projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
btrfs: sink gfp parameter to convert_extent_bit
[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,
09cbfeaf 197 PAGE_SIZE);
c8b97818 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,
09cbfeaf 211 start >> PAGE_SHIFT);
c8b97818 212 btrfs_set_file_extent_compression(leaf, ei, 0);
7ac687d9 213 kaddr = kmap_atomic(page);
09cbfeaf 214 offset = start & (PAGE_SIZE - 1);
c8b97818 215 write_extent_buffer(leaf, kaddr + offset, ptr, size);
7ac687d9 216 kunmap_atomic(kaddr);
09cbfeaf 217 put_page(page);
c8b97818
CM		 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 ||
0c29ba99 266 actual_end > root->sectorsize ||
354877be 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 */
09cbfeaf 325 btrfs_qgroup_free_data(inode, 0, PAGE_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;
09cbfeaf
KS		 438 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
439 nr_pages = min_t(unsigned long, nr_pages, SZ_128K / PAGE_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 &
09cbfeaf 517 (PAGE_SIZE - 1);
c8b97818
CM		 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 526 memset(kaddr + offset, 0,
09cbfeaf 527 PAGE_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 */
09cbfeaf 583 total_in = ALIGN(total_in, PAGE_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);
09cbfeaf 597 put_page(pages[i]);
c8b97818
CM		 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);
09cbfeaf 653 put_page(pages[i]);
771ed689 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);
09cbfeaf 667 put_page(async_extent->pages[i]);
40ae837b
FM		 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 968 *nr_written = *nr_written +
09cbfeaf 969 (end - start + PAGE_SIZE) / PAGE_SIZE;
771ed689 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;
09cbfeaf
KS		 1109 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1110 PAGE_SHIFT;
771ed689 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
09cbfeaf
KS		 1167 nr_pages = (cur_end - start + PAGE_SIZE) >>
1168 PAGE_SHIFT;
771ed689
CM		 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{
09cbfeaf 1963 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
ea8c2819 1964 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
7cd8c752 1965 cached_state);
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);
09cbfeaf 1996 page_end = page_offset(page) + PAGE_SIZE - 1;
247e743c 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 2004
dbfdb6d1 2005 ordered = btrfs_lookup_ordered_range(inode, page_start,
09cbfeaf 2006 PAGE_SIZE);
4a096752 2007 if (ordered) {
2ac55d41
JB		 2008 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2009 page_end, &cached_state, GFP_NOFS);
4a096752
CM		 2010 unlock_page(page);
2011 btrfs_start_ordered_extent(inode, ordered, 1);
87826df0 2012 btrfs_put_ordered_extent(ordered);
4a096752
CM		 2013 goto again;
2014 }
247e743c 2015
7cf5b976 2016 ret = btrfs_delalloc_reserve_space(inode, page_start,
09cbfeaf 2017 PAGE_SIZE);
87826df0
JM		 2018 if (ret) {
2019 mapping_set_error(page->mapping, ret);
2020 end_extent_writepage(page, ret, page_start, page_end);
2021 ClearPageChecked(page);
2022 goto out;
2023 }
2024
2ac55d41 2025 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
247e743c 2026 ClearPageChecked(page);
87826df0 2027 set_page_dirty(page);
247e743c 2028out:
2ac55d41
JB		 2029 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2030 &cached_state, GFP_NOFS);
247e743c
CM		 2031out_page:
2032 unlock_page(page);
09cbfeaf 2033 put_page(page);
b897abec 2034 kfree(fixup);
247e743c
CM		 2035}
2036
2037/*
2038 * There are a few paths in the higher layers of the kernel that directly
2039 * set the page dirty bit without asking the filesystem if it is a
2040 * good idea. This causes problems because we want to make sure COW
2041 * properly happens and the data=ordered rules are followed.
2042 *
c8b97818 2043 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 2044 * hasn't been properly setup for IO. We kick off an async process
2045 * to fix it up. The async helper will wait for ordered extents, set
2046 * the delalloc bit and make it safe to write the page.
2047 */
b2950863 2048static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
247e743c
CM		 2049{
2050 struct inode *inode = page->mapping->host;
2051 struct btrfs_writepage_fixup *fixup;
2052 struct btrfs_root *root = BTRFS_I(inode)->root;
247e743c 2053
8b62b72b
CM		 2054 /* this page is properly in the ordered list */
2055 if (TestClearPagePrivate2(page))
247e743c
CM		 2056 return 0;
2057
2058 if (PageChecked(page))
2059 return -EAGAIN;
2060
2061 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2062 if (!fixup)
2063 return -EAGAIN;
f421950f 2064
247e743c 2065 SetPageChecked(page);
09cbfeaf 2066 get_page(page);
9e0af237
LB		 2067 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2068 btrfs_writepage_fixup_worker, NULL, NULL);
247e743c 2069 fixup->page = page;
dc6e3209 2070 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
87826df0 2071 return -EBUSY;
247e743c
CM		 2072}
2073
d899e052
YZ		 2074static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2075 struct inode *inode, u64 file_pos,
2076 u64 disk_bytenr, u64 disk_num_bytes,
2077 u64 num_bytes, u64 ram_bytes,
2078 u8 compression, u8 encryption,
2079 u16 other_encoding, int extent_type)
2080{
2081 struct btrfs_root *root = BTRFS_I(inode)->root;
2082 struct btrfs_file_extent_item *fi;
2083 struct btrfs_path *path;
2084 struct extent_buffer *leaf;
2085 struct btrfs_key ins;
1acae57b 2086 int extent_inserted = 0;
d899e052
YZ		 2087 int ret;
2088
2089 path = btrfs_alloc_path();
d8926bb3
MF		 2090 if (!path)
2091 return -ENOMEM;
d899e052 2092
a1ed835e
CM		 2093 /*
2094 * we may be replacing one extent in the tree with another.
2095 * The new extent is pinned in the extent map, and we don't want
2096 * to drop it from the cache until it is completely in the btree.
2097 *
2098 * So, tell btrfs_drop_extents to leave this extent in the cache.
2099 * the caller is expected to unpin it and allow it to be merged
2100 * with the others.
2101 */
1acae57b
FDBM		 2102 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2103 file_pos + num_bytes, NULL, 0,
2104 1, sizeof(*fi), &extent_inserted);
79787eaa
JM		 2105 if (ret)
2106 goto out;
d899e052 2107
1acae57b
FDBM		 2108 if (!extent_inserted) {
2109 ins.objectid = btrfs_ino(inode);
2110 ins.offset = file_pos;
2111 ins.type = BTRFS_EXTENT_DATA_KEY;
2112
2113 path->leave_spinning = 1;
2114 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2115 sizeof(*fi));
2116 if (ret)
2117 goto out;
2118 }
d899e052
YZ		 2119 leaf = path->nodes[0];
2120 fi = btrfs_item_ptr(leaf, path->slots[0],
2121 struct btrfs_file_extent_item);
2122 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2123 btrfs_set_file_extent_type(leaf, fi, extent_type);
2124 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2125 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2126 btrfs_set_file_extent_offset(leaf, fi, 0);
2127 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2128 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2129 btrfs_set_file_extent_compression(leaf, fi, compression);
2130 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2131 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
b9473439 2132
d899e052 2133 btrfs_mark_buffer_dirty(leaf);
ce195332 2134 btrfs_release_path(path);
d899e052
YZ		 2135
2136 inode_add_bytes(inode, num_bytes);
d899e052
YZ		 2137
2138 ins.objectid = disk_bytenr;
2139 ins.offset = disk_num_bytes;
2140 ins.type = BTRFS_EXTENT_ITEM_KEY;
5d4f98a2
YZ		 2141 ret = btrfs_alloc_reserved_file_extent(trans, root,
2142 root->root_key.objectid,
5846a3c2
QW		 2143 btrfs_ino(inode), file_pos,
2144 ram_bytes, &ins);
297d750b 2145 /*
5846a3c2
QW		 2146 * Release the reserved range from inode dirty range map, as it is
2147 * already moved into delayed_ref_head
297d750b
QW		 2148 */
2149 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
79787eaa 2150out:
d899e052 2151 btrfs_free_path(path);
b9473439 2152
79787eaa 2153 return ret;
d899e052
YZ		 2154}
2155
38c227d8
LB		 2156/* snapshot-aware defrag */
2157struct sa_defrag_extent_backref {
2158 struct rb_node node;
2159 struct old_sa_defrag_extent *old;
2160 u64 root_id;
2161 u64 inum;
2162 u64 file_pos;
2163 u64 extent_offset;
2164 u64 num_bytes;
2165 u64 generation;
2166};
2167
2168struct old_sa_defrag_extent {
2169 struct list_head list;
2170 struct new_sa_defrag_extent *new;
2171
2172 u64 extent_offset;
2173 u64 bytenr;
2174 u64 offset;
2175 u64 len;
2176 int count;
2177};
2178
2179struct new_sa_defrag_extent {
2180 struct rb_root root;
2181 struct list_head head;
2182 struct btrfs_path *path;
2183 struct inode *inode;
2184 u64 file_pos;
2185 u64 len;
2186 u64 bytenr;
2187 u64 disk_len;
2188 u8 compress_type;
2189};
2190
2191static int backref_comp(struct sa_defrag_extent_backref *b1,
2192 struct sa_defrag_extent_backref *b2)
2193{
2194 if (b1->root_id < b2->root_id)
2195 return -1;
2196 else if (b1->root_id > b2->root_id)
2197 return 1;
2198
2199 if (b1->inum < b2->inum)
2200 return -1;
2201 else if (b1->inum > b2->inum)
2202 return 1;
2203
2204 if (b1->file_pos < b2->file_pos)
2205 return -1;
2206 else if (b1->file_pos > b2->file_pos)
2207 return 1;
2208
2209 /*
2210 * [------------------------------] ===> (a range of space)
2211 * |<--->| |<---->| =============> (fs/file tree A)
2212 * |<---------------------------->| ===> (fs/file tree B)
2213 *
2214 * A range of space can refer to two file extents in one tree while
2215 * refer to only one file extent in another tree.
2216 *
2217 * So we may process a disk offset more than one time(two extents in A)
2218 * and locate at the same extent(one extent in B), then insert two same
2219 * backrefs(both refer to the extent in B).
2220 */
2221 return 0;
2222}
2223
2224static void backref_insert(struct rb_root *root,
2225 struct sa_defrag_extent_backref *backref)
2226{
2227 struct rb_node **p = &root->rb_node;
2228 struct rb_node *parent = NULL;
2229 struct sa_defrag_extent_backref *entry;
2230 int ret;
2231
2232 while (*p) {
2233 parent = *p;
2234 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2235
2236 ret = backref_comp(backref, entry);
2237 if (ret < 0)
2238 p = &(*p)->rb_left;
2239 else
2240 p = &(*p)->rb_right;
2241 }
2242
2243 rb_link_node(&backref->node, parent, p);
2244 rb_insert_color(&backref->node, root);
2245}
2246
2247/*
2248 * Note the backref might has changed, and in this case we just return 0.
2249 */
2250static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2251 void *ctx)
2252{
2253 struct btrfs_file_extent_item *extent;
2254 struct btrfs_fs_info *fs_info;
2255 struct old_sa_defrag_extent *old = ctx;
2256 struct new_sa_defrag_extent *new = old->new;
2257 struct btrfs_path *path = new->path;
2258 struct btrfs_key key;
2259 struct btrfs_root *root;
2260 struct sa_defrag_extent_backref *backref;
2261 struct extent_buffer *leaf;
2262 struct inode *inode = new->inode;
2263 int slot;
2264 int ret;
2265 u64 extent_offset;
2266 u64 num_bytes;
2267
2268 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2269 inum == btrfs_ino(inode))
2270 return 0;
2271
2272 key.objectid = root_id;
2273 key.type = BTRFS_ROOT_ITEM_KEY;
2274 key.offset = (u64)-1;
2275
2276 fs_info = BTRFS_I(inode)->root->fs_info;
2277 root = btrfs_read_fs_root_no_name(fs_info, &key);
2278 if (IS_ERR(root)) {
2279 if (PTR_ERR(root) == -ENOENT)
2280 return 0;
2281 WARN_ON(1);
2282 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2283 inum, offset, root_id);
2284 return PTR_ERR(root);
2285 }
2286
2287 key.objectid = inum;
2288 key.type = BTRFS_EXTENT_DATA_KEY;
2289 if (offset > (u64)-1 << 32)
2290 key.offset = 0;
2291 else
2292 key.offset = offset;
2293
2294 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
fae7f21c 2295 if (WARN_ON(ret < 0))
38c227d8 2296 return ret;
50f1319c 2297 ret = 0;
38c227d8
LB		 2298
2299 while (1) {
2300 cond_resched();
2301
2302 leaf = path->nodes[0];
2303 slot = path->slots[0];
2304
2305 if (slot >= btrfs_header_nritems(leaf)) {
2306 ret = btrfs_next_leaf(root, path);
2307 if (ret < 0) {
2308 goto out;
2309 } else if (ret > 0) {
2310 ret = 0;
2311 goto out;
2312 }
2313 continue;
2314 }
2315
2316 path->slots[0]++;
2317
2318 btrfs_item_key_to_cpu(leaf, &key, slot);
2319
2320 if (key.objectid > inum)
2321 goto out;
2322
2323 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2324 continue;
2325
2326 extent = btrfs_item_ptr(leaf, slot,
2327 struct btrfs_file_extent_item);
2328
2329 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2330 continue;
2331
e68afa49
LB		 2332 /*
2333 * 'offset' refers to the exact key.offset,
2334 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2335 * (key.offset - extent_offset).
2336 */
2337 if (key.offset != offset)
38c227d8
LB		 2338 continue;
2339
e68afa49 2340 extent_offset = btrfs_file_extent_offset(leaf, extent);
38c227d8 2341 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
e68afa49 2342
38c227d8
LB		 2343 if (extent_offset >= old->extent_offset + old->offset +
2344 old->len || extent_offset + num_bytes <=
2345 old->extent_offset + old->offset)
2346 continue;
38c227d8
LB		 2347 break;
2348 }
2349
2350 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2351 if (!backref) {
2352 ret = -ENOENT;
2353 goto out;
2354 }
2355
2356 backref->root_id = root_id;
2357 backref->inum = inum;
e68afa49 2358 backref->file_pos = offset;
38c227d8
LB		 2359 backref->num_bytes = num_bytes;
2360 backref->extent_offset = extent_offset;
2361 backref->generation = btrfs_file_extent_generation(leaf, extent);
2362 backref->old = old;
2363 backref_insert(&new->root, backref);
2364 old->count++;
2365out:
2366 btrfs_release_path(path);
2367 WARN_ON(ret);
2368 return ret;
2369}
2370
2371static noinline bool record_extent_backrefs(struct btrfs_path *path,
2372 struct new_sa_defrag_extent *new)
2373{
2374 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2375 struct old_sa_defrag_extent *old, *tmp;
2376 int ret;
2377
2378 new->path = path;
2379
2380 list_for_each_entry_safe(old, tmp, &new->head, list) {
e68afa49
LB		 2381 ret = iterate_inodes_from_logical(old->bytenr +
2382 old->extent_offset, fs_info,
38c227d8
LB		 2383 path, record_one_backref,
2384 old);
4724b106
JB		 2385 if (ret < 0 && ret != -ENOENT)
2386 return false;
38c227d8
LB		 2387
2388 /* no backref to be processed for this extent */
2389 if (!old->count) {
2390 list_del(&old->list);
2391 kfree(old);
2392 }
2393 }
2394
2395 if (list_empty(&new->head))
2396 return false;
2397
2398 return true;
2399}
2400
2401static int relink_is_mergable(struct extent_buffer *leaf,
2402 struct btrfs_file_extent_item *fi,
116e0024 2403 struct new_sa_defrag_extent *new)
38c227d8 2404{
116e0024 2405 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
38c227d8
LB		 2406 return 0;
2407
2408 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2409 return 0;
2410
116e0024
LB		 2411 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2412 return 0;
2413
2414 if (btrfs_file_extent_encryption(leaf, fi) ||
38c227d8
LB		 2415 btrfs_file_extent_other_encoding(leaf, fi))
2416 return 0;
2417
2418 return 1;
2419}
2420
2421/*
2422 * Note the backref might has changed, and in this case we just return 0.
2423 */
2424static noinline int relink_extent_backref(struct btrfs_path *path,
2425 struct sa_defrag_extent_backref *prev,
2426 struct sa_defrag_extent_backref *backref)
2427{
2428 struct btrfs_file_extent_item *extent;
2429 struct btrfs_file_extent_item *item;
2430 struct btrfs_ordered_extent *ordered;
2431 struct btrfs_trans_handle *trans;
2432 struct btrfs_fs_info *fs_info;
2433 struct btrfs_root *root;
2434 struct btrfs_key key;
2435 struct extent_buffer *leaf;
2436 struct old_sa_defrag_extent *old = backref->old;
2437 struct new_sa_defrag_extent *new = old->new;
2438 struct inode *src_inode = new->inode;
2439 struct inode *inode;
2440 struct extent_state *cached = NULL;
2441 int ret = 0;
2442 u64 start;
2443 u64 len;
2444 u64 lock_start;
2445 u64 lock_end;
2446 bool merge = false;
2447 int index;
2448
2449 if (prev && prev->root_id == backref->root_id &&
2450 prev->inum == backref->inum &&
2451 prev->file_pos + prev->num_bytes == backref->file_pos)
2452 merge = true;
2453
2454 /* step 1: get root */
2455 key.objectid = backref->root_id;
2456 key.type = BTRFS_ROOT_ITEM_KEY;
2457 key.offset = (u64)-1;
2458
2459 fs_info = BTRFS_I(src_inode)->root->fs_info;
2460 index = srcu_read_lock(&fs_info->subvol_srcu);
2461
2462 root = btrfs_read_fs_root_no_name(fs_info, &key);
2463 if (IS_ERR(root)) {
2464 srcu_read_unlock(&fs_info->subvol_srcu, index);
2465 if (PTR_ERR(root) == -ENOENT)
2466 return 0;
2467 return PTR_ERR(root);
2468 }
38c227d8 2469
bcbba5e6
WS		 2470 if (btrfs_root_readonly(root)) {
2471 srcu_read_unlock(&fs_info->subvol_srcu, index);
2472 return 0;
2473 }
2474
38c227d8
LB		 2475 /* step 2: get inode */
2476 key.objectid = backref->inum;
2477 key.type = BTRFS_INODE_ITEM_KEY;
2478 key.offset = 0;
2479
2480 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2481 if (IS_ERR(inode)) {
2482 srcu_read_unlock(&fs_info->subvol_srcu, index);
2483 return 0;
2484 }
2485
2486 srcu_read_unlock(&fs_info->subvol_srcu, index);
2487
2488 /* step 3: relink backref */
2489 lock_start = backref->file_pos;
2490 lock_end = backref->file_pos + backref->num_bytes - 1;
2491 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
ff13db41 2492 &cached);
38c227d8
LB		 2493
2494 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2495 if (ordered) {
2496 btrfs_put_ordered_extent(ordered);
2497 goto out_unlock;
2498 }
2499
2500 trans = btrfs_join_transaction(root);
2501 if (IS_ERR(trans)) {
2502 ret = PTR_ERR(trans);
2503 goto out_unlock;
2504 }
2505
2506 key.objectid = backref->inum;
2507 key.type = BTRFS_EXTENT_DATA_KEY;
2508 key.offset = backref->file_pos;
2509
2510 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2511 if (ret < 0) {
2512 goto out_free_path;
2513 } else if (ret > 0) {
2514 ret = 0;
2515 goto out_free_path;
2516 }
2517
2518 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2519 struct btrfs_file_extent_item);
2520
2521 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2522 backref->generation)
2523 goto out_free_path;
2524
2525 btrfs_release_path(path);
2526
2527 start = backref->file_pos;
2528 if (backref->extent_offset < old->extent_offset + old->offset)
2529 start += old->extent_offset + old->offset -
2530 backref->extent_offset;
2531
2532 len = min(backref->extent_offset + backref->num_bytes,
2533 old->extent_offset + old->offset + old->len);
2534 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2535
2536 ret = btrfs_drop_extents(trans, root, inode, start,
2537 start + len, 1);
2538 if (ret)
2539 goto out_free_path;
2540again:
2541 key.objectid = btrfs_ino(inode);
2542 key.type = BTRFS_EXTENT_DATA_KEY;
2543 key.offset = start;
2544
a09a0a70 2545 path->leave_spinning = 1;
38c227d8
LB		 2546 if (merge) {
2547 struct btrfs_file_extent_item *fi;
2548 u64 extent_len;
2549 struct btrfs_key found_key;
2550
3c9665df 2551 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
38c227d8
LB		 2552 if (ret < 0)
2553 goto out_free_path;
2554
2555 path->slots[0]--;
2556 leaf = path->nodes[0];
2557 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2558
2559 fi = btrfs_item_ptr(leaf, path->slots[0],
2560 struct btrfs_file_extent_item);
2561 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2562
116e0024
LB		 2563 if (extent_len + found_key.offset == start &&
2564 relink_is_mergable(leaf, fi, new)) {
38c227d8
LB		 2565 btrfs_set_file_extent_num_bytes(leaf, fi,
2566 extent_len + len);
2567 btrfs_mark_buffer_dirty(leaf);
2568 inode_add_bytes(inode, len);
2569
2570 ret = 1;
2571 goto out_free_path;
2572 } else {
2573 merge = false;
2574 btrfs_release_path(path);
2575 goto again;
2576 }
2577 }
2578
2579 ret = btrfs_insert_empty_item(trans, root, path, &key,
2580 sizeof(*extent));
2581 if (ret) {
2582 btrfs_abort_transaction(trans, root, ret);
2583 goto out_free_path;
2584 }
2585
2586 leaf = path->nodes[0];
2587 item = btrfs_item_ptr(leaf, path->slots[0],
2588 struct btrfs_file_extent_item);
2589 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2590 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2591 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2592 btrfs_set_file_extent_num_bytes(leaf, item, len);
2593 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2594 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2595 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2596 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2597 btrfs_set_file_extent_encryption(leaf, item, 0);
2598 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2599
2600 btrfs_mark_buffer_dirty(leaf);
2601 inode_add_bytes(inode, len);
a09a0a70 2602 btrfs_release_path(path);
38c227d8
LB		 2603
2604 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2605 new->disk_len, 0,
2606 backref->root_id, backref->inum,
b06c4bf5 2607 new->file_pos); /* start - extent_offset */
38c227d8
LB		 2608 if (ret) {
2609 btrfs_abort_transaction(trans, root, ret);
2610 goto out_free_path;
2611 }
2612
2613 ret = 1;
2614out_free_path:
2615 btrfs_release_path(path);
a09a0a70 2616 path->leave_spinning = 0;
38c227d8
LB		 2617 btrfs_end_transaction(trans, root);
2618out_unlock:
2619 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2620 &cached, GFP_NOFS);
2621 iput(inode);
2622 return ret;
2623}
2624
6f519564
LB		 2625static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2626{
2627 struct old_sa_defrag_extent *old, *tmp;
2628
2629 if (!new)
2630 return;
2631
2632 list_for_each_entry_safe(old, tmp, &new->head, list) {
6f519564
LB		 2633 kfree(old);
2634 }
2635 kfree(new);
2636}
2637
38c227d8
LB		 2638static void relink_file_extents(struct new_sa_defrag_extent *new)
2639{
2640 struct btrfs_path *path;
38c227d8
LB		 2641 struct sa_defrag_extent_backref *backref;
2642 struct sa_defrag_extent_backref *prev = NULL;
2643 struct inode *inode;
2644 struct btrfs_root *root;
2645 struct rb_node *node;
2646 int ret;
2647
2648 inode = new->inode;
2649 root = BTRFS_I(inode)->root;
2650
2651 path = btrfs_alloc_path();
2652 if (!path)
2653 return;
2654
2655 if (!record_extent_backrefs(path, new)) {
2656 btrfs_free_path(path);
2657 goto out;
2658 }
2659 btrfs_release_path(path);
2660
2661 while (1) {
2662 node = rb_first(&new->root);
2663 if (!node)
2664 break;
2665 rb_erase(node, &new->root);
2666
2667 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2668
2669 ret = relink_extent_backref(path, prev, backref);
2670 WARN_ON(ret < 0);
2671
2672 kfree(prev);
2673
2674 if (ret == 1)
2675 prev = backref;
2676 else
2677 prev = NULL;
2678 cond_resched();
2679 }
2680 kfree(prev);
2681
2682 btrfs_free_path(path);
38c227d8 2683out:
6f519564
LB		 2684 free_sa_defrag_extent(new);
2685
38c227d8
LB		 2686 atomic_dec(&root->fs_info->defrag_running);
2687 wake_up(&root->fs_info->transaction_wait);
38c227d8
LB		 2688}
2689
2690static struct new_sa_defrag_extent *
2691record_old_file_extents(struct inode *inode,
2692 struct btrfs_ordered_extent *ordered)
2693{
2694 struct btrfs_root *root = BTRFS_I(inode)->root;
2695 struct btrfs_path *path;
2696 struct btrfs_key key;
6f519564 2697 struct old_sa_defrag_extent *old;
38c227d8
LB		 2698 struct new_sa_defrag_extent *new;
2699 int ret;
2700
2701 new = kmalloc(sizeof(*new), GFP_NOFS);
2702 if (!new)
2703 return NULL;
2704
2705 new->inode = inode;
2706 new->file_pos = ordered->file_offset;
2707 new->len = ordered->len;
2708 new->bytenr = ordered->start;
2709 new->disk_len = ordered->disk_len;
2710 new->compress_type = ordered->compress_type;
2711 new->root = RB_ROOT;
2712 INIT_LIST_HEAD(&new->head);
2713
2714 path = btrfs_alloc_path();
2715 if (!path)
2716 goto out_kfree;
2717
2718 key.objectid = btrfs_ino(inode);
2719 key.type = BTRFS_EXTENT_DATA_KEY;
2720 key.offset = new->file_pos;
2721
2722 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2723 if (ret < 0)
2724 goto out_free_path;
2725 if (ret > 0 && path->slots[0] > 0)
2726 path->slots[0]--;
2727
2728 /* find out all the old extents for the file range */
2729 while (1) {
2730 struct btrfs_file_extent_item *extent;
2731 struct extent_buffer *l;
2732 int slot;
2733 u64 num_bytes;
2734 u64 offset;
2735 u64 end;
2736 u64 disk_bytenr;
2737 u64 extent_offset;
2738
2739 l = path->nodes[0];
2740 slot = path->slots[0];
2741
2742 if (slot >= btrfs_header_nritems(l)) {
2743 ret = btrfs_next_leaf(root, path);
2744 if (ret < 0)
6f519564 2745 goto out_free_path;
38c227d8
LB		 2746 else if (ret > 0)
2747 break;
2748 continue;
2749 }
2750
2751 btrfs_item_key_to_cpu(l, &key, slot);
2752
2753 if (key.objectid != btrfs_ino(inode))
2754 break;
2755 if (key.type != BTRFS_EXTENT_DATA_KEY)
2756 break;
2757 if (key.offset >= new->file_pos + new->len)
2758 break;
2759
2760 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2761
2762 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2763 if (key.offset + num_bytes < new->file_pos)
2764 goto next;
2765
2766 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2767 if (!disk_bytenr)
2768 goto next;
2769
2770 extent_offset = btrfs_file_extent_offset(l, extent);
2771
2772 old = kmalloc(sizeof(*old), GFP_NOFS);
2773 if (!old)
6f519564 2774 goto out_free_path;
38c227d8
LB		 2775
2776 offset = max(new->file_pos, key.offset);
2777 end = min(new->file_pos + new->len, key.offset + num_bytes);
2778
2779 old->bytenr = disk_bytenr;
2780 old->extent_offset = extent_offset;
2781 old->offset = offset - key.offset;
2782 old->len = end - offset;
2783 old->new = new;
2784 old->count = 0;
2785 list_add_tail(&old->list, &new->head);
2786next:
2787 path->slots[0]++;
2788 cond_resched();
2789 }
2790
2791 btrfs_free_path(path);
2792 atomic_inc(&root->fs_info->defrag_running);
2793
2794 return new;
2795
38c227d8
LB		 2796out_free_path:
2797 btrfs_free_path(path);
2798out_kfree:
6f519564 2799 free_sa_defrag_extent(new);
38c227d8
LB		 2800 return NULL;
2801}
2802
e570fd27
MX		 2803static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2804 u64 start, u64 len)
2805{
2806 struct btrfs_block_group_cache *cache;
2807
2808 cache = btrfs_lookup_block_group(root->fs_info, start);
2809 ASSERT(cache);
2810
2811 spin_lock(&cache->lock);
2812 cache->delalloc_bytes -= len;
2813 spin_unlock(&cache->lock);
2814
2815 btrfs_put_block_group(cache);
2816}
2817
d352ac68
CM		 2818/* as ordered data IO finishes, this gets called so we can finish
2819 * an ordered extent if the range of bytes in the file it covers are
2820 * fully written.
2821 */
5fd02043 2822static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
e6dcd2dc 2823{
5fd02043 2824 struct inode *inode = ordered_extent->inode;
e6dcd2dc 2825 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 2826 struct btrfs_trans_handle *trans = NULL;
e6dcd2dc 2827 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2ac55d41 2828 struct extent_state *cached_state = NULL;
38c227d8 2829 struct new_sa_defrag_extent *new = NULL;
261507a0 2830 int compress_type = 0;
77cef2ec
JB		 2831 int ret = 0;
2832 u64 logical_len = ordered_extent->len;
82d5902d 2833 bool nolock;
77cef2ec 2834 bool truncated = false;
e6dcd2dc 2835
83eea1f1 2836 nolock = btrfs_is_free_space_inode(inode);
0cb59c99 2837
5fd02043
JB		 2838 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2839 ret = -EIO;
2840 goto out;
2841 }
2842
f612496b
MX		 2843 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2844 ordered_extent->file_offset +
2845 ordered_extent->len - 1);
2846
77cef2ec
JB		 2847 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2848 truncated = true;
2849 logical_len = ordered_extent->truncated_len;
2850 /* Truncated the entire extent, don't bother adding */
2851 if (!logical_len)
2852 goto out;
2853 }
2854
c2167754 2855 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
79787eaa 2856 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
94ed938a
QW		 2857
2858 /*
2859 * For mwrite(mmap + memset to write) case, we still reserve
2860 * space for NOCOW range.
2861 * As NOCOW won't cause a new delayed ref, just free the space
2862 */
2863 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2864 ordered_extent->len);
6c760c07
JB		 2865 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2866 if (nolock)
2867 trans = btrfs_join_transaction_nolock(root);
2868 else
2869 trans = btrfs_join_transaction(root);
2870 if (IS_ERR(trans)) {
2871 ret = PTR_ERR(trans);
2872 trans = NULL;
2873 goto out;
c2167754 2874 }
6c760c07
JB		 2875 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2876 ret = btrfs_update_inode_fallback(trans, root, inode);
2877 if (ret) /* -ENOMEM or corruption */
2878 btrfs_abort_transaction(trans, root, ret);
c2167754
YZ		 2879 goto out;
2880 }
e6dcd2dc 2881
2ac55d41
JB		 2882 lock_extent_bits(io_tree, ordered_extent->file_offset,
2883 ordered_extent->file_offset + ordered_extent->len - 1,
ff13db41 2884 &cached_state);
e6dcd2dc 2885
38c227d8
LB		 2886 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2887 ordered_extent->file_offset + ordered_extent->len - 1,
2888 EXTENT_DEFRAG, 1, cached_state);
2889 if (ret) {
2890 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
8101c8db 2891 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
38c227d8
LB		 2892 /* the inode is shared */
2893 new = record_old_file_extents(inode, ordered_extent);
2894
2895 clear_extent_bit(io_tree, ordered_extent->file_offset,
2896 ordered_extent->file_offset + ordered_extent->len - 1,
2897 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2898 }
2899
0cb59c99 2900 if (nolock)
7a7eaa40 2901 trans = btrfs_join_transaction_nolock(root);
0cb59c99 2902 else
7a7eaa40 2903 trans = btrfs_join_transaction(root);
79787eaa
JM		 2904 if (IS_ERR(trans)) {
2905 ret = PTR_ERR(trans);
2906 trans = NULL;
2907 goto out_unlock;
2908 }
a79b7d4b 2909
0ca1f7ce 2910 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
c2167754 2911
c8b97818 2912 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
261507a0 2913 compress_type = ordered_extent->compress_type;
d899e052 2914 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
261507a0 2915 BUG_ON(compress_type);
920bbbfb 2916 ret = btrfs_mark_extent_written(trans, inode,
d899e052
YZ		 2917 ordered_extent->file_offset,
2918 ordered_extent->file_offset +
77cef2ec 2919 logical_len);
d899e052 2920 } else {
0af3d00b 2921 BUG_ON(root == root->fs_info->tree_root);
d899e052
YZ		 2922 ret = insert_reserved_file_extent(trans, inode,
2923 ordered_extent->file_offset,
2924 ordered_extent->start,
2925 ordered_extent->disk_len,
77cef2ec 2926 logical_len, logical_len,
261507a0 2927 compress_type, 0, 0,
d899e052 2928 BTRFS_FILE_EXTENT_REG);
e570fd27
MX		 2929 if (!ret)
2930 btrfs_release_delalloc_bytes(root,
2931 ordered_extent->start,
2932 ordered_extent->disk_len);
d899e052 2933 }
5dc562c5
JB		 2934 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2935 ordered_extent->file_offset, ordered_extent->len,
2936 trans->transid);
79787eaa
JM		 2937 if (ret < 0) {
2938 btrfs_abort_transaction(trans, root, ret);
5fd02043 2939 goto out_unlock;
79787eaa 2940 }
2ac55d41 2941
e6dcd2dc
CM		 2942 add_pending_csums(trans, inode, ordered_extent->file_offset,
2943 &ordered_extent->list);
2944
6c760c07
JB		 2945 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2946 ret = btrfs_update_inode_fallback(trans, root, inode);
2947 if (ret) { /* -ENOMEM or corruption */
2948 btrfs_abort_transaction(trans, root, ret);
2949 goto out_unlock;
1ef30be1
JB		 2950 }
2951 ret = 0;
5fd02043
JB		 2952out_unlock:
2953 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2954 ordered_extent->file_offset +
2955 ordered_extent->len - 1, &cached_state, GFP_NOFS);
c2167754 2956out:
5b0e95bf 2957 if (root != root->fs_info->tree_root)
0cb59c99 2958 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
a698d075
MX		 2959 if (trans)
2960 btrfs_end_transaction(trans, root);
0cb59c99 2961
77cef2ec
JB		 2962 if (ret || truncated) {
2963 u64 start, end;
2964
2965 if (truncated)
2966 start = ordered_extent->file_offset + logical_len;
2967 else
2968 start = ordered_extent->file_offset;
2969 end = ordered_extent->file_offset + ordered_extent->len - 1;
2970 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2971
2972 /* Drop the cache for the part of the extent we didn't write. */
2973 btrfs_drop_extent_cache(inode, start, end, 0);
5fd02043 2974
0bec9ef5
JB		 2975 /*
2976 * If the ordered extent had an IOERR or something else went
2977 * wrong we need to return the space for this ordered extent
77cef2ec
JB		 2978 * back to the allocator. We only free the extent in the
2979 * truncated case if we didn't write out the extent at all.
0bec9ef5 2980 */
77cef2ec
JB		 2981 if ((ret || !logical_len) &&
2982 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
0bec9ef5
JB		 2983 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2984 btrfs_free_reserved_extent(root, ordered_extent->start,
e570fd27 2985 ordered_extent->disk_len, 1);
0bec9ef5
JB		 2986 }
2987
2988
5fd02043 2989 /*
8bad3c02
LB		 2990 * This needs to be done to make sure anybody waiting knows we are done
2991 * updating everything for this ordered extent.
5fd02043
JB		 2992 */
2993 btrfs_remove_ordered_extent(inode, ordered_extent);
2994
38c227d8 2995 /* for snapshot-aware defrag */
6f519564
LB		 2996 if (new) {
2997 if (ret) {
2998 free_sa_defrag_extent(new);
2999 atomic_dec(&root->fs_info->defrag_running);
3000 } else {
3001 relink_file_extents(new);
3002 }
3003 }
38c227d8 3004
e6dcd2dc
CM		 3005 /* once for us */
3006 btrfs_put_ordered_extent(ordered_extent);
3007 /* once for the tree */
3008 btrfs_put_ordered_extent(ordered_extent);
3009
5fd02043
JB		 3010 return ret;
3011}
3012
3013static void finish_ordered_fn(struct btrfs_work *work)
3014{
3015 struct btrfs_ordered_extent *ordered_extent;
3016 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3017 btrfs_finish_ordered_io(ordered_extent);
e6dcd2dc
CM		 3018}
3019
b2950863 3020static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM		 3021 struct extent_state *state, int uptodate)
3022{
5fd02043
JB		 3023 struct inode *inode = page->mapping->host;
3024 struct btrfs_root *root = BTRFS_I(inode)->root;
3025 struct btrfs_ordered_extent *ordered_extent = NULL;
9e0af237
LB		 3026 struct btrfs_workqueue *wq;
3027 btrfs_work_func_t func;
5fd02043 3028
1abe9b8a 3029 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3030
8b62b72b 3031 ClearPagePrivate2(page);
5fd02043
JB		 3032 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3033 end - start + 1, uptodate))
3034 return 0;
3035
9e0af237
LB		 3036 if (btrfs_is_free_space_inode(inode)) {
3037 wq = root->fs_info->endio_freespace_worker;
3038 func = btrfs_freespace_write_helper;
3039 } else {
3040 wq = root->fs_info->endio_write_workers;
3041 func = btrfs_endio_write_helper;
3042 }
5fd02043 3043
9e0af237
LB		 3044 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3045 NULL);
3046 btrfs_queue_work(wq, &ordered_extent->work);
5fd02043
JB		 3047
3048 return 0;
211f90e6
CM		 3049}
3050
dc380aea
MX		 3051static int __readpage_endio_check(struct inode *inode,
3052 struct btrfs_io_bio *io_bio,
3053 int icsum, struct page *page,
3054 int pgoff, u64 start, size_t len)
3055{
3056 char *kaddr;
3057 u32 csum_expected;
3058 u32 csum = ~(u32)0;
dc380aea
MX		 3059
3060 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3061
3062 kaddr = kmap_atomic(page);
3063 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3064 btrfs_csum_final(csum, (char *)&csum);
3065 if (csum != csum_expected)
3066 goto zeroit;
3067
3068 kunmap_atomic(kaddr);
3069 return 0;
3070zeroit:
94647322
DS		 3071 btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3072 "csum failed ino %llu off %llu csum %u expected csum %u",
dc380aea
MX		 3073 btrfs_ino(inode), start, csum, csum_expected);
3074 memset(kaddr + pgoff, 1, len);
3075 flush_dcache_page(page);
3076 kunmap_atomic(kaddr);
3077 if (csum_expected == 0)
3078 return 0;
3079 return -EIO;
3080}
3081
d352ac68
CM		 3082/*
3083 * when reads are done, we need to check csums to verify the data is correct
4a54c8c1
JS		 3084 * if there's a match, we allow the bio to finish. If not, the code in
3085 * extent_io.c will try to find good copies for us.
d352ac68 3086 */
facc8a22
MX		 3087static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3088 u64 phy_offset, struct page *page,
3089 u64 start, u64 end, int mirror)
07157aac 3090{
4eee4fa4 3091 size_t offset = start - page_offset(page);
07157aac 3092 struct inode *inode = page->mapping->host;
d1310b2e 3093 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
ff79f819 3094 struct btrfs_root *root = BTRFS_I(inode)->root;
d1310b2e 3095
d20f7043
CM		 3096 if (PageChecked(page)) {
3097 ClearPageChecked(page);
dc380aea 3098 return 0;
d20f7043 3099 }
6cbff00f
CH		 3100
3101 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
dc380aea 3102 return 0;
17d217fe
YZ		 3103
3104 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
9655d298 3105 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
91166212 3106 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
b6cda9bc 3107 return 0;
17d217fe 3108 }
d20f7043 3109
facc8a22 3110 phy_offset >>= inode->i_sb->s_blocksize_bits;
dc380aea
MX		 3111 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3112 start, (size_t)(end - start + 1));
07157aac 3113}
b888db2b 3114
24bbcf04
YZ		 3115void btrfs_add_delayed_iput(struct inode *inode)
3116{
3117 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
8089fe62 3118 struct btrfs_inode *binode = BTRFS_I(inode);
24bbcf04
YZ		 3119
3120 if (atomic_add_unless(&inode->i_count, -1, 1))
3121 return;
3122
24bbcf04 3123 spin_lock(&fs_info->delayed_iput_lock);
8089fe62
DS		 3124 if (binode->delayed_iput_count == 0) {
3125 ASSERT(list_empty(&binode->delayed_iput));
3126 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3127 } else {
3128 binode->delayed_iput_count++;
3129 }
24bbcf04
YZ		 3130 spin_unlock(&fs_info->delayed_iput_lock);
3131}
3132
3133void btrfs_run_delayed_iputs(struct btrfs_root *root)
3134{
24bbcf04 3135 struct btrfs_fs_info *fs_info = root->fs_info;
24bbcf04 3136
24bbcf04 3137 spin_lock(&fs_info->delayed_iput_lock);
8089fe62
DS		 3138 while (!list_empty(&fs_info->delayed_iputs)) {
3139 struct btrfs_inode *inode;
3140
3141 inode = list_first_entry(&fs_info->delayed_iputs,
3142 struct btrfs_inode, delayed_iput);
3143 if (inode->delayed_iput_count) {
3144 inode->delayed_iput_count--;
3145 list_move_tail(&inode->delayed_iput,
3146 &fs_info->delayed_iputs);
3147 } else {
3148 list_del_init(&inode->delayed_iput);
3149 }
3150 spin_unlock(&fs_info->delayed_iput_lock);
3151 iput(&inode->vfs_inode);
3152 spin_lock(&fs_info->delayed_iput_lock);
24bbcf04 3153 }
8089fe62 3154 spin_unlock(&fs_info->delayed_iput_lock);
24bbcf04
YZ		 3155}
3156
d68fc57b 3157/*
42b2aa86 3158 * This is called in transaction commit time. If there are no orphan
d68fc57b
YZ		 3159 * files in the subvolume, it removes orphan item and frees block_rsv
3160 * structure.
3161 */
3162void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3163 struct btrfs_root *root)
3164{
90290e19 3165 struct btrfs_block_rsv *block_rsv;
d68fc57b
YZ		 3166 int ret;
3167
8a35d95f 3168 if (atomic_read(&root->orphan_inodes) ||
d68fc57b
YZ		 3169 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3170 return;
3171
90290e19 3172 spin_lock(&root->orphan_lock);
8a35d95f 3173 if (atomic_read(&root->orphan_inodes)) {
90290e19
JB		 3174 spin_unlock(&root->orphan_lock);
3175 return;
3176 }
3177
3178 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3179 spin_unlock(&root->orphan_lock);
3180 return;
3181 }
3182
3183 block_rsv = root->orphan_block_rsv;
3184 root->orphan_block_rsv = NULL;
3185 spin_unlock(&root->orphan_lock);
3186
27cdeb70 3187 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
d68fc57b
YZ		 3188 btrfs_root_refs(&root->root_item) > 0) {
3189 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3190 root->root_key.objectid);
4ef31a45
JB		 3191 if (ret)
3192 btrfs_abort_transaction(trans, root, ret);
3193 else
27cdeb70
MX		 3194 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3195 &root->state);
d68fc57b
YZ		 3196 }
3197
90290e19
JB		 3198 if (block_rsv) {
3199 WARN_ON(block_rsv->size > 0);
3200 btrfs_free_block_rsv(root, block_rsv);
d68fc57b
YZ		 3201 }
3202}
3203
7b128766
JB		 3204/*
3205 * This creates an orphan entry for the given inode in case something goes
3206 * wrong in the middle of an unlink/truncate.
d68fc57b
YZ		 3207 *
3208 * NOTE: caller of this function should reserve 5 units of metadata for
3209 * this function.
7b128766
JB		 3210 */
3211int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3212{
3213 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 3214 struct btrfs_block_rsv *block_rsv = NULL;
3215 int reserve = 0;
3216 int insert = 0;
3217 int ret;
7b128766 3218
d68fc57b 3219 if (!root->orphan_block_rsv) {
66d8f3dd 3220 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
b532402e
TI		 3221 if (!block_rsv)
3222 return -ENOMEM;
d68fc57b 3223 }
7b128766 3224
d68fc57b
YZ		 3225 spin_lock(&root->orphan_lock);
3226 if (!root->orphan_block_rsv) {
3227 root->orphan_block_rsv = block_rsv;
3228 } else if (block_rsv) {
3229 btrfs_free_block_rsv(root, block_rsv);
3230 block_rsv = NULL;
7b128766 3231 }
7b128766 3232
8a35d95f
JB		 3233 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3234 &BTRFS_I(inode)->runtime_flags)) {
d68fc57b
YZ		 3235#if 0
3236 /*
3237 * For proper ENOSPC handling, we should do orphan
3238 * cleanup when mounting. But this introduces backward
3239 * compatibility issue.
3240 */
3241 if (!xchg(&root->orphan_item_inserted, 1))
3242 insert = 2;
3243 else
3244 insert = 1;
3245#endif
3246 insert = 1;
321f0e70 3247 atomic_inc(&root->orphan_inodes);
7b128766
JB		 3248 }
3249
72ac3c0d
JB		 3250 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3251 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3252 reserve = 1;
d68fc57b 3253 spin_unlock(&root->orphan_lock);
7b128766 3254
d68fc57b
YZ		 3255 /* grab metadata reservation from transaction handle */
3256 if (reserve) {
3257 ret = btrfs_orphan_reserve_metadata(trans, inode);
79787eaa 3258 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
d68fc57b 3259 }
7b128766 3260
d68fc57b
YZ		 3261 /* insert an orphan item to track this unlinked/truncated file */
3262 if (insert >= 1) {
33345d01 3263 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
4ef31a45 3264 if (ret) {
703c88e0 3265 atomic_dec(&root->orphan_inodes);
4ef31a45
JB		 3266 if (reserve) {
3267 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3268 &BTRFS_I(inode)->runtime_flags);
3269 btrfs_orphan_release_metadata(inode);
3270 }
3271 if (ret != -EEXIST) {
e8e7cff6
JB		 3272 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3273 &BTRFS_I(inode)->runtime_flags);
4ef31a45
JB		 3274 btrfs_abort_transaction(trans, root, ret);
3275 return ret;
3276 }
79787eaa
JM		 3277 }
3278 ret = 0;
d68fc57b
YZ		 3279 }
3280
3281 /* insert an orphan item to track subvolume contains orphan files */
3282 if (insert >= 2) {
3283 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3284 root->root_key.objectid);
79787eaa
JM		 3285 if (ret && ret != -EEXIST) {
3286 btrfs_abort_transaction(trans, root, ret);
3287 return ret;
3288 }
d68fc57b
YZ		 3289 }
3290 return 0;
7b128766
JB		 3291}
3292
3293/*
3294 * We have done the truncate/delete so we can go ahead and remove the orphan
3295 * item for this particular inode.
3296 */
48a3b636
ES		 3297static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3298 struct inode *inode)
7b128766
JB		 3299{
3300 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 3301 int delete_item = 0;
3302 int release_rsv = 0;
7b128766
JB		 3303 int ret = 0;
3304
d68fc57b 3305 spin_lock(&root->orphan_lock);
8a35d95f
JB		 3306 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3307 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3308 delete_item = 1;
7b128766 3309
72ac3c0d
JB		 3310 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3311 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3312 release_rsv = 1;
d68fc57b 3313 spin_unlock(&root->orphan_lock);
7b128766 3314
703c88e0 3315 if (delete_item) {
8a35d95f 3316 atomic_dec(&root->orphan_inodes);
703c88e0
FDBM		 3317 if (trans)
3318 ret = btrfs_del_orphan_item(trans, root,
3319 btrfs_ino(inode));
8a35d95f 3320 }
7b128766 3321
703c88e0
FDBM		 3322 if (release_rsv)
3323 btrfs_orphan_release_metadata(inode);
3324
4ef31a45 3325 return ret;
7b128766
JB		 3326}
3327
3328/*
3329 * this cleans up any orphans that may be left on the list from the last use
3330 * of this root.
3331 */
66b4ffd1 3332int btrfs_orphan_cleanup(struct btrfs_root *root)
7b128766
JB		 3333{
3334 struct btrfs_path *path;
3335 struct extent_buffer *leaf;
7b128766
JB		 3336 struct btrfs_key key, found_key;
3337 struct btrfs_trans_handle *trans;
3338 struct inode *inode;
8f6d7f4f 3339 u64 last_objectid = 0;
7b128766
JB		 3340 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3341
d68fc57b 3342 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
66b4ffd1 3343 return 0;
c71bf099
YZ		 3344
3345 path = btrfs_alloc_path();
66b4ffd1
JB		 3346 if (!path) {
3347 ret = -ENOMEM;
3348 goto out;
3349 }
e4058b54 3350 path->reada = READA_BACK;
7b128766
JB		 3351
3352 key.objectid = BTRFS_ORPHAN_OBJECTID;
962a298f 3353 key.type = BTRFS_ORPHAN_ITEM_KEY;
7b128766
JB		 3354 key.offset = (u64)-1;
3355
7b128766
JB		 3356 while (1) {
3357 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
66b4ffd1
JB		 3358 if (ret < 0)
3359 goto out;
7b128766
JB		 3360
3361 /*
3362 * if ret == 0 means we found what we were searching for, which
25985edc 3363 * is weird, but possible, so only screw with path if we didn't
7b128766
JB		 3364 * find the key and see if we have stuff that matches
3365 */
3366 if (ret > 0) {
66b4ffd1 3367 ret = 0;
7b128766
JB		 3368 if (path->slots[0] == 0)
3369 break;
3370 path->slots[0]--;
3371 }
3372
3373 /* pull out the item */
3374 leaf = path->nodes[0];
7b128766
JB		 3375 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3376
3377 /* make sure the item matches what we want */
3378 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3379 break;
962a298f 3380 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
7b128766
JB		 3381 break;
3382
3383 /* release the path since we're done with it */
b3b4aa74 3384 btrfs_release_path(path);
7b128766
JB		 3385
3386 /*
3387 * this is where we are basically btrfs_lookup, without the
3388 * crossing root thing. we store the inode number in the
3389 * offset of the orphan item.
3390 */
8f6d7f4f
JB		 3391
3392 if (found_key.offset == last_objectid) {
c2cf52eb
SK		 3393 btrfs_err(root->fs_info,
3394 "Error removing orphan entry, stopping orphan cleanup");
8f6d7f4f
JB		 3395 ret = -EINVAL;
3396 goto out;
3397 }
3398
3399 last_objectid = found_key.offset;
3400
5d4f98a2
YZ		 3401 found_key.objectid = found_key.offset;
3402 found_key.type = BTRFS_INODE_ITEM_KEY;
3403 found_key.offset = 0;
73f73415 3404 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
8c6ffba0 3405 ret = PTR_ERR_OR_ZERO(inode);
a8c9e576 3406 if (ret && ret != -ESTALE)
66b4ffd1 3407 goto out;
7b128766 3408
f8e9e0b0
AJ		 3409 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3410 struct btrfs_root *dead_root;
3411 struct btrfs_fs_info *fs_info = root->fs_info;
3412 int is_dead_root = 0;
3413
3414 /*
3415 * this is an orphan in the tree root. Currently these
3416 * could come from 2 sources:
3417 * a) a snapshot deletion in progress
3418 * b) a free space cache inode
3419 * We need to distinguish those two, as the snapshot
3420 * orphan must not get deleted.
3421 * find_dead_roots already ran before us, so if this
3422 * is a snapshot deletion, we should find the root
3423 * in the dead_roots list
3424 */
3425 spin_lock(&fs_info->trans_lock);
3426 list_for_each_entry(dead_root, &fs_info->dead_roots,
3427 root_list) {
3428 if (dead_root->root_key.objectid ==
3429 found_key.objectid) {
3430 is_dead_root = 1;
3431 break;
3432 }
3433 }
3434 spin_unlock(&fs_info->trans_lock);
3435 if (is_dead_root) {
3436 /* prevent this orphan from being found again */
3437 key.offset = found_key.objectid - 1;
3438 continue;
3439 }
3440 }
7b128766 3441 /*
a8c9e576
JB		 3442 * Inode is already gone but the orphan item is still there,
3443 * kill the orphan item.
7b128766 3444 */
a8c9e576
JB		 3445 if (ret == -ESTALE) {
3446 trans = btrfs_start_transaction(root, 1);
66b4ffd1
JB		 3447 if (IS_ERR(trans)) {
3448 ret = PTR_ERR(trans);
3449 goto out;
3450 }
c2cf52eb
SK		 3451 btrfs_debug(root->fs_info, "auto deleting %Lu",
3452 found_key.objectid);
a8c9e576
JB		 3453 ret = btrfs_del_orphan_item(trans, root,
3454 found_key.objectid);
5b21f2ed 3455 btrfs_end_transaction(trans, root);
4ef31a45
JB		 3456 if (ret)
3457 goto out;
7b128766
JB		 3458 continue;
3459 }
3460
a8c9e576
JB		 3461 /*
3462 * add this inode to the orphan list so btrfs_orphan_del does
3463 * the proper thing when we hit it
3464 */
8a35d95f
JB		 3465 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3466 &BTRFS_I(inode)->runtime_flags);
925396ec 3467 atomic_inc(&root->orphan_inodes);
a8c9e576 3468
7b128766
JB		 3469 /* if we have links, this was a truncate, lets do that */
3470 if (inode->i_nlink) {
fae7f21c 3471 if (WARN_ON(!S_ISREG(inode->i_mode))) {
a41ad394
JB		 3472 iput(inode);
3473 continue;
3474 }
7b128766 3475 nr_truncate++;
f3fe820c
JB		 3476
3477 /* 1 for the orphan item deletion. */
3478 trans = btrfs_start_transaction(root, 1);
3479 if (IS_ERR(trans)) {
c69b26b0 3480 iput(inode);
f3fe820c
JB		 3481 ret = PTR_ERR(trans);
3482 goto out;
3483 }
3484 ret = btrfs_orphan_add(trans, inode);
3485 btrfs_end_transaction(trans, root);
c69b26b0
JB		 3486 if (ret) {
3487 iput(inode);
f3fe820c 3488 goto out;
c69b26b0 3489 }
f3fe820c 3490
66b4ffd1 3491 ret = btrfs_truncate(inode);
4a7d0f68
JB		 3492 if (ret)
3493 btrfs_orphan_del(NULL, inode);
7b128766
JB		 3494 } else {
3495 nr_unlink++;
3496 }
3497
3498 /* this will do delete_inode and everything for us */
3499 iput(inode);
66b4ffd1
JB		 3500 if (ret)
3501 goto out;
7b128766 3502 }
3254c876
MX		 3503 /* release the path since we're done with it */
3504 btrfs_release_path(path);
3505
d68fc57b
YZ		 3506 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3507
3508 if (root->orphan_block_rsv)
3509 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3510 (u64)-1);
3511
27cdeb70
MX		 3512 if (root->orphan_block_rsv ||
3513 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
7a7eaa40 3514 trans = btrfs_join_transaction(root);
66b4ffd1
JB		 3515 if (!IS_ERR(trans))
3516 btrfs_end_transaction(trans, root);
d68fc57b 3517 }
7b128766
JB		 3518
3519 if (nr_unlink)
4884b476 3520 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
7b128766 3521 if (nr_truncate)
4884b476 3522 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
66b4ffd1
JB		 3523
3524out:
3525 if (ret)
68b663d1 3526 btrfs_err(root->fs_info,
c2cf52eb 3527 "could not do orphan cleanup %d", ret);
66b4ffd1
JB		 3528 btrfs_free_path(path);
3529 return ret;
7b128766
JB		 3530}
3531
46a53cca
CM		 3532/*
3533 * very simple check to peek ahead in the leaf looking for xattrs. If we
3534 * don't find any xattrs, we know there can't be any acls.
3535 *
3536 * slot is the slot the inode is in, objectid is the objectid of the inode
3537 */
3538static noinline int acls_after_inode_item(struct extent_buffer *leaf,
63541927
FDBM		 3539 int slot, u64 objectid,
3540 int *first_xattr_slot)
46a53cca
CM		 3541{
3542 u32 nritems = btrfs_header_nritems(leaf);
3543 struct btrfs_key found_key;
f23b5a59
JB		 3544 static u64 xattr_access = 0;
3545 static u64 xattr_default = 0;
46a53cca
CM		 3546 int scanned = 0;
3547
f23b5a59 3548 if (!xattr_access) {
97d79299
AG		 3549 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3550 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3551 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3552 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
f23b5a59
JB		 3553 }
3554
46a53cca 3555 slot++;
63541927 3556 *first_xattr_slot = -1;
46a53cca
CM		 3557 while (slot < nritems) {
3558 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3559
3560 /* we found a different objectid, there must not be acls */
3561 if (found_key.objectid != objectid)
3562 return 0;
3563
3564 /* we found an xattr, assume we've got an acl */
f23b5a59 3565 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
63541927
FDBM		 3566 if (*first_xattr_slot == -1)
3567 *first_xattr_slot = slot;
f23b5a59
JB		 3568 if (found_key.offset == xattr_access ||
3569 found_key.offset == xattr_default)
3570 return 1;
3571 }
46a53cca
CM		 3572
3573 /*
3574 * we found a key greater than an xattr key, there can't
3575 * be any acls later on
3576 */
3577 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3578 return 0;
3579
3580 slot++;
3581 scanned++;
3582
3583 /*
3584 * it goes inode, inode backrefs, xattrs, extents,
3585 * so if there are a ton of hard links to an inode there can
3586 * be a lot of backrefs. Don't waste time searching too hard,
3587 * this is just an optimization
3588 */
3589 if (scanned >= 8)
3590 break;
3591 }
3592 /* we hit the end of the leaf before we found an xattr or
3593 * something larger than an xattr. We have to assume the inode
3594 * has acls
3595 */
63541927
FDBM		 3596 if (*first_xattr_slot == -1)
3597 *first_xattr_slot = slot;
46a53cca
CM		 3598 return 1;
3599}
3600
d352ac68
CM		 3601/*
3602 * read an inode from the btree into the in-memory inode
3603 */
5d4f98a2 3604static void btrfs_read_locked_inode(struct inode *inode)
39279cc3
CM		 3605{
3606 struct btrfs_path *path;
5f39d397 3607 struct extent_buffer *leaf;
39279cc3
CM		 3608 struct btrfs_inode_item *inode_item;
3609 struct btrfs_root *root = BTRFS_I(inode)->root;
3610 struct btrfs_key location;
67de1176 3611 unsigned long ptr;
46a53cca 3612 int maybe_acls;
618e21d5 3613 u32 rdev;
39279cc3 3614 int ret;
2f7e33d4 3615 bool filled = false;
63541927 3616 int first_xattr_slot;
2f7e33d4
MX		 3617
3618 ret = btrfs_fill_inode(inode, &rdev);
3619 if (!ret)
3620 filled = true;
39279cc3
CM		 3621
3622 path = btrfs_alloc_path();
1748f843
MF		 3623 if (!path)
3624 goto make_bad;
3625
39279cc3 3626 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 3627
39279cc3 3628 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 3629 if (ret)
39279cc3 3630 goto make_bad;
39279cc3 3631
5f39d397 3632 leaf = path->nodes[0];
2f7e33d4
MX		 3633
3634 if (filled)
67de1176 3635 goto cache_index;
2f7e33d4 3636
5f39d397
CM		 3637 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3638 struct btrfs_inode_item);
5f39d397 3639 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
bfe86848 3640 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2f2f43d3
EB		 3641 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3642 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
dbe674a9 3643 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397 3644
a937b979
DS		 3645 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3646 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
5f39d397 3647
a937b979
DS		 3648 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3649 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
5f39d397 3650
a937b979
DS		 3651 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3652 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
5f39d397 3653
9cc97d64 3654 BTRFS_I(inode)->i_otime.tv_sec =
3655 btrfs_timespec_sec(leaf, &inode_item->otime);
3656 BTRFS_I(inode)->i_otime.tv_nsec =
3657 btrfs_timespec_nsec(leaf, &inode_item->otime);
5f39d397 3658
a76a3cd4 3659 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 3660 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
5dc562c5
JB		 3661 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3662
6e17d30b
YD		 3663 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3664 inode->i_generation = BTRFS_I(inode)->generation;
3665 inode->i_rdev = 0;
3666 rdev = btrfs_inode_rdev(leaf, inode_item);
3667
3668 BTRFS_I(inode)->index_cnt = (u64)-1;
3669 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3670
3671cache_index:
5dc562c5
JB		 3672 /*
3673 * If we were modified in the current generation and evicted from memory
3674 * and then re-read we need to do a full sync since we don't have any
3675 * idea about which extents were modified before we were evicted from
3676 * cache.
6e17d30b
YD		 3677 *
3678 * This is required for both inode re-read from disk and delayed inode
3679 * in delayed_nodes_tree.
5dc562c5
JB		 3680 */
3681 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3682 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3683 &BTRFS_I(inode)->runtime_flags);
3684
bde6c242
FM		 3685 /*
3686 * We don't persist the id of the transaction where an unlink operation
3687 * against the inode was last made. So here we assume the inode might
3688 * have been evicted, and therefore the exact value of last_unlink_trans
3689 * lost, and set it to last_trans to avoid metadata inconsistencies
3690 * between the inode and its parent if the inode is fsync'ed and the log
3691 * replayed. For example, in the scenario:
3692 *
3693 * touch mydir/foo
3694 * ln mydir/foo mydir/bar
3695 * sync
3696 * unlink mydir/bar
3697 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3698 * xfs_io -c fsync mydir/foo
3699 * <power failure>
3700 * mount fs, triggers fsync log replay
3701 *
3702 * We must make sure that when we fsync our inode foo we also log its
3703 * parent inode, otherwise after log replay the parent still has the
3704 * dentry with the "bar" name but our inode foo has a link count of 1
3705 * and doesn't have an inode ref with the name "bar" anymore.
3706 *
3707 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3708 * but it guarantees correctness at the expense of ocassional full
3709 * transaction commits on fsync if our inode is a directory, or if our
3710 * inode is not a directory, logging its parent unnecessarily.
3711 */
3712 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3713
67de1176
MX		 3714 path->slots[0]++;
3715 if (inode->i_nlink != 1 ||
3716 path->slots[0] >= btrfs_header_nritems(leaf))
3717 goto cache_acl;
3718
3719 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3720 if (location.objectid != btrfs_ino(inode))
3721 goto cache_acl;
3722
3723 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3724 if (location.type == BTRFS_INODE_REF_KEY) {
3725 struct btrfs_inode_ref *ref;
3726
3727 ref = (struct btrfs_inode_ref *)ptr;
3728 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3729 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3730 struct btrfs_inode_extref *extref;
3731
3732 extref = (struct btrfs_inode_extref *)ptr;
3733 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3734 extref);
3735 }
2f7e33d4 3736cache_acl:
46a53cca
CM		 3737 /*
3738 * try to precache a NULL acl entry for files that don't have
3739 * any xattrs or acls
3740 */
33345d01 3741 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
63541927
FDBM		 3742 btrfs_ino(inode), &first_xattr_slot);
3743 if (first_xattr_slot != -1) {
3744 path->slots[0] = first_xattr_slot;
3745 ret = btrfs_load_inode_props(inode, path);
3746 if (ret)
3747 btrfs_err(root->fs_info,
351fd353 3748 "error loading props for ino %llu (root %llu): %d",
63541927
FDBM		 3749 btrfs_ino(inode),
3750 root->root_key.objectid, ret);
3751 }
3752 btrfs_free_path(path);
3753
72c04902
AV		 3754 if (!maybe_acls)
3755 cache_no_acl(inode);
46a53cca 3756
39279cc3 3757 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 3758 case S_IFREG:
3759 inode->i_mapping->a_ops = &btrfs_aops;
d1310b2e 3760 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 3761 inode->i_fop = &btrfs_file_operations;
3762 inode->i_op = &btrfs_file_inode_operations;
3763 break;
3764 case S_IFDIR:
3765 inode->i_fop = &btrfs_dir_file_operations;
3766 if (root == root->fs_info->tree_root)
3767 inode->i_op = &btrfs_dir_ro_inode_operations;
3768 else
3769 inode->i_op = &btrfs_dir_inode_operations;
3770 break;
3771 case S_IFLNK:
3772 inode->i_op = &btrfs_symlink_inode_operations;
21fc61c7 3773 inode_nohighmem(inode);
39279cc3
CM		 3774 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3775 break;
618e21d5 3776 default:
0279b4cd 3777 inode->i_op = &btrfs_special_inode_operations;
618e21d5
JB		 3778 init_special_inode(inode, inode->i_mode, rdev);
3779 break;
39279cc3 3780 }
6cbff00f
CH		 3781
3782 btrfs_update_iflags(inode);
39279cc3
CM		 3783 return;
3784
3785make_bad:
39279cc3 3786 btrfs_free_path(path);
39279cc3
CM		 3787 make_bad_inode(inode);
3788}
3789
d352ac68
CM		 3790/*
3791 * given a leaf and an inode, copy the inode fields into the leaf
3792 */
e02119d5
CM		 3793static void fill_inode_item(struct btrfs_trans_handle *trans,
3794 struct extent_buffer *leaf,
5f39d397 3795 struct btrfs_inode_item *item,
39279cc3
CM		 3796 struct inode *inode)
3797{
51fab693
LB		 3798 struct btrfs_map_token token;
3799
3800 btrfs_init_map_token(&token);
5f39d397 3801
51fab693
LB		 3802 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3803 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3804 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3805 &token);
3806 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3807 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
5f39d397 3808
a937b979 3809 btrfs_set_token_timespec_sec(leaf, &item->atime,
51fab693 3810 inode->i_atime.tv_sec, &token);
a937b979 3811 btrfs_set_token_timespec_nsec(leaf, &item->atime,
51fab693 3812 inode->i_atime.tv_nsec, &token);
5f39d397 3813
a937b979 3814 btrfs_set_token_timespec_sec(leaf, &item->mtime,
51fab693 3815 inode->i_mtime.tv_sec, &token);
a937b979 3816 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
51fab693 3817 inode->i_mtime.tv_nsec, &token);
5f39d397 3818
a937b979 3819 btrfs_set_token_timespec_sec(leaf, &item->ctime,
51fab693 3820 inode->i_ctime.tv_sec, &token);
a937b979 3821 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
51fab693 3822 inode->i_ctime.tv_nsec, &token);
5f39d397 3823
9cc97d64 3824 btrfs_set_token_timespec_sec(leaf, &item->otime,
3825 BTRFS_I(inode)->i_otime.tv_sec, &token);
3826 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3827 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3828
51fab693
LB		 3829 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3830 &token);
3831 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3832 &token);
3833 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3834 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3835 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3836 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3837 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
39279cc3
CM		 3838}
3839
d352ac68
CM		 3840/*
3841 * copy everything in the in-memory inode into the btree.
3842 */
2115133f 3843static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
d397712b 3844 struct btrfs_root *root, struct inode *inode)
39279cc3
CM		 3845{
3846 struct btrfs_inode_item *inode_item;
3847 struct btrfs_path *path;
5f39d397 3848 struct extent_buffer *leaf;
39279cc3
CM		 3849 int ret;
3850
3851 path = btrfs_alloc_path();
16cdcec7
MX		 3852 if (!path)
3853 return -ENOMEM;
3854
b9473439 3855 path->leave_spinning = 1;
16cdcec7
MX		 3856 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3857 1);
39279cc3
CM		 3858 if (ret) {
3859 if (ret > 0)
3860 ret = -ENOENT;
3861 goto failed;
3862 }
3863
5f39d397
CM		 3864 leaf = path->nodes[0];
3865 inode_item = btrfs_item_ptr(leaf, path->slots[0],
16cdcec7 3866 struct btrfs_inode_item);
39279cc3 3867
e02119d5 3868 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 3869 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 3870 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 3871 ret = 0;
3872failed:
39279cc3
CM		 3873 btrfs_free_path(path);
3874 return ret;
3875}
3876
2115133f
CM		 3877/*
3878 * copy everything in the in-memory inode into the btree.
3879 */
3880noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3881 struct btrfs_root *root, struct inode *inode)
3882{
3883 int ret;
3884
3885 /*
3886 * If the inode is a free space inode, we can deadlock during commit
3887 * if we put it into the delayed code.
3888 *
3889 * The data relocation inode should also be directly updated
3890 * without delay
3891 */
83eea1f1 3892 if (!btrfs_is_free_space_inode(inode)
1d52c78a
JB		 3893 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3894 && !root->fs_info->log_root_recovering) {
8ea05e3a
AB		 3895 btrfs_update_root_times(trans, root);
3896
2115133f
CM		 3897 ret = btrfs_delayed_update_inode(trans, root, inode);
3898 if (!ret)
3899 btrfs_set_inode_last_trans(trans, inode);
3900 return ret;
3901 }
3902
3903 return btrfs_update_inode_item(trans, root, inode);
3904}
3905
be6aef60
JB		 3906noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3907 struct btrfs_root *root,
3908 struct inode *inode)
2115133f
CM		 3909{
3910 int ret;
3911
3912 ret = btrfs_update_inode(trans, root, inode);
3913 if (ret == -ENOSPC)
3914 return btrfs_update_inode_item(trans, root, inode);
3915 return ret;
3916}
3917
d352ac68
CM		 3918/*
3919 * unlink helper that gets used here in inode.c and in the tree logging
3920 * recovery code. It remove a link in a directory with a given name, and
3921 * also drops the back refs in the inode to the directory
3922 */
92986796
AV		 3923static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3924 struct btrfs_root *root,
3925 struct inode *dir, struct inode *inode,
3926 const char *name, int name_len)
39279cc3
CM		 3927{
3928 struct btrfs_path *path;
39279cc3 3929 int ret = 0;
5f39d397 3930 struct extent_buffer *leaf;
39279cc3 3931 struct btrfs_dir_item *di;
5f39d397 3932 struct btrfs_key key;
aec7477b 3933 u64 index;
33345d01
LZ		 3934 u64 ino = btrfs_ino(inode);
3935 u64 dir_ino = btrfs_ino(dir);
39279cc3
CM		 3936
3937 path = btrfs_alloc_path();
54aa1f4d
CM		 3938 if (!path) {
3939 ret = -ENOMEM;
554233a6 3940 goto out;
54aa1f4d
CM		 3941 }
3942
b9473439 3943 path->leave_spinning = 1;
33345d01 3944 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
39279cc3
CM		 3945 name, name_len, -1);
3946 if (IS_ERR(di)) {
3947 ret = PTR_ERR(di);
3948 goto err;
3949 }
3950 if (!di) {
3951 ret = -ENOENT;
3952 goto err;
3953 }
5f39d397
CM		 3954 leaf = path->nodes[0];
3955 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 3956 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 3957 if (ret)
3958 goto err;
b3b4aa74 3959 btrfs_release_path(path);
39279cc3 3960
67de1176
MX		 3961 /*
3962 * If we don't have dir index, we have to get it by looking up
3963 * the inode ref, since we get the inode ref, remove it directly,
3964 * it is unnecessary to do delayed deletion.
3965 *
3966 * But if we have dir index, needn't search inode ref to get it.
3967 * Since the inode ref is close to the inode item, it is better
3968 * that we delay to delete it, and just do this deletion when
3969 * we update the inode item.
3970 */
3971 if (BTRFS_I(inode)->dir_index) {
3972 ret = btrfs_delayed_delete_inode_ref(inode);
3973 if (!ret) {
3974 index = BTRFS_I(inode)->dir_index;
3975 goto skip_backref;
3976 }
3977 }
3978
33345d01
LZ		 3979 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3980 dir_ino, &index);
aec7477b 3981 if (ret) {
c2cf52eb
SK		 3982 btrfs_info(root->fs_info,
3983 "failed to delete reference to %.*s, inode %llu parent %llu",
c1c9ff7c 3984 name_len, name, ino, dir_ino);
79787eaa 3985 btrfs_abort_transaction(trans, root, ret);
aec7477b
JB		 3986 goto err;
3987 }
67de1176 3988skip_backref:
16cdcec7 3989 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 3990 if (ret) {
3991 btrfs_abort_transaction(trans, root, ret);
39279cc3 3992 goto err;
79787eaa 3993 }
39279cc3 3994
e02119d5 3995 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
33345d01 3996 inode, dir_ino);
79787eaa
JM		 3997 if (ret != 0 && ret != -ENOENT) {
3998 btrfs_abort_transaction(trans, root, ret);
3999 goto err;
4000 }
e02119d5
CM		 4001
4002 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4003 dir, index);
6418c961
CM		 4004 if (ret == -ENOENT)
4005 ret = 0;
d4e3991b
ZB		 4006 else if (ret)
4007 btrfs_abort_transaction(trans, root, ret);
39279cc3
CM		 4008err:
4009 btrfs_free_path(path);
e02119d5
CM		 4010 if (ret)
4011 goto out;
4012
4013 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95
JB		 4014 inode_inc_iversion(inode);
4015 inode_inc_iversion(dir);
04b285f3
DD		 4016 inode->i_ctime = dir->i_mtime =
4017 dir->i_ctime = current_fs_time(inode->i_sb);
b9959295 4018 ret = btrfs_update_inode(trans, root, dir);
e02119d5 4019out:
39279cc3
CM		 4020 return ret;
4021}
4022
92986796
AV		 4023int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4024 struct btrfs_root *root,
4025 struct inode *dir, struct inode *inode,
4026 const char *name, int name_len)
4027{
4028 int ret;
4029 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4030 if (!ret) {
8b558c5f 4031 drop_nlink(inode);
92986796
AV		 4032 ret = btrfs_update_inode(trans, root, inode);
4033 }
4034 return ret;
4035}
39279cc3 4036
a22285a6
YZ		 4037/*
4038 * helper to start transaction for unlink and rmdir.
4039 *
d52be818
JB		 4040 * unlink and rmdir are special in btrfs, they do not always free space, so
4041 * if we cannot make our reservations the normal way try and see if there is
4042 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4043 * allow the unlink to occur.
a22285a6 4044 */
d52be818 4045static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4df27c4d 4046{
a22285a6 4047 struct btrfs_root *root = BTRFS_I(dir)->root;
4df27c4d 4048
e70bea5f
JB		 4049 /*
4050 * 1 for the possible orphan item
4051 * 1 for the dir item
4052 * 1 for the dir index
4053 * 1 for the inode ref
e70bea5f
JB		 4054 * 1 for the inode
4055 */
8eab77ff 4056 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
a22285a6
YZ		 4057}
4058
4059static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4060{
4061 struct btrfs_root *root = BTRFS_I(dir)->root;
4062 struct btrfs_trans_handle *trans;
2b0143b5 4063 struct inode *inode = d_inode(dentry);
a22285a6 4064 int ret;
a22285a6 4065
d52be818 4066 trans = __unlink_start_trans(dir);
a22285a6
YZ		 4067 if (IS_ERR(trans))
4068 return PTR_ERR(trans);
5f39d397 4069
2b0143b5 4070 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
12fcfd22 4071
2b0143b5 4072 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
e02119d5 4073 dentry->d_name.name, dentry->d_name.len);
b532402e
TI		 4074 if (ret)
4075 goto out;
7b128766 4076
a22285a6 4077 if (inode->i_nlink == 0) {
7b128766 4078 ret = btrfs_orphan_add(trans, inode);
b532402e
TI		 4079 if (ret)
4080 goto out;
a22285a6 4081 }
7b128766 4082
b532402e 4083out:
d52be818 4084 btrfs_end_transaction(trans, root);
b53d3f5d 4085 btrfs_btree_balance_dirty(root);
39279cc3
CM		 4086 return ret;
4087}
4088
4df27c4d
YZ		 4089int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4090 struct btrfs_root *root,
4091 struct inode *dir, u64 objectid,
4092 const char *name, int name_len)
4093{
4094 struct btrfs_path *path;
4095 struct extent_buffer *leaf;
4096 struct btrfs_dir_item *di;
4097 struct btrfs_key key;
4098 u64 index;
4099 int ret;
33345d01 4100 u64 dir_ino = btrfs_ino(dir);
4df27c4d
YZ		 4101
4102 path = btrfs_alloc_path();
4103 if (!path)
4104 return -ENOMEM;
4105
33345d01 4106 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4df27c4d 4107 name, name_len, -1);
79787eaa
JM		 4108 if (IS_ERR_OR_NULL(di)) {
4109 if (!di)
4110 ret = -ENOENT;
4111 else
4112 ret = PTR_ERR(di);
4113 goto out;
4114 }
4df27c4d
YZ		 4115
4116 leaf = path->nodes[0];
4117 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4118 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4119 ret = btrfs_delete_one_dir_name(trans, root, path, di);
79787eaa
JM		 4120 if (ret) {
4121 btrfs_abort_transaction(trans, root, ret);
4122 goto out;
4123 }
b3b4aa74 4124 btrfs_release_path(path);
4df27c4d
YZ		 4125
4126 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4127 objectid, root->root_key.objectid,
33345d01 4128 dir_ino, &index, name, name_len);
4df27c4d 4129 if (ret < 0) {
79787eaa
JM		 4130 if (ret != -ENOENT) {
4131 btrfs_abort_transaction(trans, root, ret);
4132 goto out;
4133 }
33345d01 4134 di = btrfs_search_dir_index_item(root, path, dir_ino,
4df27c4d 4135 name, name_len);
79787eaa
JM		 4136 if (IS_ERR_OR_NULL(di)) {
4137 if (!di)
4138 ret = -ENOENT;
4139 else
4140 ret = PTR_ERR(di);
4141 btrfs_abort_transaction(trans, root, ret);
4142 goto out;
4143 }
4df27c4d
YZ		 4144
4145 leaf = path->nodes[0];
4146 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
b3b4aa74 4147 btrfs_release_path(path);
4df27c4d
YZ		 4148 index = key.offset;
4149 }
945d8962 4150 btrfs_release_path(path);
4df27c4d 4151
16cdcec7 4152 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 4153 if (ret) {
4154 btrfs_abort_transaction(trans, root, ret);
4155 goto out;
4156 }
4df27c4d
YZ		 4157
4158 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95 4159 inode_inc_iversion(dir);
04b285f3 4160 dir->i_mtime = dir->i_ctime = current_fs_time(dir->i_sb);
5a24e84c 4161 ret = btrfs_update_inode_fallback(trans, root, dir);
79787eaa
JM		 4162 if (ret)
4163 btrfs_abort_transaction(trans, root, ret);
4164out:
71d7aed0 4165 btrfs_free_path(path);
79787eaa 4166 return ret;
4df27c4d
YZ		 4167}
4168
39279cc3
CM		 4169static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4170{
2b0143b5 4171 struct inode *inode = d_inode(dentry);
1832a6d5 4172 int err = 0;
39279cc3 4173 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 4174 struct btrfs_trans_handle *trans;
39279cc3 4175
b3ae244e 4176 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
134d4512 4177 return -ENOTEMPTY;
b3ae244e
DS		 4178 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4179 return -EPERM;
134d4512 4180
d52be818 4181 trans = __unlink_start_trans(dir);
a22285a6 4182 if (IS_ERR(trans))
5df6a9f6 4183 return PTR_ERR(trans);
5df6a9f6 4184
33345d01 4185 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4df27c4d
YZ		 4186 err = btrfs_unlink_subvol(trans, root, dir,
4187 BTRFS_I(inode)->location.objectid,
4188 dentry->d_name.name,
4189 dentry->d_name.len);
4190 goto out;
4191 }
4192
7b128766
JB		 4193 err = btrfs_orphan_add(trans, inode);
4194 if (err)
4df27c4d 4195 goto out;
7b128766 4196
39279cc3 4197 /* now the directory is empty */
2b0143b5 4198 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
e02119d5 4199 dentry->d_name.name, dentry->d_name.len);
d397712b 4200 if (!err)
dbe674a9 4201 btrfs_i_size_write(inode, 0);
4df27c4d 4202out:
d52be818 4203 btrfs_end_transaction(trans, root);
b53d3f5d 4204 btrfs_btree_balance_dirty(root);
3954401f 4205
39279cc3
CM		 4206 return err;
4207}
4208
28f75a0e
CM		 4209static int truncate_space_check(struct btrfs_trans_handle *trans,
4210 struct btrfs_root *root,
4211 u64 bytes_deleted)
4212{
4213 int ret;
4214
dc95f7bf
JB		 4215 /*
4216 * This is only used to apply pressure to the enospc system, we don't
4217 * intend to use this reservation at all.
4218 */
28f75a0e 4219 bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
dc95f7bf 4220 bytes_deleted *= root->nodesize;
28f75a0e
CM		 4221 ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4222 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
dc95f7bf
JB		 4223 if (!ret) {
4224 trace_btrfs_space_reservation(root->fs_info, "transaction",
4225 trans->transid,
4226 bytes_deleted, 1);
28f75a0e 4227 trans->bytes_reserved += bytes_deleted;
dc95f7bf 4228 }
28f75a0e
CM		 4229 return ret;
4230
4231}
4232
0305cd5f
FM		 4233static int truncate_inline_extent(struct inode *inode,
4234 struct btrfs_path *path,
4235 struct btrfs_key *found_key,
4236 const u64 item_end,
4237 const u64 new_size)
4238{
4239 struct extent_buffer *leaf = path->nodes[0];
4240 int slot = path->slots[0];
4241 struct btrfs_file_extent_item *fi;
4242 u32 size = (u32)(new_size - found_key->offset);
4243 struct btrfs_root *root = BTRFS_I(inode)->root;
4244
4245 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4246
4247 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4248 loff_t offset = new_size;
09cbfeaf 4249 loff_t page_end = ALIGN(offset, PAGE_SIZE);
0305cd5f
FM		 4250
4251 /*
4252 * Zero out the remaining of the last page of our inline extent,
4253 * instead of directly truncating our inline extent here - that
4254 * would be much more complex (decompressing all the data, then
4255 * compressing the truncated data, which might be bigger than
4256 * the size of the inline extent, resize the extent, etc).
4257 * We release the path because to get the page we might need to
4258 * read the extent item from disk (data not in the page cache).
4259 */
4260 btrfs_release_path(path);
9703fefe
CR		 4261 return btrfs_truncate_block(inode, offset, page_end - offset,
4262 0);
0305cd5f
FM		 4263 }
4264
4265 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4266 size = btrfs_file_extent_calc_inline_size(size);
4267 btrfs_truncate_item(root, path, size, 1);
4268
4269 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4270 inode_sub_bytes(inode, item_end + 1 - new_size);
4271
4272 return 0;
4273}
4274
39279cc3
CM		 4275/*
4276 * this can truncate away extent items, csum items and directory items.
4277 * It starts at a high offset and removes keys until it can't find
d352ac68 4278 * any higher than new_size
39279cc3
CM		 4279 *
4280 * csum items that cross the new i_size are truncated to the new size
4281 * as well.
7b128766
JB		 4282 *
4283 * min_type is the minimum key type to truncate down to. If set to 0, this
4284 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 4285 */
8082510e
YZ		 4286int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4287 struct btrfs_root *root,
4288 struct inode *inode,
4289 u64 new_size, u32 min_type)
39279cc3 4290{
39279cc3 4291 struct btrfs_path *path;
5f39d397 4292 struct extent_buffer *leaf;
39279cc3 4293 struct btrfs_file_extent_item *fi;
8082510e
YZ		 4294 struct btrfs_key key;
4295 struct btrfs_key found_key;
39279cc3 4296 u64 extent_start = 0;
db94535d 4297 u64 extent_num_bytes = 0;
5d4f98a2 4298 u64 extent_offset = 0;
39279cc3 4299 u64 item_end = 0;
c1aa4575 4300 u64 last_size = new_size;
8082510e 4301 u32 found_type = (u8)-1;
39279cc3
CM		 4302 int found_extent;
4303 int del_item;
85e21bac
CM		 4304 int pending_del_nr = 0;
4305 int pending_del_slot = 0;
179e29e4 4306 int extent_type = -1;
8082510e
YZ		 4307 int ret;
4308 int err = 0;
33345d01 4309 u64 ino = btrfs_ino(inode);
28ed1345 4310 u64 bytes_deleted = 0;
1262133b
JB		 4311 bool be_nice = 0;
4312 bool should_throttle = 0;
28f75a0e 4313 bool should_end = 0;
8082510e
YZ		 4314
4315 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
39279cc3 4316
28ed1345
CM		 4317 /*
4318 * for non-free space inodes and ref cows, we want to back off from
4319 * time to time
4320 */
4321 if (!btrfs_is_free_space_inode(inode) &&
4322 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4323 be_nice = 1;
4324
0eb0e19c
MF		 4325 path = btrfs_alloc_path();
4326 if (!path)
4327 return -ENOMEM;
e4058b54 4328 path->reada = READA_BACK;
0eb0e19c 4329
5dc562c5
JB		 4330 /*
4331 * We want to drop from the next block forward in case this new size is
4332 * not block aligned since we will be keeping the last block of the
4333 * extent just the way it is.
4334 */
27cdeb70
MX		 4335 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4336 root == root->fs_info->tree_root)
fda2832f
QW		 4337 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4338 root->sectorsize), (u64)-1, 0);
8082510e 4339
16cdcec7
MX		 4340 /*
4341 * This function is also used to drop the items in the log tree before
4342 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4343 * it is used to drop the loged items. So we shouldn't kill the delayed
4344 * items.
4345 */
4346 if (min_type == 0 && root == BTRFS_I(inode)->root)
4347 btrfs_kill_delayed_inode_items(inode);
4348
33345d01 4349 key.objectid = ino;
39279cc3 4350 key.offset = (u64)-1;
5f39d397
CM		 4351 key.type = (u8)-1;
4352
85e21bac 4353search_again:
28ed1345
CM		 4354 /*
4355 * with a 16K leaf size and 128MB extents, you can actually queue
4356 * up a huge file in a single leaf. Most of the time that
4357 * bytes_deleted is > 0, it will be huge by the time we get here
4358 */
ee22184b 4359 if (be_nice && bytes_deleted > SZ_32M) {
28ed1345
CM		 4360 if (btrfs_should_end_transaction(trans, root)) {
4361 err = -EAGAIN;
4362 goto error;
4363 }
4364 }
4365
4366
b9473439 4367 path->leave_spinning = 1;
85e21bac 4368 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8082510e
YZ		 4369 if (ret < 0) {
4370 err = ret;
4371 goto out;
4372 }
d397712b 4373
85e21bac 4374 if (ret > 0) {
e02119d5
CM		 4375 /* there are no items in the tree for us to truncate, we're
4376 * done
4377 */
8082510e
YZ		 4378 if (path->slots[0] == 0)
4379 goto out;
85e21bac
CM		 4380 path->slots[0]--;
4381 }
4382
d397712b 4383 while (1) {
39279cc3 4384 fi = NULL;
5f39d397
CM		 4385 leaf = path->nodes[0];
4386 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
962a298f 4387 found_type = found_key.type;
39279cc3 4388
33345d01 4389 if (found_key.objectid != ino)
39279cc3 4390 break;
5f39d397 4391
85e21bac 4392 if (found_type < min_type)
39279cc3
CM		 4393 break;
4394
5f39d397 4395 item_end = found_key.offset;
39279cc3 4396 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 4397 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 4398 struct btrfs_file_extent_item);
179e29e4
CM		 4399 extent_type = btrfs_file_extent_type(leaf, fi);
4400 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 4401 item_end +=
db94535d 4402 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 4403 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 4404 item_end += btrfs_file_extent_inline_len(leaf,
514ac8ad 4405 path->slots[0], fi);
39279cc3 4406 }
008630c1 4407 item_end--;
39279cc3 4408 }
8082510e
YZ		 4409 if (found_type > min_type) {
4410 del_item = 1;
4411 } else {
4412 if (item_end < new_size)
b888db2b 4413 break;
8082510e
YZ		 4414 if (found_key.offset >= new_size)
4415 del_item = 1;
4416 else
4417 del_item = 0;
39279cc3 4418 }
39279cc3 4419 found_extent = 0;
39279cc3 4420 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 4421 if (found_type != BTRFS_EXTENT_DATA_KEY)
4422 goto delete;
4423
7f4f6e0a
JB		 4424 if (del_item)
4425 last_size = found_key.offset;
4426 else
4427 last_size = new_size;
4428
179e29e4 4429 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 4430 u64 num_dec;
db94535d 4431 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
f70a9a6b 4432 if (!del_item) {
db94535d
CM		 4433 u64 orig_num_bytes =
4434 btrfs_file_extent_num_bytes(leaf, fi);
fda2832f
QW		 4435 extent_num_bytes = ALIGN(new_size -
4436 found_key.offset,
4437 root->sectorsize);
db94535d
CM		 4438 btrfs_set_file_extent_num_bytes(leaf, fi,
4439 extent_num_bytes);
4440 num_dec = (orig_num_bytes -
9069218d 4441 extent_num_bytes);
27cdeb70
MX		 4442 if (test_bit(BTRFS_ROOT_REF_COWS,
4443 &root->state) &&
4444 extent_start != 0)
a76a3cd4 4445 inode_sub_bytes(inode, num_dec);
5f39d397 4446 btrfs_mark_buffer_dirty(leaf);
39279cc3 4447 } else {
db94535d
CM		 4448 extent_num_bytes =
4449 btrfs_file_extent_disk_num_bytes(leaf,
4450 fi);
5d4f98a2
YZ		 4451 extent_offset = found_key.offset -
4452 btrfs_file_extent_offset(leaf, fi);
4453
39279cc3 4454 /* FIXME blocksize != 4096 */
9069218d 4455 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 4456 if (extent_start != 0) {
4457 found_extent = 1;
27cdeb70
MX		 4458 if (test_bit(BTRFS_ROOT_REF_COWS,
4459 &root->state))
a76a3cd4 4460 inode_sub_bytes(inode, num_dec);
e02119d5 4461 }
39279cc3 4462 }
9069218d 4463 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 4464 /*
4465 * we can't truncate inline items that have had
4466 * special encodings
4467 */
4468 if (!del_item &&
c8b97818
CM		 4469 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4470 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
514ac8ad
CM		 4471
4472 /*
0305cd5f
FM		 4473 * Need to release path in order to truncate a
4474 * compressed extent. So delete any accumulated
4475 * extent items so far.
514ac8ad 4476 */
0305cd5f
FM		 4477 if (btrfs_file_extent_compression(leaf, fi) !=
4478 BTRFS_COMPRESS_NONE && pending_del_nr) {
4479 err = btrfs_del_items(trans, root, path,
4480 pending_del_slot,
4481 pending_del_nr);
4482 if (err) {
4483 btrfs_abort_transaction(trans,
4484 root,
4485 err);
4486 goto error;
4487 }
4488 pending_del_nr = 0;
4489 }
4490
4491 err = truncate_inline_extent(inode, path,
4492 &found_key,
4493 item_end,
4494 new_size);
4495 if (err) {
4496 btrfs_abort_transaction(trans,
4497 root, err);
4498 goto error;
4499 }
27cdeb70
MX		 4500 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4501 &root->state)) {
0305cd5f 4502 inode_sub_bytes(inode, item_end + 1 - new_size);
9069218d 4503 }
39279cc3 4504 }
179e29e4 4505delete:
39279cc3 4506 if (del_item) {
85e21bac
CM		 4507 if (!pending_del_nr) {
4508 /* no pending yet, add ourselves */
4509 pending_del_slot = path->slots[0];
4510 pending_del_nr = 1;
4511 } else if (pending_del_nr &&
4512 path->slots[0] + 1 == pending_del_slot) {
4513 /* hop on the pending chunk */
4514 pending_del_nr++;
4515 pending_del_slot = path->slots[0];
4516 } else {
d397712b 4517 BUG();
85e21bac 4518 }
39279cc3
CM		 4519 } else {
4520 break;
4521 }
28f75a0e
CM		 4522 should_throttle = 0;
4523
27cdeb70
MX		 4524 if (found_extent &&
4525 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4526 root == root->fs_info->tree_root)) {
b9473439 4527 btrfs_set_path_blocking(path);
28ed1345 4528 bytes_deleted += extent_num_bytes;
39279cc3 4529 ret = btrfs_free_extent(trans, root, extent_start,
5d4f98a2
YZ		 4530 extent_num_bytes, 0,
4531 btrfs_header_owner(leaf),
b06c4bf5 4532 ino, extent_offset);
39279cc3 4533 BUG_ON(ret);
1262133b 4534 if (btrfs_should_throttle_delayed_refs(trans, root))
28ed1345
CM		 4535 btrfs_async_run_delayed_refs(root,
4536 trans->delayed_ref_updates * 2, 0);
28f75a0e
CM		 4537 if (be_nice) {
4538 if (truncate_space_check(trans, root,
4539 extent_num_bytes)) {
4540 should_end = 1;
4541 }
4542 if (btrfs_should_throttle_delayed_refs(trans,
4543 root)) {
4544 should_throttle = 1;
4545 }
4546 }
39279cc3 4547 }
85e21bac 4548
8082510e
YZ		 4549 if (found_type == BTRFS_INODE_ITEM_KEY)
4550 break;
4551
4552 if (path->slots[0] == 0 ||
1262133b 4553 path->slots[0] != pending_del_slot ||
28f75a0e 4554 should_throttle || should_end) {
8082510e
YZ		 4555 if (pending_del_nr) {
4556 ret = btrfs_del_items(trans, root, path,
4557 pending_del_slot,
4558 pending_del_nr);
79787eaa
JM		 4559 if (ret) {
4560 btrfs_abort_transaction(trans,
4561 root, ret);
4562 goto error;
4563 }
8082510e
YZ		 4564 pending_del_nr = 0;
4565 }
b3b4aa74 4566 btrfs_release_path(path);
28f75a0e 4567 if (should_throttle) {
1262133b
JB		 4568 unsigned long updates = trans->delayed_ref_updates;
4569 if (updates) {
4570 trans->delayed_ref_updates = 0;
4571 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4572 if (ret && !err)
4573 err = ret;
4574 }
4575 }
28f75a0e
CM		 4576 /*
4577 * if we failed to refill our space rsv, bail out
4578 * and let the transaction restart
4579 */
4580 if (should_end) {
4581 err = -EAGAIN;
4582 goto error;
4583 }
85e21bac 4584 goto search_again;
8082510e
YZ		 4585 } else {
4586 path->slots[0]--;
85e21bac 4587 }
39279cc3 4588 }
8082510e 4589out:
85e21bac
CM		 4590 if (pending_del_nr) {
4591 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4592 pending_del_nr);
79787eaa
JM		 4593 if (ret)
4594 btrfs_abort_transaction(trans, root, ret);
85e21bac 4595 }
79787eaa 4596error:
c1aa4575 4597 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
7f4f6e0a 4598 btrfs_ordered_update_i_size(inode, last_size, NULL);
28ed1345 4599
39279cc3 4600 btrfs_free_path(path);
28ed1345 4601
ee22184b 4602 if (be_nice && bytes_deleted > SZ_32M) {
28ed1345
CM		 4603 unsigned long updates = trans->delayed_ref_updates;
4604 if (updates) {
4605 trans->delayed_ref_updates = 0;
4606 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4607 if (ret && !err)
4608 err = ret;
4609 }
4610 }
8082510e 4611 return err;
39279cc3
CM		 4612}
4613
4614/*
9703fefe 4615 * btrfs_truncate_block - read, zero a chunk and write a block
2aaa6655
JB		 4616 * @inode - inode that we're zeroing
4617 * @from - the offset to start zeroing
4618 * @len - the length to zero, 0 to zero the entire range respective to the
4619 * offset
4620 * @front - zero up to the offset instead of from the offset on
4621 *
9703fefe 4622 * This will find the block for the "from" offset and cow the block and zero the
2aaa6655 4623 * part we want to zero. This is used with truncate and hole punching.
39279cc3 4624 */
9703fefe 4625int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
2aaa6655 4626 int front)
39279cc3 4627{
2aaa6655 4628 struct address_space *mapping = inode->i_mapping;
db94535d 4629 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 4630 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4631 struct btrfs_ordered_extent *ordered;
2ac55d41 4632 struct extent_state *cached_state = NULL;
e6dcd2dc 4633 char *kaddr;
db94535d 4634 u32 blocksize = root->sectorsize;
09cbfeaf 4635 pgoff_t index = from >> PAGE_SHIFT;
9703fefe 4636 unsigned offset = from & (blocksize - 1);
39279cc3 4637 struct page *page;
3b16a4e3 4638 gfp_t mask = btrfs_alloc_write_mask(mapping);
39279cc3 4639 int ret = 0;
9703fefe
CR		 4640 u64 block_start;
4641 u64 block_end;
39279cc3 4642
2aaa6655
JB		 4643 if ((offset & (blocksize - 1)) == 0 &&
4644 (!len || ((len & (blocksize - 1)) == 0)))
39279cc3 4645 goto out;
9703fefe 4646
7cf5b976 4647 ret = btrfs_delalloc_reserve_space(inode,
9703fefe 4648 round_down(from, blocksize), blocksize);
5d5e103a
JB		 4649 if (ret)
4650 goto out;
39279cc3 4651
211c17f5 4652again:
3b16a4e3 4653 page = find_or_create_page(mapping, index, mask);
5d5e103a 4654 if (!page) {
7cf5b976 4655 btrfs_delalloc_release_space(inode,
9703fefe
CR		 4656 round_down(from, blocksize),
4657 blocksize);
ac6a2b36 4658 ret = -ENOMEM;
39279cc3 4659 goto out;
5d5e103a 4660 }
e6dcd2dc 4661
9703fefe
CR		 4662 block_start = round_down(from, blocksize);
4663 block_end = block_start + blocksize - 1;
e6dcd2dc 4664
39279cc3 4665 if (!PageUptodate(page)) {
9ebefb18 4666 ret = btrfs_readpage(NULL, page);
39279cc3 4667 lock_page(page);
211c17f5
CM		 4668 if (page->mapping != mapping) {
4669 unlock_page(page);
09cbfeaf 4670 put_page(page);
211c17f5
CM		 4671 goto again;
4672 }
39279cc3
CM		 4673 if (!PageUptodate(page)) {
4674 ret = -EIO;
89642229 4675 goto out_unlock;
39279cc3
CM		 4676 }
4677 }
211c17f5 4678 wait_on_page_writeback(page);
e6dcd2dc 4679
9703fefe 4680 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
e6dcd2dc
CM		 4681 set_page_extent_mapped(page);
4682
9703fefe 4683 ordered = btrfs_lookup_ordered_extent(inode, block_start);
e6dcd2dc 4684 if (ordered) {
9703fefe 4685 unlock_extent_cached(io_tree, block_start, block_end,
2ac55d41 4686 &cached_state, GFP_NOFS);
e6dcd2dc 4687 unlock_page(page);
09cbfeaf 4688 put_page(page);
eb84ae03 4689 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 4690 btrfs_put_ordered_extent(ordered);
4691 goto again;
4692 }
4693
9703fefe 4694 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
9e8a4a8b
LB		 4695 EXTENT_DIRTY | EXTENT_DELALLOC |
4696 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 4697 0, 0, &cached_state, GFP_NOFS);
5d5e103a 4698
9703fefe 4699 ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
2ac55d41 4700 &cached_state);
9ed74f2d 4701 if (ret) {
9703fefe 4702 unlock_extent_cached(io_tree, block_start, block_end,
2ac55d41 4703 &cached_state, GFP_NOFS);
9ed74f2d
JB		 4704 goto out_unlock;
4705 }
4706
9703fefe 4707 if (offset != blocksize) {
2aaa6655 4708 if (!len)
9703fefe 4709 len = blocksize - offset;
e6dcd2dc 4710 kaddr = kmap(page);
2aaa6655 4711 if (front)
9703fefe
CR		 4712 memset(kaddr + (block_start - page_offset(page)),
4713 0, offset);
2aaa6655 4714 else
9703fefe
CR		 4715 memset(kaddr + (block_start - page_offset(page)) + offset,
4716 0, len);
e6dcd2dc
CM		 4717 flush_dcache_page(page);
4718 kunmap(page);
4719 }
247e743c 4720 ClearPageChecked(page);
e6dcd2dc 4721 set_page_dirty(page);
9703fefe 4722 unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
2ac55d41 4723 GFP_NOFS);
39279cc3 4724
89642229 4725out_unlock:
5d5e103a 4726 if (ret)
9703fefe
CR		 4727 btrfs_delalloc_release_space(inode, block_start,
4728 blocksize);
39279cc3 4729 unlock_page(page);
09cbfeaf 4730 put_page(page);
39279cc3
CM		 4731out:
4732 return ret;
4733}
4734
16e7549f
JB		 4735static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4736 u64 offset, u64 len)
4737{
4738 struct btrfs_trans_handle *trans;
4739 int ret;
4740
4741 /*
4742 * Still need to make sure the inode looks like it's been updated so
4743 * that any holes get logged if we fsync.
4744 */
4745 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4746 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4747 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4748 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4749 return 0;
4750 }
4751
4752 /*
4753 * 1 - for the one we're dropping
4754 * 1 - for the one we're adding
4755 * 1 - for updating the inode.
4756 */
4757 trans = btrfs_start_transaction(root, 3);
4758 if (IS_ERR(trans))
4759 return PTR_ERR(trans);
4760
4761 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4762 if (ret) {
4763 btrfs_abort_transaction(trans, root, ret);
4764 btrfs_end_transaction(trans, root);
4765 return ret;
4766 }
4767
4768 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4769 0, 0, len, 0, len, 0, 0, 0);
4770 if (ret)
4771 btrfs_abort_transaction(trans, root, ret);
4772 else
4773 btrfs_update_inode(trans, root, inode);
4774 btrfs_end_transaction(trans, root);
4775 return ret;
4776}
4777
695a0d0d
JB		 4778/*
4779 * This function puts in dummy file extents for the area we're creating a hole
4780 * for. So if we are truncating this file to a larger size we need to insert
4781 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4782 * the range between oldsize and size
4783 */
a41ad394 4784int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
39279cc3 4785{
9036c102
YZ		 4786 struct btrfs_root *root = BTRFS_I(inode)->root;
4787 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a22285a6 4788 struct extent_map *em = NULL;
2ac55d41 4789 struct extent_state *cached_state = NULL;
5dc562c5 4790 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
fda2832f
QW		 4791 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4792 u64 block_end = ALIGN(size, root->sectorsize);
9036c102
YZ		 4793 u64 last_byte;
4794 u64 cur_offset;
4795 u64 hole_size;
9ed74f2d 4796 int err = 0;
39279cc3 4797
a71754fc 4798 /*
9703fefe
CR		 4799 * If our size started in the middle of a block we need to zero out the
4800 * rest of the block before we expand the i_size, otherwise we could
a71754fc
JB		 4801 * expose stale data.
4802 */
9703fefe 4803 err = btrfs_truncate_block(inode, oldsize, 0, 0);
a71754fc
JB		 4804 if (err)
4805 return err;
4806
9036c102
YZ		 4807 if (size <= hole_start)
4808 return 0;
4809
9036c102
YZ		 4810 while (1) {
4811 struct btrfs_ordered_extent *ordered;
fa7c1494 4812
ff13db41 4813 lock_extent_bits(io_tree, hole_start, block_end - 1,
d0082371 4814 &cached_state);
fa7c1494
MX		 4815 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4816 block_end - hole_start);
9036c102
YZ		 4817 if (!ordered)
4818 break;
2ac55d41
JB		 4819 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4820 &cached_state, GFP_NOFS);
fa7c1494 4821 btrfs_start_ordered_extent(inode, ordered, 1);
9036c102
YZ		 4822 btrfs_put_ordered_extent(ordered);
4823 }
39279cc3 4824
9036c102
YZ		 4825 cur_offset = hole_start;
4826 while (1) {
4827 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4828 block_end - cur_offset, 0);
79787eaa
JM		 4829 if (IS_ERR(em)) {
4830 err = PTR_ERR(em);
f2767956 4831 em = NULL;
79787eaa
JM		 4832 break;
4833 }
9036c102 4834 last_byte = min(extent_map_end(em), block_end);
fda2832f 4835 last_byte = ALIGN(last_byte , root->sectorsize);
8082510e 4836 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5dc562c5 4837 struct extent_map *hole_em;
9036c102 4838 hole_size = last_byte - cur_offset;
9ed74f2d 4839
16e7549f
JB		 4840 err = maybe_insert_hole(root, inode, cur_offset,
4841 hole_size);
4842 if (err)
3893e33b 4843 break;
5dc562c5
JB		 4844 btrfs_drop_extent_cache(inode, cur_offset,
4845 cur_offset + hole_size - 1, 0);
4846 hole_em = alloc_extent_map();
4847 if (!hole_em) {
4848 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4849 &BTRFS_I(inode)->runtime_flags);
4850 goto next;
4851 }
4852 hole_em->start = cur_offset;
4853 hole_em->len = hole_size;
4854 hole_em->orig_start = cur_offset;
8082510e 4855
5dc562c5
JB		 4856 hole_em->block_start = EXTENT_MAP_HOLE;
4857 hole_em->block_len = 0;
b4939680 4858 hole_em->orig_block_len = 0;
cc95bef6 4859 hole_em->ram_bytes = hole_size;
5dc562c5
JB		 4860 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4861 hole_em->compress_type = BTRFS_COMPRESS_NONE;
16e7549f 4862 hole_em->generation = root->fs_info->generation;
8082510e 4863
5dc562c5
JB		 4864 while (1) {
4865 write_lock(&em_tree->lock);
09a2a8f9 4866 err = add_extent_mapping(em_tree, hole_em, 1);
5dc562c5
JB		 4867 write_unlock(&em_tree->lock);
4868 if (err != -EEXIST)
4869 break;
4870 btrfs_drop_extent_cache(inode, cur_offset,
4871 cur_offset +
4872 hole_size - 1, 0);
4873 }
4874 free_extent_map(hole_em);
9036c102 4875 }
16e7549f 4876next:
9036c102 4877 free_extent_map(em);
a22285a6 4878 em = NULL;
9036c102 4879 cur_offset = last_byte;
8082510e 4880 if (cur_offset >= block_end)
9036c102
YZ		 4881 break;
4882 }
a22285a6 4883 free_extent_map(em);
2ac55d41
JB		 4884 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4885 GFP_NOFS);
9036c102
YZ		 4886 return err;
4887}
39279cc3 4888
3972f260 4889static int btrfs_setsize(struct inode *inode, struct iattr *attr)
8082510e 4890{
f4a2f4c5
MX		 4891 struct btrfs_root *root = BTRFS_I(inode)->root;
4892 struct btrfs_trans_handle *trans;
a41ad394 4893 loff_t oldsize = i_size_read(inode);
3972f260
ES		 4894 loff_t newsize = attr->ia_size;
4895 int mask = attr->ia_valid;
8082510e
YZ		 4896 int ret;
4897
3972f260
ES		 4898 /*
4899 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4900 * special case where we need to update the times despite not having
4901 * these flags set. For all other operations the VFS set these flags
4902 * explicitly if it wants a timestamp update.
4903 */
dff6efc3
CH		 4904 if (newsize != oldsize) {
4905 inode_inc_iversion(inode);
4906 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4907 inode->i_ctime = inode->i_mtime =
4908 current_fs_time(inode->i_sb);
4909 }
3972f260 4910
a41ad394 4911 if (newsize > oldsize) {
9ea24bbe
FM		 4912 /*
4913 * Don't do an expanding truncate while snapshoting is ongoing.
4914 * This is to ensure the snapshot captures a fully consistent
4915 * state of this file - if the snapshot captures this expanding
4916 * truncation, it must capture all writes that happened before
4917 * this truncation.
4918 */
0bc19f90 4919 btrfs_wait_for_snapshot_creation(root);
a41ad394 4920 ret = btrfs_cont_expand(inode, oldsize, newsize);
9ea24bbe
FM		 4921 if (ret) {
4922 btrfs_end_write_no_snapshoting(root);
8082510e 4923 return ret;
9ea24bbe 4924 }
8082510e 4925
f4a2f4c5 4926 trans = btrfs_start_transaction(root, 1);
9ea24bbe
FM		 4927 if (IS_ERR(trans)) {
4928 btrfs_end_write_no_snapshoting(root);
f4a2f4c5 4929 return PTR_ERR(trans);
9ea24bbe 4930 }
f4a2f4c5
MX		 4931
4932 i_size_write(inode, newsize);
4933 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
27772b68 4934 pagecache_isize_extended(inode, oldsize, newsize);
f4a2f4c5 4935 ret = btrfs_update_inode(trans, root, inode);
9ea24bbe 4936 btrfs_end_write_no_snapshoting(root);
7ad85bb7 4937 btrfs_end_transaction(trans, root);
a41ad394 4938 } else {
8082510e 4939
a41ad394
JB		 4940 /*
4941 * We're truncating a file that used to have good data down to
4942 * zero. Make sure it gets into the ordered flush list so that
4943 * any new writes get down to disk quickly.
4944 */
4945 if (newsize == 0)
72ac3c0d
JB		 4946 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4947 &BTRFS_I(inode)->runtime_flags);
8082510e 4948
f3fe820c
JB		 4949 /*
4950 * 1 for the orphan item we're going to add
4951 * 1 for the orphan item deletion.
4952 */
4953 trans = btrfs_start_transaction(root, 2);
4954 if (IS_ERR(trans))
4955 return PTR_ERR(trans);
4956
4957 /*
4958 * We need to do this in case we fail at _any_ point during the
4959 * actual truncate. Once we do the truncate_setsize we could
4960 * invalidate pages which forces any outstanding ordered io to
4961 * be instantly completed which will give us extents that need
4962 * to be truncated. If we fail to get an orphan inode down we
4963 * could have left over extents that were never meant to live,
4964 * so we need to garuntee from this point on that everything
4965 * will be consistent.
4966 */
4967 ret = btrfs_orphan_add(trans, inode);
4968 btrfs_end_transaction(trans, root);
4969 if (ret)
4970 return ret;
4971
a41ad394
JB		 4972 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4973 truncate_setsize(inode, newsize);
2e60a51e
MX		 4974
4975 /* Disable nonlocked read DIO to avoid the end less truncate */
4976 btrfs_inode_block_unlocked_dio(inode);
4977 inode_dio_wait(inode);
4978 btrfs_inode_resume_unlocked_dio(inode);
4979
a41ad394 4980 ret = btrfs_truncate(inode);
7f4f6e0a
JB		 4981 if (ret && inode->i_nlink) {
4982 int err;
4983
4984 /*
4985 * failed to truncate, disk_i_size is only adjusted down
4986 * as we remove extents, so it should represent the true
4987 * size of the inode, so reset the in memory size and
4988 * delete our orphan entry.
4989 */
4990 trans = btrfs_join_transaction(root);
4991 if (IS_ERR(trans)) {
4992 btrfs_orphan_del(NULL, inode);
4993 return ret;
4994 }
4995 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4996 err = btrfs_orphan_del(trans, inode);
4997 if (err)
4998 btrfs_abort_transaction(trans, root, err);
4999 btrfs_end_transaction(trans, root);
5000 }
8082510e
YZ		 5001 }
5002
a41ad394 5003 return ret;
8082510e
YZ		 5004}
5005
9036c102
YZ		 5006static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5007{
2b0143b5 5008 struct inode *inode = d_inode(dentry);
b83cc969 5009 struct btrfs_root *root = BTRFS_I(inode)->root;
9036c102 5010 int err;
39279cc3 5011
b83cc969
LZ		 5012 if (btrfs_root_readonly(root))
5013 return -EROFS;
5014
9036c102
YZ		 5015 err = inode_change_ok(inode, attr);
5016 if (err)
5017 return err;
2bf5a725 5018
5a3f23d5 5019 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3972f260 5020 err = btrfs_setsize(inode, attr);
8082510e
YZ		 5021 if (err)
5022 return err;
39279cc3 5023 }
9036c102 5024
1025774c
CH		 5025 if (attr->ia_valid) {
5026 setattr_copy(inode, attr);
0c4d2d95 5027 inode_inc_iversion(inode);
22c44fe6 5028 err = btrfs_dirty_inode(inode);
1025774c 5029
22c44fe6 5030 if (!err && attr->ia_valid & ATTR_MODE)
996a710d 5031 err = posix_acl_chmod(inode, inode->i_mode);
1025774c 5032 }
33268eaf 5033
39279cc3
CM		 5034 return err;
5035}
61295eb8 5036
131e404a
FDBM		 5037/*
5038 * While truncating the inode pages during eviction, we get the VFS calling
5039 * btrfs_invalidatepage() against each page of the inode. This is slow because
5040 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5041 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5042 * extent_state structures over and over, wasting lots of time.
5043 *
5044 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5045 * those expensive operations on a per page basis and do only the ordered io
5046 * finishing, while we release here the extent_map and extent_state structures,
5047 * without the excessive merging and splitting.
5048 */
5049static void evict_inode_truncate_pages(struct inode *inode)
5050{
5051 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5052 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5053 struct rb_node *node;
5054
5055 ASSERT(inode->i_state & I_FREEING);
91b0abe3 5056 truncate_inode_pages_final(&inode->i_data);
131e404a
FDBM		 5057
5058 write_lock(&map_tree->lock);
5059 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5060 struct extent_map *em;
5061
5062 node = rb_first(&map_tree->map);
5063 em = rb_entry(node, struct extent_map, rb_node);
180589ef
WS		 5064 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5065 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
131e404a
FDBM		 5066 remove_extent_mapping(map_tree, em);
5067 free_extent_map(em);
7064dd5c
FM		 5068 if (need_resched()) {
5069 write_unlock(&map_tree->lock);
5070 cond_resched();
5071 write_lock(&map_tree->lock);
5072 }
131e404a
FDBM		 5073 }
5074 write_unlock(&map_tree->lock);
5075
6ca07097
FM		 5076 /*
5077 * Keep looping until we have no more ranges in the io tree.
5078 * We can have ongoing bios started by readpages (called from readahead)
9c6429d9
FM		 5079 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5080 * still in progress (unlocked the pages in the bio but did not yet
5081 * unlocked the ranges in the io tree). Therefore this means some
6ca07097
FM		 5082 * ranges can still be locked and eviction started because before
5083 * submitting those bios, which are executed by a separate task (work
5084 * queue kthread), inode references (inode->i_count) were not taken
5085 * (which would be dropped in the end io callback of each bio).
5086 * Therefore here we effectively end up waiting for those bios and
5087 * anyone else holding locked ranges without having bumped the inode's
5088 * reference count - if we don't do it, when they access the inode's
5089 * io_tree to unlock a range it may be too late, leading to an
5090 * use-after-free issue.
5091 */
131e404a
FDBM		 5092 spin_lock(&io_tree->lock);
5093 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5094 struct extent_state *state;
5095 struct extent_state *cached_state = NULL;
6ca07097
FM		 5096 u64 start;
5097 u64 end;
131e404a
FDBM		 5098
5099 node = rb_first(&io_tree->state);
5100 state = rb_entry(node, struct extent_state, rb_node);
6ca07097
FM		 5101 start = state->start;
5102 end = state->end;
131e404a
FDBM		 5103 spin_unlock(&io_tree->lock);
5104
ff13db41 5105 lock_extent_bits(io_tree, start, end, &cached_state);
b9d0b389
QW		 5106
5107 /*
5108 * If still has DELALLOC flag, the extent didn't reach disk,
5109 * and its reserved space won't be freed by delayed_ref.
5110 * So we need to free its reserved space here.
5111 * (Refer to comment in btrfs_invalidatepage, case 2)
5112 *
5113 * Note, end is the bytenr of last byte, so we need + 1 here.
5114 */
5115 if (state->state & EXTENT_DELALLOC)
5116 btrfs_qgroup_free_data(inode, start, end - start + 1);
5117
6ca07097 5118 clear_extent_bit(io_tree, start, end,
131e404a
FDBM		 5119 EXTENT_LOCKED | EXTENT_DIRTY |
5120 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5121 EXTENT_DEFRAG, 1, 1,
5122 &cached_state, GFP_NOFS);
131e404a 5123
7064dd5c 5124 cond_resched();
131e404a
FDBM		 5125 spin_lock(&io_tree->lock);
5126 }
5127 spin_unlock(&io_tree->lock);
5128}
5129
bd555975 5130void btrfs_evict_inode(struct inode *inode)
39279cc3
CM		 5131{
5132 struct btrfs_trans_handle *trans;
5133 struct btrfs_root *root = BTRFS_I(inode)->root;
726c35fa 5134 struct btrfs_block_rsv *rsv, *global_rsv;
3bce876f 5135 int steal_from_global = 0;
07127184 5136 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3
CM		 5137 int ret;
5138
1abe9b8a 5139 trace_btrfs_inode_evict(inode);
5140
131e404a
FDBM		 5141 evict_inode_truncate_pages(inode);
5142
69e9c6c6
SB		 5143 if (inode->i_nlink &&
5144 ((btrfs_root_refs(&root->root_item) != 0 &&
5145 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5146 btrfs_is_free_space_inode(inode)))
bd555975
AV		 5147 goto no_delete;
5148
39279cc3 5149 if (is_bad_inode(inode)) {
7b128766 5150 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 5151 goto no_delete;
5152 }
bd555975 5153 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
a30e577c
JM		 5154 if (!special_file(inode->i_mode))
5155 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 5156
f612496b
MX		 5157 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5158
c71bf099 5159 if (root->fs_info->log_root_recovering) {
6bf02314 5160 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8a35d95f 5161 &BTRFS_I(inode)->runtime_flags));
c71bf099
YZ		 5162 goto no_delete;
5163 }
5164
76dda93c 5165 if (inode->i_nlink > 0) {
69e9c6c6
SB		 5166 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5167 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
76dda93c
YZ		 5168 goto no_delete;
5169 }
5170
0e8c36a9
MX		 5171 ret = btrfs_commit_inode_delayed_inode(inode);
5172 if (ret) {
5173 btrfs_orphan_del(NULL, inode);
5174 goto no_delete;
5175 }
5176
66d8f3dd 5177 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4289a667
JB		 5178 if (!rsv) {
5179 btrfs_orphan_del(NULL, inode);
5180 goto no_delete;
5181 }
4a338542 5182 rsv->size = min_size;
ca7e70f5 5183 rsv->failfast = 1;
726c35fa 5184 global_rsv = &root->fs_info->global_block_rsv;
4289a667 5185
dbe674a9 5186 btrfs_i_size_write(inode, 0);
5f39d397 5187
4289a667 5188 /*
8407aa46
MX		 5189 * This is a bit simpler than btrfs_truncate since we've already
5190 * reserved our space for our orphan item in the unlink, so we just
5191 * need to reserve some slack space in case we add bytes and update
5192 * inode item when doing the truncate.
4289a667 5193 */
8082510e 5194 while (1) {
08e007d2
MX		 5195 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5196 BTRFS_RESERVE_FLUSH_LIMIT);
726c35fa
JB		 5197
5198 /*
5199 * Try and steal from the global reserve since we will
5200 * likely not use this space anyway, we want to try as
5201 * hard as possible to get this to work.
5202 */
5203 if (ret)
3bce876f
JB		 5204 steal_from_global++;
5205 else
5206 steal_from_global = 0;
5207 ret = 0;
d68fc57b 5208
3bce876f
JB		 5209 /*
5210 * steal_from_global == 0: we reserved stuff, hooray!
5211 * steal_from_global == 1: we didn't reserve stuff, boo!
5212 * steal_from_global == 2: we've committed, still not a lot of
5213 * room but maybe we'll have room in the global reserve this
5214 * time.
5215 * steal_from_global == 3: abandon all hope!
5216 */
5217 if (steal_from_global > 2) {
c2cf52eb
SK		 5218 btrfs_warn(root->fs_info,
5219 "Could not get space for a delete, will truncate on mount %d",
5220 ret);
4289a667
JB		 5221 btrfs_orphan_del(NULL, inode);
5222 btrfs_free_block_rsv(root, rsv);
5223 goto no_delete;
d68fc57b 5224 }
7b128766 5225
0e8c36a9 5226 trans = btrfs_join_transaction(root);
4289a667
JB		 5227 if (IS_ERR(trans)) {
5228 btrfs_orphan_del(NULL, inode);
5229 btrfs_free_block_rsv(root, rsv);
5230 goto no_delete;
d68fc57b 5231 }
7b128766 5232
3bce876f
JB		 5233 /*
5234 * We can't just steal from the global reserve, we need tomake
5235 * sure there is room to do it, if not we need to commit and try
5236 * again.
5237 */
5238 if (steal_from_global) {
5239 if (!btrfs_check_space_for_delayed_refs(trans, root))
5240 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5241 min_size);
5242 else
5243 ret = -ENOSPC;
5244 }
5245
5246 /*
5247 * Couldn't steal from the global reserve, we have too much
5248 * pending stuff built up, commit the transaction and try it
5249 * again.
5250 */
5251 if (ret) {
5252 ret = btrfs_commit_transaction(trans, root);
5253 if (ret) {
5254 btrfs_orphan_del(NULL, inode);
5255 btrfs_free_block_rsv(root, rsv);
5256 goto no_delete;
5257 }
5258 continue;
5259 } else {
5260 steal_from_global = 0;
5261 }
5262
4289a667
JB		 5263 trans->block_rsv = rsv;
5264
d68fc57b 5265 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
28ed1345 5266 if (ret != -ENOSPC && ret != -EAGAIN)
8082510e 5267 break;
85e21bac 5268
8407aa46 5269 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e
YZ		 5270 btrfs_end_transaction(trans, root);
5271 trans = NULL;
b53d3f5d 5272 btrfs_btree_balance_dirty(root);
8082510e 5273 }
5f39d397 5274
4289a667
JB		 5275 btrfs_free_block_rsv(root, rsv);
5276
4ef31a45
JB		 5277 /*
5278 * Errors here aren't a big deal, it just means we leave orphan items
5279 * in the tree. They will be cleaned up on the next mount.
5280 */
8082510e 5281 if (ret == 0) {
4289a667 5282 trans->block_rsv = root->orphan_block_rsv;
4ef31a45
JB		 5283 btrfs_orphan_del(trans, inode);
5284 } else {
5285 btrfs_orphan_del(NULL, inode);
8082510e 5286 }
54aa1f4d 5287
4289a667 5288 trans->block_rsv = &root->fs_info->trans_block_rsv;
581bb050
LZ		 5289 if (!(root == root->fs_info->tree_root ||
5290 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
33345d01 5291 btrfs_return_ino(root, btrfs_ino(inode));
581bb050 5292
54aa1f4d 5293 btrfs_end_transaction(trans, root);
b53d3f5d 5294 btrfs_btree_balance_dirty(root);
39279cc3 5295no_delete:
89042e5a 5296 btrfs_remove_delayed_node(inode);
dbd5768f 5297 clear_inode(inode);
39279cc3
CM		 5298}
5299
5300/*
5301 * this returns the key found in the dir entry in the location pointer.
5302 * If no dir entries were found, location->objectid is 0.
5303 */
5304static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5305 struct btrfs_key *location)
5306{
5307 const char *name = dentry->d_name.name;
5308 int namelen = dentry->d_name.len;
5309 struct btrfs_dir_item *di;
5310 struct btrfs_path *path;
5311 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 5312 int ret = 0;
39279cc3
CM		 5313
5314 path = btrfs_alloc_path();
d8926bb3
MF		 5315 if (!path)
5316 return -ENOMEM;
3954401f 5317
33345d01 5318 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
39279cc3 5319 namelen, 0);
0d9f7f3e
Y		 5320 if (IS_ERR(di))
5321 ret = PTR_ERR(di);
d397712b 5322
c704005d 5323 if (IS_ERR_OR_NULL(di))
3954401f 5324 goto out_err;
d397712b 5325
5f39d397 5326 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 5327out:
39279cc3
CM		 5328 btrfs_free_path(path);
5329 return ret;
3954401f
CM		 5330out_err:
5331 location->objectid = 0;
5332 goto out;
39279cc3
CM		 5333}
5334
5335/*
5336 * when we hit a tree root in a directory, the btrfs part of the inode
5337 * needs to be changed to reflect the root directory of the tree root. This
5338 * is kind of like crossing a mount point.
5339 */
5340static int fixup_tree_root_location(struct btrfs_root *root,
4df27c4d
YZ		 5341 struct inode *dir,
5342 struct dentry *dentry,
5343 struct btrfs_key *location,
5344 struct btrfs_root **sub_root)
39279cc3 5345{
4df27c4d
YZ		 5346 struct btrfs_path *path;
5347 struct btrfs_root *new_root;
5348 struct btrfs_root_ref *ref;
5349 struct extent_buffer *leaf;
1d4c08e0 5350 struct btrfs_key key;
4df27c4d
YZ		 5351 int ret;
5352 int err = 0;
39279cc3 5353
4df27c4d
YZ		 5354 path = btrfs_alloc_path();
5355 if (!path) {
5356 err = -ENOMEM;
5357 goto out;
5358 }
39279cc3 5359
4df27c4d 5360 err = -ENOENT;
1d4c08e0
DS		 5361 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5362 key.type = BTRFS_ROOT_REF_KEY;
5363 key.offset = location->objectid;
5364
5365 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5366 0, 0);
4df27c4d
YZ		 5367 if (ret) {
5368 if (ret < 0)
5369 err = ret;
5370 goto out;
5371 }
39279cc3 5372
4df27c4d
YZ		 5373 leaf = path->nodes[0];
5374 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
33345d01 5375 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4df27c4d
YZ		 5376 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5377 goto out;
39279cc3 5378
4df27c4d
YZ		 5379 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5380 (unsigned long)(ref + 1),
5381 dentry->d_name.len);
5382 if (ret)
5383 goto out;
5384
b3b4aa74 5385 btrfs_release_path(path);
4df27c4d
YZ		 5386
5387 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5388 if (IS_ERR(new_root)) {
5389 err = PTR_ERR(new_root);
5390 goto out;
5391 }
5392
4df27c4d
YZ		 5393 *sub_root = new_root;
5394 location->objectid = btrfs_root_dirid(&new_root->root_item);
5395 location->type = BTRFS_INODE_ITEM_KEY;
5396 location->offset = 0;
5397 err = 0;
5398out:
5399 btrfs_free_path(path);
5400 return err;
39279cc3
CM		 5401}
5402
5d4f98a2
YZ		 5403static void inode_tree_add(struct inode *inode)
5404{
5405 struct btrfs_root *root = BTRFS_I(inode)->root;
5406 struct btrfs_inode *entry;
03e860bd
FNP		 5407 struct rb_node **p;
5408 struct rb_node *parent;
cef21937 5409 struct rb_node *new = &BTRFS_I(inode)->rb_node;
33345d01 5410 u64 ino = btrfs_ino(inode);
5d4f98a2 5411
1d3382cb 5412 if (inode_unhashed(inode))
76dda93c 5413 return;
e1409cef 5414 parent = NULL;
5d4f98a2 5415 spin_lock(&root->inode_lock);
e1409cef 5416 p = &root->inode_tree.rb_node;
5d4f98a2
YZ		 5417 while (*p) {
5418 parent = *p;
5419 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5420
33345d01 5421 if (ino < btrfs_ino(&entry->vfs_inode))
03e860bd 5422 p = &parent->rb_left;
33345d01 5423 else if (ino > btrfs_ino(&entry->vfs_inode))
03e860bd 5424 p = &parent->rb_right;
5d4f98a2
YZ		 5425 else {
5426 WARN_ON(!(entry->vfs_inode.i_state &
a4ffdde6 5427 (I_WILL_FREE | I_FREEING)));
cef21937 5428 rb_replace_node(parent, new, &root->inode_tree);
03e860bd
FNP		 5429 RB_CLEAR_NODE(parent);
5430 spin_unlock(&root->inode_lock);
cef21937 5431 return;
5d4f98a2
YZ		 5432 }
5433 }
cef21937
FDBM		 5434 rb_link_node(new, parent, p);
5435 rb_insert_color(new, &root->inode_tree);
5d4f98a2
YZ		 5436 spin_unlock(&root->inode_lock);
5437}
5438
5439static void inode_tree_del(struct inode *inode)
5440{
5441 struct btrfs_root *root = BTRFS_I(inode)->root;
76dda93c 5442 int empty = 0;
5d4f98a2 5443
03e860bd 5444 spin_lock(&root->inode_lock);
5d4f98a2 5445 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5d4f98a2 5446 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5d4f98a2 5447 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
76dda93c 5448 empty = RB_EMPTY_ROOT(&root->inode_tree);
5d4f98a2 5449 }
03e860bd 5450 spin_unlock(&root->inode_lock);
76dda93c 5451
69e9c6c6 5452 if (empty && btrfs_root_refs(&root->root_item) == 0) {
76dda93c
YZ		 5453 synchronize_srcu(&root->fs_info->subvol_srcu);
5454 spin_lock(&root->inode_lock);
5455 empty = RB_EMPTY_ROOT(&root->inode_tree);
5456 spin_unlock(&root->inode_lock);
5457 if (empty)
5458 btrfs_add_dead_root(root);
5459 }
5460}
5461
143bede5 5462void btrfs_invalidate_inodes(struct btrfs_root *root)
76dda93c
YZ		 5463{
5464 struct rb_node *node;
5465 struct rb_node *prev;
5466 struct btrfs_inode *entry;
5467 struct inode *inode;
5468 u64 objectid = 0;
5469
7813b3db
LB		 5470 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5471 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
76dda93c
YZ		 5472
5473 spin_lock(&root->inode_lock);
5474again:
5475 node = root->inode_tree.rb_node;
5476 prev = NULL;
5477 while (node) {
5478 prev = node;
5479 entry = rb_entry(node, struct btrfs_inode, rb_node);
5480
33345d01 5481 if (objectid < btrfs_ino(&entry->vfs_inode))
76dda93c 5482 node = node->rb_left;
33345d01 5483 else if (objectid > btrfs_ino(&entry->vfs_inode))
76dda93c
YZ		 5484 node = node->rb_right;
5485 else
5486 break;
5487 }
5488 if (!node) {
5489 while (prev) {
5490 entry = rb_entry(prev, struct btrfs_inode, rb_node);
33345d01 5491 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
76dda93c
YZ		 5492 node = prev;
5493 break;
5494 }
5495 prev = rb_next(prev);
5496 }
5497 }
5498 while (node) {
5499 entry = rb_entry(node, struct btrfs_inode, rb_node);
33345d01 5500 objectid = btrfs_ino(&entry->vfs_inode) + 1;
76dda93c
YZ		 5501 inode = igrab(&entry->vfs_inode);
5502 if (inode) {
5503 spin_unlock(&root->inode_lock);
5504 if (atomic_read(&inode->i_count) > 1)
5505 d_prune_aliases(inode);
5506 /*
45321ac5 5507 * btrfs_drop_inode will have it removed from
76dda93c
YZ		 5508 * the inode cache when its usage count
5509 * hits zero.
5510 */
5511 iput(inode);
5512 cond_resched();
5513 spin_lock(&root->inode_lock);
5514 goto again;
5515 }
5516
5517 if (cond_resched_lock(&root->inode_lock))
5518 goto again;
5519
5520 node = rb_next(node);
5521 }
5522 spin_unlock(&root->inode_lock);
5d4f98a2
YZ		 5523}
5524
e02119d5
CM		 5525static int btrfs_init_locked_inode(struct inode *inode, void *p)
5526{
5527 struct btrfs_iget_args *args = p;
90d3e592
CM		 5528 inode->i_ino = args->location->objectid;
5529 memcpy(&BTRFS_I(inode)->location, args->location,
5530 sizeof(*args->location));
e02119d5 5531 BTRFS_I(inode)->root = args->root;
39279cc3
CM		 5532 return 0;
5533}
5534
5535static int btrfs_find_actor(struct inode *inode, void *opaque)
5536{
5537 struct btrfs_iget_args *args = opaque;
90d3e592 5538 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
d397712b 5539 args->root == BTRFS_I(inode)->root;
39279cc3
CM		 5540}
5541
5d4f98a2 5542static struct inode *btrfs_iget_locked(struct super_block *s,
90d3e592 5543 struct btrfs_key *location,
5d4f98a2 5544 struct btrfs_root *root)
39279cc3
CM		 5545{
5546 struct inode *inode;
5547 struct btrfs_iget_args args;
90d3e592 5548 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
778ba82b 5549
90d3e592 5550 args.location = location;
39279cc3
CM		 5551 args.root = root;
5552
778ba82b 5553 inode = iget5_locked(s, hashval, btrfs_find_actor,
39279cc3
CM		 5554 btrfs_init_locked_inode,
5555 (void *)&args);
5556 return inode;
5557}
5558
1a54ef8c
BR		 5559/* Get an inode object given its location and corresponding root.
5560 * Returns in *is_new if the inode was read from disk
5561 */
5562struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
73f73415 5563 struct btrfs_root *root, int *new)
1a54ef8c
BR		 5564{
5565 struct inode *inode;
5566
90d3e592 5567 inode = btrfs_iget_locked(s, location, root);
1a54ef8c 5568 if (!inode)
5d4f98a2 5569 return ERR_PTR(-ENOMEM);
1a54ef8c
BR		 5570
5571 if (inode->i_state & I_NEW) {
1a54ef8c 5572 btrfs_read_locked_inode(inode);
1748f843
MF		 5573 if (!is_bad_inode(inode)) {
5574 inode_tree_add(inode);
5575 unlock_new_inode(inode);
5576 if (new)
5577 *new = 1;
5578 } else {
e0b6d65b
ST		 5579 unlock_new_inode(inode);
5580 iput(inode);
5581 inode = ERR_PTR(-ESTALE);
1748f843
MF		 5582 }
5583 }
5584
1a54ef8c
BR		 5585 return inode;
5586}
5587
4df27c4d
YZ		 5588static struct inode *new_simple_dir(struct super_block *s,
5589 struct btrfs_key *key,
5590 struct btrfs_root *root)
5591{
5592 struct inode *inode = new_inode(s);
5593
5594 if (!inode)
5595 return ERR_PTR(-ENOMEM);
5596
4df27c4d
YZ		 5597 BTRFS_I(inode)->root = root;
5598 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
72ac3c0d 5599 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4df27c4d
YZ		 5600
5601 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
848cce0d 5602 inode->i_op = &btrfs_dir_ro_inode_operations;
4df27c4d
YZ		 5603 inode->i_fop = &simple_dir_operations;
5604 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
04b285f3 5605 inode->i_mtime = current_fs_time(inode->i_sb);
9cc97d64 5606 inode->i_atime = inode->i_mtime;
5607 inode->i_ctime = inode->i_mtime;
5608 BTRFS_I(inode)->i_otime = inode->i_mtime;
4df27c4d
YZ		 5609
5610 return inode;
5611}
5612
3de4586c 5613struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3 5614{
d397712b 5615 struct inode *inode;
4df27c4d 5616 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3
CM		 5617 struct btrfs_root *sub_root = root;
5618 struct btrfs_key location;
76dda93c 5619 int index;
b4aff1f8 5620 int ret = 0;
39279cc3
CM		 5621
5622 if (dentry->d_name.len > BTRFS_NAME_LEN)
5623 return ERR_PTR(-ENAMETOOLONG);
5f39d397 5624
39e3c955 5625 ret = btrfs_inode_by_name(dir, dentry, &location);
39279cc3
CM		 5626 if (ret < 0)
5627 return ERR_PTR(ret);
5f39d397 5628
4df27c4d 5629 if (location.objectid == 0)
5662344b 5630 return ERR_PTR(-ENOENT);
4df27c4d
YZ		 5631
5632 if (location.type == BTRFS_INODE_ITEM_KEY) {
73f73415 5633 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4df27c4d
YZ		 5634 return inode;
5635 }
5636
5637 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5638
76dda93c 5639 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4df27c4d
YZ		 5640 ret = fixup_tree_root_location(root, dir, dentry,
5641 &location, &sub_root);
5642 if (ret < 0) {
5643 if (ret != -ENOENT)
5644 inode = ERR_PTR(ret);
5645 else
5646 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5647 } else {
73f73415 5648 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
39279cc3 5649 }
76dda93c
YZ		 5650 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5651
34d19bad 5652 if (!IS_ERR(inode) && root != sub_root) {
c71bf099
YZ		 5653 down_read(&root->fs_info->cleanup_work_sem);
5654 if (!(inode->i_sb->s_flags & MS_RDONLY))
66b4ffd1 5655 ret = btrfs_orphan_cleanup(sub_root);
c71bf099 5656 up_read(&root->fs_info->cleanup_work_sem);
01cd3367
JB		 5657 if (ret) {
5658 iput(inode);
66b4ffd1 5659 inode = ERR_PTR(ret);
01cd3367 5660 }
c71bf099
YZ		 5661 }
5662
3de4586c
CM		 5663 return inode;
5664}
5665
fe15ce44 5666static int btrfs_dentry_delete(const struct dentry *dentry)
76dda93c
YZ		 5667{
5668 struct btrfs_root *root;
2b0143b5 5669 struct inode *inode = d_inode(dentry);
76dda93c 5670
848cce0d 5671 if (!inode && !IS_ROOT(dentry))
2b0143b5 5672 inode = d_inode(dentry->d_parent);
76dda93c 5673
848cce0d
LZ		 5674 if (inode) {
5675 root = BTRFS_I(inode)->root;
efefb143
YZ		 5676 if (btrfs_root_refs(&root->root_item) == 0)
5677 return 1;
848cce0d
LZ		 5678
5679 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5680 return 1;
efefb143 5681 }
76dda93c
YZ		 5682 return 0;
5683}
5684
b4aff1f8
JB		 5685static void btrfs_dentry_release(struct dentry *dentry)
5686{
944a4515 5687 kfree(dentry->d_fsdata);
b4aff1f8
JB		 5688}
5689
3de4586c 5690static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
00cd8dd3 5691 unsigned int flags)
3de4586c 5692{
5662344b 5693 struct inode *inode;
a66e7cc6 5694
5662344b
TI		 5695 inode = btrfs_lookup_dentry(dir, dentry);
5696 if (IS_ERR(inode)) {
5697 if (PTR_ERR(inode) == -ENOENT)
5698 inode = NULL;
5699 else
5700 return ERR_CAST(inode);
5701 }
5702
41d28bca 5703 return d_splice_alias(inode, dentry);
39279cc3
CM		 5704}
5705
16cdcec7 5706unsigned char btrfs_filetype_table[] = {
39279cc3
CM		 5707 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5708};
5709
9cdda8d3 5710static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
39279cc3 5711{
9cdda8d3 5712 struct inode *inode = file_inode(file);
39279cc3
CM		 5713 struct btrfs_root *root = BTRFS_I(inode)->root;
5714 struct btrfs_item *item;
5715 struct btrfs_dir_item *di;
5716 struct btrfs_key key;
5f39d397 5717 struct btrfs_key found_key;
39279cc3 5718 struct btrfs_path *path;
16cdcec7
MX		 5719 struct list_head ins_list;
5720 struct list_head del_list;
39279cc3 5721 int ret;
5f39d397 5722 struct extent_buffer *leaf;
39279cc3 5723 int slot;
39279cc3
CM		 5724 unsigned char d_type;
5725 int over = 0;
5726 u32 di_cur;
5727 u32 di_total;
5728 u32 di_len;
5729 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 5730 char tmp_name[32];
5731 char *name_ptr;
5732 int name_len;
9cdda8d3 5733 int is_curr = 0; /* ctx->pos points to the current index? */
bc4ef759 5734 bool emitted;
39279cc3
CM		 5735
5736 /* FIXME, use a real flag for deciding about the key type */
5737 if (root->fs_info->tree_root == root)
5738 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 5739
9cdda8d3
AV		 5740 if (!dir_emit_dots(file, ctx))
5741 return 0;
5742
49593bfa 5743 path = btrfs_alloc_path();
16cdcec7
MX		 5744 if (!path)
5745 return -ENOMEM;
ff5714cc 5746
e4058b54 5747 path->reada = READA_FORWARD;
49593bfa 5748
16cdcec7
MX		 5749 if (key_type == BTRFS_DIR_INDEX_KEY) {
5750 INIT_LIST_HEAD(&ins_list);
5751 INIT_LIST_HEAD(&del_list);
5752 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5753 }
5754
962a298f 5755 key.type = key_type;
9cdda8d3 5756 key.offset = ctx->pos;
33345d01 5757 key.objectid = btrfs_ino(inode);
5f39d397 5758
39279cc3
CM		 5759 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5760 if (ret < 0)
5761 goto err;
49593bfa 5762
bc4ef759 5763 emitted = false;
49593bfa 5764 while (1) {
5f39d397 5765 leaf = path->nodes[0];
39279cc3 5766 slot = path->slots[0];
b9e03af0
LZ		 5767 if (slot >= btrfs_header_nritems(leaf)) {
5768 ret = btrfs_next_leaf(root, path);
5769 if (ret < 0)
5770 goto err;
5771 else if (ret > 0)
5772 break;
5773 continue;
39279cc3 5774 }
3de4586c 5775
dd3cc16b 5776 item = btrfs_item_nr(slot);
5f39d397
CM		 5777 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5778
5779 if (found_key.objectid != key.objectid)
39279cc3 5780 break;
962a298f 5781 if (found_key.type != key_type)
39279cc3 5782 break;
9cdda8d3 5783 if (found_key.offset < ctx->pos)
b9e03af0 5784 goto next;
16cdcec7
MX		 5785 if (key_type == BTRFS_DIR_INDEX_KEY &&
5786 btrfs_should_delete_dir_index(&del_list,
5787 found_key.offset))
5788 goto next;
5f39d397 5789
9cdda8d3 5790 ctx->pos = found_key.offset;
16cdcec7 5791 is_curr = 1;
49593bfa 5792
39279cc3
CM		 5793 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5794 di_cur = 0;
5f39d397 5795 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 5796
5797 while (di_cur < di_total) {
5f39d397
CM		 5798 struct btrfs_key location;
5799
22a94d44
JB		 5800 if (verify_dir_item(root, leaf, di))
5801 break;
5802
5f39d397 5803 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 5804 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 5805 name_ptr = tmp_name;
5806 } else {
49e350a4 5807 name_ptr = kmalloc(name_len, GFP_KERNEL);
49593bfa
DW		 5808 if (!name_ptr) {
5809 ret = -ENOMEM;
5810 goto err;
5811 }
5f39d397
CM		 5812 }
5813 read_extent_buffer(leaf, name_ptr,
5814 (unsigned long)(di + 1), name_len);
5815
5816 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5817 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c 5818
fede766f 5819
3de4586c 5820 /* is this a reference to our own snapshot? If so
8c9c2bf7
AJ		 5821 * skip it.
5822 *
5823 * In contrast to old kernels, we insert the snapshot's
5824 * dir item and dir index after it has been created, so
5825 * we won't find a reference to our own snapshot. We
5826 * still keep the following code for backward
5827 * compatibility.
3de4586c
CM		 5828 */
5829 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5830 location.objectid == root->root_key.objectid) {
5831 over = 0;
5832 goto skip;
5833 }
9cdda8d3
AV		 5834 over = !dir_emit(ctx, name_ptr, name_len,
5835 location.objectid, d_type);
5f39d397 5836
3de4586c 5837skip:
5f39d397
CM		 5838 if (name_ptr != tmp_name)
5839 kfree(name_ptr);
5840
39279cc3
CM		 5841 if (over)
5842 goto nopos;
bc4ef759 5843 emitted = true;
5103e947 5844 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 5845 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 5846 di_cur += di_len;
5847 di = (struct btrfs_dir_item *)((char *)di + di_len);
5848 }
b9e03af0
LZ		 5849next:
5850 path->slots[0]++;
39279cc3 5851 }
49593bfa 5852
16cdcec7
MX		 5853 if (key_type == BTRFS_DIR_INDEX_KEY) {
5854 if (is_curr)
9cdda8d3 5855 ctx->pos++;
bc4ef759 5856 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list, &emitted);
16cdcec7
MX		 5857 if (ret)
5858 goto nopos;
5859 }
5860
bc4ef759
DS		 5861 /*
5862 * If we haven't emitted any dir entry, we must not touch ctx->pos as
5863 * it was was set to the termination value in previous call. We assume
5864 * that "." and ".." were emitted if we reach this point and set the
5865 * termination value as well for an empty directory.
5866 */
5867 if (ctx->pos > 2 && !emitted)
5868 goto nopos;
5869
49593bfa 5870 /* Reached end of directory/root. Bump pos past the last item. */
db62efbb
ZB		 5871 ctx->pos++;
5872
5873 /*
5874 * Stop new entries from being returned after we return the last
5875 * entry.
5876 *
5877 * New directory entries are assigned a strictly increasing
5878 * offset. This means that new entries created during readdir
5879 * are *guaranteed* to be seen in the future by that readdir.
5880 * This has broken buggy programs which operate on names as
5881 * they're returned by readdir. Until we re-use freed offsets
5882 * we have this hack to stop new entries from being returned
5883 * under the assumption that they'll never reach this huge
5884 * offset.
5885 *
5886 * This is being careful not to overflow 32bit loff_t unless the
5887 * last entry requires it because doing so has broken 32bit apps
5888 * in the past.
5889 */
5890 if (key_type == BTRFS_DIR_INDEX_KEY) {
5891 if (ctx->pos >= INT_MAX)
5892 ctx->pos = LLONG_MAX;
5893 else
5894 ctx->pos = INT_MAX;
5895 }
39279cc3
CM		 5896nopos:
5897 ret = 0;
5898err:
16cdcec7
MX		 5899 if (key_type == BTRFS_DIR_INDEX_KEY)
5900 btrfs_put_delayed_items(&ins_list, &del_list);
39279cc3 5901 btrfs_free_path(path);
39279cc3
CM		 5902 return ret;
5903}
5904
a9185b41 5905int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
39279cc3
CM		 5906{
5907 struct btrfs_root *root = BTRFS_I(inode)->root;
5908 struct btrfs_trans_handle *trans;
5909 int ret = 0;
0af3d00b 5910 bool nolock = false;
39279cc3 5911
72ac3c0d 5912 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4ca8b41e
CM		 5913 return 0;
5914
83eea1f1 5915 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
82d5902d 5916 nolock = true;
0af3d00b 5917
a9185b41 5918 if (wbc->sync_mode == WB_SYNC_ALL) {
0af3d00b 5919 if (nolock)
7a7eaa40 5920 trans = btrfs_join_transaction_nolock(root);
0af3d00b 5921 else
7a7eaa40 5922 trans = btrfs_join_transaction(root);
3612b495
TI		 5923 if (IS_ERR(trans))
5924 return PTR_ERR(trans);
a698d075 5925 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 5926 }
5927 return ret;
5928}
5929
5930/*
54aa1f4d 5931 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 5932 * inode changes. But, it is most likely to find the inode in cache.
5933 * FIXME, needs more benchmarking...there are no reasons other than performance
5934 * to keep or drop this code.
5935 */
48a3b636 5936static int btrfs_dirty_inode(struct inode *inode)
39279cc3
CM		 5937{
5938 struct btrfs_root *root = BTRFS_I(inode)->root;
5939 struct btrfs_trans_handle *trans;
8929ecfa
YZ		 5940 int ret;
5941
72ac3c0d 5942 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
22c44fe6 5943 return 0;
39279cc3 5944
7a7eaa40 5945 trans = btrfs_join_transaction(root);
22c44fe6
JB		 5946 if (IS_ERR(trans))
5947 return PTR_ERR(trans);
8929ecfa
YZ		 5948
5949 ret = btrfs_update_inode(trans, root, inode);
94b60442
CM		 5950 if (ret && ret == -ENOSPC) {
5951 /* whoops, lets try again with the full transaction */
5952 btrfs_end_transaction(trans, root);
5953 trans = btrfs_start_transaction(root, 1);
22c44fe6
JB		 5954 if (IS_ERR(trans))
5955 return PTR_ERR(trans);
8929ecfa 5956
94b60442 5957 ret = btrfs_update_inode(trans, root, inode);
94b60442 5958 }
39279cc3 5959 btrfs_end_transaction(trans, root);
16cdcec7
MX		 5960 if (BTRFS_I(inode)->delayed_node)
5961 btrfs_balance_delayed_items(root);
22c44fe6
JB		 5962
5963 return ret;
5964}
5965
5966/*
5967 * This is a copy of file_update_time. We need this so we can return error on
5968 * ENOSPC for updating the inode in the case of file write and mmap writes.
5969 */
e41f941a
JB		 5970static int btrfs_update_time(struct inode *inode, struct timespec *now,
5971 int flags)
22c44fe6 5972{
2bc55652
AB		 5973 struct btrfs_root *root = BTRFS_I(inode)->root;
5974
5975 if (btrfs_root_readonly(root))
5976 return -EROFS;
5977
e41f941a 5978 if (flags & S_VERSION)
22c44fe6 5979 inode_inc_iversion(inode);
e41f941a
JB		 5980 if (flags & S_CTIME)
5981 inode->i_ctime = *now;
5982 if (flags & S_MTIME)
5983 inode->i_mtime = *now;
5984 if (flags & S_ATIME)
5985 inode->i_atime = *now;
5986 return btrfs_dirty_inode(inode);
39279cc3
CM		 5987}
5988
d352ac68
CM		 5989/*
5990 * find the highest existing sequence number in a directory
5991 * and then set the in-memory index_cnt variable to reflect
5992 * free sequence numbers
5993 */
aec7477b
JB		 5994static int btrfs_set_inode_index_count(struct inode *inode)
5995{
5996 struct btrfs_root *root = BTRFS_I(inode)->root;
5997 struct btrfs_key key, found_key;
5998 struct btrfs_path *path;
5999 struct extent_buffer *leaf;
6000 int ret;
6001
33345d01 6002 key.objectid = btrfs_ino(inode);
962a298f 6003 key.type = BTRFS_DIR_INDEX_KEY;
aec7477b
JB		 6004 key.offset = (u64)-1;
6005
6006 path = btrfs_alloc_path();
6007 if (!path)
6008 return -ENOMEM;
6009
6010 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6011 if (ret < 0)
6012 goto out;
6013 /* FIXME: we should be able to handle this */
6014 if (ret == 0)
6015 goto out;
6016 ret = 0;
6017
6018 /*
6019 * MAGIC NUMBER EXPLANATION:
6020 * since we search a directory based on f_pos we have to start at 2
6021 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6022 * else has to start at 2
6023 */
6024 if (path->slots[0] == 0) {
6025 BTRFS_I(inode)->index_cnt = 2;
6026 goto out;
6027 }
6028
6029 path->slots[0]--;
6030
6031 leaf = path->nodes[0];
6032 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6033
33345d01 6034 if (found_key.objectid != btrfs_ino(inode) ||
962a298f 6035 found_key.type != BTRFS_DIR_INDEX_KEY) {
aec7477b
JB		 6036 BTRFS_I(inode)->index_cnt = 2;
6037 goto out;
6038 }
6039
6040 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
6041out:
6042 btrfs_free_path(path);
6043 return ret;
6044}
6045
d352ac68
CM		 6046/*
6047 * helper to find a free sequence number in a given directory. This current
6048 * code is very simple, later versions will do smarter things in the btree
6049 */
3de4586c 6050int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 6051{
6052 int ret = 0;
6053
6054 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
16cdcec7
MX		 6055 ret = btrfs_inode_delayed_dir_index_count(dir);
6056 if (ret) {
6057 ret = btrfs_set_inode_index_count(dir);
6058 if (ret)
6059 return ret;
6060 }
aec7477b
JB		 6061 }
6062
00e4e6b3 6063 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 6064 BTRFS_I(dir)->index_cnt++;
6065
6066 return ret;
6067}
6068
b0d5d10f
CM		 6069static int btrfs_insert_inode_locked(struct inode *inode)
6070{
6071 struct btrfs_iget_args args;
6072 args.location = &BTRFS_I(inode)->location;
6073 args.root = BTRFS_I(inode)->root;
6074
6075 return insert_inode_locked4(inode,
6076 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6077 btrfs_find_actor, &args);
6078}
6079
39279cc3
CM		 6080static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6081 struct btrfs_root *root,
aec7477b 6082 struct inode *dir,
9c58309d 6083 const char *name, int name_len,
175a4eb7
AV		 6084 u64 ref_objectid, u64 objectid,
6085 umode_t mode, u64 *index)
39279cc3
CM		 6086{
6087 struct inode *inode;
5f39d397 6088 struct btrfs_inode_item *inode_item;
39279cc3 6089 struct btrfs_key *location;
5f39d397 6090 struct btrfs_path *path;
9c58309d
CM		 6091 struct btrfs_inode_ref *ref;
6092 struct btrfs_key key[2];
6093 u32 sizes[2];
ef3b9af5 6094 int nitems = name ? 2 : 1;
9c58309d 6095 unsigned long ptr;
39279cc3 6096 int ret;
39279cc3 6097
5f39d397 6098 path = btrfs_alloc_path();
d8926bb3
MF		 6099 if (!path)
6100 return ERR_PTR(-ENOMEM);
5f39d397 6101
39279cc3 6102 inode = new_inode(root->fs_info->sb);
8fb27640
YS		 6103 if (!inode) {
6104 btrfs_free_path(path);
39279cc3 6105 return ERR_PTR(-ENOMEM);
8fb27640 6106 }
39279cc3 6107
5762b5c9
FM		 6108 /*
6109 * O_TMPFILE, set link count to 0, so that after this point,
6110 * we fill in an inode item with the correct link count.
6111 */
6112 if (!name)
6113 set_nlink(inode, 0);
6114
581bb050
LZ		 6115 /*
6116 * we have to initialize this early, so we can reclaim the inode
6117 * number if we fail afterwards in this function.
6118 */
6119 inode->i_ino = objectid;
6120
ef3b9af5 6121 if (dir && name) {
1abe9b8a 6122 trace_btrfs_inode_request(dir);
6123
3de4586c 6124 ret = btrfs_set_inode_index(dir, index);
09771430 6125 if (ret) {
8fb27640 6126 btrfs_free_path(path);
09771430 6127 iput(inode);
aec7477b 6128 return ERR_PTR(ret);
09771430 6129 }
ef3b9af5
FM		 6130 } else if (dir) {
6131 *index = 0;
aec7477b
JB		 6132 }
6133 /*
6134 * index_cnt is ignored for everything but a dir,
6135 * btrfs_get_inode_index_count has an explanation for the magic
6136 * number
6137 */
6138 BTRFS_I(inode)->index_cnt = 2;
67de1176 6139 BTRFS_I(inode)->dir_index = *index;
39279cc3 6140 BTRFS_I(inode)->root = root;
e02119d5 6141 BTRFS_I(inode)->generation = trans->transid;
76195853 6142 inode->i_generation = BTRFS_I(inode)->generation;
b888db2b 6143
5dc562c5
JB		 6144 /*
6145 * We could have gotten an inode number from somebody who was fsynced
6146 * and then removed in this same transaction, so let's just set full
6147 * sync since it will be a full sync anyway and this will blow away the
6148 * old info in the log.
6149 */
6150 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6151
9c58309d 6152 key[0].objectid = objectid;
962a298f 6153 key[0].type = BTRFS_INODE_ITEM_KEY;
9c58309d
CM		 6154 key[0].offset = 0;
6155
9c58309d 6156 sizes[0] = sizeof(struct btrfs_inode_item);
ef3b9af5
FM		 6157
6158 if (name) {
6159 /*
6160 * Start new inodes with an inode_ref. This is slightly more
6161 * efficient for small numbers of hard links since they will
6162 * be packed into one item. Extended refs will kick in if we
6163 * add more hard links than can fit in the ref item.
6164 */
6165 key[1].objectid = objectid;
962a298f 6166 key[1].type = BTRFS_INODE_REF_KEY;
ef3b9af5
FM		 6167 key[1].offset = ref_objectid;
6168
6169 sizes[1] = name_len + sizeof(*ref);
6170 }
9c58309d 6171
b0d5d10f
CM		 6172 location = &BTRFS_I(inode)->location;
6173 location->objectid = objectid;
6174 location->offset = 0;
962a298f 6175 location->type = BTRFS_INODE_ITEM_KEY;
b0d5d10f
CM		 6176
6177 ret = btrfs_insert_inode_locked(inode);
6178 if (ret < 0)
6179 goto fail;
6180
b9473439 6181 path->leave_spinning = 1;
ef3b9af5 6182 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
9c58309d 6183 if (ret != 0)
b0d5d10f 6184 goto fail_unlock;
5f39d397 6185
ecc11fab 6186 inode_init_owner(inode, dir, mode);
a76a3cd4 6187 inode_set_bytes(inode, 0);
9cc97d64 6188
04b285f3 6189 inode->i_mtime = current_fs_time(inode->i_sb);
9cc97d64 6190 inode->i_atime = inode->i_mtime;
6191 inode->i_ctime = inode->i_mtime;
6192 BTRFS_I(inode)->i_otime = inode->i_mtime;
6193
5f39d397
CM		 6194 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6195 struct btrfs_inode_item);
293f7e07
LZ		 6196 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6197 sizeof(*inode_item));
e02119d5 6198 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d 6199
ef3b9af5
FM		 6200 if (name) {
6201 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6202 struct btrfs_inode_ref);
6203 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6204 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6205 ptr = (unsigned long)(ref + 1);
6206 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6207 }
9c58309d 6208
5f39d397
CM		 6209 btrfs_mark_buffer_dirty(path->nodes[0]);
6210 btrfs_free_path(path);
6211
6cbff00f
CH		 6212 btrfs_inherit_iflags(inode, dir);
6213
569254b0 6214 if (S_ISREG(mode)) {
94272164
CM		 6215 if (btrfs_test_opt(root, NODATASUM))
6216 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
213490b3 6217 if (btrfs_test_opt(root, NODATACOW))
f2bdf9a8
JB		 6218 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6219 BTRFS_INODE_NODATASUM;
94272164
CM		 6220 }
6221
5d4f98a2 6222 inode_tree_add(inode);
1abe9b8a 6223
6224 trace_btrfs_inode_new(inode);
1973f0fa 6225 btrfs_set_inode_last_trans(trans, inode);
1abe9b8a 6226
8ea05e3a
AB		 6227 btrfs_update_root_times(trans, root);
6228
63541927
FDBM		 6229 ret = btrfs_inode_inherit_props(trans, inode, dir);
6230 if (ret)
6231 btrfs_err(root->fs_info,
6232 "error inheriting props for ino %llu (root %llu): %d",
6233 btrfs_ino(inode), root->root_key.objectid, ret);
6234
39279cc3 6235 return inode;
b0d5d10f
CM		 6236
6237fail_unlock:
6238 unlock_new_inode(inode);
5f39d397 6239fail:
ef3b9af5 6240 if (dir && name)
aec7477b 6241 BTRFS_I(dir)->index_cnt--;
5f39d397 6242 btrfs_free_path(path);
09771430 6243 iput(inode);
5f39d397 6244 return ERR_PTR(ret);
39279cc3
CM		 6245}
6246
6247static inline u8 btrfs_inode_type(struct inode *inode)
6248{
6249 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6250}
6251
d352ac68
CM		 6252/*
6253 * utility function to add 'inode' into 'parent_inode' with
6254 * a give name and a given sequence number.
6255 * if 'add_backref' is true, also insert a backref from the
6256 * inode to the parent directory.
6257 */
e02119d5
CM		 6258int btrfs_add_link(struct btrfs_trans_handle *trans,
6259 struct inode *parent_inode, struct inode *inode,
6260 const char *name, int name_len, int add_backref, u64 index)
39279cc3 6261{
4df27c4d 6262 int ret = 0;
39279cc3 6263 struct btrfs_key key;
e02119d5 6264 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
33345d01
LZ		 6265 u64 ino = btrfs_ino(inode);
6266 u64 parent_ino = btrfs_ino(parent_inode);
5f39d397 6267
33345d01 6268 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 6269 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6270 } else {
33345d01 6271 key.objectid = ino;
962a298f 6272 key.type = BTRFS_INODE_ITEM_KEY;
4df27c4d
YZ		 6273 key.offset = 0;
6274 }
6275
33345d01 6276 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 6277 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6278 key.objectid, root->root_key.objectid,
33345d01 6279 parent_ino, index, name, name_len);
4df27c4d 6280 } else if (add_backref) {
33345d01
LZ		 6281 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6282 parent_ino, index);
4df27c4d 6283 }
39279cc3 6284
79787eaa
JM		 6285 /* Nothing to clean up yet */
6286 if (ret)
6287 return ret;
4df27c4d 6288
79787eaa
JM		 6289 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6290 parent_inode, &key,
6291 btrfs_inode_type(inode), index);
9c52057c 6292 if (ret == -EEXIST || ret == -EOVERFLOW)
79787eaa
JM		 6293 goto fail_dir_item;
6294 else if (ret) {
6295 btrfs_abort_transaction(trans, root, ret);
6296 return ret;
39279cc3 6297 }
79787eaa
JM		 6298
6299 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6300 name_len * 2);
0c4d2d95 6301 inode_inc_iversion(parent_inode);
04b285f3
DD		 6302 parent_inode->i_mtime = parent_inode->i_ctime =
6303 current_fs_time(parent_inode->i_sb);
79787eaa
JM		 6304 ret = btrfs_update_inode(trans, root, parent_inode);
6305 if (ret)
6306 btrfs_abort_transaction(trans, root, ret);
39279cc3 6307 return ret;
fe66a05a
CM		 6308
6309fail_dir_item:
6310 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6311 u64 local_index;
6312 int err;
6313 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6314 key.objectid, root->root_key.objectid,
6315 parent_ino, &local_index, name, name_len);
6316
6317 } else if (add_backref) {
6318 u64 local_index;
6319 int err;
6320
6321 err = btrfs_del_inode_ref(trans, root, name, name_len,
6322 ino, parent_ino, &local_index);
6323 }
6324 return ret;
39279cc3
CM		 6325}
6326
6327static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
a1b075d2
JB		 6328 struct inode *dir, struct dentry *dentry,
6329 struct inode *inode, int backref, u64 index)
39279cc3 6330{
a1b075d2
JB		 6331 int err = btrfs_add_link(trans, dir, inode,
6332 dentry->d_name.name, dentry->d_name.len,
6333 backref, index);
39279cc3
CM		 6334 if (err > 0)
6335 err = -EEXIST;
6336 return err;
6337}
6338
618e21d5 6339static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1a67aafb 6340 umode_t mode, dev_t rdev)
618e21d5
JB		 6341{
6342 struct btrfs_trans_handle *trans;
6343 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 6344 struct inode *inode = NULL;
618e21d5
JB		 6345 int err;
6346 int drop_inode = 0;
6347 u64 objectid;
00e4e6b3 6348 u64 index = 0;
618e21d5 6349
9ed74f2d
JB		 6350 /*
6351 * 2 for inode item and ref
6352 * 2 for dir items
6353 * 1 for xattr if selinux is on
6354 */
a22285a6
YZ		 6355 trans = btrfs_start_transaction(root, 5);
6356 if (IS_ERR(trans))
6357 return PTR_ERR(trans);
1832a6d5 6358
581bb050
LZ		 6359 err = btrfs_find_free_ino(root, &objectid);
6360 if (err)
6361 goto out_unlock;
6362
aec7477b 6363 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 6364 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 6365 mode, &index);
7cf96da3
TI		 6366 if (IS_ERR(inode)) {
6367 err = PTR_ERR(inode);
618e21d5 6368 goto out_unlock;
7cf96da3 6369 }
618e21d5 6370
ad19db71
CS		 6371 /*
6372 * If the active LSM wants to access the inode during
6373 * d_instantiate it needs these. Smack checks to see
6374 * if the filesystem supports xattrs by looking at the
6375 * ops vector.
6376 */
ad19db71 6377 inode->i_op = &btrfs_special_inode_operations;
b0d5d10f
CM		 6378 init_special_inode(inode, inode->i_mode, rdev);
6379
6380 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
618e21d5 6381 if (err)
b0d5d10f
CM		 6382 goto out_unlock_inode;
6383
6384 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6385 if (err) {
6386 goto out_unlock_inode;
6387 } else {
1b4ab1bb 6388 btrfs_update_inode(trans, root, inode);
b0d5d10f 6389 unlock_new_inode(inode);
08c422c2 6390 d_instantiate(dentry, inode);
618e21d5 6391 }
b0d5d10f 6392
618e21d5 6393out_unlock:
7ad85bb7 6394 btrfs_end_transaction(trans, root);
c581afc8 6395 btrfs_balance_delayed_items(root);
b53d3f5d 6396 btrfs_btree_balance_dirty(root);
618e21d5
JB		 6397 if (drop_inode) {
6398 inode_dec_link_count(inode);
6399 iput(inode);
6400 }
618e21d5 6401 return err;
b0d5d10f
CM		 6402
6403out_unlock_inode:
6404 drop_inode = 1;
6405 unlock_new_inode(inode);
6406 goto out_unlock;
6407
618e21d5
JB		 6408}
6409
39279cc3 6410static int btrfs_create(struct inode *dir, struct dentry *dentry,
ebfc3b49 6411 umode_t mode, bool excl)
39279cc3
CM		 6412{
6413 struct btrfs_trans_handle *trans;
6414 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 6415 struct inode *inode = NULL;
43baa579 6416 int drop_inode_on_err = 0;
a22285a6 6417 int err;
39279cc3 6418 u64 objectid;
00e4e6b3 6419 u64 index = 0;
39279cc3 6420
9ed74f2d
JB		 6421 /*
6422 * 2 for inode item and ref
6423 * 2 for dir items
6424 * 1 for xattr if selinux is on
6425 */
a22285a6
YZ		 6426 trans = btrfs_start_transaction(root, 5);
6427 if (IS_ERR(trans))
6428 return PTR_ERR(trans);
9ed74f2d 6429
581bb050
LZ		 6430 err = btrfs_find_free_ino(root, &objectid);
6431 if (err)
6432 goto out_unlock;
6433
aec7477b 6434 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 6435 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 6436 mode, &index);
7cf96da3
TI		 6437 if (IS_ERR(inode)) {
6438 err = PTR_ERR(inode);
39279cc3 6439 goto out_unlock;
7cf96da3 6440 }
43baa579 6441 drop_inode_on_err = 1;
ad19db71
CS		 6442 /*
6443 * If the active LSM wants to access the inode during
6444 * d_instantiate it needs these. Smack checks to see
6445 * if the filesystem supports xattrs by looking at the
6446 * ops vector.
6447 */
6448 inode->i_fop = &btrfs_file_operations;
6449 inode->i_op = &btrfs_file_inode_operations;
b0d5d10f 6450 inode->i_mapping->a_ops = &btrfs_aops;
b0d5d10f
CM		 6451
6452 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6453 if (err)
6454 goto out_unlock_inode;
6455
6456 err = btrfs_update_inode(trans, root, inode);
6457 if (err)
6458 goto out_unlock_inode;
ad19db71 6459
a1b075d2 6460 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3 6461 if (err)
b0d5d10f 6462 goto out_unlock_inode;
43baa579 6463
43baa579 6464 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
b0d5d10f 6465 unlock_new_inode(inode);
43baa579
FB		 6466 d_instantiate(dentry, inode);
6467
39279cc3 6468out_unlock:
7ad85bb7 6469 btrfs_end_transaction(trans, root);
43baa579 6470 if (err && drop_inode_on_err) {
39279cc3
CM		 6471 inode_dec_link_count(inode);
6472 iput(inode);
6473 }
c581afc8 6474 btrfs_balance_delayed_items(root);
b53d3f5d 6475 btrfs_btree_balance_dirty(root);
39279cc3 6476 return err;
b0d5d10f
CM		 6477
6478out_unlock_inode:
6479 unlock_new_inode(inode);
6480 goto out_unlock;
6481
39279cc3
CM		 6482}
6483
6484static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6485 struct dentry *dentry)
6486{
271dba45 6487 struct btrfs_trans_handle *trans = NULL;
39279cc3 6488 struct btrfs_root *root = BTRFS_I(dir)->root;
2b0143b5 6489 struct inode *inode = d_inode(old_dentry);
00e4e6b3 6490 u64 index;
39279cc3
CM		 6491 int err;
6492 int drop_inode = 0;
6493
4a8be425
TH		 6494 /* do not allow sys_link's with other subvols of the same device */
6495 if (root->objectid != BTRFS_I(inode)->root->objectid)
3ab3564f 6496 return -EXDEV;
4a8be425 6497
f186373f 6498 if (inode->i_nlink >= BTRFS_LINK_MAX)
c055e99e 6499 return -EMLINK;
4a8be425 6500
3de4586c 6501 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 6502 if (err)
6503 goto fail;
6504
a22285a6 6505 /*
7e6b6465 6506 * 2 items for inode and inode ref
a22285a6 6507 * 2 items for dir items
7e6b6465 6508 * 1 item for parent inode
a22285a6 6509 */
7e6b6465 6510 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 6511 if (IS_ERR(trans)) {
6512 err = PTR_ERR(trans);
271dba45 6513 trans = NULL;
a22285a6
YZ		 6514 goto fail;
6515 }
5f39d397 6516
67de1176
MX		 6517 /* There are several dir indexes for this inode, clear the cache. */
6518 BTRFS_I(inode)->dir_index = 0ULL;
8b558c5f 6519 inc_nlink(inode);
0c4d2d95 6520 inode_inc_iversion(inode);
04b285f3 6521 inode->i_ctime = current_fs_time(inode->i_sb);
7de9c6ee 6522 ihold(inode);
e9976151 6523 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
aec7477b 6524
a1b075d2 6525 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5f39d397 6526
a5719521 6527 if (err) {
54aa1f4d 6528 drop_inode = 1;
a5719521 6529 } else {
10d9f309 6530 struct dentry *parent = dentry->d_parent;
a5719521 6531 err = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 6532 if (err)
6533 goto fail;
ef3b9af5
FM		 6534 if (inode->i_nlink == 1) {
6535 /*
6536 * If new hard link count is 1, it's a file created
6537 * with open(2) O_TMPFILE flag.
6538 */
6539 err = btrfs_orphan_del(trans, inode);
6540 if (err)
6541 goto fail;
6542 }
08c422c2 6543 d_instantiate(dentry, inode);
6a912213 6544 btrfs_log_new_name(trans, inode, NULL, parent);
a5719521 6545 }
39279cc3 6546
c581afc8 6547 btrfs_balance_delayed_items(root);
1832a6d5 6548fail:
271dba45
FM		 6549 if (trans)
6550 btrfs_end_transaction(trans, root);
39279cc3
CM		 6551 if (drop_inode) {
6552 inode_dec_link_count(inode);
6553 iput(inode);
6554 }
b53d3f5d 6555 btrfs_btree_balance_dirty(root);
39279cc3
CM		 6556 return err;
6557}
6558
18bb1db3 6559static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
39279cc3 6560{
b9d86667 6561 struct inode *inode = NULL;
39279cc3
CM		 6562 struct btrfs_trans_handle *trans;
6563 struct btrfs_root *root = BTRFS_I(dir)->root;
6564 int err = 0;
6565 int drop_on_err = 0;
b9d86667 6566 u64 objectid = 0;
00e4e6b3 6567 u64 index = 0;
39279cc3 6568
9ed74f2d
JB		 6569 /*
6570 * 2 items for inode and ref
6571 * 2 items for dir items
6572 * 1 for xattr if selinux is on
6573 */
a22285a6
YZ		 6574 trans = btrfs_start_transaction(root, 5);
6575 if (IS_ERR(trans))
6576 return PTR_ERR(trans);
39279cc3 6577
581bb050
LZ		 6578 err = btrfs_find_free_ino(root, &objectid);
6579 if (err)
6580 goto out_fail;
6581
aec7477b 6582 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 6583 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 6584 S_IFDIR | mode, &index);
39279cc3
CM		 6585 if (IS_ERR(inode)) {
6586 err = PTR_ERR(inode);
6587 goto out_fail;
6588 }
5f39d397 6589
39279cc3 6590 drop_on_err = 1;
b0d5d10f
CM		 6591 /* these must be set before we unlock the inode */
6592 inode->i_op = &btrfs_dir_inode_operations;
6593 inode->i_fop = &btrfs_dir_file_operations;
33268eaf 6594
2a7dba39 6595 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf 6596 if (err)
b0d5d10f 6597 goto out_fail_inode;
39279cc3 6598
dbe674a9 6599 btrfs_i_size_write(inode, 0);
39279cc3
CM		 6600 err = btrfs_update_inode(trans, root, inode);
6601 if (err)
b0d5d10f 6602 goto out_fail_inode;
5f39d397 6603
a1b075d2
JB		 6604 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6605 dentry->d_name.len, 0, index);
39279cc3 6606 if (err)
b0d5d10f 6607 goto out_fail_inode;
5f39d397 6608
39279cc3 6609 d_instantiate(dentry, inode);
b0d5d10f
CM		 6610 /*
6611 * mkdir is special. We're unlocking after we call d_instantiate
6612 * to avoid a race with nfsd calling d_instantiate.
6613 */
6614 unlock_new_inode(inode);
39279cc3 6615 drop_on_err = 0;
39279cc3
CM		 6616
6617out_fail:
7ad85bb7 6618 btrfs_end_transaction(trans, root);
c7cfb8a5
WS		 6619 if (drop_on_err) {
6620 inode_dec_link_count(inode);
39279cc3 6621 iput(inode);
c7cfb8a5 6622 }
c581afc8 6623 btrfs_balance_delayed_items(root);
b53d3f5d 6624 btrfs_btree_balance_dirty(root);
39279cc3 6625 return err;
b0d5d10f
CM		 6626
6627out_fail_inode:
6628 unlock_new_inode(inode);
6629 goto out_fail;
39279cc3
CM		 6630}
6631
e6c4efd8
QW		 6632/* Find next extent map of a given extent map, caller needs to ensure locks */
6633static struct extent_map *next_extent_map(struct extent_map *em)
6634{
6635 struct rb_node *next;
6636
6637 next = rb_next(&em->rb_node);
6638 if (!next)
6639 return NULL;
6640 return container_of(next, struct extent_map, rb_node);
6641}
6642
6643static struct extent_map *prev_extent_map(struct extent_map *em)
6644{
6645 struct rb_node *prev;
6646
6647 prev = rb_prev(&em->rb_node);
6648 if (!prev)
6649 return NULL;
6650 return container_of(prev, struct extent_map, rb_node);
6651}
6652
d352ac68 6653/* helper for btfs_get_extent. Given an existing extent in the tree,
e6c4efd8 6654 * the existing extent is the nearest extent to map_start,
d352ac68 6655 * and an extent that you want to insert, deal with overlap and insert
e6c4efd8 6656 * the best fitted new extent into the tree.
d352ac68 6657 */
3b951516
CM		 6658static int merge_extent_mapping(struct extent_map_tree *em_tree,
6659 struct extent_map *existing,
e6dcd2dc 6660 struct extent_map *em,
51f395ad 6661 u64 map_start)
3b951516 6662{
e6c4efd8
QW		 6663 struct extent_map *prev;
6664 struct extent_map *next;
6665 u64 start;
6666 u64 end;
3b951516 6667 u64 start_diff;
3b951516 6668
e6dcd2dc 6669 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
e6c4efd8
QW		 6670
6671 if (existing->start > map_start) {
6672 next = existing;
6673 prev = prev_extent_map(next);
6674 } else {
6675 prev = existing;
6676 next = next_extent_map(prev);
6677 }
6678
6679 start = prev ? extent_map_end(prev) : em->start;
6680 start = max_t(u64, start, em->start);
6681 end = next ? next->start : extent_map_end(em);
6682 end = min_t(u64, end, extent_map_end(em));
6683 start_diff = start - em->start;
6684 em->start = start;
6685 em->len = end - start;
c8b97818
CM		 6686 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6687 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 6688 em->block_start += start_diff;
c8b97818
CM		 6689 em->block_len -= start_diff;
6690 }
09a2a8f9 6691 return add_extent_mapping(em_tree, em, 0);
3b951516
CM		 6692}
6693
c8b97818 6694static noinline int uncompress_inline(struct btrfs_path *path,
e40da0e5 6695 struct page *page,
c8b97818
CM		 6696 size_t pg_offset, u64 extent_offset,
6697 struct btrfs_file_extent_item *item)
6698{
6699 int ret;
6700 struct extent_buffer *leaf = path->nodes[0];
6701 char *tmp;
6702 size_t max_size;
6703 unsigned long inline_size;
6704 unsigned long ptr;
261507a0 6705 int compress_type;
c8b97818
CM		 6706
6707 WARN_ON(pg_offset != 0);
261507a0 6708 compress_type = btrfs_file_extent_compression(leaf, item);
c8b97818
CM		 6709 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6710 inline_size = btrfs_file_extent_inline_item_len(leaf,
dd3cc16b 6711 btrfs_item_nr(path->slots[0]));
c8b97818 6712 tmp = kmalloc(inline_size, GFP_NOFS);
8d413713
TI		 6713 if (!tmp)
6714 return -ENOMEM;
c8b97818
CM		 6715 ptr = btrfs_file_extent_inline_start(item);
6716
6717 read_extent_buffer(leaf, tmp, ptr, inline_size);
6718
09cbfeaf 6719 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
261507a0
LZ		 6720 ret = btrfs_decompress(compress_type, tmp, page,
6721 extent_offset, inline_size, max_size);
c8b97818 6722 kfree(tmp);
166ae5a4 6723 return ret;
c8b97818
CM		 6724}
6725
d352ac68
CM		 6726/*
6727 * a bit scary, this does extent mapping from logical file offset to the disk.
d397712b
CM		 6728 * the ugly parts come from merging extents from the disk with the in-ram
6729 * representation. This gets more complex because of the data=ordered code,
d352ac68
CM		 6730 * where the in-ram extents might be locked pending data=ordered completion.
6731 *
6732 * This also copies inline extents directly into the page.
6733 */
d397712b 6734
a52d9a80 6735struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 6736 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 6737 int create)
6738{
6739 int ret;
6740 int err = 0;
a52d9a80
CM		 6741 u64 extent_start = 0;
6742 u64 extent_end = 0;
33345d01 6743 u64 objectid = btrfs_ino(inode);
a52d9a80 6744 u32 found_type;
f421950f 6745 struct btrfs_path *path = NULL;
a52d9a80
CM		 6746 struct btrfs_root *root = BTRFS_I(inode)->root;
6747 struct btrfs_file_extent_item *item;
5f39d397
CM		 6748 struct extent_buffer *leaf;
6749 struct btrfs_key found_key;
a52d9a80
CM		 6750 struct extent_map *em = NULL;
6751 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 6752 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 6753 struct btrfs_trans_handle *trans = NULL;
7ffbb598 6754 const bool new_inline = !page || create;
a52d9a80 6755
a52d9a80 6756again:
890871be 6757 read_lock(&em_tree->lock);
d1310b2e 6758 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 6759 if (em)
6760 em->bdev = root->fs_info->fs_devices->latest_bdev;
890871be 6761 read_unlock(&em_tree->lock);
d1310b2e 6762
a52d9a80 6763 if (em) {
e1c4b745
CM		 6764 if (em->start > start || em->start + em->len <= start)
6765 free_extent_map(em);
6766 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 6767 free_extent_map(em);
6768 else
6769 goto out;
a52d9a80 6770 }
172ddd60 6771 em = alloc_extent_map();
a52d9a80 6772 if (!em) {
d1310b2e
CM		 6773 err = -ENOMEM;
6774 goto out;
a52d9a80 6775 }
e6dcd2dc 6776 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 6777 em->start = EXTENT_MAP_HOLE;
445a6944 6778 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 6779 em->len = (u64)-1;
c8b97818 6780 em->block_len = (u64)-1;
f421950f
CM		 6781
6782 if (!path) {
6783 path = btrfs_alloc_path();
026fd317
JB		 6784 if (!path) {
6785 err = -ENOMEM;
6786 goto out;
6787 }
6788 /*
6789 * Chances are we'll be called again, so go ahead and do
6790 * readahead
6791 */
e4058b54 6792 path->reada = READA_FORWARD;
f421950f
CM		 6793 }
6794
179e29e4
CM		 6795 ret = btrfs_lookup_file_extent(trans, root, path,
6796 objectid, start, trans != NULL);
a52d9a80
CM		 6797 if (ret < 0) {
6798 err = ret;
6799 goto out;
6800 }
6801
6802 if (ret != 0) {
6803 if (path->slots[0] == 0)
6804 goto not_found;
6805 path->slots[0]--;
6806 }
6807
5f39d397
CM		 6808 leaf = path->nodes[0];
6809 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 6810 struct btrfs_file_extent_item);
a52d9a80 6811 /* are we inside the extent that was found? */
5f39d397 6812 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
962a298f 6813 found_type = found_key.type;
5f39d397 6814 if (found_key.objectid != objectid ||
a52d9a80 6815 found_type != BTRFS_EXTENT_DATA_KEY) {
25a50341
JB		 6816 /*
6817 * If we backup past the first extent we want to move forward
6818 * and see if there is an extent in front of us, otherwise we'll
6819 * say there is a hole for our whole search range which can
6820 * cause problems.
6821 */
6822 extent_end = start;
6823 goto next;
a52d9a80
CM		 6824 }
6825
5f39d397
CM		 6826 found_type = btrfs_file_extent_type(leaf, item);
6827 extent_start = found_key.offset;
d899e052
YZ		 6828 if (found_type == BTRFS_FILE_EXTENT_REG ||
6829 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 6830 extent_end = extent_start +
db94535d 6831 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 6832 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6833 size_t size;
514ac8ad 6834 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
fda2832f 6835 extent_end = ALIGN(extent_start + size, root->sectorsize);
9036c102 6836 }
25a50341 6837next:
9036c102
YZ		 6838 if (start >= extent_end) {
6839 path->slots[0]++;
6840 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6841 ret = btrfs_next_leaf(root, path);
6842 if (ret < 0) {
6843 err = ret;
6844 goto out;
a52d9a80 6845 }
9036c102
YZ		 6846 if (ret > 0)
6847 goto not_found;
6848 leaf = path->nodes[0];
a52d9a80 6849 }
9036c102
YZ		 6850 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6851 if (found_key.objectid != objectid ||
6852 found_key.type != BTRFS_EXTENT_DATA_KEY)
6853 goto not_found;
6854 if (start + len <= found_key.offset)
6855 goto not_found;
e2eca69d
WS		 6856 if (start > found_key.offset)
6857 goto next;
9036c102 6858 em->start = start;
70c8a91c 6859 em->orig_start = start;
9036c102
YZ		 6860 em->len = found_key.offset - start;
6861 goto not_found_em;
6862 }
6863
7ffbb598
FM		 6864 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6865
d899e052
YZ		 6866 if (found_type == BTRFS_FILE_EXTENT_REG ||
6867 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80
CM		 6868 goto insert;
6869 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 6870 unsigned long ptr;
a52d9a80 6871 char *map;
3326d1b0
CM		 6872 size_t size;
6873 size_t extent_offset;
6874 size_t copy_size;
a52d9a80 6875
7ffbb598 6876 if (new_inline)
689f9346 6877 goto out;
5f39d397 6878
514ac8ad 6879 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
9036c102 6880 extent_offset = page_offset(page) + pg_offset - extent_start;
09cbfeaf
KS		 6881 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6882 size - extent_offset);
3326d1b0 6883 em->start = extent_start + extent_offset;
fda2832f 6884 em->len = ALIGN(copy_size, root->sectorsize);
b4939680 6885 em->orig_block_len = em->len;
70c8a91c 6886 em->orig_start = em->start;
689f9346 6887 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 6888 if (create == 0 && !PageUptodate(page)) {
261507a0
LZ		 6889 if (btrfs_file_extent_compression(leaf, item) !=
6890 BTRFS_COMPRESS_NONE) {
e40da0e5 6891 ret = uncompress_inline(path, page, pg_offset,
c8b97818 6892 extent_offset, item);
166ae5a4
ZB		 6893 if (ret) {
6894 err = ret;
6895 goto out;
6896 }
c8b97818
CM		 6897 } else {
6898 map = kmap(page);
6899 read_extent_buffer(leaf, map + pg_offset, ptr,
6900 copy_size);
09cbfeaf 6901 if (pg_offset + copy_size < PAGE_SIZE) {
93c82d57 6902 memset(map + pg_offset + copy_size, 0,
09cbfeaf 6903 PAGE_SIZE - pg_offset -
93c82d57
CM		 6904 copy_size);
6905 }
c8b97818
CM		 6906 kunmap(page);
6907 }
179e29e4
CM		 6908 flush_dcache_page(page);
6909 } else if (create && PageUptodate(page)) {
6bf7e080 6910 BUG();
179e29e4
CM		 6911 if (!trans) {
6912 kunmap(page);
6913 free_extent_map(em);
6914 em = NULL;
ff5714cc 6915
b3b4aa74 6916 btrfs_release_path(path);
7a7eaa40 6917 trans = btrfs_join_transaction(root);
ff5714cc 6918
3612b495
TI		 6919 if (IS_ERR(trans))
6920 return ERR_CAST(trans);
179e29e4
CM		 6921 goto again;
6922 }
c8b97818 6923 map = kmap(page);
70dec807 6924 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 6925 copy_size);
c8b97818 6926 kunmap(page);
179e29e4 6927 btrfs_mark_buffer_dirty(leaf);
a52d9a80 6928 }
d1310b2e 6929 set_extent_uptodate(io_tree, em->start,
507903b8 6930 extent_map_end(em) - 1, NULL, GFP_NOFS);
a52d9a80 6931 goto insert;
a52d9a80
CM		 6932 }
6933not_found:
6934 em->start = start;
70c8a91c 6935 em->orig_start = start;
d1310b2e 6936 em->len = len;
a52d9a80 6937not_found_em:
5f39d397 6938 em->block_start = EXTENT_MAP_HOLE;
9036c102 6939 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80 6940insert:
b3b4aa74 6941 btrfs_release_path(path);
d1310b2e 6942 if (em->start > start || extent_map_end(em) <= start) {
c2cf52eb 6943 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
c1c9ff7c 6944 em->start, em->len, start, len);
a52d9a80
CM		 6945 err = -EIO;
6946 goto out;
6947 }
d1310b2e
CM		 6948
6949 err = 0;
890871be 6950 write_lock(&em_tree->lock);
09a2a8f9 6951 ret = add_extent_mapping(em_tree, em, 0);
3b951516
CM		 6952 /* it is possible that someone inserted the extent into the tree
6953 * while we had the lock dropped. It is also possible that
6954 * an overlapping map exists in the tree
6955 */
a52d9a80 6956 if (ret == -EEXIST) {
3b951516 6957 struct extent_map *existing;
e6dcd2dc
CM		 6958
6959 ret = 0;
6960
e6c4efd8
QW		 6961 existing = search_extent_mapping(em_tree, start, len);
6962 /*
6963 * existing will always be non-NULL, since there must be
6964 * extent causing the -EEXIST.
6965 */
6966 if (start >= extent_map_end(existing) ||
32be3a1a 6967 start <= existing->start) {
e6c4efd8
QW		 6968 /*
6969 * The existing extent map is the one nearest to
6970 * the [start, start + len) range which overlaps
6971 */
6972 err = merge_extent_mapping(em_tree, existing,
6973 em, start);
e1c4b745 6974 free_extent_map(existing);
e6c4efd8 6975 if (err) {
3b951516
CM		 6976 free_extent_map(em);
6977 em = NULL;
6978 }
6979 } else {
6980 free_extent_map(em);
6981 em = existing;
e6dcd2dc 6982 err = 0;
a52d9a80 6983 }
a52d9a80 6984 }
890871be 6985 write_unlock(&em_tree->lock);
a52d9a80 6986out:
1abe9b8a 6987
4cd8587c 6988 trace_btrfs_get_extent(root, em);
1abe9b8a 6989
527afb44 6990 btrfs_free_path(path);
a52d9a80
CM		 6991 if (trans) {
6992 ret = btrfs_end_transaction(trans, root);
d397712b 6993 if (!err)
a52d9a80
CM		 6994 err = ret;
6995 }
a52d9a80
CM		 6996 if (err) {
6997 free_extent_map(em);
a52d9a80
CM		 6998 return ERR_PTR(err);
6999 }
79787eaa 7000 BUG_ON(!em); /* Error is always set */
a52d9a80
CM		 7001 return em;
7002}
7003
ec29ed5b
CM		 7004struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
7005 size_t pg_offset, u64 start, u64 len,
7006 int create)
7007{
7008 struct extent_map *em;
7009 struct extent_map *hole_em = NULL;
7010 u64 range_start = start;
7011 u64 end;
7012 u64 found;
7013 u64 found_end;
7014 int err = 0;
7015
7016 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7017 if (IS_ERR(em))
7018 return em;
7019 if (em) {
7020 /*
f9e4fb53
LB		 7021 * if our em maps to
7022 * - a hole or
7023 * - a pre-alloc extent,
7024 * there might actually be delalloc bytes behind it.
ec29ed5b 7025 */
f9e4fb53
LB		 7026 if (em->block_start != EXTENT_MAP_HOLE &&
7027 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
ec29ed5b
CM		 7028 return em;
7029 else
7030 hole_em = em;
7031 }
7032
7033 /* check to see if we've wrapped (len == -1 or similar) */
7034 end = start + len;
7035 if (end < start)
7036 end = (u64)-1;
7037 else
7038 end -= 1;
7039
7040 em = NULL;
7041
7042 /* ok, we didn't find anything, lets look for delalloc */
7043 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
7044 end, len, EXTENT_DELALLOC, 1);
7045 found_end = range_start + found;
7046 if (found_end < range_start)
7047 found_end = (u64)-1;
7048
7049 /*
7050 * we didn't find anything useful, return
7051 * the original results from get_extent()
7052 */
7053 if (range_start > end || found_end <= start) {
7054 em = hole_em;
7055 hole_em = NULL;
7056 goto out;
7057 }
7058
7059 /* adjust the range_start to make sure it doesn't
7060 * go backwards from the start they passed in
7061 */
67871254 7062 range_start = max(start, range_start);
ec29ed5b
CM		 7063 found = found_end - range_start;
7064
7065 if (found > 0) {
7066 u64 hole_start = start;
7067 u64 hole_len = len;
7068
172ddd60 7069 em = alloc_extent_map();
ec29ed5b
CM		 7070 if (!em) {
7071 err = -ENOMEM;
7072 goto out;
7073 }
7074 /*
7075 * when btrfs_get_extent can't find anything it
7076 * returns one huge hole
7077 *
7078 * make sure what it found really fits our range, and
7079 * adjust to make sure it is based on the start from
7080 * the caller
7081 */
7082 if (hole_em) {
7083 u64 calc_end = extent_map_end(hole_em);
7084
7085 if (calc_end <= start || (hole_em->start > end)) {
7086 free_extent_map(hole_em);
7087 hole_em = NULL;
7088 } else {
7089 hole_start = max(hole_em->start, start);
7090 hole_len = calc_end - hole_start;
7091 }
7092 }
7093 em->bdev = NULL;
7094 if (hole_em && range_start > hole_start) {
7095 /* our hole starts before our delalloc, so we
7096 * have to return just the parts of the hole
7097 * that go until the delalloc starts
7098 */
7099 em->len = min(hole_len,
7100 range_start - hole_start);
7101 em->start = hole_start;
7102 em->orig_start = hole_start;
7103 /*
7104 * don't adjust block start at all,
7105 * it is fixed at EXTENT_MAP_HOLE
7106 */
7107 em->block_start = hole_em->block_start;
7108 em->block_len = hole_len;
f9e4fb53
LB		 7109 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7110 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
ec29ed5b
CM		 7111 } else {
7112 em->start = range_start;
7113 em->len = found;
7114 em->orig_start = range_start;
7115 em->block_start = EXTENT_MAP_DELALLOC;
7116 em->block_len = found;
7117 }
7118 } else if (hole_em) {
7119 return hole_em;
7120 }
7121out:
7122
7123 free_extent_map(hole_em);
7124 if (err) {
7125 free_extent_map(em);
7126 return ERR_PTR(err);
7127 }
7128 return em;
7129}
7130
4b46fce2
JB		 7131static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7132 u64 start, u64 len)
7133{
7134 struct btrfs_root *root = BTRFS_I(inode)->root;
70c8a91c 7135 struct extent_map *em;
4b46fce2
JB		 7136 struct btrfs_key ins;
7137 u64 alloc_hint;
7138 int ret;
4b46fce2 7139
4b46fce2 7140 alloc_hint = get_extent_allocation_hint(inode, start, len);
00361589 7141 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
e570fd27 7142 alloc_hint, &ins, 1, 1);
00361589
JB		 7143 if (ret)
7144 return ERR_PTR(ret);
4b46fce2 7145
de0ee0ed
FM		 7146 /*
7147 * Create the ordered extent before the extent map. This is to avoid
7148 * races with the fast fsync path that would lead to it logging file
7149 * extent items that point to disk extents that were not yet written to.
7150 * The fast fsync path collects ordered extents into a local list and
7151 * then collects all the new extent maps, so we must create the ordered
7152 * extent first and make sure the fast fsync path collects any new
7153 * ordered extents after collecting new extent maps as well.
7154 * The fsync path simply can not rely on inode_dio_wait() because it
7155 * causes deadlock with AIO.
7156 */
4b46fce2
JB		 7157 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
7158 ins.offset, ins.offset, 0);
7159 if (ret) {
e570fd27 7160 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
00361589 7161 return ERR_PTR(ret);
4b46fce2 7162 }
00361589 7163
de0ee0ed
FM		 7164 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
7165 ins.offset, ins.offset, ins.offset, 0);
7166 if (IS_ERR(em)) {
7167 struct btrfs_ordered_extent *oe;
7168
7169 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7170 oe = btrfs_lookup_ordered_extent(inode, start);
7171 ASSERT(oe);
7172 if (WARN_ON(!oe))
7173 return em;
7174 set_bit(BTRFS_ORDERED_IOERR, &oe->flags);
7175 set_bit(BTRFS_ORDERED_IO_DONE, &oe->flags);
7176 btrfs_remove_ordered_extent(inode, oe);
7177 /* Once for our lookup and once for the ordered extents tree. */
7178 btrfs_put_ordered_extent(oe);
7179 btrfs_put_ordered_extent(oe);
7180 }
4b46fce2
JB		 7181 return em;
7182}
7183
46bfbb5c
CM		 7184/*
7185 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7186 * block must be cow'd
7187 */
00361589 7188noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7ee9e440
JB		 7189 u64 *orig_start, u64 *orig_block_len,
7190 u64 *ram_bytes)
46bfbb5c 7191{
00361589 7192 struct btrfs_trans_handle *trans;
46bfbb5c
CM		 7193 struct btrfs_path *path;
7194 int ret;
7195 struct extent_buffer *leaf;
7196 struct btrfs_root *root = BTRFS_I(inode)->root;
7b2b7085 7197 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
46bfbb5c
CM		 7198 struct btrfs_file_extent_item *fi;
7199 struct btrfs_key key;
7200 u64 disk_bytenr;
7201 u64 backref_offset;
7202 u64 extent_end;
7203 u64 num_bytes;
7204 int slot;
7205 int found_type;
7ee9e440 7206 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
e77751aa 7207
46bfbb5c
CM		 7208 path = btrfs_alloc_path();
7209 if (!path)
7210 return -ENOMEM;
7211
00361589 7212 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
46bfbb5c
CM		 7213 offset, 0);
7214 if (ret < 0)
7215 goto out;
7216
7217 slot = path->slots[0];
7218 if (ret == 1) {
7219 if (slot == 0) {
7220 /* can't find the item, must cow */
7221 ret = 0;
7222 goto out;
7223 }
7224 slot--;
7225 }
7226 ret = 0;
7227 leaf = path->nodes[0];
7228 btrfs_item_key_to_cpu(leaf, &key, slot);
33345d01 7229 if (key.objectid != btrfs_ino(inode) ||
46bfbb5c
CM		 7230 key.type != BTRFS_EXTENT_DATA_KEY) {
7231 /* not our file or wrong item type, must cow */
7232 goto out;
7233 }
7234
7235 if (key.offset > offset) {
7236 /* Wrong offset, must cow */
7237 goto out;
7238 }
7239
7240 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7241 found_type = btrfs_file_extent_type(leaf, fi);
7242 if (found_type != BTRFS_FILE_EXTENT_REG &&
7243 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7244 /* not a regular extent, must cow */
7245 goto out;
7246 }
7ee9e440
JB		 7247
7248 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7249 goto out;
7250
e77751aa
MX		 7251 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7252 if (extent_end <= offset)
7253 goto out;
7254
46bfbb5c 7255 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7ee9e440
JB		 7256 if (disk_bytenr == 0)
7257 goto out;
7258
7259 if (btrfs_file_extent_compression(leaf, fi) ||
7260 btrfs_file_extent_encryption(leaf, fi) ||
7261 btrfs_file_extent_other_encoding(leaf, fi))
7262 goto out;
7263
46bfbb5c
CM		 7264 backref_offset = btrfs_file_extent_offset(leaf, fi);
7265
7ee9e440
JB		 7266 if (orig_start) {
7267 *orig_start = key.offset - backref_offset;
7268 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7269 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7270 }
eb384b55 7271
46bfbb5c
CM		 7272 if (btrfs_extent_readonly(root, disk_bytenr))
7273 goto out;
7b2b7085
MX		 7274
7275 num_bytes = min(offset + *len, extent_end) - offset;
7276 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7277 u64 range_end;
7278
7279 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7280 ret = test_range_bit(io_tree, offset, range_end,
7281 EXTENT_DELALLOC, 0, NULL);
7282 if (ret) {
7283 ret = -EAGAIN;
7284 goto out;
7285 }
7286 }
7287
1bda19eb 7288 btrfs_release_path(path);
46bfbb5c
CM		 7289
7290 /*
7291 * look for other files referencing this extent, if we
7292 * find any we must cow
7293 */
00361589
JB		 7294 trans = btrfs_join_transaction(root);
7295 if (IS_ERR(trans)) {
7296 ret = 0;
46bfbb5c 7297 goto out;
00361589
JB		 7298 }
7299
7300 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7301 key.offset - backref_offset, disk_bytenr);
7302 btrfs_end_transaction(trans, root);
7303 if (ret) {
7304 ret = 0;
7305 goto out;
7306 }
46bfbb5c
CM		 7307
7308 /*
7309 * adjust disk_bytenr and num_bytes to cover just the bytes
7310 * in this extent we are about to write. If there
7311 * are any csums in that range we have to cow in order
7312 * to keep the csums correct
7313 */
7314 disk_bytenr += backref_offset;
7315 disk_bytenr += offset - key.offset;
46bfbb5c
CM		 7316 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7317 goto out;
7318 /*
7319 * all of the above have passed, it is safe to overwrite this extent
7320 * without cow
7321 */
eb384b55 7322 *len = num_bytes;
46bfbb5c
CM		 7323 ret = 1;
7324out:
7325 btrfs_free_path(path);
7326 return ret;
7327}
7328
fc4adbff
AG		 7329bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7330{
7331 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7332 int found = false;
7333 void **pagep = NULL;
7334 struct page *page = NULL;
7335 int start_idx;
7336 int end_idx;
7337
09cbfeaf 7338 start_idx = start >> PAGE_SHIFT;
fc4adbff
AG		 7339
7340 /*
7341 * end is the last byte in the last page. end == start is legal
7342 */
09cbfeaf 7343 end_idx = end >> PAGE_SHIFT;
fc4adbff
AG		 7344
7345 rcu_read_lock();
7346
7347 /* Most of the code in this while loop is lifted from
7348 * find_get_page. It's been modified to begin searching from a
7349 * page and return just the first page found in that range. If the
7350 * found idx is less than or equal to the end idx then we know that
7351 * a page exists. If no pages are found or if those pages are
7352 * outside of the range then we're fine (yay!) */
7353 while (page == NULL &&
7354 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7355 page = radix_tree_deref_slot(pagep);
7356 if (unlikely(!page))
7357 break;
7358
7359 if (radix_tree_exception(page)) {
809f9016
FM		 7360 if (radix_tree_deref_retry(page)) {
7361 page = NULL;
fc4adbff 7362 continue;
809f9016 7363 }
fc4adbff
AG		 7364 /*
7365 * Otherwise, shmem/tmpfs must be storing a swap entry
7366 * here as an exceptional entry: so return it without
7367 * attempting to raise page count.
7368 */
6fdef6d4 7369 page = NULL;
fc4adbff
AG		 7370 break; /* TODO: Is this relevant for this use case? */
7371 }
7372
91405151
FM		 7373 if (!page_cache_get_speculative(page)) {
7374 page = NULL;
fc4adbff 7375 continue;
91405151 7376 }
fc4adbff
AG		 7377
7378 /*
7379 * Has the page moved?
7380 * This is part of the lockless pagecache protocol. See
7381 * include/linux/pagemap.h for details.
7382 */
7383 if (unlikely(page != *pagep)) {
09cbfeaf 7384 put_page(page);
fc4adbff
AG		 7385 page = NULL;
7386 }
7387 }
7388
7389 if (page) {
7390 if (page->index <= end_idx)
7391 found = true;
09cbfeaf 7392 put_page(page);
fc4adbff
AG		 7393 }
7394
7395 rcu_read_unlock();
7396 return found;
7397}
7398
eb838e73
JB		 7399static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7400 struct extent_state **cached_state, int writing)
7401{
7402 struct btrfs_ordered_extent *ordered;
7403 int ret = 0;
7404
7405 while (1) {
7406 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
ff13db41 7407 cached_state);
eb838e73
JB		 7408 /*
7409 * We're concerned with the entire range that we're going to be
7410 * doing DIO to, so we need to make sure theres no ordered
7411 * extents in this range.
7412 */
7413 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7414 lockend - lockstart + 1);
7415
7416 /*
7417 * We need to make sure there are no buffered pages in this
7418 * range either, we could have raced between the invalidate in
7419 * generic_file_direct_write and locking the extent. The
7420 * invalidate needs to happen so that reads after a write do not
7421 * get stale data.
7422 */
fc4adbff
AG		 7423 if (!ordered &&
7424 (!writing ||
7425 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
eb838e73
JB		 7426 break;
7427
7428 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7429 cached_state, GFP_NOFS);
7430
7431 if (ordered) {
ade77029
FM		 7432 /*
7433 * If we are doing a DIO read and the ordered extent we
7434 * found is for a buffered write, we can not wait for it
7435 * to complete and retry, because if we do so we can
7436 * deadlock with concurrent buffered writes on page
7437 * locks. This happens only if our DIO read covers more
7438 * than one extent map, if at this point has already
7439 * created an ordered extent for a previous extent map
7440 * and locked its range in the inode's io tree, and a
7441 * concurrent write against that previous extent map's
7442 * range and this range started (we unlock the ranges
7443 * in the io tree only when the bios complete and
7444 * buffered writes always lock pages before attempting
7445 * to lock range in the io tree).
7446 */
7447 if (writing ||
7448 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7449 btrfs_start_ordered_extent(inode, ordered, 1);
7450 else
7451 ret = -ENOTBLK;
eb838e73
JB		 7452 btrfs_put_ordered_extent(ordered);
7453 } else {
eb838e73 7454 /*
b850ae14
FM		 7455 * We could trigger writeback for this range (and wait
7456 * for it to complete) and then invalidate the pages for
7457 * this range (through invalidate_inode_pages2_range()),
7458 * but that can lead us to a deadlock with a concurrent
7459 * call to readpages() (a buffered read or a defrag call
7460 * triggered a readahead) on a page lock due to an
7461 * ordered dio extent we created before but did not have
7462 * yet a corresponding bio submitted (whence it can not
7463 * complete), which makes readpages() wait for that
7464 * ordered extent to complete while holding a lock on
7465 * that page.
eb838e73 7466 */
b850ae14 7467 ret = -ENOTBLK;
eb838e73
JB		 7468 }
7469
ade77029
FM		 7470 if (ret)
7471 break;
7472
eb838e73
JB		 7473 cond_resched();
7474 }
7475
7476 return ret;
7477}
7478
69ffb543
JB		 7479static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7480 u64 len, u64 orig_start,
7481 u64 block_start, u64 block_len,
cc95bef6
JB		 7482 u64 orig_block_len, u64 ram_bytes,
7483 int type)
69ffb543
JB		 7484{
7485 struct extent_map_tree *em_tree;
7486 struct extent_map *em;
7487 struct btrfs_root *root = BTRFS_I(inode)->root;
7488 int ret;
7489
7490 em_tree = &BTRFS_I(inode)->extent_tree;
7491 em = alloc_extent_map();
7492 if (!em)
7493 return ERR_PTR(-ENOMEM);
7494
7495 em->start = start;
7496 em->orig_start = orig_start;
2ab28f32
JB		 7497 em->mod_start = start;
7498 em->mod_len = len;
69ffb543
JB		 7499 em->len = len;
7500 em->block_len = block_len;
7501 em->block_start = block_start;
7502 em->bdev = root->fs_info->fs_devices->latest_bdev;
b4939680 7503 em->orig_block_len = orig_block_len;
cc95bef6 7504 em->ram_bytes = ram_bytes;
70c8a91c 7505 em->generation = -1;
69ffb543
JB		 7506 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7507 if (type == BTRFS_ORDERED_PREALLOC)
b11e234d 7508 set_bit(EXTENT_FLAG_FILLING, &em->flags);
69ffb543
JB		 7509
7510 do {
7511 btrfs_drop_extent_cache(inode, em->start,
7512 em->start + em->len - 1, 0);
7513 write_lock(&em_tree->lock);
09a2a8f9 7514 ret = add_extent_mapping(em_tree, em, 1);
69ffb543
JB		 7515 write_unlock(&em_tree->lock);
7516 } while (ret == -EEXIST);
7517
7518 if (ret) {
7519 free_extent_map(em);
7520 return ERR_PTR(ret);
7521 }
7522
7523 return em;
7524}
7525
9c9464cc
FM		 7526static void adjust_dio_outstanding_extents(struct inode *inode,
7527 struct btrfs_dio_data *dio_data,
7528 const u64 len)
7529{
7530 unsigned num_extents;
7531
7532 num_extents = (unsigned) div64_u64(len + BTRFS_MAX_EXTENT_SIZE - 1,
7533 BTRFS_MAX_EXTENT_SIZE);
7534 /*
7535 * If we have an outstanding_extents count still set then we're
7536 * within our reservation, otherwise we need to adjust our inode
7537 * counter appropriately.
7538 */
7539 if (dio_data->outstanding_extents) {
7540 dio_data->outstanding_extents -= num_extents;
7541 } else {
7542 spin_lock(&BTRFS_I(inode)->lock);
7543 BTRFS_I(inode)->outstanding_extents += num_extents;
7544 spin_unlock(&BTRFS_I(inode)->lock);
7545 }
7546}
7547
4b46fce2
JB		 7548static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7549 struct buffer_head *bh_result, int create)
7550{
7551 struct extent_map *em;
7552 struct btrfs_root *root = BTRFS_I(inode)->root;
eb838e73 7553 struct extent_state *cached_state = NULL;
50745b0a 7554 struct btrfs_dio_data *dio_data = NULL;
4b46fce2 7555 u64 start = iblock << inode->i_blkbits;
eb838e73 7556 u64 lockstart, lockend;
4b46fce2 7557 u64 len = bh_result->b_size;
eb838e73 7558 int unlock_bits = EXTENT_LOCKED;
0934856d 7559 int ret = 0;
eb838e73 7560
172a5049 7561 if (create)
3266789f 7562 unlock_bits |= EXTENT_DIRTY;
172a5049 7563 else
c329861d 7564 len = min_t(u64, len, root->sectorsize);
eb838e73 7565
c329861d
JB		 7566 lockstart = start;
7567 lockend = start + len - 1;
7568
e1cbbfa5
JB		 7569 if (current->journal_info) {
7570 /*
7571 * Need to pull our outstanding extents and set journal_info to NULL so
7572 * that anything that needs to check if there's a transction doesn't get
7573 * confused.
7574 */
50745b0a 7575 dio_data = current->journal_info;
e1cbbfa5
JB		 7576 current->journal_info = NULL;
7577 }
7578
eb838e73
JB		 7579 /*
7580 * If this errors out it's because we couldn't invalidate pagecache for
7581 * this range and we need to fallback to buffered.
7582 */
9c9464cc
FM		 7583 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7584 create)) {
7585 ret = -ENOTBLK;
7586 goto err;
7587 }
eb838e73 7588
4b46fce2 7589 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
eb838e73
JB		 7590 if (IS_ERR(em)) {
7591 ret = PTR_ERR(em);
7592 goto unlock_err;
7593 }
4b46fce2
JB		 7594
7595 /*
7596 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7597 * io. INLINE is special, and we could probably kludge it in here, but
7598 * it's still buffered so for safety lets just fall back to the generic
7599 * buffered path.
7600 *
7601 * For COMPRESSED we _have_ to read the entire extent in so we can
7602 * decompress it, so there will be buffering required no matter what we
7603 * do, so go ahead and fallback to buffered.
7604 *
7605 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7606 * to buffered IO. Don't blame me, this is the price we pay for using
7607 * the generic code.
7608 */
7609 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7610 em->block_start == EXTENT_MAP_INLINE) {
7611 free_extent_map(em);
eb838e73
JB		 7612 ret = -ENOTBLK;
7613 goto unlock_err;
4b46fce2
JB		 7614 }
7615
7616 /* Just a good old fashioned hole, return */
7617 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7618 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7619 free_extent_map(em);
eb838e73 7620 goto unlock_err;
4b46fce2
JB		 7621 }
7622
7623 /*
7624 * We don't allocate a new extent in the following cases
7625 *
7626 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7627 * existing extent.
7628 * 2) The extent is marked as PREALLOC. We're good to go here and can
7629 * just use the extent.
7630 *
7631 */
46bfbb5c 7632 if (!create) {
eb838e73
JB		 7633 len = min(len, em->len - (start - em->start));
7634 lockstart = start + len;
7635 goto unlock;
46bfbb5c 7636 }
4b46fce2
JB		 7637
7638 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7639 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7640 em->block_start != EXTENT_MAP_HOLE)) {
4b46fce2 7641 int type;
eb384b55 7642 u64 block_start, orig_start, orig_block_len, ram_bytes;
4b46fce2
JB		 7643
7644 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7645 type = BTRFS_ORDERED_PREALLOC;
7646 else
7647 type = BTRFS_ORDERED_NOCOW;
46bfbb5c 7648 len = min(len, em->len - (start - em->start));
4b46fce2 7649 block_start = em->block_start + (start - em->start);
46bfbb5c 7650
00361589 7651 if (can_nocow_extent(inode, start, &len, &orig_start,
7ee9e440 7652 &orig_block_len, &ram_bytes) == 1) {
69ffb543
JB		 7653 if (type == BTRFS_ORDERED_PREALLOC) {
7654 free_extent_map(em);
7655 em = create_pinned_em(inode, start, len,
7656 orig_start,
b4939680 7657 block_start, len,
cc95bef6
JB		 7658 orig_block_len,
7659 ram_bytes, type);
555e1286
FM		 7660 if (IS_ERR(em)) {
7661 ret = PTR_ERR(em);
69ffb543 7662 goto unlock_err;
555e1286 7663 }
69ffb543
JB		 7664 }
7665
46bfbb5c
CM		 7666 ret = btrfs_add_ordered_extent_dio(inode, start,
7667 block_start, len, len, type);
46bfbb5c
CM		 7668 if (ret) {
7669 free_extent_map(em);
eb838e73 7670 goto unlock_err;
46bfbb5c
CM		 7671 }
7672 goto unlock;
4b46fce2 7673 }
4b46fce2 7674 }
00361589 7675
46bfbb5c
CM		 7676 /*
7677 * this will cow the extent, reset the len in case we changed
7678 * it above
7679 */
7680 len = bh_result->b_size;
70c8a91c
JB		 7681 free_extent_map(em);
7682 em = btrfs_new_extent_direct(inode, start, len);
eb838e73
JB		 7683 if (IS_ERR(em)) {
7684 ret = PTR_ERR(em);
7685 goto unlock_err;
7686 }
46bfbb5c
CM		 7687 len = min(len, em->len - (start - em->start));
7688unlock:
4b46fce2
JB		 7689 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7690 inode->i_blkbits;
46bfbb5c 7691 bh_result->b_size = len;
4b46fce2
JB		 7692 bh_result->b_bdev = em->bdev;
7693 set_buffer_mapped(bh_result);
c3473e83
JB		 7694 if (create) {
7695 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7696 set_buffer_new(bh_result);
7697
7698 /*
7699 * Need to update the i_size under the extent lock so buffered
7700 * readers will get the updated i_size when we unlock.
7701 */
7702 if (start + len > i_size_read(inode))
7703 i_size_write(inode, start + len);
0934856d 7704
9c9464cc 7705 adjust_dio_outstanding_extents(inode, dio_data, len);
7cf5b976 7706 btrfs_free_reserved_data_space(inode, start, len);
50745b0a 7707 WARN_ON(dio_data->reserve < len);
7708 dio_data->reserve -= len;
f28a4928 7709 dio_data->unsubmitted_oe_range_end = start + len;
50745b0a 7710 current->journal_info = dio_data;
c3473e83 7711 }
4b46fce2 7712
eb838e73
JB		 7713 /*
7714 * In the case of write we need to clear and unlock the entire range,
7715 * in the case of read we need to unlock only the end area that we
7716 * aren't using if there is any left over space.
7717 */
24c03fa5 7718 if (lockstart < lockend) {
0934856d
MX		 7719 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7720 lockend, unlock_bits, 1, 0,
7721 &cached_state, GFP_NOFS);
24c03fa5 7722 } else {
eb838e73 7723 free_extent_state(cached_state);
24c03fa5 7724 }
eb838e73 7725
4b46fce2
JB		 7726 free_extent_map(em);
7727
7728 return 0;
eb838e73
JB		 7729
7730unlock_err:
eb838e73
JB		 7731 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7732 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
9c9464cc 7733err:
50745b0a 7734 if (dio_data)
7735 current->journal_info = dio_data;
9c9464cc
FM		 7736 /*
7737 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7738 * write less data then expected, so that we don't underflow our inode's
7739 * outstanding extents counter.
7740 */
7741 if (create && dio_data)
7742 adjust_dio_outstanding_extents(inode, dio_data, len);
7743
eb838e73 7744 return ret;
4b46fce2
JB		 7745}
7746
8b110e39
MX		 7747static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7748 int rw, int mirror_num)
7749{
7750 struct btrfs_root *root = BTRFS_I(inode)->root;
7751 int ret;
7752
7753 BUG_ON(rw & REQ_WRITE);
7754
7755 bio_get(bio);
7756
7757 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7758 BTRFS_WQ_ENDIO_DIO_REPAIR);
7759 if (ret)
7760 goto err;
7761
7762 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7763err:
7764 bio_put(bio);
7765 return ret;
7766}
7767
7768static int btrfs_check_dio_repairable(struct inode *inode,
7769 struct bio *failed_bio,
7770 struct io_failure_record *failrec,
7771 int failed_mirror)
7772{
7773 int num_copies;
7774
7775 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7776 failrec->logical, failrec->len);
7777 if (num_copies == 1) {
7778 /*
7779 * we only have a single copy of the data, so don't bother with
7780 * all the retry and error correction code that follows. no
7781 * matter what the error is, it is very likely to persist.
7782 */
7783 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7784 num_copies, failrec->this_mirror, failed_mirror);
7785 return 0;
7786 }
7787
7788 failrec->failed_mirror = failed_mirror;
7789 failrec->this_mirror++;
7790 if (failrec->this_mirror == failed_mirror)
7791 failrec->this_mirror++;
7792
7793 if (failrec->this_mirror > num_copies) {
7794 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7795 num_copies, failrec->this_mirror, failed_mirror);
7796 return 0;
7797 }
7798
7799 return 1;
7800}
7801
7802static int dio_read_error(struct inode *inode, struct bio *failed_bio,
2dabb324
CR		 7803 struct page *page, unsigned int pgoff,
7804 u64 start, u64 end, int failed_mirror,
7805 bio_end_io_t *repair_endio, void *repair_arg)
8b110e39
MX		 7806{
7807 struct io_failure_record *failrec;
7808 struct bio *bio;
7809 int isector;
7810 int read_mode;
7811 int ret;
7812
7813 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7814
7815 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7816 if (ret)
7817 return ret;
7818
7819 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7820 failed_mirror);
7821 if (!ret) {
7822 free_io_failure(inode, failrec);
7823 return -EIO;
7824 }
7825
2dabb324
CR		 7826 if ((failed_bio->bi_vcnt > 1)
7827 || (failed_bio->bi_io_vec->bv_len
7828 > BTRFS_I(inode)->root->sectorsize))
8b110e39
MX		 7829 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7830 else
7831 read_mode = READ_SYNC;
7832
7833 isector = start - btrfs_io_bio(failed_bio)->logical;
7834 isector >>= inode->i_sb->s_blocksize_bits;
7835 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
2dabb324 7836 pgoff, isector, repair_endio, repair_arg);
8b110e39
MX		 7837 if (!bio) {
7838 free_io_failure(inode, failrec);
7839 return -EIO;
7840 }
7841
7842 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7843 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7844 read_mode, failrec->this_mirror, failrec->in_validation);
7845
7846 ret = submit_dio_repair_bio(inode, bio, read_mode,
7847 failrec->this_mirror);
7848 if (ret) {
7849 free_io_failure(inode, failrec);
7850 bio_put(bio);
7851 }
7852
7853 return ret;
7854}
7855
7856struct btrfs_retry_complete {
7857 struct completion done;
7858 struct inode *inode;
7859 u64 start;
7860 int uptodate;
7861};
7862
4246a0b6 7863static void btrfs_retry_endio_nocsum(struct bio *bio)
8b110e39
MX		 7864{
7865 struct btrfs_retry_complete *done = bio->bi_private;
2dabb324 7866 struct inode *inode;
8b110e39
MX		 7867 struct bio_vec *bvec;
7868 int i;
7869
4246a0b6 7870 if (bio->bi_error)
8b110e39
MX		 7871 goto end;
7872
2dabb324
CR		 7873 ASSERT(bio->bi_vcnt == 1);
7874 inode = bio->bi_io_vec->bv_page->mapping->host;
7875 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
7876
8b110e39
MX		 7877 done->uptodate = 1;
7878 bio_for_each_segment_all(bvec, bio, i)
7879 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7880end:
7881 complete(&done->done);
7882 bio_put(bio);
7883}
7884
7885static int __btrfs_correct_data_nocsum(struct inode *inode,
7886 struct btrfs_io_bio *io_bio)
4b46fce2 7887{
2dabb324 7888 struct btrfs_fs_info *fs_info;
2c30c71b 7889 struct bio_vec *bvec;
8b110e39 7890 struct btrfs_retry_complete done;
4b46fce2 7891 u64 start;
2dabb324
CR		 7892 unsigned int pgoff;
7893 u32 sectorsize;
7894 int nr_sectors;
2c30c71b 7895 int i;
c1dc0896 7896 int ret;
4b46fce2 7897
2dabb324
CR		 7898 fs_info = BTRFS_I(inode)->root->fs_info;
7899 sectorsize = BTRFS_I(inode)->root->sectorsize;
7900
8b110e39
MX		 7901 start = io_bio->logical;
7902 done.inode = inode;
7903
7904 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
2dabb324
CR		 7905 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
7906 pgoff = bvec->bv_offset;
7907
7908next_block_or_try_again:
8b110e39
MX		 7909 done.uptodate = 0;
7910 done.start = start;
7911 init_completion(&done.done);
7912
2dabb324
CR		 7913 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
7914 pgoff, start, start + sectorsize - 1,
7915 io_bio->mirror_num,
7916 btrfs_retry_endio_nocsum, &done);
8b110e39
MX		 7917 if (ret)
7918 return ret;
7919
7920 wait_for_completion(&done.done);
7921
7922 if (!done.uptodate) {
7923 /* We might have another mirror, so try again */
2dabb324 7924 goto next_block_or_try_again;
8b110e39
MX		 7925 }
7926
2dabb324
CR		 7927 start += sectorsize;
7928
7929 if (nr_sectors--) {
7930 pgoff += sectorsize;
7931 goto next_block_or_try_again;
7932 }
8b110e39
MX		 7933 }
7934
7935 return 0;
7936}
7937
4246a0b6 7938static void btrfs_retry_endio(struct bio *bio)
8b110e39
MX		 7939{
7940 struct btrfs_retry_complete *done = bio->bi_private;
7941 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2dabb324 7942 struct inode *inode;
8b110e39 7943 struct bio_vec *bvec;
2dabb324 7944 u64 start;
8b110e39
MX		 7945 int uptodate;
7946 int ret;
7947 int i;
7948
4246a0b6 7949 if (bio->bi_error)
8b110e39
MX		 7950 goto end;
7951
7952 uptodate = 1;
2dabb324
CR		 7953
7954 start = done->start;
7955
7956 ASSERT(bio->bi_vcnt == 1);
7957 inode = bio->bi_io_vec->bv_page->mapping->host;
7958 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
7959
8b110e39
MX		 7960 bio_for_each_segment_all(bvec, bio, i) {
7961 ret = __readpage_endio_check(done->inode, io_bio, i,
2dabb324
CR		 7962 bvec->bv_page, bvec->bv_offset,
7963 done->start, bvec->bv_len);
8b110e39
MX		 7964 if (!ret)
7965 clean_io_failure(done->inode, done->start,
2dabb324 7966 bvec->bv_page, bvec->bv_offset);
8b110e39
MX		 7967 else
7968 uptodate = 0;
7969 }
7970
7971 done->uptodate = uptodate;
7972end:
7973 complete(&done->done);
7974 bio_put(bio);
7975}
7976
7977static int __btrfs_subio_endio_read(struct inode *inode,
7978 struct btrfs_io_bio *io_bio, int err)
7979{
2dabb324 7980 struct btrfs_fs_info *fs_info;
8b110e39
MX		 7981 struct bio_vec *bvec;
7982 struct btrfs_retry_complete done;
7983 u64 start;
7984 u64 offset = 0;
2dabb324
CR		 7985 u32 sectorsize;
7986 int nr_sectors;
7987 unsigned int pgoff;
7988 int csum_pos;
8b110e39
MX		 7989 int i;
7990 int ret;
dc380aea 7991
2dabb324
CR		 7992 fs_info = BTRFS_I(inode)->root->fs_info;
7993 sectorsize = BTRFS_I(inode)->root->sectorsize;
7994
8b110e39 7995 err = 0;
c1dc0896 7996 start = io_bio->logical;
8b110e39
MX		 7997 done.inode = inode;
7998
c1dc0896 7999 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
2dabb324
CR		 8000 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
8001
8002 pgoff = bvec->bv_offset;
8003next_block:
8004 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8005 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8006 bvec->bv_page, pgoff, start,
8007 sectorsize);
8b110e39
MX		 8008 if (likely(!ret))
8009 goto next;
8010try_again:
8011 done.uptodate = 0;
8012 done.start = start;
8013 init_completion(&done.done);
8014
2dabb324
CR		 8015 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
8016 pgoff, start, start + sectorsize - 1,
8017 io_bio->mirror_num,
8018 btrfs_retry_endio, &done);
8b110e39
MX		 8019 if (ret) {
8020 err = ret;
8021 goto next;
8022 }
8023
8024 wait_for_completion(&done.done);
8025
8026 if (!done.uptodate) {
8027 /* We might have another mirror, so try again */
8028 goto try_again;
8029 }
8030next:
2dabb324
CR		 8031 offset += sectorsize;
8032 start += sectorsize;
8033
8034 ASSERT(nr_sectors);
8035
8036 if (--nr_sectors) {
8037 pgoff += sectorsize;
8038 goto next_block;
8039 }
2c30c71b 8040 }
c1dc0896
MX		 8041
8042 return err;
8043}
8044
8b110e39
MX		 8045static int btrfs_subio_endio_read(struct inode *inode,
8046 struct btrfs_io_bio *io_bio, int err)
8047{
8048 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8049
8050 if (skip_csum) {
8051 if (unlikely(err))
8052 return __btrfs_correct_data_nocsum(inode, io_bio);
8053 else
8054 return 0;
8055 } else {
8056 return __btrfs_subio_endio_read(inode, io_bio, err);
8057 }
8058}
8059
4246a0b6 8060static void btrfs_endio_direct_read(struct bio *bio)
c1dc0896
MX		 8061{
8062 struct btrfs_dio_private *dip = bio->bi_private;
8063 struct inode *inode = dip->inode;
8064 struct bio *dio_bio;
8065 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
4246a0b6 8066 int err = bio->bi_error;
c1dc0896 8067
8b110e39
MX		 8068 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8069 err = btrfs_subio_endio_read(inode, io_bio, err);
c1dc0896 8070
4b46fce2 8071 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
d0082371 8072 dip->logical_offset + dip->bytes - 1);
9be3395b 8073 dio_bio = dip->dio_bio;
4b46fce2 8074
4b46fce2 8075 kfree(dip);
c0da7aa1 8076
1636d1d7 8077 dio_bio->bi_error = bio->bi_error;
4246a0b6 8078 dio_end_io(dio_bio, bio->bi_error);
23ea8e5a
MX		 8079
8080 if (io_bio->end_io)
8081 io_bio->end_io(io_bio, err);
9be3395b 8082 bio_put(bio);
4b46fce2
JB		 8083}
8084
14543774
FM		 8085static void btrfs_endio_direct_write_update_ordered(struct inode *inode,
8086 const u64 offset,
8087 const u64 bytes,
8088 const int uptodate)
4b46fce2 8089{
4b46fce2 8090 struct btrfs_root *root = BTRFS_I(inode)->root;
4b46fce2 8091 struct btrfs_ordered_extent *ordered = NULL;
14543774
FM		 8092 u64 ordered_offset = offset;
8093 u64 ordered_bytes = bytes;
4b46fce2
JB		 8094 int ret;
8095
163cf09c
CM		 8096again:
8097 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8098 &ordered_offset,
4246a0b6 8099 ordered_bytes,
14543774 8100 uptodate);
4b46fce2 8101 if (!ret)
163cf09c 8102 goto out_test;
4b46fce2 8103
9e0af237
LB		 8104 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8105 finish_ordered_fn, NULL, NULL);
fccb5d86
QW		 8106 btrfs_queue_work(root->fs_info->endio_write_workers,
8107 &ordered->work);
163cf09c
CM		 8108out_test:
8109 /*
8110 * our bio might span multiple ordered extents. If we haven't
8111 * completed the accounting for the whole dio, go back and try again
8112 */
14543774
FM		 8113 if (ordered_offset < offset + bytes) {
8114 ordered_bytes = offset + bytes - ordered_offset;
5fd02043 8115 ordered = NULL;
163cf09c
CM		 8116 goto again;
8117 }
14543774
FM		 8118}
8119
8120static void btrfs_endio_direct_write(struct bio *bio)
8121{
8122 struct btrfs_dio_private *dip = bio->bi_private;
8123 struct bio *dio_bio = dip->dio_bio;
8124
8125 btrfs_endio_direct_write_update_ordered(dip->inode,
8126 dip->logical_offset,
8127 dip->bytes,
8128 !bio->bi_error);
4b46fce2 8129
4b46fce2 8130 kfree(dip);
c0da7aa1 8131
1636d1d7 8132 dio_bio->bi_error = bio->bi_error;
4246a0b6 8133 dio_end_io(dio_bio, bio->bi_error);
9be3395b 8134 bio_put(bio);
4b46fce2
JB		 8135}
8136
eaf25d93
CM		 8137static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
8138 struct bio *bio, int mirror_num,
8139 unsigned long bio_flags, u64 offset)
8140{
8141 int ret;
8142 struct btrfs_root *root = BTRFS_I(inode)->root;
8143 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
79787eaa 8144 BUG_ON(ret); /* -ENOMEM */
eaf25d93
CM		 8145 return 0;
8146}
8147
4246a0b6 8148static void btrfs_end_dio_bio(struct bio *bio)
e65e1535
MX		 8149{
8150 struct btrfs_dio_private *dip = bio->bi_private;
4246a0b6 8151 int err = bio->bi_error;
e65e1535 8152
8b110e39
MX		 8153 if (err)
8154 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8155 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
8156 btrfs_ino(dip->inode), bio->bi_rw,
8157 (unsigned long long)bio->bi_iter.bi_sector,
8158 bio->bi_iter.bi_size, err);
8159
8160 if (dip->subio_endio)
8161 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
c1dc0896
MX		 8162
8163 if (err) {
e65e1535
MX		 8164 dip->errors = 1;
8165
8166 /*
8167 * before atomic variable goto zero, we must make sure
8168 * dip->errors is perceived to be set.
8169 */
4e857c58 8170 smp_mb__before_atomic();
e65e1535
MX		 8171 }
8172
8173 /* if there are more bios still pending for this dio, just exit */
8174 if (!atomic_dec_and_test(&dip->pending_bios))
8175 goto out;
8176
9be3395b 8177 if (dip->errors) {
e65e1535 8178 bio_io_error(dip->orig_bio);
9be3395b 8179 } else {
4246a0b6
CH		 8180 dip->dio_bio->bi_error = 0;
8181 bio_endio(dip->orig_bio);
e65e1535
MX		 8182 }
8183out:
8184 bio_put(bio);
8185}
8186
8187static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8188 u64 first_sector, gfp_t gfp_flags)
8189{
da2f0f74 8190 struct bio *bio;
22365979 8191 bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
da2f0f74
CM		 8192 if (bio)
8193 bio_associate_current(bio);
8194 return bio;
e65e1535
MX		 8195}
8196
c1dc0896
MX		 8197static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
8198 struct inode *inode,
8199 struct btrfs_dio_private *dip,
8200 struct bio *bio,
8201 u64 file_offset)
8202{
8203 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8204 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8205 int ret;
8206
8207 /*
8208 * We load all the csum data we need when we submit
8209 * the first bio to reduce the csum tree search and
8210 * contention.
8211 */
8212 if (dip->logical_offset == file_offset) {
8213 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
8214 file_offset);
8215 if (ret)
8216 return ret;
8217 }
8218
8219 if (bio == dip->orig_bio)
8220 return 0;
8221
8222 file_offset -= dip->logical_offset;
8223 file_offset >>= inode->i_sb->s_blocksize_bits;
8224 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8225
8226 return 0;
8227}
8228
e65e1535
MX		 8229static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8230 int rw, u64 file_offset, int skip_sum,
c329861d 8231 int async_submit)
e65e1535 8232{
facc8a22 8233 struct btrfs_dio_private *dip = bio->bi_private;
e65e1535
MX		 8234 int write = rw & REQ_WRITE;
8235 struct btrfs_root *root = BTRFS_I(inode)->root;
8236 int ret;
8237
b812ce28
JB		 8238 if (async_submit)
8239 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8240
e65e1535 8241 bio_get(bio);
5fd02043
JB		 8242
8243 if (!write) {
bfebd8b5
DS		 8244 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8245 BTRFS_WQ_ENDIO_DATA);
5fd02043
JB		 8246 if (ret)
8247 goto err;
8248 }
e65e1535 8249
1ae39938
JB		 8250 if (skip_sum)
8251 goto map;
8252
8253 if (write && async_submit) {
e65e1535
MX		 8254 ret = btrfs_wq_submit_bio(root->fs_info,
8255 inode, rw, bio, 0, 0,
8256 file_offset,
8257 __btrfs_submit_bio_start_direct_io,
8258 __btrfs_submit_bio_done);
8259 goto err;
1ae39938
JB		 8260 } else if (write) {
8261 /*
8262 * If we aren't doing async submit, calculate the csum of the
8263 * bio now.
8264 */
8265 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8266 if (ret)
8267 goto err;
23ea8e5a 8268 } else {
c1dc0896
MX		 8269 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8270 file_offset);
c2db1073
TI		 8271 if (ret)
8272 goto err;
8273 }
1ae39938
JB		 8274map:
8275 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
e65e1535
MX		 8276err:
8277 bio_put(bio);
8278 return ret;
8279}
8280
8281static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
8282 int skip_sum)
8283{
8284 struct inode *inode = dip->inode;
8285 struct btrfs_root *root = BTRFS_I(inode)->root;
e65e1535
MX		 8286 struct bio *bio;
8287 struct bio *orig_bio = dip->orig_bio;
8288 struct bio_vec *bvec = orig_bio->bi_io_vec;
4f024f37 8289 u64 start_sector = orig_bio->bi_iter.bi_sector;
e65e1535
MX		 8290 u64 file_offset = dip->logical_offset;
8291 u64 submit_len = 0;
8292 u64 map_length;
5f4dc8fc 8293 u32 blocksize = root->sectorsize;
1ae39938 8294 int async_submit = 0;
5f4dc8fc
CR		 8295 int nr_sectors;
8296 int ret;
8297 int i;
e65e1535 8298
4f024f37 8299 map_length = orig_bio->bi_iter.bi_size;
53b381b3 8300 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
e65e1535 8301 &map_length, NULL, 0);
7a5c3c9b 8302 if (ret)
e65e1535 8303 return -EIO;
facc8a22 8304
4f024f37 8305 if (map_length >= orig_bio->bi_iter.bi_size) {
02f57c7a 8306 bio = orig_bio;
c1dc0896 8307 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
02f57c7a
JB		 8308 goto submit;
8309 }
8310
53b381b3 8311 /* async crcs make it difficult to collect full stripe writes. */
ffe2d203 8312 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
53b381b3
DW		 8313 async_submit = 0;
8314 else
8315 async_submit = 1;
8316
02f57c7a
JB		 8317 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8318 if (!bio)
8319 return -ENOMEM;
7a5c3c9b 8320
02f57c7a
JB		 8321 bio->bi_private = dip;
8322 bio->bi_end_io = btrfs_end_dio_bio;
c1dc0896 8323 btrfs_io_bio(bio)->logical = file_offset;
02f57c7a
JB		 8324 atomic_inc(&dip->pending_bios);
8325
e65e1535 8326 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5f4dc8fc
CR		 8327 nr_sectors = BTRFS_BYTES_TO_BLKS(root->fs_info, bvec->bv_len);
8328 i = 0;
8329next_block:
8330 if (unlikely(map_length < submit_len + blocksize ||
8331 bio_add_page(bio, bvec->bv_page, blocksize,
8332 bvec->bv_offset + (i * blocksize)) < blocksize)) {
e65e1535
MX		 8333 /*
8334 * inc the count before we submit the bio so
8335 * we know the end IO handler won't happen before
8336 * we inc the count. Otherwise, the dip might get freed
8337 * before we're done setting it up
8338 */
8339 atomic_inc(&dip->pending_bios);
8340 ret = __btrfs_submit_dio_bio(bio, inode, rw,
8341 file_offset, skip_sum,
c329861d 8342 async_submit);
e65e1535
MX		 8343 if (ret) {
8344 bio_put(bio);
8345 atomic_dec(&dip->pending_bios);
8346 goto out_err;
8347 }
8348
e65e1535
MX		 8349 start_sector += submit_len >> 9;
8350 file_offset += submit_len;
8351
8352 submit_len = 0;
e65e1535
MX		 8353
8354 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8355 start_sector, GFP_NOFS);
8356 if (!bio)
8357 goto out_err;
8358 bio->bi_private = dip;
8359 bio->bi_end_io = btrfs_end_dio_bio;
c1dc0896 8360 btrfs_io_bio(bio)->logical = file_offset;
e65e1535 8361
4f024f37 8362 map_length = orig_bio->bi_iter.bi_size;
53b381b3 8363 ret = btrfs_map_block(root->fs_info, rw,
3ec706c8 8364 start_sector << 9,
e65e1535
MX		 8365 &map_length, NULL, 0);
8366 if (ret) {
8367 bio_put(bio);
8368 goto out_err;
8369 }
5f4dc8fc
CR		 8370
8371 goto next_block;
e65e1535 8372 } else {
5f4dc8fc
CR		 8373 submit_len += blocksize;
8374 if (--nr_sectors) {
8375 i++;
8376 goto next_block;
8377 }
e65e1535
MX		 8378 bvec++;
8379 }
8380 }
8381
02f57c7a 8382submit:
e65e1535 8383 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
c329861d 8384 async_submit);
e65e1535
MX		 8385 if (!ret)
8386 return 0;
8387
8388 bio_put(bio);
8389out_err:
8390 dip->errors = 1;
8391 /*
8392 * before atomic variable goto zero, we must
8393 * make sure dip->errors is perceived to be set.
8394 */
4e857c58 8395 smp_mb__before_atomic();
e65e1535
MX		 8396 if (atomic_dec_and_test(&dip->pending_bios))
8397 bio_io_error(dip->orig_bio);
8398
8399 /* bio_end_io() will handle error, so we needn't return it */
8400 return 0;
8401}
8402
9be3395b
CM		 8403static void btrfs_submit_direct(int rw, struct bio *dio_bio,
8404 struct inode *inode, loff_t file_offset)
4b46fce2 8405{
61de718f
FM		 8406 struct btrfs_dio_private *dip = NULL;
8407 struct bio *io_bio = NULL;
23ea8e5a 8408 struct btrfs_io_bio *btrfs_bio;
4b46fce2 8409 int skip_sum;
7b6d91da 8410 int write = rw & REQ_WRITE;
4b46fce2
JB		 8411 int ret = 0;
8412
8413 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8414
9be3395b 8415 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
9be3395b
CM		 8416 if (!io_bio) {
8417 ret = -ENOMEM;
8418 goto free_ordered;
8419 }
8420
c1dc0896 8421 dip = kzalloc(sizeof(*dip), GFP_NOFS);
4b46fce2
JB		 8422 if (!dip) {
8423 ret = -ENOMEM;
61de718f 8424 goto free_ordered;
4b46fce2 8425 }
4b46fce2 8426
9be3395b 8427 dip->private = dio_bio->bi_private;
4b46fce2
JB		 8428 dip->inode = inode;
8429 dip->logical_offset = file_offset;
4f024f37
KO		 8430 dip->bytes = dio_bio->bi_iter.bi_size;
8431 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
9be3395b 8432 io_bio->bi_private = dip;
9be3395b
CM		 8433 dip->orig_bio = io_bio;
8434 dip->dio_bio = dio_bio;
e65e1535 8435 atomic_set(&dip->pending_bios, 0);
c1dc0896
MX		 8436 btrfs_bio = btrfs_io_bio(io_bio);
8437 btrfs_bio->logical = file_offset;
4b46fce2 8438
c1dc0896 8439 if (write) {
9be3395b 8440 io_bio->bi_end_io = btrfs_endio_direct_write;
c1dc0896 8441 } else {
9be3395b 8442 io_bio->bi_end_io = btrfs_endio_direct_read;
c1dc0896
MX		 8443 dip->subio_endio = btrfs_subio_endio_read;
8444 }
4b46fce2 8445
f28a4928
FM		 8446 /*
8447 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8448 * even if we fail to submit a bio, because in such case we do the
8449 * corresponding error handling below and it must not be done a second
8450 * time by btrfs_direct_IO().
8451 */
8452 if (write) {
8453 struct btrfs_dio_data *dio_data = current->journal_info;
8454
8455 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8456 dip->bytes;
8457 dio_data->unsubmitted_oe_range_start =
8458 dio_data->unsubmitted_oe_range_end;
8459 }
8460
e65e1535
MX		 8461 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8462 if (!ret)
eaf25d93 8463 return;
9be3395b 8464
23ea8e5a
MX		 8465 if (btrfs_bio->end_io)
8466 btrfs_bio->end_io(btrfs_bio, ret);
9be3395b 8467
4b46fce2
JB		 8468free_ordered:
8469 /*
61de718f
FM		 8470 * If we arrived here it means either we failed to submit the dip
8471 * or we either failed to clone the dio_bio or failed to allocate the
8472 * dip. If we cloned the dio_bio and allocated the dip, we can just
8473 * call bio_endio against our io_bio so that we get proper resource
8474 * cleanup if we fail to submit the dip, otherwise, we must do the
8475 * same as btrfs_endio_direct_[write|read] because we can't call these
8476 * callbacks - they require an allocated dip and a clone of dio_bio.
4b46fce2 8477 */
61de718f 8478 if (io_bio && dip) {
4246a0b6
CH		 8479 io_bio->bi_error = -EIO;
8480 bio_endio(io_bio);
61de718f
FM		 8481 /*
8482 * The end io callbacks free our dip, do the final put on io_bio
8483 * and all the cleanup and final put for dio_bio (through
8484 * dio_end_io()).
8485 */
8486 dip = NULL;
8487 io_bio = NULL;
8488 } else {
14543774
FM		 8489 if (write)
8490 btrfs_endio_direct_write_update_ordered(inode,
8491 file_offset,
8492 dio_bio->bi_iter.bi_size,
8493 0);
8494 else
61de718f
FM		 8495 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8496 file_offset + dio_bio->bi_iter.bi_size - 1);
14543774 8497
4246a0b6 8498 dio_bio->bi_error = -EIO;
61de718f
FM		 8499 /*
8500 * Releases and cleans up our dio_bio, no need to bio_put()
8501 * nor bio_endio()/bio_io_error() against dio_bio.
8502 */
8503 dio_end_io(dio_bio, ret);
4b46fce2 8504 }
61de718f
FM		 8505 if (io_bio)
8506 bio_put(io_bio);
8507 kfree(dip);
4b46fce2
JB		 8508}
8509
6f673763 8510static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
28060d5d 8511 const struct iov_iter *iter, loff_t offset)
5a5f79b5
CM		 8512{
8513 int seg;
a1b75f7d 8514 int i;
5a5f79b5
CM		 8515 unsigned blocksize_mask = root->sectorsize - 1;
8516 ssize_t retval = -EINVAL;
5a5f79b5
CM		 8517
8518 if (offset & blocksize_mask)
8519 goto out;
8520
28060d5d
AV		 8521 if (iov_iter_alignment(iter) & blocksize_mask)
8522 goto out;
a1b75f7d 8523
28060d5d 8524 /* If this is a write we don't need to check anymore */
6f673763 8525 if (iov_iter_rw(iter) == WRITE)
28060d5d
AV		 8526 return 0;
8527 /*
8528 * Check to make sure we don't have duplicate iov_base's in this
8529 * iovec, if so return EINVAL, otherwise we'll get csum errors
8530 * when reading back.
8531 */
8532 for (seg = 0; seg < iter->nr_segs; seg++) {
8533 for (i = seg + 1; i < iter->nr_segs; i++) {
8534 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
a1b75f7d
JB		 8535 goto out;
8536 }
5a5f79b5
CM		 8537 }
8538 retval = 0;
8539out:
8540 return retval;
8541}
eb838e73 8542
22c6186e
OS		 8543static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
8544 loff_t offset)
16432985 8545{
4b46fce2
JB		 8546 struct file *file = iocb->ki_filp;
8547 struct inode *inode = file->f_mapping->host;
50745b0a 8548 struct btrfs_root *root = BTRFS_I(inode)->root;
8549 struct btrfs_dio_data dio_data = { 0 };
0934856d 8550 size_t count = 0;
2e60a51e 8551 int flags = 0;
38851cc1
MX		 8552 bool wakeup = true;
8553 bool relock = false;
0934856d 8554 ssize_t ret;
4b46fce2 8555
6f673763 8556 if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
5a5f79b5 8557 return 0;
3f7c579c 8558
fe0f07d0 8559 inode_dio_begin(inode);
4e857c58 8560 smp_mb__after_atomic();
38851cc1 8561
0e267c44 8562 /*
41bd9ca4
MX		 8563 * The generic stuff only does filemap_write_and_wait_range, which
8564 * isn't enough if we've written compressed pages to this area, so
8565 * we need to flush the dirty pages again to make absolutely sure
8566 * that any outstanding dirty pages are on disk.
0e267c44 8567 */
a6cbcd4a 8568 count = iov_iter_count(iter);
41bd9ca4
MX		 8569 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8570 &BTRFS_I(inode)->runtime_flags))
9a025a08
WS		 8571 filemap_fdatawrite_range(inode->i_mapping, offset,
8572 offset + count - 1);
0e267c44 8573
6f673763 8574 if (iov_iter_rw(iter) == WRITE) {
38851cc1
MX		 8575 /*
8576 * If the write DIO is beyond the EOF, we need update
8577 * the isize, but it is protected by i_mutex. So we can
8578 * not unlock the i_mutex at this case.
8579 */
8580 if (offset + count <= inode->i_size) {
5955102c 8581 inode_unlock(inode);
38851cc1
MX		 8582 relock = true;
8583 }
7cf5b976 8584 ret = btrfs_delalloc_reserve_space(inode, offset, count);
0934856d 8585 if (ret)
38851cc1 8586 goto out;
50745b0a 8587 dio_data.outstanding_extents = div64_u64(count +
e1cbbfa5
JB		 8588 BTRFS_MAX_EXTENT_SIZE - 1,
8589 BTRFS_MAX_EXTENT_SIZE);
8590
8591 /*
8592 * We need to know how many extents we reserved so that we can
8593 * do the accounting properly if we go over the number we
8594 * originally calculated. Abuse current->journal_info for this.
8595 */
50745b0a 8596 dio_data.reserve = round_up(count, root->sectorsize);
f28a4928
FM		 8597 dio_data.unsubmitted_oe_range_start = (u64)offset;
8598 dio_data.unsubmitted_oe_range_end = (u64)offset;
50745b0a 8599 current->journal_info = &dio_data;
ee39b432
DS		 8600 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8601 &BTRFS_I(inode)->runtime_flags)) {
fe0f07d0 8602 inode_dio_end(inode);
38851cc1
MX		 8603 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8604 wakeup = false;
0934856d
MX		 8605 }
8606
17f8c842
OS		 8607 ret = __blockdev_direct_IO(iocb, inode,
8608 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8609 iter, offset, btrfs_get_blocks_direct, NULL,
8610 btrfs_submit_direct, flags);
6f673763 8611 if (iov_iter_rw(iter) == WRITE) {
e1cbbfa5 8612 current->journal_info = NULL;
ddba1bfc 8613 if (ret < 0 && ret != -EIOCBQUEUED) {
50745b0a 8614 if (dio_data.reserve)
7cf5b976
QW		 8615 btrfs_delalloc_release_space(inode, offset,
8616 dio_data.reserve);
f28a4928
FM		 8617 /*
8618 * On error we might have left some ordered extents
8619 * without submitting corresponding bios for them, so
8620 * cleanup them up to avoid other tasks getting them
8621 * and waiting for them to complete forever.
8622 */
8623 if (dio_data.unsubmitted_oe_range_start <
8624 dio_data.unsubmitted_oe_range_end)
8625 btrfs_endio_direct_write_update_ordered(inode,
8626 dio_data.unsubmitted_oe_range_start,
8627 dio_data.unsubmitted_oe_range_end -
8628 dio_data.unsubmitted_oe_range_start,
8629 0);
ddba1bfc 8630 } else if (ret >= 0 && (size_t)ret < count)
7cf5b976
QW		 8631 btrfs_delalloc_release_space(inode, offset,
8632 count - (size_t)ret);
0934856d 8633 }
38851cc1 8634out:
2e60a51e 8635 if (wakeup)
fe0f07d0 8636 inode_dio_end(inode);
38851cc1 8637 if (relock)
5955102c 8638 inode_lock(inode);
0934856d
MX		 8639
8640 return ret;
16432985
CM		 8641}
8642
05dadc09
TI		 8643#define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8644
1506fcc8
YS		 8645static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8646 __u64 start, __u64 len)
8647{
05dadc09
TI		 8648 int ret;
8649
8650 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8651 if (ret)
8652 return ret;
8653
ec29ed5b 8654 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
1506fcc8
YS		 8655}
8656
a52d9a80 8657int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 8658{
d1310b2e
CM		 8659 struct extent_io_tree *tree;
8660 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 8661 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
9ebefb18 8662}
1832a6d5 8663
a52d9a80 8664static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 8665{
d1310b2e 8666 struct extent_io_tree *tree;
be7bd730
JB		 8667 struct inode *inode = page->mapping->host;
8668 int ret;
b888db2b
CM		 8669
8670 if (current->flags & PF_MEMALLOC) {
8671 redirty_page_for_writepage(wbc, page);
8672 unlock_page(page);
8673 return 0;
8674 }
be7bd730
JB		 8675
8676 /*
8677 * If we are under memory pressure we will call this directly from the
8678 * VM, we need to make sure we have the inode referenced for the ordered
8679 * extent. If not just return like we didn't do anything.
8680 */
8681 if (!igrab(inode)) {
8682 redirty_page_for_writepage(wbc, page);
8683 return AOP_WRITEPAGE_ACTIVATE;
8684 }
d1310b2e 8685 tree = &BTRFS_I(page->mapping->host)->io_tree;
be7bd730
JB		 8686 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8687 btrfs_add_delayed_iput(inode);
8688 return ret;
9ebefb18
CM		 8689}
8690
48a3b636
ES		 8691static int btrfs_writepages(struct address_space *mapping,
8692 struct writeback_control *wbc)
b293f02e 8693{
d1310b2e 8694 struct extent_io_tree *tree;
771ed689 8695
d1310b2e 8696 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 8697 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8698}
8699
3ab2fb5a
CM		 8700static int
8701btrfs_readpages(struct file *file, struct address_space *mapping,
8702 struct list_head *pages, unsigned nr_pages)
8703{
d1310b2e
CM		 8704 struct extent_io_tree *tree;
8705 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 8706 return extent_readpages(tree, mapping, pages, nr_pages,
8707 btrfs_get_extent);
8708}
e6dcd2dc 8709static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 8710{
d1310b2e
CM		 8711 struct extent_io_tree *tree;
8712 struct extent_map_tree *map;
a52d9a80 8713 int ret;
8c2383c3 8714
d1310b2e
CM		 8715 tree = &BTRFS_I(page->mapping->host)->io_tree;
8716 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 8717 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 8718 if (ret == 1) {
8719 ClearPagePrivate(page);
8720 set_page_private(page, 0);
09cbfeaf 8721 put_page(page);
39279cc3 8722 }
a52d9a80 8723 return ret;
39279cc3
CM		 8724}
8725
e6dcd2dc
CM		 8726static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8727{
98509cfc
CM		 8728 if (PageWriteback(page) || PageDirty(page))
8729 return 0;
b335b003 8730 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
e6dcd2dc
CM		 8731}
8732
d47992f8
LC		 8733static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8734 unsigned int length)
39279cc3 8735{
5fd02043 8736 struct inode *inode = page->mapping->host;
d1310b2e 8737 struct extent_io_tree *tree;
e6dcd2dc 8738 struct btrfs_ordered_extent *ordered;
2ac55d41 8739 struct extent_state *cached_state = NULL;
e6dcd2dc 8740 u64 page_start = page_offset(page);
09cbfeaf 8741 u64 page_end = page_start + PAGE_SIZE - 1;
dbfdb6d1
CR		 8742 u64 start;
8743 u64 end;
131e404a 8744 int inode_evicting = inode->i_state & I_FREEING;
39279cc3 8745
8b62b72b
CM		 8746 /*
8747 * we have the page locked, so new writeback can't start,
8748 * and the dirty bit won't be cleared while we are here.
8749 *
8750 * Wait for IO on this page so that we can safely clear
8751 * the PagePrivate2 bit and do ordered accounting
8752 */
e6dcd2dc 8753 wait_on_page_writeback(page);
8b62b72b 8754
5fd02043 8755 tree = &BTRFS_I(inode)->io_tree;
e6dcd2dc
CM		 8756 if (offset) {
8757 btrfs_releasepage(page, GFP_NOFS);
8758 return;
8759 }
131e404a
FDBM		 8760
8761 if (!inode_evicting)
ff13db41 8762 lock_extent_bits(tree, page_start, page_end, &cached_state);
dbfdb6d1
CR		 8763again:
8764 start = page_start;
8765 ordered = btrfs_lookup_ordered_range(inode, start,
8766 page_end - start + 1);
e6dcd2dc 8767 if (ordered) {
dbfdb6d1 8768 end = min(page_end, ordered->file_offset + ordered->len - 1);
eb84ae03
CM		 8769 /*
8770 * IO on this page will never be started, so we need
8771 * to account for any ordered extents now
8772 */
131e404a 8773 if (!inode_evicting)
dbfdb6d1 8774 clear_extent_bit(tree, start, end,
131e404a
FDBM		 8775 EXTENT_DIRTY | EXTENT_DELALLOC |
8776 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8777 EXTENT_DEFRAG, 1, 0, &cached_state,
8778 GFP_NOFS);
8b62b72b
CM		 8779 /*
8780 * whoever cleared the private bit is responsible
8781 * for the finish_ordered_io
8782 */
77cef2ec
JB		 8783 if (TestClearPagePrivate2(page)) {
8784 struct btrfs_ordered_inode_tree *tree;
8785 u64 new_len;
8786
8787 tree = &BTRFS_I(inode)->ordered_tree;
8788
8789 spin_lock_irq(&tree->lock);
8790 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
dbfdb6d1 8791 new_len = start - ordered->file_offset;
77cef2ec
JB		 8792 if (new_len < ordered->truncated_len)
8793 ordered->truncated_len = new_len;
8794 spin_unlock_irq(&tree->lock);
8795
8796 if (btrfs_dec_test_ordered_pending(inode, &ordered,
dbfdb6d1
CR		 8797 start,
8798 end - start + 1, 1))
77cef2ec 8799 btrfs_finish_ordered_io(ordered);
8b62b72b 8800 }
e6dcd2dc 8801 btrfs_put_ordered_extent(ordered);
131e404a
FDBM		 8802 if (!inode_evicting) {
8803 cached_state = NULL;
dbfdb6d1 8804 lock_extent_bits(tree, start, end,
131e404a
FDBM		 8805 &cached_state);
8806 }
dbfdb6d1
CR		 8807
8808 start = end + 1;
8809 if (start < page_end)
8810 goto again;
131e404a
FDBM		 8811 }
8812
b9d0b389
QW		 8813 /*
8814 * Qgroup reserved space handler
8815 * Page here will be either
8816 * 1) Already written to disk
8817 * In this case, its reserved space is released from data rsv map
8818 * and will be freed by delayed_ref handler finally.
8819 * So even we call qgroup_free_data(), it won't decrease reserved
8820 * space.
8821 * 2) Not written to disk
8822 * This means the reserved space should be freed here.
8823 */
09cbfeaf 8824 btrfs_qgroup_free_data(inode, page_start, PAGE_SIZE);
131e404a
FDBM		 8825 if (!inode_evicting) {
8826 clear_extent_bit(tree, page_start, page_end,
8827 EXTENT_LOCKED | EXTENT_DIRTY |
8828 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8829 EXTENT_DEFRAG, 1, 1,
8830 &cached_state, GFP_NOFS);
8831
8832 __btrfs_releasepage(page, GFP_NOFS);
e6dcd2dc 8833 }
e6dcd2dc 8834
4a096752 8835 ClearPageChecked(page);
9ad6b7bc 8836 if (PagePrivate(page)) {
9ad6b7bc
CM		 8837 ClearPagePrivate(page);
8838 set_page_private(page, 0);
09cbfeaf 8839 put_page(page);
9ad6b7bc 8840 }
39279cc3
CM		 8841}
8842
9ebefb18
CM		 8843/*
8844 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8845 * called from a page fault handler when a page is first dirtied. Hence we must
8846 * be careful to check for EOF conditions here. We set the page up correctly
8847 * for a written page which means we get ENOSPC checking when writing into
8848 * holes and correct delalloc and unwritten extent mapping on filesystems that
8849 * support these features.
8850 *
8851 * We are not allowed to take the i_mutex here so we have to play games to
8852 * protect against truncate races as the page could now be beyond EOF. Because
8853 * vmtruncate() writes the inode size before removing pages, once we have the
8854 * page lock we can determine safely if the page is beyond EOF. If it is not
8855 * beyond EOF, then the page is guaranteed safe against truncation until we
8856 * unlock the page.
8857 */
c2ec175c 8858int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
9ebefb18 8859{
c2ec175c 8860 struct page *page = vmf->page;
496ad9aa 8861 struct inode *inode = file_inode(vma->vm_file);
1832a6d5 8862 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 8863 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8864 struct btrfs_ordered_extent *ordered;
2ac55d41 8865 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 8866 char *kaddr;
8867 unsigned long zero_start;
9ebefb18 8868 loff_t size;
1832a6d5 8869 int ret;
9998eb70 8870 int reserved = 0;
d0b7da88 8871 u64 reserved_space;
a52d9a80 8872 u64 page_start;
e6dcd2dc 8873 u64 page_end;
d0b7da88
CR		 8874 u64 end;
8875
09cbfeaf 8876 reserved_space = PAGE_SIZE;
9ebefb18 8877
b2b5ef5c 8878 sb_start_pagefault(inode->i_sb);
df480633 8879 page_start = page_offset(page);
09cbfeaf 8880 page_end = page_start + PAGE_SIZE - 1;
d0b7da88 8881 end = page_end;
df480633 8882
d0b7da88
CR		 8883 /*
8884 * Reserving delalloc space after obtaining the page lock can lead to
8885 * deadlock. For example, if a dirty page is locked by this function
8886 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8887 * dirty page write out, then the btrfs_writepage() function could
8888 * end up waiting indefinitely to get a lock on the page currently
8889 * being processed by btrfs_page_mkwrite() function.
8890 */
7cf5b976 8891 ret = btrfs_delalloc_reserve_space(inode, page_start,
d0b7da88 8892 reserved_space);
9998eb70 8893 if (!ret) {
e41f941a 8894 ret = file_update_time(vma->vm_file);
9998eb70
CM		 8895 reserved = 1;
8896 }
56a76f82
NP		 8897 if (ret) {
8898 if (ret == -ENOMEM)
8899 ret = VM_FAULT_OOM;
8900 else /* -ENOSPC, -EIO, etc */
8901 ret = VM_FAULT_SIGBUS;
9998eb70
CM		 8902 if (reserved)
8903 goto out;
8904 goto out_noreserve;
56a76f82 8905 }
1832a6d5 8906
56a76f82 8907 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
e6dcd2dc 8908again:
9ebefb18 8909 lock_page(page);
9ebefb18 8910 size = i_size_read(inode);
a52d9a80 8911
9ebefb18 8912 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 8913 (page_start >= size)) {
9ebefb18
CM		 8914 /* page got truncated out from underneath us */
8915 goto out_unlock;
8916 }
e6dcd2dc
CM		 8917 wait_on_page_writeback(page);
8918
ff13db41 8919 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
e6dcd2dc
CM		 8920 set_page_extent_mapped(page);
8921
eb84ae03
CM		 8922 /*
8923 * we can't set the delalloc bits if there are pending ordered
8924 * extents. Drop our locks and wait for them to finish
8925 */
d0b7da88 8926 ordered = btrfs_lookup_ordered_range(inode, page_start, page_end);
e6dcd2dc 8927 if (ordered) {
2ac55d41
JB		 8928 unlock_extent_cached(io_tree, page_start, page_end,
8929 &cached_state, GFP_NOFS);
e6dcd2dc 8930 unlock_page(page);
eb84ae03 8931 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 8932 btrfs_put_ordered_extent(ordered);
8933 goto again;
8934 }
8935
09cbfeaf 8936 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
d0b7da88 8937 reserved_space = round_up(size - page_start, root->sectorsize);
09cbfeaf 8938 if (reserved_space < PAGE_SIZE) {
d0b7da88
CR		 8939 end = page_start + reserved_space - 1;
8940 spin_lock(&BTRFS_I(inode)->lock);
8941 BTRFS_I(inode)->outstanding_extents++;
8942 spin_unlock(&BTRFS_I(inode)->lock);
8943 btrfs_delalloc_release_space(inode, page_start,
09cbfeaf 8944 PAGE_SIZE - reserved_space);
d0b7da88
CR		 8945 }
8946 }
8947
fbf19087
JB		 8948 /*
8949 * XXX - page_mkwrite gets called every time the page is dirtied, even
8950 * if it was already dirty, so for space accounting reasons we need to
8951 * clear any delalloc bits for the range we are fixing to save. There
8952 * is probably a better way to do this, but for now keep consistent with
8953 * prepare_pages in the normal write path.
8954 */
d0b7da88 8955 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
9e8a4a8b
LB		 8956 EXTENT_DIRTY | EXTENT_DELALLOC |
8957 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 8958 0, 0, &cached_state, GFP_NOFS);
fbf19087 8959
d0b7da88 8960 ret = btrfs_set_extent_delalloc(inode, page_start, end,
2ac55d41 8961 &cached_state);
9ed74f2d 8962 if (ret) {
2ac55d41
JB		 8963 unlock_extent_cached(io_tree, page_start, page_end,
8964 &cached_state, GFP_NOFS);
9ed74f2d
JB		 8965 ret = VM_FAULT_SIGBUS;
8966 goto out_unlock;
8967 }
e6dcd2dc 8968 ret = 0;
9ebefb18
CM		 8969
8970 /* page is wholly or partially inside EOF */
09cbfeaf
KS		 8971 if (page_start + PAGE_SIZE > size)
8972 zero_start = size & ~PAGE_MASK;
9ebefb18 8973 else
09cbfeaf 8974 zero_start = PAGE_SIZE;
9ebefb18 8975
09cbfeaf 8976 if (zero_start != PAGE_SIZE) {
e6dcd2dc 8977 kaddr = kmap(page);
09cbfeaf 8978 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
e6dcd2dc
CM		 8979 flush_dcache_page(page);
8980 kunmap(page);
8981 }
247e743c 8982 ClearPageChecked(page);
e6dcd2dc 8983 set_page_dirty(page);
50a9b214 8984 SetPageUptodate(page);
5a3f23d5 8985
257c62e1
CM		 8986 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8987 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
46d8bc34 8988 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
257c62e1 8989
2ac55d41 8990 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9ebefb18
CM		 8991
8992out_unlock:
b2b5ef5c
JK		 8993 if (!ret) {
8994 sb_end_pagefault(inode->i_sb);
50a9b214 8995 return VM_FAULT_LOCKED;
b2b5ef5c 8996 }
9ebefb18 8997 unlock_page(page);
1832a6d5 8998out:
d0b7da88 8999 btrfs_delalloc_release_space(inode, page_start, reserved_space);
9998eb70 9000out_noreserve:
b2b5ef5c 9001 sb_end_pagefault(inode->i_sb);
9ebefb18
CM		 9002 return ret;
9003}
9004
a41ad394 9005static int btrfs_truncate(struct inode *inode)
39279cc3
CM		 9006{
9007 struct btrfs_root *root = BTRFS_I(inode)->root;
fcb80c2a 9008 struct btrfs_block_rsv *rsv;
a71754fc 9009 int ret = 0;
3893e33b 9010 int err = 0;
39279cc3 9011 struct btrfs_trans_handle *trans;
dbe674a9 9012 u64 mask = root->sectorsize - 1;
07127184 9013 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3 9014
0ef8b726
JB		 9015 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9016 (u64)-1);
9017 if (ret)
9018 return ret;
39279cc3 9019
fcb80c2a
JB		 9020 /*
9021 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
9022 * 3 things going on here
9023 *
9024 * 1) We need to reserve space for our orphan item and the space to
9025 * delete our orphan item. Lord knows we don't want to have a dangling
9026 * orphan item because we didn't reserve space to remove it.
9027 *
9028 * 2) We need to reserve space to update our inode.
9029 *
9030 * 3) We need to have something to cache all the space that is going to
9031 * be free'd up by the truncate operation, but also have some slack
9032 * space reserved in case it uses space during the truncate (thank you
9033 * very much snapshotting).
9034 *
9035 * And we need these to all be seperate. The fact is we can use alot of
9036 * space doing the truncate, and we have no earthly idea how much space
9037 * we will use, so we need the truncate reservation to be seperate so it
9038 * doesn't end up using space reserved for updating the inode or
9039 * removing the orphan item. We also need to be able to stop the
9040 * transaction and start a new one, which means we need to be able to
9041 * update the inode several times, and we have no idea of knowing how
9042 * many times that will be, so we can't just reserve 1 item for the
9043 * entirety of the opration, so that has to be done seperately as well.
9044 * Then there is the orphan item, which does indeed need to be held on
9045 * to for the whole operation, and we need nobody to touch this reserved
9046 * space except the orphan code.
9047 *
9048 * So that leaves us with
9049 *
9050 * 1) root->orphan_block_rsv - for the orphan deletion.
9051 * 2) rsv - for the truncate reservation, which we will steal from the
9052 * transaction reservation.
9053 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9054 * updating the inode.
9055 */
66d8f3dd 9056 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
fcb80c2a
JB		 9057 if (!rsv)
9058 return -ENOMEM;
4a338542 9059 rsv->size = min_size;
ca7e70f5 9060 rsv->failfast = 1;
f0cd846e 9061
907cbceb 9062 /*
07127184 9063 * 1 for the truncate slack space
907cbceb
JB		 9064 * 1 for updating the inode.
9065 */
f3fe820c 9066 trans = btrfs_start_transaction(root, 2);
fcb80c2a
JB		 9067 if (IS_ERR(trans)) {
9068 err = PTR_ERR(trans);
9069 goto out;
9070 }
f0cd846e 9071
907cbceb
JB		 9072 /* Migrate the slack space for the truncate to our reserve */
9073 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
9074 min_size);
fcb80c2a 9075 BUG_ON(ret);
f0cd846e 9076
5dc562c5
JB		 9077 /*
9078 * So if we truncate and then write and fsync we normally would just
9079 * write the extents that changed, which is a problem if we need to
9080 * first truncate that entire inode. So set this flag so we write out
9081 * all of the extents in the inode to the sync log so we're completely
9082 * safe.
9083 */
9084 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
ca7e70f5 9085 trans->block_rsv = rsv;
907cbceb 9086
8082510e
YZ		 9087 while (1) {
9088 ret = btrfs_truncate_inode_items(trans, root, inode,
9089 inode->i_size,
9090 BTRFS_EXTENT_DATA_KEY);
28ed1345 9091 if (ret != -ENOSPC && ret != -EAGAIN) {
3893e33b 9092 err = ret;
8082510e 9093 break;
3893e33b 9094 }
39279cc3 9095
fcb80c2a 9096 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e 9097 ret = btrfs_update_inode(trans, root, inode);
3893e33b
JB		 9098 if (ret) {
9099 err = ret;
9100 break;
9101 }
ca7e70f5 9102
8082510e 9103 btrfs_end_transaction(trans, root);
b53d3f5d 9104 btrfs_btree_balance_dirty(root);
ca7e70f5
JB		 9105
9106 trans = btrfs_start_transaction(root, 2);
9107 if (IS_ERR(trans)) {
9108 ret = err = PTR_ERR(trans);
9109 trans = NULL;
9110 break;
9111 }
9112
9113 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
9114 rsv, min_size);
9115 BUG_ON(ret); /* shouldn't happen */
9116 trans->block_rsv = rsv;
8082510e
YZ		 9117 }
9118
9119 if (ret == 0 && inode->i_nlink > 0) {
fcb80c2a 9120 trans->block_rsv = root->orphan_block_rsv;
8082510e 9121 ret = btrfs_orphan_del(trans, inode);
3893e33b
JB		 9122 if (ret)
9123 err = ret;
8082510e
YZ		 9124 }
9125
917c16b2
CM		 9126 if (trans) {
9127 trans->block_rsv = &root->fs_info->trans_block_rsv;
9128 ret = btrfs_update_inode(trans, root, inode);
9129 if (ret && !err)
9130 err = ret;
7b128766 9131
7ad85bb7 9132 ret = btrfs_end_transaction(trans, root);
b53d3f5d 9133 btrfs_btree_balance_dirty(root);
917c16b2 9134 }
fcb80c2a
JB		 9135
9136out:
9137 btrfs_free_block_rsv(root, rsv);
9138
3893e33b
JB		 9139 if (ret && !err)
9140 err = ret;
a41ad394 9141
3893e33b 9142 return err;
39279cc3
CM		 9143}
9144
d352ac68
CM		 9145/*
9146 * create a new subvolume directory/inode (helper for the ioctl).
9147 */
d2fb3437 9148int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
63541927
FDBM		 9149 struct btrfs_root *new_root,
9150 struct btrfs_root *parent_root,
9151 u64 new_dirid)
39279cc3 9152{
39279cc3 9153 struct inode *inode;
76dda93c 9154 int err;
00e4e6b3 9155 u64 index = 0;
39279cc3 9156
12fc9d09
FA		 9157 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9158 new_dirid, new_dirid,
9159 S_IFDIR | (~current_umask() & S_IRWXUGO),
9160 &index);
54aa1f4d 9161 if (IS_ERR(inode))
f46b5a66 9162 return PTR_ERR(inode);
39279cc3
CM		 9163 inode->i_op = &btrfs_dir_inode_operations;
9164 inode->i_fop = &btrfs_dir_file_operations;
9165
bfe86848 9166 set_nlink(inode, 1);
dbe674a9 9167 btrfs_i_size_write(inode, 0);
b0d5d10f 9168 unlock_new_inode(inode);
3b96362c 9169
63541927
FDBM		 9170 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9171 if (err)
9172 btrfs_err(new_root->fs_info,
351fd353 9173 "error inheriting subvolume %llu properties: %d",
63541927
FDBM		 9174 new_root->root_key.objectid, err);
9175
76dda93c 9176 err = btrfs_update_inode(trans, new_root, inode);
cb8e7090 9177
76dda93c 9178 iput(inode);
ce598979 9179 return err;
39279cc3
CM		 9180}
9181
39279cc3
CM		 9182struct inode *btrfs_alloc_inode(struct super_block *sb)
9183{
9184 struct btrfs_inode *ei;
2ead6ae7 9185 struct inode *inode;
39279cc3
CM		 9186
9187 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9188 if (!ei)
9189 return NULL;
2ead6ae7
YZ		 9190
9191 ei->root = NULL;
2ead6ae7 9192 ei->generation = 0;
15ee9bc7 9193 ei->last_trans = 0;
257c62e1 9194 ei->last_sub_trans = 0;
e02119d5 9195 ei->logged_trans = 0;
2ead6ae7 9196 ei->delalloc_bytes = 0;
47059d93 9197 ei->defrag_bytes = 0;
2ead6ae7
YZ		 9198 ei->disk_i_size = 0;
9199 ei->flags = 0;
7709cde3 9200 ei->csum_bytes = 0;
2ead6ae7 9201 ei->index_cnt = (u64)-1;
67de1176 9202 ei->dir_index = 0;
2ead6ae7 9203 ei->last_unlink_trans = 0;
46d8bc34 9204 ei->last_log_commit = 0;
8089fe62 9205 ei->delayed_iput_count = 0;
2ead6ae7 9206
9e0baf60
JB		 9207 spin_lock_init(&ei->lock);
9208 ei->outstanding_extents = 0;
9209 ei->reserved_extents = 0;
2ead6ae7 9210
72ac3c0d 9211 ei->runtime_flags = 0;
261507a0 9212 ei->force_compress = BTRFS_COMPRESS_NONE;
2ead6ae7 9213
16cdcec7
MX		 9214 ei->delayed_node = NULL;
9215
9cc97d64 9216 ei->i_otime.tv_sec = 0;
9217 ei->i_otime.tv_nsec = 0;
9218
2ead6ae7 9219 inode = &ei->vfs_inode;
a8067e02 9220 extent_map_tree_init(&ei->extent_tree);
f993c883
DS		 9221 extent_io_tree_init(&ei->io_tree, &inode->i_data);
9222 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
0b32f4bb
JB		 9223 ei->io_tree.track_uptodate = 1;
9224 ei->io_failure_tree.track_uptodate = 1;
b812ce28 9225 atomic_set(&ei->sync_writers, 0);
2ead6ae7 9226 mutex_init(&ei->log_mutex);
f248679e 9227 mutex_init(&ei->delalloc_mutex);
e6dcd2dc 9228 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
2ead6ae7 9229 INIT_LIST_HEAD(&ei->delalloc_inodes);
8089fe62 9230 INIT_LIST_HEAD(&ei->delayed_iput);
2ead6ae7
YZ		 9231 RB_CLEAR_NODE(&ei->rb_node);
9232
9233 return inode;
39279cc3
CM		 9234}
9235
aaedb55b
JB		 9236#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9237void btrfs_test_destroy_inode(struct inode *inode)
9238{
9239 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9240 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9241}
9242#endif
9243
fa0d7e3d
NP		 9244static void btrfs_i_callback(struct rcu_head *head)
9245{
9246 struct inode *inode = container_of(head, struct inode, i_rcu);
fa0d7e3d
NP		 9247 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9248}
9249
39279cc3
CM		 9250void btrfs_destroy_inode(struct inode *inode)
9251{
e6dcd2dc 9252 struct btrfs_ordered_extent *ordered;
5a3f23d5
CM		 9253 struct btrfs_root *root = BTRFS_I(inode)->root;
9254
b3d9b7a3 9255 WARN_ON(!hlist_empty(&inode->i_dentry));
39279cc3 9256 WARN_ON(inode->i_data.nrpages);
9e0baf60
JB		 9257 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9258 WARN_ON(BTRFS_I(inode)->reserved_extents);
7709cde3
JB		 9259 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9260 WARN_ON(BTRFS_I(inode)->csum_bytes);
47059d93 9261 WARN_ON(BTRFS_I(inode)->defrag_bytes);
39279cc3 9262
a6dbd429
JB		 9263 /*
9264 * This can happen where we create an inode, but somebody else also
9265 * created the same inode and we need to destroy the one we already
9266 * created.
9267 */
9268 if (!root)
9269 goto free;
9270
8a35d95f
JB		 9271 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9272 &BTRFS_I(inode)->runtime_flags)) {
c2cf52eb 9273 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
c1c9ff7c 9274 btrfs_ino(inode));
8a35d95f 9275 atomic_dec(&root->orphan_inodes);
7b128766 9276 }
7b128766 9277
d397712b 9278 while (1) {
e6dcd2dc
CM		 9279 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9280 if (!ordered)
9281 break;
9282 else {
c2cf52eb 9283 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
c1c9ff7c 9284 ordered->file_offset, ordered->len);
e6dcd2dc
CM		 9285 btrfs_remove_ordered_extent(inode, ordered);
9286 btrfs_put_ordered_extent(ordered);
9287 btrfs_put_ordered_extent(ordered);
9288 }
9289 }
56fa9d07 9290 btrfs_qgroup_check_reserved_leak(inode);
5d4f98a2 9291 inode_tree_del(inode);
5b21f2ed 9292 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
a6dbd429 9293free:
fa0d7e3d 9294 call_rcu(&inode->i_rcu, btrfs_i_callback);
39279cc3
CM		 9295}
9296
45321ac5 9297int btrfs_drop_inode(struct inode *inode)
76dda93c
YZ		 9298{
9299 struct btrfs_root *root = BTRFS_I(inode)->root;
45321ac5 9300
6379ef9f
NA		 9301 if (root == NULL)
9302 return 1;
9303
fa6ac876 9304 /* the snap/subvol tree is on deleting */
69e9c6c6 9305 if (btrfs_root_refs(&root->root_item) == 0)
45321ac5 9306 return 1;
76dda93c 9307 else
45321ac5 9308 return generic_drop_inode(inode);
76dda93c
YZ		 9309}
9310
0ee0fda0 9311static void init_once(void *foo)
39279cc3
CM		 9312{
9313 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9314
9315 inode_init_once(&ei->vfs_inode);
9316}
9317
9318void btrfs_destroy_cachep(void)
9319{
8c0a8537
KS		 9320 /*
9321 * Make sure all delayed rcu free inodes are flushed before we
9322 * destroy cache.
9323 */
9324 rcu_barrier();
5598e900
KM		 9325 kmem_cache_destroy(btrfs_inode_cachep);
9326 kmem_cache_destroy(btrfs_trans_handle_cachep);
9327 kmem_cache_destroy(btrfs_transaction_cachep);
9328 kmem_cache_destroy(btrfs_path_cachep);
9329 kmem_cache_destroy(btrfs_free_space_cachep);
39279cc3
CM		 9330}
9331
9332int btrfs_init_cachep(void)
9333{
837e1972 9334 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9601e3f6 9335 sizeof(struct btrfs_inode), 0,
5d097056
VD		 9336 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9337 init_once);
39279cc3
CM		 9338 if (!btrfs_inode_cachep)
9339 goto fail;
9601e3f6 9340
837e1972 9341 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9601e3f6
CH		 9342 sizeof(struct btrfs_trans_handle), 0,
9343 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 9344 if (!btrfs_trans_handle_cachep)
9345 goto fail;
9601e3f6 9346
837e1972 9347 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9601e3f6
CH		 9348 sizeof(struct btrfs_transaction), 0,
9349 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 9350 if (!btrfs_transaction_cachep)
9351 goto fail;
9601e3f6 9352
837e1972 9353 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9601e3f6
CH		 9354 sizeof(struct btrfs_path), 0,
9355 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 9356 if (!btrfs_path_cachep)
9357 goto fail;
9601e3f6 9358
837e1972 9359 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
dc89e982
JB		 9360 sizeof(struct btrfs_free_space), 0,
9361 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9362 if (!btrfs_free_space_cachep)
9363 goto fail;
9364
39279cc3
CM		 9365 return 0;
9366fail:
9367 btrfs_destroy_cachep();
9368 return -ENOMEM;
9369}
9370
9371static int btrfs_getattr(struct vfsmount *mnt,
9372 struct dentry *dentry, struct kstat *stat)
9373{
df0af1a5 9374 u64 delalloc_bytes;
2b0143b5 9375 struct inode *inode = d_inode(dentry);
fadc0d8b
DS		 9376 u32 blocksize = inode->i_sb->s_blocksize;
9377
39279cc3 9378 generic_fillattr(inode, stat);
0ee5dc67 9379 stat->dev = BTRFS_I(inode)->root->anon_dev;
df0af1a5
MX		 9380
9381 spin_lock(&BTRFS_I(inode)->lock);
9382 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9383 spin_unlock(&BTRFS_I(inode)->lock);
fadc0d8b 9384 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
df0af1a5 9385 ALIGN(delalloc_bytes, blocksize)) >> 9;
39279cc3
CM		 9386 return 0;
9387}
9388
d397712b
CM		 9389static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9390 struct inode *new_dir, struct dentry *new_dentry)
39279cc3
CM		 9391{
9392 struct btrfs_trans_handle *trans;
9393 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4df27c4d 9394 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
2b0143b5
DH		 9395 struct inode *new_inode = d_inode(new_dentry);
9396 struct inode *old_inode = d_inode(old_dentry);
00e4e6b3 9397 u64 index = 0;
4df27c4d 9398 u64 root_objectid;
39279cc3 9399 int ret;
33345d01 9400 u64 old_ino = btrfs_ino(old_inode);
39279cc3 9401
33345d01 9402 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
f679a840
YZ		 9403 return -EPERM;
9404
4df27c4d 9405 /* we only allow rename subvolume link between subvolumes */
33345d01 9406 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
3394e160
CM		 9407 return -EXDEV;
9408
33345d01
LZ		 9409 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9410 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
39279cc3 9411 return -ENOTEMPTY;
5f39d397 9412
4df27c4d
YZ		 9413 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9414 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9415 return -ENOTEMPTY;
9c52057c
CM		 9416
9417
9418 /* check for collisions, even if the name isn't there */
4871c158 9419 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9c52057c
CM		 9420 new_dentry->d_name.name,
9421 new_dentry->d_name.len);
9422
9423 if (ret) {
9424 if (ret == -EEXIST) {
9425 /* we shouldn't get
9426 * eexist without a new_inode */
fae7f21c 9427 if (WARN_ON(!new_inode)) {
9c52057c
CM		 9428 return ret;
9429 }
9430 } else {
9431 /* maybe -EOVERFLOW */
9432 return ret;
9433 }
9434 }
9435 ret = 0;
9436
5a3f23d5 9437 /*
8d875f95
CM		 9438 * we're using rename to replace one file with another. Start IO on it
9439 * now so we don't add too much work to the end of the transaction
5a3f23d5 9440 */
8d875f95 9441 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
5a3f23d5
CM		 9442 filemap_flush(old_inode->i_mapping);
9443
76dda93c 9444 /* close the racy window with snapshot create/destroy ioctl */
33345d01 9445 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 9446 down_read(&root->fs_info->subvol_sem);
a22285a6
YZ		 9447 /*
9448 * We want to reserve the absolute worst case amount of items. So if
9449 * both inodes are subvols and we need to unlink them then that would
9450 * require 4 item modifications, but if they are both normal inodes it
9451 * would require 5 item modifications, so we'll assume their normal
9452 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9453 * should cover the worst case number of items we'll modify.
9454 */
6e137ed3 9455 trans = btrfs_start_transaction(root, 11);
b44c59a8
JL		 9456 if (IS_ERR(trans)) {
9457 ret = PTR_ERR(trans);
9458 goto out_notrans;
9459 }
76dda93c 9460
4df27c4d
YZ		 9461 if (dest != root)
9462 btrfs_record_root_in_trans(trans, dest);
5f39d397 9463
a5719521
YZ		 9464 ret = btrfs_set_inode_index(new_dir, &index);
9465 if (ret)
9466 goto out_fail;
5a3f23d5 9467
67de1176 9468 BTRFS_I(old_inode)->dir_index = 0ULL;
33345d01 9469 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d 9470 /* force full log commit if subvolume involved. */
995946dd 9471 btrfs_set_log_full_commit(root->fs_info, trans);
4df27c4d 9472 } else {
a5719521
YZ		 9473 ret = btrfs_insert_inode_ref(trans, dest,
9474 new_dentry->d_name.name,
9475 new_dentry->d_name.len,
33345d01
LZ		 9476 old_ino,
9477 btrfs_ino(new_dir), index);
a5719521
YZ		 9478 if (ret)
9479 goto out_fail;
4df27c4d
YZ		 9480 /*
9481 * this is an ugly little race, but the rename is required
9482 * to make sure that if we crash, the inode is either at the
9483 * old name or the new one. pinning the log transaction lets
9484 * us make sure we don't allow a log commit to come in after
9485 * we unlink the name but before we add the new name back in.
9486 */
9487 btrfs_pin_log_trans(root);
9488 }
5a3f23d5 9489
0c4d2d95
JB		 9490 inode_inc_iversion(old_dir);
9491 inode_inc_iversion(new_dir);
9492 inode_inc_iversion(old_inode);
04b285f3
DD		 9493 old_dir->i_ctime = old_dir->i_mtime =
9494 new_dir->i_ctime = new_dir->i_mtime =
9495 old_inode->i_ctime = current_fs_time(old_dir->i_sb);
5f39d397 9496
12fcfd22
CM		 9497 if (old_dentry->d_parent != new_dentry->d_parent)
9498 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9499
33345d01 9500 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 9501 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9502 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9503 old_dentry->d_name.name,
9504 old_dentry->d_name.len);
9505 } else {
92986796 9506 ret = __btrfs_unlink_inode(trans, root, old_dir,
2b0143b5 9507 d_inode(old_dentry),
92986796
AV		 9508 old_dentry->d_name.name,
9509 old_dentry->d_name.len);
9510 if (!ret)
9511 ret = btrfs_update_inode(trans, root, old_inode);
4df27c4d 9512 }
79787eaa
JM		 9513 if (ret) {
9514 btrfs_abort_transaction(trans, root, ret);
9515 goto out_fail;
9516 }
39279cc3
CM		 9517
9518 if (new_inode) {
0c4d2d95 9519 inode_inc_iversion(new_inode);
04b285f3 9520 new_inode->i_ctime = current_fs_time(new_inode->i_sb);
33345d01 9521 if (unlikely(btrfs_ino(new_inode) ==
4df27c4d
YZ		 9522 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9523 root_objectid = BTRFS_I(new_inode)->location.objectid;
9524 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9525 root_objectid,
9526 new_dentry->d_name.name,
9527 new_dentry->d_name.len);
9528 BUG_ON(new_inode->i_nlink == 0);
9529 } else {
9530 ret = btrfs_unlink_inode(trans, dest, new_dir,
2b0143b5 9531 d_inode(new_dentry),
4df27c4d
YZ		 9532 new_dentry->d_name.name,
9533 new_dentry->d_name.len);
9534 }
4ef31a45 9535 if (!ret && new_inode->i_nlink == 0)
2b0143b5 9536 ret = btrfs_orphan_add(trans, d_inode(new_dentry));
79787eaa
JM		 9537 if (ret) {
9538 btrfs_abort_transaction(trans, root, ret);
9539 goto out_fail;
9540 }
39279cc3 9541 }
aec7477b 9542
4df27c4d
YZ		 9543 ret = btrfs_add_link(trans, new_dir, old_inode,
9544 new_dentry->d_name.name,
a5719521 9545 new_dentry->d_name.len, 0, index);
79787eaa
JM		 9546 if (ret) {
9547 btrfs_abort_transaction(trans, root, ret);
9548 goto out_fail;
9549 }
39279cc3 9550
67de1176
MX		 9551 if (old_inode->i_nlink == 1)
9552 BTRFS_I(old_inode)->dir_index = index;
9553
33345d01 9554 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
10d9f309 9555 struct dentry *parent = new_dentry->d_parent;
6a912213 9556 btrfs_log_new_name(trans, old_inode, old_dir, parent);
4df27c4d
YZ		 9557 btrfs_end_log_trans(root);
9558 }
39279cc3 9559out_fail:
7ad85bb7 9560 btrfs_end_transaction(trans, root);
b44c59a8 9561out_notrans:
33345d01 9562 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 9563 up_read(&root->fs_info->subvol_sem);
9ed74f2d 9564
39279cc3
CM		 9565 return ret;
9566}
9567
80ace85c
MS		 9568static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9569 struct inode *new_dir, struct dentry *new_dentry,
9570 unsigned int flags)
9571{
9572 if (flags & ~RENAME_NOREPLACE)
9573 return -EINVAL;
9574
9575 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
9576}
9577
8ccf6f19
MX		 9578static void btrfs_run_delalloc_work(struct btrfs_work *work)
9579{
9580 struct btrfs_delalloc_work *delalloc_work;
9f23e289 9581 struct inode *inode;
8ccf6f19
MX		 9582
9583 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9584 work);
9f23e289 9585 inode = delalloc_work->inode;
30424601
DS		 9586 filemap_flush(inode->i_mapping);
9587 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9588 &BTRFS_I(inode)->runtime_flags))
9f23e289 9589 filemap_flush(inode->i_mapping);
8ccf6f19
MX		 9590
9591 if (delalloc_work->delay_iput)
9f23e289 9592 btrfs_add_delayed_iput(inode);
8ccf6f19 9593 else
9f23e289 9594 iput(inode);
8ccf6f19
MX		 9595 complete(&delalloc_work->completion);
9596}
9597
9598struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
651d494a 9599 int delay_iput)
8ccf6f19
MX		 9600{
9601 struct btrfs_delalloc_work *work;
9602
100d5702 9603 work = kmalloc(sizeof(*work), GFP_NOFS);
8ccf6f19
MX		 9604 if (!work)
9605 return NULL;
9606
9607 init_completion(&work->completion);
9608 INIT_LIST_HEAD(&work->list);
9609 work->inode = inode;
8ccf6f19 9610 work->delay_iput = delay_iput;
9e0af237
LB		 9611 WARN_ON_ONCE(!inode);
9612 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9613 btrfs_run_delalloc_work, NULL, NULL);
8ccf6f19
MX		 9614
9615 return work;
9616}
9617
9618void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9619{
9620 wait_for_completion(&work->completion);
100d5702 9621 kfree(work);
8ccf6f19
MX		 9622}
9623
d352ac68
CM		 9624/*
9625 * some fairly slow code that needs optimization. This walks the list
9626 * of all the inodes with pending delalloc and forces them to disk.
9627 */
6c255e67
MX		 9628static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9629 int nr)
ea8c2819 9630{
ea8c2819 9631 struct btrfs_inode *binode;
5b21f2ed 9632 struct inode *inode;
8ccf6f19
MX		 9633 struct btrfs_delalloc_work *work, *next;
9634 struct list_head works;
1eafa6c7 9635 struct list_head splice;
8ccf6f19 9636 int ret = 0;
ea8c2819 9637
8ccf6f19 9638 INIT_LIST_HEAD(&works);
1eafa6c7 9639 INIT_LIST_HEAD(&splice);
63607cc8 9640
573bfb72 9641 mutex_lock(&root->delalloc_mutex);
eb73c1b7
MX		 9642 spin_lock(&root->delalloc_lock);
9643 list_splice_init(&root->delalloc_inodes, &splice);
1eafa6c7
MX		 9644 while (!list_empty(&splice)) {
9645 binode = list_entry(splice.next, struct btrfs_inode,
ea8c2819 9646 delalloc_inodes);
1eafa6c7 9647
eb73c1b7
MX		 9648 list_move_tail(&binode->delalloc_inodes,
9649 &root->delalloc_inodes);
5b21f2ed 9650 inode = igrab(&binode->vfs_inode);
df0af1a5 9651 if (!inode) {
eb73c1b7 9652 cond_resched_lock(&root->delalloc_lock);
1eafa6c7 9653 continue;
df0af1a5 9654 }
eb73c1b7 9655 spin_unlock(&root->delalloc_lock);
1eafa6c7 9656
651d494a 9657 work = btrfs_alloc_delalloc_work(inode, delay_iput);
5d99a998 9658 if (!work) {
f4ab9ea7
JB		 9659 if (delay_iput)
9660 btrfs_add_delayed_iput(inode);
9661 else
9662 iput(inode);
1eafa6c7 9663 ret = -ENOMEM;
a1ecaabb 9664 goto out;
5b21f2ed 9665 }
1eafa6c7 9666 list_add_tail(&work->list, &works);
a44903ab
QW		 9667 btrfs_queue_work(root->fs_info->flush_workers,
9668 &work->work);
6c255e67
MX		 9669 ret++;
9670 if (nr != -1 && ret >= nr)
a1ecaabb 9671 goto out;
5b21f2ed 9672 cond_resched();
eb73c1b7 9673 spin_lock(&root->delalloc_lock);
ea8c2819 9674 }
eb73c1b7 9675 spin_unlock(&root->delalloc_lock);
8c8bee1d 9676
a1ecaabb 9677out:
eb73c1b7
MX		 9678 list_for_each_entry_safe(work, next, &works, list) {
9679 list_del_init(&work->list);
9680 btrfs_wait_and_free_delalloc_work(work);
9681 }
9682
9683 if (!list_empty_careful(&splice)) {
9684 spin_lock(&root->delalloc_lock);
9685 list_splice_tail(&splice, &root->delalloc_inodes);
9686 spin_unlock(&root->delalloc_lock);
9687 }
573bfb72 9688 mutex_unlock(&root->delalloc_mutex);
eb73c1b7
MX		 9689 return ret;
9690}
1eafa6c7 9691
eb73c1b7
MX		 9692int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9693{
9694 int ret;
1eafa6c7 9695
2c21b4d7 9696 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
eb73c1b7
MX		 9697 return -EROFS;
9698
6c255e67
MX		 9699 ret = __start_delalloc_inodes(root, delay_iput, -1);
9700 if (ret > 0)
9701 ret = 0;
eb73c1b7
MX		 9702 /*
9703 * the filemap_flush will queue IO into the worker threads, but
8c8bee1d
CM		 9704 * we have to make sure the IO is actually started and that
9705 * ordered extents get created before we return
9706 */
9707 atomic_inc(&root->fs_info->async_submit_draining);
d397712b 9708 while (atomic_read(&root->fs_info->nr_async_submits) ||
771ed689 9709 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 9710 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 9711 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9712 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 9713 }
9714 atomic_dec(&root->fs_info->async_submit_draining);
eb73c1b7
MX		 9715 return ret;
9716}
9717
6c255e67
MX		 9718int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9719 int nr)
eb73c1b7
MX		 9720{
9721 struct btrfs_root *root;
9722 struct list_head splice;
9723 int ret;
9724
2c21b4d7 9725 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
eb73c1b7
MX		 9726 return -EROFS;
9727
9728 INIT_LIST_HEAD(&splice);
9729
573bfb72 9730 mutex_lock(&fs_info->delalloc_root_mutex);
eb73c1b7
MX		 9731 spin_lock(&fs_info->delalloc_root_lock);
9732 list_splice_init(&fs_info->delalloc_roots, &splice);
6c255e67 9733 while (!list_empty(&splice) && nr) {
eb73c1b7
MX		 9734 root = list_first_entry(&splice, struct btrfs_root,
9735 delalloc_root);
9736 root = btrfs_grab_fs_root(root);
9737 BUG_ON(!root);
9738 list_move_tail(&root->delalloc_root,
9739 &fs_info->delalloc_roots);
9740 spin_unlock(&fs_info->delalloc_root_lock);
9741
6c255e67 9742 ret = __start_delalloc_inodes(root, delay_iput, nr);
eb73c1b7 9743 btrfs_put_fs_root(root);
6c255e67 9744 if (ret < 0)
eb73c1b7
MX		 9745 goto out;
9746
6c255e67
MX		 9747 if (nr != -1) {
9748 nr -= ret;
9749 WARN_ON(nr < 0);
9750 }
eb73c1b7 9751 spin_lock(&fs_info->delalloc_root_lock);
8ccf6f19 9752 }
eb73c1b7 9753 spin_unlock(&fs_info->delalloc_root_lock);
1eafa6c7 9754
6c255e67 9755 ret = 0;
eb73c1b7
MX		 9756 atomic_inc(&fs_info->async_submit_draining);
9757 while (atomic_read(&fs_info->nr_async_submits) ||
9758 atomic_read(&fs_info->async_delalloc_pages)) {
9759 wait_event(fs_info->async_submit_wait,
9760 (atomic_read(&fs_info->nr_async_submits) == 0 &&
9761 atomic_read(&fs_info->async_delalloc_pages) == 0));
9762 }
9763 atomic_dec(&fs_info->async_submit_draining);
eb73c1b7 9764out:
1eafa6c7 9765 if (!list_empty_careful(&splice)) {
eb73c1b7
MX		 9766 spin_lock(&fs_info->delalloc_root_lock);
9767 list_splice_tail(&splice, &fs_info->delalloc_roots);
9768 spin_unlock(&fs_info->delalloc_root_lock);
1eafa6c7 9769 }
573bfb72 9770 mutex_unlock(&fs_info->delalloc_root_mutex);
8ccf6f19 9771 return ret;
ea8c2819
CM		 9772}
9773
39279cc3
CM		 9774static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9775 const char *symname)
9776{
9777 struct btrfs_trans_handle *trans;
9778 struct btrfs_root *root = BTRFS_I(dir)->root;
9779 struct btrfs_path *path;
9780 struct btrfs_key key;
1832a6d5 9781 struct inode *inode = NULL;
39279cc3
CM		 9782 int err;
9783 int drop_inode = 0;
9784 u64 objectid;
67871254 9785 u64 index = 0;
39279cc3
CM		 9786 int name_len;
9787 int datasize;
5f39d397 9788 unsigned long ptr;
39279cc3 9789 struct btrfs_file_extent_item *ei;
5f39d397 9790 struct extent_buffer *leaf;
39279cc3 9791
f06becc4 9792 name_len = strlen(symname);
39279cc3
CM		 9793 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9794 return -ENAMETOOLONG;
1832a6d5 9795
9ed74f2d
JB		 9796 /*
9797 * 2 items for inode item and ref
9798 * 2 items for dir items
9269d12b
FM		 9799 * 1 item for updating parent inode item
9800 * 1 item for the inline extent item
9ed74f2d
JB		 9801 * 1 item for xattr if selinux is on
9802 */
9269d12b 9803 trans = btrfs_start_transaction(root, 7);
a22285a6
YZ		 9804 if (IS_ERR(trans))
9805 return PTR_ERR(trans);
1832a6d5 9806
581bb050
LZ		 9807 err = btrfs_find_free_ino(root, &objectid);
9808 if (err)
9809 goto out_unlock;
9810
aec7477b 9811 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 9812 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 9813 S_IFLNK|S_IRWXUGO, &index);
7cf96da3
TI		 9814 if (IS_ERR(inode)) {
9815 err = PTR_ERR(inode);
39279cc3 9816 goto out_unlock;
7cf96da3 9817 }
39279cc3 9818
ad19db71
CS		 9819 /*
9820 * If the active LSM wants to access the inode during
9821 * d_instantiate it needs these. Smack checks to see
9822 * if the filesystem supports xattrs by looking at the
9823 * ops vector.
9824 */
9825 inode->i_fop = &btrfs_file_operations;
9826 inode->i_op = &btrfs_file_inode_operations;
b0d5d10f 9827 inode->i_mapping->a_ops = &btrfs_aops;
b0d5d10f
CM		 9828 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9829
9830 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9831 if (err)
9832 goto out_unlock_inode;
ad19db71 9833
39279cc3 9834 path = btrfs_alloc_path();
d8926bb3
MF		 9835 if (!path) {
9836 err = -ENOMEM;
b0d5d10f 9837 goto out_unlock_inode;
d8926bb3 9838 }
33345d01 9839 key.objectid = btrfs_ino(inode);
39279cc3 9840 key.offset = 0;
962a298f 9841 key.type = BTRFS_EXTENT_DATA_KEY;
39279cc3
CM		 9842 datasize = btrfs_file_extent_calc_inline_size(name_len);
9843 err = btrfs_insert_empty_item(trans, root, path, &key,
9844 datasize);
54aa1f4d 9845 if (err) {
b0839166 9846 btrfs_free_path(path);
b0d5d10f 9847 goto out_unlock_inode;
54aa1f4d 9848 }
5f39d397
CM		 9849 leaf = path->nodes[0];
9850 ei = btrfs_item_ptr(leaf, path->slots[0],
9851 struct btrfs_file_extent_item);
9852 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9853 btrfs_set_file_extent_type(leaf, ei,
39279cc3 9854 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 9855 btrfs_set_file_extent_encryption(leaf, ei, 0);
9856 btrfs_set_file_extent_compression(leaf, ei, 0);
9857 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9858 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9859
39279cc3 9860 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 9861 write_extent_buffer(leaf, symname, ptr, name_len);
9862 btrfs_mark_buffer_dirty(leaf);
39279cc3 9863 btrfs_free_path(path);
5f39d397 9864
39279cc3 9865 inode->i_op = &btrfs_symlink_inode_operations;
21fc61c7 9866 inode_nohighmem(inode);
39279cc3 9867 inode->i_mapping->a_ops = &btrfs_symlink_aops;
d899e052 9868 inode_set_bytes(inode, name_len);
f06becc4 9869 btrfs_i_size_write(inode, name_len);
54aa1f4d 9870 err = btrfs_update_inode(trans, root, inode);
d50866d0
FM		 9871 /*
9872 * Last step, add directory indexes for our symlink inode. This is the
9873 * last step to avoid extra cleanup of these indexes if an error happens
9874 * elsewhere above.
9875 */
9876 if (!err)
9877 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
b0d5d10f 9878 if (err) {
54aa1f4d 9879 drop_inode = 1;
b0d5d10f
CM		 9880 goto out_unlock_inode;
9881 }
9882
9883 unlock_new_inode(inode);
9884 d_instantiate(dentry, inode);
39279cc3
CM		 9885
9886out_unlock:
7ad85bb7 9887 btrfs_end_transaction(trans, root);
39279cc3
CM		 9888 if (drop_inode) {
9889 inode_dec_link_count(inode);
9890 iput(inode);
9891 }
b53d3f5d 9892 btrfs_btree_balance_dirty(root);
39279cc3 9893 return err;
b0d5d10f
CM		 9894
9895out_unlock_inode:
9896 drop_inode = 1;
9897 unlock_new_inode(inode);
9898 goto out_unlock;
39279cc3 9899}
16432985 9900
0af3d00b
JB		 9901static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9902 u64 start, u64 num_bytes, u64 min_size,
9903 loff_t actual_len, u64 *alloc_hint,
9904 struct btrfs_trans_handle *trans)
d899e052 9905{
5dc562c5
JB		 9906 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9907 struct extent_map *em;
d899e052
YZ		 9908 struct btrfs_root *root = BTRFS_I(inode)->root;
9909 struct btrfs_key ins;
d899e052 9910 u64 cur_offset = start;
55a61d1d 9911 u64 i_size;
154ea289 9912 u64 cur_bytes;
0b670dc4 9913 u64 last_alloc = (u64)-1;
d899e052 9914 int ret = 0;
0af3d00b 9915 bool own_trans = true;
d899e052 9916
0af3d00b
JB		 9917 if (trans)
9918 own_trans = false;
d899e052 9919 while (num_bytes > 0) {
0af3d00b
JB		 9920 if (own_trans) {
9921 trans = btrfs_start_transaction(root, 3);
9922 if (IS_ERR(trans)) {
9923 ret = PTR_ERR(trans);
9924 break;
9925 }
5a303d5d
YZ		 9926 }
9927
ee22184b 9928 cur_bytes = min_t(u64, num_bytes, SZ_256M);
154ea289 9929 cur_bytes = max(cur_bytes, min_size);
0b670dc4
JB		 9930 /*
9931 * If we are severely fragmented we could end up with really
9932 * small allocations, so if the allocator is returning small
9933 * chunks lets make its job easier by only searching for those
9934 * sized chunks.
9935 */
9936 cur_bytes = min(cur_bytes, last_alloc);
00361589 9937 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
e570fd27 9938 *alloc_hint, &ins, 1, 0);
5a303d5d 9939 if (ret) {
0af3d00b
JB		 9940 if (own_trans)
9941 btrfs_end_transaction(trans, root);
a22285a6 9942 break;
d899e052 9943 }
5a303d5d 9944
0b670dc4 9945 last_alloc = ins.offset;
d899e052
YZ		 9946 ret = insert_reserved_file_extent(trans, inode,
9947 cur_offset, ins.objectid,
9948 ins.offset, ins.offset,
920bbbfb 9949 ins.offset, 0, 0, 0,
d899e052 9950 BTRFS_FILE_EXTENT_PREALLOC);
79787eaa 9951 if (ret) {
857cc2fc 9952 btrfs_free_reserved_extent(root, ins.objectid,
e570fd27 9953 ins.offset, 0);
79787eaa
JM		 9954 btrfs_abort_transaction(trans, root, ret);
9955 if (own_trans)
9956 btrfs_end_transaction(trans, root);
9957 break;
9958 }
31193213 9959
a1ed835e
CM		 9960 btrfs_drop_extent_cache(inode, cur_offset,
9961 cur_offset + ins.offset -1, 0);
5a303d5d 9962
5dc562c5
JB		 9963 em = alloc_extent_map();
9964 if (!em) {
9965 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9966 &BTRFS_I(inode)->runtime_flags);
9967 goto next;
9968 }
9969
9970 em->start = cur_offset;
9971 em->orig_start = cur_offset;
9972 em->len = ins.offset;
9973 em->block_start = ins.objectid;
9974 em->block_len = ins.offset;
b4939680 9975 em->orig_block_len = ins.offset;
cc95bef6 9976 em->ram_bytes = ins.offset;
5dc562c5
JB		 9977 em->bdev = root->fs_info->fs_devices->latest_bdev;
9978 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9979 em->generation = trans->transid;
9980
9981 while (1) {
9982 write_lock(&em_tree->lock);
09a2a8f9 9983 ret = add_extent_mapping(em_tree, em, 1);
5dc562c5
JB		 9984 write_unlock(&em_tree->lock);
9985 if (ret != -EEXIST)
9986 break;
9987 btrfs_drop_extent_cache(inode, cur_offset,
9988 cur_offset + ins.offset - 1,
9989 0);
9990 }
9991 free_extent_map(em);
9992next:
d899e052
YZ		 9993 num_bytes -= ins.offset;
9994 cur_offset += ins.offset;
efa56464 9995 *alloc_hint = ins.objectid + ins.offset;
5a303d5d 9996
0c4d2d95 9997 inode_inc_iversion(inode);
04b285f3 9998 inode->i_ctime = current_fs_time(inode->i_sb);
6cbff00f 9999 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
d899e052 10000 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
efa56464
YZ		 10001 (actual_len > inode->i_size) &&
10002 (cur_offset > inode->i_size)) {
d1ea6a61 10003 if (cur_offset > actual_len)
55a61d1d 10004 i_size = actual_len;
d1ea6a61 10005 else
55a61d1d
JB		 10006 i_size = cur_offset;
10007 i_size_write(inode, i_size);
10008 btrfs_ordered_update_i_size(inode, i_size, NULL);
5a303d5d
YZ		 10009 }
10010
d899e052 10011 ret = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 10012
10013 if (ret) {
10014 btrfs_abort_transaction(trans, root, ret);
10015 if (own_trans)
10016 btrfs_end_transaction(trans, root);
10017 break;
10018 }
d899e052 10019
0af3d00b
JB		 10020 if (own_trans)
10021 btrfs_end_transaction(trans, root);
5a303d5d 10022 }
d899e052
YZ		 10023 return ret;
10024}
10025
0af3d00b
JB		 10026int btrfs_prealloc_file_range(struct inode *inode, int mode,
10027 u64 start, u64 num_bytes, u64 min_size,
10028 loff_t actual_len, u64 *alloc_hint)
10029{
10030 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10031 min_size, actual_len, alloc_hint,
10032 NULL);
10033}
10034
10035int btrfs_prealloc_file_range_trans(struct inode *inode,
10036 struct btrfs_trans_handle *trans, int mode,
10037 u64 start, u64 num_bytes, u64 min_size,
10038 loff_t actual_len, u64 *alloc_hint)
10039{
10040 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10041 min_size, actual_len, alloc_hint, trans);
10042}
10043
e6dcd2dc
CM		 10044static int btrfs_set_page_dirty(struct page *page)
10045{
e6dcd2dc
CM		 10046 return __set_page_dirty_nobuffers(page);
10047}
10048
10556cb2 10049static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd 10050{
b83cc969 10051 struct btrfs_root *root = BTRFS_I(inode)->root;
cb6db4e5 10052 umode_t mode = inode->i_mode;
b83cc969 10053
cb6db4e5
JM		 10054 if (mask & MAY_WRITE &&
10055 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10056 if (btrfs_root_readonly(root))
10057 return -EROFS;
10058 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10059 return -EACCES;
10060 }
2830ba7f 10061 return generic_permission(inode, mask);
fdebe2bd 10062}
39279cc3 10063
ef3b9af5
FM		 10064static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10065{
10066 struct btrfs_trans_handle *trans;
10067 struct btrfs_root *root = BTRFS_I(dir)->root;
10068 struct inode *inode = NULL;
10069 u64 objectid;
10070 u64 index;
10071 int ret = 0;
10072
10073 /*
10074 * 5 units required for adding orphan entry
10075 */
10076 trans = btrfs_start_transaction(root, 5);
10077 if (IS_ERR(trans))
10078 return PTR_ERR(trans);
10079
10080 ret = btrfs_find_free_ino(root, &objectid);
10081 if (ret)
10082 goto out;
10083
10084 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10085 btrfs_ino(dir), objectid, mode, &index);
10086 if (IS_ERR(inode)) {
10087 ret = PTR_ERR(inode);
10088 inode = NULL;
10089 goto out;
10090 }
10091
ef3b9af5
FM		 10092 inode->i_fop = &btrfs_file_operations;
10093 inode->i_op = &btrfs_file_inode_operations;
10094
10095 inode->i_mapping->a_ops = &btrfs_aops;
ef3b9af5
FM		 10096 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10097
b0d5d10f
CM		 10098 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10099 if (ret)
10100 goto out_inode;
10101
10102 ret = btrfs_update_inode(trans, root, inode);
10103 if (ret)
10104 goto out_inode;
ef3b9af5
FM		 10105 ret = btrfs_orphan_add(trans, inode);
10106 if (ret)
b0d5d10f 10107 goto out_inode;
ef3b9af5 10108
5762b5c9
FM		 10109 /*
10110 * We set number of links to 0 in btrfs_new_inode(), and here we set
10111 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10112 * through:
10113 *
10114 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10115 */
10116 set_nlink(inode, 1);
b0d5d10f 10117 unlock_new_inode(inode);
ef3b9af5
FM		 10118 d_tmpfile(dentry, inode);
10119 mark_inode_dirty(inode);
10120
10121out:
10122 btrfs_end_transaction(trans, root);
10123 if (ret)
10124 iput(inode);
10125 btrfs_balance_delayed_items(root);
10126 btrfs_btree_balance_dirty(root);
ef3b9af5 10127 return ret;
b0d5d10f
CM		 10128
10129out_inode:
10130 unlock_new_inode(inode);
10131 goto out;
10132
ef3b9af5
FM		 10133}
10134
b38ef71c
FM		 10135/* Inspired by filemap_check_errors() */
10136int btrfs_inode_check_errors(struct inode *inode)
10137{
10138 int ret = 0;
10139
10140 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
10141 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
10142 ret = -ENOSPC;
10143 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
10144 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
10145 ret = -EIO;
10146
10147 return ret;
10148}
10149
6e1d5dcc 10150static const struct inode_operations btrfs_dir_inode_operations = {
3394e160 10151 .getattr = btrfs_getattr,
39279cc3
CM		 10152 .lookup = btrfs_lookup,
10153 .create = btrfs_create,
10154 .unlink = btrfs_unlink,
10155 .link = btrfs_link,
10156 .mkdir = btrfs_mkdir,
10157 .rmdir = btrfs_rmdir,
80ace85c 10158 .rename2 = btrfs_rename2,
39279cc3
CM		 10159 .symlink = btrfs_symlink,
10160 .setattr = btrfs_setattr,
618e21d5 10161 .mknod = btrfs_mknod,
95819c05 10162 .setxattr = btrfs_setxattr,
9172abbc 10163 .getxattr = generic_getxattr,
5103e947 10164 .listxattr = btrfs_listxattr,
95819c05 10165 .removexattr = btrfs_removexattr,
fdebe2bd 10166 .permission = btrfs_permission,
4e34e719 10167 .get_acl = btrfs_get_acl,
996a710d 10168 .set_acl = btrfs_set_acl,
93fd63c2 10169 .update_time = btrfs_update_time,
ef3b9af5 10170 .tmpfile = btrfs_tmpfile,
39279cc3 10171};
6e1d5dcc 10172static const struct inode_operations btrfs_dir_ro_inode_operations = {
39279cc3 10173 .lookup = btrfs_lookup,
fdebe2bd 10174 .permission = btrfs_permission,
4e34e719 10175 .get_acl = btrfs_get_acl,
996a710d 10176 .set_acl = btrfs_set_acl,
93fd63c2 10177 .update_time = btrfs_update_time,
39279cc3 10178};
76dda93c 10179
828c0950 10180static const struct file_operations btrfs_dir_file_operations = {
39279cc3
CM		 10181 .llseek = generic_file_llseek,
10182 .read = generic_read_dir,
9cdda8d3 10183 .iterate = btrfs_real_readdir,
34287aa3 10184 .unlocked_ioctl = btrfs_ioctl,
39279cc3 10185#ifdef CONFIG_COMPAT
34287aa3 10186 .compat_ioctl = btrfs_ioctl,
39279cc3 10187#endif
6bf13c0c 10188 .release = btrfs_release_file,
e02119d5 10189 .fsync = btrfs_sync_file,
39279cc3
CM		 10190};
10191
20e5506b 10192static const struct extent_io_ops btrfs_extent_io_ops = {
07157aac 10193 .fill_delalloc = run_delalloc_range,
065631f6 10194 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 10195 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 10196 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 10197 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 10198 .writepage_start_hook = btrfs_writepage_start_hook,
b0c68f8b
CM		 10199 .set_bit_hook = btrfs_set_bit_hook,
10200 .clear_bit_hook = btrfs_clear_bit_hook,
9ed74f2d
JB		 10201 .merge_extent_hook = btrfs_merge_extent_hook,
10202 .split_extent_hook = btrfs_split_extent_hook,
07157aac
CM		 10203};
10204
35054394
CM		 10205/*
10206 * btrfs doesn't support the bmap operation because swapfiles
10207 * use bmap to make a mapping of extents in the file. They assume
10208 * these extents won't change over the life of the file and they
10209 * use the bmap result to do IO directly to the drive.
10210 *
10211 * the btrfs bmap call would return logical addresses that aren't
10212 * suitable for IO and they also will change frequently as COW
10213 * operations happen. So, swapfile + btrfs == corruption.
10214 *
10215 * For now we're avoiding this by dropping bmap.
10216 */
7f09410b 10217static const struct address_space_operations btrfs_aops = {
39279cc3
CM		 10218 .readpage = btrfs_readpage,
10219 .writepage = btrfs_writepage,
b293f02e 10220 .writepages = btrfs_writepages,
3ab2fb5a 10221 .readpages = btrfs_readpages,
16432985 10222 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 10223 .invalidatepage = btrfs_invalidatepage,
10224 .releasepage = btrfs_releasepage,
e6dcd2dc 10225 .set_page_dirty = btrfs_set_page_dirty,
465fdd97 10226 .error_remove_page = generic_error_remove_page,
39279cc3
CM		 10227};
10228
7f09410b 10229static const struct address_space_operations btrfs_symlink_aops = {
39279cc3
CM		 10230 .readpage = btrfs_readpage,
10231 .writepage = btrfs_writepage,
2bf5a725
CM		 10232 .invalidatepage = btrfs_invalidatepage,
10233 .releasepage = btrfs_releasepage,
39279cc3
CM		 10234};
10235
6e1d5dcc 10236static const struct inode_operations btrfs_file_inode_operations = {
39279cc3
CM		 10237 .getattr = btrfs_getattr,
10238 .setattr = btrfs_setattr,
95819c05 10239 .setxattr = btrfs_setxattr,
9172abbc 10240 .getxattr = generic_getxattr,
5103e947 10241 .listxattr = btrfs_listxattr,
95819c05 10242 .removexattr = btrfs_removexattr,
fdebe2bd 10243 .permission = btrfs_permission,
1506fcc8 10244 .fiemap = btrfs_fiemap,
4e34e719 10245 .get_acl = btrfs_get_acl,
996a710d 10246 .set_acl = btrfs_set_acl,
e41f941a 10247 .update_time = btrfs_update_time,
39279cc3 10248};
6e1d5dcc 10249static const struct inode_operations btrfs_special_inode_operations = {
618e21d5
JB		 10250 .getattr = btrfs_getattr,
10251 .setattr = btrfs_setattr,
fdebe2bd 10252 .permission = btrfs_permission,
95819c05 10253 .setxattr = btrfs_setxattr,
9172abbc 10254 .getxattr = generic_getxattr,
33268eaf 10255 .listxattr = btrfs_listxattr,
95819c05 10256 .removexattr = btrfs_removexattr,
4e34e719 10257 .get_acl = btrfs_get_acl,
996a710d 10258 .set_acl = btrfs_set_acl,
e41f941a 10259 .update_time = btrfs_update_time,
618e21d5 10260};
6e1d5dcc 10261static const struct inode_operations btrfs_symlink_inode_operations = {
39279cc3 10262 .readlink = generic_readlink,
6b255391 10263 .get_link = page_get_link,
f209561a 10264 .getattr = btrfs_getattr,
22c44fe6 10265 .setattr = btrfs_setattr,
fdebe2bd 10266 .permission = btrfs_permission,
0279b4cd 10267 .setxattr = btrfs_setxattr,
9172abbc 10268 .getxattr = generic_getxattr,
0279b4cd
JO		 10269 .listxattr = btrfs_listxattr,
10270 .removexattr = btrfs_removexattr,
e41f941a 10271 .update_time = btrfs_update_time,
39279cc3 10272};
76dda93c 10273
82d339d9 10274const struct dentry_operations btrfs_dentry_operations = {
76dda93c 10275 .d_delete = btrfs_dentry_delete,
b4aff1f8 10276 .d_release = btrfs_dentry_release,
76dda93c 10277};
Linux 6.x block layer and io_uring tree(s)