item:
6286 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6287 u64 local_index;
6288 int err;
6289 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6290 key.objectid, root->root_key.objectid,
6291 parent_ino, &local_index, name, name_len);
6292
6293 } else if (add_backref) {
6294 u64 local_index;
6295 int err;
6296
6297 err = btrfs_del_inode_ref(trans, root, name, name_len,
6298 ino, parent_ino, &local_index);
6299 }
6300 return ret;
39279cc3
CM		 6301}
6302
6303static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
a1b075d2
JB		 6304 struct inode *dir, struct dentry *dentry,
6305 struct inode *inode, int backref, u64 index)
39279cc3 6306{
a1b075d2
JB		 6307 int err = btrfs_add_link(trans, dir, inode,
6308 dentry->d_name.name, dentry->d_name.len,
6309 backref, index);
39279cc3
CM		 6310 if (err > 0)
6311 err = -EEXIST;
6312 return err;
6313}
6314
618e21d5 6315static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1a67aafb 6316 umode_t mode, dev_t rdev)
618e21d5
JB		 6317{
6318 struct btrfs_trans_handle *trans;
6319 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 6320 struct inode *inode = NULL;
618e21d5
JB		 6321 int err;
6322 int drop_inode = 0;
6323 u64 objectid;
00e4e6b3 6324 u64 index = 0;
618e21d5 6325
9ed74f2d
JB		 6326 /*
6327 * 2 for inode item and ref
6328 * 2 for dir items
6329 * 1 for xattr if selinux is on
6330 */
a22285a6
YZ		 6331 trans = btrfs_start_transaction(root, 5);
6332 if (IS_ERR(trans))
6333 return PTR_ERR(trans);
1832a6d5 6334
581bb050
LZ		 6335 err = btrfs_find_free_ino(root, &objectid);
6336 if (err)
6337 goto out_unlock;
6338
aec7477b 6339 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 6340 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 6341 mode, &index);
7cf96da3
TI		 6342 if (IS_ERR(inode)) {
6343 err = PTR_ERR(inode);
618e21d5 6344 goto out_unlock;
7cf96da3 6345 }
618e21d5 6346
ad19db71
CS		 6347 /*
6348 * If the active LSM wants to access the inode during
6349 * d_instantiate it needs these. Smack checks to see
6350 * if the filesystem supports xattrs by looking at the
6351 * ops vector.
6352 */
ad19db71 6353 inode->i_op = &btrfs_special_inode_operations;
b0d5d10f
CM		 6354 init_special_inode(inode, inode->i_mode, rdev);
6355
6356 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
618e21d5 6357 if (err)
b0d5d10f
CM		 6358 goto out_unlock_inode;
6359
6360 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6361 if (err) {
6362 goto out_unlock_inode;
6363 } else {
1b4ab1bb 6364 btrfs_update_inode(trans, root, inode);
b0d5d10f 6365 unlock_new_inode(inode);
08c422c2 6366 d_instantiate(dentry, inode);
618e21d5 6367 }
b0d5d10f 6368
618e21d5 6369out_unlock:
7ad85bb7 6370 btrfs_end_transaction(trans, root);
c581afc8 6371 btrfs_balance_delayed_items(root);
b53d3f5d 6372 btrfs_btree_balance_dirty(root);
618e21d5
JB		 6373 if (drop_inode) {
6374 inode_dec_link_count(inode);
6375 iput(inode);
6376 }
618e21d5 6377 return err;
b0d5d10f
CM		 6378
6379out_unlock_inode:
6380 drop_inode = 1;
6381 unlock_new_inode(inode);
6382 goto out_unlock;
6383
618e21d5
JB		 6384}
6385
39279cc3 6386static int btrfs_create(struct inode *dir, struct dentry *dentry,
ebfc3b49 6387 umode_t mode, bool excl)
39279cc3
CM		 6388{
6389 struct btrfs_trans_handle *trans;
6390 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 6391 struct inode *inode = NULL;
43baa579 6392 int drop_inode_on_err = 0;
a22285a6 6393 int err;
39279cc3 6394 u64 objectid;
00e4e6b3 6395 u64 index = 0;
39279cc3 6396
9ed74f2d
JB		 6397 /*
6398 * 2 for inode item and ref
6399 * 2 for dir items
6400 * 1 for xattr if selinux is on
6401 */
a22285a6
YZ		 6402 trans = btrfs_start_transaction(root, 5);
6403 if (IS_ERR(trans))
6404 return PTR_ERR(trans);
9ed74f2d 6405
581bb050
LZ		 6406 err = btrfs_find_free_ino(root, &objectid);
6407 if (err)
6408 goto out_unlock;
6409
aec7477b 6410 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 6411 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 6412 mode, &index);
7cf96da3
TI		 6413 if (IS_ERR(inode)) {
6414 err = PTR_ERR(inode);
39279cc3 6415 goto out_unlock;
7cf96da3 6416 }
43baa579 6417 drop_inode_on_err = 1;
ad19db71
CS		 6418 /*
6419 * If the active LSM wants to access the inode during
6420 * d_instantiate it needs these. Smack checks to see
6421 * if the filesystem supports xattrs by looking at the
6422 * ops vector.
6423 */
6424 inode->i_fop = &btrfs_file_operations;
6425 inode->i_op = &btrfs_file_inode_operations;
b0d5d10f 6426 inode->i_mapping->a_ops = &btrfs_aops;
b0d5d10f
CM		 6427
6428 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6429 if (err)
6430 goto out_unlock_inode;
6431
6432 err = btrfs_update_inode(trans, root, inode);
6433 if (err)
6434 goto out_unlock_inode;
ad19db71 6435
a1b075d2 6436 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3 6437 if (err)
b0d5d10f 6438 goto out_unlock_inode;
43baa579 6439
43baa579 6440 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
b0d5d10f 6441 unlock_new_inode(inode);
43baa579
FB		 6442 d_instantiate(dentry, inode);
6443
39279cc3 6444out_unlock:
7ad85bb7 6445 btrfs_end_transaction(trans, root);
43baa579 6446 if (err && drop_inode_on_err) {
39279cc3
CM		 6447 inode_dec_link_count(inode);
6448 iput(inode);
6449 }
c581afc8 6450 btrfs_balance_delayed_items(root);
b53d3f5d 6451 btrfs_btree_balance_dirty(root);
39279cc3 6452 return err;
b0d5d10f
CM		 6453
6454out_unlock_inode:
6455 unlock_new_inode(inode);
6456 goto out_unlock;
6457
39279cc3
CM		 6458}
6459
6460static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6461 struct dentry *dentry)
6462{
271dba45 6463 struct btrfs_trans_handle *trans = NULL;
39279cc3 6464 struct btrfs_root *root = BTRFS_I(dir)->root;
2b0143b5 6465 struct inode *inode = d_inode(old_dentry);
00e4e6b3 6466 u64 index;
39279cc3
CM		 6467 int err;
6468 int drop_inode = 0;
6469
4a8be425
TH		 6470 /* do not allow sys_link's with other subvols of the same device */
6471 if (root->objectid != BTRFS_I(inode)->root->objectid)
3ab3564f 6472 return -EXDEV;
4a8be425 6473
f186373f 6474 if (inode->i_nlink >= BTRFS_LINK_MAX)
c055e99e 6475 return -EMLINK;
4a8be425 6476
3de4586c 6477 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 6478 if (err)
6479 goto fail;
6480
a22285a6 6481 /*
7e6b6465 6482 * 2 items for inode and inode ref
a22285a6 6483 * 2 items for dir items
7e6b6465 6484 * 1 item for parent inode
a22285a6 6485 */
7e6b6465 6486 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 6487 if (IS_ERR(trans)) {
6488 err = PTR_ERR(trans);
271dba45 6489 trans = NULL;
a22285a6
YZ		 6490 goto fail;
6491 }
5f39d397 6492
67de1176
MX		 6493 /* There are several dir indexes for this inode, clear the cache. */
6494 BTRFS_I(inode)->dir_index = 0ULL;
8b558c5f 6495 inc_nlink(inode);
0c4d2d95 6496 inode_inc_iversion(inode);
3153495d 6497 inode->i_ctime = CURRENT_TIME;
7de9c6ee 6498 ihold(inode);
e9976151 6499 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
aec7477b 6500
a1b075d2 6501 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5f39d397 6502
a5719521 6503 if (err) {
54aa1f4d 6504 drop_inode = 1;
a5719521 6505 } else {
10d9f309 6506 struct dentry *parent = dentry->d_parent;
a5719521 6507 err = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 6508 if (err)
6509 goto fail;
ef3b9af5
FM		 6510 if (inode->i_nlink == 1) {
6511 /*
6512 * If new hard link count is 1, it's a file created
6513 * with open(2) O_TMPFILE flag.
6514 */
6515 err = btrfs_orphan_del(trans, inode);
6516 if (err)
6517 goto fail;
6518 }
08c422c2 6519 d_instantiate(dentry, inode);
6a912213 6520 btrfs_log_new_name(trans, inode, NULL, parent);
a5719521 6521 }
39279cc3 6522
c581afc8 6523 btrfs_balance_delayed_items(root);
1832a6d5 6524fail:
271dba45
FM		 6525 if (trans)
6526 btrfs_end_transaction(trans, root);
39279cc3
CM		 6527 if (drop_inode) {
6528 inode_dec_link_count(inode);
6529 iput(inode);
6530 }
b53d3f5d 6531 btrfs_btree_balance_dirty(root);
39279cc3
CM		 6532 return err;
6533}
6534
18bb1db3 6535static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
39279cc3 6536{
b9d86667 6537 struct inode *inode = NULL;
39279cc3
CM		 6538 struct btrfs_trans_handle *trans;
6539 struct btrfs_root *root = BTRFS_I(dir)->root;
6540 int err = 0;
6541 int drop_on_err = 0;
b9d86667 6542 u64 objectid = 0;
00e4e6b3 6543 u64 index = 0;
39279cc3 6544
9ed74f2d
JB		 6545 /*
6546 * 2 items for inode and ref
6547 * 2 items for dir items
6548 * 1 for xattr if selinux is on
6549 */
a22285a6
YZ		 6550 trans = btrfs_start_transaction(root, 5);
6551 if (IS_ERR(trans))
6552 return PTR_ERR(trans);
39279cc3 6553
581bb050
LZ		 6554 err = btrfs_find_free_ino(root, &objectid);
6555 if (err)
6556 goto out_fail;
6557
aec7477b 6558 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 6559 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 6560 S_IFDIR | mode, &index);
39279cc3
CM		 6561 if (IS_ERR(inode)) {
6562 err = PTR_ERR(inode);
6563 goto out_fail;
6564 }
5f39d397 6565
39279cc3 6566 drop_on_err = 1;
b0d5d10f
CM		 6567 /* these must be set before we unlock the inode */
6568 inode->i_op = &btrfs_dir_inode_operations;
6569 inode->i_fop = &btrfs_dir_file_operations;
33268eaf 6570
2a7dba39 6571 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf 6572 if (err)
b0d5d10f 6573 goto out_fail_inode;
39279cc3 6574
dbe674a9 6575 btrfs_i_size_write(inode, 0);
39279cc3
CM		 6576 err = btrfs_update_inode(trans, root, inode);
6577 if (err)
b0d5d10f 6578 goto out_fail_inode;
5f39d397 6579
a1b075d2
JB		 6580 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6581 dentry->d_name.len, 0, index);
39279cc3 6582 if (err)
b0d5d10f 6583 goto out_fail_inode;
5f39d397 6584
39279cc3 6585 d_instantiate(dentry, inode);
b0d5d10f
CM		 6586 /*
6587 * mkdir is special. We're unlocking after we call d_instantiate
6588 * to avoid a race with nfsd calling d_instantiate.
6589 */
6590 unlock_new_inode(inode);
39279cc3 6591 drop_on_err = 0;
39279cc3
CM		 6592
6593out_fail:
7ad85bb7 6594 btrfs_end_transaction(trans, root);
c7cfb8a5
WS		 6595 if (drop_on_err) {
6596 inode_dec_link_count(inode);
39279cc3 6597 iput(inode);
c7cfb8a5 6598 }
c581afc8 6599 btrfs_balance_delayed_items(root);
b53d3f5d 6600 btrfs_btree_balance_dirty(root);
39279cc3 6601 return err;
b0d5d10f
CM		 6602
6603out_fail_inode:
6604 unlock_new_inode(inode);
6605 goto out_fail;
39279cc3
CM		 6606}
6607
e6c4efd8
QW		 6608/* Find next extent map of a given extent map, caller needs to ensure locks */
6609static struct extent_map *next_extent_map(struct extent_map *em)
6610{
6611 struct rb_node *next;
6612
6613 next = rb_next(&em->rb_node);
6614 if (!next)
6615 return NULL;
6616 return container_of(next, struct extent_map, rb_node);
6617}
6618
6619static struct extent_map *prev_extent_map(struct extent_map *em)
6620{
6621 struct rb_node *prev;
6622
6623 prev = rb_prev(&em->rb_node);
6624 if (!prev)
6625 return NULL;
6626 return container_of(prev, struct extent_map, rb_node);
6627}
6628
d352ac68 6629/* helper for btfs_get_extent. Given an existing extent in the tree,
e6c4efd8 6630 * the existing extent is the nearest extent to map_start,
d352ac68 6631 * and an extent that you want to insert, deal with overlap and insert
e6c4efd8 6632 * the best fitted new extent into the tree.
d352ac68 6633 */
3b951516
CM		 6634static int merge_extent_mapping(struct extent_map_tree *em_tree,
6635 struct extent_map *existing,
e6dcd2dc 6636 struct extent_map *em,
51f395ad 6637 u64 map_start)
3b951516 6638{
e6c4efd8
QW		 6639 struct extent_map *prev;
6640 struct extent_map *next;
6641 u64 start;
6642 u64 end;
3b951516 6643 u64 start_diff;
3b951516 6644
e6dcd2dc 6645 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
e6c4efd8
QW		 6646
6647 if (existing->start > map_start) {
6648 next = existing;
6649 prev = prev_extent_map(next);
6650 } else {
6651 prev = existing;
6652 next = next_extent_map(prev);
6653 }
6654
6655 start = prev ? extent_map_end(prev) : em->start;
6656 start = max_t(u64, start, em->start);
6657 end = next ? next->start : extent_map_end(em);
6658 end = min_t(u64, end, extent_map_end(em));
6659 start_diff = start - em->start;
6660 em->start = start;
6661 em->len = end - start;
c8b97818
CM		 6662 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6663 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 6664 em->block_start += start_diff;
c8b97818
CM		 6665 em->block_len -= start_diff;
6666 }
09a2a8f9 6667 return add_extent_mapping(em_tree, em, 0);
3b951516
CM		 6668}
6669
c8b97818 6670static noinline int uncompress_inline(struct btrfs_path *path,
e40da0e5 6671 struct page *page,
c8b97818
CM		 6672 size_t pg_offset, u64 extent_offset,
6673 struct btrfs_file_extent_item *item)
6674{
6675 int ret;
6676 struct extent_buffer *leaf = path->nodes[0];
6677 char *tmp;
6678 size_t max_size;
6679 unsigned long inline_size;
6680 unsigned long ptr;
261507a0 6681 int compress_type;
c8b97818
CM		 6682
6683 WARN_ON(pg_offset != 0);
261507a0 6684 compress_type = btrfs_file_extent_compression(leaf, item);
c8b97818
CM		 6685 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6686 inline_size = btrfs_file_extent_inline_item_len(leaf,
dd3cc16b 6687 btrfs_item_nr(path->slots[0]));
c8b97818 6688 tmp = kmalloc(inline_size, GFP_NOFS);
8d413713
TI		 6689 if (!tmp)
6690 return -ENOMEM;
c8b97818
CM		 6691 ptr = btrfs_file_extent_inline_start(item);
6692
6693 read_extent_buffer(leaf, tmp, ptr, inline_size);
6694
5b050f04 6695 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
261507a0
LZ		 6696 ret = btrfs_decompress(compress_type, tmp, page,
6697 extent_offset, inline_size, max_size);
c8b97818 6698 kfree(tmp);
166ae5a4 6699 return ret;
c8b97818
CM		 6700}
6701
d352ac68
CM		 6702/*
6703 * a bit scary, this does extent mapping from logical file offset to the disk.
d397712b
CM		 6704 * the ugly parts come from merging extents from the disk with the in-ram
6705 * representation. This gets more complex because of the data=ordered code,
d352ac68
CM		 6706 * where the in-ram extents might be locked pending data=ordered completion.
6707 *
6708 * This also copies inline extents directly into the page.
6709 */
d397712b 6710
a52d9a80 6711struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 6712 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 6713 int create)
6714{
6715 int ret;
6716 int err = 0;
a52d9a80
CM		 6717 u64 extent_start = 0;
6718 u64 extent_end = 0;
33345d01 6719 u64 objectid = btrfs_ino(inode);
a52d9a80 6720 u32 found_type;
f421950f 6721 struct btrfs_path *path = NULL;
a52d9a80
CM		 6722 struct btrfs_root *root = BTRFS_I(inode)->root;
6723 struct btrfs_file_extent_item *item;
5f39d397
CM		 6724 struct extent_buffer *leaf;
6725 struct btrfs_key found_key;
a52d9a80
CM		 6726 struct extent_map *em = NULL;
6727 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 6728 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 6729 struct btrfs_trans_handle *trans = NULL;
7ffbb598 6730 const bool new_inline = !page || create;
a52d9a80 6731
a52d9a80 6732again:
890871be 6733 read_lock(&em_tree->lock);
d1310b2e 6734 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 6735 if (em)
6736 em->bdev = root->fs_info->fs_devices->latest_bdev;
890871be 6737 read_unlock(&em_tree->lock);
d1310b2e 6738
a52d9a80 6739 if (em) {
e1c4b745
CM		 6740 if (em->start > start || em->start + em->len <= start)
6741 free_extent_map(em);
6742 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 6743 free_extent_map(em);
6744 else
6745 goto out;
a52d9a80 6746 }
172ddd60 6747 em = alloc_extent_map();
a52d9a80 6748 if (!em) {
d1310b2e
CM		 6749 err = -ENOMEM;
6750 goto out;
a52d9a80 6751 }
e6dcd2dc 6752 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 6753 em->start = EXTENT_MAP_HOLE;
445a6944 6754 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 6755 em->len = (u64)-1;
c8b97818 6756 em->block_len = (u64)-1;
f421950f
CM		 6757
6758 if (!path) {
6759 path = btrfs_alloc_path();
026fd317
JB		 6760 if (!path) {
6761 err = -ENOMEM;
6762 goto out;
6763 }
6764 /*
6765 * Chances are we'll be called again, so go ahead and do
6766 * readahead
6767 */
e4058b54 6768 path->reada = READA_FORWARD;
f421950f
CM		 6769 }
6770
179e29e4
CM		 6771 ret = btrfs_lookup_file_extent(trans, root, path,
6772 objectid, start, trans != NULL);
a52d9a80
CM		 6773 if (ret < 0) {
6774 err = ret;
6775 goto out;
6776 }
6777
6778 if (ret != 0) {
6779 if (path->slots[0] == 0)
6780 goto not_found;
6781 path->slots[0]--;
6782 }
6783
5f39d397
CM		 6784 leaf = path->nodes[0];
6785 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 6786 struct btrfs_file_extent_item);
a52d9a80 6787 /* are we inside the extent that was found? */
5f39d397 6788 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
962a298f 6789 found_type = found_key.type;
5f39d397 6790 if (found_key.objectid != objectid ||
a52d9a80 6791 found_type != BTRFS_EXTENT_DATA_KEY) {
25a50341
JB		 6792 /*
6793 * If we backup past the first extent we want to move forward
6794 * and see if there is an extent in front of us, otherwise we'll
6795 * say there is a hole for our whole search range which can
6796 * cause problems.
6797 */
6798 extent_end = start;
6799 goto next;
a52d9a80
CM		 6800 }
6801
5f39d397
CM		 6802 found_type = btrfs_file_extent_type(leaf, item);
6803 extent_start = found_key.offset;
d899e052
YZ		 6804 if (found_type == BTRFS_FILE_EXTENT_REG ||
6805 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 6806 extent_end = extent_start +
db94535d 6807 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 6808 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6809 size_t size;
514ac8ad 6810 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
fda2832f 6811 extent_end = ALIGN(extent_start + size, root->sectorsize);
9036c102 6812 }
25a50341 6813next:
9036c102
YZ		 6814 if (start >= extent_end) {
6815 path->slots[0]++;
6816 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6817 ret = btrfs_next_leaf(root, path);
6818 if (ret < 0) {
6819 err = ret;
6820 goto out;
a52d9a80 6821 }
9036c102
YZ		 6822 if (ret > 0)
6823 goto not_found;
6824 leaf = path->nodes[0];
a52d9a80 6825 }
9036c102
YZ		 6826 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6827 if (found_key.objectid != objectid ||
6828 found_key.type != BTRFS_EXTENT_DATA_KEY)
6829 goto not_found;
6830 if (start + len <= found_key.offset)
6831 goto not_found;
e2eca69d
WS		 6832 if (start > found_key.offset)
6833 goto next;
9036c102 6834 em->start = start;
70c8a91c 6835 em->orig_start = start;
9036c102
YZ		 6836 em->len = found_key.offset - start;
6837 goto not_found_em;
6838 }
6839
7ffbb598
FM		 6840 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6841
d899e052
YZ		 6842 if (found_type == BTRFS_FILE_EXTENT_REG ||
6843 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80
CM		 6844 goto insert;
6845 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 6846 unsigned long ptr;
a52d9a80 6847 char *map;
3326d1b0
CM		 6848 size_t size;
6849 size_t extent_offset;
6850 size_t copy_size;
a52d9a80 6851
7ffbb598 6852 if (new_inline)
689f9346 6853 goto out;
5f39d397 6854
514ac8ad 6855 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
9036c102 6856 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 6857 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 6858 size - extent_offset);
3326d1b0 6859 em->start = extent_start + extent_offset;
fda2832f 6860 em->len = ALIGN(copy_size, root->sectorsize);
b4939680 6861 em->orig_block_len = em->len;
70c8a91c 6862 em->orig_start = em->start;
689f9346 6863 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 6864 if (create == 0 && !PageUptodate(page)) {
261507a0
LZ		 6865 if (btrfs_file_extent_compression(leaf, item) !=
6866 BTRFS_COMPRESS_NONE) {
e40da0e5 6867 ret = uncompress_inline(path, page, pg_offset,
c8b97818 6868 extent_offset, item);
166ae5a4
ZB		 6869 if (ret) {
6870 err = ret;
6871 goto out;
6872 }
c8b97818
CM		 6873 } else {
6874 map = kmap(page);
6875 read_extent_buffer(leaf, map + pg_offset, ptr,
6876 copy_size);
93c82d57
CM		 6877 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6878 memset(map + pg_offset + copy_size, 0,
6879 PAGE_CACHE_SIZE - pg_offset -
6880 copy_size);
6881 }
c8b97818
CM		 6882 kunmap(page);
6883 }
179e29e4
CM		 6884 flush_dcache_page(page);
6885 } else if (create && PageUptodate(page)) {
6bf7e080 6886 BUG();
179e29e4
CM		 6887 if (!trans) {
6888 kunmap(page);
6889 free_extent_map(em);
6890 em = NULL;
ff5714cc 6891
b3b4aa74 6892 btrfs_release_path(path);
7a7eaa40 6893 trans = btrfs_join_transaction(root);
ff5714cc 6894
3612b495
TI		 6895 if (IS_ERR(trans))
6896 return ERR_CAST(trans);
179e29e4
CM		 6897 goto again;
6898 }
c8b97818 6899 map = kmap(page);
70dec807 6900 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 6901 copy_size);
c8b97818 6902 kunmap(page);
179e29e4 6903 btrfs_mark_buffer_dirty(leaf);
a52d9a80 6904 }
d1310b2e 6905 set_extent_uptodate(io_tree, em->start,
507903b8 6906 extent_map_end(em) - 1, NULL, GFP_NOFS);
a52d9a80 6907 goto insert;
a52d9a80
CM		 6908 }
6909not_found:
6910 em->start = start;
70c8a91c 6911 em->orig_start = start;
d1310b2e 6912 em->len = len;
a52d9a80 6913not_found_em:
5f39d397 6914 em->block_start = EXTENT_MAP_HOLE;
9036c102 6915 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80 6916insert:
b3b4aa74 6917 btrfs_release_path(path);
d1310b2e 6918 if (em->start > start || extent_map_end(em) <= start) {
c2cf52eb 6919 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
c1c9ff7c 6920 em->start, em->len, start, len);
a52d9a80
CM		 6921 err = -EIO;
6922 goto out;
6923 }
d1310b2e
CM		 6924
6925 err = 0;
890871be 6926 write_lock(&em_tree->lock);
09a2a8f9 6927 ret = add_extent_mapping(em_tree, em, 0);
3b951516
CM		 6928 /* it is possible that someone inserted the extent into the tree
6929 * while we had the lock dropped. It is also possible that
6930 * an overlapping map exists in the tree
6931 */
a52d9a80 6932 if (ret == -EEXIST) {
3b951516 6933 struct extent_map *existing;
e6dcd2dc
CM		 6934
6935 ret = 0;
6936
e6c4efd8
QW		 6937 existing = search_extent_mapping(em_tree, start, len);
6938 /*
6939 * existing will always be non-NULL, since there must be
6940 * extent causing the -EEXIST.
6941 */
6942 if (start >= extent_map_end(existing) ||
32be3a1a 6943 start <= existing->start) {
e6c4efd8
QW		 6944 /*
6945 * The existing extent map is the one nearest to
6946 * the [start, start + len) range which overlaps
6947 */
6948 err = merge_extent_mapping(em_tree, existing,
6949 em, start);
e1c4b745 6950 free_extent_map(existing);
e6c4efd8 6951 if (err) {
3b951516
CM		 6952 free_extent_map(em);
6953 em = NULL;
6954 }
6955 } else {
6956 free_extent_map(em);
6957 em = existing;
e6dcd2dc 6958 err = 0;
a52d9a80 6959 }
a52d9a80 6960 }
890871be 6961 write_unlock(&em_tree->lock);
a52d9a80 6962out:
1abe9b8a 6963
4cd8587c 6964 trace_btrfs_get_extent(root, em);
1abe9b8a 6965
527afb44 6966 btrfs_free_path(path);
a52d9a80
CM		 6967 if (trans) {
6968 ret = btrfs_end_transaction(trans, root);
d397712b 6969 if (!err)
a52d9a80
CM		 6970 err = ret;
6971 }
a52d9a80
CM		 6972 if (err) {
6973 free_extent_map(em);
a52d9a80
CM		 6974 return ERR_PTR(err);
6975 }
79787eaa 6976 BUG_ON(!em); /* Error is always set */
a52d9a80
CM		 6977 return em;
6978}
6979
ec29ed5b
CM		 6980struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6981 size_t pg_offset, u64 start, u64 len,
6982 int create)
6983{
6984 struct extent_map *em;
6985 struct extent_map *hole_em = NULL;
6986 u64 range_start = start;
6987 u64 end;
6988 u64 found;
6989 u64 found_end;
6990 int err = 0;
6991
6992 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6993 if (IS_ERR(em))
6994 return em;
6995 if (em) {
6996 /*
f9e4fb53
LB		 6997 * if our em maps to
6998 * - a hole or
6999 * - a pre-alloc extent,
7000 * there might actually be delalloc bytes behind it.
ec29ed5b 7001 */
f9e4fb53
LB		 7002 if (em->block_start != EXTENT_MAP_HOLE &&
7003 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
ec29ed5b
CM		 7004 return em;
7005 else
7006 hole_em = em;
7007 }
7008
7009 /* check to see if we've wrapped (len == -1 or similar) */
7010 end = start + len;
7011 if (end < start)
7012 end = (u64)-1;
7013 else
7014 end -= 1;
7015
7016 em = NULL;
7017
7018 /* ok, we didn't find anything, lets look for delalloc */
7019 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
7020 end, len, EXTENT_DELALLOC, 1);
7021 found_end = range_start + found;
7022 if (found_end < range_start)
7023 found_end = (u64)-1;
7024
7025 /*
7026 * we didn't find anything useful, return
7027 * the original results from get_extent()
7028 */
7029 if (range_start > end || found_end <= start) {
7030 em = hole_em;
7031 hole_em = NULL;
7032 goto out;
7033 }
7034
7035 /* adjust the range_start to make sure it doesn't
7036 * go backwards from the start they passed in
7037 */
67871254 7038 range_start = max(start, range_start);
ec29ed5b
CM		 7039 found = found_end - range_start;
7040
7041 if (found > 0) {
7042 u64 hole_start = start;
7043 u64 hole_len = len;
7044
172ddd60 7045 em = alloc_extent_map();
ec29ed5b
CM		 7046 if (!em) {
7047 err = -ENOMEM;
7048 goto out;
7049 }
7050 /*
7051 * when btrfs_get_extent can't find anything it
7052 * returns one huge hole
7053 *
7054 * make sure what it found really fits our range, and
7055 * adjust to make sure it is based on the start from
7056 * the caller
7057 */
7058 if (hole_em) {
7059 u64 calc_end = extent_map_end(hole_em);
7060
7061 if (calc_end <= start || (hole_em->start > end)) {
7062 free_extent_map(hole_em);
7063 hole_em = NULL;
7064 } else {
7065 hole_start = max(hole_em->start, start);
7066 hole_len = calc_end - hole_start;
7067 }
7068 }
7069 em->bdev = NULL;
7070 if (hole_em && range_start > hole_start) {
7071 /* our hole starts before our delalloc, so we
7072 * have to return just the parts of the hole
7073 * that go until the delalloc starts
7074 */
7075 em->len = min(hole_len,
7076 range_start - hole_start);
7077 em->start = hole_start;
7078 em->orig_start = hole_start;
7079 /*
7080 * don't adjust block start at all,
7081 * it is fixed at EXTENT_MAP_HOLE
7082 */
7083 em->block_start = hole_em->block_start;
7084 em->block_len = hole_len;
f9e4fb53
LB		 7085 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7086 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
ec29ed5b
CM		 7087 } else {
7088 em->start = range_start;
7089 em->len = found;
7090 em->orig_start = range_start;
7091 em->block_start = EXTENT_MAP_DELALLOC;
7092 em->block_len = found;
7093 }
7094 } else if (hole_em) {
7095 return hole_em;
7096 }
7097out:
7098
7099 free_extent_map(hole_em);
7100 if (err) {
7101 free_extent_map(em);
7102 return ERR_PTR(err);
7103 }
7104 return em;
7105}
7106
4b46fce2
JB		 7107static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7108 u64 start, u64 len)
7109{
7110 struct btrfs_root *root = BTRFS_I(inode)->root;
70c8a91c 7111 struct extent_map *em;
4b46fce2
JB		 7112 struct btrfs_key ins;
7113 u64 alloc_hint;
7114 int ret;
4b46fce2 7115
4b46fce2 7116 alloc_hint = get_extent_allocation_hint(inode, start, len);
00361589 7117 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
e570fd27 7118 alloc_hint, &ins, 1, 1);
00361589
JB		 7119 if (ret)
7120 return ERR_PTR(ret);
4b46fce2 7121
de0ee0ed
FM		 7122 /*
7123 * Create the ordered extent before the extent map. This is to avoid
7124 * races with the fast fsync path that would lead to it logging file
7125 * extent items that point to disk extents that were not yet written to.
7126 * The fast fsync path collects ordered extents into a local list and
7127 * then collects all the new extent maps, so we must create the ordered
7128 * extent first and make sure the fast fsync path collects any new
7129 * ordered extents after collecting new extent maps as well.
7130 * The fsync path simply can not rely on inode_dio_wait() because it
7131 * causes deadlock with AIO.
7132 */
4b46fce2
JB		 7133 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
7134 ins.offset, ins.offset, 0);
7135 if (ret) {
e570fd27 7136 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
00361589 7137 return ERR_PTR(ret);
4b46fce2 7138 }
00361589 7139
de0ee0ed
FM		 7140 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
7141 ins.offset, ins.offset, ins.offset, 0);
7142 if (IS_ERR(em)) {
7143 struct btrfs_ordered_extent *oe;
7144
7145 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7146 oe = btrfs_lookup_ordered_extent(inode, start);
7147 ASSERT(oe);
7148 if (WARN_ON(!oe))
7149 return em;
7150 set_bit(BTRFS_ORDERED_IOERR, &oe->flags);
7151 set_bit(BTRFS_ORDERED_IO_DONE, &oe->flags);
7152 btrfs_remove_ordered_extent(inode, oe);
7153 /* Once for our lookup and once for the ordered extents tree. */
7154 btrfs_put_ordered_extent(oe);
7155 btrfs_put_ordered_extent(oe);
7156 }
4b46fce2
JB		 7157 return em;
7158}
7159
46bfbb5c
CM		 7160/*
7161 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7162 * block must be cow'd
7163 */
00361589 7164noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7ee9e440
JB		 7165 u64 *orig_start, u64 *orig_block_len,
7166 u64 *ram_bytes)
46bfbb5c 7167{
00361589 7168 struct btrfs_trans_handle *trans;
46bfbb5c
CM		 7169 struct btrfs_path *path;
7170 int ret;
7171 struct extent_buffer *leaf;
7172 struct btrfs_root *root = BTRFS_I(inode)->root;
7b2b7085 7173 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
46bfbb5c
CM		 7174 struct btrfs_file_extent_item *fi;
7175 struct btrfs_key key;
7176 u64 disk_bytenr;
7177 u64 backref_offset;
7178 u64 extent_end;
7179 u64 num_bytes;
7180 int slot;
7181 int found_type;
7ee9e440 7182 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
e77751aa 7183
46bfbb5c
CM		 7184 path = btrfs_alloc_path();
7185 if (!path)
7186 return -ENOMEM;
7187
00361589 7188 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
46bfbb5c
CM		 7189 offset, 0);
7190 if (ret < 0)
7191 goto out;
7192
7193 slot = path->slots[0];
7194 if (ret == 1) {
7195 if (slot == 0) {
7196 /* can't find the item, must cow */
7197 ret = 0;
7198 goto out;
7199 }
7200 slot--;
7201 }
7202 ret = 0;
7203 leaf = path->nodes[0];
7204 btrfs_item_key_to_cpu(leaf, &key, slot);
33345d01 7205 if (key.objectid != btrfs_ino(inode) ||
46bfbb5c
CM		 7206 key.type != BTRFS_EXTENT_DATA_KEY) {
7207 /* not our file or wrong item type, must cow */
7208 goto out;
7209 }
7210
7211 if (key.offset > offset) {
7212 /* Wrong offset, must cow */
7213 goto out;
7214 }
7215
7216 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7217 found_type = btrfs_file_extent_type(leaf, fi);
7218 if (found_type != BTRFS_FILE_EXTENT_REG &&
7219 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7220 /* not a regular extent, must cow */
7221 goto out;
7222 }
7ee9e440
JB		 7223
7224 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7225 goto out;
7226
e77751aa
MX		 7227 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7228 if (extent_end <= offset)
7229 goto out;
7230
46bfbb5c 7231 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7ee9e440
JB		 7232 if (disk_bytenr == 0)
7233 goto out;
7234
7235 if (btrfs_file_extent_compression(leaf, fi) ||
7236 btrfs_file_extent_encryption(leaf, fi) ||
7237 btrfs_file_extent_other_encoding(leaf, fi))
7238 goto out;
7239
46bfbb5c
CM		 7240 backref_offset = btrfs_file_extent_offset(leaf, fi);
7241
7ee9e440
JB		 7242 if (orig_start) {
7243 *orig_start = key.offset - backref_offset;
7244 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7245 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7246 }
eb384b55 7247
46bfbb5c
CM		 7248 if (btrfs_extent_readonly(root, disk_bytenr))
7249 goto out;
7b2b7085
MX		 7250
7251 num_bytes = min(offset + *len, extent_end) - offset;
7252 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7253 u64 range_end;
7254
7255 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7256 ret = test_range_bit(io_tree, offset, range_end,
7257 EXTENT_DELALLOC, 0, NULL);
7258 if (ret) {
7259 ret = -EAGAIN;
7260 goto out;
7261 }
7262 }
7263
1bda19eb 7264 btrfs_release_path(path);
46bfbb5c
CM		 7265
7266 /*
7267 * look for other files referencing this extent, if we
7268 * find any we must cow
7269 */
00361589
JB		 7270 trans = btrfs_join_transaction(root);
7271 if (IS_ERR(trans)) {
7272 ret = 0;
46bfbb5c 7273 goto out;
00361589
JB		 7274 }
7275
7276 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7277 key.offset - backref_offset, disk_bytenr);
7278 btrfs_end_transaction(trans, root);
7279 if (ret) {
7280 ret = 0;
7281 goto out;
7282 }
46bfbb5c
CM		 7283
7284 /*
7285 * adjust disk_bytenr and num_bytes to cover just the bytes
7286 * in this extent we are about to write. If there
7287 * are any csums in that range we have to cow in order
7288 * to keep the csums correct
7289 */
7290 disk_bytenr += backref_offset;
7291 disk_bytenr += offset - key.offset;
46bfbb5c
CM		 7292 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7293 goto out;
7294 /*
7295 * all of the above have passed, it is safe to overwrite this extent
7296 * without cow
7297 */
eb384b55 7298 *len = num_bytes;
46bfbb5c
CM		 7299 ret = 1;
7300out:
7301 btrfs_free_path(path);
7302 return ret;
7303}
7304
fc4adbff
AG		 7305bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7306{
7307 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7308 int found = false;
7309 void **pagep = NULL;
7310 struct page *page = NULL;
7311 int start_idx;
7312 int end_idx;
7313
7314 start_idx = start >> PAGE_CACHE_SHIFT;
7315
7316 /*
7317 * end is the last byte in the last page. end == start is legal
7318 */
7319 end_idx = end >> PAGE_CACHE_SHIFT;
7320
7321 rcu_read_lock();
7322
7323 /* Most of the code in this while loop is lifted from
7324 * find_get_page. It's been modified to begin searching from a
7325 * page and return just the first page found in that range. If the
7326 * found idx is less than or equal to the end idx then we know that
7327 * a page exists. If no pages are found or if those pages are
7328 * outside of the range then we're fine (yay!) */
7329 while (page == NULL &&
7330 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7331 page = radix_tree_deref_slot(pagep);
7332 if (unlikely(!page))
7333 break;
7334
7335 if (radix_tree_exception(page)) {
809f9016
FM		 7336 if (radix_tree_deref_retry(page)) {
7337 page = NULL;
fc4adbff 7338 continue;
809f9016 7339 }
fc4adbff
AG		 7340 /*
7341 * Otherwise, shmem/tmpfs must be storing a swap entry
7342 * here as an exceptional entry: so return it without
7343 * attempting to raise page count.
7344 */
6fdef6d4 7345 page = NULL;
fc4adbff
AG		 7346 break; /* TODO: Is this relevant for this use case? */
7347 }
7348
91405151
FM		 7349 if (!page_cache_get_speculative(page)) {
7350 page = NULL;
fc4adbff 7351 continue;
91405151 7352 }
fc4adbff
AG		 7353
7354 /*
7355 * Has the page moved?
7356 * This is part of the lockless pagecache protocol. See
7357 * include/linux/pagemap.h for details.
7358 */
7359 if (unlikely(page != *pagep)) {
7360 page_cache_release(page);
7361 page = NULL;
7362 }
7363 }
7364
7365 if (page) {
7366 if (page->index <= end_idx)
7367 found = true;
7368 page_cache_release(page);
7369 }
7370
7371 rcu_read_unlock();
7372 return found;
7373}
7374
eb838e73
JB		 7375static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7376 struct extent_state **cached_state, int writing)
7377{
7378 struct btrfs_ordered_extent *ordered;
7379 int ret = 0;
7380
7381 while (1) {
7382 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
ff13db41 7383 cached_state);
eb838e73
JB		 7384 /*
7385 * We're concerned with the entire range that we're going to be
7386 * doing DIO to, so we need to make sure theres no ordered
7387 * extents in this range.
7388 */
7389 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7390 lockend - lockstart + 1);
7391
7392 /*
7393 * We need to make sure there are no buffered pages in this
7394 * range either, we could have raced between the invalidate in
7395 * generic_file_direct_write and locking the extent. The
7396 * invalidate needs to happen so that reads after a write do not
7397 * get stale data.
7398 */
fc4adbff
AG		 7399 if (!ordered &&
7400 (!writing ||
7401 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
eb838e73
JB		 7402 break;
7403
7404 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7405 cached_state, GFP_NOFS);
7406
7407 if (ordered) {
7408 btrfs_start_ordered_extent(inode, ordered, 1);
7409 btrfs_put_ordered_extent(ordered);
7410 } else {
eb838e73 7411 /*
b850ae14
FM		 7412 * We could trigger writeback for this range (and wait
7413 * for it to complete) and then invalidate the pages for
7414 * this range (through invalidate_inode_pages2_range()),
7415 * but that can lead us to a deadlock with a concurrent
7416 * call to readpages() (a buffered read or a defrag call
7417 * triggered a readahead) on a page lock due to an
7418 * ordered dio extent we created before but did not have
7419 * yet a corresponding bio submitted (whence it can not
7420 * complete), which makes readpages() wait for that
7421 * ordered extent to complete while holding a lock on
7422 * that page.
eb838e73 7423 */
b850ae14
FM		 7424 ret = -ENOTBLK;
7425 break;
eb838e73
JB		 7426 }
7427
7428 cond_resched();
7429 }
7430
7431 return ret;
7432}
7433
69ffb543
JB		 7434static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7435 u64 len, u64 orig_start,
7436 u64 block_start, u64 block_len,
cc95bef6
JB		 7437 u64 orig_block_len, u64 ram_bytes,
7438 int type)
69ffb543
JB		 7439{
7440 struct extent_map_tree *em_tree;
7441 struct extent_map *em;
7442 struct btrfs_root *root = BTRFS_I(inode)->root;
7443 int ret;
7444
7445 em_tree = &BTRFS_I(inode)->extent_tree;
7446 em = alloc_extent_map();
7447 if (!em)
7448 return ERR_PTR(-ENOMEM);
7449
7450 em->start = start;
7451 em->orig_start = orig_start;
2ab28f32
JB		 7452 em->mod_start = start;
7453 em->mod_len = len;
69ffb543
JB		 7454 em->len = len;
7455 em->block_len = block_len;
7456 em->block_start = block_start;
7457 em->bdev = root->fs_info->fs_devices->latest_bdev;
b4939680 7458 em->orig_block_len = orig_block_len;
cc95bef6 7459 em->ram_bytes = ram_bytes;
70c8a91c 7460 em->generation = -1;
69ffb543
JB		 7461 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7462 if (type == BTRFS_ORDERED_PREALLOC)
b11e234d 7463 set_bit(EXTENT_FLAG_FILLING, &em->flags);
69ffb543
JB		 7464
7465 do {
7466 btrfs_drop_extent_cache(inode, em->start,
7467 em->start + em->len - 1, 0);
7468 write_lock(&em_tree->lock);
09a2a8f9 7469 ret = add_extent_mapping(em_tree, em, 1);
69ffb543
JB		 7470 write_unlock(&em_tree->lock);
7471 } while (ret == -EEXIST);
7472
7473 if (ret) {
7474 free_extent_map(em);
7475 return ERR_PTR(ret);
7476 }
7477
7478 return em;
7479}
7480
9c9464cc
FM		 7481static void adjust_dio_outstanding_extents(struct inode *inode,
7482 struct btrfs_dio_data *dio_data,
7483 const u64 len)
7484{
7485 unsigned num_extents;
7486
7487 num_extents = (unsigned) div64_u64(len + BTRFS_MAX_EXTENT_SIZE - 1,
7488 BTRFS_MAX_EXTENT_SIZE);
7489 /*
7490 * If we have an outstanding_extents count still set then we're
7491 * within our reservation, otherwise we need to adjust our inode
7492 * counter appropriately.
7493 */
7494 if (dio_data->outstanding_extents) {
7495 dio_data->outstanding_extents -= num_extents;
7496 } else {
7497 spin_lock(&BTRFS_I(inode)->lock);
7498 BTRFS_I(inode)->outstanding_extents += num_extents;
7499 spin_unlock(&BTRFS_I(inode)->lock);
7500 }
7501}
7502
4b46fce2
JB		 7503static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7504 struct buffer_head *bh_result, int create)
7505{
7506 struct extent_map *em;
7507 struct btrfs_root *root = BTRFS_I(inode)->root;
eb838e73 7508 struct extent_state *cached_state = NULL;
50745b0a 7509 struct btrfs_dio_data *dio_data = NULL;
4b46fce2 7510 u64 start = iblock << inode->i_blkbits;
eb838e73 7511 u64 lockstart, lockend;
4b46fce2 7512 u64 len = bh_result->b_size;
eb838e73 7513 int unlock_bits = EXTENT_LOCKED;
0934856d 7514 int ret = 0;
eb838e73 7515
172a5049 7516 if (create)
3266789f 7517 unlock_bits |= EXTENT_DIRTY;
172a5049 7518 else
c329861d 7519 len = min_t(u64, len, root->sectorsize);
eb838e73 7520
c329861d
JB		 7521 lockstart = start;
7522 lockend = start + len - 1;
7523
e1cbbfa5
JB		 7524 if (current->journal_info) {
7525 /*
7526 * Need to pull our outstanding extents and set journal_info to NULL so
7527 * that anything that needs to check if there's a transction doesn't get
7528 * confused.
7529 */
50745b0a 7530 dio_data = current->journal_info;
e1cbbfa5
JB		 7531 current->journal_info = NULL;
7532 }
7533
eb838e73
JB		 7534 /*
7535 * If this errors out it's because we couldn't invalidate pagecache for
7536 * this range and we need to fallback to buffered.
7537 */
9c9464cc
FM		 7538 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7539 create)) {
7540 ret = -ENOTBLK;
7541 goto err;
7542 }
eb838e73 7543
4b46fce2 7544 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
eb838e73
JB		 7545 if (IS_ERR(em)) {
7546 ret = PTR_ERR(em);
7547 goto unlock_err;
7548 }
4b46fce2
JB		 7549
7550 /*
7551 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7552 * io. INLINE is special, and we could probably kludge it in here, but
7553 * it's still buffered so for safety lets just fall back to the generic
7554 * buffered path.
7555 *
7556 * For COMPRESSED we _have_ to read the entire extent in so we can
7557 * decompress it, so there will be buffering required no matter what we
7558 * do, so go ahead and fallback to buffered.
7559 *
7560 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7561 * to buffered IO. Don't blame me, this is the price we pay for using
7562 * the generic code.
7563 */
7564 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7565 em->block_start == EXTENT_MAP_INLINE) {
7566 free_extent_map(em);
eb838e73
JB		 7567 ret = -ENOTBLK;
7568 goto unlock_err;
4b46fce2
JB		 7569 }
7570
7571 /* Just a good old fashioned hole, return */
7572 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7573 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7574 free_extent_map(em);
eb838e73 7575 goto unlock_err;
4b46fce2
JB		 7576 }
7577
7578 /*
7579 * We don't allocate a new extent in the following cases
7580 *
7581 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7582 * existing extent.
7583 * 2) The extent is marked as PREALLOC. We're good to go here and can
7584 * just use the extent.
7585 *
7586 */
46bfbb5c 7587 if (!create) {
eb838e73
JB		 7588 len = min(len, em->len - (start - em->start));
7589 lockstart = start + len;
7590 goto unlock;
46bfbb5c 7591 }
4b46fce2
JB		 7592
7593 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7594 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7595 em->block_start != EXTENT_MAP_HOLE)) {
4b46fce2 7596 int type;
eb384b55 7597 u64 block_start, orig_start, orig_block_len, ram_bytes;
4b46fce2
JB		 7598
7599 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7600 type = BTRFS_ORDERED_PREALLOC;
7601 else
7602 type = BTRFS_ORDERED_NOCOW;
46bfbb5c 7603 len = min(len, em->len - (start - em->start));
4b46fce2 7604 block_start = em->block_start + (start - em->start);
46bfbb5c 7605
00361589 7606 if (can_nocow_extent(inode, start, &len, &orig_start,
7ee9e440 7607 &orig_block_len, &ram_bytes) == 1) {
69ffb543
JB		 7608 if (type == BTRFS_ORDERED_PREALLOC) {
7609 free_extent_map(em);
7610 em = create_pinned_em(inode, start, len,
7611 orig_start,
b4939680 7612 block_start, len,
cc95bef6
JB		 7613 orig_block_len,
7614 ram_bytes, type);
555e1286
FM		 7615 if (IS_ERR(em)) {
7616 ret = PTR_ERR(em);
69ffb543 7617 goto unlock_err;
555e1286 7618 }
69ffb543
JB		 7619 }
7620
46bfbb5c
CM		 7621 ret = btrfs_add_ordered_extent_dio(inode, start,
7622 block_start, len, len, type);
46bfbb5c
CM		 7623 if (ret) {
7624 free_extent_map(em);
eb838e73 7625 goto unlock_err;
46bfbb5c
CM		 7626 }
7627 goto unlock;
4b46fce2 7628 }
4b46fce2 7629 }
00361589 7630
46bfbb5c
CM		 7631 /*
7632 * this will cow the extent, reset the len in case we changed
7633 * it above
7634 */
7635 len = bh_result->b_size;
70c8a91c
JB		 7636 free_extent_map(em);
7637 em = btrfs_new_extent_direct(inode, start, len);
eb838e73
JB		 7638 if (IS_ERR(em)) {
7639 ret = PTR_ERR(em);
7640 goto unlock_err;
7641 }
46bfbb5c
CM		 7642 len = min(len, em->len - (start - em->start));
7643unlock:
4b46fce2
JB		 7644 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7645 inode->i_blkbits;
46bfbb5c 7646 bh_result->b_size = len;
4b46fce2
JB		 7647 bh_result->b_bdev = em->bdev;
7648 set_buffer_mapped(bh_result);
c3473e83
JB		 7649 if (create) {
7650 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7651 set_buffer_new(bh_result);
7652
7653 /*
7654 * Need to update the i_size under the extent lock so buffered
7655 * readers will get the updated i_size when we unlock.
7656 */
7657 if (start + len > i_size_read(inode))
7658 i_size_write(inode, start + len);
0934856d 7659
9c9464cc 7660 adjust_dio_outstanding_extents(inode, dio_data, len);
7cf5b976 7661 btrfs_free_reserved_data_space(inode, start, len);
50745b0a 7662 WARN_ON(dio_data->reserve < len);
7663 dio_data->reserve -= len;
f28a4928 7664 dio_data->unsubmitted_oe_range_end = start + len;
50745b0a 7665 current->journal_info = dio_data;
c3473e83 7666 }
4b46fce2 7667
eb838e73
JB		 7668 /*
7669 * In the case of write we need to clear and unlock the entire range,
7670 * in the case of read we need to unlock only the end area that we
7671 * aren't using if there is any left over space.
7672 */
24c03fa5 7673 if (lockstart < lockend) {
0934856d
MX		 7674 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7675 lockend, unlock_bits, 1, 0,
7676 &cached_state, GFP_NOFS);
24c03fa5 7677 } else {
eb838e73 7678 free_extent_state(cached_state);
24c03fa5 7679 }
eb838e73 7680
4b46fce2
JB		 7681 free_extent_map(em);
7682
7683 return 0;
eb838e73
JB		 7684
7685unlock_err:
eb838e73
JB		 7686 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7687 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
9c9464cc 7688err:
50745b0a 7689 if (dio_data)
7690 current->journal_info = dio_data;
9c9464cc
FM		 7691 /*
7692 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7693 * write less data then expected, so that we don't underflow our inode's
7694 * outstanding extents counter.
7695 */
7696 if (create && dio_data)
7697 adjust_dio_outstanding_extents(inode, dio_data, len);
7698
eb838e73 7699 return ret;
4b46fce2
JB		 7700}
7701
8b110e39
MX		 7702static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7703 int rw, int mirror_num)
7704{
7705 struct btrfs_root *root = BTRFS_I(inode)->root;
7706 int ret;
7707
7708 BUG_ON(rw & REQ_WRITE);
7709
7710 bio_get(bio);
7711
7712 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7713 BTRFS_WQ_ENDIO_DIO_REPAIR);
7714 if (ret)
7715 goto err;
7716
7717 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7718err:
7719 bio_put(bio);
7720 return ret;
7721}
7722
7723static int btrfs_check_dio_repairable(struct inode *inode,
7724 struct bio *failed_bio,
7725 struct io_failure_record *failrec,
7726 int failed_mirror)
7727{
7728 int num_copies;
7729
7730 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7731 failrec->logical, failrec->len);
7732 if (num_copies == 1) {
7733 /*
7734 * we only have a single copy of the data, so don't bother with
7735 * all the retry and error correction code that follows. no
7736 * matter what the error is, it is very likely to persist.
7737 */
7738 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7739 num_copies, failrec->this_mirror, failed_mirror);
7740 return 0;
7741 }
7742
7743 failrec->failed_mirror = failed_mirror;
7744 failrec->this_mirror++;
7745 if (failrec->this_mirror == failed_mirror)
7746 failrec->this_mirror++;
7747
7748 if (failrec->this_mirror > num_copies) {
7749 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7750 num_copies, failrec->this_mirror, failed_mirror);
7751 return 0;
7752 }
7753
7754 return 1;
7755}
7756
7757static int dio_read_error(struct inode *inode, struct bio *failed_bio,
2dabb324
CR		 7758 struct page *page, unsigned int pgoff,
7759 u64 start, u64 end, int failed_mirror,
7760 bio_end_io_t *repair_endio, void *repair_arg)
8b110e39
MX		 7761{
7762 struct io_failure_record *failrec;
7763 struct bio *bio;
7764 int isector;
7765 int read_mode;
7766 int ret;
7767
7768 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7769
7770 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7771 if (ret)
7772 return ret;
7773
7774 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7775 failed_mirror);
7776 if (!ret) {
7777 free_io_failure(inode, failrec);
7778 return -EIO;
7779 }
7780
2dabb324
CR		 7781 if ((failed_bio->bi_vcnt > 1)
7782 || (failed_bio->bi_io_vec->bv_len
7783 > BTRFS_I(inode)->root->sectorsize))
8b110e39
MX		 7784 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7785 else
7786 read_mode = READ_SYNC;
7787
7788 isector = start - btrfs_io_bio(failed_bio)->logical;
7789 isector >>= inode->i_sb->s_blocksize_bits;
7790 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
2dabb324 7791 pgoff, isector, repair_endio, repair_arg);
8b110e39
MX		 7792 if (!bio) {
7793 free_io_failure(inode, failrec);
7794 return -EIO;
7795 }
7796
7797 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7798 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7799 read_mode, failrec->this_mirror, failrec->in_validation);
7800
7801 ret = submit_dio_repair_bio(inode, bio, read_mode,
7802 failrec->this_mirror);
7803 if (ret) {
7804 free_io_failure(inode, failrec);
7805 bio_put(bio);
7806 }
7807
7808 return ret;
7809}
7810
7811struct btrfs_retry_complete {
7812 struct completion done;
7813 struct inode *inode;
7814 u64 start;
7815 int uptodate;
7816};
7817
4246a0b6 7818static void btrfs_retry_endio_nocsum(struct bio *bio)
8b110e39
MX		 7819{
7820 struct btrfs_retry_complete *done = bio->bi_private;
2dabb324 7821 struct inode *inode;
8b110e39
MX		 7822 struct bio_vec *bvec;
7823 int i;
7824
4246a0b6 7825 if (bio->bi_error)
8b110e39
MX		 7826 goto end;
7827
2dabb324
CR		 7828 ASSERT(bio->bi_vcnt == 1);
7829 inode = bio->bi_io_vec->bv_page->mapping->host;
7830 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
7831
8b110e39
MX		 7832 done->uptodate = 1;
7833 bio_for_each_segment_all(bvec, bio, i)
7834 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7835end:
7836 complete(&done->done);
7837 bio_put(bio);
7838}
7839
7840static int __btrfs_correct_data_nocsum(struct inode *inode,
7841 struct btrfs_io_bio *io_bio)
4b46fce2 7842{
2dabb324 7843 struct btrfs_fs_info *fs_info;
2c30c71b 7844 struct bio_vec *bvec;
8b110e39 7845 struct btrfs_retry_complete done;
4b46fce2 7846 u64 start;
2dabb324
CR		 7847 unsigned int pgoff;
7848 u32 sectorsize;
7849 int nr_sectors;
2c30c71b 7850 int i;
c1dc0896 7851 int ret;
4b46fce2 7852
2dabb324
CR		 7853 fs_info = BTRFS_I(inode)->root->fs_info;
7854 sectorsize = BTRFS_I(inode)->root->sectorsize;
7855
8b110e39
MX		 7856 start = io_bio->logical;
7857 done.inode = inode;
7858
7859 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
2dabb324
CR		 7860 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
7861 pgoff = bvec->bv_offset;
7862
7863next_block_or_try_again:
8b110e39
MX		 7864 done.uptodate = 0;
7865 done.start = start;
7866 init_completion(&done.done);
7867
2dabb324
CR		 7868 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
7869 pgoff, start, start + sectorsize - 1,
7870 io_bio->mirror_num,
7871 btrfs_retry_endio_nocsum, &done);
8b110e39
MX		 7872 if (ret)
7873 return ret;
7874
7875 wait_for_completion(&done.done);
7876
7877 if (!done.uptodate) {
7878 /* We might have another mirror, so try again */
2dabb324 7879 goto next_block_or_try_again;
8b110e39
MX		 7880 }
7881
2dabb324
CR		 7882 start += sectorsize;
7883
7884 if (nr_sectors--) {
7885 pgoff += sectorsize;
7886 goto next_block_or_try_again;
7887 }
8b110e39
MX		 7888 }
7889
7890 return 0;
7891}
7892
4246a0b6 7893static void btrfs_retry_endio(struct bio *bio)
8b110e39
MX		 7894{
7895 struct btrfs_retry_complete *done = bio->bi_private;
7896 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2dabb324 7897 struct inode *inode;
8b110e39 7898 struct bio_vec *bvec;
2dabb324 7899 u64 start;
8b110e39
MX		 7900 int uptodate;
7901 int ret;
7902 int i;
7903
4246a0b6 7904 if (bio->bi_error)
8b110e39
MX		 7905 goto end;
7906
7907 uptodate = 1;
2dabb324
CR		 7908
7909 start = done->start;
7910
7911 ASSERT(bio->bi_vcnt == 1);
7912 inode = bio->bi_io_vec->bv_page->mapping->host;
7913 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
7914
8b110e39
MX		 7915 bio_for_each_segment_all(bvec, bio, i) {
7916 ret = __readpage_endio_check(done->inode, io_bio, i,
2dabb324
CR		 7917 bvec->bv_page, bvec->bv_offset,
7918 done->start, bvec->bv_len);
8b110e39
MX		 7919 if (!ret)
7920 clean_io_failure(done->inode, done->start,
2dabb324 7921 bvec->bv_page, bvec->bv_offset);
8b110e39
MX		 7922 else
7923 uptodate = 0;
7924 }
7925
7926 done->uptodate = uptodate;
7927end:
7928 complete(&done->done);
7929 bio_put(bio);
7930}
7931
7932static int __btrfs_subio_endio_read(struct inode *inode,
7933 struct btrfs_io_bio *io_bio, int err)
7934{
2dabb324 7935 struct btrfs_fs_info *fs_info;
8b110e39
MX		 7936 struct bio_vec *bvec;
7937 struct btrfs_retry_complete done;
7938 u64 start;
7939 u64 offset = 0;
2dabb324
CR		 7940 u32 sectorsize;
7941 int nr_sectors;
7942 unsigned int pgoff;
7943 int csum_pos;
8b110e39
MX		 7944 int i;
7945 int ret;
dc380aea 7946
2dabb324
CR		 7947 fs_info = BTRFS_I(inode)->root->fs_info;
7948 sectorsize = BTRFS_I(inode)->root->sectorsize;
7949
8b110e39 7950 err = 0;
c1dc0896 7951 start = io_bio->logical;
8b110e39
MX		 7952 done.inode = inode;
7953
c1dc0896 7954 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
2dabb324
CR		 7955 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
7956
7957 pgoff = bvec->bv_offset;
7958next_block:
7959 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
7960 ret = __readpage_endio_check(inode, io_bio, csum_pos,
7961 bvec->bv_page, pgoff, start,
7962 sectorsize);
8b110e39
MX		 7963 if (likely(!ret))
7964 goto next;
7965try_again:
7966 done.uptodate = 0;
7967 done.start = start;
7968 init_completion(&done.done);
7969
2dabb324
CR		 7970 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
7971 pgoff, start, start + sectorsize - 1,
7972 io_bio->mirror_num,
7973 btrfs_retry_endio, &done);
8b110e39
MX		 7974 if (ret) {
7975 err = ret;
7976 goto next;
7977 }
7978
7979 wait_for_completion(&done.done);
7980
7981 if (!done.uptodate) {
7982 /* We might have another mirror, so try again */
7983 goto try_again;
7984 }
7985next:
2dabb324
CR		 7986 offset += sectorsize;
7987 start += sectorsize;
7988
7989 ASSERT(nr_sectors);
7990
7991 if (--nr_sectors) {
7992 pgoff += sectorsize;
7993 goto next_block;
7994 }
2c30c71b 7995 }
c1dc0896
MX		 7996
7997 return err;
7998}
7999
8b110e39
MX		 8000static int btrfs_subio_endio_read(struct inode *inode,
8001 struct btrfs_io_bio *io_bio, int err)
8002{
8003 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8004
8005 if (skip_csum) {
8006 if (unlikely(err))
8007 return __btrfs_correct_data_nocsum(inode, io_bio);
8008 else
8009 return 0;
8010 } else {
8011 return __btrfs_subio_endio_read(inode, io_bio, err);
8012 }
8013}
8014
4246a0b6 8015static void btrfs_endio_direct_read(struct bio *bio)
c1dc0896
MX		 8016{
8017 struct btrfs_dio_private *dip = bio->bi_private;
8018 struct inode *inode = dip->inode;
8019 struct bio *dio_bio;
8020 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
4246a0b6 8021 int err = bio->bi_error;
c1dc0896 8022
8b110e39
MX		 8023 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8024 err = btrfs_subio_endio_read(inode, io_bio, err);
c1dc0896 8025
4b46fce2 8026 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
d0082371 8027 dip->logical_offset + dip->bytes - 1);
9be3395b 8028 dio_bio = dip->dio_bio;
4b46fce2 8029
4b46fce2 8030 kfree(dip);
c0da7aa1 8031
4246a0b6 8032 dio_end_io(dio_bio, bio->bi_error);
23ea8e5a
MX		 8033
8034 if (io_bio->end_io)
8035 io_bio->end_io(io_bio, err);
9be3395b 8036 bio_put(bio);
4b46fce2
JB		 8037}
8038
14543774
FM		 8039static void btrfs_endio_direct_write_update_ordered(struct inode *inode,
8040 const u64 offset,
8041 const u64 bytes,
8042 const int uptodate)
4b46fce2 8043{
4b46fce2 8044 struct btrfs_root *root = BTRFS_I(inode)->root;
4b46fce2 8045 struct btrfs_ordered_extent *ordered = NULL;
14543774
FM		 8046 u64 ordered_offset = offset;
8047 u64 ordered_bytes = bytes;
4b46fce2
JB		 8048 int ret;
8049
163cf09c
CM		 8050again:
8051 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8052 &ordered_offset,
4246a0b6 8053 ordered_bytes,
14543774 8054 uptodate);
4b46fce2 8055 if (!ret)
163cf09c 8056 goto out_test;
4b46fce2 8057
9e0af237
LB		 8058 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8059 finish_ordered_fn, NULL, NULL);
fccb5d86
QW		 8060 btrfs_queue_work(root->fs_info->endio_write_workers,
8061 &ordered->work);
163cf09c
CM		 8062out_test:
8063 /*
8064 * our bio might span multiple ordered extents. If we haven't
8065 * completed the accounting for the whole dio, go back and try again
8066 */
14543774
FM		 8067 if (ordered_offset < offset + bytes) {
8068 ordered_bytes = offset + bytes - ordered_offset;
5fd02043 8069 ordered = NULL;
163cf09c
CM		 8070 goto again;
8071 }
14543774
FM		 8072}
8073
8074static void btrfs_endio_direct_write(struct bio *bio)
8075{
8076 struct btrfs_dio_private *dip = bio->bi_private;
8077 struct bio *dio_bio = dip->dio_bio;
8078
8079 btrfs_endio_direct_write_update_ordered(dip->inode,
8080 dip->logical_offset,
8081 dip->bytes,
8082 !bio->bi_error);
4b46fce2 8083
4b46fce2 8084 kfree(dip);
c0da7aa1 8085
4246a0b6 8086 dio_end_io(dio_bio, bio->bi_error);
9be3395b 8087 bio_put(bio);
4b46fce2
JB		 8088}
8089
eaf25d93
CM		 8090static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
8091 struct bio *bio, int mirror_num,
8092 unsigned long bio_flags, u64 offset)
8093{
8094 int ret;
8095 struct btrfs_root *root = BTRFS_I(inode)->root;
8096 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
79787eaa 8097 BUG_ON(ret); /* -ENOMEM */
eaf25d93
CM		 8098 return 0;
8099}
8100
4246a0b6 8101static void btrfs_end_dio_bio(struct bio *bio)
e65e1535
MX		 8102{
8103 struct btrfs_dio_private *dip = bio->bi_private;
4246a0b6 8104 int err = bio->bi_error;
e65e1535 8105
8b110e39
MX		 8106 if (err)
8107 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8108 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
8109 btrfs_ino(dip->inode), bio->bi_rw,
8110 (unsigned long long)bio->bi_iter.bi_sector,
8111 bio->bi_iter.bi_size, err);
8112
8113 if (dip->subio_endio)
8114 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
c1dc0896
MX		 8115
8116 if (err) {
e65e1535
MX		 8117 dip->errors = 1;
8118
8119 /*
8120 * before atomic variable goto zero, we must make sure
8121 * dip->errors is perceived to be set.
8122 */
4e857c58 8123 smp_mb__before_atomic();
e65e1535
MX		 8124 }
8125
8126 /* if there are more bios still pending for this dio, just exit */
8127 if (!atomic_dec_and_test(&dip->pending_bios))
8128 goto out;
8129
9be3395b 8130 if (dip->errors) {
e65e1535 8131 bio_io_error(dip->orig_bio);
9be3395b 8132 } else {
4246a0b6
CH		 8133 dip->dio_bio->bi_error = 0;
8134 bio_endio(dip->orig_bio);
e65e1535
MX		 8135 }
8136out:
8137 bio_put(bio);
8138}
8139
8140static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8141 u64 first_sector, gfp_t gfp_flags)
8142{
da2f0f74 8143 struct bio *bio;
22365979 8144 bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
da2f0f74
CM		 8145 if (bio)
8146 bio_associate_current(bio);
8147 return bio;
e65e1535
MX		 8148}
8149
c1dc0896
MX		 8150static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
8151 struct inode *inode,
8152 struct btrfs_dio_private *dip,
8153 struct bio *bio,
8154 u64 file_offset)
8155{
8156 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8157 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8158 int ret;
8159
8160 /*
8161 * We load all the csum data we need when we submit
8162 * the first bio to reduce the csum tree search and
8163 * contention.
8164 */
8165 if (dip->logical_offset == file_offset) {
8166 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
8167 file_offset);
8168 if (ret)
8169 return ret;
8170 }
8171
8172 if (bio == dip->orig_bio)
8173 return 0;
8174
8175 file_offset -= dip->logical_offset;
8176 file_offset >>= inode->i_sb->s_blocksize_bits;
8177 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8178
8179 return 0;
8180}
8181
e65e1535
MX		 8182static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8183 int rw, u64 file_offset, int skip_sum,
c329861d 8184 int async_submit)
e65e1535 8185{
facc8a22 8186 struct btrfs_dio_private *dip = bio->bi_private;
e65e1535
MX		 8187 int write = rw & REQ_WRITE;
8188 struct btrfs_root *root = BTRFS_I(inode)->root;
8189 int ret;
8190
b812ce28
JB		 8191 if (async_submit)
8192 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8193
e65e1535 8194 bio_get(bio);
5fd02043
JB		 8195
8196 if (!write) {
bfebd8b5
DS		 8197 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8198 BTRFS_WQ_ENDIO_DATA);
5fd02043
JB		 8199 if (ret)
8200 goto err;
8201 }
e65e1535 8202
1ae39938
JB		 8203 if (skip_sum)
8204 goto map;
8205
8206 if (write && async_submit) {
e65e1535
MX		 8207 ret = btrfs_wq_submit_bio(root->fs_info,
8208 inode, rw, bio, 0, 0,
8209 file_offset,
8210 __btrfs_submit_bio_start_direct_io,
8211 __btrfs_submit_bio_done);
8212 goto err;
1ae39938
JB		 8213 } else if (write) {
8214 /*
8215 * If we aren't doing async submit, calculate the csum of the
8216 * bio now.
8217 */
8218 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8219 if (ret)
8220 goto err;
23ea8e5a 8221 } else {
c1dc0896
MX		 8222 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8223 file_offset);
c2db1073
TI		 8224 if (ret)
8225 goto err;
8226 }
1ae39938
JB		 8227map:
8228 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
e65e1535
MX		 8229err:
8230 bio_put(bio);
8231 return ret;
8232}
8233
8234static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
8235 int skip_sum)
8236{
8237 struct inode *inode = dip->inode;
8238 struct btrfs_root *root = BTRFS_I(inode)->root;
e65e1535
MX		 8239 struct bio *bio;
8240 struct bio *orig_bio = dip->orig_bio;
8241 struct bio_vec *bvec = orig_bio->bi_io_vec;
4f024f37 8242 u64 start_sector = orig_bio->bi_iter.bi_sector;
e65e1535
MX		 8243 u64 file_offset = dip->logical_offset;
8244 u64 submit_len = 0;
8245 u64 map_length;
8246 int nr_pages = 0;
23ea8e5a 8247 int ret;
1ae39938 8248 int async_submit = 0;
e65e1535 8249
4f024f37 8250 map_length = orig_bio->bi_iter.bi_size;
53b381b3 8251 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
e65e1535 8252 &map_length, NULL, 0);
7a5c3c9b 8253 if (ret)
e65e1535 8254 return -EIO;
facc8a22 8255
4f024f37 8256 if (map_length >= orig_bio->bi_iter.bi_size) {
02f57c7a 8257 bio = orig_bio;
c1dc0896 8258 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
02f57c7a
JB		 8259 goto submit;
8260 }
8261
53b381b3 8262 /* async crcs make it difficult to collect full stripe writes. */
ffe2d203 8263 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
53b381b3
DW		 8264 async_submit = 0;
8265 else
8266 async_submit = 1;
8267
02f57c7a
JB		 8268 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8269 if (!bio)
8270 return -ENOMEM;
7a5c3c9b 8271
02f57c7a
JB		 8272 bio->bi_private = dip;
8273 bio->bi_end_io = btrfs_end_dio_bio;
c1dc0896 8274 btrfs_io_bio(bio)->logical = file_offset;
02f57c7a
JB		 8275 atomic_inc(&dip->pending_bios);
8276
e65e1535 8277 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
ee39b432 8278 if (map_length < submit_len + bvec->bv_len ||
e65e1535 8279 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
ee39b432 8280 bvec->bv_offset) < bvec->bv_len) {
e65e1535
MX		 8281 /*
8282 * inc the count before we submit the bio so
8283 * we know the end IO handler won't happen before
8284 * we inc the count. Otherwise, the dip might get freed
8285 * before we're done setting it up
8286 */
8287 atomic_inc(&dip->pending_bios);
8288 ret = __btrfs_submit_dio_bio(bio, inode, rw,
8289 file_offset, skip_sum,
c329861d 8290 async_submit);
e65e1535
MX		 8291 if (ret) {
8292 bio_put(bio);
8293 atomic_dec(&dip->pending_bios);
8294 goto out_err;
8295 }
8296
e65e1535
MX		 8297 start_sector += submit_len >> 9;
8298 file_offset += submit_len;
8299
8300 submit_len = 0;
8301 nr_pages = 0;
8302
8303 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8304 start_sector, GFP_NOFS);
8305 if (!bio)
8306 goto out_err;
8307 bio->bi_private = dip;
8308 bio->bi_end_io = btrfs_end_dio_bio;
c1dc0896 8309 btrfs_io_bio(bio)->logical = file_offset;
e65e1535 8310
4f024f37 8311 map_length = orig_bio->bi_iter.bi_size;
53b381b3 8312 ret = btrfs_map_block(root->fs_info, rw,
3ec706c8 8313 start_sector << 9,
e65e1535
MX		 8314 &map_length, NULL, 0);
8315 if (ret) {
8316 bio_put(bio);
8317 goto out_err;
8318 }
8319 } else {
8320 submit_len += bvec->bv_len;
67871254 8321 nr_pages++;
e65e1535
MX		 8322 bvec++;
8323 }
8324 }
8325
02f57c7a 8326submit:
e65e1535 8327 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
c329861d 8328 async_submit);
e65e1535
MX		 8329 if (!ret)
8330 return 0;
8331
8332 bio_put(bio);
8333out_err:
8334 dip->errors = 1;
8335 /*
8336 * before atomic variable goto zero, we must
8337 * make sure dip->errors is perceived to be set.
8338 */
4e857c58 8339 smp_mb__before_atomic();
e65e1535
MX		 8340 if (atomic_dec_and_test(&dip->pending_bios))
8341 bio_io_error(dip->orig_bio);
8342
8343 /* bio_end_io() will handle error, so we needn't return it */
8344 return 0;
8345}
8346
9be3395b
CM		 8347static void btrfs_submit_direct(int rw, struct bio *dio_bio,
8348 struct inode *inode, loff_t file_offset)
4b46fce2 8349{
61de718f
FM		 8350 struct btrfs_dio_private *dip = NULL;
8351 struct bio *io_bio = NULL;
23ea8e5a 8352 struct btrfs_io_bio *btrfs_bio;
4b46fce2 8353 int skip_sum;
7b6d91da 8354 int write = rw & REQ_WRITE;
4b46fce2
JB		 8355 int ret = 0;
8356
8357 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8358
9be3395b 8359 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
9be3395b
CM		 8360 if (!io_bio) {
8361 ret = -ENOMEM;
8362 goto free_ordered;
8363 }
8364
c1dc0896 8365 dip = kzalloc(sizeof(*dip), GFP_NOFS);
4b46fce2
JB		 8366 if (!dip) {
8367 ret = -ENOMEM;
61de718f 8368 goto free_ordered;
4b46fce2 8369 }
4b46fce2 8370
9be3395b 8371 dip->private = dio_bio->bi_private;
4b46fce2
JB		 8372 dip->inode = inode;
8373 dip->logical_offset = file_offset;
4f024f37
KO		 8374 dip->bytes = dio_bio->bi_iter.bi_size;
8375 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
9be3395b 8376 io_bio->bi_private = dip;
9be3395b
CM		 8377 dip->orig_bio = io_bio;
8378 dip->dio_bio = dio_bio;
e65e1535 8379 atomic_set(&dip->pending_bios, 0);
c1dc0896
MX		 8380 btrfs_bio = btrfs_io_bio(io_bio);
8381 btrfs_bio->logical = file_offset;
4b46fce2 8382
c1dc0896 8383 if (write) {
9be3395b 8384 io_bio->bi_end_io = btrfs_endio_direct_write;
c1dc0896 8385 } else {
9be3395b 8386 io_bio->bi_end_io = btrfs_endio_direct_read;
c1dc0896
MX		 8387 dip->subio_endio = btrfs_subio_endio_read;
8388 }
4b46fce2 8389
f28a4928
FM		 8390 /*
8391 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8392 * even if we fail to submit a bio, because in such case we do the
8393 * corresponding error handling below and it must not be done a second
8394 * time by btrfs_direct_IO().
8395 */
8396 if (write) {
8397 struct btrfs_dio_data *dio_data = current->journal_info;
8398
8399 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8400 dip->bytes;
8401 dio_data->unsubmitted_oe_range_start =
8402 dio_data->unsubmitted_oe_range_end;
8403 }
8404
e65e1535
MX		 8405 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8406 if (!ret)
eaf25d93 8407 return;
9be3395b 8408
23ea8e5a
MX		 8409 if (btrfs_bio->end_io)
8410 btrfs_bio->end_io(btrfs_bio, ret);
9be3395b 8411
4b46fce2
JB		 8412free_ordered:
8413 /*
61de718f
FM		 8414 * If we arrived here it means either we failed to submit the dip
8415 * or we either failed to clone the dio_bio or failed to allocate the
8416 * dip. If we cloned the dio_bio and allocated the dip, we can just
8417 * call bio_endio against our io_bio so that we get proper resource
8418 * cleanup if we fail to submit the dip, otherwise, we must do the
8419 * same as btrfs_endio_direct_[write|read] because we can't call these
8420 * callbacks - they require an allocated dip and a clone of dio_bio.
4b46fce2 8421 */
61de718f 8422 if (io_bio && dip) {
4246a0b6
CH		 8423 io_bio->bi_error = -EIO;
8424 bio_endio(io_bio);
61de718f
FM		 8425 /*
8426 * The end io callbacks free our dip, do the final put on io_bio
8427 * and all the cleanup and final put for dio_bio (through
8428 * dio_end_io()).
8429 */
8430 dip = NULL;
8431 io_bio = NULL;
8432 } else {
14543774
FM		 8433 if (write)
8434 btrfs_endio_direct_write_update_ordered(inode,
8435 file_offset,
8436 dio_bio->bi_iter.bi_size,
8437 0);
8438 else
61de718f
FM		 8439 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8440 file_offset + dio_bio->bi_iter.bi_size - 1);
14543774 8441
4246a0b6 8442 dio_bio->bi_error = -EIO;
61de718f
FM		 8443 /*
8444 * Releases and cleans up our dio_bio, no need to bio_put()
8445 * nor bio_endio()/bio_io_error() against dio_bio.
8446 */
8447 dio_end_io(dio_bio, ret);
4b46fce2 8448 }
61de718f
FM		 8449 if (io_bio)
8450 bio_put(io_bio);
8451 kfree(dip);
4b46fce2
JB		 8452}
8453
6f673763 8454static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
28060d5d 8455 const struct iov_iter *iter, loff_t offset)
5a5f79b5
CM		 8456{
8457 int seg;
a1b75f7d 8458 int i;
5a5f79b5
CM		 8459 unsigned blocksize_mask = root->sectorsize - 1;
8460 ssize_t retval = -EINVAL;
5a5f79b5
CM		 8461
8462 if (offset & blocksize_mask)
8463 goto out;
8464
28060d5d
AV		 8465 if (iov_iter_alignment(iter) & blocksize_mask)
8466 goto out;
a1b75f7d 8467
28060d5d 8468 /* If this is a write we don't need to check anymore */
6f673763 8469 if (iov_iter_rw(iter) == WRITE)
28060d5d
AV		 8470 return 0;
8471 /*
8472 * Check to make sure we don't have duplicate iov_base's in this
8473 * iovec, if so return EINVAL, otherwise we'll get csum errors
8474 * when reading back.
8475 */
8476 for (seg = 0; seg < iter->nr_segs; seg++) {
8477 for (i = seg + 1; i < iter->nr_segs; i++) {
8478 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
a1b75f7d
JB		 8479 goto out;
8480 }
5a5f79b5
CM		 8481 }
8482 retval = 0;
8483out:
8484 return retval;
8485}
eb838e73 8486
22c6186e
OS		 8487static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
8488 loff_t offset)
16432985 8489{
4b46fce2
JB		 8490 struct file *file = iocb->ki_filp;
8491 struct inode *inode = file->f_mapping->host;
50745b0a 8492 struct btrfs_root *root = BTRFS_I(inode)->root;
8493 struct btrfs_dio_data dio_data = { 0 };
0934856d 8494 size_t count = 0;
2e60a51e 8495 int flags = 0;
38851cc1
MX		 8496 bool wakeup = true;
8497 bool relock = false;
0934856d 8498 ssize_t ret;
4b46fce2 8499
6f673763 8500 if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
5a5f79b5 8501 return 0;
3f7c579c 8502
fe0f07d0 8503 inode_dio_begin(inode);
4e857c58 8504 smp_mb__after_atomic();
38851cc1 8505
0e267c44 8506 /*
41bd9ca4
MX		 8507 * The generic stuff only does filemap_write_and_wait_range, which
8508 * isn't enough if we've written compressed pages to this area, so
8509 * we need to flush the dirty pages again to make absolutely sure
8510 * that any outstanding dirty pages are on disk.
0e267c44 8511 */
a6cbcd4a 8512 count = iov_iter_count(iter);
41bd9ca4
MX		 8513 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8514 &BTRFS_I(inode)->runtime_flags))
9a025a08
WS		 8515 filemap_fdatawrite_range(inode->i_mapping, offset,
8516 offset + count - 1);
0e267c44 8517
6f673763 8518 if (iov_iter_rw(iter) == WRITE) {
38851cc1
MX		 8519 /*
8520 * If the write DIO is beyond the EOF, we need update
8521 * the isize, but it is protected by i_mutex. So we can
8522 * not unlock the i_mutex at this case.
8523 */
8524 if (offset + count <= inode->i_size) {
8525 mutex_unlock(&inode->i_mutex);
8526 relock = true;
8527 }
7cf5b976 8528 ret = btrfs_delalloc_reserve_space(inode, offset, count);
0934856d 8529 if (ret)
38851cc1 8530 goto out;
50745b0a 8531 dio_data.outstanding_extents = div64_u64(count +
e1cbbfa5
JB		 8532 BTRFS_MAX_EXTENT_SIZE - 1,
8533 BTRFS_MAX_EXTENT_SIZE);
8534
8535 /*
8536 * We need to know how many extents we reserved so that we can
8537 * do the accounting properly if we go over the number we
8538 * originally calculated. Abuse current->journal_info for this.
8539 */
50745b0a 8540 dio_data.reserve = round_up(count, root->sectorsize);
f28a4928
FM		 8541 dio_data.unsubmitted_oe_range_start = (u64)offset;
8542 dio_data.unsubmitted_oe_range_end = (u64)offset;
50745b0a 8543 current->journal_info = &dio_data;
ee39b432
DS		 8544 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8545 &BTRFS_I(inode)->runtime_flags)) {
fe0f07d0 8546 inode_dio_end(inode);
38851cc1
MX		 8547 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8548 wakeup = false;
0934856d
MX		 8549 }
8550
17f8c842
OS		 8551 ret = __blockdev_direct_IO(iocb, inode,
8552 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8553 iter, offset, btrfs_get_blocks_direct, NULL,
8554 btrfs_submit_direct, flags);
6f673763 8555 if (iov_iter_rw(iter) == WRITE) {
e1cbbfa5 8556 current->journal_info = NULL;
ddba1bfc 8557 if (ret < 0 && ret != -EIOCBQUEUED) {
50745b0a 8558 if (dio_data.reserve)
7cf5b976
QW		 8559 btrfs_delalloc_release_space(inode, offset,
8560 dio_data.reserve);
f28a4928
FM		 8561 /*
8562 * On error we might have left some ordered extents
8563 * without submitting corresponding bios for them, so
8564 * cleanup them up to avoid other tasks getting them
8565 * and waiting for them to complete forever.
8566 */
8567 if (dio_data.unsubmitted_oe_range_start <
8568 dio_data.unsubmitted_oe_range_end)
8569 btrfs_endio_direct_write_update_ordered(inode,
8570 dio_data.unsubmitted_oe_range_start,
8571 dio_data.unsubmitted_oe_range_end -
8572 dio_data.unsubmitted_oe_range_start,
8573 0);
ddba1bfc 8574 } else if (ret >= 0 && (size_t)ret < count)
7cf5b976
QW		 8575 btrfs_delalloc_release_space(inode, offset,
8576 count - (size_t)ret);
0934856d 8577 }
38851cc1 8578out:
2e60a51e 8579 if (wakeup)
fe0f07d0 8580 inode_dio_end(inode);
38851cc1
MX		 8581 if (relock)
8582 mutex_lock(&inode->i_mutex);
0934856d
MX		 8583
8584 return ret;
16432985
CM		 8585}
8586
05dadc09
TI		 8587#define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8588
1506fcc8
YS		 8589static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8590 __u64 start, __u64 len)
8591{
05dadc09
TI		 8592 int ret;
8593
8594 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8595 if (ret)
8596 return ret;
8597
ec29ed5b 8598 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
1506fcc8
YS		 8599}
8600
a52d9a80 8601int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 8602{
d1310b2e
CM		 8603 struct extent_io_tree *tree;
8604 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 8605 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
9ebefb18 8606}
1832a6d5 8607
a52d9a80 8608static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 8609{
d1310b2e 8610 struct extent_io_tree *tree;
be7bd730
JB		 8611 struct inode *inode = page->mapping->host;
8612 int ret;
b888db2b
CM		 8613
8614 if (current->flags & PF_MEMALLOC) {
8615 redirty_page_for_writepage(wbc, page);
8616 unlock_page(page);
8617 return 0;
8618 }
be7bd730
JB		 8619
8620 /*
8621 * If we are under memory pressure we will call this directly from the
8622 * VM, we need to make sure we have the inode referenced for the ordered
8623 * extent. If not just return like we didn't do anything.
8624 */
8625 if (!igrab(inode)) {
8626 redirty_page_for_writepage(wbc, page);
8627 return AOP_WRITEPAGE_ACTIVATE;
8628 }
d1310b2e 8629 tree = &BTRFS_I(page->mapping->host)->io_tree;
be7bd730
JB		 8630 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8631 btrfs_add_delayed_iput(inode);
8632 return ret;
9ebefb18
CM		 8633}
8634
48a3b636
ES		 8635static int btrfs_writepages(struct address_space *mapping,
8636 struct writeback_control *wbc)
b293f02e 8637{
d1310b2e 8638 struct extent_io_tree *tree;
771ed689 8639
d1310b2e 8640 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 8641 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8642}
8643
3ab2fb5a
CM		 8644static int
8645btrfs_readpages(struct file *file, struct address_space *mapping,
8646 struct list_head *pages, unsigned nr_pages)
8647{
d1310b2e
CM		 8648 struct extent_io_tree *tree;
8649 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 8650 return extent_readpages(tree, mapping, pages, nr_pages,
8651 btrfs_get_extent);
8652}
e6dcd2dc 8653static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 8654{
d1310b2e
CM		 8655 struct extent_io_tree *tree;
8656 struct extent_map_tree *map;
a52d9a80 8657 int ret;
8c2383c3 8658
d1310b2e
CM		 8659 tree = &BTRFS_I(page->mapping->host)->io_tree;
8660 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 8661 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 8662 if (ret == 1) {
8663 ClearPagePrivate(page);
8664 set_page_private(page, 0);
8665 page_cache_release(page);
39279cc3 8666 }
a52d9a80 8667 return ret;
39279cc3
CM		 8668}
8669
e6dcd2dc
CM		 8670static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8671{
98509cfc
CM		 8672 if (PageWriteback(page) || PageDirty(page))
8673 return 0;
b335b003 8674 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
e6dcd2dc
CM		 8675}
8676
d47992f8
LC		 8677static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8678 unsigned int length)
39279cc3 8679{
5fd02043 8680 struct inode *inode = page->mapping->host;
d1310b2e 8681 struct extent_io_tree *tree;
e6dcd2dc 8682 struct btrfs_ordered_extent *ordered;
2ac55d41 8683 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 8684 u64 page_start = page_offset(page);
8685 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
131e404a 8686 int inode_evicting = inode->i_state & I_FREEING;
39279cc3 8687
8b62b72b
CM		 8688 /*
8689 * we have the page locked, so new writeback can't start,
8690 * and the dirty bit won't be cleared while we are here.
8691 *
8692 * Wait for IO on this page so that we can safely clear
8693 * the PagePrivate2 bit and do ordered accounting
8694 */
e6dcd2dc 8695 wait_on_page_writeback(page);
8b62b72b 8696
5fd02043 8697 tree = &BTRFS_I(inode)->io_tree;
e6dcd2dc
CM		 8698 if (offset) {
8699 btrfs_releasepage(page, GFP_NOFS);
8700 return;
8701 }
131e404a
FDBM		 8702
8703 if (!inode_evicting)
ff13db41 8704 lock_extent_bits(tree, page_start, page_end, &cached_state);
131e404a 8705 ordered = btrfs_lookup_ordered_extent(inode, page_start);
e6dcd2dc 8706 if (ordered) {
eb84ae03
CM		 8707 /*
8708 * IO on this page will never be started, so we need
8709 * to account for any ordered extents now
8710 */
131e404a
FDBM		 8711 if (!inode_evicting)
8712 clear_extent_bit(tree, page_start, page_end,
8713 EXTENT_DIRTY | EXTENT_DELALLOC |
8714 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8715 EXTENT_DEFRAG, 1, 0, &cached_state,
8716 GFP_NOFS);
8b62b72b
CM		 8717 /*
8718 * whoever cleared the private bit is responsible
8719 * for the finish_ordered_io
8720 */
77cef2ec
JB		 8721 if (TestClearPagePrivate2(page)) {
8722 struct btrfs_ordered_inode_tree *tree;
8723 u64 new_len;
8724
8725 tree = &BTRFS_I(inode)->ordered_tree;
8726
8727 spin_lock_irq(&tree->lock);
8728 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8729 new_len = page_start - ordered->file_offset;
8730 if (new_len < ordered->truncated_len)
8731 ordered->truncated_len = new_len;
8732 spin_unlock_irq(&tree->lock);
8733
8734 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8735 page_start,
8736 PAGE_CACHE_SIZE, 1))
8737 btrfs_finish_ordered_io(ordered);
8b62b72b 8738 }
e6dcd2dc 8739 btrfs_put_ordered_extent(ordered);
131e404a
FDBM		 8740 if (!inode_evicting) {
8741 cached_state = NULL;
ff13db41 8742 lock_extent_bits(tree, page_start, page_end,
131e404a
FDBM		 8743 &cached_state);
8744 }
8745 }
8746
b9d0b389
QW		 8747 /*
8748 * Qgroup reserved space handler
8749 * Page here will be either
8750 * 1) Already written to disk
8751 * In this case, its reserved space is released from data rsv map
8752 * and will be freed by delayed_ref handler finally.
8753 * So even we call qgroup_free_data(), it won't decrease reserved
8754 * space.
8755 * 2) Not written to disk
8756 * This means the reserved space should be freed here.
8757 */
8758 btrfs_qgroup_free_data(inode, page_start, PAGE_CACHE_SIZE);
131e404a
FDBM		 8759 if (!inode_evicting) {
8760 clear_extent_bit(tree, page_start, page_end,
8761 EXTENT_LOCKED | EXTENT_DIRTY |
8762 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8763 EXTENT_DEFRAG, 1, 1,
8764 &cached_state, GFP_NOFS);
8765
8766 __btrfs_releasepage(page, GFP_NOFS);
e6dcd2dc 8767 }
e6dcd2dc 8768
4a096752 8769 ClearPageChecked(page);
9ad6b7bc 8770 if (PagePrivate(page)) {
9ad6b7bc
CM		 8771 ClearPagePrivate(page);
8772 set_page_private(page, 0);
8773 page_cache_release(page);
8774 }
39279cc3
CM		 8775}
8776
9ebefb18
CM		 8777/*
8778 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8779 * called from a page fault handler when a page is first dirtied. Hence we must
8780 * be careful to check for EOF conditions here. We set the page up correctly
8781 * for a written page which means we get ENOSPC checking when writing into
8782 * holes and correct delalloc and unwritten extent mapping on filesystems that
8783 * support these features.
8784 *
8785 * We are not allowed to take the i_mutex here so we have to play games to
8786 * protect against truncate races as the page could now be beyond EOF. Because
8787 * vmtruncate() writes the inode size before removing pages, once we have the
8788 * page lock we can determine safely if the page is beyond EOF. If it is not
8789 * beyond EOF, then the page is guaranteed safe against truncation until we
8790 * unlock the page.
8791 */
c2ec175c 8792int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
9ebefb18 8793{
c2ec175c 8794 struct page *page = vmf->page;
496ad9aa 8795 struct inode *inode = file_inode(vma->vm_file);
1832a6d5 8796 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 8797 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8798 struct btrfs_ordered_extent *ordered;
2ac55d41 8799 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 8800 char *kaddr;
8801 unsigned long zero_start;
9ebefb18 8802 loff_t size;
1832a6d5 8803 int ret;
9998eb70 8804 int reserved = 0;
a52d9a80 8805 u64 page_start;
e6dcd2dc 8806 u64 page_end;
9ebefb18 8807
b2b5ef5c 8808 sb_start_pagefault(inode->i_sb);
df480633
QW		 8809 page_start = page_offset(page);
8810 page_end = page_start + PAGE_CACHE_SIZE - 1;
8811
7cf5b976
QW		 8812 ret = btrfs_delalloc_reserve_space(inode, page_start,
8813 PAGE_CACHE_SIZE);
9998eb70 8814 if (!ret) {
e41f941a 8815 ret = file_update_time(vma->vm_file);
9998eb70
CM		 8816 reserved = 1;
8817 }
56a76f82
NP		 8818 if (ret) {
8819 if (ret == -ENOMEM)
8820 ret = VM_FAULT_OOM;
8821 else /* -ENOSPC, -EIO, etc */
8822 ret = VM_FAULT_SIGBUS;
9998eb70
CM		 8823 if (reserved)
8824 goto out;
8825 goto out_noreserve;
56a76f82 8826 }
1832a6d5 8827
56a76f82 8828 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
e6dcd2dc 8829again:
9ebefb18 8830 lock_page(page);
9ebefb18 8831 size = i_size_read(inode);
a52d9a80 8832
9ebefb18 8833 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 8834 (page_start >= size)) {
9ebefb18
CM		 8835 /* page got truncated out from underneath us */
8836 goto out_unlock;
8837 }
e6dcd2dc
CM		 8838 wait_on_page_writeback(page);
8839
ff13db41 8840 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
e6dcd2dc
CM		 8841 set_page_extent_mapped(page);
8842
eb84ae03
CM		 8843 /*
8844 * we can't set the delalloc bits if there are pending ordered
8845 * extents. Drop our locks and wait for them to finish
8846 */
e6dcd2dc
CM		 8847 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8848 if (ordered) {
2ac55d41
JB		 8849 unlock_extent_cached(io_tree, page_start, page_end,
8850 &cached_state, GFP_NOFS);
e6dcd2dc 8851 unlock_page(page);
eb84ae03 8852 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 8853 btrfs_put_ordered_extent(ordered);
8854 goto again;
8855 }
8856
fbf19087
JB		 8857 /*
8858 * XXX - page_mkwrite gets called every time the page is dirtied, even
8859 * if it was already dirty, so for space accounting reasons we need to
8860 * clear any delalloc bits for the range we are fixing to save. There
8861 * is probably a better way to do this, but for now keep consistent with
8862 * prepare_pages in the normal write path.
8863 */
2ac55d41 8864 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 8865 EXTENT_DIRTY | EXTENT_DELALLOC |
8866 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 8867 0, 0, &cached_state, GFP_NOFS);
fbf19087 8868
2ac55d41
JB		 8869 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
8870 &cached_state);
9ed74f2d 8871 if (ret) {
2ac55d41
JB		 8872 unlock_extent_cached(io_tree, page_start, page_end,
8873 &cached_state, GFP_NOFS);
9ed74f2d
JB		 8874 ret = VM_FAULT_SIGBUS;
8875 goto out_unlock;
8876 }
e6dcd2dc 8877 ret = 0;
9ebefb18
CM		 8878
8879 /* page is wholly or partially inside EOF */
a52d9a80 8880 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 8881 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 8882 else
e6dcd2dc 8883 zero_start = PAGE_CACHE_SIZE;
9ebefb18 8884
e6dcd2dc
CM		 8885 if (zero_start != PAGE_CACHE_SIZE) {
8886 kaddr = kmap(page);
8887 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
8888 flush_dcache_page(page);
8889 kunmap(page);
8890 }
247e743c 8891 ClearPageChecked(page);
e6dcd2dc 8892 set_page_dirty(page);
50a9b214 8893 SetPageUptodate(page);
5a3f23d5 8894
257c62e1
CM		 8895 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8896 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
46d8bc34 8897 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
257c62e1 8898
2ac55d41 8899 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9ebefb18
CM		 8900
8901out_unlock:
b2b5ef5c
JK		 8902 if (!ret) {
8903 sb_end_pagefault(inode->i_sb);
50a9b214 8904 return VM_FAULT_LOCKED;
b2b5ef5c 8905 }
9ebefb18 8906 unlock_page(page);
1832a6d5 8907out:
7cf5b976 8908 btrfs_delalloc_release_space(inode, page_start, PAGE_CACHE_SIZE);
9998eb70 8909out_noreserve:
b2b5ef5c 8910 sb_end_pagefault(inode->i_sb);
9ebefb18
CM		 8911 return ret;
8912}
8913
a41ad394 8914static int btrfs_truncate(struct inode *inode)
39279cc3
CM		 8915{
8916 struct btrfs_root *root = BTRFS_I(inode)->root;
fcb80c2a 8917 struct btrfs_block_rsv *rsv;
a71754fc 8918 int ret = 0;
3893e33b 8919 int err = 0;
39279cc3 8920 struct btrfs_trans_handle *trans;
dbe674a9 8921 u64 mask = root->sectorsize - 1;
07127184 8922 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3 8923
0ef8b726
JB		 8924 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8925 (u64)-1);
8926 if (ret)
8927 return ret;
39279cc3 8928
fcb80c2a
JB		 8929 /*
8930 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8931 * 3 things going on here
8932 *
8933 * 1) We need to reserve space for our orphan item and the space to
8934 * delete our orphan item. Lord knows we don't want to have a dangling
8935 * orphan item because we didn't reserve space to remove it.
8936 *
8937 * 2) We need to reserve space to update our inode.
8938 *
8939 * 3) We need to have something to cache all the space that is going to
8940 * be free'd up by the truncate operation, but also have some slack
8941 * space reserved in case it uses space during the truncate (thank you
8942 * very much snapshotting).
8943 *
8944 * And we need these to all be seperate. The fact is we can use alot of
8945 * space doing the truncate, and we have no earthly idea how much space
8946 * we will use, so we need the truncate reservation to be seperate so it
8947 * doesn't end up using space reserved for updating the inode or
8948 * removing the orphan item. We also need to be able to stop the
8949 * transaction and start a new one, which means we need to be able to
8950 * update the inode several times, and we have no idea of knowing how
8951 * many times that will be, so we can't just reserve 1 item for the
8952 * entirety of the opration, so that has to be done seperately as well.
8953 * Then there is the orphan item, which does indeed need to be held on
8954 * to for the whole operation, and we need nobody to touch this reserved
8955 * space except the orphan code.
8956 *
8957 * So that leaves us with
8958 *
8959 * 1) root->orphan_block_rsv - for the orphan deletion.
8960 * 2) rsv - for the truncate reservation, which we will steal from the
8961 * transaction reservation.
8962 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8963 * updating the inode.
8964 */
66d8f3dd 8965 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
fcb80c2a
JB		 8966 if (!rsv)
8967 return -ENOMEM;
4a338542 8968 rsv->size = min_size;
ca7e70f5 8969 rsv->failfast = 1;
f0cd846e 8970
907cbceb 8971 /*
07127184 8972 * 1 for the truncate slack space
907cbceb
JB		 8973 * 1 for updating the inode.
8974 */
f3fe820c 8975 trans = btrfs_start_transaction(root, 2);
fcb80c2a
JB		 8976 if (IS_ERR(trans)) {
8977 err = PTR_ERR(trans);
8978 goto out;
8979 }
f0cd846e 8980
907cbceb
JB		 8981 /* Migrate the slack space for the truncate to our reserve */
8982 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8983 min_size);
fcb80c2a 8984 BUG_ON(ret);
f0cd846e 8985
5dc562c5
JB		 8986 /*
8987 * So if we truncate and then write and fsync we normally would just
8988 * write the extents that changed, which is a problem if we need to
8989 * first truncate that entire inode. So set this flag so we write out
8990 * all of the extents in the inode to the sync log so we're completely
8991 * safe.
8992 */
8993 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
ca7e70f5 8994 trans->block_rsv = rsv;
907cbceb 8995
8082510e
YZ		 8996 while (1) {
8997 ret = btrfs_truncate_inode_items(trans, root, inode,
8998 inode->i_size,
8999 BTRFS_EXTENT_DATA_KEY);
28ed1345 9000 if (ret != -ENOSPC && ret != -EAGAIN) {
3893e33b 9001 err = ret;
8082510e 9002 break;
3893e33b 9003 }
39279cc3 9004
fcb80c2a 9005 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e 9006 ret = btrfs_update_inode(trans, root, inode);
3893e33b
JB		 9007 if (ret) {
9008 err = ret;
9009 break;
9010 }
ca7e70f5 9011
8082510e 9012 btrfs_end_transaction(trans, root);
b53d3f5d 9013 btrfs_btree_balance_dirty(root);
ca7e70f5
JB		 9014
9015 trans = btrfs_start_transaction(root, 2);
9016 if (IS_ERR(trans)) {
9017 ret = err = PTR_ERR(trans);
9018 trans = NULL;
9019 break;
9020 }
9021
9022 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
9023 rsv, min_size);
9024 BUG_ON(ret); /* shouldn't happen */
9025 trans->block_rsv = rsv;
8082510e
YZ		 9026 }
9027
9028 if (ret == 0 && inode->i_nlink > 0) {
fcb80c2a 9029 trans->block_rsv = root->orphan_block_rsv;
8082510e 9030 ret = btrfs_orphan_del(trans, inode);
3893e33b
JB		 9031 if (ret)
9032 err = ret;
8082510e
YZ		 9033 }
9034
917c16b2
CM		 9035 if (trans) {
9036 trans->block_rsv = &root->fs_info->trans_block_rsv;
9037 ret = btrfs_update_inode(trans, root, inode);
9038 if (ret && !err)
9039 err = ret;
7b128766 9040
7ad85bb7 9041 ret = btrfs_end_transaction(trans, root);
b53d3f5d 9042 btrfs_btree_balance_dirty(root);
917c16b2 9043 }
fcb80c2a
JB		 9044
9045out:
9046 btrfs_free_block_rsv(root, rsv);
9047
3893e33b
JB		 9048 if (ret && !err)
9049 err = ret;
a41ad394 9050
3893e33b 9051 return err;
39279cc3
CM		 9052}
9053
d352ac68
CM		 9054/*
9055 * create a new subvolume directory/inode (helper for the ioctl).
9056 */
d2fb3437 9057int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
63541927
FDBM		 9058 struct btrfs_root *new_root,
9059 struct btrfs_root *parent_root,
9060 u64 new_dirid)
39279cc3 9061{
39279cc3 9062 struct inode *inode;
76dda93c 9063 int err;
00e4e6b3 9064 u64 index = 0;
39279cc3 9065
12fc9d09
FA		 9066 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9067 new_dirid, new_dirid,
9068 S_IFDIR | (~current_umask() & S_IRWXUGO),
9069 &index);
54aa1f4d 9070 if (IS_ERR(inode))
f46b5a66 9071 return PTR_ERR(inode);
39279cc3
CM		 9072 inode->i_op = &btrfs_dir_inode_operations;
9073 inode->i_fop = &btrfs_dir_file_operations;
9074
bfe86848 9075 set_nlink(inode, 1);
dbe674a9 9076 btrfs_i_size_write(inode, 0);
b0d5d10f 9077 unlock_new_inode(inode);
3b96362c 9078
63541927
FDBM		 9079 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9080 if (err)
9081 btrfs_err(new_root->fs_info,
351fd353 9082 "error inheriting subvolume %llu properties: %d",
63541927
FDBM		 9083 new_root->root_key.objectid, err);
9084
76dda93c 9085 err = btrfs_update_inode(trans, new_root, inode);
cb8e7090 9086
76dda93c 9087 iput(inode);
ce598979 9088 return err;
39279cc3
CM		 9089}
9090
39279cc3
CM		 9091struct inode *btrfs_alloc_inode(struct super_block *sb)
9092{
9093 struct btrfs_inode *ei;
2ead6ae7 9094 struct inode *inode;
39279cc3
CM		 9095
9096 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9097 if (!ei)
9098 return NULL;
2ead6ae7
YZ		 9099
9100 ei->root = NULL;
2ead6ae7 9101 ei->generation = 0;
15ee9bc7 9102 ei->last_trans = 0;
257c62e1 9103 ei->last_sub_trans = 0;
e02119d5 9104 ei->logged_trans = 0;
2ead6ae7 9105 ei->delalloc_bytes = 0;
47059d93 9106 ei->defrag_bytes = 0;
2ead6ae7
YZ		 9107 ei->disk_i_size = 0;
9108 ei->flags = 0;
7709cde3 9109 ei->csum_bytes = 0;
2ead6ae7 9110 ei->index_cnt = (u64)-1;
67de1176 9111 ei->dir_index = 0;
2ead6ae7 9112 ei->last_unlink_trans = 0;
46d8bc34 9113 ei->last_log_commit = 0;
8089fe62 9114 ei->delayed_iput_count = 0;
2ead6ae7 9115
9e0baf60
JB		 9116 spin_lock_init(&ei->lock);
9117 ei->outstanding_extents = 0;
9118 ei->reserved_extents = 0;
2ead6ae7 9119
72ac3c0d 9120 ei->runtime_flags = 0;
261507a0 9121 ei->force_compress = BTRFS_COMPRESS_NONE;
2ead6ae7 9122
16cdcec7
MX		 9123 ei->delayed_node = NULL;
9124
9cc97d64 9125 ei->i_otime.tv_sec = 0;
9126 ei->i_otime.tv_nsec = 0;
9127
2ead6ae7 9128 inode = &ei->vfs_inode;
a8067e02 9129 extent_map_tree_init(&ei->extent_tree);
f993c883
DS		 9130 extent_io_tree_init(&ei->io_tree, &inode->i_data);
9131 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
0b32f4bb
JB		 9132 ei->io_tree.track_uptodate = 1;
9133 ei->io_failure_tree.track_uptodate = 1;
b812ce28 9134 atomic_set(&ei->sync_writers, 0);
2ead6ae7 9135 mutex_init(&ei->log_mutex);
f248679e 9136 mutex_init(&ei->delalloc_mutex);
e6dcd2dc 9137 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
2ead6ae7 9138 INIT_LIST_HEAD(&ei->delalloc_inodes);
8089fe62 9139 INIT_LIST_HEAD(&ei->delayed_iput);
2ead6ae7
YZ		 9140 RB_CLEAR_NODE(&ei->rb_node);
9141
9142 return inode;
39279cc3
CM		 9143}
9144
aaedb55b
JB		 9145#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9146void btrfs_test_destroy_inode(struct inode *inode)
9147{
9148 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9149 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9150}
9151#endif
9152
fa0d7e3d
NP		 9153static void btrfs_i_callback(struct rcu_head *head)
9154{
9155 struct inode *inode = container_of(head, struct inode, i_rcu);
fa0d7e3d
NP		 9156 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9157}
9158
39279cc3
CM		 9159void btrfs_destroy_inode(struct inode *inode)
9160{
e6dcd2dc 9161 struct btrfs_ordered_extent *ordered;
5a3f23d5
CM		 9162 struct btrfs_root *root = BTRFS_I(inode)->root;
9163
b3d9b7a3 9164 WARN_ON(!hlist_empty(&inode->i_dentry));
39279cc3 9165 WARN_ON(inode->i_data.nrpages);
9e0baf60
JB		 9166 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9167 WARN_ON(BTRFS_I(inode)->reserved_extents);
7709cde3
JB		 9168 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9169 WARN_ON(BTRFS_I(inode)->csum_bytes);
47059d93 9170 WARN_ON(BTRFS_I(inode)->defrag_bytes);
39279cc3 9171
a6dbd429
JB		 9172 /*
9173 * This can happen where we create an inode, but somebody else also
9174 * created the same inode and we need to destroy the one we already
9175 * created.
9176 */
9177 if (!root)
9178 goto free;
9179
8a35d95f
JB		 9180 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9181 &BTRFS_I(inode)->runtime_flags)) {
c2cf52eb 9182 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
c1c9ff7c 9183 btrfs_ino(inode));
8a35d95f 9184 atomic_dec(&root->orphan_inodes);
7b128766 9185 }
7b128766 9186
d397712b 9187 while (1) {
e6dcd2dc
CM		 9188 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9189 if (!ordered)
9190 break;
9191 else {
c2cf52eb 9192 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
c1c9ff7c 9193 ordered->file_offset, ordered->len);
e6dcd2dc
CM		 9194 btrfs_remove_ordered_extent(inode, ordered);
9195 btrfs_put_ordered_extent(ordered);
9196 btrfs_put_ordered_extent(ordered);
9197 }
9198 }
56fa9d07 9199 btrfs_qgroup_check_reserved_leak(inode);
5d4f98a2 9200 inode_tree_del(inode);
5b21f2ed 9201 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
a6dbd429 9202free:
fa0d7e3d 9203 call_rcu(&inode->i_rcu, btrfs_i_callback);
39279cc3
CM		 9204}
9205
45321ac5 9206int btrfs_drop_inode(struct inode *inode)
76dda93c
YZ		 9207{
9208 struct btrfs_root *root = BTRFS_I(inode)->root;
45321ac5 9209
6379ef9f
NA		 9210 if (root == NULL)
9211 return 1;
9212
fa6ac876 9213 /* the snap/subvol tree is on deleting */
69e9c6c6 9214 if (btrfs_root_refs(&root->root_item) == 0)
45321ac5 9215 return 1;
76dda93c 9216 else
45321ac5 9217 return generic_drop_inode(inode);
76dda93c
YZ		 9218}
9219
0ee0fda0 9220static void init_once(void *foo)
39279cc3
CM		 9221{
9222 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9223
9224 inode_init_once(&ei->vfs_inode);
9225}
9226
9227void btrfs_destroy_cachep(void)
9228{
8c0a8537
KS		 9229 /*
9230 * Make sure all delayed rcu free inodes are flushed before we
9231 * destroy cache.
9232 */
9233 rcu_barrier();
39279cc3
CM		 9234 if (btrfs_inode_cachep)
9235 kmem_cache_destroy(btrfs_inode_cachep);
9236 if (btrfs_trans_handle_cachep)
9237 kmem_cache_destroy(btrfs_trans_handle_cachep);
9238 if (btrfs_transaction_cachep)
9239 kmem_cache_destroy(btrfs_transaction_cachep);
39279cc3
CM		 9240 if (btrfs_path_cachep)
9241 kmem_cache_destroy(btrfs_path_cachep);
dc89e982
JB		 9242 if (btrfs_free_space_cachep)
9243 kmem_cache_destroy(btrfs_free_space_cachep);
39279cc3
CM		 9244}
9245
9246int btrfs_init_cachep(void)
9247{
837e1972 9248 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9601e3f6
CH		 9249 sizeof(struct btrfs_inode), 0,
9250 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
39279cc3
CM		 9251 if (!btrfs_inode_cachep)
9252 goto fail;
9601e3f6 9253
837e1972 9254 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9601e3f6
CH		 9255 sizeof(struct btrfs_trans_handle), 0,
9256 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 9257 if (!btrfs_trans_handle_cachep)
9258 goto fail;
9601e3f6 9259
837e1972 9260 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9601e3f6
CH		 9261 sizeof(struct btrfs_transaction), 0,
9262 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 9263 if (!btrfs_transaction_cachep)
9264 goto fail;
9601e3f6 9265
837e1972 9266 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9601e3f6
CH		 9267 sizeof(struct btrfs_path), 0,
9268 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 9269 if (!btrfs_path_cachep)
9270 goto fail;
9601e3f6 9271
837e1972 9272 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
dc89e982
JB		 9273 sizeof(struct btrfs_free_space), 0,
9274 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9275 if (!btrfs_free_space_cachep)
9276 goto fail;
9277
39279cc3
CM		 9278 return 0;
9279fail:
9280 btrfs_destroy_cachep();
9281 return -ENOMEM;
9282}
9283
9284static int btrfs_getattr(struct vfsmount *mnt,
9285 struct dentry *dentry, struct kstat *stat)
9286{
df0af1a5 9287 u64 delalloc_bytes;
2b0143b5 9288 struct inode *inode = d_inode(dentry);
fadc0d8b
DS		 9289 u32 blocksize = inode->i_sb->s_blocksize;
9290
39279cc3 9291 generic_fillattr(inode, stat);
0ee5dc67 9292 stat->dev = BTRFS_I(inode)->root->anon_dev;
d6667462 9293 stat->blksize = PAGE_CACHE_SIZE;
df0af1a5
MX		 9294
9295 spin_lock(&BTRFS_I(inode)->lock);
9296 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9297 spin_unlock(&BTRFS_I(inode)->lock);
fadc0d8b 9298 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
df0af1a5 9299 ALIGN(delalloc_bytes, blocksize)) >> 9;
39279cc3
CM		 9300 return 0;
9301}
9302
d397712b
CM		 9303static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9304 struct inode *new_dir, struct dentry *new_dentry)
39279cc3
CM		 9305{
9306 struct btrfs_trans_handle *trans;
9307 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4df27c4d 9308 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
2b0143b5
DH		 9309 struct inode *new_inode = d_inode(new_dentry);
9310 struct inode *old_inode = d_inode(old_dentry);
39279cc3 9311 struct timespec ctime = CURRENT_TIME;
00e4e6b3 9312 u64 index = 0;
4df27c4d 9313 u64 root_objectid;
39279cc3 9314 int ret;
33345d01 9315 u64 old_ino = btrfs_ino(old_inode);
39279cc3 9316
33345d01 9317 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
f679a840
YZ		 9318 return -EPERM;
9319
4df27c4d 9320 /* we only allow rename subvolume link between subvolumes */
33345d01 9321 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
3394e160
CM		 9322 return -EXDEV;
9323
33345d01
LZ		 9324 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9325 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
39279cc3 9326 return -ENOTEMPTY;
5f39d397 9327
4df27c4d
YZ		 9328 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9329 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9330 return -ENOTEMPTY;
9c52057c
CM		 9331
9332
9333 /* check for collisions, even if the name isn't there */
4871c158 9334 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9c52057c
CM		 9335 new_dentry->d_name.name,
9336 new_dentry->d_name.len);
9337
9338 if (ret) {
9339 if (ret == -EEXIST) {
9340 /* we shouldn't get
9341 * eexist without a new_inode */
fae7f21c 9342 if (WARN_ON(!new_inode)) {
9c52057c
CM		 9343 return ret;
9344 }
9345 } else {
9346 /* maybe -EOVERFLOW */
9347 return ret;
9348 }
9349 }
9350 ret = 0;
9351
5a3f23d5 9352 /*
8d875f95
CM		 9353 * we're using rename to replace one file with another. Start IO on it
9354 * now so we don't add too much work to the end of the transaction
5a3f23d5 9355 */
8d875f95 9356 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
5a3f23d5
CM		 9357 filemap_flush(old_inode->i_mapping);
9358
76dda93c 9359 /* close the racy window with snapshot create/destroy ioctl */
33345d01 9360 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 9361 down_read(&root->fs_info->subvol_sem);
a22285a6
YZ		 9362 /*
9363 * We want to reserve the absolute worst case amount of items. So if
9364 * both inodes are subvols and we need to unlink them then that would
9365 * require 4 item modifications, but if they are both normal inodes it
9366 * would require 5 item modifications, so we'll assume their normal
9367 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9368 * should cover the worst case number of items we'll modify.
9369 */
6e137ed3 9370 trans = btrfs_start_transaction(root, 11);
b44c59a8
JL		 9371 if (IS_ERR(trans)) {
9372 ret = PTR_ERR(trans);
9373 goto out_notrans;
9374 }
76dda93c 9375
4df27c4d
YZ		 9376 if (dest != root)
9377 btrfs_record_root_in_trans(trans, dest);
5f39d397 9378
a5719521
YZ		 9379 ret = btrfs_set_inode_index(new_dir, &index);
9380 if (ret)
9381 goto out_fail;
5a3f23d5 9382
67de1176 9383 BTRFS_I(old_inode)->dir_index = 0ULL;
33345d01 9384 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d 9385 /* force full log commit if subvolume involved. */
995946dd 9386 btrfs_set_log_full_commit(root->fs_info, trans);
4df27c4d 9387 } else {
a5719521
YZ		 9388 ret = btrfs_insert_inode_ref(trans, dest,
9389 new_dentry->d_name.name,
9390 new_dentry->d_name.len,
33345d01
LZ		 9391 old_ino,
9392 btrfs_ino(new_dir), index);
a5719521
YZ		 9393 if (ret)
9394 goto out_fail;
4df27c4d
YZ		 9395 /*
9396 * this is an ugly little race, but the rename is required
9397 * to make sure that if we crash, the inode is either at the
9398 * old name or the new one. pinning the log transaction lets
9399 * us make sure we don't allow a log commit to come in after
9400 * we unlink the name but before we add the new name back in.
9401 */
9402 btrfs_pin_log_trans(root);
9403 }
5a3f23d5 9404
0c4d2d95
JB		 9405 inode_inc_iversion(old_dir);
9406 inode_inc_iversion(new_dir);
9407 inode_inc_iversion(old_inode);
39279cc3
CM		 9408 old_dir->i_ctime = old_dir->i_mtime = ctime;
9409 new_dir->i_ctime = new_dir->i_mtime = ctime;
9410 old_inode->i_ctime = ctime;
5f39d397 9411
12fcfd22
CM		 9412 if (old_dentry->d_parent != new_dentry->d_parent)
9413 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9414
33345d01 9415 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 9416 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9417 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9418 old_dentry->d_name.name,
9419 old_dentry->d_name.len);
9420 } else {
92986796 9421 ret = __btrfs_unlink_inode(trans, root, old_dir,
2b0143b5 9422 d_inode(old_dentry),
92986796
AV		 9423 old_dentry->d_name.name,
9424 old_dentry->d_name.len);
9425 if (!ret)
9426 ret = btrfs_update_inode(trans, root, old_inode);
4df27c4d 9427 }
79787eaa
JM		 9428 if (ret) {
9429 btrfs_abort_transaction(trans, root, ret);
9430 goto out_fail;
9431 }
39279cc3
CM		 9432
9433 if (new_inode) {
0c4d2d95 9434 inode_inc_iversion(new_inode);
39279cc3 9435 new_inode->i_ctime = CURRENT_TIME;
33345d01 9436 if (unlikely(btrfs_ino(new_inode) ==
4df27c4d
YZ		 9437 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9438 root_objectid = BTRFS_I(new_inode)->location.objectid;
9439 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9440 root_objectid,
9441 new_dentry->d_name.name,
9442 new_dentry->d_name.len);
9443 BUG_ON(new_inode->i_nlink == 0);
9444 } else {
9445 ret = btrfs_unlink_inode(trans, dest, new_dir,
2b0143b5 9446 d_inode(new_dentry),
4df27c4d
YZ		 9447 new_dentry->d_name.name,
9448 new_dentry->d_name.len);
9449 }
4ef31a45 9450 if (!ret && new_inode->i_nlink == 0)
2b0143b5 9451 ret = btrfs_orphan_add(trans, d_inode(new_dentry));
79787eaa
JM		 9452 if (ret) {
9453 btrfs_abort_transaction(trans, root, ret);
9454 goto out_fail;
9455 }
39279cc3 9456 }
aec7477b 9457
4df27c4d
YZ		 9458 ret = btrfs_add_link(trans, new_dir, old_inode,
9459 new_dentry->d_name.name,
a5719521 9460 new_dentry->d_name.len, 0, index);
79787eaa
JM		 9461 if (ret) {
9462 btrfs_abort_transaction(trans, root, ret);
9463 goto out_fail;
9464 }
39279cc3 9465
67de1176
MX		 9466 if (old_inode->i_nlink == 1)
9467 BTRFS_I(old_inode)->dir_index = index;
9468
33345d01 9469 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
10d9f309 9470 struct dentry *parent = new_dentry->d_parent;
6a912213 9471 btrfs_log_new_name(trans, old_inode, old_dir, parent);
4df27c4d
YZ		 9472 btrfs_end_log_trans(root);
9473 }
39279cc3 9474out_fail:
7ad85bb7 9475 btrfs_end_transaction(trans, root);
b44c59a8 9476out_notrans:
33345d01 9477 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 9478 up_read(&root->fs_info->subvol_sem);
9ed74f2d 9479
39279cc3
CM		 9480 return ret;
9481}
9482
80ace85c
MS		 9483static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9484 struct inode *new_dir, struct dentry *new_dentry,
9485 unsigned int flags)
9486{
9487 if (flags & ~RENAME_NOREPLACE)
9488 return -EINVAL;
9489
9490 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
9491}
9492
8ccf6f19
MX		 9493static void btrfs_run_delalloc_work(struct btrfs_work *work)
9494{
9495 struct btrfs_delalloc_work *delalloc_work;
9f23e289 9496 struct inode *inode;
8ccf6f19
MX		 9497
9498 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9499 work);
9f23e289 9500 inode = delalloc_work->inode;
30424601
DS		 9501 filemap_flush(inode->i_mapping);
9502 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9503 &BTRFS_I(inode)->runtime_flags))
9f23e289 9504 filemap_flush(inode->i_mapping);
8ccf6f19
MX		 9505
9506 if (delalloc_work->delay_iput)
9f23e289 9507 btrfs_add_delayed_iput(inode);
8ccf6f19 9508 else
9f23e289 9509 iput(inode);
8ccf6f19
MX		 9510 complete(&delalloc_work->completion);
9511}
9512
9513struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
651d494a 9514 int delay_iput)
8ccf6f19
MX		 9515{
9516 struct btrfs_delalloc_work *work;
9517
100d5702 9518 work = kmalloc(sizeof(*work), GFP_NOFS);
8ccf6f19
MX		 9519 if (!work)
9520 return NULL;
9521
9522 init_completion(&work->completion);
9523 INIT_LIST_HEAD(&work->list);
9524 work->inode = inode;
8ccf6f19 9525 work->delay_iput = delay_iput;
9e0af237
LB		 9526 WARN_ON_ONCE(!inode);
9527 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9528 btrfs_run_delalloc_work, NULL, NULL);
8ccf6f19
MX		 9529
9530 return work;
9531}
9532
9533void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9534{
9535 wait_for_completion(&work->completion);
100d5702 9536 kfree(work);
8ccf6f19
MX		 9537}
9538
d352ac68
CM		 9539/*
9540 * some fairly slow code that needs optimization. This walks the list
9541 * of all the inodes with pending delalloc and forces them to disk.
9542 */
6c255e67
MX		 9543static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9544 int nr)
ea8c2819 9545{
ea8c2819 9546 struct btrfs_inode *binode;
5b21f2ed 9547 struct inode *inode;
8ccf6f19
MX		 9548 struct btrfs_delalloc_work *work, *next;
9549 struct list_head works;
1eafa6c7 9550 struct list_head splice;
8ccf6f19 9551 int ret = 0;
ea8c2819 9552
8ccf6f19 9553 INIT_LIST_HEAD(&works);
1eafa6c7 9554 INIT_LIST_HEAD(&splice);
63607cc8 9555
573bfb72 9556 mutex_lock(&root->delalloc_mutex);
eb73c1b7
MX		 9557 spin_lock(&root->delalloc_lock);
9558 list_splice_init(&root->delalloc_inodes, &splice);
1eafa6c7
MX		 9559 while (!list_empty(&splice)) {
9560 binode = list_entry(splice.next, struct btrfs_inode,
ea8c2819 9561 delalloc_inodes);
1eafa6c7 9562
eb73c1b7
MX		 9563 list_move_tail(&binode->delalloc_inodes,
9564 &root->delalloc_inodes);
5b21f2ed 9565 inode = igrab(&binode->vfs_inode);
df0af1a5 9566 if (!inode) {
eb73c1b7 9567 cond_resched_lock(&root->delalloc_lock);
1eafa6c7 9568 continue;
df0af1a5 9569 }
eb73c1b7 9570 spin_unlock(&root->delalloc_lock);
1eafa6c7 9571
651d494a 9572 work = btrfs_alloc_delalloc_work(inode, delay_iput);
5d99a998 9573 if (!work) {
f4ab9ea7
JB		 9574 if (delay_iput)
9575 btrfs_add_delayed_iput(inode);
9576 else
9577 iput(inode);
1eafa6c7 9578 ret = -ENOMEM;
a1ecaabb 9579 goto out;
5b21f2ed 9580 }
1eafa6c7 9581 list_add_tail(&work->list, &works);
a44903ab
QW		 9582 btrfs_queue_work(root->fs_info->flush_workers,
9583 &work->work);
6c255e67
MX		 9584 ret++;
9585 if (nr != -1 && ret >= nr)
a1ecaabb 9586 goto out;
5b21f2ed 9587 cond_resched();
eb73c1b7 9588 spin_lock(&root->delalloc_lock);
ea8c2819 9589 }
eb73c1b7 9590 spin_unlock(&root->delalloc_lock);
8c8bee1d 9591
a1ecaabb 9592out:
eb73c1b7
MX		 9593 list_for_each_entry_safe(work, next, &works, list) {
9594 list_del_init(&work->list);
9595 btrfs_wait_and_free_delalloc_work(work);
9596 }
9597
9598 if (!list_empty_careful(&splice)) {
9599 spin_lock(&root->delalloc_lock);
9600 list_splice_tail(&splice, &root->delalloc_inodes);
9601 spin_unlock(&root->delalloc_lock);
9602 }
573bfb72 9603 mutex_unlock(&root->delalloc_mutex);
eb73c1b7
MX		 9604 return ret;
9605}
1eafa6c7 9606
eb73c1b7
MX		 9607int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9608{
9609 int ret;
1eafa6c7 9610
2c21b4d7 9611 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
eb73c1b7
MX		 9612 return -EROFS;
9613
6c255e67
MX		 9614 ret = __start_delalloc_inodes(root, delay_iput, -1);
9615 if (ret > 0)
9616 ret = 0;
eb73c1b7
MX		 9617 /*
9618 * the filemap_flush will queue IO into the worker threads, but
8c8bee1d
CM		 9619 * we have to make sure the IO is actually started and that
9620 * ordered extents get created before we return
9621 */
9622 atomic_inc(&root->fs_info->async_submit_draining);
d397712b 9623 while (atomic_read(&root->fs_info->nr_async_submits) ||
771ed689 9624 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 9625 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 9626 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9627 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 9628 }
9629 atomic_dec(&root->fs_info->async_submit_draining);
eb73c1b7
MX		 9630 return ret;
9631}
9632
6c255e67
MX		 9633int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9634 int nr)
eb73c1b7
MX		 9635{
9636 struct btrfs_root *root;
9637 struct list_head splice;
9638 int ret;
9639
2c21b4d7 9640 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
eb73c1b7
MX		 9641 return -EROFS;
9642
9643 INIT_LIST_HEAD(&splice);
9644
573bfb72 9645 mutex_lock(&fs_info->delalloc_root_mutex);
eb73c1b7
MX		 9646 spin_lock(&fs_info->delalloc_root_lock);
9647 list_splice_init(&fs_info->delalloc_roots, &splice);
6c255e67 9648 while (!list_empty(&splice) && nr) {
eb73c1b7
MX		 9649 root = list_first_entry(&splice, struct btrfs_root,
9650 delalloc_root);
9651 root = btrfs_grab_fs_root(root);
9652 BUG_ON(!root);
9653 list_move_tail(&root->delalloc_root,
9654 &fs_info->delalloc_roots);
9655 spin_unlock(&fs_info->delalloc_root_lock);
9656
6c255e67 9657 ret = __start_delalloc_inodes(root, delay_iput, nr);
eb73c1b7 9658 btrfs_put_fs_root(root);
6c255e67 9659 if (ret < 0)
eb73c1b7
MX		 9660 goto out;
9661
6c255e67
MX		 9662 if (nr != -1) {
9663 nr -= ret;
9664 WARN_ON(nr < 0);
9665 }
eb73c1b7 9666 spin_lock(&fs_info->delalloc_root_lock);
8ccf6f19 9667 }
eb73c1b7 9668 spin_unlock(&fs_info->delalloc_root_lock);
1eafa6c7 9669
6c255e67 9670 ret = 0;
eb73c1b7
MX		 9671 atomic_inc(&fs_info->async_submit_draining);
9672 while (atomic_read(&fs_info->nr_async_submits) ||
9673 atomic_read(&fs_info->async_delalloc_pages)) {
9674 wait_event(fs_info->async_submit_wait,
9675 (atomic_read(&fs_info->nr_async_submits) == 0 &&
9676 atomic_read(&fs_info->async_delalloc_pages) == 0));
9677 }
9678 atomic_dec(&fs_info->async_submit_draining);
eb73c1b7 9679out:
1eafa6c7 9680 if (!list_empty_careful(&splice)) {
eb73c1b7
MX		 9681 spin_lock(&fs_info->delalloc_root_lock);
9682 list_splice_tail(&splice, &fs_info->delalloc_roots);
9683 spin_unlock(&fs_info->delalloc_root_lock);
1eafa6c7 9684 }
573bfb72 9685 mutex_unlock(&fs_info->delalloc_root_mutex);
8ccf6f19 9686 return ret;
ea8c2819
CM		 9687}
9688
39279cc3
CM		 9689static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9690 const char *symname)
9691{
9692 struct btrfs_trans_handle *trans;
9693 struct btrfs_root *root = BTRFS_I(dir)->root;
9694 struct btrfs_path *path;
9695 struct btrfs_key key;
1832a6d5 9696 struct inode *inode = NULL;
39279cc3
CM		 9697 int err;
9698 int drop_inode = 0;
9699 u64 objectid;
67871254 9700 u64 index = 0;
39279cc3
CM		 9701 int name_len;
9702 int datasize;
5f39d397 9703 unsigned long ptr;
39279cc3 9704 struct btrfs_file_extent_item *ei;
5f39d397 9705 struct extent_buffer *leaf;
39279cc3 9706
f06becc4 9707 name_len = strlen(symname);
39279cc3
CM		 9708 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9709 return -ENAMETOOLONG;
1832a6d5 9710
9ed74f2d
JB		 9711 /*
9712 * 2 items for inode item and ref
9713 * 2 items for dir items
9269d12b
FM		 9714 * 1 item for updating parent inode item
9715 * 1 item for the inline extent item
9ed74f2d
JB		 9716 * 1 item for xattr if selinux is on
9717 */
9269d12b 9718 trans = btrfs_start_transaction(root, 7);
a22285a6
YZ		 9719 if (IS_ERR(trans))
9720 return PTR_ERR(trans);
1832a6d5 9721
581bb050
LZ		 9722 err = btrfs_find_free_ino(root, &objectid);
9723 if (err)
9724 goto out_unlock;
9725
aec7477b 9726 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 9727 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 9728 S_IFLNK|S_IRWXUGO, &index);
7cf96da3
TI		 9729 if (IS_ERR(inode)) {
9730 err = PTR_ERR(inode);
39279cc3 9731 goto out_unlock;
7cf96da3 9732 }
39279cc3 9733
ad19db71
CS		 9734 /*
9735 * If the active LSM wants to access the inode during
9736 * d_instantiate it needs these. Smack checks to see
9737 * if the filesystem supports xattrs by looking at the
9738 * ops vector.
9739 */
9740 inode->i_fop = &btrfs_file_operations;
9741 inode->i_op = &btrfs_file_inode_operations;
b0d5d10f 9742 inode->i_mapping->a_ops = &btrfs_aops;
b0d5d10f
CM		 9743 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9744
9745 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9746 if (err)
9747 goto out_unlock_inode;
ad19db71 9748
39279cc3 9749 path = btrfs_alloc_path();
d8926bb3
MF		 9750 if (!path) {
9751 err = -ENOMEM;
b0d5d10f 9752 goto out_unlock_inode;
d8926bb3 9753 }
33345d01 9754 key.objectid = btrfs_ino(inode);
39279cc3 9755 key.offset = 0;
962a298f 9756 key.type = BTRFS_EXTENT_DATA_KEY;
39279cc3
CM		 9757 datasize = btrfs_file_extent_calc_inline_size(name_len);
9758 err = btrfs_insert_empty_item(trans, root, path, &key,
9759 datasize);
54aa1f4d 9760 if (err) {
b0839166 9761 btrfs_free_path(path);
b0d5d10f 9762 goto out_unlock_inode;
54aa1f4d 9763 }
5f39d397
CM		 9764 leaf = path->nodes[0];
9765 ei = btrfs_item_ptr(leaf, path->slots[0],
9766 struct btrfs_file_extent_item);
9767 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9768 btrfs_set_file_extent_type(leaf, ei,
39279cc3 9769 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 9770 btrfs_set_file_extent_encryption(leaf, ei, 0);
9771 btrfs_set_file_extent_compression(leaf, ei, 0);
9772 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9773 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9774
39279cc3 9775 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 9776 write_extent_buffer(leaf, symname, ptr, name_len);
9777 btrfs_mark_buffer_dirty(leaf);
39279cc3 9778 btrfs_free_path(path);
5f39d397 9779
39279cc3
CM		 9780 inode->i_op = &btrfs_symlink_inode_operations;
9781 inode->i_mapping->a_ops = &btrfs_symlink_aops;
d899e052 9782 inode_set_bytes(inode, name_len);
f06becc4 9783 btrfs_i_size_write(inode, name_len);
54aa1f4d 9784 err = btrfs_update_inode(trans, root, inode);
d50866d0
FM		 9785 /*
9786 * Last step, add directory indexes for our symlink inode. This is the
9787 * last step to avoid extra cleanup of these indexes if an error happens
9788 * elsewhere above.
9789 */
9790 if (!err)
9791 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
b0d5d10f 9792 if (err) {
54aa1f4d 9793 drop_inode = 1;
b0d5d10f
CM		 9794 goto out_unlock_inode;
9795 }
9796
9797 unlock_new_inode(inode);
9798 d_instantiate(dentry, inode);
39279cc3
CM		 9799
9800out_unlock:
7ad85bb7 9801 btrfs_end_transaction(trans, root);
39279cc3
CM		 9802 if (drop_inode) {
9803 inode_dec_link_count(inode);
9804 iput(inode);
9805 }
b53d3f5d 9806 btrfs_btree_balance_dirty(root);
39279cc3 9807 return err;
b0d5d10f
CM		 9808
9809out_unlock_inode:
9810 drop_inode = 1;
9811 unlock_new_inode(inode);
9812 goto out_unlock;
39279cc3 9813}
16432985 9814
0af3d00b
JB		 9815static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9816 u64 start, u64 num_bytes, u64 min_size,
9817 loff_t actual_len, u64 *alloc_hint,
9818 struct btrfs_trans_handle *trans)
d899e052 9819{
5dc562c5
JB		 9820 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9821 struct extent_map *em;
d899e052
YZ		 9822 struct btrfs_root *root = BTRFS_I(inode)->root;
9823 struct btrfs_key ins;
d899e052 9824 u64 cur_offset = start;
55a61d1d 9825 u64 i_size;
154ea289 9826 u64 cur_bytes;
0b670dc4 9827 u64 last_alloc = (u64)-1;
d899e052 9828 int ret = 0;
0af3d00b 9829 bool own_trans = true;
d899e052 9830
0af3d00b
JB		 9831 if (trans)
9832 own_trans = false;
d899e052 9833 while (num_bytes > 0) {
0af3d00b
JB		 9834 if (own_trans) {
9835 trans = btrfs_start_transaction(root, 3);
9836 if (IS_ERR(trans)) {
9837 ret = PTR_ERR(trans);
9838 break;
9839 }
5a303d5d
YZ		 9840 }
9841
ee22184b 9842 cur_bytes = min_t(u64, num_bytes, SZ_256M);
154ea289 9843 cur_bytes = max(cur_bytes, min_size);
0b670dc4
JB		 9844 /*
9845 * If we are severely fragmented we could end up with really
9846 * small allocations, so if the allocator is returning small
9847 * chunks lets make its job easier by only searching for those
9848 * sized chunks.
9849 */
9850 cur_bytes = min(cur_bytes, last_alloc);
00361589 9851 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
e570fd27 9852 *alloc_hint, &ins, 1, 0);
5a303d5d 9853 if (ret) {
0af3d00b
JB		 9854 if (own_trans)
9855 btrfs_end_transaction(trans, root);
a22285a6 9856 break;
d899e052 9857 }
5a303d5d 9858
0b670dc4 9859 last_alloc = ins.offset;
d899e052
YZ		 9860 ret = insert_reserved_file_extent(trans, inode,
9861 cur_offset, ins.objectid,
9862 ins.offset, ins.offset,
920bbbfb 9863 ins.offset, 0, 0, 0,
d899e052 9864 BTRFS_FILE_EXTENT_PREALLOC);
79787eaa 9865 if (ret) {
857cc2fc 9866 btrfs_free_reserved_extent(root, ins.objectid,
e570fd27 9867 ins.offset, 0);
79787eaa
JM		 9868 btrfs_abort_transaction(trans, root, ret);
9869 if (own_trans)
9870 btrfs_end_transaction(trans, root);
9871 break;
9872 }
31193213 9873
a1ed835e
CM		 9874 btrfs_drop_extent_cache(inode, cur_offset,
9875 cur_offset + ins.offset -1, 0);
5a303d5d 9876
5dc562c5
JB		 9877 em = alloc_extent_map();
9878 if (!em) {
9879 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9880 &BTRFS_I(inode)->runtime_flags);
9881 goto next;
9882 }
9883
9884 em->start = cur_offset;
9885 em->orig_start = cur_offset;
9886 em->len = ins.offset;
9887 em->block_start = ins.objectid;
9888 em->block_len = ins.offset;
b4939680 9889 em->orig_block_len = ins.offset;
cc95bef6 9890 em->ram_bytes = ins.offset;
5dc562c5
JB		 9891 em->bdev = root->fs_info->fs_devices->latest_bdev;
9892 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9893 em->generation = trans->transid;
9894
9895 while (1) {
9896 write_lock(&em_tree->lock);
09a2a8f9 9897 ret = add_extent_mapping(em_tree, em, 1);
5dc562c5
JB		 9898 write_unlock(&em_tree->lock);
9899 if (ret != -EEXIST)
9900 break;
9901 btrfs_drop_extent_cache(inode, cur_offset,
9902 cur_offset + ins.offset - 1,
9903 0);
9904 }
9905 free_extent_map(em);
9906next:
d899e052
YZ		 9907 num_bytes -= ins.offset;
9908 cur_offset += ins.offset;
efa56464 9909 *alloc_hint = ins.objectid + ins.offset;
5a303d5d 9910
0c4d2d95 9911 inode_inc_iversion(inode);
d899e052 9912 inode->i_ctime = CURRENT_TIME;
6cbff00f 9913 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
d899e052 9914 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
efa56464
YZ		 9915 (actual_len > inode->i_size) &&
9916 (cur_offset > inode->i_size)) {
d1ea6a61 9917 if (cur_offset > actual_len)
55a61d1d 9918 i_size = actual_len;
d1ea6a61 9919 else
55a61d1d
JB		 9920 i_size = cur_offset;
9921 i_size_write(inode, i_size);
9922 btrfs_ordered_update_i_size(inode, i_size, NULL);
5a303d5d
YZ		 9923 }
9924
d899e052 9925 ret = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 9926
9927 if (ret) {
9928 btrfs_abort_transaction(trans, root, ret);
9929 if (own_trans)
9930 btrfs_end_transaction(trans, root);
9931 break;
9932 }
d899e052 9933
0af3d00b
JB		 9934 if (own_trans)
9935 btrfs_end_transaction(trans, root);
5a303d5d 9936 }
d899e052
YZ		 9937 return ret;
9938}
9939
0af3d00b
JB		 9940int btrfs_prealloc_file_range(struct inode *inode, int mode,
9941 u64 start, u64 num_bytes, u64 min_size,
9942 loff_t actual_len, u64 *alloc_hint)
9943{
9944 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9945 min_size, actual_len, alloc_hint,
9946 NULL);
9947}
9948
9949int btrfs_prealloc_file_range_trans(struct inode *inode,
9950 struct btrfs_trans_handle *trans, int mode,
9951 u64 start, u64 num_bytes, u64 min_size,
9952 loff_t actual_len, u64 *alloc_hint)
9953{
9954 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9955 min_size, actual_len, alloc_hint, trans);
9956}
9957
e6dcd2dc
CM		 9958static int btrfs_set_page_dirty(struct page *page)
9959{
e6dcd2dc
CM		 9960 return __set_page_dirty_nobuffers(page);
9961}
9962
10556cb2 9963static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd 9964{
b83cc969 9965 struct btrfs_root *root = BTRFS_I(inode)->root;
cb6db4e5 9966 umode_t mode = inode->i_mode;
b83cc969 9967
cb6db4e5
JM		 9968 if (mask & MAY_WRITE &&
9969 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9970 if (btrfs_root_readonly(root))
9971 return -EROFS;
9972 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9973 return -EACCES;
9974 }
2830ba7f 9975 return generic_permission(inode, mask);
fdebe2bd 9976}
39279cc3 9977
ef3b9af5
FM		 9978static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9979{
9980 struct btrfs_trans_handle *trans;
9981 struct btrfs_root *root = BTRFS_I(dir)->root;
9982 struct inode *inode = NULL;
9983 u64 objectid;
9984 u64 index;
9985 int ret = 0;
9986
9987 /*
9988 * 5 units required for adding orphan entry
9989 */
9990 trans = btrfs_start_transaction(root, 5);
9991 if (IS_ERR(trans))
9992 return PTR_ERR(trans);
9993
9994 ret = btrfs_find_free_ino(root, &objectid);
9995 if (ret)
9996 goto out;
9997
9998 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9999 btrfs_ino(dir), objectid, mode, &index);
10000 if (IS_ERR(inode)) {
10001 ret = PTR_ERR(inode);
10002 inode = NULL;
10003 goto out;
10004 }
10005
ef3b9af5
FM		 10006 inode->i_fop = &btrfs_file_operations;
10007 inode->i_op = &btrfs_file_inode_operations;
10008
10009 inode->i_mapping->a_ops = &btrfs_aops;
ef3b9af5
FM		 10010 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10011
b0d5d10f
CM		 10012 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10013 if (ret)
10014 goto out_inode;
10015
10016 ret = btrfs_update_inode(trans, root, inode);
10017 if (ret)
10018 goto out_inode;
ef3b9af5
FM		 10019 ret = btrfs_orphan_add(trans, inode);
10020 if (ret)
b0d5d10f 10021 goto out_inode;
ef3b9af5 10022
5762b5c9
FM		 10023 /*
10024 * We set number of links to 0 in btrfs_new_inode(), and here we set
10025 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10026 * through:
10027 *
10028 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10029 */
10030 set_nlink(inode, 1);
b0d5d10f 10031 unlock_new_inode(inode);
ef3b9af5
FM		 10032 d_tmpfile(dentry, inode);
10033 mark_inode_dirty(inode);
10034
10035out:
10036 btrfs_end_transaction(trans, root);
10037 if (ret)
10038 iput(inode);
10039 btrfs_balance_delayed_items(root);
10040 btrfs_btree_balance_dirty(root);
ef3b9af5 10041 return ret;
b0d5d10f
CM		 10042
10043out_inode:
10044 unlock_new_inode(inode);
10045 goto out;
10046
ef3b9af5
FM		 10047}
10048
b38ef71c
FM		 10049/* Inspired by filemap_check_errors() */
10050int btrfs_inode_check_errors(struct inode *inode)
10051{
10052 int ret = 0;
10053
10054 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
10055 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
10056 ret = -ENOSPC;
10057 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
10058 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
10059 ret = -EIO;
10060
10061 return ret;
10062}
10063
6e1d5dcc 10064static const struct inode_operations btrfs_dir_inode_operations = {
3394e160 10065 .getattr = btrfs_getattr,
39279cc3
CM		 10066 .lookup = btrfs_lookup,
10067 .create = btrfs_create,
10068 .unlink = btrfs_unlink,
10069 .link = btrfs_link,
10070 .mkdir = btrfs_mkdir,
10071 .rmdir = btrfs_rmdir,
80ace85c 10072 .rename2 = btrfs_rename2,
39279cc3
CM		 10073 .symlink = btrfs_symlink,
10074 .setattr = btrfs_setattr,
618e21d5 10075 .mknod = btrfs_mknod,
95819c05
CH		 10076 .setxattr = btrfs_setxattr,
10077 .getxattr = btrfs_getxattr,
5103e947 10078 .listxattr = btrfs_listxattr,
95819c05 10079 .removexattr = btrfs_removexattr,
fdebe2bd 10080 .permission = btrfs_permission,
4e34e719 10081 .get_acl = btrfs_get_acl,
996a710d 10082 .set_acl = btrfs_set_acl,
93fd63c2 10083 .update_time = btrfs_update_time,
ef3b9af5 10084 .tmpfile = btrfs_tmpfile,
39279cc3 10085};
6e1d5dcc 10086static const struct inode_operations btrfs_dir_ro_inode_operations = {
39279cc3 10087 .lookup = btrfs_lookup,
fdebe2bd 10088 .permission = btrfs_permission,
4e34e719 10089 .get_acl = btrfs_get_acl,
996a710d 10090 .set_acl = btrfs_set_acl,
93fd63c2 10091 .update_time = btrfs_update_time,
39279cc3 10092};
76dda93c 10093
828c0950 10094static const struct file_operations btrfs_dir_file_operations = {
39279cc3
CM		 10095 .llseek = generic_file_llseek,
10096 .read = generic_read_dir,
9cdda8d3 10097 .iterate = btrfs_real_readdir,
34287aa3 10098 .unlocked_ioctl = btrfs_ioctl,
39279cc3 10099#ifdef CONFIG_COMPAT
34287aa3 10100 .compat_ioctl = btrfs_ioctl,
39279cc3 10101#endif
6bf13c0c 10102 .release = btrfs_release_file,
e02119d5 10103 .fsync = btrfs_sync_file,
39279cc3
CM		 10104};
10105
20e5506b 10106static const struct extent_io_ops btrfs_extent_io_ops = {
07157aac 10107 .fill_delalloc = run_delalloc_range,
065631f6 10108 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 10109 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 10110 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 10111 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 10112 .writepage_start_hook = btrfs_writepage_start_hook,
b0c68f8b
CM		 10113 .set_bit_hook = btrfs_set_bit_hook,
10114 .clear_bit_hook = btrfs_clear_bit_hook,
9ed74f2d
JB		 10115 .merge_extent_hook = btrfs_merge_extent_hook,
10116 .split_extent_hook = btrfs_split_extent_hook,
07157aac
CM		 10117};
10118
35054394
CM		 10119/*
10120 * btrfs doesn't support the bmap operation because swapfiles
10121 * use bmap to make a mapping of extents in the file. They assume
10122 * these extents won't change over the life of the file and they
10123 * use the bmap result to do IO directly to the drive.
10124 *
10125 * the btrfs bmap call would return logical addresses that aren't
10126 * suitable for IO and they also will change frequently as COW
10127 * operations happen. So, swapfile + btrfs == corruption.
10128 *
10129 * For now we're avoiding this by dropping bmap.
10130 */
7f09410b 10131static const struct address_space_operations btrfs_aops = {
39279cc3
CM		 10132 .readpage = btrfs_readpage,
10133 .writepage = btrfs_writepage,
b293f02e 10134 .writepages = btrfs_writepages,
3ab2fb5a 10135 .readpages = btrfs_readpages,
16432985 10136 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 10137 .invalidatepage = btrfs_invalidatepage,
10138 .releasepage = btrfs_releasepage,
e6dcd2dc 10139 .set_page_dirty = btrfs_set_page_dirty,
465fdd97 10140 .error_remove_page = generic_error_remove_page,
39279cc3
CM		 10141};
10142
7f09410b 10143static const struct address_space_operations btrfs_symlink_aops = {
39279cc3
CM		 10144 .readpage = btrfs_readpage,
10145 .writepage = btrfs_writepage,
2bf5a725
CM		 10146 .invalidatepage = btrfs_invalidatepage,
10147 .releasepage = btrfs_releasepage,
39279cc3
CM		 10148};
10149
6e1d5dcc 10150static const struct inode_operations btrfs_file_inode_operations = {
39279cc3
CM		 10151 .getattr = btrfs_getattr,
10152 .setattr = btrfs_setattr,
95819c05
CH		 10153 .setxattr = btrfs_setxattr,
10154 .getxattr = btrfs_getxattr,
5103e947 10155 .listxattr = btrfs_listxattr,
95819c05 10156 .removexattr = btrfs_removexattr,
fdebe2bd 10157 .permission = btrfs_permission,
1506fcc8 10158 .fiemap = btrfs_fiemap,
4e34e719 10159 .get_acl = btrfs_get_acl,
996a710d 10160 .set_acl = btrfs_set_acl,
e41f941a 10161 .update_time = btrfs_update_time,
39279cc3 10162};
6e1d5dcc 10163static const struct inode_operations btrfs_special_inode_operations = {
618e21d5
JB		 10164 .getattr = btrfs_getattr,
10165 .setattr = btrfs_setattr,
fdebe2bd 10166 .permission = btrfs_permission,
95819c05
CH		 10167 .setxattr = btrfs_setxattr,
10168 .getxattr = btrfs_getxattr,
33268eaf 10169 .listxattr = btrfs_listxattr,
95819c05 10170 .removexattr = btrfs_removexattr,
4e34e719 10171 .get_acl = btrfs_get_acl,
996a710d 10172 .set_acl = btrfs_set_acl,
e41f941a 10173 .update_time = btrfs_update_time,
618e21d5 10174};
6e1d5dcc 10175static const struct inode_operations btrfs_symlink_inode_operations = {
39279cc3
CM		 10176 .readlink = generic_readlink,
10177 .follow_link = page_follow_link_light,
10178 .put_link = page_put_link,
f209561a 10179 .getattr = btrfs_getattr,
22c44fe6 10180 .setattr = btrfs_setattr,
fdebe2bd 10181 .permission = btrfs_permission,
0279b4cd
JO		 10182 .setxattr = btrfs_setxattr,
10183 .getxattr = btrfs_getxattr,
10184 .listxattr = btrfs_listxattr,
10185 .removexattr = btrfs_removexattr,
e41f941a 10186 .update_time = btrfs_update_time,
39279cc3 10187};
76dda93c 10188
82d339d9 10189const struct dentry_operations btrfs_dentry_operations = {
76dda93c 10190 .d_delete = btrfs_dentry_delete,
b4aff1f8 10191 .d_release = btrfs_dentry_release,
76dda93c 10192};
Linux 6.x block layer and io_uring tree(s)