projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: just free dummy extent buffers
[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>
a27bb332 35#include <linux/aio.h>
9ebefb18 36#include <linux/bit_spinlock.h>
5103e947 37#include <linux/xattr.h>
33268eaf 38#include <linux/posix_acl.h>
d899e052 39#include <linux/falloc.h>
5a0e3ad6 40#include <linux/slab.h>
7a36ddec 41#include <linux/ratelimit.h>
22c44fe6 42#include <linux/mount.h>
55e301fd 43#include <linux/btrfs.h>
53b381b3 44#include <linux/blkdev.h>
f23b5a59 45#include <linux/posix_acl_xattr.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"
39279cc3
CM		 62
63struct btrfs_iget_args {
90d3e592 64 struct btrfs_key *location;
39279cc3
CM		 65 struct btrfs_root *root;
66};
67
6e1d5dcc
AD		 68static const struct inode_operations btrfs_dir_inode_operations;
69static const struct inode_operations btrfs_symlink_inode_operations;
70static const struct inode_operations btrfs_dir_ro_inode_operations;
71static const struct inode_operations btrfs_special_inode_operations;
72static const struct inode_operations btrfs_file_inode_operations;
7f09410b
AD		 73static const struct address_space_operations btrfs_aops;
74static const struct address_space_operations btrfs_symlink_aops;
828c0950 75static const struct file_operations btrfs_dir_file_operations;
d1310b2e 76static struct extent_io_ops btrfs_extent_io_ops;
39279cc3
CM		 77
78static struct kmem_cache *btrfs_inode_cachep;
8ccf6f19 79static struct kmem_cache *btrfs_delalloc_work_cachep;
39279cc3
CM		 80struct kmem_cache *btrfs_trans_handle_cachep;
81struct kmem_cache *btrfs_transaction_cachep;
39279cc3 82struct kmem_cache *btrfs_path_cachep;
dc89e982 83struct kmem_cache *btrfs_free_space_cachep;
39279cc3
CM		 84
85#define S_SHIFT 12
86static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
87 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
88 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
89 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
90 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
91 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
92 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
93 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
94};
95
3972f260 96static int btrfs_setsize(struct inode *inode, struct iattr *attr);
a41ad394 97static int btrfs_truncate(struct inode *inode);
5fd02043 98static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
771ed689
CM		 99static noinline int cow_file_range(struct inode *inode,
100 struct page *locked_page,
101 u64 start, u64 end, int *page_started,
102 unsigned long *nr_written, int unlock);
70c8a91c
JB		 103static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
104 u64 len, u64 orig_start,
105 u64 block_start, u64 block_len,
cc95bef6
JB		 106 u64 orig_block_len, u64 ram_bytes,
107 int type);
7b128766 108
48a3b636 109static int btrfs_dirty_inode(struct inode *inode);
7b128766 110
f34f57a3 111static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
2a7dba39
EP		 112 struct inode *inode, struct inode *dir,
113 const struct qstr *qstr)
0279b4cd
JO		 114{
115 int err;
116
f34f57a3 117 err = btrfs_init_acl(trans, inode, dir);
0279b4cd 118 if (!err)
2a7dba39 119 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
0279b4cd
JO		 120 return err;
121}
122
c8b97818
CM		 123/*
124 * this does all the hard work for inserting an inline extent into
125 * the btree. The caller should have done a btrfs_drop_extents so that
126 * no overlapping inline items exist in the btree
127 */
40f76580 128static int insert_inline_extent(struct btrfs_trans_handle *trans,
1acae57b 129 struct btrfs_path *path, int extent_inserted,
c8b97818
CM		 130 struct btrfs_root *root, struct inode *inode,
131 u64 start, size_t size, size_t compressed_size,
fe3f566c 132 int compress_type,
c8b97818
CM		 133 struct page **compressed_pages)
134{
c8b97818
CM		 135 struct extent_buffer *leaf;
136 struct page *page = NULL;
137 char *kaddr;
138 unsigned long ptr;
139 struct btrfs_file_extent_item *ei;
140 int err = 0;
141 int ret;
142 size_t cur_size = size;
c8b97818 143 unsigned long offset;
c8b97818 144
fe3f566c 145 if (compressed_size && compressed_pages)
c8b97818 146 cur_size = compressed_size;
c8b97818 147
1acae57b 148 inode_add_bytes(inode, size);
c8b97818 149
1acae57b
FDBM		 150 if (!extent_inserted) {
151 struct btrfs_key key;
152 size_t datasize;
c8b97818 153
1acae57b
FDBM		 154 key.objectid = btrfs_ino(inode);
155 key.offset = start;
962a298f 156 key.type = BTRFS_EXTENT_DATA_KEY;
c8b97818 157
1acae57b
FDBM		 158 datasize = btrfs_file_extent_calc_inline_size(cur_size);
159 path->leave_spinning = 1;
160 ret = btrfs_insert_empty_item(trans, root, path, &key,
161 datasize);
162 if (ret) {
163 err = ret;
164 goto fail;
165 }
c8b97818
CM		 166 }
167 leaf = path->nodes[0];
168 ei = btrfs_item_ptr(leaf, path->slots[0],
169 struct btrfs_file_extent_item);
170 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
171 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
172 btrfs_set_file_extent_encryption(leaf, ei, 0);
173 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
174 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
175 ptr = btrfs_file_extent_inline_start(ei);
176
261507a0 177 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818
CM		 178 struct page *cpage;
179 int i = 0;
d397712b 180 while (compressed_size > 0) {
c8b97818 181 cpage = compressed_pages[i];
5b050f04 182 cur_size = min_t(unsigned long, compressed_size,
c8b97818
CM		 183 PAGE_CACHE_SIZE);
184
7ac687d9 185 kaddr = kmap_atomic(cpage);
c8b97818 186 write_extent_buffer(leaf, kaddr, ptr, cur_size);
7ac687d9 187 kunmap_atomic(kaddr);
c8b97818
CM		 188
189 i++;
190 ptr += cur_size;
191 compressed_size -= cur_size;
192 }
193 btrfs_set_file_extent_compression(leaf, ei,
261507a0 194 compress_type);
c8b97818
CM		 195 } else {
196 page = find_get_page(inode->i_mapping,
197 start >> PAGE_CACHE_SHIFT);
198 btrfs_set_file_extent_compression(leaf, ei, 0);
7ac687d9 199 kaddr = kmap_atomic(page);
c8b97818
CM		 200 offset = start & (PAGE_CACHE_SIZE - 1);
201 write_extent_buffer(leaf, kaddr + offset, ptr, size);
7ac687d9 202 kunmap_atomic(kaddr);
c8b97818
CM		 203 page_cache_release(page);
204 }
205 btrfs_mark_buffer_dirty(leaf);
1acae57b 206 btrfs_release_path(path);
c8b97818 207
c2167754
YZ		 208 /*
209 * we're an inline extent, so nobody can
210 * extend the file past i_size without locking
211 * a page we already have locked.
212 *
213 * We must do any isize and inode updates
214 * before we unlock the pages. Otherwise we
215 * could end up racing with unlink.
216 */
c8b97818 217 BTRFS_I(inode)->disk_i_size = inode->i_size;
79787eaa 218 ret = btrfs_update_inode(trans, root, inode);
c2167754 219
79787eaa 220 return ret;
c8b97818 221fail:
c8b97818
CM		 222 return err;
223}
224
225
226/*
227 * conditionally insert an inline extent into the file. This
228 * does the checks required to make sure the data is small enough
229 * to fit as an inline extent.
230 */
00361589
JB		 231static noinline int cow_file_range_inline(struct btrfs_root *root,
232 struct inode *inode, u64 start,
233 u64 end, size_t compressed_size,
234 int compress_type,
235 struct page **compressed_pages)
c8b97818 236{
00361589 237 struct btrfs_trans_handle *trans;
c8b97818
CM		 238 u64 isize = i_size_read(inode);
239 u64 actual_end = min(end + 1, isize);
240 u64 inline_len = actual_end - start;
fda2832f 241 u64 aligned_end = ALIGN(end, root->sectorsize);
c8b97818
CM		 242 u64 data_len = inline_len;
243 int ret;
1acae57b
FDBM		 244 struct btrfs_path *path;
245 int extent_inserted = 0;
246 u32 extent_item_size;
c8b97818
CM		 247
248 if (compressed_size)
249 data_len = compressed_size;
250
251 if (start > 0 ||
354877be
WS		 252 actual_end > PAGE_CACHE_SIZE ||
253 data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
c8b97818
CM		 254 (!compressed_size &&
255 (actual_end & (root->sectorsize - 1)) == 0) ||
256 end + 1 < isize ||
257 data_len > root->fs_info->max_inline) {
258 return 1;
259 }
260
1acae57b
FDBM		 261 path = btrfs_alloc_path();
262 if (!path)
263 return -ENOMEM;
264
00361589 265 trans = btrfs_join_transaction(root);
1acae57b
FDBM		 266 if (IS_ERR(trans)) {
267 btrfs_free_path(path);
00361589 268 return PTR_ERR(trans);
1acae57b 269 }
00361589
JB		 270 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
271
1acae57b
FDBM		 272 if (compressed_size && compressed_pages)
273 extent_item_size = btrfs_file_extent_calc_inline_size(
274 compressed_size);
275 else
276 extent_item_size = btrfs_file_extent_calc_inline_size(
277 inline_len);
278
279 ret = __btrfs_drop_extents(trans, root, inode, path,
280 start, aligned_end, NULL,
281 1, 1, extent_item_size, &extent_inserted);
00361589
JB		 282 if (ret) {
283 btrfs_abort_transaction(trans, root, ret);
284 goto out;
285 }
c8b97818
CM		 286
287 if (isize > actual_end)
288 inline_len = min_t(u64, isize, actual_end);
1acae57b
FDBM		 289 ret = insert_inline_extent(trans, path, extent_inserted,
290 root, inode, start,
c8b97818 291 inline_len, compressed_size,
fe3f566c 292 compress_type, compressed_pages);
2adcac1a 293 if (ret && ret != -ENOSPC) {
79787eaa 294 btrfs_abort_transaction(trans, root, ret);
00361589 295 goto out;
2adcac1a 296 } else if (ret == -ENOSPC) {
00361589
JB		 297 ret = 1;
298 goto out;
79787eaa 299 }
2adcac1a 300
bdc20e67 301 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
0ca1f7ce 302 btrfs_delalloc_release_metadata(inode, end + 1 - start);
a1ed835e 303 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
00361589 304out:
1acae57b 305 btrfs_free_path(path);
00361589
JB		 306 btrfs_end_transaction(trans, root);
307 return ret;
c8b97818
CM		 308}
309
771ed689
CM		 310struct async_extent {
311 u64 start;
312 u64 ram_size;
313 u64 compressed_size;
314 struct page **pages;
315 unsigned long nr_pages;
261507a0 316 int compress_type;
771ed689
CM		 317 struct list_head list;
318};
319
320struct async_cow {
321 struct inode *inode;
322 struct btrfs_root *root;
323 struct page *locked_page;
324 u64 start;
325 u64 end;
326 struct list_head extents;
327 struct btrfs_work work;
328};
329
330static noinline int add_async_extent(struct async_cow *cow,
331 u64 start, u64 ram_size,
332 u64 compressed_size,
333 struct page **pages,
261507a0
LZ		 334 unsigned long nr_pages,
335 int compress_type)
771ed689
CM		 336{
337 struct async_extent *async_extent;
338
339 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
79787eaa 340 BUG_ON(!async_extent); /* -ENOMEM */
771ed689
CM		 341 async_extent->start = start;
342 async_extent->ram_size = ram_size;
343 async_extent->compressed_size = compressed_size;
344 async_extent->pages = pages;
345 async_extent->nr_pages = nr_pages;
261507a0 346 async_extent->compress_type = compress_type;
771ed689
CM		 347 list_add_tail(&async_extent->list, &cow->extents);
348 return 0;
349}
350
f79707b0
WS		 351static inline int inode_need_compress(struct inode *inode)
352{
353 struct btrfs_root *root = BTRFS_I(inode)->root;
354
355 /* force compress */
356 if (btrfs_test_opt(root, FORCE_COMPRESS))
357 return 1;
358 /* bad compression ratios */
359 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
360 return 0;
361 if (btrfs_test_opt(root, COMPRESS) ||
362 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
363 BTRFS_I(inode)->force_compress)
364 return 1;
365 return 0;
366}
367
d352ac68 368/*
771ed689
CM		 369 * we create compressed extents in two phases. The first
370 * phase compresses a range of pages that have already been
371 * locked (both pages and state bits are locked).
c8b97818 372 *
771ed689
CM		 373 * This is done inside an ordered work queue, and the compression
374 * is spread across many cpus. The actual IO submission is step
375 * two, and the ordered work queue takes care of making sure that
376 * happens in the same order things were put onto the queue by
377 * writepages and friends.
c8b97818 378 *
771ed689
CM		 379 * If this code finds it can't get good compression, it puts an
380 * entry onto the work queue to write the uncompressed bytes. This
381 * makes sure that both compressed inodes and uncompressed inodes
b2570314
AB		 382 * are written in the same order that the flusher thread sent them
383 * down.
d352ac68 384 */
c44f649e 385static noinline void compress_file_range(struct inode *inode,
771ed689
CM		 386 struct page *locked_page,
387 u64 start, u64 end,
388 struct async_cow *async_cow,
389 int *num_added)
b888db2b
CM		 390{
391 struct btrfs_root *root = BTRFS_I(inode)->root;
db94535d 392 u64 num_bytes;
db94535d 393 u64 blocksize = root->sectorsize;
c8b97818 394 u64 actual_end;
42dc7bab 395 u64 isize = i_size_read(inode);
e6dcd2dc 396 int ret = 0;
c8b97818
CM		 397 struct page **pages = NULL;
398 unsigned long nr_pages;
399 unsigned long nr_pages_ret = 0;
400 unsigned long total_compressed = 0;
401 unsigned long total_in = 0;
402 unsigned long max_compressed = 128 * 1024;
771ed689 403 unsigned long max_uncompressed = 128 * 1024;
c8b97818
CM		 404 int i;
405 int will_compress;
261507a0 406 int compress_type = root->fs_info->compress_type;
4adaa611 407 int redirty = 0;
b888db2b 408
4cb13e5d
LB		 409 /* if this is a small write inside eof, kick off a defrag */
410 if ((end - start + 1) < 16 * 1024 &&
411 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
4cb5300b
CM		 412 btrfs_add_inode_defrag(NULL, inode);
413
42dc7bab 414 actual_end = min_t(u64, isize, end + 1);
c8b97818
CM		 415again:
416 will_compress = 0;
417 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
418 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
be20aa9d 419
f03d9301
CM		 420 /*
421 * we don't want to send crud past the end of i_size through
422 * compression, that's just a waste of CPU time. So, if the
423 * end of the file is before the start of our current
424 * requested range of bytes, we bail out to the uncompressed
425 * cleanup code that can deal with all of this.
426 *
427 * It isn't really the fastest way to fix things, but this is a
428 * very uncommon corner.
429 */
430 if (actual_end <= start)
431 goto cleanup_and_bail_uncompressed;
432
c8b97818
CM		 433 total_compressed = actual_end - start;
434
4bcbb332
SW		 435 /*
436 * skip compression for a small file range(<=blocksize) that
437 * isn't an inline extent, since it dosen't save disk space at all.
438 */
439 if (total_compressed <= blocksize &&
440 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
441 goto cleanup_and_bail_uncompressed;
442
c8b97818
CM		 443 /* we want to make sure that amount of ram required to uncompress
444 * an extent is reasonable, so we limit the total size in ram
771ed689
CM		 445 * of a compressed extent to 128k. This is a crucial number
446 * because it also controls how easily we can spread reads across
447 * cpus for decompression.
448 *
449 * We also want to make sure the amount of IO required to do
450 * a random read is reasonably small, so we limit the size of
451 * a compressed extent to 128k.
c8b97818
CM		 452 */
453 total_compressed = min(total_compressed, max_uncompressed);
fda2832f 454 num_bytes = ALIGN(end - start + 1, blocksize);
be20aa9d 455 num_bytes = max(blocksize, num_bytes);
c8b97818
CM		 456 total_in = 0;
457 ret = 0;
db94535d 458
771ed689
CM		 459 /*
460 * we do compression for mount -o compress and when the
461 * inode has not been flagged as nocompress. This flag can
462 * change at any time if we discover bad compression ratios.
c8b97818 463 */
f79707b0 464 if (inode_need_compress(inode)) {
c8b97818 465 WARN_ON(pages);
cfbc246e 466 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
560f7d75
LZ		 467 if (!pages) {
468 /* just bail out to the uncompressed code */
469 goto cont;
470 }
c8b97818 471
261507a0
LZ		 472 if (BTRFS_I(inode)->force_compress)
473 compress_type = BTRFS_I(inode)->force_compress;
474
4adaa611
CM		 475 /*
476 * we need to call clear_page_dirty_for_io on each
477 * page in the range. Otherwise applications with the file
478 * mmap'd can wander in and change the page contents while
479 * we are compressing them.
480 *
481 * If the compression fails for any reason, we set the pages
482 * dirty again later on.
483 */
484 extent_range_clear_dirty_for_io(inode, start, end);
485 redirty = 1;
261507a0
LZ		 486 ret = btrfs_compress_pages(compress_type,
487 inode->i_mapping, start,
488 total_compressed, pages,
489 nr_pages, &nr_pages_ret,
490 &total_in,
491 &total_compressed,
492 max_compressed);
c8b97818
CM		 493
494 if (!ret) {
495 unsigned long offset = total_compressed &
496 (PAGE_CACHE_SIZE - 1);
497 struct page *page = pages[nr_pages_ret - 1];
498 char *kaddr;
499
500 /* zero the tail end of the last page, we might be
501 * sending it down to disk
502 */
503 if (offset) {
7ac687d9 504 kaddr = kmap_atomic(page);
c8b97818
CM		 505 memset(kaddr + offset, 0,
506 PAGE_CACHE_SIZE - offset);
7ac687d9 507 kunmap_atomic(kaddr);
c8b97818
CM		 508 }
509 will_compress = 1;
510 }
511 }
560f7d75 512cont:
c8b97818
CM		 513 if (start == 0) {
514 /* lets try to make an inline extent */
771ed689 515 if (ret || total_in < (actual_end - start)) {
c8b97818 516 /* we didn't compress the entire range, try
771ed689 517 * to make an uncompressed inline extent.
c8b97818 518 */
00361589
JB		 519 ret = cow_file_range_inline(root, inode, start, end,
520 0, 0, NULL);
c8b97818 521 } else {
771ed689 522 /* try making a compressed inline extent */
00361589 523 ret = cow_file_range_inline(root, inode, start, end,
fe3f566c
LZ		 524 total_compressed,
525 compress_type, pages);
c8b97818 526 }
79787eaa 527 if (ret <= 0) {
151a41bc
JB		 528 unsigned long clear_flags = EXTENT_DELALLOC |
529 EXTENT_DEFRAG;
e6eb4314
FM		 530 unsigned long page_error_op;
531
151a41bc 532 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
e6eb4314 533 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
151a41bc 534
771ed689 535 /*
79787eaa
JM		 536 * inline extent creation worked or returned error,
537 * we don't need to create any more async work items.
538 * Unlock and free up our temp pages.
771ed689 539 */
c2790a2e 540 extent_clear_unlock_delalloc(inode, start, end, NULL,
151a41bc 541 clear_flags, PAGE_UNLOCK |
c2790a2e
JB		 542 PAGE_CLEAR_DIRTY |
543 PAGE_SET_WRITEBACK |
e6eb4314 544 page_error_op |
c2790a2e 545 PAGE_END_WRITEBACK);
c8b97818
CM		 546 goto free_pages_out;
547 }
548 }
549
550 if (will_compress) {
551 /*
552 * we aren't doing an inline extent round the compressed size
553 * up to a block size boundary so the allocator does sane
554 * things
555 */
fda2832f 556 total_compressed = ALIGN(total_compressed, blocksize);
c8b97818
CM		 557
558 /*
559 * one last check to make sure the compression is really a
560 * win, compare the page count read with the blocks on disk
561 */
fda2832f 562 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
c8b97818
CM		 563 if (total_compressed >= total_in) {
564 will_compress = 0;
565 } else {
c8b97818
CM		 566 num_bytes = total_in;
567 }
568 }
569 if (!will_compress && pages) {
570 /*
571 * the compression code ran but failed to make things smaller,
572 * free any pages it allocated and our page pointer array
573 */
574 for (i = 0; i < nr_pages_ret; i++) {
70b99e69 575 WARN_ON(pages[i]->mapping);
c8b97818
CM		 576 page_cache_release(pages[i]);
577 }
578 kfree(pages);
579 pages = NULL;
580 total_compressed = 0;
581 nr_pages_ret = 0;
582
583 /* flag the file so we don't compress in the future */
1e701a32
CM		 584 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
585 !(BTRFS_I(inode)->force_compress)) {
a555f810 586 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
1e701a32 587 }
c8b97818 588 }
771ed689
CM		 589 if (will_compress) {
590 *num_added += 1;
c8b97818 591
771ed689
CM		 592 /* the async work queues will take care of doing actual
593 * allocation on disk for these compressed pages,
594 * and will submit them to the elevator.
595 */
596 add_async_extent(async_cow, start, num_bytes,
261507a0
LZ		 597 total_compressed, pages, nr_pages_ret,
598 compress_type);
179e29e4 599
24ae6365 600 if (start + num_bytes < end) {
771ed689
CM		 601 start += num_bytes;
602 pages = NULL;
603 cond_resched();
604 goto again;
605 }
606 } else {
f03d9301 607cleanup_and_bail_uncompressed:
771ed689
CM		 608 /*
609 * No compression, but we still need to write the pages in
610 * the file we've been given so far. redirty the locked
611 * page if it corresponds to our extent and set things up
612 * for the async work queue to run cow_file_range to do
613 * the normal delalloc dance
614 */
615 if (page_offset(locked_page) >= start &&
616 page_offset(locked_page) <= end) {
617 __set_page_dirty_nobuffers(locked_page);
618 /* unlocked later on in the async handlers */
619 }
4adaa611
CM		 620 if (redirty)
621 extent_range_redirty_for_io(inode, start, end);
261507a0
LZ		 622 add_async_extent(async_cow, start, end - start + 1,
623 0, NULL, 0, BTRFS_COMPRESS_NONE);
771ed689
CM		 624 *num_added += 1;
625 }
3b951516 626
c44f649e 627 return;
771ed689
CM		 628
629free_pages_out:
630 for (i = 0; i < nr_pages_ret; i++) {
631 WARN_ON(pages[i]->mapping);
632 page_cache_release(pages[i]);
633 }
d397712b 634 kfree(pages);
771ed689 635}
771ed689 636
40ae837b
FM		 637static void free_async_extent_pages(struct async_extent *async_extent)
638{
639 int i;
640
641 if (!async_extent->pages)
642 return;
643
644 for (i = 0; i < async_extent->nr_pages; i++) {
645 WARN_ON(async_extent->pages[i]->mapping);
646 page_cache_release(async_extent->pages[i]);
647 }
648 kfree(async_extent->pages);
649 async_extent->nr_pages = 0;
650 async_extent->pages = NULL;
771ed689
CM		 651}
652
653/*
654 * phase two of compressed writeback. This is the ordered portion
655 * of the code, which only gets called in the order the work was
656 * queued. We walk all the async extents created by compress_file_range
657 * and send them down to the disk.
658 */
dec8f175 659static noinline void submit_compressed_extents(struct inode *inode,
771ed689
CM		 660 struct async_cow *async_cow)
661{
662 struct async_extent *async_extent;
663 u64 alloc_hint = 0;
771ed689
CM		 664 struct btrfs_key ins;
665 struct extent_map *em;
666 struct btrfs_root *root = BTRFS_I(inode)->root;
667 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
668 struct extent_io_tree *io_tree;
f5a84ee3 669 int ret = 0;
771ed689 670
3e04e7f1 671again:
d397712b 672 while (!list_empty(&async_cow->extents)) {
771ed689
CM		 673 async_extent = list_entry(async_cow->extents.next,
674 struct async_extent, list);
675 list_del(&async_extent->list);
c8b97818 676
771ed689
CM		 677 io_tree = &BTRFS_I(inode)->io_tree;
678
f5a84ee3 679retry:
771ed689
CM		 680 /* did the compression code fall back to uncompressed IO? */
681 if (!async_extent->pages) {
682 int page_started = 0;
683 unsigned long nr_written = 0;
684
685 lock_extent(io_tree, async_extent->start,
2ac55d41 686 async_extent->start +
d0082371 687 async_extent->ram_size - 1);
771ed689
CM		 688
689 /* allocate blocks */
f5a84ee3
JB		 690 ret = cow_file_range(inode, async_cow->locked_page,
691 async_extent->start,
692 async_extent->start +
693 async_extent->ram_size - 1,
694 &page_started, &nr_written, 0);
771ed689 695
79787eaa
JM		 696 /* JDM XXX */
697
771ed689
CM		 698 /*
699 * if page_started, cow_file_range inserted an
700 * inline extent and took care of all the unlocking
701 * and IO for us. Otherwise, we need to submit
702 * all those pages down to the drive.
703 */
f5a84ee3 704 if (!page_started && !ret)
771ed689
CM		 705 extent_write_locked_range(io_tree,
706 inode, async_extent->start,
d397712b 707 async_extent->start +
771ed689
CM		 708 async_extent->ram_size - 1,
709 btrfs_get_extent,
710 WB_SYNC_ALL);
3e04e7f1
JB		 711 else if (ret)
712 unlock_page(async_cow->locked_page);
771ed689
CM		 713 kfree(async_extent);
714 cond_resched();
715 continue;
716 }
717
718 lock_extent(io_tree, async_extent->start,
d0082371 719 async_extent->start + async_extent->ram_size - 1);
771ed689 720
00361589 721 ret = btrfs_reserve_extent(root,
771ed689
CM		 722 async_extent->compressed_size,
723 async_extent->compressed_size,
e570fd27 724 0, alloc_hint, &ins, 1, 1);
f5a84ee3 725 if (ret) {
40ae837b 726 free_async_extent_pages(async_extent);
3e04e7f1 727
fdf8e2ea
JB		 728 if (ret == -ENOSPC) {
729 unlock_extent(io_tree, async_extent->start,
730 async_extent->start +
731 async_extent->ram_size - 1);
ce62003f
LB		 732
733 /*
734 * we need to redirty the pages if we decide to
735 * fallback to uncompressed IO, otherwise we
736 * will not submit these pages down to lower
737 * layers.
738 */
739 extent_range_redirty_for_io(inode,
740 async_extent->start,
741 async_extent->start +
742 async_extent->ram_size - 1);
743
79787eaa 744 goto retry;
fdf8e2ea 745 }
3e04e7f1 746 goto out_free;
f5a84ee3
JB		 747 }
748
c2167754
YZ		 749 /*
750 * here we're doing allocation and writeback of the
751 * compressed pages
752 */
753 btrfs_drop_extent_cache(inode, async_extent->start,
754 async_extent->start +
755 async_extent->ram_size - 1, 0);
756
172ddd60 757 em = alloc_extent_map();
b9aa55be
LB		 758 if (!em) {
759 ret = -ENOMEM;
3e04e7f1 760 goto out_free_reserve;
b9aa55be 761 }
771ed689
CM		 762 em->start = async_extent->start;
763 em->len = async_extent->ram_size;
445a6944 764 em->orig_start = em->start;
2ab28f32
JB		 765 em->mod_start = em->start;
766 em->mod_len = em->len;
c8b97818 767
771ed689
CM		 768 em->block_start = ins.objectid;
769 em->block_len = ins.offset;
b4939680 770 em->orig_block_len = ins.offset;
cc95bef6 771 em->ram_bytes = async_extent->ram_size;
771ed689 772 em->bdev = root->fs_info->fs_devices->latest_bdev;
261507a0 773 em->compress_type = async_extent->compress_type;
771ed689
CM		 774 set_bit(EXTENT_FLAG_PINNED, &em->flags);
775 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
70c8a91c 776 em->generation = -1;
771ed689 777
d397712b 778 while (1) {
890871be 779 write_lock(&em_tree->lock);
09a2a8f9 780 ret = add_extent_mapping(em_tree, em, 1);
890871be 781 write_unlock(&em_tree->lock);
771ed689
CM		 782 if (ret != -EEXIST) {
783 free_extent_map(em);
784 break;
785 }
786 btrfs_drop_extent_cache(inode, async_extent->start,
787 async_extent->start +
788 async_extent->ram_size - 1, 0);
789 }
790
3e04e7f1
JB		 791 if (ret)
792 goto out_free_reserve;
793
261507a0
LZ		 794 ret = btrfs_add_ordered_extent_compress(inode,
795 async_extent->start,
796 ins.objectid,
797 async_extent->ram_size,
798 ins.offset,
799 BTRFS_ORDERED_COMPRESSED,
800 async_extent->compress_type);
d9f85963
FM		 801 if (ret) {
802 btrfs_drop_extent_cache(inode, async_extent->start,
803 async_extent->start +
804 async_extent->ram_size - 1, 0);
3e04e7f1 805 goto out_free_reserve;
d9f85963 806 }
771ed689 807
771ed689
CM		 808 /*
809 * clear dirty, set writeback and unlock the pages.
810 */
c2790a2e 811 extent_clear_unlock_delalloc(inode, async_extent->start,
a791e35e
CM		 812 async_extent->start +
813 async_extent->ram_size - 1,
151a41bc
JB		 814 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
815 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
c2790a2e 816 PAGE_SET_WRITEBACK);
771ed689 817 ret = btrfs_submit_compressed_write(inode,
d397712b
CM		 818 async_extent->start,
819 async_extent->ram_size,
820 ins.objectid,
821 ins.offset, async_extent->pages,
822 async_extent->nr_pages);
fce2a4e6
FM		 823 if (ret) {
824 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
825 struct page *p = async_extent->pages[0];
826 const u64 start = async_extent->start;
827 const u64 end = start + async_extent->ram_size - 1;
828
829 p->mapping = inode->i_mapping;
830 tree->ops->writepage_end_io_hook(p, start, end,
831 NULL, 0);
832 p->mapping = NULL;
833 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
834 PAGE_END_WRITEBACK |
835 PAGE_SET_ERROR);
40ae837b 836 free_async_extent_pages(async_extent);
fce2a4e6 837 }
771ed689
CM		 838 alloc_hint = ins.objectid + ins.offset;
839 kfree(async_extent);
840 cond_resched();
841 }
dec8f175 842 return;
3e04e7f1 843out_free_reserve:
e570fd27 844 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
79787eaa 845out_free:
c2790a2e 846 extent_clear_unlock_delalloc(inode, async_extent->start,
3e04e7f1
JB		 847 async_extent->start +
848 async_extent->ram_size - 1,
c2790a2e 849 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc
JB		 850 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
851 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
704de49d
FM		 852 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
853 PAGE_SET_ERROR);
40ae837b 854 free_async_extent_pages(async_extent);
79787eaa 855 kfree(async_extent);
3e04e7f1 856 goto again;
771ed689
CM		 857}
858
4b46fce2
JB		 859static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
860 u64 num_bytes)
861{
862 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
863 struct extent_map *em;
864 u64 alloc_hint = 0;
865
866 read_lock(&em_tree->lock);
867 em = search_extent_mapping(em_tree, start, num_bytes);
868 if (em) {
869 /*
870 * if block start isn't an actual block number then find the
871 * first block in this inode and use that as a hint. If that
872 * block is also bogus then just don't worry about it.
873 */
874 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
875 free_extent_map(em);
876 em = search_extent_mapping(em_tree, 0, 0);
877 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
878 alloc_hint = em->block_start;
879 if (em)
880 free_extent_map(em);
881 } else {
882 alloc_hint = em->block_start;
883 free_extent_map(em);
884 }
885 }
886 read_unlock(&em_tree->lock);
887
888 return alloc_hint;
889}
890
771ed689
CM		 891/*
892 * when extent_io.c finds a delayed allocation range in the file,
893 * the call backs end up in this code. The basic idea is to
894 * allocate extents on disk for the range, and create ordered data structs
895 * in ram to track those extents.
896 *
897 * locked_page is the page that writepage had locked already. We use
898 * it to make sure we don't do extra locks or unlocks.
899 *
900 * *page_started is set to one if we unlock locked_page and do everything
901 * required to start IO on it. It may be clean and already done with
902 * IO when we return.
903 */
00361589
JB		 904static noinline int cow_file_range(struct inode *inode,
905 struct page *locked_page,
906 u64 start, u64 end, int *page_started,
907 unsigned long *nr_written,
908 int unlock)
771ed689 909{
00361589 910 struct btrfs_root *root = BTRFS_I(inode)->root;
771ed689
CM		 911 u64 alloc_hint = 0;
912 u64 num_bytes;
913 unsigned long ram_size;
914 u64 disk_num_bytes;
915 u64 cur_alloc_size;
916 u64 blocksize = root->sectorsize;
771ed689
CM		 917 struct btrfs_key ins;
918 struct extent_map *em;
919 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
920 int ret = 0;
921
02ecd2c2
JB		 922 if (btrfs_is_free_space_inode(inode)) {
923 WARN_ON_ONCE(1);
29bce2f3
JB		 924 ret = -EINVAL;
925 goto out_unlock;
02ecd2c2 926 }
771ed689 927
fda2832f 928 num_bytes = ALIGN(end - start + 1, blocksize);
771ed689
CM		 929 num_bytes = max(blocksize, num_bytes);
930 disk_num_bytes = num_bytes;
771ed689 931
4cb5300b 932 /* if this is a small write inside eof, kick off defrag */
4cb13e5d
LB		 933 if (num_bytes < 64 * 1024 &&
934 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
00361589 935 btrfs_add_inode_defrag(NULL, inode);
4cb5300b 936
771ed689
CM		 937 if (start == 0) {
938 /* lets try to make an inline extent */
00361589
JB		 939 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
940 NULL);
771ed689 941 if (ret == 0) {
c2790a2e
JB		 942 extent_clear_unlock_delalloc(inode, start, end, NULL,
943 EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc 944 EXTENT_DEFRAG, PAGE_UNLOCK |
c2790a2e
JB		 945 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
946 PAGE_END_WRITEBACK);
c2167754 947
771ed689
CM		 948 *nr_written = *nr_written +
949 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
950 *page_started = 1;
771ed689 951 goto out;
79787eaa 952 } else if (ret < 0) {
79787eaa 953 goto out_unlock;
771ed689
CM		 954 }
955 }
956
957 BUG_ON(disk_num_bytes >
6c41761f 958 btrfs_super_total_bytes(root->fs_info->super_copy));
771ed689 959
4b46fce2 960 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
771ed689
CM		 961 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
962
d397712b 963 while (disk_num_bytes > 0) {
a791e35e
CM		 964 unsigned long op;
965
287a0ab9 966 cur_alloc_size = disk_num_bytes;
00361589 967 ret = btrfs_reserve_extent(root, cur_alloc_size,
771ed689 968 root->sectorsize, 0, alloc_hint,
e570fd27 969 &ins, 1, 1);
00361589 970 if (ret < 0)
79787eaa 971 goto out_unlock;
d397712b 972
172ddd60 973 em = alloc_extent_map();
b9aa55be
LB		 974 if (!em) {
975 ret = -ENOMEM;
ace68bac 976 goto out_reserve;
b9aa55be 977 }
e6dcd2dc 978 em->start = start;
445a6944 979 em->orig_start = em->start;
771ed689
CM		 980 ram_size = ins.offset;
981 em->len = ins.offset;
2ab28f32
JB		 982 em->mod_start = em->start;
983 em->mod_len = em->len;
c8b97818 984
e6dcd2dc 985 em->block_start = ins.objectid;
c8b97818 986 em->block_len = ins.offset;
b4939680 987 em->orig_block_len = ins.offset;
cc95bef6 988 em->ram_bytes = ram_size;
e6dcd2dc 989 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 990 set_bit(EXTENT_FLAG_PINNED, &em->flags);
70c8a91c 991 em->generation = -1;
c8b97818 992
d397712b 993 while (1) {
890871be 994 write_lock(&em_tree->lock);
09a2a8f9 995 ret = add_extent_mapping(em_tree, em, 1);
890871be 996 write_unlock(&em_tree->lock);
e6dcd2dc
CM		 997 if (ret != -EEXIST) {
998 free_extent_map(em);
999 break;
1000 }
1001 btrfs_drop_extent_cache(inode, start,
c8b97818 1002 start + ram_size - 1, 0);
e6dcd2dc 1003 }
ace68bac
LB		 1004 if (ret)
1005 goto out_reserve;
e6dcd2dc 1006
98d20f67 1007 cur_alloc_size = ins.offset;
e6dcd2dc 1008 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771ed689 1009 ram_size, cur_alloc_size, 0);
ace68bac 1010 if (ret)
d9f85963 1011 goto out_drop_extent_cache;
c8b97818 1012
17d217fe
YZ		 1013 if (root->root_key.objectid ==
1014 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1015 ret = btrfs_reloc_clone_csums(inode, start,
1016 cur_alloc_size);
00361589 1017 if (ret)
d9f85963 1018 goto out_drop_extent_cache;
17d217fe
YZ		 1019 }
1020
d397712b 1021 if (disk_num_bytes < cur_alloc_size)
3b951516 1022 break;
d397712b 1023
c8b97818
CM		 1024 /* we're not doing compressed IO, don't unlock the first
1025 * page (which the caller expects to stay locked), don't
1026 * clear any dirty bits and don't set any writeback bits
8b62b72b
CM		 1027 *
1028 * Do set the Private2 bit so we know this page was properly
1029 * setup for writepage
c8b97818 1030 */
c2790a2e
JB		 1031 op = unlock ? PAGE_UNLOCK : 0;
1032 op |= PAGE_SET_PRIVATE2;
a791e35e 1033
c2790a2e
JB		 1034 extent_clear_unlock_delalloc(inode, start,
1035 start + ram_size - 1, locked_page,
1036 EXTENT_LOCKED | EXTENT_DELALLOC,
1037 op);
c8b97818 1038 disk_num_bytes -= cur_alloc_size;
c59f8951
CM		 1039 num_bytes -= cur_alloc_size;
1040 alloc_hint = ins.objectid + ins.offset;
1041 start += cur_alloc_size;
b888db2b 1042 }
79787eaa 1043out:
be20aa9d 1044 return ret;
b7d5b0a8 1045
d9f85963
FM		 1046out_drop_extent_cache:
1047 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
ace68bac 1048out_reserve:
e570fd27 1049 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
79787eaa 1050out_unlock:
c2790a2e 1051 extent_clear_unlock_delalloc(inode, start, end, locked_page,
151a41bc
JB		 1052 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1053 EXTENT_DELALLOC | EXTENT_DEFRAG,
1054 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1055 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
79787eaa 1056 goto out;
771ed689 1057}
c8b97818 1058
771ed689
CM		 1059/*
1060 * work queue call back to started compression on a file and pages
1061 */
1062static noinline void async_cow_start(struct btrfs_work *work)
1063{
1064 struct async_cow *async_cow;
1065 int num_added = 0;
1066 async_cow = container_of(work, struct async_cow, work);
1067
1068 compress_file_range(async_cow->inode, async_cow->locked_page,
1069 async_cow->start, async_cow->end, async_cow,
1070 &num_added);
8180ef88 1071 if (num_added == 0) {
cb77fcd8 1072 btrfs_add_delayed_iput(async_cow->inode);
771ed689 1073 async_cow->inode = NULL;
8180ef88 1074 }
771ed689
CM		 1075}
1076
1077/*
1078 * work queue call back to submit previously compressed pages
1079 */
1080static noinline void async_cow_submit(struct btrfs_work *work)
1081{
1082 struct async_cow *async_cow;
1083 struct btrfs_root *root;
1084 unsigned long nr_pages;
1085
1086 async_cow = container_of(work, struct async_cow, work);
1087
1088 root = async_cow->root;
1089 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1090 PAGE_CACHE_SHIFT;
1091
66657b31 1092 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
287082b0 1093 5 * 1024 * 1024 &&
771ed689
CM		 1094 waitqueue_active(&root->fs_info->async_submit_wait))
1095 wake_up(&root->fs_info->async_submit_wait);
1096
d397712b 1097 if (async_cow->inode)
771ed689 1098 submit_compressed_extents(async_cow->inode, async_cow);
771ed689 1099}
c8b97818 1100
771ed689
CM		 1101static noinline void async_cow_free(struct btrfs_work *work)
1102{
1103 struct async_cow *async_cow;
1104 async_cow = container_of(work, struct async_cow, work);
8180ef88 1105 if (async_cow->inode)
cb77fcd8 1106 btrfs_add_delayed_iput(async_cow->inode);
771ed689
CM		 1107 kfree(async_cow);
1108}
1109
1110static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1111 u64 start, u64 end, int *page_started,
1112 unsigned long *nr_written)
1113{
1114 struct async_cow *async_cow;
1115 struct btrfs_root *root = BTRFS_I(inode)->root;
1116 unsigned long nr_pages;
1117 u64 cur_end;
287082b0 1118 int limit = 10 * 1024 * 1024;
771ed689 1119
a3429ab7
CM		 1120 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1121 1, 0, NULL, GFP_NOFS);
d397712b 1122 while (start < end) {
771ed689 1123 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
79787eaa 1124 BUG_ON(!async_cow); /* -ENOMEM */
8180ef88 1125 async_cow->inode = igrab(inode);
771ed689
CM		 1126 async_cow->root = root;
1127 async_cow->locked_page = locked_page;
1128 async_cow->start = start;
1129
f79707b0
WS		 1130 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1131 !btrfs_test_opt(root, FORCE_COMPRESS))
771ed689
CM		 1132 cur_end = end;
1133 else
1134 cur_end = min(end, start + 512 * 1024 - 1);
1135
1136 async_cow->end = cur_end;
1137 INIT_LIST_HEAD(&async_cow->extents);
1138
9e0af237
LB		 1139 btrfs_init_work(&async_cow->work,
1140 btrfs_delalloc_helper,
1141 async_cow_start, async_cow_submit,
1142 async_cow_free);
771ed689 1143
771ed689
CM		 1144 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1145 PAGE_CACHE_SHIFT;
1146 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1147
afe3d242
QW		 1148 btrfs_queue_work(root->fs_info->delalloc_workers,
1149 &async_cow->work);
771ed689
CM		 1150
1151 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1152 wait_event(root->fs_info->async_submit_wait,
1153 (atomic_read(&root->fs_info->async_delalloc_pages) <
1154 limit));
1155 }
1156
d397712b 1157 while (atomic_read(&root->fs_info->async_submit_draining) &&
771ed689
CM		 1158 atomic_read(&root->fs_info->async_delalloc_pages)) {
1159 wait_event(root->fs_info->async_submit_wait,
1160 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1161 0));
1162 }
1163
1164 *nr_written += nr_pages;
1165 start = cur_end + 1;
1166 }
1167 *page_started = 1;
1168 return 0;
be20aa9d
CM		 1169}
1170
d397712b 1171static noinline int csum_exist_in_range(struct btrfs_root *root,
17d217fe
YZ		 1172 u64 bytenr, u64 num_bytes)
1173{
1174 int ret;
1175 struct btrfs_ordered_sum *sums;
1176 LIST_HEAD(list);
1177
07d400a6 1178 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
a2de733c 1179 bytenr + num_bytes - 1, &list, 0);
17d217fe
YZ		 1180 if (ret == 0 && list_empty(&list))
1181 return 0;
1182
1183 while (!list_empty(&list)) {
1184 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1185 list_del(&sums->list);
1186 kfree(sums);
1187 }
1188 return 1;
1189}
1190
d352ac68
CM		 1191/*
1192 * when nowcow writeback call back. This checks for snapshots or COW copies
1193 * of the extents that exist in the file, and COWs the file as required.
1194 *
1195 * If no cow copies or snapshots exist, we write directly to the existing
1196 * blocks on disk
1197 */
7f366cfe
CM		 1198static noinline int run_delalloc_nocow(struct inode *inode,
1199 struct page *locked_page,
771ed689
CM		 1200 u64 start, u64 end, int *page_started, int force,
1201 unsigned long *nr_written)
be20aa9d 1202{
be20aa9d 1203 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 1204 struct btrfs_trans_handle *trans;
be20aa9d 1205 struct extent_buffer *leaf;
be20aa9d 1206 struct btrfs_path *path;
80ff3856 1207 struct btrfs_file_extent_item *fi;
be20aa9d 1208 struct btrfs_key found_key;
80ff3856
YZ		 1209 u64 cow_start;
1210 u64 cur_offset;
1211 u64 extent_end;
5d4f98a2 1212 u64 extent_offset;
80ff3856
YZ		 1213 u64 disk_bytenr;
1214 u64 num_bytes;
b4939680 1215 u64 disk_num_bytes;
cc95bef6 1216 u64 ram_bytes;
80ff3856 1217 int extent_type;
79787eaa 1218 int ret, err;
d899e052 1219 int type;
80ff3856
YZ		 1220 int nocow;
1221 int check_prev = 1;
82d5902d 1222 bool nolock;
33345d01 1223 u64 ino = btrfs_ino(inode);
be20aa9d
CM		 1224
1225 path = btrfs_alloc_path();
17ca04af 1226 if (!path) {
c2790a2e
JB		 1227 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1228 EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc
JB		 1229 EXTENT_DO_ACCOUNTING |
1230 EXTENT_DEFRAG, PAGE_UNLOCK |
c2790a2e
JB		 1231 PAGE_CLEAR_DIRTY |
1232 PAGE_SET_WRITEBACK |
1233 PAGE_END_WRITEBACK);
d8926bb3 1234 return -ENOMEM;
17ca04af 1235 }
82d5902d 1236
83eea1f1 1237 nolock = btrfs_is_free_space_inode(inode);
82d5902d
LZ		 1238
1239 if (nolock)
7a7eaa40 1240 trans = btrfs_join_transaction_nolock(root);
82d5902d 1241 else
7a7eaa40 1242 trans = btrfs_join_transaction(root);
ff5714cc 1243
79787eaa 1244 if (IS_ERR(trans)) {
c2790a2e
JB		 1245 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1246 EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc
JB		 1247 EXTENT_DO_ACCOUNTING |
1248 EXTENT_DEFRAG, PAGE_UNLOCK |
c2790a2e
JB		 1249 PAGE_CLEAR_DIRTY |
1250 PAGE_SET_WRITEBACK |
1251 PAGE_END_WRITEBACK);
79787eaa
JM		 1252 btrfs_free_path(path);
1253 return PTR_ERR(trans);
1254 }
1255
74b21075 1256 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
be20aa9d 1257
80ff3856
YZ		 1258 cow_start = (u64)-1;
1259 cur_offset = start;
1260 while (1) {
33345d01 1261 ret = btrfs_lookup_file_extent(trans, root, path, ino,
80ff3856 1262 cur_offset, 0);
d788a349 1263 if (ret < 0)
79787eaa 1264 goto error;
80ff3856
YZ		 1265 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1266 leaf = path->nodes[0];
1267 btrfs_item_key_to_cpu(leaf, &found_key,
1268 path->slots[0] - 1);
33345d01 1269 if (found_key.objectid == ino &&
80ff3856
YZ		 1270 found_key.type == BTRFS_EXTENT_DATA_KEY)
1271 path->slots[0]--;
1272 }
1273 check_prev = 0;
1274next_slot:
1275 leaf = path->nodes[0];
1276 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1277 ret = btrfs_next_leaf(root, path);
d788a349 1278 if (ret < 0)
79787eaa 1279 goto error;
80ff3856
YZ		 1280 if (ret > 0)
1281 break;
1282 leaf = path->nodes[0];
1283 }
be20aa9d 1284
80ff3856
YZ		 1285 nocow = 0;
1286 disk_bytenr = 0;
17d217fe 1287 num_bytes = 0;
80ff3856
YZ		 1288 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1289
33345d01 1290 if (found_key.objectid > ino ||
80ff3856
YZ		 1291 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1292 found_key.offset > end)
1293 break;
1294
1295 if (found_key.offset > cur_offset) {
1296 extent_end = found_key.offset;
e9061e21 1297 extent_type = 0;
80ff3856
YZ		 1298 goto out_check;
1299 }
1300
1301 fi = btrfs_item_ptr(leaf, path->slots[0],
1302 struct btrfs_file_extent_item);
1303 extent_type = btrfs_file_extent_type(leaf, fi);
1304
cc95bef6 1305 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
d899e052
YZ		 1306 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1307 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856 1308 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5d4f98a2 1309 extent_offset = btrfs_file_extent_offset(leaf, fi);
80ff3856
YZ		 1310 extent_end = found_key.offset +
1311 btrfs_file_extent_num_bytes(leaf, fi);
b4939680
JB		 1312 disk_num_bytes =
1313 btrfs_file_extent_disk_num_bytes(leaf, fi);
80ff3856
YZ		 1314 if (extent_end <= start) {
1315 path->slots[0]++;
1316 goto next_slot;
1317 }
17d217fe
YZ		 1318 if (disk_bytenr == 0)
1319 goto out_check;
80ff3856
YZ		 1320 if (btrfs_file_extent_compression(leaf, fi) ||
1321 btrfs_file_extent_encryption(leaf, fi) ||
1322 btrfs_file_extent_other_encoding(leaf, fi))
1323 goto out_check;
d899e052
YZ		 1324 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1325 goto out_check;
d2fb3437 1326 if (btrfs_extent_readonly(root, disk_bytenr))
80ff3856 1327 goto out_check;
33345d01 1328 if (btrfs_cross_ref_exist(trans, root, ino,
5d4f98a2
YZ		 1329 found_key.offset -
1330 extent_offset, disk_bytenr))
17d217fe 1331 goto out_check;
5d4f98a2 1332 disk_bytenr += extent_offset;
17d217fe
YZ		 1333 disk_bytenr += cur_offset - found_key.offset;
1334 num_bytes = min(end + 1, extent_end) - cur_offset;
e9894fd3
WS		 1335 /*
1336 * if there are pending snapshots for this root,
1337 * we fall into common COW way.
1338 */
1339 if (!nolock) {
9ea24bbe 1340 err = btrfs_start_write_no_snapshoting(root);
e9894fd3
WS		 1341 if (!err)
1342 goto out_check;
1343 }
17d217fe
YZ		 1344 /*
1345 * force cow if csum exists in the range.
1346 * this ensure that csum for a given extent are
1347 * either valid or do not exist.
1348 */
1349 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1350 goto out_check;
80ff3856
YZ		 1351 nocow = 1;
1352 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1353 extent_end = found_key.offset +
514ac8ad
CM		 1354 btrfs_file_extent_inline_len(leaf,
1355 path->slots[0], fi);
80ff3856
YZ		 1356 extent_end = ALIGN(extent_end, root->sectorsize);
1357 } else {
1358 BUG_ON(1);
1359 }
1360out_check:
1361 if (extent_end <= start) {
1362 path->slots[0]++;
e9894fd3 1363 if (!nolock && nocow)
9ea24bbe 1364 btrfs_end_write_no_snapshoting(root);
80ff3856
YZ		 1365 goto next_slot;
1366 }
1367 if (!nocow) {
1368 if (cow_start == (u64)-1)
1369 cow_start = cur_offset;
1370 cur_offset = extent_end;
1371 if (cur_offset > end)
1372 break;
1373 path->slots[0]++;
1374 goto next_slot;
7ea394f1
YZ		 1375 }
1376
b3b4aa74 1377 btrfs_release_path(path);
80ff3856 1378 if (cow_start != (u64)-1) {
00361589
JB		 1379 ret = cow_file_range(inode, locked_page,
1380 cow_start, found_key.offset - 1,
1381 page_started, nr_written, 1);
e9894fd3
WS		 1382 if (ret) {
1383 if (!nolock && nocow)
9ea24bbe 1384 btrfs_end_write_no_snapshoting(root);
79787eaa 1385 goto error;
e9894fd3 1386 }
80ff3856 1387 cow_start = (u64)-1;
7ea394f1 1388 }
80ff3856 1389
d899e052
YZ		 1390 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1391 struct extent_map *em;
1392 struct extent_map_tree *em_tree;
1393 em_tree = &BTRFS_I(inode)->extent_tree;
172ddd60 1394 em = alloc_extent_map();
79787eaa 1395 BUG_ON(!em); /* -ENOMEM */
d899e052 1396 em->start = cur_offset;
70c8a91c 1397 em->orig_start = found_key.offset - extent_offset;
d899e052
YZ		 1398 em->len = num_bytes;
1399 em->block_len = num_bytes;
1400 em->block_start = disk_bytenr;
b4939680 1401 em->orig_block_len = disk_num_bytes;
cc95bef6 1402 em->ram_bytes = ram_bytes;
d899e052 1403 em->bdev = root->fs_info->fs_devices->latest_bdev;
2ab28f32
JB		 1404 em->mod_start = em->start;
1405 em->mod_len = em->len;
d899e052 1406 set_bit(EXTENT_FLAG_PINNED, &em->flags);
b11e234d 1407 set_bit(EXTENT_FLAG_FILLING, &em->flags);
70c8a91c 1408 em->generation = -1;
d899e052 1409 while (1) {
890871be 1410 write_lock(&em_tree->lock);
09a2a8f9 1411 ret = add_extent_mapping(em_tree, em, 1);
890871be 1412 write_unlock(&em_tree->lock);
d899e052
YZ		 1413 if (ret != -EEXIST) {
1414 free_extent_map(em);
1415 break;
1416 }
1417 btrfs_drop_extent_cache(inode, em->start,
1418 em->start + em->len - 1, 0);
1419 }
1420 type = BTRFS_ORDERED_PREALLOC;
1421 } else {
1422 type = BTRFS_ORDERED_NOCOW;
1423 }
80ff3856
YZ		 1424
1425 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052 1426 num_bytes, num_bytes, type);
79787eaa 1427 BUG_ON(ret); /* -ENOMEM */
771ed689 1428
efa56464
YZ		 1429 if (root->root_key.objectid ==
1430 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1431 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1432 num_bytes);
e9894fd3
WS		 1433 if (ret) {
1434 if (!nolock && nocow)
9ea24bbe 1435 btrfs_end_write_no_snapshoting(root);
79787eaa 1436 goto error;
e9894fd3 1437 }
efa56464
YZ		 1438 }
1439
c2790a2e
JB		 1440 extent_clear_unlock_delalloc(inode, cur_offset,
1441 cur_offset + num_bytes - 1,
1442 locked_page, EXTENT_LOCKED |
1443 EXTENT_DELALLOC, PAGE_UNLOCK |
1444 PAGE_SET_PRIVATE2);
e9894fd3 1445 if (!nolock && nocow)
9ea24bbe 1446 btrfs_end_write_no_snapshoting(root);
80ff3856
YZ		 1447 cur_offset = extent_end;
1448 if (cur_offset > end)
1449 break;
be20aa9d 1450 }
b3b4aa74 1451 btrfs_release_path(path);
80ff3856 1452
17ca04af 1453 if (cur_offset <= end && cow_start == (u64)-1) {
80ff3856 1454 cow_start = cur_offset;
17ca04af
JB		 1455 cur_offset = end;
1456 }
1457
80ff3856 1458 if (cow_start != (u64)-1) {
00361589
JB		 1459 ret = cow_file_range(inode, locked_page, cow_start, end,
1460 page_started, nr_written, 1);
d788a349 1461 if (ret)
79787eaa 1462 goto error;
80ff3856
YZ		 1463 }
1464
79787eaa 1465error:
a698d075 1466 err = btrfs_end_transaction(trans, root);
79787eaa
JM		 1467 if (!ret)
1468 ret = err;
1469
17ca04af 1470 if (ret && cur_offset < end)
c2790a2e
JB		 1471 extent_clear_unlock_delalloc(inode, cur_offset, end,
1472 locked_page, EXTENT_LOCKED |
151a41bc
JB		 1473 EXTENT_DELALLOC | EXTENT_DEFRAG |
1474 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1475 PAGE_CLEAR_DIRTY |
c2790a2e
JB		 1476 PAGE_SET_WRITEBACK |
1477 PAGE_END_WRITEBACK);
7ea394f1 1478 btrfs_free_path(path);
79787eaa 1479 return ret;
be20aa9d
CM		 1480}
1481
47059d93
WS		 1482static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1483{
1484
1485 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1486 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1487 return 0;
1488
1489 /*
1490 * @defrag_bytes is a hint value, no spinlock held here,
1491 * if is not zero, it means the file is defragging.
1492 * Force cow if given extent needs to be defragged.
1493 */
1494 if (BTRFS_I(inode)->defrag_bytes &&
1495 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1496 EXTENT_DEFRAG, 0, NULL))
1497 return 1;
1498
1499 return 0;
1500}
1501
d352ac68
CM		 1502/*
1503 * extent_io.c call back to do delayed allocation processing
1504 */
c8b97818 1505static int run_delalloc_range(struct inode *inode, struct page *locked_page,
771ed689
CM		 1506 u64 start, u64 end, int *page_started,
1507 unsigned long *nr_written)
be20aa9d 1508{
be20aa9d 1509 int ret;
47059d93 1510 int force_cow = need_force_cow(inode, start, end);
a2135011 1511
47059d93 1512 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
c8b97818 1513 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1514 page_started, 1, nr_written);
47059d93 1515 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
d899e052 1516 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1517 page_started, 0, nr_written);
7816030e 1518 } else if (!inode_need_compress(inode)) {
7f366cfe
CM		 1519 ret = cow_file_range(inode, locked_page, start, end,
1520 page_started, nr_written, 1);
7ddf5a42
JB		 1521 } else {
1522 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1523 &BTRFS_I(inode)->runtime_flags);
771ed689 1524 ret = cow_file_range_async(inode, locked_page, start, end,
d397712b 1525 page_started, nr_written);
7ddf5a42 1526 }
b888db2b
CM		 1527 return ret;
1528}
1529
1bf85046
JM		 1530static void btrfs_split_extent_hook(struct inode *inode,
1531 struct extent_state *orig, u64 split)
9ed74f2d 1532{
dcab6a3b
JB		 1533 u64 size;
1534
0ca1f7ce 1535 /* not delalloc, ignore it */
9ed74f2d 1536 if (!(orig->state & EXTENT_DELALLOC))
1bf85046 1537 return;
9ed74f2d 1538
dcab6a3b
JB		 1539 size = orig->end - orig->start + 1;
1540 if (size > BTRFS_MAX_EXTENT_SIZE) {
1541 u64 num_extents;
1542 u64 new_size;
1543
1544 /*
ba117213
JB		 1545 * See the explanation in btrfs_merge_extent_hook, the same
1546 * applies here, just in reverse.
dcab6a3b
JB		 1547 */
1548 new_size = orig->end - split + 1;
ba117213 1549 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
dcab6a3b 1550 BTRFS_MAX_EXTENT_SIZE);
ba117213
JB		 1551 new_size = split - orig->start;
1552 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1553 BTRFS_MAX_EXTENT_SIZE);
1554 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1555 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
dcab6a3b
JB		 1556 return;
1557 }
1558
9e0baf60
JB		 1559 spin_lock(&BTRFS_I(inode)->lock);
1560 BTRFS_I(inode)->outstanding_extents++;
1561 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1562}
1563
1564/*
1565 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1566 * extents so we can keep track of new extents that are just merged onto old
1567 * extents, such as when we are doing sequential writes, so we can properly
1568 * account for the metadata space we'll need.
1569 */
1bf85046
JM		 1570static void btrfs_merge_extent_hook(struct inode *inode,
1571 struct extent_state *new,
1572 struct extent_state *other)
9ed74f2d 1573{
dcab6a3b
JB		 1574 u64 new_size, old_size;
1575 u64 num_extents;
1576
9ed74f2d
JB		 1577 /* not delalloc, ignore it */
1578 if (!(other->state & EXTENT_DELALLOC))
1bf85046 1579 return;
9ed74f2d 1580
8461a3de
JB		 1581 if (new->start > other->start)
1582 new_size = new->end - other->start + 1;
1583 else
1584 new_size = other->end - new->start + 1;
dcab6a3b
JB		 1585
1586 /* we're not bigger than the max, unreserve the space and go */
1587 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1588 spin_lock(&BTRFS_I(inode)->lock);
1589 BTRFS_I(inode)->outstanding_extents--;
1590 spin_unlock(&BTRFS_I(inode)->lock);
1591 return;
1592 }
1593
1594 /*
ba117213
JB		 1595 * We have to add up either side to figure out how many extents were
1596 * accounted for before we merged into one big extent. If the number of
1597 * extents we accounted for is <= the amount we need for the new range
1598 * then we can return, otherwise drop. Think of it like this
1599 *
1600 * [ 4k][MAX_SIZE]
1601 *
1602 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1603 * need 2 outstanding extents, on one side we have 1 and the other side
1604 * we have 1 so they are == and we can return. But in this case
1605 *
1606 * [MAX_SIZE+4k][MAX_SIZE+4k]
1607 *
1608 * Each range on their own accounts for 2 extents, but merged together
1609 * they are only 3 extents worth of accounting, so we need to drop in
1610 * this case.
dcab6a3b 1611 */
ba117213 1612 old_size = other->end - other->start + 1;
dcab6a3b
JB		 1613 num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1614 BTRFS_MAX_EXTENT_SIZE);
ba117213
JB		 1615 old_size = new->end - new->start + 1;
1616 num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1617 BTRFS_MAX_EXTENT_SIZE);
1618
dcab6a3b 1619 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
ba117213 1620 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
dcab6a3b
JB		 1621 return;
1622
9e0baf60
JB		 1623 spin_lock(&BTRFS_I(inode)->lock);
1624 BTRFS_I(inode)->outstanding_extents--;
1625 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1626}
1627
eb73c1b7
MX		 1628static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1629 struct inode *inode)
1630{
1631 spin_lock(&root->delalloc_lock);
1632 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1633 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1634 &root->delalloc_inodes);
1635 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1636 &BTRFS_I(inode)->runtime_flags);
1637 root->nr_delalloc_inodes++;
1638 if (root->nr_delalloc_inodes == 1) {
1639 spin_lock(&root->fs_info->delalloc_root_lock);
1640 BUG_ON(!list_empty(&root->delalloc_root));
1641 list_add_tail(&root->delalloc_root,
1642 &root->fs_info->delalloc_roots);
1643 spin_unlock(&root->fs_info->delalloc_root_lock);
1644 }
1645 }
1646 spin_unlock(&root->delalloc_lock);
1647}
1648
1649static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1650 struct inode *inode)
1651{
1652 spin_lock(&root->delalloc_lock);
1653 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1654 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1655 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1656 &BTRFS_I(inode)->runtime_flags);
1657 root->nr_delalloc_inodes--;
1658 if (!root->nr_delalloc_inodes) {
1659 spin_lock(&root->fs_info->delalloc_root_lock);
1660 BUG_ON(list_empty(&root->delalloc_root));
1661 list_del_init(&root->delalloc_root);
1662 spin_unlock(&root->fs_info->delalloc_root_lock);
1663 }
1664 }
1665 spin_unlock(&root->delalloc_lock);
1666}
1667
d352ac68
CM		 1668/*
1669 * extent_io.c set_bit_hook, used to track delayed allocation
1670 * bytes in this file, and to maintain the list of inodes that
1671 * have pending delalloc work to be done.
1672 */
1bf85046 1673static void btrfs_set_bit_hook(struct inode *inode,
9ee49a04 1674 struct extent_state *state, unsigned *bits)
291d673e 1675{
9ed74f2d 1676
47059d93
WS		 1677 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1678 WARN_ON(1);
75eff68e
CM		 1679 /*
1680 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1681 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1682 * bit, which is only set or cleared with irqs on
1683 */
0ca1f7ce 1684 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1685 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1686 u64 len = state->end + 1 - state->start;
83eea1f1 1687 bool do_list = !btrfs_is_free_space_inode(inode);
9ed74f2d 1688
9e0baf60 1689 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1690 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1691 } else {
1692 spin_lock(&BTRFS_I(inode)->lock);
1693 BTRFS_I(inode)->outstanding_extents++;
1694 spin_unlock(&BTRFS_I(inode)->lock);
1695 }
287a0ab9 1696
963d678b
MX		 1697 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1698 root->fs_info->delalloc_batch);
df0af1a5 1699 spin_lock(&BTRFS_I(inode)->lock);
0ca1f7ce 1700 BTRFS_I(inode)->delalloc_bytes += len;
47059d93
WS		 1701 if (*bits & EXTENT_DEFRAG)
1702 BTRFS_I(inode)->defrag_bytes += len;
df0af1a5 1703 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
eb73c1b7
MX		 1704 &BTRFS_I(inode)->runtime_flags))
1705 btrfs_add_delalloc_inodes(root, inode);
df0af1a5 1706 spin_unlock(&BTRFS_I(inode)->lock);
291d673e 1707 }
291d673e
CM		 1708}
1709
d352ac68
CM		 1710/*
1711 * extent_io.c clear_bit_hook, see set_bit_hook for why
1712 */
1bf85046 1713static void btrfs_clear_bit_hook(struct inode *inode,
41074888 1714 struct extent_state *state,
9ee49a04 1715 unsigned *bits)
291d673e 1716{
47059d93 1717 u64 len = state->end + 1 - state->start;
dcab6a3b
JB		 1718 u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1719 BTRFS_MAX_EXTENT_SIZE);
47059d93
WS		 1720
1721 spin_lock(&BTRFS_I(inode)->lock);
1722 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1723 BTRFS_I(inode)->defrag_bytes -= len;
1724 spin_unlock(&BTRFS_I(inode)->lock);
1725
75eff68e
CM		 1726 /*
1727 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1728 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1729 * bit, which is only set or cleared with irqs on
1730 */
0ca1f7ce 1731 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1732 struct btrfs_root *root = BTRFS_I(inode)->root;
83eea1f1 1733 bool do_list = !btrfs_is_free_space_inode(inode);
bcbfce8a 1734
9e0baf60 1735 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1736 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1737 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1738 spin_lock(&BTRFS_I(inode)->lock);
dcab6a3b 1739 BTRFS_I(inode)->outstanding_extents -= num_extents;
9e0baf60
JB		 1740 spin_unlock(&BTRFS_I(inode)->lock);
1741 }
0ca1f7ce 1742
b6d08f06
JB		 1743 /*
1744 * We don't reserve metadata space for space cache inodes so we
1745 * don't need to call dellalloc_release_metadata if there is an
1746 * error.
1747 */
1748 if (*bits & EXTENT_DO_ACCOUNTING &&
1749 root != root->fs_info->tree_root)
0ca1f7ce
YZ		 1750 btrfs_delalloc_release_metadata(inode, len);
1751
0cb59c99 1752 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
7ee9e440 1753 && do_list && !(state->state & EXTENT_NORESERVE))
0ca1f7ce 1754 btrfs_free_reserved_data_space(inode, len);
9ed74f2d 1755
963d678b
MX		 1756 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1757 root->fs_info->delalloc_batch);
df0af1a5 1758 spin_lock(&BTRFS_I(inode)->lock);
0ca1f7ce 1759 BTRFS_I(inode)->delalloc_bytes -= len;
0cb59c99 1760 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
df0af1a5 1761 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
eb73c1b7
MX		 1762 &BTRFS_I(inode)->runtime_flags))
1763 btrfs_del_delalloc_inode(root, inode);
df0af1a5 1764 spin_unlock(&BTRFS_I(inode)->lock);
291d673e 1765 }
291d673e
CM		 1766}
1767
d352ac68
CM		 1768/*
1769 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1770 * we don't create bios that span stripes or chunks
1771 */
64a16701 1772int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
c8b97818
CM		 1773 size_t size, struct bio *bio,
1774 unsigned long bio_flags)
239b14b3
CM		 1775{
1776 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
4f024f37 1777 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
239b14b3
CM		 1778 u64 length = 0;
1779 u64 map_length;
239b14b3
CM		 1780 int ret;
1781
771ed689
CM		 1782 if (bio_flags & EXTENT_BIO_COMPRESSED)
1783 return 0;
1784
4f024f37 1785 length = bio->bi_iter.bi_size;
239b14b3 1786 map_length = length;
64a16701 1787 ret = btrfs_map_block(root->fs_info, rw, logical,
f188591e 1788 &map_length, NULL, 0);
3ec706c8 1789 /* Will always return 0 with map_multi == NULL */
3444a972 1790 BUG_ON(ret < 0);
d397712b 1791 if (map_length < length + size)
239b14b3 1792 return 1;
3444a972 1793 return 0;
239b14b3
CM		 1794}
1795
d352ac68
CM		 1796/*
1797 * in order to insert checksums into the metadata in large chunks,
1798 * we wait until bio submission time. All the pages in the bio are
1799 * checksummed and sums are attached onto the ordered extent record.
1800 *
1801 * At IO completion time the cums attached on the ordered extent record
1802 * are inserted into the btree
1803 */
d397712b
CM		 1804static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1805 struct bio *bio, int mirror_num,
eaf25d93
CM		 1806 unsigned long bio_flags,
1807 u64 bio_offset)
065631f6 1808{
065631f6 1809 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 1810 int ret = 0;
e015640f 1811
d20f7043 1812 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
79787eaa 1813 BUG_ON(ret); /* -ENOMEM */
4a69a410
CM		 1814 return 0;
1815}
e015640f 1816
4a69a410
CM		 1817/*
1818 * in order to insert checksums into the metadata in large chunks,
1819 * we wait until bio submission time. All the pages in the bio are
1820 * checksummed and sums are attached onto the ordered extent record.
1821 *
1822 * At IO completion time the cums attached on the ordered extent record
1823 * are inserted into the btree
1824 */
b2950863 1825static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1826 int mirror_num, unsigned long bio_flags,
1827 u64 bio_offset)
4a69a410
CM		 1828{
1829 struct btrfs_root *root = BTRFS_I(inode)->root;
61891923
SB		 1830 int ret;
1831
1832 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1833 if (ret)
1834 bio_endio(bio, ret);
1835 return ret;
44b8bd7e
CM		 1836}
1837
d352ac68 1838/*
cad321ad
CM		 1839 * extent_io.c submission hook. This does the right thing for csum calculation
1840 * on write, or reading the csums from the tree before a read
d352ac68 1841 */
b2950863 1842static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1843 int mirror_num, unsigned long bio_flags,
1844 u64 bio_offset)
44b8bd7e
CM		 1845{
1846 struct btrfs_root *root = BTRFS_I(inode)->root;
1847 int ret = 0;
19b9bdb0 1848 int skip_sum;
0417341e 1849 int metadata = 0;
b812ce28 1850 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
44b8bd7e 1851
6cbff00f 1852 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
cad321ad 1853
83eea1f1 1854 if (btrfs_is_free_space_inode(inode))
0417341e
JM		 1855 metadata = 2;
1856
7b6d91da 1857 if (!(rw & REQ_WRITE)) {
5fd02043
JB		 1858 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1859 if (ret)
61891923 1860 goto out;
5fd02043 1861
d20f7043 1862 if (bio_flags & EXTENT_BIO_COMPRESSED) {
61891923
SB		 1863 ret = btrfs_submit_compressed_read(inode, bio,
1864 mirror_num,
1865 bio_flags);
1866 goto out;
c2db1073
TI		 1867 } else if (!skip_sum) {
1868 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1869 if (ret)
61891923 1870 goto out;
c2db1073 1871 }
4d1b5fb4 1872 goto mapit;
b812ce28 1873 } else if (async && !skip_sum) {
17d217fe
YZ		 1874 /* csum items have already been cloned */
1875 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1876 goto mapit;
19b9bdb0 1877 /* we're doing a write, do the async checksumming */
61891923 1878 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 1879 inode, rw, bio, mirror_num,
eaf25d93
CM		 1880 bio_flags, bio_offset,
1881 __btrfs_submit_bio_start,
4a69a410 1882 __btrfs_submit_bio_done);
61891923 1883 goto out;
b812ce28
JB		 1884 } else if (!skip_sum) {
1885 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1886 if (ret)
1887 goto out;
19b9bdb0
CM		 1888 }
1889
0b86a832 1890mapit:
61891923
SB		 1891 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1892
1893out:
1894 if (ret < 0)
1895 bio_endio(bio, ret);
1896 return ret;
065631f6 1897}
6885f308 1898
d352ac68
CM		 1899/*
1900 * given a list of ordered sums record them in the inode. This happens
1901 * at IO completion time based on sums calculated at bio submission time.
1902 */
ba1da2f4 1903static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM		 1904 struct inode *inode, u64 file_offset,
1905 struct list_head *list)
1906{
e6dcd2dc
CM		 1907 struct btrfs_ordered_sum *sum;
1908
c6e30871 1909 list_for_each_entry(sum, list, list) {
39847c4d 1910 trans->adding_csums = 1;
d20f7043
CM		 1911 btrfs_csum_file_blocks(trans,
1912 BTRFS_I(inode)->root->fs_info->csum_root, sum);
39847c4d 1913 trans->adding_csums = 0;
e6dcd2dc
CM		 1914 }
1915 return 0;
1916}
1917
2ac55d41
JB		 1918int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1919 struct extent_state **cached_state)
ea8c2819 1920{
6c1500f2 1921 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
ea8c2819 1922 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2ac55d41 1923 cached_state, GFP_NOFS);
ea8c2819
CM		 1924}
1925
d352ac68 1926/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM		 1927struct btrfs_writepage_fixup {
1928 struct page *page;
1929 struct btrfs_work work;
1930};
1931
b2950863 1932static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
247e743c
CM		 1933{
1934 struct btrfs_writepage_fixup *fixup;
1935 struct btrfs_ordered_extent *ordered;
2ac55d41 1936 struct extent_state *cached_state = NULL;
247e743c
CM		 1937 struct page *page;
1938 struct inode *inode;
1939 u64 page_start;
1940 u64 page_end;
87826df0 1941 int ret;
247e743c
CM		 1942
1943 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1944 page = fixup->page;
4a096752 1945again:
247e743c
CM		 1946 lock_page(page);
1947 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1948 ClearPageChecked(page);
1949 goto out_page;
1950 }
1951
1952 inode = page->mapping->host;
1953 page_start = page_offset(page);
1954 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1955
2ac55d41 1956 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
d0082371 1957 &cached_state);
4a096752
CM		 1958
1959 /* already ordered? We're done */
8b62b72b 1960 if (PagePrivate2(page))
247e743c 1961 goto out;
4a096752
CM		 1962
1963 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1964 if (ordered) {
2ac55d41
JB		 1965 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1966 page_end, &cached_state, GFP_NOFS);
4a096752
CM		 1967 unlock_page(page);
1968 btrfs_start_ordered_extent(inode, ordered, 1);
87826df0 1969 btrfs_put_ordered_extent(ordered);
4a096752
CM		 1970 goto again;
1971 }
247e743c 1972
87826df0
JM		 1973 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1974 if (ret) {
1975 mapping_set_error(page->mapping, ret);
1976 end_extent_writepage(page, ret, page_start, page_end);
1977 ClearPageChecked(page);
1978 goto out;
1979 }
1980
2ac55d41 1981 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
247e743c 1982 ClearPageChecked(page);
87826df0 1983 set_page_dirty(page);
247e743c 1984out:
2ac55d41
JB		 1985 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1986 &cached_state, GFP_NOFS);
247e743c
CM		 1987out_page:
1988 unlock_page(page);
1989 page_cache_release(page);
b897abec 1990 kfree(fixup);
247e743c
CM		 1991}
1992
1993/*
1994 * There are a few paths in the higher layers of the kernel that directly
1995 * set the page dirty bit without asking the filesystem if it is a
1996 * good idea. This causes problems because we want to make sure COW
1997 * properly happens and the data=ordered rules are followed.
1998 *
c8b97818 1999 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 2000 * hasn't been properly setup for IO. We kick off an async process
2001 * to fix it up. The async helper will wait for ordered extents, set
2002 * the delalloc bit and make it safe to write the page.
2003 */
b2950863 2004static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
247e743c
CM		 2005{
2006 struct inode *inode = page->mapping->host;
2007 struct btrfs_writepage_fixup *fixup;
2008 struct btrfs_root *root = BTRFS_I(inode)->root;
247e743c 2009
8b62b72b
CM		 2010 /* this page is properly in the ordered list */
2011 if (TestClearPagePrivate2(page))
247e743c
CM		 2012 return 0;
2013
2014 if (PageChecked(page))
2015 return -EAGAIN;
2016
2017 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2018 if (!fixup)
2019 return -EAGAIN;
f421950f 2020
247e743c
CM		 2021 SetPageChecked(page);
2022 page_cache_get(page);
9e0af237
LB		 2023 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2024 btrfs_writepage_fixup_worker, NULL, NULL);
247e743c 2025 fixup->page = page;
dc6e3209 2026 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
87826df0 2027 return -EBUSY;
247e743c
CM		 2028}
2029
d899e052
YZ		 2030static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2031 struct inode *inode, u64 file_pos,
2032 u64 disk_bytenr, u64 disk_num_bytes,
2033 u64 num_bytes, u64 ram_bytes,
2034 u8 compression, u8 encryption,
2035 u16 other_encoding, int extent_type)
2036{
2037 struct btrfs_root *root = BTRFS_I(inode)->root;
2038 struct btrfs_file_extent_item *fi;
2039 struct btrfs_path *path;
2040 struct extent_buffer *leaf;
2041 struct btrfs_key ins;
1acae57b 2042 int extent_inserted = 0;
d899e052
YZ		 2043 int ret;
2044
2045 path = btrfs_alloc_path();
d8926bb3
MF		 2046 if (!path)
2047 return -ENOMEM;
d899e052 2048
a1ed835e
CM		 2049 /*
2050 * we may be replacing one extent in the tree with another.
2051 * The new extent is pinned in the extent map, and we don't want
2052 * to drop it from the cache until it is completely in the btree.
2053 *
2054 * So, tell btrfs_drop_extents to leave this extent in the cache.
2055 * the caller is expected to unpin it and allow it to be merged
2056 * with the others.
2057 */
1acae57b
FDBM		 2058 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2059 file_pos + num_bytes, NULL, 0,
2060 1, sizeof(*fi), &extent_inserted);
79787eaa
JM		 2061 if (ret)
2062 goto out;
d899e052 2063
1acae57b
FDBM		 2064 if (!extent_inserted) {
2065 ins.objectid = btrfs_ino(inode);
2066 ins.offset = file_pos;
2067 ins.type = BTRFS_EXTENT_DATA_KEY;
2068
2069 path->leave_spinning = 1;
2070 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2071 sizeof(*fi));
2072 if (ret)
2073 goto out;
2074 }
d899e052
YZ		 2075 leaf = path->nodes[0];
2076 fi = btrfs_item_ptr(leaf, path->slots[0],
2077 struct btrfs_file_extent_item);
2078 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2079 btrfs_set_file_extent_type(leaf, fi, extent_type);
2080 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2081 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2082 btrfs_set_file_extent_offset(leaf, fi, 0);
2083 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2084 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2085 btrfs_set_file_extent_compression(leaf, fi, compression);
2086 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2087 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
b9473439 2088
d899e052 2089 btrfs_mark_buffer_dirty(leaf);
ce195332 2090 btrfs_release_path(path);
d899e052
YZ		 2091
2092 inode_add_bytes(inode, num_bytes);
d899e052
YZ		 2093
2094 ins.objectid = disk_bytenr;
2095 ins.offset = disk_num_bytes;
2096 ins.type = BTRFS_EXTENT_ITEM_KEY;
5d4f98a2
YZ		 2097 ret = btrfs_alloc_reserved_file_extent(trans, root,
2098 root->root_key.objectid,
33345d01 2099 btrfs_ino(inode), file_pos, &ins);
79787eaa 2100out:
d899e052 2101 btrfs_free_path(path);
b9473439 2102
79787eaa 2103 return ret;
d899e052
YZ		 2104}
2105
38c227d8
LB		 2106/* snapshot-aware defrag */
2107struct sa_defrag_extent_backref {
2108 struct rb_node node;
2109 struct old_sa_defrag_extent *old;
2110 u64 root_id;
2111 u64 inum;
2112 u64 file_pos;
2113 u64 extent_offset;
2114 u64 num_bytes;
2115 u64 generation;
2116};
2117
2118struct old_sa_defrag_extent {
2119 struct list_head list;
2120 struct new_sa_defrag_extent *new;
2121
2122 u64 extent_offset;
2123 u64 bytenr;
2124 u64 offset;
2125 u64 len;
2126 int count;
2127};
2128
2129struct new_sa_defrag_extent {
2130 struct rb_root root;
2131 struct list_head head;
2132 struct btrfs_path *path;
2133 struct inode *inode;
2134 u64 file_pos;
2135 u64 len;
2136 u64 bytenr;
2137 u64 disk_len;
2138 u8 compress_type;
2139};
2140
2141static int backref_comp(struct sa_defrag_extent_backref *b1,
2142 struct sa_defrag_extent_backref *b2)
2143{
2144 if (b1->root_id < b2->root_id)
2145 return -1;
2146 else if (b1->root_id > b2->root_id)
2147 return 1;
2148
2149 if (b1->inum < b2->inum)
2150 return -1;
2151 else if (b1->inum > b2->inum)
2152 return 1;
2153
2154 if (b1->file_pos < b2->file_pos)
2155 return -1;
2156 else if (b1->file_pos > b2->file_pos)
2157 return 1;
2158
2159 /*
2160 * [------------------------------] ===> (a range of space)
2161 * |<--->| |<---->| =============> (fs/file tree A)
2162 * |<---------------------------->| ===> (fs/file tree B)
2163 *
2164 * A range of space can refer to two file extents in one tree while
2165 * refer to only one file extent in another tree.
2166 *
2167 * So we may process a disk offset more than one time(two extents in A)
2168 * and locate at the same extent(one extent in B), then insert two same
2169 * backrefs(both refer to the extent in B).
2170 */
2171 return 0;
2172}
2173
2174static void backref_insert(struct rb_root *root,
2175 struct sa_defrag_extent_backref *backref)
2176{
2177 struct rb_node **p = &root->rb_node;
2178 struct rb_node *parent = NULL;
2179 struct sa_defrag_extent_backref *entry;
2180 int ret;
2181
2182 while (*p) {
2183 parent = *p;
2184 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2185
2186 ret = backref_comp(backref, entry);
2187 if (ret < 0)
2188 p = &(*p)->rb_left;
2189 else
2190 p = &(*p)->rb_right;
2191 }
2192
2193 rb_link_node(&backref->node, parent, p);
2194 rb_insert_color(&backref->node, root);
2195}
2196
2197/*
2198 * Note the backref might has changed, and in this case we just return 0.
2199 */
2200static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2201 void *ctx)
2202{
2203 struct btrfs_file_extent_item *extent;
2204 struct btrfs_fs_info *fs_info;
2205 struct old_sa_defrag_extent *old = ctx;
2206 struct new_sa_defrag_extent *new = old->new;
2207 struct btrfs_path *path = new->path;
2208 struct btrfs_key key;
2209 struct btrfs_root *root;
2210 struct sa_defrag_extent_backref *backref;
2211 struct extent_buffer *leaf;
2212 struct inode *inode = new->inode;
2213 int slot;
2214 int ret;
2215 u64 extent_offset;
2216 u64 num_bytes;
2217
2218 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2219 inum == btrfs_ino(inode))
2220 return 0;
2221
2222 key.objectid = root_id;
2223 key.type = BTRFS_ROOT_ITEM_KEY;
2224 key.offset = (u64)-1;
2225
2226 fs_info = BTRFS_I(inode)->root->fs_info;
2227 root = btrfs_read_fs_root_no_name(fs_info, &key);
2228 if (IS_ERR(root)) {
2229 if (PTR_ERR(root) == -ENOENT)
2230 return 0;
2231 WARN_ON(1);
2232 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2233 inum, offset, root_id);
2234 return PTR_ERR(root);
2235 }
2236
2237 key.objectid = inum;
2238 key.type = BTRFS_EXTENT_DATA_KEY;
2239 if (offset > (u64)-1 << 32)
2240 key.offset = 0;
2241 else
2242 key.offset = offset;
2243
2244 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
fae7f21c 2245 if (WARN_ON(ret < 0))
38c227d8 2246 return ret;
50f1319c 2247 ret = 0;
38c227d8
LB		 2248
2249 while (1) {
2250 cond_resched();
2251
2252 leaf = path->nodes[0];
2253 slot = path->slots[0];
2254
2255 if (slot >= btrfs_header_nritems(leaf)) {
2256 ret = btrfs_next_leaf(root, path);
2257 if (ret < 0) {
2258 goto out;
2259 } else if (ret > 0) {
2260 ret = 0;
2261 goto out;
2262 }
2263 continue;
2264 }
2265
2266 path->slots[0]++;
2267
2268 btrfs_item_key_to_cpu(leaf, &key, slot);
2269
2270 if (key.objectid > inum)
2271 goto out;
2272
2273 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2274 continue;
2275
2276 extent = btrfs_item_ptr(leaf, slot,
2277 struct btrfs_file_extent_item);
2278
2279 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2280 continue;
2281
e68afa49
LB		 2282 /*
2283 * 'offset' refers to the exact key.offset,
2284 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2285 * (key.offset - extent_offset).
2286 */
2287 if (key.offset != offset)
38c227d8
LB		 2288 continue;
2289
e68afa49 2290 extent_offset = btrfs_file_extent_offset(leaf, extent);
38c227d8 2291 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
e68afa49 2292
38c227d8
LB		 2293 if (extent_offset >= old->extent_offset + old->offset +
2294 old->len || extent_offset + num_bytes <=
2295 old->extent_offset + old->offset)
2296 continue;
38c227d8
LB		 2297 break;
2298 }
2299
2300 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2301 if (!backref) {
2302 ret = -ENOENT;
2303 goto out;
2304 }
2305
2306 backref->root_id = root_id;
2307 backref->inum = inum;
e68afa49 2308 backref->file_pos = offset;
38c227d8
LB		 2309 backref->num_bytes = num_bytes;
2310 backref->extent_offset = extent_offset;
2311 backref->generation = btrfs_file_extent_generation(leaf, extent);
2312 backref->old = old;
2313 backref_insert(&new->root, backref);
2314 old->count++;
2315out:
2316 btrfs_release_path(path);
2317 WARN_ON(ret);
2318 return ret;
2319}
2320
2321static noinline bool record_extent_backrefs(struct btrfs_path *path,
2322 struct new_sa_defrag_extent *new)
2323{
2324 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2325 struct old_sa_defrag_extent *old, *tmp;
2326 int ret;
2327
2328 new->path = path;
2329
2330 list_for_each_entry_safe(old, tmp, &new->head, list) {
e68afa49
LB		 2331 ret = iterate_inodes_from_logical(old->bytenr +
2332 old->extent_offset, fs_info,
38c227d8
LB		 2333 path, record_one_backref,
2334 old);
4724b106
JB		 2335 if (ret < 0 && ret != -ENOENT)
2336 return false;
38c227d8
LB		 2337
2338 /* no backref to be processed for this extent */
2339 if (!old->count) {
2340 list_del(&old->list);
2341 kfree(old);
2342 }
2343 }
2344
2345 if (list_empty(&new->head))
2346 return false;
2347
2348 return true;
2349}
2350
2351static int relink_is_mergable(struct extent_buffer *leaf,
2352 struct btrfs_file_extent_item *fi,
116e0024 2353 struct new_sa_defrag_extent *new)
38c227d8 2354{
116e0024 2355 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
38c227d8
LB		 2356 return 0;
2357
2358 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2359 return 0;
2360
116e0024
LB		 2361 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2362 return 0;
2363
2364 if (btrfs_file_extent_encryption(leaf, fi) ||
38c227d8
LB		 2365 btrfs_file_extent_other_encoding(leaf, fi))
2366 return 0;
2367
2368 return 1;
2369}
2370
2371/*
2372 * Note the backref might has changed, and in this case we just return 0.
2373 */
2374static noinline int relink_extent_backref(struct btrfs_path *path,
2375 struct sa_defrag_extent_backref *prev,
2376 struct sa_defrag_extent_backref *backref)
2377{
2378 struct btrfs_file_extent_item *extent;
2379 struct btrfs_file_extent_item *item;
2380 struct btrfs_ordered_extent *ordered;
2381 struct btrfs_trans_handle *trans;
2382 struct btrfs_fs_info *fs_info;
2383 struct btrfs_root *root;
2384 struct btrfs_key key;
2385 struct extent_buffer *leaf;
2386 struct old_sa_defrag_extent *old = backref->old;
2387 struct new_sa_defrag_extent *new = old->new;
2388 struct inode *src_inode = new->inode;
2389 struct inode *inode;
2390 struct extent_state *cached = NULL;
2391 int ret = 0;
2392 u64 start;
2393 u64 len;
2394 u64 lock_start;
2395 u64 lock_end;
2396 bool merge = false;
2397 int index;
2398
2399 if (prev && prev->root_id == backref->root_id &&
2400 prev->inum == backref->inum &&
2401 prev->file_pos + prev->num_bytes == backref->file_pos)
2402 merge = true;
2403
2404 /* step 1: get root */
2405 key.objectid = backref->root_id;
2406 key.type = BTRFS_ROOT_ITEM_KEY;
2407 key.offset = (u64)-1;
2408
2409 fs_info = BTRFS_I(src_inode)->root->fs_info;
2410 index = srcu_read_lock(&fs_info->subvol_srcu);
2411
2412 root = btrfs_read_fs_root_no_name(fs_info, &key);
2413 if (IS_ERR(root)) {
2414 srcu_read_unlock(&fs_info->subvol_srcu, index);
2415 if (PTR_ERR(root) == -ENOENT)
2416 return 0;
2417 return PTR_ERR(root);
2418 }
38c227d8 2419
bcbba5e6
WS		 2420 if (btrfs_root_readonly(root)) {
2421 srcu_read_unlock(&fs_info->subvol_srcu, index);
2422 return 0;
2423 }
2424
38c227d8
LB		 2425 /* step 2: get inode */
2426 key.objectid = backref->inum;
2427 key.type = BTRFS_INODE_ITEM_KEY;
2428 key.offset = 0;
2429
2430 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2431 if (IS_ERR(inode)) {
2432 srcu_read_unlock(&fs_info->subvol_srcu, index);
2433 return 0;
2434 }
2435
2436 srcu_read_unlock(&fs_info->subvol_srcu, index);
2437
2438 /* step 3: relink backref */
2439 lock_start = backref->file_pos;
2440 lock_end = backref->file_pos + backref->num_bytes - 1;
2441 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2442 0, &cached);
2443
2444 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2445 if (ordered) {
2446 btrfs_put_ordered_extent(ordered);
2447 goto out_unlock;
2448 }
2449
2450 trans = btrfs_join_transaction(root);
2451 if (IS_ERR(trans)) {
2452 ret = PTR_ERR(trans);
2453 goto out_unlock;
2454 }
2455
2456 key.objectid = backref->inum;
2457 key.type = BTRFS_EXTENT_DATA_KEY;
2458 key.offset = backref->file_pos;
2459
2460 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2461 if (ret < 0) {
2462 goto out_free_path;
2463 } else if (ret > 0) {
2464 ret = 0;
2465 goto out_free_path;
2466 }
2467
2468 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2469 struct btrfs_file_extent_item);
2470
2471 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2472 backref->generation)
2473 goto out_free_path;
2474
2475 btrfs_release_path(path);
2476
2477 start = backref->file_pos;
2478 if (backref->extent_offset < old->extent_offset + old->offset)
2479 start += old->extent_offset + old->offset -
2480 backref->extent_offset;
2481
2482 len = min(backref->extent_offset + backref->num_bytes,
2483 old->extent_offset + old->offset + old->len);
2484 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2485
2486 ret = btrfs_drop_extents(trans, root, inode, start,
2487 start + len, 1);
2488 if (ret)
2489 goto out_free_path;
2490again:
2491 key.objectid = btrfs_ino(inode);
2492 key.type = BTRFS_EXTENT_DATA_KEY;
2493 key.offset = start;
2494
a09a0a70 2495 path->leave_spinning = 1;
38c227d8
LB		 2496 if (merge) {
2497 struct btrfs_file_extent_item *fi;
2498 u64 extent_len;
2499 struct btrfs_key found_key;
2500
3c9665df 2501 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
38c227d8
LB		 2502 if (ret < 0)
2503 goto out_free_path;
2504
2505 path->slots[0]--;
2506 leaf = path->nodes[0];
2507 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2508
2509 fi = btrfs_item_ptr(leaf, path->slots[0],
2510 struct btrfs_file_extent_item);
2511 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2512
116e0024
LB		 2513 if (extent_len + found_key.offset == start &&
2514 relink_is_mergable(leaf, fi, new)) {
38c227d8
LB		 2515 btrfs_set_file_extent_num_bytes(leaf, fi,
2516 extent_len + len);
2517 btrfs_mark_buffer_dirty(leaf);
2518 inode_add_bytes(inode, len);
2519
2520 ret = 1;
2521 goto out_free_path;
2522 } else {
2523 merge = false;
2524 btrfs_release_path(path);
2525 goto again;
2526 }
2527 }
2528
2529 ret = btrfs_insert_empty_item(trans, root, path, &key,
2530 sizeof(*extent));
2531 if (ret) {
2532 btrfs_abort_transaction(trans, root, ret);
2533 goto out_free_path;
2534 }
2535
2536 leaf = path->nodes[0];
2537 item = btrfs_item_ptr(leaf, path->slots[0],
2538 struct btrfs_file_extent_item);
2539 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2540 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2541 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2542 btrfs_set_file_extent_num_bytes(leaf, item, len);
2543 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2544 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2545 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2546 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2547 btrfs_set_file_extent_encryption(leaf, item, 0);
2548 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2549
2550 btrfs_mark_buffer_dirty(leaf);
2551 inode_add_bytes(inode, len);
a09a0a70 2552 btrfs_release_path(path);
38c227d8
LB		 2553
2554 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2555 new->disk_len, 0,
2556 backref->root_id, backref->inum,
2557 new->file_pos, 0); /* start - extent_offset */
2558 if (ret) {
2559 btrfs_abort_transaction(trans, root, ret);
2560 goto out_free_path;
2561 }
2562
2563 ret = 1;
2564out_free_path:
2565 btrfs_release_path(path);
a09a0a70 2566 path->leave_spinning = 0;
38c227d8
LB		 2567 btrfs_end_transaction(trans, root);
2568out_unlock:
2569 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2570 &cached, GFP_NOFS);
2571 iput(inode);
2572 return ret;
2573}
2574
6f519564
LB		 2575static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2576{
2577 struct old_sa_defrag_extent *old, *tmp;
2578
2579 if (!new)
2580 return;
2581
2582 list_for_each_entry_safe(old, tmp, &new->head, list) {
2583 list_del(&old->list);
2584 kfree(old);
2585 }
2586 kfree(new);
2587}
2588
38c227d8
LB		 2589static void relink_file_extents(struct new_sa_defrag_extent *new)
2590{
2591 struct btrfs_path *path;
38c227d8
LB		 2592 struct sa_defrag_extent_backref *backref;
2593 struct sa_defrag_extent_backref *prev = NULL;
2594 struct inode *inode;
2595 struct btrfs_root *root;
2596 struct rb_node *node;
2597 int ret;
2598
2599 inode = new->inode;
2600 root = BTRFS_I(inode)->root;
2601
2602 path = btrfs_alloc_path();
2603 if (!path)
2604 return;
2605
2606 if (!record_extent_backrefs(path, new)) {
2607 btrfs_free_path(path);
2608 goto out;
2609 }
2610 btrfs_release_path(path);
2611
2612 while (1) {
2613 node = rb_first(&new->root);
2614 if (!node)
2615 break;
2616 rb_erase(node, &new->root);
2617
2618 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2619
2620 ret = relink_extent_backref(path, prev, backref);
2621 WARN_ON(ret < 0);
2622
2623 kfree(prev);
2624
2625 if (ret == 1)
2626 prev = backref;
2627 else
2628 prev = NULL;
2629 cond_resched();
2630 }
2631 kfree(prev);
2632
2633 btrfs_free_path(path);
38c227d8 2634out:
6f519564
LB		 2635 free_sa_defrag_extent(new);
2636
38c227d8
LB		 2637 atomic_dec(&root->fs_info->defrag_running);
2638 wake_up(&root->fs_info->transaction_wait);
38c227d8
LB		 2639}
2640
2641static struct new_sa_defrag_extent *
2642record_old_file_extents(struct inode *inode,
2643 struct btrfs_ordered_extent *ordered)
2644{
2645 struct btrfs_root *root = BTRFS_I(inode)->root;
2646 struct btrfs_path *path;
2647 struct btrfs_key key;
6f519564 2648 struct old_sa_defrag_extent *old;
38c227d8
LB		 2649 struct new_sa_defrag_extent *new;
2650 int ret;
2651
2652 new = kmalloc(sizeof(*new), GFP_NOFS);
2653 if (!new)
2654 return NULL;
2655
2656 new->inode = inode;
2657 new->file_pos = ordered->file_offset;
2658 new->len = ordered->len;
2659 new->bytenr = ordered->start;
2660 new->disk_len = ordered->disk_len;
2661 new->compress_type = ordered->compress_type;
2662 new->root = RB_ROOT;
2663 INIT_LIST_HEAD(&new->head);
2664
2665 path = btrfs_alloc_path();
2666 if (!path)
2667 goto out_kfree;
2668
2669 key.objectid = btrfs_ino(inode);
2670 key.type = BTRFS_EXTENT_DATA_KEY;
2671 key.offset = new->file_pos;
2672
2673 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2674 if (ret < 0)
2675 goto out_free_path;
2676 if (ret > 0 && path->slots[0] > 0)
2677 path->slots[0]--;
2678
2679 /* find out all the old extents for the file range */
2680 while (1) {
2681 struct btrfs_file_extent_item *extent;
2682 struct extent_buffer *l;
2683 int slot;
2684 u64 num_bytes;
2685 u64 offset;
2686 u64 end;
2687 u64 disk_bytenr;
2688 u64 extent_offset;
2689
2690 l = path->nodes[0];
2691 slot = path->slots[0];
2692
2693 if (slot >= btrfs_header_nritems(l)) {
2694 ret = btrfs_next_leaf(root, path);
2695 if (ret < 0)
6f519564 2696 goto out_free_path;
38c227d8
LB		 2697 else if (ret > 0)
2698 break;
2699 continue;
2700 }
2701
2702 btrfs_item_key_to_cpu(l, &key, slot);
2703
2704 if (key.objectid != btrfs_ino(inode))
2705 break;
2706 if (key.type != BTRFS_EXTENT_DATA_KEY)
2707 break;
2708 if (key.offset >= new->file_pos + new->len)
2709 break;
2710
2711 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2712
2713 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2714 if (key.offset + num_bytes < new->file_pos)
2715 goto next;
2716
2717 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2718 if (!disk_bytenr)
2719 goto next;
2720
2721 extent_offset = btrfs_file_extent_offset(l, extent);
2722
2723 old = kmalloc(sizeof(*old), GFP_NOFS);
2724 if (!old)
6f519564 2725 goto out_free_path;
38c227d8
LB		 2726
2727 offset = max(new->file_pos, key.offset);
2728 end = min(new->file_pos + new->len, key.offset + num_bytes);
2729
2730 old->bytenr = disk_bytenr;
2731 old->extent_offset = extent_offset;
2732 old->offset = offset - key.offset;
2733 old->len = end - offset;
2734 old->new = new;
2735 old->count = 0;
2736 list_add_tail(&old->list, &new->head);
2737next:
2738 path->slots[0]++;
2739 cond_resched();
2740 }
2741
2742 btrfs_free_path(path);
2743 atomic_inc(&root->fs_info->defrag_running);
2744
2745 return new;
2746
38c227d8
LB		 2747out_free_path:
2748 btrfs_free_path(path);
2749out_kfree:
6f519564 2750 free_sa_defrag_extent(new);
38c227d8
LB		 2751 return NULL;
2752}
2753
e570fd27
MX		 2754static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2755 u64 start, u64 len)
2756{
2757 struct btrfs_block_group_cache *cache;
2758
2759 cache = btrfs_lookup_block_group(root->fs_info, start);
2760 ASSERT(cache);
2761
2762 spin_lock(&cache->lock);
2763 cache->delalloc_bytes -= len;
2764 spin_unlock(&cache->lock);
2765
2766 btrfs_put_block_group(cache);
2767}
2768
d352ac68
CM		 2769/* as ordered data IO finishes, this gets called so we can finish
2770 * an ordered extent if the range of bytes in the file it covers are
2771 * fully written.
2772 */
5fd02043 2773static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
e6dcd2dc 2774{
5fd02043 2775 struct inode *inode = ordered_extent->inode;
e6dcd2dc 2776 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 2777 struct btrfs_trans_handle *trans = NULL;
e6dcd2dc 2778 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2ac55d41 2779 struct extent_state *cached_state = NULL;
38c227d8 2780 struct new_sa_defrag_extent *new = NULL;
261507a0 2781 int compress_type = 0;
77cef2ec
JB		 2782 int ret = 0;
2783 u64 logical_len = ordered_extent->len;
82d5902d 2784 bool nolock;
77cef2ec 2785 bool truncated = false;
e6dcd2dc 2786
83eea1f1 2787 nolock = btrfs_is_free_space_inode(inode);
0cb59c99 2788
5fd02043
JB		 2789 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2790 ret = -EIO;
2791 goto out;
2792 }
2793
f612496b
MX		 2794 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2795 ordered_extent->file_offset +
2796 ordered_extent->len - 1);
2797
77cef2ec
JB		 2798 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2799 truncated = true;
2800 logical_len = ordered_extent->truncated_len;
2801 /* Truncated the entire extent, don't bother adding */
2802 if (!logical_len)
2803 goto out;
2804 }
2805
c2167754 2806 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
79787eaa 2807 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
6c760c07
JB		 2808 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2809 if (nolock)
2810 trans = btrfs_join_transaction_nolock(root);
2811 else
2812 trans = btrfs_join_transaction(root);
2813 if (IS_ERR(trans)) {
2814 ret = PTR_ERR(trans);
2815 trans = NULL;
2816 goto out;
c2167754 2817 }
6c760c07
JB		 2818 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2819 ret = btrfs_update_inode_fallback(trans, root, inode);
2820 if (ret) /* -ENOMEM or corruption */
2821 btrfs_abort_transaction(trans, root, ret);
c2167754
YZ		 2822 goto out;
2823 }
e6dcd2dc 2824
2ac55d41
JB		 2825 lock_extent_bits(io_tree, ordered_extent->file_offset,
2826 ordered_extent->file_offset + ordered_extent->len - 1,
d0082371 2827 0, &cached_state);
e6dcd2dc 2828
38c227d8
LB		 2829 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2830 ordered_extent->file_offset + ordered_extent->len - 1,
2831 EXTENT_DEFRAG, 1, cached_state);
2832 if (ret) {
2833 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
8101c8db 2834 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
38c227d8
LB		 2835 /* the inode is shared */
2836 new = record_old_file_extents(inode, ordered_extent);
2837
2838 clear_extent_bit(io_tree, ordered_extent->file_offset,
2839 ordered_extent->file_offset + ordered_extent->len - 1,
2840 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2841 }
2842
0cb59c99 2843 if (nolock)
7a7eaa40 2844 trans = btrfs_join_transaction_nolock(root);
0cb59c99 2845 else
7a7eaa40 2846 trans = btrfs_join_transaction(root);
79787eaa
JM		 2847 if (IS_ERR(trans)) {
2848 ret = PTR_ERR(trans);
2849 trans = NULL;
2850 goto out_unlock;
2851 }
a79b7d4b 2852
0ca1f7ce 2853 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
c2167754 2854
c8b97818 2855 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
261507a0 2856 compress_type = ordered_extent->compress_type;
d899e052 2857 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
261507a0 2858 BUG_ON(compress_type);
920bbbfb 2859 ret = btrfs_mark_extent_written(trans, inode,
d899e052
YZ		 2860 ordered_extent->file_offset,
2861 ordered_extent->file_offset +
77cef2ec 2862 logical_len);
d899e052 2863 } else {
0af3d00b 2864 BUG_ON(root == root->fs_info->tree_root);
d899e052
YZ		 2865 ret = insert_reserved_file_extent(trans, inode,
2866 ordered_extent->file_offset,
2867 ordered_extent->start,
2868 ordered_extent->disk_len,
77cef2ec 2869 logical_len, logical_len,
261507a0 2870 compress_type, 0, 0,
d899e052 2871 BTRFS_FILE_EXTENT_REG);
e570fd27
MX		 2872 if (!ret)
2873 btrfs_release_delalloc_bytes(root,
2874 ordered_extent->start,
2875 ordered_extent->disk_len);
d899e052 2876 }
5dc562c5
JB		 2877 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2878 ordered_extent->file_offset, ordered_extent->len,
2879 trans->transid);
79787eaa
JM		 2880 if (ret < 0) {
2881 btrfs_abort_transaction(trans, root, ret);
5fd02043 2882 goto out_unlock;
79787eaa 2883 }
2ac55d41 2884
e6dcd2dc
CM		 2885 add_pending_csums(trans, inode, ordered_extent->file_offset,
2886 &ordered_extent->list);
2887
6c760c07
JB		 2888 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2889 ret = btrfs_update_inode_fallback(trans, root, inode);
2890 if (ret) { /* -ENOMEM or corruption */
2891 btrfs_abort_transaction(trans, root, ret);
2892 goto out_unlock;
1ef30be1
JB		 2893 }
2894 ret = 0;
5fd02043
JB		 2895out_unlock:
2896 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2897 ordered_extent->file_offset +
2898 ordered_extent->len - 1, &cached_state, GFP_NOFS);
c2167754 2899out:
5b0e95bf 2900 if (root != root->fs_info->tree_root)
0cb59c99 2901 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
a698d075
MX		 2902 if (trans)
2903 btrfs_end_transaction(trans, root);
0cb59c99 2904
77cef2ec
JB		 2905 if (ret || truncated) {
2906 u64 start, end;
2907
2908 if (truncated)
2909 start = ordered_extent->file_offset + logical_len;
2910 else
2911 start = ordered_extent->file_offset;
2912 end = ordered_extent->file_offset + ordered_extent->len - 1;
2913 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2914
2915 /* Drop the cache for the part of the extent we didn't write. */
2916 btrfs_drop_extent_cache(inode, start, end, 0);
5fd02043 2917
0bec9ef5
JB		 2918 /*
2919 * If the ordered extent had an IOERR or something else went
2920 * wrong we need to return the space for this ordered extent
77cef2ec
JB		 2921 * back to the allocator. We only free the extent in the
2922 * truncated case if we didn't write out the extent at all.
0bec9ef5 2923 */
77cef2ec
JB		 2924 if ((ret || !logical_len) &&
2925 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
0bec9ef5
JB		 2926 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2927 btrfs_free_reserved_extent(root, ordered_extent->start,
e570fd27 2928 ordered_extent->disk_len, 1);
0bec9ef5
JB		 2929 }
2930
2931
5fd02043 2932 /*
8bad3c02
LB		 2933 * This needs to be done to make sure anybody waiting knows we are done
2934 * updating everything for this ordered extent.
5fd02043
JB		 2935 */
2936 btrfs_remove_ordered_extent(inode, ordered_extent);
2937
38c227d8 2938 /* for snapshot-aware defrag */
6f519564
LB		 2939 if (new) {
2940 if (ret) {
2941 free_sa_defrag_extent(new);
2942 atomic_dec(&root->fs_info->defrag_running);
2943 } else {
2944 relink_file_extents(new);
2945 }
2946 }
38c227d8 2947
e6dcd2dc
CM		 2948 /* once for us */
2949 btrfs_put_ordered_extent(ordered_extent);
2950 /* once for the tree */
2951 btrfs_put_ordered_extent(ordered_extent);
2952
5fd02043
JB		 2953 return ret;
2954}
2955
2956static void finish_ordered_fn(struct btrfs_work *work)
2957{
2958 struct btrfs_ordered_extent *ordered_extent;
2959 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2960 btrfs_finish_ordered_io(ordered_extent);
e6dcd2dc
CM		 2961}
2962
b2950863 2963static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM		 2964 struct extent_state *state, int uptodate)
2965{
5fd02043
JB		 2966 struct inode *inode = page->mapping->host;
2967 struct btrfs_root *root = BTRFS_I(inode)->root;
2968 struct btrfs_ordered_extent *ordered_extent = NULL;
9e0af237
LB		 2969 struct btrfs_workqueue *wq;
2970 btrfs_work_func_t func;
5fd02043 2971
1abe9b8a 2972 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2973
8b62b72b 2974 ClearPagePrivate2(page);
5fd02043
JB		 2975 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2976 end - start + 1, uptodate))
2977 return 0;
2978
9e0af237
LB		 2979 if (btrfs_is_free_space_inode(inode)) {
2980 wq = root->fs_info->endio_freespace_worker;
2981 func = btrfs_freespace_write_helper;
2982 } else {
2983 wq = root->fs_info->endio_write_workers;
2984 func = btrfs_endio_write_helper;
2985 }
5fd02043 2986
9e0af237
LB		 2987 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
2988 NULL);
2989 btrfs_queue_work(wq, &ordered_extent->work);
5fd02043
JB		 2990
2991 return 0;
211f90e6
CM		 2992}
2993
dc380aea
MX		 2994static int __readpage_endio_check(struct inode *inode,
2995 struct btrfs_io_bio *io_bio,
2996 int icsum, struct page *page,
2997 int pgoff, u64 start, size_t len)
2998{
2999 char *kaddr;
3000 u32 csum_expected;
3001 u32 csum = ~(u32)0;
3002 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
3003 DEFAULT_RATELIMIT_BURST);
3004
3005 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3006
3007 kaddr = kmap_atomic(page);
3008 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3009 btrfs_csum_final(csum, (char *)&csum);
3010 if (csum != csum_expected)
3011 goto zeroit;
3012
3013 kunmap_atomic(kaddr);
3014 return 0;
3015zeroit:
3016 if (__ratelimit(&_rs))
f0954c66 3017 btrfs_warn(BTRFS_I(inode)->root->fs_info,
dc380aea
MX		 3018 "csum failed ino %llu off %llu csum %u expected csum %u",
3019 btrfs_ino(inode), start, csum, csum_expected);
3020 memset(kaddr + pgoff, 1, len);
3021 flush_dcache_page(page);
3022 kunmap_atomic(kaddr);
3023 if (csum_expected == 0)
3024 return 0;
3025 return -EIO;
3026}
3027
d352ac68
CM		 3028/*
3029 * when reads are done, we need to check csums to verify the data is correct
4a54c8c1
JS		 3030 * if there's a match, we allow the bio to finish. If not, the code in
3031 * extent_io.c will try to find good copies for us.
d352ac68 3032 */
facc8a22
MX		 3033static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3034 u64 phy_offset, struct page *page,
3035 u64 start, u64 end, int mirror)
07157aac 3036{
4eee4fa4 3037 size_t offset = start - page_offset(page);
07157aac 3038 struct inode *inode = page->mapping->host;
d1310b2e 3039 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
ff79f819 3040 struct btrfs_root *root = BTRFS_I(inode)->root;
d1310b2e 3041
d20f7043
CM		 3042 if (PageChecked(page)) {
3043 ClearPageChecked(page);
dc380aea 3044 return 0;
d20f7043 3045 }
6cbff00f
CH		 3046
3047 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
dc380aea 3048 return 0;
17d217fe
YZ		 3049
3050 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
9655d298 3051 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
17d217fe
YZ		 3052 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
3053 GFP_NOFS);
b6cda9bc 3054 return 0;
17d217fe 3055 }
d20f7043 3056
facc8a22 3057 phy_offset >>= inode->i_sb->s_blocksize_bits;
dc380aea
MX		 3058 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3059 start, (size_t)(end - start + 1));
07157aac 3060}
b888db2b 3061
24bbcf04
YZ		 3062struct delayed_iput {
3063 struct list_head list;
3064 struct inode *inode;
3065};
3066
79787eaa
JM		 3067/* JDM: If this is fs-wide, why can't we add a pointer to
3068 * btrfs_inode instead and avoid the allocation? */
24bbcf04
YZ		 3069void btrfs_add_delayed_iput(struct inode *inode)
3070{
3071 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3072 struct delayed_iput *delayed;
3073
3074 if (atomic_add_unless(&inode->i_count, -1, 1))
3075 return;
3076
3077 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
3078 delayed->inode = inode;
3079
3080 spin_lock(&fs_info->delayed_iput_lock);
3081 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
3082 spin_unlock(&fs_info->delayed_iput_lock);
3083}
3084
3085void btrfs_run_delayed_iputs(struct btrfs_root *root)
3086{
3087 LIST_HEAD(list);
3088 struct btrfs_fs_info *fs_info = root->fs_info;
3089 struct delayed_iput *delayed;
3090 int empty;
3091
3092 spin_lock(&fs_info->delayed_iput_lock);
3093 empty = list_empty(&fs_info->delayed_iputs);
3094 spin_unlock(&fs_info->delayed_iput_lock);
3095 if (empty)
3096 return;
3097
24bbcf04
YZ		 3098 spin_lock(&fs_info->delayed_iput_lock);
3099 list_splice_init(&fs_info->delayed_iputs, &list);
3100 spin_unlock(&fs_info->delayed_iput_lock);
3101
3102 while (!list_empty(&list)) {
3103 delayed = list_entry(list.next, struct delayed_iput, list);
3104 list_del(&delayed->list);
3105 iput(delayed->inode);
3106 kfree(delayed);
3107 }
24bbcf04
YZ		 3108}
3109
d68fc57b 3110/*
42b2aa86 3111 * This is called in transaction commit time. If there are no orphan
d68fc57b
YZ		 3112 * files in the subvolume, it removes orphan item and frees block_rsv
3113 * structure.
3114 */
3115void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3116 struct btrfs_root *root)
3117{
90290e19 3118 struct btrfs_block_rsv *block_rsv;
d68fc57b
YZ		 3119 int ret;
3120
8a35d95f 3121 if (atomic_read(&root->orphan_inodes) ||
d68fc57b
YZ		 3122 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3123 return;
3124
90290e19 3125 spin_lock(&root->orphan_lock);
8a35d95f 3126 if (atomic_read(&root->orphan_inodes)) {
90290e19
JB		 3127 spin_unlock(&root->orphan_lock);
3128 return;
3129 }
3130
3131 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3132 spin_unlock(&root->orphan_lock);
3133 return;
3134 }
3135
3136 block_rsv = root->orphan_block_rsv;
3137 root->orphan_block_rsv = NULL;
3138 spin_unlock(&root->orphan_lock);
3139
27cdeb70 3140 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
d68fc57b
YZ		 3141 btrfs_root_refs(&root->root_item) > 0) {
3142 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3143 root->root_key.objectid);
4ef31a45
JB		 3144 if (ret)
3145 btrfs_abort_transaction(trans, root, ret);
3146 else
27cdeb70
MX		 3147 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3148 &root->state);
d68fc57b
YZ		 3149 }
3150
90290e19
JB		 3151 if (block_rsv) {
3152 WARN_ON(block_rsv->size > 0);
3153 btrfs_free_block_rsv(root, block_rsv);
d68fc57b
YZ		 3154 }
3155}
3156
7b128766
JB		 3157/*
3158 * This creates an orphan entry for the given inode in case something goes
3159 * wrong in the middle of an unlink/truncate.
d68fc57b
YZ		 3160 *
3161 * NOTE: caller of this function should reserve 5 units of metadata for
3162 * this function.
7b128766
JB		 3163 */
3164int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3165{
3166 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 3167 struct btrfs_block_rsv *block_rsv = NULL;
3168 int reserve = 0;
3169 int insert = 0;
3170 int ret;
7b128766 3171
d68fc57b 3172 if (!root->orphan_block_rsv) {
66d8f3dd 3173 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
b532402e
TI		 3174 if (!block_rsv)
3175 return -ENOMEM;
d68fc57b 3176 }
7b128766 3177
d68fc57b
YZ		 3178 spin_lock(&root->orphan_lock);
3179 if (!root->orphan_block_rsv) {
3180 root->orphan_block_rsv = block_rsv;
3181 } else if (block_rsv) {
3182 btrfs_free_block_rsv(root, block_rsv);
3183 block_rsv = NULL;
7b128766 3184 }
7b128766 3185
8a35d95f
JB		 3186 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3187 &BTRFS_I(inode)->runtime_flags)) {
d68fc57b
YZ		 3188#if 0
3189 /*
3190 * For proper ENOSPC handling, we should do orphan
3191 * cleanup when mounting. But this introduces backward
3192 * compatibility issue.
3193 */
3194 if (!xchg(&root->orphan_item_inserted, 1))
3195 insert = 2;
3196 else
3197 insert = 1;
3198#endif
3199 insert = 1;
321f0e70 3200 atomic_inc(&root->orphan_inodes);
7b128766
JB		 3201 }
3202
72ac3c0d
JB		 3203 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3204 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3205 reserve = 1;
d68fc57b 3206 spin_unlock(&root->orphan_lock);
7b128766 3207
d68fc57b
YZ		 3208 /* grab metadata reservation from transaction handle */
3209 if (reserve) {
3210 ret = btrfs_orphan_reserve_metadata(trans, inode);
79787eaa 3211 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
d68fc57b 3212 }
7b128766 3213
d68fc57b
YZ		 3214 /* insert an orphan item to track this unlinked/truncated file */
3215 if (insert >= 1) {
33345d01 3216 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
4ef31a45 3217 if (ret) {
703c88e0 3218 atomic_dec(&root->orphan_inodes);
4ef31a45
JB		 3219 if (reserve) {
3220 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3221 &BTRFS_I(inode)->runtime_flags);
3222 btrfs_orphan_release_metadata(inode);
3223 }
3224 if (ret != -EEXIST) {
e8e7cff6
JB		 3225 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3226 &BTRFS_I(inode)->runtime_flags);
4ef31a45
JB		 3227 btrfs_abort_transaction(trans, root, ret);
3228 return ret;
3229 }
79787eaa
JM		 3230 }
3231 ret = 0;
d68fc57b
YZ		 3232 }
3233
3234 /* insert an orphan item to track subvolume contains orphan files */
3235 if (insert >= 2) {
3236 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3237 root->root_key.objectid);
79787eaa
JM		 3238 if (ret && ret != -EEXIST) {
3239 btrfs_abort_transaction(trans, root, ret);
3240 return ret;
3241 }
d68fc57b
YZ		 3242 }
3243 return 0;
7b128766
JB		 3244}
3245
3246/*
3247 * We have done the truncate/delete so we can go ahead and remove the orphan
3248 * item for this particular inode.
3249 */
48a3b636
ES		 3250static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3251 struct inode *inode)
7b128766
JB		 3252{
3253 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 3254 int delete_item = 0;
3255 int release_rsv = 0;
7b128766
JB		 3256 int ret = 0;
3257
d68fc57b 3258 spin_lock(&root->orphan_lock);
8a35d95f
JB		 3259 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3260 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3261 delete_item = 1;
7b128766 3262
72ac3c0d
JB		 3263 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3264 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3265 release_rsv = 1;
d68fc57b 3266 spin_unlock(&root->orphan_lock);
7b128766 3267
703c88e0 3268 if (delete_item) {
8a35d95f 3269 atomic_dec(&root->orphan_inodes);
703c88e0
FDBM		 3270 if (trans)
3271 ret = btrfs_del_orphan_item(trans, root,
3272 btrfs_ino(inode));
8a35d95f 3273 }
7b128766 3274
703c88e0
FDBM		 3275 if (release_rsv)
3276 btrfs_orphan_release_metadata(inode);
3277
4ef31a45 3278 return ret;
7b128766
JB		 3279}
3280
3281/*
3282 * this cleans up any orphans that may be left on the list from the last use
3283 * of this root.
3284 */
66b4ffd1 3285int btrfs_orphan_cleanup(struct btrfs_root *root)
7b128766
JB		 3286{
3287 struct btrfs_path *path;
3288 struct extent_buffer *leaf;
7b128766
JB		 3289 struct btrfs_key key, found_key;
3290 struct btrfs_trans_handle *trans;
3291 struct inode *inode;
8f6d7f4f 3292 u64 last_objectid = 0;
7b128766
JB		 3293 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3294
d68fc57b 3295 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
66b4ffd1 3296 return 0;
c71bf099
YZ		 3297
3298 path = btrfs_alloc_path();
66b4ffd1
JB		 3299 if (!path) {
3300 ret = -ENOMEM;
3301 goto out;
3302 }
7b128766
JB		 3303 path->reada = -1;
3304
3305 key.objectid = BTRFS_ORPHAN_OBJECTID;
962a298f 3306 key.type = BTRFS_ORPHAN_ITEM_KEY;
7b128766
JB		 3307 key.offset = (u64)-1;
3308
7b128766
JB		 3309 while (1) {
3310 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
66b4ffd1
JB		 3311 if (ret < 0)
3312 goto out;
7b128766
JB		 3313
3314 /*
3315 * if ret == 0 means we found what we were searching for, which
25985edc 3316 * is weird, but possible, so only screw with path if we didn't
7b128766
JB		 3317 * find the key and see if we have stuff that matches
3318 */
3319 if (ret > 0) {
66b4ffd1 3320 ret = 0;
7b128766
JB		 3321 if (path->slots[0] == 0)
3322 break;
3323 path->slots[0]--;
3324 }
3325
3326 /* pull out the item */
3327 leaf = path->nodes[0];
7b128766
JB		 3328 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3329
3330 /* make sure the item matches what we want */
3331 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3332 break;
962a298f 3333 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
7b128766
JB		 3334 break;
3335
3336 /* release the path since we're done with it */
b3b4aa74 3337 btrfs_release_path(path);
7b128766
JB		 3338
3339 /*
3340 * this is where we are basically btrfs_lookup, without the
3341 * crossing root thing. we store the inode number in the
3342 * offset of the orphan item.
3343 */
8f6d7f4f
JB		 3344
3345 if (found_key.offset == last_objectid) {
c2cf52eb
SK		 3346 btrfs_err(root->fs_info,
3347 "Error removing orphan entry, stopping orphan cleanup");
8f6d7f4f
JB		 3348 ret = -EINVAL;
3349 goto out;
3350 }
3351
3352 last_objectid = found_key.offset;
3353
5d4f98a2
YZ		 3354 found_key.objectid = found_key.offset;
3355 found_key.type = BTRFS_INODE_ITEM_KEY;
3356 found_key.offset = 0;
73f73415 3357 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
8c6ffba0 3358 ret = PTR_ERR_OR_ZERO(inode);
a8c9e576 3359 if (ret && ret != -ESTALE)
66b4ffd1 3360 goto out;
7b128766 3361
f8e9e0b0
AJ		 3362 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3363 struct btrfs_root *dead_root;
3364 struct btrfs_fs_info *fs_info = root->fs_info;
3365 int is_dead_root = 0;
3366
3367 /*
3368 * this is an orphan in the tree root. Currently these
3369 * could come from 2 sources:
3370 * a) a snapshot deletion in progress
3371 * b) a free space cache inode
3372 * We need to distinguish those two, as the snapshot
3373 * orphan must not get deleted.
3374 * find_dead_roots already ran before us, so if this
3375 * is a snapshot deletion, we should find the root
3376 * in the dead_roots list
3377 */
3378 spin_lock(&fs_info->trans_lock);
3379 list_for_each_entry(dead_root, &fs_info->dead_roots,
3380 root_list) {
3381 if (dead_root->root_key.objectid ==
3382 found_key.objectid) {
3383 is_dead_root = 1;
3384 break;
3385 }
3386 }
3387 spin_unlock(&fs_info->trans_lock);
3388 if (is_dead_root) {
3389 /* prevent this orphan from being found again */
3390 key.offset = found_key.objectid - 1;
3391 continue;
3392 }
3393 }
7b128766 3394 /*
a8c9e576
JB		 3395 * Inode is already gone but the orphan item is still there,
3396 * kill the orphan item.
7b128766 3397 */
a8c9e576
JB		 3398 if (ret == -ESTALE) {
3399 trans = btrfs_start_transaction(root, 1);
66b4ffd1
JB		 3400 if (IS_ERR(trans)) {
3401 ret = PTR_ERR(trans);
3402 goto out;
3403 }
c2cf52eb
SK		 3404 btrfs_debug(root->fs_info, "auto deleting %Lu",
3405 found_key.objectid);
a8c9e576
JB		 3406 ret = btrfs_del_orphan_item(trans, root,
3407 found_key.objectid);
5b21f2ed 3408 btrfs_end_transaction(trans, root);
4ef31a45
JB		 3409 if (ret)
3410 goto out;
7b128766
JB		 3411 continue;
3412 }
3413
a8c9e576
JB		 3414 /*
3415 * add this inode to the orphan list so btrfs_orphan_del does
3416 * the proper thing when we hit it
3417 */
8a35d95f
JB		 3418 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3419 &BTRFS_I(inode)->runtime_flags);
925396ec 3420 atomic_inc(&root->orphan_inodes);
a8c9e576 3421
7b128766
JB		 3422 /* if we have links, this was a truncate, lets do that */
3423 if (inode->i_nlink) {
fae7f21c 3424 if (WARN_ON(!S_ISREG(inode->i_mode))) {
a41ad394
JB		 3425 iput(inode);
3426 continue;
3427 }
7b128766 3428 nr_truncate++;
f3fe820c
JB		 3429
3430 /* 1 for the orphan item deletion. */
3431 trans = btrfs_start_transaction(root, 1);
3432 if (IS_ERR(trans)) {
c69b26b0 3433 iput(inode);
f3fe820c
JB		 3434 ret = PTR_ERR(trans);
3435 goto out;
3436 }
3437 ret = btrfs_orphan_add(trans, inode);
3438 btrfs_end_transaction(trans, root);
c69b26b0
JB		 3439 if (ret) {
3440 iput(inode);
f3fe820c 3441 goto out;
c69b26b0 3442 }
f3fe820c 3443
66b4ffd1 3444 ret = btrfs_truncate(inode);
4a7d0f68
JB		 3445 if (ret)
3446 btrfs_orphan_del(NULL, inode);
7b128766
JB		 3447 } else {
3448 nr_unlink++;
3449 }
3450
3451 /* this will do delete_inode and everything for us */
3452 iput(inode);
66b4ffd1
JB		 3453 if (ret)
3454 goto out;
7b128766 3455 }
3254c876
MX		 3456 /* release the path since we're done with it */
3457 btrfs_release_path(path);
3458
d68fc57b
YZ		 3459 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3460
3461 if (root->orphan_block_rsv)
3462 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3463 (u64)-1);
3464
27cdeb70
MX		 3465 if (root->orphan_block_rsv ||
3466 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
7a7eaa40 3467 trans = btrfs_join_transaction(root);
66b4ffd1
JB		 3468 if (!IS_ERR(trans))
3469 btrfs_end_transaction(trans, root);
d68fc57b 3470 }
7b128766
JB		 3471
3472 if (nr_unlink)
4884b476 3473 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
7b128766 3474 if (nr_truncate)
4884b476 3475 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
66b4ffd1
JB		 3476
3477out:
3478 if (ret)
68b663d1 3479 btrfs_err(root->fs_info,
c2cf52eb 3480 "could not do orphan cleanup %d", ret);
66b4ffd1
JB		 3481 btrfs_free_path(path);
3482 return ret;
7b128766
JB		 3483}
3484
46a53cca
CM		 3485/*
3486 * very simple check to peek ahead in the leaf looking for xattrs. If we
3487 * don't find any xattrs, we know there can't be any acls.
3488 *
3489 * slot is the slot the inode is in, objectid is the objectid of the inode
3490 */
3491static noinline int acls_after_inode_item(struct extent_buffer *leaf,
63541927
FDBM		 3492 int slot, u64 objectid,
3493 int *first_xattr_slot)
46a53cca
CM		 3494{
3495 u32 nritems = btrfs_header_nritems(leaf);
3496 struct btrfs_key found_key;
f23b5a59
JB		 3497 static u64 xattr_access = 0;
3498 static u64 xattr_default = 0;
46a53cca
CM		 3499 int scanned = 0;
3500
f23b5a59
JB		 3501 if (!xattr_access) {
3502 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3503 strlen(POSIX_ACL_XATTR_ACCESS));
3504 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3505 strlen(POSIX_ACL_XATTR_DEFAULT));
3506 }
3507
46a53cca 3508 slot++;
63541927 3509 *first_xattr_slot = -1;
46a53cca
CM		 3510 while (slot < nritems) {
3511 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3512
3513 /* we found a different objectid, there must not be acls */
3514 if (found_key.objectid != objectid)
3515 return 0;
3516
3517 /* we found an xattr, assume we've got an acl */
f23b5a59 3518 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
63541927
FDBM		 3519 if (*first_xattr_slot == -1)
3520 *first_xattr_slot = slot;
f23b5a59
JB		 3521 if (found_key.offset == xattr_access ||
3522 found_key.offset == xattr_default)
3523 return 1;
3524 }
46a53cca
CM		 3525
3526 /*
3527 * we found a key greater than an xattr key, there can't
3528 * be any acls later on
3529 */
3530 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3531 return 0;
3532
3533 slot++;
3534 scanned++;
3535
3536 /*
3537 * it goes inode, inode backrefs, xattrs, extents,
3538 * so if there are a ton of hard links to an inode there can
3539 * be a lot of backrefs. Don't waste time searching too hard,
3540 * this is just an optimization
3541 */
3542 if (scanned >= 8)
3543 break;
3544 }
3545 /* we hit the end of the leaf before we found an xattr or
3546 * something larger than an xattr. We have to assume the inode
3547 * has acls
3548 */
63541927
FDBM		 3549 if (*first_xattr_slot == -1)
3550 *first_xattr_slot = slot;
46a53cca
CM		 3551 return 1;
3552}
3553
d352ac68
CM		 3554/*
3555 * read an inode from the btree into the in-memory inode
3556 */
5d4f98a2 3557static void btrfs_read_locked_inode(struct inode *inode)
39279cc3
CM		 3558{
3559 struct btrfs_path *path;
5f39d397 3560 struct extent_buffer *leaf;
39279cc3
CM		 3561 struct btrfs_inode_item *inode_item;
3562 struct btrfs_root *root = BTRFS_I(inode)->root;
3563 struct btrfs_key location;
67de1176 3564 unsigned long ptr;
46a53cca 3565 int maybe_acls;
618e21d5 3566 u32 rdev;
39279cc3 3567 int ret;
2f7e33d4 3568 bool filled = false;
63541927 3569 int first_xattr_slot;
2f7e33d4
MX		 3570
3571 ret = btrfs_fill_inode(inode, &rdev);
3572 if (!ret)
3573 filled = true;
39279cc3
CM		 3574
3575 path = btrfs_alloc_path();
1748f843
MF		 3576 if (!path)
3577 goto make_bad;
3578
39279cc3 3579 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 3580
39279cc3 3581 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 3582 if (ret)
39279cc3 3583 goto make_bad;
39279cc3 3584
5f39d397 3585 leaf = path->nodes[0];
2f7e33d4
MX		 3586
3587 if (filled)
67de1176 3588 goto cache_index;
2f7e33d4 3589
5f39d397
CM		 3590 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3591 struct btrfs_inode_item);
5f39d397 3592 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
bfe86848 3593 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2f2f43d3
EB		 3594 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3595 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
dbe674a9 3596 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397 3597
a937b979
DS		 3598 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3599 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
5f39d397 3600
a937b979
DS		 3601 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3602 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
5f39d397 3603
a937b979
DS		 3604 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3605 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
5f39d397 3606
9cc97d64 3607 BTRFS_I(inode)->i_otime.tv_sec =
3608 btrfs_timespec_sec(leaf, &inode_item->otime);
3609 BTRFS_I(inode)->i_otime.tv_nsec =
3610 btrfs_timespec_nsec(leaf, &inode_item->otime);
3611
a76a3cd4 3612 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 3613 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
5dc562c5
JB		 3614 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3615
3616 /*
3617 * If we were modified in the current generation and evicted from memory
3618 * and then re-read we need to do a full sync since we don't have any
3619 * idea about which extents were modified before we were evicted from
3620 * cache.
3621 */
3622 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3623 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3624 &BTRFS_I(inode)->runtime_flags);
3625
0c4d2d95 3626 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
e02119d5 3627 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 3628 inode->i_rdev = 0;
5f39d397
CM		 3629 rdev = btrfs_inode_rdev(leaf, inode_item);
3630
aec7477b 3631 BTRFS_I(inode)->index_cnt = (u64)-1;
d2fb3437 3632 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
67de1176
MX		 3633
3634cache_index:
3635 path->slots[0]++;
3636 if (inode->i_nlink != 1 ||
3637 path->slots[0] >= btrfs_header_nritems(leaf))
3638 goto cache_acl;
3639
3640 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3641 if (location.objectid != btrfs_ino(inode))
3642 goto cache_acl;
3643
3644 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3645 if (location.type == BTRFS_INODE_REF_KEY) {
3646 struct btrfs_inode_ref *ref;
3647
3648 ref = (struct btrfs_inode_ref *)ptr;
3649 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3650 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3651 struct btrfs_inode_extref *extref;
3652
3653 extref = (struct btrfs_inode_extref *)ptr;
3654 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3655 extref);
3656 }
2f7e33d4 3657cache_acl:
46a53cca
CM		 3658 /*
3659 * try to precache a NULL acl entry for files that don't have
3660 * any xattrs or acls
3661 */
33345d01 3662 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
63541927
FDBM		 3663 btrfs_ino(inode), &first_xattr_slot);
3664 if (first_xattr_slot != -1) {
3665 path->slots[0] = first_xattr_slot;
3666 ret = btrfs_load_inode_props(inode, path);
3667 if (ret)
3668 btrfs_err(root->fs_info,
351fd353 3669 "error loading props for ino %llu (root %llu): %d",
63541927
FDBM		 3670 btrfs_ino(inode),
3671 root->root_key.objectid, ret);
3672 }
3673 btrfs_free_path(path);
3674
72c04902
AV		 3675 if (!maybe_acls)
3676 cache_no_acl(inode);
46a53cca 3677
39279cc3 3678 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 3679 case S_IFREG:
3680 inode->i_mapping->a_ops = &btrfs_aops;
04160088 3681 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 3682 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 3683 inode->i_fop = &btrfs_file_operations;
3684 inode->i_op = &btrfs_file_inode_operations;
3685 break;
3686 case S_IFDIR:
3687 inode->i_fop = &btrfs_dir_file_operations;
3688 if (root == root->fs_info->tree_root)
3689 inode->i_op = &btrfs_dir_ro_inode_operations;
3690 else
3691 inode->i_op = &btrfs_dir_inode_operations;
3692 break;
3693 case S_IFLNK:
3694 inode->i_op = &btrfs_symlink_inode_operations;
3695 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 3696 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3 3697 break;
618e21d5 3698 default:
0279b4cd 3699 inode->i_op = &btrfs_special_inode_operations;
618e21d5
JB		 3700 init_special_inode(inode, inode->i_mode, rdev);
3701 break;
39279cc3 3702 }
6cbff00f
CH		 3703
3704 btrfs_update_iflags(inode);
39279cc3
CM		 3705 return;
3706
3707make_bad:
39279cc3 3708 btrfs_free_path(path);
39279cc3
CM		 3709 make_bad_inode(inode);
3710}
3711
d352ac68
CM		 3712/*
3713 * given a leaf and an inode, copy the inode fields into the leaf
3714 */
e02119d5
CM		 3715static void fill_inode_item(struct btrfs_trans_handle *trans,
3716 struct extent_buffer *leaf,
5f39d397 3717 struct btrfs_inode_item *item,
39279cc3
CM		 3718 struct inode *inode)
3719{
51fab693
LB		 3720 struct btrfs_map_token token;
3721
3722 btrfs_init_map_token(&token);
5f39d397 3723
51fab693
LB		 3724 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3725 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3726 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3727 &token);
3728 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3729 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
5f39d397 3730
a937b979 3731 btrfs_set_token_timespec_sec(leaf, &item->atime,
51fab693 3732 inode->i_atime.tv_sec, &token);
a937b979 3733 btrfs_set_token_timespec_nsec(leaf, &item->atime,
51fab693 3734 inode->i_atime.tv_nsec, &token);
5f39d397 3735
a937b979 3736 btrfs_set_token_timespec_sec(leaf, &item->mtime,
51fab693 3737 inode->i_mtime.tv_sec, &token);
a937b979 3738 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
51fab693 3739 inode->i_mtime.tv_nsec, &token);
5f39d397 3740
a937b979 3741 btrfs_set_token_timespec_sec(leaf, &item->ctime,
51fab693 3742 inode->i_ctime.tv_sec, &token);
a937b979 3743 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
51fab693 3744 inode->i_ctime.tv_nsec, &token);
5f39d397 3745
9cc97d64 3746 btrfs_set_token_timespec_sec(leaf, &item->otime,
3747 BTRFS_I(inode)->i_otime.tv_sec, &token);
3748 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3749 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3750
51fab693
LB		 3751 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3752 &token);
3753 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3754 &token);
3755 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3756 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3757 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3758 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3759 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
39279cc3
CM		 3760}
3761
d352ac68
CM		 3762/*
3763 * copy everything in the in-memory inode into the btree.
3764 */
2115133f 3765static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
d397712b 3766 struct btrfs_root *root, struct inode *inode)
39279cc3
CM		 3767{
3768 struct btrfs_inode_item *inode_item;
3769 struct btrfs_path *path;
5f39d397 3770 struct extent_buffer *leaf;
39279cc3
CM		 3771 int ret;
3772
3773 path = btrfs_alloc_path();
16cdcec7
MX		 3774 if (!path)
3775 return -ENOMEM;
3776
b9473439 3777 path->leave_spinning = 1;
16cdcec7
MX		 3778 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3779 1);
39279cc3
CM		 3780 if (ret) {
3781 if (ret > 0)
3782 ret = -ENOENT;
3783 goto failed;
3784 }
3785
5f39d397
CM		 3786 leaf = path->nodes[0];
3787 inode_item = btrfs_item_ptr(leaf, path->slots[0],
16cdcec7 3788 struct btrfs_inode_item);
39279cc3 3789
e02119d5 3790 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 3791 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 3792 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 3793 ret = 0;
3794failed:
39279cc3
CM		 3795 btrfs_free_path(path);
3796 return ret;
3797}
3798
2115133f
CM		 3799/*
3800 * copy everything in the in-memory inode into the btree.
3801 */
3802noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3803 struct btrfs_root *root, struct inode *inode)
3804{
3805 int ret;
3806
3807 /*
3808 * If the inode is a free space inode, we can deadlock during commit
3809 * if we put it into the delayed code.
3810 *
3811 * The data relocation inode should also be directly updated
3812 * without delay
3813 */
83eea1f1 3814 if (!btrfs_is_free_space_inode(inode)
1d52c78a
JB		 3815 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3816 && !root->fs_info->log_root_recovering) {
8ea05e3a
AB		 3817 btrfs_update_root_times(trans, root);
3818
2115133f
CM		 3819 ret = btrfs_delayed_update_inode(trans, root, inode);
3820 if (!ret)
3821 btrfs_set_inode_last_trans(trans, inode);
3822 return ret;
3823 }
3824
3825 return btrfs_update_inode_item(trans, root, inode);
3826}
3827
be6aef60
JB		 3828noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3829 struct btrfs_root *root,
3830 struct inode *inode)
2115133f
CM		 3831{
3832 int ret;
3833
3834 ret = btrfs_update_inode(trans, root, inode);
3835 if (ret == -ENOSPC)
3836 return btrfs_update_inode_item(trans, root, inode);
3837 return ret;
3838}
3839
d352ac68
CM		 3840/*
3841 * unlink helper that gets used here in inode.c and in the tree logging
3842 * recovery code. It remove a link in a directory with a given name, and
3843 * also drops the back refs in the inode to the directory
3844 */
92986796
AV		 3845static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3846 struct btrfs_root *root,
3847 struct inode *dir, struct inode *inode,
3848 const char *name, int name_len)
39279cc3
CM		 3849{
3850 struct btrfs_path *path;
39279cc3 3851 int ret = 0;
5f39d397 3852 struct extent_buffer *leaf;
39279cc3 3853 struct btrfs_dir_item *di;
5f39d397 3854 struct btrfs_key key;
aec7477b 3855 u64 index;
33345d01
LZ		 3856 u64 ino = btrfs_ino(inode);
3857 u64 dir_ino = btrfs_ino(dir);
39279cc3
CM		 3858
3859 path = btrfs_alloc_path();
54aa1f4d
CM		 3860 if (!path) {
3861 ret = -ENOMEM;
554233a6 3862 goto out;
54aa1f4d
CM		 3863 }
3864
b9473439 3865 path->leave_spinning = 1;
33345d01 3866 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
39279cc3
CM		 3867 name, name_len, -1);
3868 if (IS_ERR(di)) {
3869 ret = PTR_ERR(di);
3870 goto err;
3871 }
3872 if (!di) {
3873 ret = -ENOENT;
3874 goto err;
3875 }
5f39d397
CM		 3876 leaf = path->nodes[0];
3877 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 3878 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 3879 if (ret)
3880 goto err;
b3b4aa74 3881 btrfs_release_path(path);
39279cc3 3882
67de1176
MX		 3883 /*
3884 * If we don't have dir index, we have to get it by looking up
3885 * the inode ref, since we get the inode ref, remove it directly,
3886 * it is unnecessary to do delayed deletion.
3887 *
3888 * But if we have dir index, needn't search inode ref to get it.
3889 * Since the inode ref is close to the inode item, it is better
3890 * that we delay to delete it, and just do this deletion when
3891 * we update the inode item.
3892 */
3893 if (BTRFS_I(inode)->dir_index) {
3894 ret = btrfs_delayed_delete_inode_ref(inode);
3895 if (!ret) {
3896 index = BTRFS_I(inode)->dir_index;
3897 goto skip_backref;
3898 }
3899 }
3900
33345d01
LZ		 3901 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3902 dir_ino, &index);
aec7477b 3903 if (ret) {
c2cf52eb
SK		 3904 btrfs_info(root->fs_info,
3905 "failed to delete reference to %.*s, inode %llu parent %llu",
c1c9ff7c 3906 name_len, name, ino, dir_ino);
79787eaa 3907 btrfs_abort_transaction(trans, root, ret);
aec7477b
JB		 3908 goto err;
3909 }
67de1176 3910skip_backref:
16cdcec7 3911 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 3912 if (ret) {
3913 btrfs_abort_transaction(trans, root, ret);
39279cc3 3914 goto err;
79787eaa 3915 }
39279cc3 3916
e02119d5 3917 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
33345d01 3918 inode, dir_ino);
79787eaa
JM		 3919 if (ret != 0 && ret != -ENOENT) {
3920 btrfs_abort_transaction(trans, root, ret);
3921 goto err;
3922 }
e02119d5
CM		 3923
3924 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3925 dir, index);
6418c961
CM		 3926 if (ret == -ENOENT)
3927 ret = 0;
d4e3991b
ZB		 3928 else if (ret)
3929 btrfs_abort_transaction(trans, root, ret);
39279cc3
CM		 3930err:
3931 btrfs_free_path(path);
e02119d5
CM		 3932 if (ret)
3933 goto out;
3934
3935 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95
JB		 3936 inode_inc_iversion(inode);
3937 inode_inc_iversion(dir);
e02119d5 3938 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
b9959295 3939 ret = btrfs_update_inode(trans, root, dir);
e02119d5 3940out:
39279cc3
CM		 3941 return ret;
3942}
3943
92986796
AV		 3944int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3945 struct btrfs_root *root,
3946 struct inode *dir, struct inode *inode,
3947 const char *name, int name_len)
3948{
3949 int ret;
3950 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3951 if (!ret) {
8b558c5f 3952 drop_nlink(inode);
92986796
AV		 3953 ret = btrfs_update_inode(trans, root, inode);
3954 }
3955 return ret;
3956}
39279cc3 3957
a22285a6
YZ		 3958/*
3959 * helper to start transaction for unlink and rmdir.
3960 *
d52be818
JB		 3961 * unlink and rmdir are special in btrfs, they do not always free space, so
3962 * if we cannot make our reservations the normal way try and see if there is
3963 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3964 * allow the unlink to occur.
a22285a6 3965 */
d52be818 3966static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4df27c4d 3967{
39279cc3 3968 struct btrfs_trans_handle *trans;
a22285a6 3969 struct btrfs_root *root = BTRFS_I(dir)->root;
4df27c4d
YZ		 3970 int ret;
3971
e70bea5f
JB		 3972 /*
3973 * 1 for the possible orphan item
3974 * 1 for the dir item
3975 * 1 for the dir index
3976 * 1 for the inode ref
e70bea5f
JB		 3977 * 1 for the inode
3978 */
6e137ed3 3979 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 3980 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3981 return trans;
4df27c4d 3982
d52be818
JB		 3983 if (PTR_ERR(trans) == -ENOSPC) {
3984 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4df27c4d 3985
d52be818
JB		 3986 trans = btrfs_start_transaction(root, 0);
3987 if (IS_ERR(trans))
3988 return trans;
3989 ret = btrfs_cond_migrate_bytes(root->fs_info,
3990 &root->fs_info->trans_block_rsv,
3991 num_bytes, 5);
3992 if (ret) {
3993 btrfs_end_transaction(trans, root);
3994 return ERR_PTR(ret);
a22285a6 3995 }
5a77d76c 3996 trans->block_rsv = &root->fs_info->trans_block_rsv;
d52be818 3997 trans->bytes_reserved = num_bytes;
a22285a6 3998 }
d52be818 3999 return trans;
a22285a6
YZ		 4000}
4001
4002static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4003{
4004 struct btrfs_root *root = BTRFS_I(dir)->root;
4005 struct btrfs_trans_handle *trans;
4006 struct inode *inode = dentry->d_inode;
4007 int ret;
a22285a6 4008
d52be818 4009 trans = __unlink_start_trans(dir);
a22285a6
YZ		 4010 if (IS_ERR(trans))
4011 return PTR_ERR(trans);
5f39d397 4012
12fcfd22
CM		 4013 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
4014
e02119d5
CM		 4015 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
4016 dentry->d_name.name, dentry->d_name.len);
b532402e
TI		 4017 if (ret)
4018 goto out;
7b128766 4019
a22285a6 4020 if (inode->i_nlink == 0) {
7b128766 4021 ret = btrfs_orphan_add(trans, inode);
b532402e
TI		 4022 if (ret)
4023 goto out;
a22285a6 4024 }
7b128766 4025
b532402e 4026out:
d52be818 4027 btrfs_end_transaction(trans, root);
b53d3f5d 4028 btrfs_btree_balance_dirty(root);
39279cc3
CM		 4029 return ret;
4030}
4031
4df27c4d
YZ		 4032int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4033 struct btrfs_root *root,
4034 struct inode *dir, u64 objectid,
4035 const char *name, int name_len)
4036{
4037 struct btrfs_path *path;
4038 struct extent_buffer *leaf;
4039 struct btrfs_dir_item *di;
4040 struct btrfs_key key;
4041 u64 index;
4042 int ret;
33345d01 4043 u64 dir_ino = btrfs_ino(dir);
4df27c4d
YZ		 4044
4045 path = btrfs_alloc_path();
4046 if (!path)
4047 return -ENOMEM;
4048
33345d01 4049 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4df27c4d 4050 name, name_len, -1);
79787eaa
JM		 4051 if (IS_ERR_OR_NULL(di)) {
4052 if (!di)
4053 ret = -ENOENT;
4054 else
4055 ret = PTR_ERR(di);
4056 goto out;
4057 }
4df27c4d
YZ		 4058
4059 leaf = path->nodes[0];
4060 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4061 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4062 ret = btrfs_delete_one_dir_name(trans, root, path, di);
79787eaa
JM		 4063 if (ret) {
4064 btrfs_abort_transaction(trans, root, ret);
4065 goto out;
4066 }
b3b4aa74 4067 btrfs_release_path(path);
4df27c4d
YZ		 4068
4069 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4070 objectid, root->root_key.objectid,
33345d01 4071 dir_ino, &index, name, name_len);
4df27c4d 4072 if (ret < 0) {
79787eaa
JM		 4073 if (ret != -ENOENT) {
4074 btrfs_abort_transaction(trans, root, ret);
4075 goto out;
4076 }
33345d01 4077 di = btrfs_search_dir_index_item(root, path, dir_ino,
4df27c4d 4078 name, name_len);
79787eaa
JM		 4079 if (IS_ERR_OR_NULL(di)) {
4080 if (!di)
4081 ret = -ENOENT;
4082 else
4083 ret = PTR_ERR(di);
4084 btrfs_abort_transaction(trans, root, ret);
4085 goto out;
4086 }
4df27c4d
YZ		 4087
4088 leaf = path->nodes[0];
4089 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
b3b4aa74 4090 btrfs_release_path(path);
4df27c4d
YZ		 4091 index = key.offset;
4092 }
945d8962 4093 btrfs_release_path(path);
4df27c4d 4094
16cdcec7 4095 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 4096 if (ret) {
4097 btrfs_abort_transaction(trans, root, ret);
4098 goto out;
4099 }
4df27c4d
YZ		 4100
4101 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95 4102 inode_inc_iversion(dir);
4df27c4d 4103 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
5a24e84c 4104 ret = btrfs_update_inode_fallback(trans, root, dir);
79787eaa
JM		 4105 if (ret)
4106 btrfs_abort_transaction(trans, root, ret);
4107out:
71d7aed0 4108 btrfs_free_path(path);
79787eaa 4109 return ret;
4df27c4d
YZ		 4110}
4111
39279cc3
CM		 4112static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4113{
4114 struct inode *inode = dentry->d_inode;
1832a6d5 4115 int err = 0;
39279cc3 4116 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 4117 struct btrfs_trans_handle *trans;
39279cc3 4118
b3ae244e 4119 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
134d4512 4120 return -ENOTEMPTY;
b3ae244e
DS		 4121 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4122 return -EPERM;
134d4512 4123
d52be818 4124 trans = __unlink_start_trans(dir);
a22285a6 4125 if (IS_ERR(trans))
5df6a9f6 4126 return PTR_ERR(trans);
5df6a9f6 4127
33345d01 4128 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4df27c4d
YZ		 4129 err = btrfs_unlink_subvol(trans, root, dir,
4130 BTRFS_I(inode)->location.objectid,
4131 dentry->d_name.name,
4132 dentry->d_name.len);
4133 goto out;
4134 }
4135
7b128766
JB		 4136 err = btrfs_orphan_add(trans, inode);
4137 if (err)
4df27c4d 4138 goto out;
7b128766 4139
39279cc3 4140 /* now the directory is empty */
e02119d5
CM		 4141 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
4142 dentry->d_name.name, dentry->d_name.len);
d397712b 4143 if (!err)
dbe674a9 4144 btrfs_i_size_write(inode, 0);
4df27c4d 4145out:
d52be818 4146 btrfs_end_transaction(trans, root);
b53d3f5d 4147 btrfs_btree_balance_dirty(root);
3954401f 4148
39279cc3
CM		 4149 return err;
4150}
4151
39279cc3
CM		 4152/*
4153 * this can truncate away extent items, csum items and directory items.
4154 * It starts at a high offset and removes keys until it can't find
d352ac68 4155 * any higher than new_size
39279cc3
CM		 4156 *
4157 * csum items that cross the new i_size are truncated to the new size
4158 * as well.
7b128766
JB		 4159 *
4160 * min_type is the minimum key type to truncate down to. If set to 0, this
4161 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 4162 */
8082510e
YZ		 4163int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4164 struct btrfs_root *root,
4165 struct inode *inode,
4166 u64 new_size, u32 min_type)
39279cc3 4167{
39279cc3 4168 struct btrfs_path *path;
5f39d397 4169 struct extent_buffer *leaf;
39279cc3 4170 struct btrfs_file_extent_item *fi;
8082510e
YZ		 4171 struct btrfs_key key;
4172 struct btrfs_key found_key;
39279cc3 4173 u64 extent_start = 0;
db94535d 4174 u64 extent_num_bytes = 0;
5d4f98a2 4175 u64 extent_offset = 0;
39279cc3 4176 u64 item_end = 0;
7f4f6e0a 4177 u64 last_size = (u64)-1;
8082510e 4178 u32 found_type = (u8)-1;
39279cc3
CM		 4179 int found_extent;
4180 int del_item;
85e21bac
CM		 4181 int pending_del_nr = 0;
4182 int pending_del_slot = 0;
179e29e4 4183 int extent_type = -1;
8082510e
YZ		 4184 int ret;
4185 int err = 0;
33345d01 4186 u64 ino = btrfs_ino(inode);
8082510e
YZ		 4187
4188 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
39279cc3 4189
0eb0e19c
MF		 4190 path = btrfs_alloc_path();
4191 if (!path)
4192 return -ENOMEM;
4193 path->reada = -1;
4194
5dc562c5
JB		 4195 /*
4196 * We want to drop from the next block forward in case this new size is
4197 * not block aligned since we will be keeping the last block of the
4198 * extent just the way it is.
4199 */
27cdeb70
MX		 4200 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4201 root == root->fs_info->tree_root)
fda2832f
QW		 4202 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4203 root->sectorsize), (u64)-1, 0);
8082510e 4204
16cdcec7
MX		 4205 /*
4206 * This function is also used to drop the items in the log tree before
4207 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4208 * it is used to drop the loged items. So we shouldn't kill the delayed
4209 * items.
4210 */
4211 if (min_type == 0 && root == BTRFS_I(inode)->root)
4212 btrfs_kill_delayed_inode_items(inode);
4213
33345d01 4214 key.objectid = ino;
39279cc3 4215 key.offset = (u64)-1;
5f39d397
CM		 4216 key.type = (u8)-1;
4217
85e21bac 4218search_again:
b9473439 4219 path->leave_spinning = 1;
85e21bac 4220 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8082510e
YZ		 4221 if (ret < 0) {
4222 err = ret;
4223 goto out;
4224 }
d397712b 4225
85e21bac 4226 if (ret > 0) {
e02119d5
CM		 4227 /* there are no items in the tree for us to truncate, we're
4228 * done
4229 */
8082510e
YZ		 4230 if (path->slots[0] == 0)
4231 goto out;
85e21bac
CM		 4232 path->slots[0]--;
4233 }
4234
d397712b 4235 while (1) {
39279cc3 4236 fi = NULL;
5f39d397
CM		 4237 leaf = path->nodes[0];
4238 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
962a298f 4239 found_type = found_key.type;
39279cc3 4240
33345d01 4241 if (found_key.objectid != ino)
39279cc3 4242 break;
5f39d397 4243
85e21bac 4244 if (found_type < min_type)
39279cc3
CM		 4245 break;
4246
5f39d397 4247 item_end = found_key.offset;
39279cc3 4248 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 4249 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 4250 struct btrfs_file_extent_item);
179e29e4
CM		 4251 extent_type = btrfs_file_extent_type(leaf, fi);
4252 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 4253 item_end +=
db94535d 4254 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 4255 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 4256 item_end += btrfs_file_extent_inline_len(leaf,
514ac8ad 4257 path->slots[0], fi);
39279cc3 4258 }
008630c1 4259 item_end--;
39279cc3 4260 }
8082510e
YZ		 4261 if (found_type > min_type) {
4262 del_item = 1;
4263 } else {
4264 if (item_end < new_size)
b888db2b 4265 break;
8082510e
YZ		 4266 if (found_key.offset >= new_size)
4267 del_item = 1;
4268 else
4269 del_item = 0;
39279cc3 4270 }
39279cc3 4271 found_extent = 0;
39279cc3 4272 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 4273 if (found_type != BTRFS_EXTENT_DATA_KEY)
4274 goto delete;
4275
7f4f6e0a
JB		 4276 if (del_item)
4277 last_size = found_key.offset;
4278 else
4279 last_size = new_size;
4280
179e29e4 4281 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 4282 u64 num_dec;
db94535d 4283 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
f70a9a6b 4284 if (!del_item) {
db94535d
CM		 4285 u64 orig_num_bytes =
4286 btrfs_file_extent_num_bytes(leaf, fi);
fda2832f
QW		 4287 extent_num_bytes = ALIGN(new_size -
4288 found_key.offset,
4289 root->sectorsize);
db94535d
CM		 4290 btrfs_set_file_extent_num_bytes(leaf, fi,
4291 extent_num_bytes);
4292 num_dec = (orig_num_bytes -
9069218d 4293 extent_num_bytes);
27cdeb70
MX		 4294 if (test_bit(BTRFS_ROOT_REF_COWS,
4295 &root->state) &&
4296 extent_start != 0)
a76a3cd4 4297 inode_sub_bytes(inode, num_dec);
5f39d397 4298 btrfs_mark_buffer_dirty(leaf);
39279cc3 4299 } else {
db94535d
CM		 4300 extent_num_bytes =
4301 btrfs_file_extent_disk_num_bytes(leaf,
4302 fi);
5d4f98a2
YZ		 4303 extent_offset = found_key.offset -
4304 btrfs_file_extent_offset(leaf, fi);
4305
39279cc3 4306 /* FIXME blocksize != 4096 */
9069218d 4307 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 4308 if (extent_start != 0) {
4309 found_extent = 1;
27cdeb70
MX		 4310 if (test_bit(BTRFS_ROOT_REF_COWS,
4311 &root->state))
a76a3cd4 4312 inode_sub_bytes(inode, num_dec);
e02119d5 4313 }
39279cc3 4314 }
9069218d 4315 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 4316 /*
4317 * we can't truncate inline items that have had
4318 * special encodings
4319 */
4320 if (!del_item &&
4321 btrfs_file_extent_compression(leaf, fi) == 0 &&
4322 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4323 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM		 4324 u32 size = new_size - found_key.offset;
4325
27cdeb70 4326 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
a76a3cd4
YZ		 4327 inode_sub_bytes(inode, item_end + 1 -
4328 new_size);
514ac8ad
CM		 4329
4330 /*
4331 * update the ram bytes to properly reflect
4332 * the new size of our item
4333 */
4334 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
e02119d5
CM		 4335 size =
4336 btrfs_file_extent_calc_inline_size(size);
afe5fea7 4337 btrfs_truncate_item(root, path, size, 1);
27cdeb70
MX		 4338 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4339 &root->state)) {
a76a3cd4
YZ		 4340 inode_sub_bytes(inode, item_end + 1 -
4341 found_key.offset);
9069218d 4342 }
39279cc3 4343 }
179e29e4 4344delete:
39279cc3 4345 if (del_item) {
85e21bac
CM		 4346 if (!pending_del_nr) {
4347 /* no pending yet, add ourselves */
4348 pending_del_slot = path->slots[0];
4349 pending_del_nr = 1;
4350 } else if (pending_del_nr &&
4351 path->slots[0] + 1 == pending_del_slot) {
4352 /* hop on the pending chunk */
4353 pending_del_nr++;
4354 pending_del_slot = path->slots[0];
4355 } else {
d397712b 4356 BUG();
85e21bac 4357 }
39279cc3
CM		 4358 } else {
4359 break;
4360 }
27cdeb70
MX		 4361 if (found_extent &&
4362 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4363 root == root->fs_info->tree_root)) {
b9473439 4364 btrfs_set_path_blocking(path);
39279cc3 4365 ret = btrfs_free_extent(trans, root, extent_start,
5d4f98a2
YZ		 4366 extent_num_bytes, 0,
4367 btrfs_header_owner(leaf),
66d7e7f0 4368 ino, extent_offset, 0);
39279cc3
CM		 4369 BUG_ON(ret);
4370 }
85e21bac 4371
8082510e
YZ		 4372 if (found_type == BTRFS_INODE_ITEM_KEY)
4373 break;
4374
4375 if (path->slots[0] == 0 ||
4376 path->slots[0] != pending_del_slot) {
8082510e
YZ		 4377 if (pending_del_nr) {
4378 ret = btrfs_del_items(trans, root, path,
4379 pending_del_slot,
4380 pending_del_nr);
79787eaa
JM		 4381 if (ret) {
4382 btrfs_abort_transaction(trans,
4383 root, ret);
4384 goto error;
4385 }
8082510e
YZ		 4386 pending_del_nr = 0;
4387 }
b3b4aa74 4388 btrfs_release_path(path);
85e21bac 4389 goto search_again;
8082510e
YZ		 4390 } else {
4391 path->slots[0]--;
85e21bac 4392 }
39279cc3 4393 }
8082510e 4394out:
85e21bac
CM		 4395 if (pending_del_nr) {
4396 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4397 pending_del_nr);
79787eaa
JM		 4398 if (ret)
4399 btrfs_abort_transaction(trans, root, ret);
85e21bac 4400 }
79787eaa 4401error:
dac5705c
FM		 4402 if (last_size != (u64)-1 &&
4403 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
7f4f6e0a 4404 btrfs_ordered_update_i_size(inode, last_size, NULL);
39279cc3 4405 btrfs_free_path(path);
8082510e 4406 return err;
39279cc3
CM		 4407}
4408
4409/*
2aaa6655
JB		 4410 * btrfs_truncate_page - read, zero a chunk and write a page
4411 * @inode - inode that we're zeroing
4412 * @from - the offset to start zeroing
4413 * @len - the length to zero, 0 to zero the entire range respective to the
4414 * offset
4415 * @front - zero up to the offset instead of from the offset on
4416 *
4417 * This will find the page for the "from" offset and cow the page and zero the
4418 * part we want to zero. This is used with truncate and hole punching.
39279cc3 4419 */
2aaa6655
JB		 4420int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4421 int front)
39279cc3 4422{
2aaa6655 4423 struct address_space *mapping = inode->i_mapping;
db94535d 4424 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 4425 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4426 struct btrfs_ordered_extent *ordered;
2ac55d41 4427 struct extent_state *cached_state = NULL;
e6dcd2dc 4428 char *kaddr;
db94535d 4429 u32 blocksize = root->sectorsize;
39279cc3
CM		 4430 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4431 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4432 struct page *page;
3b16a4e3 4433 gfp_t mask = btrfs_alloc_write_mask(mapping);
39279cc3 4434 int ret = 0;
a52d9a80 4435 u64 page_start;
e6dcd2dc 4436 u64 page_end;
39279cc3 4437
2aaa6655
JB		 4438 if ((offset & (blocksize - 1)) == 0 &&
4439 (!len || ((len & (blocksize - 1)) == 0)))
39279cc3 4440 goto out;
0ca1f7ce 4441 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
5d5e103a
JB		 4442 if (ret)
4443 goto out;
39279cc3 4444
211c17f5 4445again:
3b16a4e3 4446 page = find_or_create_page(mapping, index, mask);
5d5e103a 4447 if (!page) {
0ca1f7ce 4448 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
ac6a2b36 4449 ret = -ENOMEM;
39279cc3 4450 goto out;
5d5e103a 4451 }
e6dcd2dc
CM		 4452
4453 page_start = page_offset(page);
4454 page_end = page_start + PAGE_CACHE_SIZE - 1;
4455
39279cc3 4456 if (!PageUptodate(page)) {
9ebefb18 4457 ret = btrfs_readpage(NULL, page);
39279cc3 4458 lock_page(page);
211c17f5
CM		 4459 if (page->mapping != mapping) {
4460 unlock_page(page);
4461 page_cache_release(page);
4462 goto again;
4463 }
39279cc3
CM		 4464 if (!PageUptodate(page)) {
4465 ret = -EIO;
89642229 4466 goto out_unlock;
39279cc3
CM		 4467 }
4468 }
211c17f5 4469 wait_on_page_writeback(page);
e6dcd2dc 4470
d0082371 4471 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 4472 set_page_extent_mapped(page);
4473
4474 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4475 if (ordered) {
2ac55d41
JB		 4476 unlock_extent_cached(io_tree, page_start, page_end,
4477 &cached_state, GFP_NOFS);
e6dcd2dc
CM		 4478 unlock_page(page);
4479 page_cache_release(page);
eb84ae03 4480 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 4481 btrfs_put_ordered_extent(ordered);
4482 goto again;
4483 }
4484
2ac55d41 4485 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 4486 EXTENT_DIRTY | EXTENT_DELALLOC |
4487 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 4488 0, 0, &cached_state, GFP_NOFS);
5d5e103a 4489
2ac55d41
JB		 4490 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4491 &cached_state);
9ed74f2d 4492 if (ret) {
2ac55d41
JB		 4493 unlock_extent_cached(io_tree, page_start, page_end,
4494 &cached_state, GFP_NOFS);
9ed74f2d
JB		 4495 goto out_unlock;
4496 }
4497
e6dcd2dc 4498 if (offset != PAGE_CACHE_SIZE) {
2aaa6655
JB		 4499 if (!len)
4500 len = PAGE_CACHE_SIZE - offset;
e6dcd2dc 4501 kaddr = kmap(page);
2aaa6655
JB		 4502 if (front)
4503 memset(kaddr, 0, offset);
4504 else
4505 memset(kaddr + offset, 0, len);
e6dcd2dc
CM		 4506 flush_dcache_page(page);
4507 kunmap(page);
4508 }
247e743c 4509 ClearPageChecked(page);
e6dcd2dc 4510 set_page_dirty(page);
2ac55d41
JB		 4511 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4512 GFP_NOFS);
39279cc3 4513
89642229 4514out_unlock:
5d5e103a 4515 if (ret)
0ca1f7ce 4516 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
39279cc3
CM		 4517 unlock_page(page);
4518 page_cache_release(page);
4519out:
4520 return ret;
4521}
4522
16e7549f
JB		 4523static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4524 u64 offset, u64 len)
4525{
4526 struct btrfs_trans_handle *trans;
4527 int ret;
4528
4529 /*
4530 * Still need to make sure the inode looks like it's been updated so
4531 * that any holes get logged if we fsync.
4532 */
4533 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4534 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4535 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4536 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4537 return 0;
4538 }
4539
4540 /*
4541 * 1 - for the one we're dropping
4542 * 1 - for the one we're adding
4543 * 1 - for updating the inode.
4544 */
4545 trans = btrfs_start_transaction(root, 3);
4546 if (IS_ERR(trans))
4547 return PTR_ERR(trans);
4548
4549 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4550 if (ret) {
4551 btrfs_abort_transaction(trans, root, ret);
4552 btrfs_end_transaction(trans, root);
4553 return ret;
4554 }
4555
4556 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4557 0, 0, len, 0, len, 0, 0, 0);
4558 if (ret)
4559 btrfs_abort_transaction(trans, root, ret);
4560 else
4561 btrfs_update_inode(trans, root, inode);
4562 btrfs_end_transaction(trans, root);
4563 return ret;
4564}
4565
695a0d0d
JB		 4566/*
4567 * This function puts in dummy file extents for the area we're creating a hole
4568 * for. So if we are truncating this file to a larger size we need to insert
4569 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4570 * the range between oldsize and size
4571 */
a41ad394 4572int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
39279cc3 4573{
9036c102
YZ		 4574 struct btrfs_root *root = BTRFS_I(inode)->root;
4575 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a22285a6 4576 struct extent_map *em = NULL;
2ac55d41 4577 struct extent_state *cached_state = NULL;
5dc562c5 4578 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
fda2832f
QW		 4579 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4580 u64 block_end = ALIGN(size, root->sectorsize);
9036c102
YZ		 4581 u64 last_byte;
4582 u64 cur_offset;
4583 u64 hole_size;
9ed74f2d 4584 int err = 0;
39279cc3 4585
a71754fc
JB		 4586 /*
4587 * If our size started in the middle of a page we need to zero out the
4588 * rest of the page before we expand the i_size, otherwise we could
4589 * expose stale data.
4590 */
4591 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4592 if (err)
4593 return err;
4594
9036c102
YZ		 4595 if (size <= hole_start)
4596 return 0;
4597
9036c102
YZ		 4598 while (1) {
4599 struct btrfs_ordered_extent *ordered;
fa7c1494 4600
2ac55d41 4601 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
d0082371 4602 &cached_state);
fa7c1494
MX		 4603 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4604 block_end - hole_start);
9036c102
YZ		 4605 if (!ordered)
4606 break;
2ac55d41
JB		 4607 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4608 &cached_state, GFP_NOFS);
fa7c1494 4609 btrfs_start_ordered_extent(inode, ordered, 1);
9036c102
YZ		 4610 btrfs_put_ordered_extent(ordered);
4611 }
39279cc3 4612
9036c102
YZ		 4613 cur_offset = hole_start;
4614 while (1) {
4615 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4616 block_end - cur_offset, 0);
79787eaa
JM		 4617 if (IS_ERR(em)) {
4618 err = PTR_ERR(em);
f2767956 4619 em = NULL;
79787eaa
JM		 4620 break;
4621 }
9036c102 4622 last_byte = min(extent_map_end(em), block_end);
fda2832f 4623 last_byte = ALIGN(last_byte , root->sectorsize);
8082510e 4624 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5dc562c5 4625 struct extent_map *hole_em;
9036c102 4626 hole_size = last_byte - cur_offset;
9ed74f2d 4627
16e7549f
JB		 4628 err = maybe_insert_hole(root, inode, cur_offset,
4629 hole_size);
4630 if (err)
3893e33b 4631 break;
5dc562c5
JB		 4632 btrfs_drop_extent_cache(inode, cur_offset,
4633 cur_offset + hole_size - 1, 0);
4634 hole_em = alloc_extent_map();
4635 if (!hole_em) {
4636 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4637 &BTRFS_I(inode)->runtime_flags);
4638 goto next;
4639 }
4640 hole_em->start = cur_offset;
4641 hole_em->len = hole_size;
4642 hole_em->orig_start = cur_offset;
8082510e 4643
5dc562c5
JB		 4644 hole_em->block_start = EXTENT_MAP_HOLE;
4645 hole_em->block_len = 0;
b4939680 4646 hole_em->orig_block_len = 0;
cc95bef6 4647 hole_em->ram_bytes = hole_size;
5dc562c5
JB		 4648 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4649 hole_em->compress_type = BTRFS_COMPRESS_NONE;
16e7549f 4650 hole_em->generation = root->fs_info->generation;
8082510e 4651
5dc562c5
JB		 4652 while (1) {
4653 write_lock(&em_tree->lock);
09a2a8f9 4654 err = add_extent_mapping(em_tree, hole_em, 1);
5dc562c5
JB		 4655 write_unlock(&em_tree->lock);
4656 if (err != -EEXIST)
4657 break;
4658 btrfs_drop_extent_cache(inode, cur_offset,
4659 cur_offset +
4660 hole_size - 1, 0);
4661 }
4662 free_extent_map(hole_em);
9036c102 4663 }
16e7549f 4664next:
9036c102 4665 free_extent_map(em);
a22285a6 4666 em = NULL;
9036c102 4667 cur_offset = last_byte;
8082510e 4668 if (cur_offset >= block_end)
9036c102
YZ		 4669 break;
4670 }
a22285a6 4671 free_extent_map(em);
2ac55d41
JB		 4672 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4673 GFP_NOFS);
9036c102
YZ		 4674 return err;
4675}
39279cc3 4676
9ea24bbe
FM		 4677static int wait_snapshoting_atomic_t(atomic_t *a)
4678{
4679 schedule();
4680 return 0;
4681}
4682
4683static void wait_for_snapshot_creation(struct btrfs_root *root)
4684{
4685 while (true) {
4686 int ret;
4687
4688 ret = btrfs_start_write_no_snapshoting(root);
4689 if (ret)
4690 break;
4691 wait_on_atomic_t(&root->will_be_snapshoted,
4692 wait_snapshoting_atomic_t,
4693 TASK_UNINTERRUPTIBLE);
4694 }
4695}
4696
3972f260 4697static int btrfs_setsize(struct inode *inode, struct iattr *attr)
8082510e 4698{
f4a2f4c5
MX		 4699 struct btrfs_root *root = BTRFS_I(inode)->root;
4700 struct btrfs_trans_handle *trans;
a41ad394 4701 loff_t oldsize = i_size_read(inode);
3972f260
ES		 4702 loff_t newsize = attr->ia_size;
4703 int mask = attr->ia_valid;
8082510e
YZ		 4704 int ret;
4705
3972f260
ES		 4706 /*
4707 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4708 * special case where we need to update the times despite not having
4709 * these flags set. For all other operations the VFS set these flags
4710 * explicitly if it wants a timestamp update.
4711 */
dff6efc3
CH		 4712 if (newsize != oldsize) {
4713 inode_inc_iversion(inode);
4714 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4715 inode->i_ctime = inode->i_mtime =
4716 current_fs_time(inode->i_sb);
4717 }
3972f260 4718
a41ad394 4719 if (newsize > oldsize) {
7caef267 4720 truncate_pagecache(inode, newsize);
9ea24bbe
FM		 4721 /*
4722 * Don't do an expanding truncate while snapshoting is ongoing.
4723 * This is to ensure the snapshot captures a fully consistent
4724 * state of this file - if the snapshot captures this expanding
4725 * truncation, it must capture all writes that happened before
4726 * this truncation.
4727 */
4728 wait_for_snapshot_creation(root);
a41ad394 4729 ret = btrfs_cont_expand(inode, oldsize, newsize);
9ea24bbe
FM		 4730 if (ret) {
4731 btrfs_end_write_no_snapshoting(root);
8082510e 4732 return ret;
9ea24bbe 4733 }
8082510e 4734
f4a2f4c5 4735 trans = btrfs_start_transaction(root, 1);
9ea24bbe
FM		 4736 if (IS_ERR(trans)) {
4737 btrfs_end_write_no_snapshoting(root);
f4a2f4c5 4738 return PTR_ERR(trans);
9ea24bbe 4739 }
f4a2f4c5
MX		 4740
4741 i_size_write(inode, newsize);
4742 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4743 ret = btrfs_update_inode(trans, root, inode);
9ea24bbe 4744 btrfs_end_write_no_snapshoting(root);
7ad85bb7 4745 btrfs_end_transaction(trans, root);
a41ad394 4746 } else {
8082510e 4747
a41ad394
JB		 4748 /*
4749 * We're truncating a file that used to have good data down to
4750 * zero. Make sure it gets into the ordered flush list so that
4751 * any new writes get down to disk quickly.
4752 */
4753 if (newsize == 0)
72ac3c0d
JB		 4754 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4755 &BTRFS_I(inode)->runtime_flags);
8082510e 4756
f3fe820c
JB		 4757 /*
4758 * 1 for the orphan item we're going to add
4759 * 1 for the orphan item deletion.
4760 */
4761 trans = btrfs_start_transaction(root, 2);
4762 if (IS_ERR(trans))
4763 return PTR_ERR(trans);
4764
4765 /*
4766 * We need to do this in case we fail at _any_ point during the
4767 * actual truncate. Once we do the truncate_setsize we could
4768 * invalidate pages which forces any outstanding ordered io to
4769 * be instantly completed which will give us extents that need
4770 * to be truncated. If we fail to get an orphan inode down we
4771 * could have left over extents that were never meant to live,
4772 * so we need to garuntee from this point on that everything
4773 * will be consistent.
4774 */
4775 ret = btrfs_orphan_add(trans, inode);
4776 btrfs_end_transaction(trans, root);
4777 if (ret)
4778 return ret;
4779
a41ad394
JB		 4780 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4781 truncate_setsize(inode, newsize);
2e60a51e
MX		 4782
4783 /* Disable nonlocked read DIO to avoid the end less truncate */
4784 btrfs_inode_block_unlocked_dio(inode);
4785 inode_dio_wait(inode);
4786 btrfs_inode_resume_unlocked_dio(inode);
4787
a41ad394 4788 ret = btrfs_truncate(inode);
7f4f6e0a
JB		 4789 if (ret && inode->i_nlink) {
4790 int err;
4791
4792 /*
4793 * failed to truncate, disk_i_size is only adjusted down
4794 * as we remove extents, so it should represent the true
4795 * size of the inode, so reset the in memory size and
4796 * delete our orphan entry.
4797 */
4798 trans = btrfs_join_transaction(root);
4799 if (IS_ERR(trans)) {
4800 btrfs_orphan_del(NULL, inode);
4801 return ret;
4802 }
4803 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4804 err = btrfs_orphan_del(trans, inode);
4805 if (err)
4806 btrfs_abort_transaction(trans, root, err);
4807 btrfs_end_transaction(trans, root);
4808 }
8082510e
YZ		 4809 }
4810
a41ad394 4811 return ret;
8082510e
YZ		 4812}
4813
9036c102
YZ		 4814static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4815{
4816 struct inode *inode = dentry->d_inode;
b83cc969 4817 struct btrfs_root *root = BTRFS_I(inode)->root;
9036c102 4818 int err;
39279cc3 4819
b83cc969
LZ		 4820 if (btrfs_root_readonly(root))
4821 return -EROFS;
4822
9036c102
YZ		 4823 err = inode_change_ok(inode, attr);
4824 if (err)
4825 return err;
2bf5a725 4826
5a3f23d5 4827 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3972f260 4828 err = btrfs_setsize(inode, attr);
8082510e
YZ		 4829 if (err)
4830 return err;
39279cc3 4831 }
9036c102 4832
1025774c
CH		 4833 if (attr->ia_valid) {
4834 setattr_copy(inode, attr);
0c4d2d95 4835 inode_inc_iversion(inode);
22c44fe6 4836 err = btrfs_dirty_inode(inode);
1025774c 4837
22c44fe6 4838 if (!err && attr->ia_valid & ATTR_MODE)
996a710d 4839 err = posix_acl_chmod(inode, inode->i_mode);
1025774c 4840 }
33268eaf 4841
39279cc3
CM		 4842 return err;
4843}
61295eb8 4844
131e404a
FDBM		 4845/*
4846 * While truncating the inode pages during eviction, we get the VFS calling
4847 * btrfs_invalidatepage() against each page of the inode. This is slow because
4848 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4849 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4850 * extent_state structures over and over, wasting lots of time.
4851 *
4852 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4853 * those expensive operations on a per page basis and do only the ordered io
4854 * finishing, while we release here the extent_map and extent_state structures,
4855 * without the excessive merging and splitting.
4856 */
4857static void evict_inode_truncate_pages(struct inode *inode)
4858{
4859 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4860 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4861 struct rb_node *node;
4862
4863 ASSERT(inode->i_state & I_FREEING);
91b0abe3 4864 truncate_inode_pages_final(&inode->i_data);
131e404a
FDBM		 4865
4866 write_lock(&map_tree->lock);
4867 while (!RB_EMPTY_ROOT(&map_tree->map)) {
4868 struct extent_map *em;
4869
4870 node = rb_first(&map_tree->map);
4871 em = rb_entry(node, struct extent_map, rb_node);
180589ef
WS		 4872 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4873 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
131e404a
FDBM		 4874 remove_extent_mapping(map_tree, em);
4875 free_extent_map(em);
7064dd5c
FM		 4876 if (need_resched()) {
4877 write_unlock(&map_tree->lock);
4878 cond_resched();
4879 write_lock(&map_tree->lock);
4880 }
131e404a
FDBM		 4881 }
4882 write_unlock(&map_tree->lock);
4883
4884 spin_lock(&io_tree->lock);
4885 while (!RB_EMPTY_ROOT(&io_tree->state)) {
4886 struct extent_state *state;
4887 struct extent_state *cached_state = NULL;
4888
4889 node = rb_first(&io_tree->state);
4890 state = rb_entry(node, struct extent_state, rb_node);
4891 atomic_inc(&state->refs);
4892 spin_unlock(&io_tree->lock);
4893
4894 lock_extent_bits(io_tree, state->start, state->end,
4895 0, &cached_state);
4896 clear_extent_bit(io_tree, state->start, state->end,
4897 EXTENT_LOCKED | EXTENT_DIRTY |
4898 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
4899 EXTENT_DEFRAG, 1, 1,
4900 &cached_state, GFP_NOFS);
4901 free_extent_state(state);
4902
7064dd5c 4903 cond_resched();
131e404a
FDBM		 4904 spin_lock(&io_tree->lock);
4905 }
4906 spin_unlock(&io_tree->lock);
4907}
4908
bd555975 4909void btrfs_evict_inode(struct inode *inode)
39279cc3
CM		 4910{
4911 struct btrfs_trans_handle *trans;
4912 struct btrfs_root *root = BTRFS_I(inode)->root;
726c35fa 4913 struct btrfs_block_rsv *rsv, *global_rsv;
07127184 4914 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3
CM		 4915 int ret;
4916
1abe9b8a 4917 trace_btrfs_inode_evict(inode);
4918
131e404a
FDBM		 4919 evict_inode_truncate_pages(inode);
4920
69e9c6c6
SB		 4921 if (inode->i_nlink &&
4922 ((btrfs_root_refs(&root->root_item) != 0 &&
4923 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4924 btrfs_is_free_space_inode(inode)))
bd555975
AV		 4925 goto no_delete;
4926
39279cc3 4927 if (is_bad_inode(inode)) {
7b128766 4928 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 4929 goto no_delete;
4930 }
bd555975 4931 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4a096752 4932 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 4933
f612496b
MX		 4934 btrfs_free_io_failure_record(inode, 0, (u64)-1);
4935
c71bf099 4936 if (root->fs_info->log_root_recovering) {
6bf02314 4937 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8a35d95f 4938 &BTRFS_I(inode)->runtime_flags));
c71bf099
YZ		 4939 goto no_delete;
4940 }
4941
76dda93c 4942 if (inode->i_nlink > 0) {
69e9c6c6
SB		 4943 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4944 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
76dda93c
YZ		 4945 goto no_delete;
4946 }
4947
0e8c36a9
MX		 4948 ret = btrfs_commit_inode_delayed_inode(inode);
4949 if (ret) {
4950 btrfs_orphan_del(NULL, inode);
4951 goto no_delete;
4952 }
4953
66d8f3dd 4954 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4289a667
JB		 4955 if (!rsv) {
4956 btrfs_orphan_del(NULL, inode);
4957 goto no_delete;
4958 }
4a338542 4959 rsv->size = min_size;
ca7e70f5 4960 rsv->failfast = 1;
726c35fa 4961 global_rsv = &root->fs_info->global_block_rsv;
4289a667 4962
dbe674a9 4963 btrfs_i_size_write(inode, 0);
5f39d397 4964
4289a667 4965 /*
8407aa46
MX		 4966 * This is a bit simpler than btrfs_truncate since we've already
4967 * reserved our space for our orphan item in the unlink, so we just
4968 * need to reserve some slack space in case we add bytes and update
4969 * inode item when doing the truncate.
4289a667 4970 */
8082510e 4971 while (1) {
08e007d2
MX		 4972 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4973 BTRFS_RESERVE_FLUSH_LIMIT);
726c35fa
JB		 4974
4975 /*
4976 * Try and steal from the global reserve since we will
4977 * likely not use this space anyway, we want to try as
4978 * hard as possible to get this to work.
4979 */
4980 if (ret)
4981 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
d68fc57b 4982
d68fc57b 4983 if (ret) {
c2cf52eb
SK		 4984 btrfs_warn(root->fs_info,
4985 "Could not get space for a delete, will truncate on mount %d",
4986 ret);
4289a667
JB		 4987 btrfs_orphan_del(NULL, inode);
4988 btrfs_free_block_rsv(root, rsv);
4989 goto no_delete;
d68fc57b 4990 }
7b128766 4991
0e8c36a9 4992 trans = btrfs_join_transaction(root);
4289a667
JB		 4993 if (IS_ERR(trans)) {
4994 btrfs_orphan_del(NULL, inode);
4995 btrfs_free_block_rsv(root, rsv);
4996 goto no_delete;
d68fc57b 4997 }
7b128766 4998
4289a667
JB		 4999 trans->block_rsv = rsv;
5000
d68fc57b 5001 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
ca7e70f5 5002 if (ret != -ENOSPC)
8082510e 5003 break;
85e21bac 5004
8407aa46 5005 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e
YZ		 5006 btrfs_end_transaction(trans, root);
5007 trans = NULL;
b53d3f5d 5008 btrfs_btree_balance_dirty(root);
8082510e 5009 }
5f39d397 5010
4289a667
JB		 5011 btrfs_free_block_rsv(root, rsv);
5012
4ef31a45
JB		 5013 /*
5014 * Errors here aren't a big deal, it just means we leave orphan items
5015 * in the tree. They will be cleaned up on the next mount.
5016 */
8082510e 5017 if (ret == 0) {
4289a667 5018 trans->block_rsv = root->orphan_block_rsv;
4ef31a45
JB		 5019 btrfs_orphan_del(trans, inode);
5020 } else {
5021 btrfs_orphan_del(NULL, inode);
8082510e 5022 }
54aa1f4d 5023
4289a667 5024 trans->block_rsv = &root->fs_info->trans_block_rsv;
581bb050
LZ		 5025 if (!(root == root->fs_info->tree_root ||
5026 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
33345d01 5027 btrfs_return_ino(root, btrfs_ino(inode));
581bb050 5028
54aa1f4d 5029 btrfs_end_transaction(trans, root);
b53d3f5d 5030 btrfs_btree_balance_dirty(root);
39279cc3 5031no_delete:
89042e5a 5032 btrfs_remove_delayed_node(inode);
dbd5768f 5033 clear_inode(inode);
8082510e 5034 return;
39279cc3
CM		 5035}
5036
5037/*
5038 * this returns the key found in the dir entry in the location pointer.
5039 * If no dir entries were found, location->objectid is 0.
5040 */
5041static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5042 struct btrfs_key *location)
5043{
5044 const char *name = dentry->d_name.name;
5045 int namelen = dentry->d_name.len;
5046 struct btrfs_dir_item *di;
5047 struct btrfs_path *path;
5048 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 5049 int ret = 0;
39279cc3
CM		 5050
5051 path = btrfs_alloc_path();
d8926bb3
MF		 5052 if (!path)
5053 return -ENOMEM;
3954401f 5054
33345d01 5055 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
39279cc3 5056 namelen, 0);
0d9f7f3e
Y		 5057 if (IS_ERR(di))
5058 ret = PTR_ERR(di);
d397712b 5059
c704005d 5060 if (IS_ERR_OR_NULL(di))
3954401f 5061 goto out_err;
d397712b 5062
5f39d397 5063 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 5064out:
39279cc3
CM		 5065 btrfs_free_path(path);
5066 return ret;
3954401f
CM		 5067out_err:
5068 location->objectid = 0;
5069 goto out;
39279cc3
CM		 5070}
5071
5072/*
5073 * when we hit a tree root in a directory, the btrfs part of the inode
5074 * needs to be changed to reflect the root directory of the tree root. This
5075 * is kind of like crossing a mount point.
5076 */
5077static int fixup_tree_root_location(struct btrfs_root *root,
4df27c4d
YZ		 5078 struct inode *dir,
5079 struct dentry *dentry,
5080 struct btrfs_key *location,
5081 struct btrfs_root **sub_root)
39279cc3 5082{
4df27c4d
YZ		 5083 struct btrfs_path *path;
5084 struct btrfs_root *new_root;
5085 struct btrfs_root_ref *ref;
5086 struct extent_buffer *leaf;
1d4c08e0 5087 struct btrfs_key key;
4df27c4d
YZ		 5088 int ret;
5089 int err = 0;
39279cc3 5090
4df27c4d
YZ		 5091 path = btrfs_alloc_path();
5092 if (!path) {
5093 err = -ENOMEM;
5094 goto out;
5095 }
39279cc3 5096
4df27c4d 5097 err = -ENOENT;
1d4c08e0
DS		 5098 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5099 key.type = BTRFS_ROOT_REF_KEY;
5100 key.offset = location->objectid;
5101
5102 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5103 0, 0);
4df27c4d
YZ		 5104 if (ret) {
5105 if (ret < 0)
5106 err = ret;
5107 goto out;
5108 }
39279cc3 5109
4df27c4d
YZ		 5110 leaf = path->nodes[0];
5111 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
33345d01 5112 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4df27c4d
YZ		 5113 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5114 goto out;
39279cc3 5115
4df27c4d
YZ		 5116 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5117 (unsigned long)(ref + 1),
5118 dentry->d_name.len);
5119 if (ret)
5120 goto out;
5121
b3b4aa74 5122 btrfs_release_path(path);
4df27c4d
YZ		 5123
5124 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5125 if (IS_ERR(new_root)) {
5126 err = PTR_ERR(new_root);
5127 goto out;
5128 }
5129
4df27c4d
YZ		 5130 *sub_root = new_root;
5131 location->objectid = btrfs_root_dirid(&new_root->root_item);
5132 location->type = BTRFS_INODE_ITEM_KEY;
5133 location->offset = 0;
5134 err = 0;
5135out:
5136 btrfs_free_path(path);
5137 return err;
39279cc3
CM		 5138}
5139
5d4f98a2
YZ		 5140static void inode_tree_add(struct inode *inode)
5141{
5142 struct btrfs_root *root = BTRFS_I(inode)->root;
5143 struct btrfs_inode *entry;
03e860bd
FNP		 5144 struct rb_node **p;
5145 struct rb_node *parent;
cef21937 5146 struct rb_node *new = &BTRFS_I(inode)->rb_node;
33345d01 5147 u64 ino = btrfs_ino(inode);
5d4f98a2 5148
1d3382cb 5149 if (inode_unhashed(inode))
76dda93c 5150 return;
e1409cef 5151 parent = NULL;
5d4f98a2 5152 spin_lock(&root->inode_lock);
e1409cef 5153 p = &root->inode_tree.rb_node;
5d4f98a2
YZ		 5154 while (*p) {
5155 parent = *p;
5156 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5157
33345d01 5158 if (ino < btrfs_ino(&entry->vfs_inode))
03e860bd 5159 p = &parent->rb_left;
33345d01 5160 else if (ino > btrfs_ino(&entry->vfs_inode))
03e860bd 5161 p = &parent->rb_right;
5d4f98a2
YZ		 5162 else {
5163 WARN_ON(!(entry->vfs_inode.i_state &
a4ffdde6 5164 (I_WILL_FREE | I_FREEING)));
cef21937 5165 rb_replace_node(parent, new, &root->inode_tree);
03e860bd
FNP		 5166 RB_CLEAR_NODE(parent);
5167 spin_unlock(&root->inode_lock);
cef21937 5168 return;
5d4f98a2
YZ		 5169 }
5170 }
cef21937
FDBM		 5171 rb_link_node(new, parent, p);
5172 rb_insert_color(new, &root->inode_tree);
5d4f98a2
YZ		 5173 spin_unlock(&root->inode_lock);
5174}
5175
5176static void inode_tree_del(struct inode *inode)
5177{
5178 struct btrfs_root *root = BTRFS_I(inode)->root;
76dda93c 5179 int empty = 0;
5d4f98a2 5180
03e860bd 5181 spin_lock(&root->inode_lock);
5d4f98a2 5182 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5d4f98a2 5183 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5d4f98a2 5184 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
76dda93c 5185 empty = RB_EMPTY_ROOT(&root->inode_tree);
5d4f98a2 5186 }
03e860bd 5187 spin_unlock(&root->inode_lock);
76dda93c 5188
69e9c6c6 5189 if (empty && btrfs_root_refs(&root->root_item) == 0) {
76dda93c
YZ		 5190 synchronize_srcu(&root->fs_info->subvol_srcu);
5191 spin_lock(&root->inode_lock);
5192 empty = RB_EMPTY_ROOT(&root->inode_tree);
5193 spin_unlock(&root->inode_lock);
5194 if (empty)
5195 btrfs_add_dead_root(root);
5196 }
5197}
5198
143bede5 5199void btrfs_invalidate_inodes(struct btrfs_root *root)
76dda93c
YZ		 5200{
5201 struct rb_node *node;
5202 struct rb_node *prev;
5203 struct btrfs_inode *entry;
5204 struct inode *inode;
5205 u64 objectid = 0;
5206
7813b3db
LB		 5207 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5208 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
76dda93c
YZ		 5209
5210 spin_lock(&root->inode_lock);
5211again:
5212 node = root->inode_tree.rb_node;
5213 prev = NULL;
5214 while (node) {
5215 prev = node;
5216 entry = rb_entry(node, struct btrfs_inode, rb_node);
5217
33345d01 5218 if (objectid < btrfs_ino(&entry->vfs_inode))
76dda93c 5219 node = node->rb_left;
33345d01 5220 else if (objectid > btrfs_ino(&entry->vfs_inode))
76dda93c
YZ		 5221 node = node->rb_right;
5222 else
5223 break;
5224 }
5225 if (!node) {
5226 while (prev) {
5227 entry = rb_entry(prev, struct btrfs_inode, rb_node);
33345d01 5228 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
76dda93c
YZ		 5229 node = prev;
5230 break;
5231 }
5232 prev = rb_next(prev);
5233 }
5234 }
5235 while (node) {
5236 entry = rb_entry(node, struct btrfs_inode, rb_node);
33345d01 5237 objectid = btrfs_ino(&entry->vfs_inode) + 1;
76dda93c
YZ		 5238 inode = igrab(&entry->vfs_inode);
5239 if (inode) {
5240 spin_unlock(&root->inode_lock);
5241 if (atomic_read(&inode->i_count) > 1)
5242 d_prune_aliases(inode);
5243 /*
45321ac5 5244 * btrfs_drop_inode will have it removed from
76dda93c
YZ		 5245 * the inode cache when its usage count
5246 * hits zero.
5247 */
5248 iput(inode);
5249 cond_resched();
5250 spin_lock(&root->inode_lock);
5251 goto again;
5252 }
5253
5254 if (cond_resched_lock(&root->inode_lock))
5255 goto again;
5256
5257 node = rb_next(node);
5258 }
5259 spin_unlock(&root->inode_lock);
5d4f98a2
YZ		 5260}
5261
e02119d5
CM		 5262static int btrfs_init_locked_inode(struct inode *inode, void *p)
5263{
5264 struct btrfs_iget_args *args = p;
90d3e592
CM		 5265 inode->i_ino = args->location->objectid;
5266 memcpy(&BTRFS_I(inode)->location, args->location,
5267 sizeof(*args->location));
e02119d5 5268 BTRFS_I(inode)->root = args->root;
39279cc3
CM		 5269 return 0;
5270}
5271
5272static int btrfs_find_actor(struct inode *inode, void *opaque)
5273{
5274 struct btrfs_iget_args *args = opaque;
90d3e592 5275 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
d397712b 5276 args->root == BTRFS_I(inode)->root;
39279cc3
CM		 5277}
5278
5d4f98a2 5279static struct inode *btrfs_iget_locked(struct super_block *s,
90d3e592 5280 struct btrfs_key *location,
5d4f98a2 5281 struct btrfs_root *root)
39279cc3
CM		 5282{
5283 struct inode *inode;
5284 struct btrfs_iget_args args;
90d3e592 5285 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
778ba82b 5286
90d3e592 5287 args.location = location;
39279cc3
CM		 5288 args.root = root;
5289
778ba82b 5290 inode = iget5_locked(s, hashval, btrfs_find_actor,
39279cc3
CM		 5291 btrfs_init_locked_inode,
5292 (void *)&args);
5293 return inode;
5294}
5295
1a54ef8c
BR		 5296/* Get an inode object given its location and corresponding root.
5297 * Returns in *is_new if the inode was read from disk
5298 */
5299struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
73f73415 5300 struct btrfs_root *root, int *new)
1a54ef8c
BR		 5301{
5302 struct inode *inode;
5303
90d3e592 5304 inode = btrfs_iget_locked(s, location, root);
1a54ef8c 5305 if (!inode)
5d4f98a2 5306 return ERR_PTR(-ENOMEM);
1a54ef8c
BR		 5307
5308 if (inode->i_state & I_NEW) {
1a54ef8c 5309 btrfs_read_locked_inode(inode);
1748f843
MF		 5310 if (!is_bad_inode(inode)) {
5311 inode_tree_add(inode);
5312 unlock_new_inode(inode);
5313 if (new)
5314 *new = 1;
5315 } else {
e0b6d65b
ST		 5316 unlock_new_inode(inode);
5317 iput(inode);
5318 inode = ERR_PTR(-ESTALE);
1748f843
MF		 5319 }
5320 }
5321
1a54ef8c
BR		 5322 return inode;
5323}
5324
4df27c4d
YZ		 5325static struct inode *new_simple_dir(struct super_block *s,
5326 struct btrfs_key *key,
5327 struct btrfs_root *root)
5328{
5329 struct inode *inode = new_inode(s);
5330
5331 if (!inode)
5332 return ERR_PTR(-ENOMEM);
5333
4df27c4d
YZ		 5334 BTRFS_I(inode)->root = root;
5335 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
72ac3c0d 5336 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4df27c4d
YZ		 5337
5338 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
848cce0d 5339 inode->i_op = &btrfs_dir_ro_inode_operations;
4df27c4d
YZ		 5340 inode->i_fop = &simple_dir_operations;
5341 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
9cc97d64 5342 inode->i_mtime = CURRENT_TIME;
5343 inode->i_atime = inode->i_mtime;
5344 inode->i_ctime = inode->i_mtime;
5345 BTRFS_I(inode)->i_otime = inode->i_mtime;
4df27c4d
YZ		 5346
5347 return inode;
5348}
5349
3de4586c 5350struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3 5351{
d397712b 5352 struct inode *inode;
4df27c4d 5353 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3
CM		 5354 struct btrfs_root *sub_root = root;
5355 struct btrfs_key location;
76dda93c 5356 int index;
b4aff1f8 5357 int ret = 0;
39279cc3
CM		 5358
5359 if (dentry->d_name.len > BTRFS_NAME_LEN)
5360 return ERR_PTR(-ENAMETOOLONG);
5f39d397 5361
39e3c955 5362 ret = btrfs_inode_by_name(dir, dentry, &location);
39279cc3
CM		 5363 if (ret < 0)
5364 return ERR_PTR(ret);
5f39d397 5365
4df27c4d 5366 if (location.objectid == 0)
5662344b 5367 return ERR_PTR(-ENOENT);
4df27c4d
YZ		 5368
5369 if (location.type == BTRFS_INODE_ITEM_KEY) {
73f73415 5370 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4df27c4d
YZ		 5371 return inode;
5372 }
5373
5374 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5375
76dda93c 5376 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4df27c4d
YZ		 5377 ret = fixup_tree_root_location(root, dir, dentry,
5378 &location, &sub_root);
5379 if (ret < 0) {
5380 if (ret != -ENOENT)
5381 inode = ERR_PTR(ret);
5382 else
5383 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5384 } else {
73f73415 5385 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
39279cc3 5386 }
76dda93c
YZ		 5387 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5388
34d19bad 5389 if (!IS_ERR(inode) && root != sub_root) {
c71bf099
YZ		 5390 down_read(&root->fs_info->cleanup_work_sem);
5391 if (!(inode->i_sb->s_flags & MS_RDONLY))
66b4ffd1 5392 ret = btrfs_orphan_cleanup(sub_root);
c71bf099 5393 up_read(&root->fs_info->cleanup_work_sem);
01cd3367
JB		 5394 if (ret) {
5395 iput(inode);
66b4ffd1 5396 inode = ERR_PTR(ret);
01cd3367 5397 }
c71bf099
YZ		 5398 }
5399
3de4586c
CM		 5400 return inode;
5401}
5402
fe15ce44 5403static int btrfs_dentry_delete(const struct dentry *dentry)
76dda93c
YZ		 5404{
5405 struct btrfs_root *root;
848cce0d 5406 struct inode *inode = dentry->d_inode;
76dda93c 5407
848cce0d
LZ		 5408 if (!inode && !IS_ROOT(dentry))
5409 inode = dentry->d_parent->d_inode;
76dda93c 5410
848cce0d
LZ		 5411 if (inode) {
5412 root = BTRFS_I(inode)->root;
efefb143
YZ		 5413 if (btrfs_root_refs(&root->root_item) == 0)
5414 return 1;
848cce0d
LZ		 5415
5416 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5417 return 1;
efefb143 5418 }
76dda93c
YZ		 5419 return 0;
5420}
5421
b4aff1f8
JB		 5422static void btrfs_dentry_release(struct dentry *dentry)
5423{
944a4515 5424 kfree(dentry->d_fsdata);
b4aff1f8
JB		 5425}
5426
3de4586c 5427static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
00cd8dd3 5428 unsigned int flags)
3de4586c 5429{
5662344b 5430 struct inode *inode;
a66e7cc6 5431
5662344b
TI		 5432 inode = btrfs_lookup_dentry(dir, dentry);
5433 if (IS_ERR(inode)) {
5434 if (PTR_ERR(inode) == -ENOENT)
5435 inode = NULL;
5436 else
5437 return ERR_CAST(inode);
5438 }
5439
41d28bca 5440 return d_splice_alias(inode, dentry);
39279cc3
CM		 5441}
5442
16cdcec7 5443unsigned char btrfs_filetype_table[] = {
39279cc3
CM		 5444 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5445};
5446
9cdda8d3 5447static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
39279cc3 5448{
9cdda8d3 5449 struct inode *inode = file_inode(file);
39279cc3
CM		 5450 struct btrfs_root *root = BTRFS_I(inode)->root;
5451 struct btrfs_item *item;
5452 struct btrfs_dir_item *di;
5453 struct btrfs_key key;
5f39d397 5454 struct btrfs_key found_key;
39279cc3 5455 struct btrfs_path *path;
16cdcec7
MX		 5456 struct list_head ins_list;
5457 struct list_head del_list;
39279cc3 5458 int ret;
5f39d397 5459 struct extent_buffer *leaf;
39279cc3 5460 int slot;
39279cc3
CM		 5461 unsigned char d_type;
5462 int over = 0;
5463 u32 di_cur;
5464 u32 di_total;
5465 u32 di_len;
5466 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 5467 char tmp_name[32];
5468 char *name_ptr;
5469 int name_len;
9cdda8d3 5470 int is_curr = 0; /* ctx->pos points to the current index? */
39279cc3
CM		 5471
5472 /* FIXME, use a real flag for deciding about the key type */
5473 if (root->fs_info->tree_root == root)
5474 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 5475
9cdda8d3
AV		 5476 if (!dir_emit_dots(file, ctx))
5477 return 0;
5478
49593bfa 5479 path = btrfs_alloc_path();
16cdcec7
MX		 5480 if (!path)
5481 return -ENOMEM;
ff5714cc 5482
026fd317 5483 path->reada = 1;
49593bfa 5484
16cdcec7
MX		 5485 if (key_type == BTRFS_DIR_INDEX_KEY) {
5486 INIT_LIST_HEAD(&ins_list);
5487 INIT_LIST_HEAD(&del_list);
5488 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5489 }
5490
962a298f 5491 key.type = key_type;
9cdda8d3 5492 key.offset = ctx->pos;
33345d01 5493 key.objectid = btrfs_ino(inode);
5f39d397 5494
39279cc3
CM		 5495 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5496 if (ret < 0)
5497 goto err;
49593bfa
DW		 5498
5499 while (1) {
5f39d397 5500 leaf = path->nodes[0];
39279cc3 5501 slot = path->slots[0];
b9e03af0
LZ		 5502 if (slot >= btrfs_header_nritems(leaf)) {
5503 ret = btrfs_next_leaf(root, path);
5504 if (ret < 0)
5505 goto err;
5506 else if (ret > 0)
5507 break;
5508 continue;
39279cc3 5509 }
3de4586c 5510
dd3cc16b 5511 item = btrfs_item_nr(slot);
5f39d397
CM		 5512 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5513
5514 if (found_key.objectid != key.objectid)
39279cc3 5515 break;
962a298f 5516 if (found_key.type != key_type)
39279cc3 5517 break;
9cdda8d3 5518 if (found_key.offset < ctx->pos)
b9e03af0 5519 goto next;
16cdcec7
MX		 5520 if (key_type == BTRFS_DIR_INDEX_KEY &&
5521 btrfs_should_delete_dir_index(&del_list,
5522 found_key.offset))
5523 goto next;
5f39d397 5524
9cdda8d3 5525 ctx->pos = found_key.offset;
16cdcec7 5526 is_curr = 1;
49593bfa 5527
39279cc3
CM		 5528 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5529 di_cur = 0;
5f39d397 5530 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 5531
5532 while (di_cur < di_total) {
5f39d397
CM		 5533 struct btrfs_key location;
5534
22a94d44
JB		 5535 if (verify_dir_item(root, leaf, di))
5536 break;
5537
5f39d397 5538 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 5539 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 5540 name_ptr = tmp_name;
5541 } else {
5542 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 5543 if (!name_ptr) {
5544 ret = -ENOMEM;
5545 goto err;
5546 }
5f39d397
CM		 5547 }
5548 read_extent_buffer(leaf, name_ptr,
5549 (unsigned long)(di + 1), name_len);
5550
5551 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5552 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c 5553
fede766f 5554
3de4586c 5555 /* is this a reference to our own snapshot? If so
8c9c2bf7
AJ		 5556 * skip it.
5557 *
5558 * In contrast to old kernels, we insert the snapshot's
5559 * dir item and dir index after it has been created, so
5560 * we won't find a reference to our own snapshot. We
5561 * still keep the following code for backward
5562 * compatibility.
3de4586c
CM		 5563 */
5564 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5565 location.objectid == root->root_key.objectid) {
5566 over = 0;
5567 goto skip;
5568 }
9cdda8d3
AV		 5569 over = !dir_emit(ctx, name_ptr, name_len,
5570 location.objectid, d_type);
5f39d397 5571
3de4586c 5572skip:
5f39d397
CM		 5573 if (name_ptr != tmp_name)
5574 kfree(name_ptr);
5575
39279cc3
CM		 5576 if (over)
5577 goto nopos;
5103e947 5578 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 5579 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 5580 di_cur += di_len;
5581 di = (struct btrfs_dir_item *)((char *)di + di_len);
5582 }
b9e03af0
LZ		 5583next:
5584 path->slots[0]++;
39279cc3 5585 }
49593bfa 5586
16cdcec7
MX		 5587 if (key_type == BTRFS_DIR_INDEX_KEY) {
5588 if (is_curr)
9cdda8d3
AV		 5589 ctx->pos++;
5590 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
16cdcec7
MX		 5591 if (ret)
5592 goto nopos;
5593 }
5594
49593bfa 5595 /* Reached end of directory/root. Bump pos past the last item. */
db62efbb
ZB		 5596 ctx->pos++;
5597
5598 /*
5599 * Stop new entries from being returned after we return the last
5600 * entry.
5601 *
5602 * New directory entries are assigned a strictly increasing
5603 * offset. This means that new entries created during readdir
5604 * are *guaranteed* to be seen in the future by that readdir.
5605 * This has broken buggy programs which operate on names as
5606 * they're returned by readdir. Until we re-use freed offsets
5607 * we have this hack to stop new entries from being returned
5608 * under the assumption that they'll never reach this huge
5609 * offset.
5610 *
5611 * This is being careful not to overflow 32bit loff_t unless the
5612 * last entry requires it because doing so has broken 32bit apps
5613 * in the past.
5614 */
5615 if (key_type == BTRFS_DIR_INDEX_KEY) {
5616 if (ctx->pos >= INT_MAX)
5617 ctx->pos = LLONG_MAX;
5618 else
5619 ctx->pos = INT_MAX;
5620 }
39279cc3
CM		 5621nopos:
5622 ret = 0;
5623err:
16cdcec7
MX		 5624 if (key_type == BTRFS_DIR_INDEX_KEY)
5625 btrfs_put_delayed_items(&ins_list, &del_list);
39279cc3 5626 btrfs_free_path(path);
39279cc3
CM		 5627 return ret;
5628}
5629
a9185b41 5630int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
39279cc3
CM		 5631{
5632 struct btrfs_root *root = BTRFS_I(inode)->root;
5633 struct btrfs_trans_handle *trans;
5634 int ret = 0;
0af3d00b 5635 bool nolock = false;
39279cc3 5636
72ac3c0d 5637 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4ca8b41e
CM		 5638 return 0;
5639
83eea1f1 5640 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
82d5902d 5641 nolock = true;
0af3d00b 5642
a9185b41 5643 if (wbc->sync_mode == WB_SYNC_ALL) {
0af3d00b 5644 if (nolock)
7a7eaa40 5645 trans = btrfs_join_transaction_nolock(root);
0af3d00b 5646 else
7a7eaa40 5647 trans = btrfs_join_transaction(root);
3612b495
TI		 5648 if (IS_ERR(trans))
5649 return PTR_ERR(trans);
a698d075 5650 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 5651 }
5652 return ret;
5653}
5654
5655/*
54aa1f4d 5656 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 5657 * inode changes. But, it is most likely to find the inode in cache.
5658 * FIXME, needs more benchmarking...there are no reasons other than performance
5659 * to keep or drop this code.
5660 */
48a3b636 5661static int btrfs_dirty_inode(struct inode *inode)
39279cc3
CM		 5662{
5663 struct btrfs_root *root = BTRFS_I(inode)->root;
5664 struct btrfs_trans_handle *trans;
8929ecfa
YZ		 5665 int ret;
5666
72ac3c0d 5667 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
22c44fe6 5668 return 0;
39279cc3 5669
7a7eaa40 5670 trans = btrfs_join_transaction(root);
22c44fe6
JB		 5671 if (IS_ERR(trans))
5672 return PTR_ERR(trans);
8929ecfa
YZ		 5673
5674 ret = btrfs_update_inode(trans, root, inode);
94b60442
CM		 5675 if (ret && ret == -ENOSPC) {
5676 /* whoops, lets try again with the full transaction */
5677 btrfs_end_transaction(trans, root);
5678 trans = btrfs_start_transaction(root, 1);
22c44fe6
JB		 5679 if (IS_ERR(trans))
5680 return PTR_ERR(trans);
8929ecfa 5681
94b60442 5682 ret = btrfs_update_inode(trans, root, inode);
94b60442 5683 }
39279cc3 5684 btrfs_end_transaction(trans, root);
16cdcec7
MX		 5685 if (BTRFS_I(inode)->delayed_node)
5686 btrfs_balance_delayed_items(root);
22c44fe6
JB		 5687
5688 return ret;
5689}
5690
5691/*
5692 * This is a copy of file_update_time. We need this so we can return error on
5693 * ENOSPC for updating the inode in the case of file write and mmap writes.
5694 */
e41f941a
JB		 5695static int btrfs_update_time(struct inode *inode, struct timespec *now,
5696 int flags)
22c44fe6 5697{
2bc55652
AB		 5698 struct btrfs_root *root = BTRFS_I(inode)->root;
5699
5700 if (btrfs_root_readonly(root))
5701 return -EROFS;
5702
e41f941a 5703 if (flags & S_VERSION)
22c44fe6 5704 inode_inc_iversion(inode);
e41f941a
JB		 5705 if (flags & S_CTIME)
5706 inode->i_ctime = *now;
5707 if (flags & S_MTIME)
5708 inode->i_mtime = *now;
5709 if (flags & S_ATIME)
5710 inode->i_atime = *now;
5711 return btrfs_dirty_inode(inode);
39279cc3
CM		 5712}
5713
d352ac68
CM		 5714/*
5715 * find the highest existing sequence number in a directory
5716 * and then set the in-memory index_cnt variable to reflect
5717 * free sequence numbers
5718 */
aec7477b
JB		 5719static int btrfs_set_inode_index_count(struct inode *inode)
5720{
5721 struct btrfs_root *root = BTRFS_I(inode)->root;
5722 struct btrfs_key key, found_key;
5723 struct btrfs_path *path;
5724 struct extent_buffer *leaf;
5725 int ret;
5726
33345d01 5727 key.objectid = btrfs_ino(inode);
962a298f 5728 key.type = BTRFS_DIR_INDEX_KEY;
aec7477b
JB		 5729 key.offset = (u64)-1;
5730
5731 path = btrfs_alloc_path();
5732 if (!path)
5733 return -ENOMEM;
5734
5735 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5736 if (ret < 0)
5737 goto out;
5738 /* FIXME: we should be able to handle this */
5739 if (ret == 0)
5740 goto out;
5741 ret = 0;
5742
5743 /*
5744 * MAGIC NUMBER EXPLANATION:
5745 * since we search a directory based on f_pos we have to start at 2
5746 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5747 * else has to start at 2
5748 */
5749 if (path->slots[0] == 0) {
5750 BTRFS_I(inode)->index_cnt = 2;
5751 goto out;
5752 }
5753
5754 path->slots[0]--;
5755
5756 leaf = path->nodes[0];
5757 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5758
33345d01 5759 if (found_key.objectid != btrfs_ino(inode) ||
962a298f 5760 found_key.type != BTRFS_DIR_INDEX_KEY) {
aec7477b
JB		 5761 BTRFS_I(inode)->index_cnt = 2;
5762 goto out;
5763 }
5764
5765 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5766out:
5767 btrfs_free_path(path);
5768 return ret;
5769}
5770
d352ac68
CM		 5771/*
5772 * helper to find a free sequence number in a given directory. This current
5773 * code is very simple, later versions will do smarter things in the btree
5774 */
3de4586c 5775int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 5776{
5777 int ret = 0;
5778
5779 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
16cdcec7
MX		 5780 ret = btrfs_inode_delayed_dir_index_count(dir);
5781 if (ret) {
5782 ret = btrfs_set_inode_index_count(dir);
5783 if (ret)
5784 return ret;
5785 }
aec7477b
JB		 5786 }
5787
00e4e6b3 5788 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 5789 BTRFS_I(dir)->index_cnt++;
5790
5791 return ret;
5792}
5793
b0d5d10f
CM		 5794static int btrfs_insert_inode_locked(struct inode *inode)
5795{
5796 struct btrfs_iget_args args;
5797 args.location = &BTRFS_I(inode)->location;
5798 args.root = BTRFS_I(inode)->root;
5799
5800 return insert_inode_locked4(inode,
5801 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
5802 btrfs_find_actor, &args);
5803}
5804
39279cc3
CM		 5805static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5806 struct btrfs_root *root,
aec7477b 5807 struct inode *dir,
9c58309d 5808 const char *name, int name_len,
175a4eb7
AV		 5809 u64 ref_objectid, u64 objectid,
5810 umode_t mode, u64 *index)
39279cc3
CM		 5811{
5812 struct inode *inode;
5f39d397 5813 struct btrfs_inode_item *inode_item;
39279cc3 5814 struct btrfs_key *location;
5f39d397 5815 struct btrfs_path *path;
9c58309d
CM		 5816 struct btrfs_inode_ref *ref;
5817 struct btrfs_key key[2];
5818 u32 sizes[2];
ef3b9af5 5819 int nitems = name ? 2 : 1;
9c58309d 5820 unsigned long ptr;
39279cc3 5821 int ret;
39279cc3 5822
5f39d397 5823 path = btrfs_alloc_path();
d8926bb3
MF		 5824 if (!path)
5825 return ERR_PTR(-ENOMEM);
5f39d397 5826
39279cc3 5827 inode = new_inode(root->fs_info->sb);
8fb27640
YS		 5828 if (!inode) {
5829 btrfs_free_path(path);
39279cc3 5830 return ERR_PTR(-ENOMEM);
8fb27640 5831 }
39279cc3 5832
5762b5c9
FM		 5833 /*
5834 * O_TMPFILE, set link count to 0, so that after this point,
5835 * we fill in an inode item with the correct link count.
5836 */
5837 if (!name)
5838 set_nlink(inode, 0);
5839
581bb050
LZ		 5840 /*
5841 * we have to initialize this early, so we can reclaim the inode
5842 * number if we fail afterwards in this function.
5843 */
5844 inode->i_ino = objectid;
5845
ef3b9af5 5846 if (dir && name) {
1abe9b8a 5847 trace_btrfs_inode_request(dir);
5848
3de4586c 5849 ret = btrfs_set_inode_index(dir, index);
09771430 5850 if (ret) {
8fb27640 5851 btrfs_free_path(path);
09771430 5852 iput(inode);
aec7477b 5853 return ERR_PTR(ret);
09771430 5854 }
ef3b9af5
FM		 5855 } else if (dir) {
5856 *index = 0;
aec7477b
JB		 5857 }
5858 /*
5859 * index_cnt is ignored for everything but a dir,
5860 * btrfs_get_inode_index_count has an explanation for the magic
5861 * number
5862 */
5863 BTRFS_I(inode)->index_cnt = 2;
67de1176 5864 BTRFS_I(inode)->dir_index = *index;
39279cc3 5865 BTRFS_I(inode)->root = root;
e02119d5 5866 BTRFS_I(inode)->generation = trans->transid;
76195853 5867 inode->i_generation = BTRFS_I(inode)->generation;
b888db2b 5868
5dc562c5
JB		 5869 /*
5870 * We could have gotten an inode number from somebody who was fsynced
5871 * and then removed in this same transaction, so let's just set full
5872 * sync since it will be a full sync anyway and this will blow away the
5873 * old info in the log.
5874 */
5875 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5876
9c58309d 5877 key[0].objectid = objectid;
962a298f 5878 key[0].type = BTRFS_INODE_ITEM_KEY;
9c58309d
CM		 5879 key[0].offset = 0;
5880
9c58309d 5881 sizes[0] = sizeof(struct btrfs_inode_item);
ef3b9af5
FM		 5882
5883 if (name) {
5884 /*
5885 * Start new inodes with an inode_ref. This is slightly more
5886 * efficient for small numbers of hard links since they will
5887 * be packed into one item. Extended refs will kick in if we
5888 * add more hard links than can fit in the ref item.
5889 */
5890 key[1].objectid = objectid;
962a298f 5891 key[1].type = BTRFS_INODE_REF_KEY;
ef3b9af5
FM		 5892 key[1].offset = ref_objectid;
5893
5894 sizes[1] = name_len + sizeof(*ref);
5895 }
9c58309d 5896
b0d5d10f
CM		 5897 location = &BTRFS_I(inode)->location;
5898 location->objectid = objectid;
5899 location->offset = 0;
962a298f 5900 location->type = BTRFS_INODE_ITEM_KEY;
b0d5d10f
CM		 5901
5902 ret = btrfs_insert_inode_locked(inode);
5903 if (ret < 0)
5904 goto fail;
5905
b9473439 5906 path->leave_spinning = 1;
ef3b9af5 5907 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
9c58309d 5908 if (ret != 0)
b0d5d10f 5909 goto fail_unlock;
5f39d397 5910
ecc11fab 5911 inode_init_owner(inode, dir, mode);
a76a3cd4 5912 inode_set_bytes(inode, 0);
9cc97d64 5913
5914 inode->i_mtime = CURRENT_TIME;
5915 inode->i_atime = inode->i_mtime;
5916 inode->i_ctime = inode->i_mtime;
5917 BTRFS_I(inode)->i_otime = inode->i_mtime;
5918
5f39d397
CM		 5919 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5920 struct btrfs_inode_item);
293f7e07
LZ		 5921 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5922 sizeof(*inode_item));
e02119d5 5923 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d 5924
ef3b9af5
FM		 5925 if (name) {
5926 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5927 struct btrfs_inode_ref);
5928 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5929 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5930 ptr = (unsigned long)(ref + 1);
5931 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5932 }
9c58309d 5933
5f39d397
CM		 5934 btrfs_mark_buffer_dirty(path->nodes[0]);
5935 btrfs_free_path(path);
5936
6cbff00f
CH		 5937 btrfs_inherit_iflags(inode, dir);
5938
569254b0 5939 if (S_ISREG(mode)) {
94272164
CM		 5940 if (btrfs_test_opt(root, NODATASUM))
5941 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
213490b3 5942 if (btrfs_test_opt(root, NODATACOW))
f2bdf9a8
JB		 5943 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5944 BTRFS_INODE_NODATASUM;
94272164
CM		 5945 }
5946
5d4f98a2 5947 inode_tree_add(inode);
1abe9b8a 5948
5949 trace_btrfs_inode_new(inode);
1973f0fa 5950 btrfs_set_inode_last_trans(trans, inode);
1abe9b8a 5951
8ea05e3a
AB		 5952 btrfs_update_root_times(trans, root);
5953
63541927
FDBM		 5954 ret = btrfs_inode_inherit_props(trans, inode, dir);
5955 if (ret)
5956 btrfs_err(root->fs_info,
5957 "error inheriting props for ino %llu (root %llu): %d",
5958 btrfs_ino(inode), root->root_key.objectid, ret);
5959
39279cc3 5960 return inode;
b0d5d10f
CM		 5961
5962fail_unlock:
5963 unlock_new_inode(inode);
5f39d397 5964fail:
ef3b9af5 5965 if (dir && name)
aec7477b 5966 BTRFS_I(dir)->index_cnt--;
5f39d397 5967 btrfs_free_path(path);
09771430 5968 iput(inode);
5f39d397 5969 return ERR_PTR(ret);
39279cc3
CM		 5970}
5971
5972static inline u8 btrfs_inode_type(struct inode *inode)
5973{
5974 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5975}
5976
d352ac68
CM		 5977/*
5978 * utility function to add 'inode' into 'parent_inode' with
5979 * a give name and a given sequence number.
5980 * if 'add_backref' is true, also insert a backref from the
5981 * inode to the parent directory.
5982 */
e02119d5
CM		 5983int btrfs_add_link(struct btrfs_trans_handle *trans,
5984 struct inode *parent_inode, struct inode *inode,
5985 const char *name, int name_len, int add_backref, u64 index)
39279cc3 5986{
4df27c4d 5987 int ret = 0;
39279cc3 5988 struct btrfs_key key;
e02119d5 5989 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
33345d01
LZ		 5990 u64 ino = btrfs_ino(inode);
5991 u64 parent_ino = btrfs_ino(parent_inode);
5f39d397 5992
33345d01 5993 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 5994 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5995 } else {
33345d01 5996 key.objectid = ino;
962a298f 5997 key.type = BTRFS_INODE_ITEM_KEY;
4df27c4d
YZ		 5998 key.offset = 0;
5999 }
6000
33345d01 6001 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 6002 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6003 key.objectid, root->root_key.objectid,
33345d01 6004 parent_ino, index, name, name_len);
4df27c4d 6005 } else if (add_backref) {
33345d01
LZ		 6006 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6007 parent_ino, index);
4df27c4d 6008 }
39279cc3 6009
79787eaa
JM		 6010 /* Nothing to clean up yet */
6011 if (ret)
6012 return ret;
4df27c4d 6013
79787eaa
JM		 6014 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6015 parent_inode, &key,
6016 btrfs_inode_type(inode), index);
9c52057c 6017 if (ret == -EEXIST || ret == -EOVERFLOW)
79787eaa
JM		 6018 goto fail_dir_item;
6019 else if (ret) {
6020 btrfs_abort_transaction(trans, root, ret);
6021 return ret;
39279cc3 6022 }
79787eaa
JM		 6023
6024 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6025 name_len * 2);
0c4d2d95 6026 inode_inc_iversion(parent_inode);
79787eaa
JM		 6027 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
6028 ret = btrfs_update_inode(trans, root, parent_inode);
6029 if (ret)
6030 btrfs_abort_transaction(trans, root, ret);
39279cc3 6031 return ret;
fe66a05a
CM		 6032
6033fail_dir_item:
6034 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6035 u64 local_index;
6036 int err;
6037 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6038 key.objectid, root->root_key.objectid,
6039 parent_ino, &local_index, name, name_len);
6040
6041 } else if (add_backref) {
6042 u64 local_index;
6043 int err;
6044
6045 err = btrfs_del_inode_ref(trans, root, name, name_len,
6046 ino, parent_ino, &local_index);
6047 }
6048 return ret;
39279cc3
CM		 6049}
6050
6051static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
a1b075d2
JB		 6052 struct inode *dir, struct dentry *dentry,
6053 struct inode *inode, int backref, u64 index)
39279cc3 6054{
a1b075d2
JB		 6055 int err = btrfs_add_link(trans, dir, inode,
6056 dentry->d_name.name, dentry->d_name.len,
6057 backref, index);
39279cc3
CM		 6058 if (err > 0)
6059 err = -EEXIST;
6060 return err;
6061}
6062
618e21d5 6063static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1a67aafb 6064 umode_t mode, dev_t rdev)
618e21d5
JB		 6065{
6066 struct btrfs_trans_handle *trans;
6067 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 6068 struct inode *inode = NULL;
618e21d5
JB		 6069 int err;
6070 int drop_inode = 0;
6071 u64 objectid;
00e4e6b3 6072 u64 index = 0;
618e21d5
JB		 6073
6074 if (!new_valid_dev(rdev))
6075 return -EINVAL;
6076
9ed74f2d
JB		 6077 /*
6078 * 2 for inode item and ref
6079 * 2 for dir items
6080 * 1 for xattr if selinux is on
6081 */
a22285a6
YZ		 6082 trans = btrfs_start_transaction(root, 5);
6083 if (IS_ERR(trans))
6084 return PTR_ERR(trans);
1832a6d5 6085
581bb050
LZ		 6086 err = btrfs_find_free_ino(root, &objectid);
6087 if (err)
6088 goto out_unlock;
6089
aec7477b 6090 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 6091 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 6092 mode, &index);
7cf96da3
TI		 6093 if (IS_ERR(inode)) {
6094 err = PTR_ERR(inode);
618e21d5 6095 goto out_unlock;
7cf96da3 6096 }
618e21d5 6097
ad19db71
CS		 6098 /*
6099 * If the active LSM wants to access the inode during
6100 * d_instantiate it needs these. Smack checks to see
6101 * if the filesystem supports xattrs by looking at the
6102 * ops vector.
6103 */
ad19db71 6104 inode->i_op = &btrfs_special_inode_operations;
b0d5d10f
CM		 6105 init_special_inode(inode, inode->i_mode, rdev);
6106
6107 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
618e21d5 6108 if (err)
b0d5d10f
CM		 6109 goto out_unlock_inode;
6110
6111 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6112 if (err) {
6113 goto out_unlock_inode;
6114 } else {
1b4ab1bb 6115 btrfs_update_inode(trans, root, inode);
b0d5d10f 6116 unlock_new_inode(inode);
08c422c2 6117 d_instantiate(dentry, inode);
618e21d5 6118 }
b0d5d10f 6119
618e21d5 6120out_unlock:
7ad85bb7 6121 btrfs_end_transaction(trans, root);
c581afc8 6122 btrfs_balance_delayed_items(root);
b53d3f5d 6123 btrfs_btree_balance_dirty(root);
618e21d5
JB		 6124 if (drop_inode) {
6125 inode_dec_link_count(inode);
6126 iput(inode);
6127 }
618e21d5 6128 return err;
b0d5d10f
CM		 6129
6130out_unlock_inode:
6131 drop_inode = 1;
6132 unlock_new_inode(inode);
6133 goto out_unlock;
6134
618e21d5
JB		 6135}
6136
39279cc3 6137static int btrfs_create(struct inode *dir, struct dentry *dentry,
ebfc3b49 6138 umode_t mode, bool excl)
39279cc3
CM		 6139{
6140 struct btrfs_trans_handle *trans;
6141 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 6142 struct inode *inode = NULL;
43baa579 6143 int drop_inode_on_err = 0;
a22285a6 6144 int err;
39279cc3 6145 u64 objectid;
00e4e6b3 6146 u64 index = 0;
39279cc3 6147
9ed74f2d
JB		 6148 /*
6149 * 2 for inode item and ref
6150 * 2 for dir items
6151 * 1 for xattr if selinux is on
6152 */
a22285a6
YZ		 6153 trans = btrfs_start_transaction(root, 5);
6154 if (IS_ERR(trans))
6155 return PTR_ERR(trans);
9ed74f2d 6156
581bb050
LZ		 6157 err = btrfs_find_free_ino(root, &objectid);
6158 if (err)
6159 goto out_unlock;
6160
aec7477b 6161 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 6162 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 6163 mode, &index);
7cf96da3
TI		 6164 if (IS_ERR(inode)) {
6165 err = PTR_ERR(inode);
39279cc3 6166 goto out_unlock;
7cf96da3 6167 }
43baa579 6168 drop_inode_on_err = 1;
ad19db71
CS		 6169 /*
6170 * If the active LSM wants to access the inode during
6171 * d_instantiate it needs these. Smack checks to see
6172 * if the filesystem supports xattrs by looking at the
6173 * ops vector.
6174 */
6175 inode->i_fop = &btrfs_file_operations;
6176 inode->i_op = &btrfs_file_inode_operations;
b0d5d10f
CM		 6177 inode->i_mapping->a_ops = &btrfs_aops;
6178 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
6179
6180 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6181 if (err)
6182 goto out_unlock_inode;
6183
6184 err = btrfs_update_inode(trans, root, inode);
6185 if (err)
6186 goto out_unlock_inode;
ad19db71 6187
a1b075d2 6188 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3 6189 if (err)
b0d5d10f 6190 goto out_unlock_inode;
43baa579 6191
43baa579 6192 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
b0d5d10f 6193 unlock_new_inode(inode);
43baa579
FB		 6194 d_instantiate(dentry, inode);
6195
39279cc3 6196out_unlock:
7ad85bb7 6197 btrfs_end_transaction(trans, root);
43baa579 6198 if (err && drop_inode_on_err) {
39279cc3
CM		 6199 inode_dec_link_count(inode);
6200 iput(inode);
6201 }
c581afc8 6202 btrfs_balance_delayed_items(root);
b53d3f5d 6203 btrfs_btree_balance_dirty(root);
39279cc3 6204 return err;
b0d5d10f
CM		 6205
6206out_unlock_inode:
6207 unlock_new_inode(inode);
6208 goto out_unlock;
6209
39279cc3
CM		 6210}
6211
6212static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6213 struct dentry *dentry)
6214{
6215 struct btrfs_trans_handle *trans;
6216 struct btrfs_root *root = BTRFS_I(dir)->root;
6217 struct inode *inode = old_dentry->d_inode;
00e4e6b3 6218 u64 index;
39279cc3
CM		 6219 int err;
6220 int drop_inode = 0;
6221
4a8be425
TH		 6222 /* do not allow sys_link's with other subvols of the same device */
6223 if (root->objectid != BTRFS_I(inode)->root->objectid)
3ab3564f 6224 return -EXDEV;
4a8be425 6225
f186373f 6226 if (inode->i_nlink >= BTRFS_LINK_MAX)
c055e99e 6227 return -EMLINK;
4a8be425 6228
3de4586c 6229 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 6230 if (err)
6231 goto fail;
6232
a22285a6 6233 /*
7e6b6465 6234 * 2 items for inode and inode ref
a22285a6 6235 * 2 items for dir items
7e6b6465 6236 * 1 item for parent inode
a22285a6 6237 */
7e6b6465 6238 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 6239 if (IS_ERR(trans)) {
6240 err = PTR_ERR(trans);
6241 goto fail;
6242 }
5f39d397 6243
67de1176
MX		 6244 /* There are several dir indexes for this inode, clear the cache. */
6245 BTRFS_I(inode)->dir_index = 0ULL;
8b558c5f 6246 inc_nlink(inode);
0c4d2d95 6247 inode_inc_iversion(inode);
3153495d 6248 inode->i_ctime = CURRENT_TIME;
7de9c6ee 6249 ihold(inode);
e9976151 6250 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
aec7477b 6251
a1b075d2 6252 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5f39d397 6253
a5719521 6254 if (err) {
54aa1f4d 6255 drop_inode = 1;
a5719521 6256 } else {
10d9f309 6257 struct dentry *parent = dentry->d_parent;
a5719521 6258 err = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 6259 if (err)
6260 goto fail;
ef3b9af5
FM		 6261 if (inode->i_nlink == 1) {
6262 /*
6263 * If new hard link count is 1, it's a file created
6264 * with open(2) O_TMPFILE flag.
6265 */
6266 err = btrfs_orphan_del(trans, inode);
6267 if (err)
6268 goto fail;
6269 }
08c422c2 6270 d_instantiate(dentry, inode);
6a912213 6271 btrfs_log_new_name(trans, inode, NULL, parent);
a5719521 6272 }
39279cc3 6273
7ad85bb7 6274 btrfs_end_transaction(trans, root);
c581afc8 6275 btrfs_balance_delayed_items(root);
1832a6d5 6276fail:
39279cc3
CM		 6277 if (drop_inode) {
6278 inode_dec_link_count(inode);
6279 iput(inode);
6280 }
b53d3f5d 6281 btrfs_btree_balance_dirty(root);
39279cc3
CM		 6282 return err;
6283}
6284
18bb1db3 6285static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
39279cc3 6286{
b9d86667 6287 struct inode *inode = NULL;
39279cc3
CM		 6288 struct btrfs_trans_handle *trans;
6289 struct btrfs_root *root = BTRFS_I(dir)->root;
6290 int err = 0;
6291 int drop_on_err = 0;
b9d86667 6292 u64 objectid = 0;
00e4e6b3 6293 u64 index = 0;
39279cc3 6294
9ed74f2d
JB		 6295 /*
6296 * 2 items for inode and ref
6297 * 2 items for dir items
6298 * 1 for xattr if selinux is on
6299 */
a22285a6
YZ		 6300 trans = btrfs_start_transaction(root, 5);
6301 if (IS_ERR(trans))
6302 return PTR_ERR(trans);
39279cc3 6303
581bb050
LZ		 6304 err = btrfs_find_free_ino(root, &objectid);
6305 if (err)
6306 goto out_fail;
6307
aec7477b 6308 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 6309 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 6310 S_IFDIR | mode, &index);
39279cc3
CM		 6311 if (IS_ERR(inode)) {
6312 err = PTR_ERR(inode);
6313 goto out_fail;
6314 }
5f39d397 6315
39279cc3 6316 drop_on_err = 1;
b0d5d10f
CM		 6317 /* these must be set before we unlock the inode */
6318 inode->i_op = &btrfs_dir_inode_operations;
6319 inode->i_fop = &btrfs_dir_file_operations;
33268eaf 6320
2a7dba39 6321 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf 6322 if (err)
b0d5d10f 6323 goto out_fail_inode;
39279cc3 6324
dbe674a9 6325 btrfs_i_size_write(inode, 0);
39279cc3
CM		 6326 err = btrfs_update_inode(trans, root, inode);
6327 if (err)
b0d5d10f 6328 goto out_fail_inode;
5f39d397 6329
a1b075d2
JB		 6330 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6331 dentry->d_name.len, 0, index);
39279cc3 6332 if (err)
b0d5d10f 6333 goto out_fail_inode;
5f39d397 6334
39279cc3 6335 d_instantiate(dentry, inode);
b0d5d10f
CM		 6336 /*
6337 * mkdir is special. We're unlocking after we call d_instantiate
6338 * to avoid a race with nfsd calling d_instantiate.
6339 */
6340 unlock_new_inode(inode);
39279cc3 6341 drop_on_err = 0;
39279cc3
CM		 6342
6343out_fail:
7ad85bb7 6344 btrfs_end_transaction(trans, root);
c7cfb8a5
WS		 6345 if (drop_on_err) {
6346 inode_dec_link_count(inode);
39279cc3 6347 iput(inode);
c7cfb8a5 6348 }
c581afc8 6349 btrfs_balance_delayed_items(root);
b53d3f5d 6350 btrfs_btree_balance_dirty(root);
39279cc3 6351 return err;
b0d5d10f
CM		 6352
6353out_fail_inode:
6354 unlock_new_inode(inode);
6355 goto out_fail;
39279cc3
CM		 6356}
6357
e6c4efd8
QW		 6358/* Find next extent map of a given extent map, caller needs to ensure locks */
6359static struct extent_map *next_extent_map(struct extent_map *em)
6360{
6361 struct rb_node *next;
6362
6363 next = rb_next(&em->rb_node);
6364 if (!next)
6365 return NULL;
6366 return container_of(next, struct extent_map, rb_node);
6367}
6368
6369static struct extent_map *prev_extent_map(struct extent_map *em)
6370{
6371 struct rb_node *prev;
6372
6373 prev = rb_prev(&em->rb_node);
6374 if (!prev)
6375 return NULL;
6376 return container_of(prev, struct extent_map, rb_node);
6377}
6378
d352ac68 6379/* helper for btfs_get_extent. Given an existing extent in the tree,
e6c4efd8 6380 * the existing extent is the nearest extent to map_start,
d352ac68 6381 * and an extent that you want to insert, deal with overlap and insert
e6c4efd8 6382 * the best fitted new extent into the tree.
d352ac68 6383 */
3b951516
CM		 6384static int merge_extent_mapping(struct extent_map_tree *em_tree,
6385 struct extent_map *existing,
e6dcd2dc 6386 struct extent_map *em,
51f395ad 6387 u64 map_start)
3b951516 6388{
e6c4efd8
QW		 6389 struct extent_map *prev;
6390 struct extent_map *next;
6391 u64 start;
6392 u64 end;
3b951516 6393 u64 start_diff;
3b951516 6394
e6dcd2dc 6395 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
e6c4efd8
QW		 6396
6397 if (existing->start > map_start) {
6398 next = existing;
6399 prev = prev_extent_map(next);
6400 } else {
6401 prev = existing;
6402 next = next_extent_map(prev);
6403 }
6404
6405 start = prev ? extent_map_end(prev) : em->start;
6406 start = max_t(u64, start, em->start);
6407 end = next ? next->start : extent_map_end(em);
6408 end = min_t(u64, end, extent_map_end(em));
6409 start_diff = start - em->start;
6410 em->start = start;
6411 em->len = end - start;
c8b97818
CM		 6412 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6413 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 6414 em->block_start += start_diff;
c8b97818
CM		 6415 em->block_len -= start_diff;
6416 }
09a2a8f9 6417 return add_extent_mapping(em_tree, em, 0);
3b951516
CM		 6418}
6419
c8b97818
CM		 6420static noinline int uncompress_inline(struct btrfs_path *path,
6421 struct inode *inode, struct page *page,
6422 size_t pg_offset, u64 extent_offset,
6423 struct btrfs_file_extent_item *item)
6424{
6425 int ret;
6426 struct extent_buffer *leaf = path->nodes[0];
6427 char *tmp;
6428 size_t max_size;
6429 unsigned long inline_size;
6430 unsigned long ptr;
261507a0 6431 int compress_type;
c8b97818
CM		 6432
6433 WARN_ON(pg_offset != 0);
261507a0 6434 compress_type = btrfs_file_extent_compression(leaf, item);
c8b97818
CM		 6435 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6436 inline_size = btrfs_file_extent_inline_item_len(leaf,
dd3cc16b 6437 btrfs_item_nr(path->slots[0]));
c8b97818 6438 tmp = kmalloc(inline_size, GFP_NOFS);
8d413713
TI		 6439 if (!tmp)
6440 return -ENOMEM;
c8b97818
CM		 6441 ptr = btrfs_file_extent_inline_start(item);
6442
6443 read_extent_buffer(leaf, tmp, ptr, inline_size);
6444
5b050f04 6445 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
261507a0
LZ		 6446 ret = btrfs_decompress(compress_type, tmp, page,
6447 extent_offset, inline_size, max_size);
c8b97818 6448 kfree(tmp);
166ae5a4 6449 return ret;
c8b97818
CM		 6450}
6451
d352ac68
CM		 6452/*
6453 * a bit scary, this does extent mapping from logical file offset to the disk.
d397712b
CM		 6454 * the ugly parts come from merging extents from the disk with the in-ram
6455 * representation. This gets more complex because of the data=ordered code,
d352ac68
CM		 6456 * where the in-ram extents might be locked pending data=ordered completion.
6457 *
6458 * This also copies inline extents directly into the page.
6459 */
d397712b 6460
a52d9a80 6461struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 6462 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 6463 int create)
6464{
6465 int ret;
6466 int err = 0;
a52d9a80
CM		 6467 u64 extent_start = 0;
6468 u64 extent_end = 0;
33345d01 6469 u64 objectid = btrfs_ino(inode);
a52d9a80 6470 u32 found_type;
f421950f 6471 struct btrfs_path *path = NULL;
a52d9a80
CM		 6472 struct btrfs_root *root = BTRFS_I(inode)->root;
6473 struct btrfs_file_extent_item *item;
5f39d397
CM		 6474 struct extent_buffer *leaf;
6475 struct btrfs_key found_key;
a52d9a80
CM		 6476 struct extent_map *em = NULL;
6477 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 6478 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 6479 struct btrfs_trans_handle *trans = NULL;
7ffbb598 6480 const bool new_inline = !page || create;
a52d9a80 6481
a52d9a80 6482again:
890871be 6483 read_lock(&em_tree->lock);
d1310b2e 6484 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 6485 if (em)
6486 em->bdev = root->fs_info->fs_devices->latest_bdev;
890871be 6487 read_unlock(&em_tree->lock);
d1310b2e 6488
a52d9a80 6489 if (em) {
e1c4b745
CM		 6490 if (em->start > start || em->start + em->len <= start)
6491 free_extent_map(em);
6492 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 6493 free_extent_map(em);
6494 else
6495 goto out;
a52d9a80 6496 }
172ddd60 6497 em = alloc_extent_map();
a52d9a80 6498 if (!em) {
d1310b2e
CM		 6499 err = -ENOMEM;
6500 goto out;
a52d9a80 6501 }
e6dcd2dc 6502 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 6503 em->start = EXTENT_MAP_HOLE;
445a6944 6504 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 6505 em->len = (u64)-1;
c8b97818 6506 em->block_len = (u64)-1;
f421950f
CM		 6507
6508 if (!path) {
6509 path = btrfs_alloc_path();
026fd317
JB		 6510 if (!path) {
6511 err = -ENOMEM;
6512 goto out;
6513 }
6514 /*
6515 * Chances are we'll be called again, so go ahead and do
6516 * readahead
6517 */
6518 path->reada = 1;
f421950f
CM		 6519 }
6520
179e29e4
CM		 6521 ret = btrfs_lookup_file_extent(trans, root, path,
6522 objectid, start, trans != NULL);
a52d9a80
CM		 6523 if (ret < 0) {
6524 err = ret;
6525 goto out;
6526 }
6527
6528 if (ret != 0) {
6529 if (path->slots[0] == 0)
6530 goto not_found;
6531 path->slots[0]--;
6532 }
6533
5f39d397
CM		 6534 leaf = path->nodes[0];
6535 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 6536 struct btrfs_file_extent_item);
a52d9a80 6537 /* are we inside the extent that was found? */
5f39d397 6538 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
962a298f 6539 found_type = found_key.type;
5f39d397 6540 if (found_key.objectid != objectid ||
a52d9a80 6541 found_type != BTRFS_EXTENT_DATA_KEY) {
25a50341
JB		 6542 /*
6543 * If we backup past the first extent we want to move forward
6544 * and see if there is an extent in front of us, otherwise we'll
6545 * say there is a hole for our whole search range which can
6546 * cause problems.
6547 */
6548 extent_end = start;
6549 goto next;
a52d9a80
CM		 6550 }
6551
5f39d397
CM		 6552 found_type = btrfs_file_extent_type(leaf, item);
6553 extent_start = found_key.offset;
d899e052
YZ		 6554 if (found_type == BTRFS_FILE_EXTENT_REG ||
6555 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 6556 extent_end = extent_start +
db94535d 6557 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 6558 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6559 size_t size;
514ac8ad 6560 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
fda2832f 6561 extent_end = ALIGN(extent_start + size, root->sectorsize);
9036c102 6562 }
25a50341 6563next:
9036c102
YZ		 6564 if (start >= extent_end) {
6565 path->slots[0]++;
6566 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6567 ret = btrfs_next_leaf(root, path);
6568 if (ret < 0) {
6569 err = ret;
6570 goto out;
a52d9a80 6571 }
9036c102
YZ		 6572 if (ret > 0)
6573 goto not_found;
6574 leaf = path->nodes[0];
a52d9a80 6575 }
9036c102
YZ		 6576 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6577 if (found_key.objectid != objectid ||
6578 found_key.type != BTRFS_EXTENT_DATA_KEY)
6579 goto not_found;
6580 if (start + len <= found_key.offset)
6581 goto not_found;
e2eca69d
WS		 6582 if (start > found_key.offset)
6583 goto next;
9036c102 6584 em->start = start;
70c8a91c 6585 em->orig_start = start;
9036c102
YZ		 6586 em->len = found_key.offset - start;
6587 goto not_found_em;
6588 }
6589
7ffbb598
FM		 6590 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6591
d899e052
YZ		 6592 if (found_type == BTRFS_FILE_EXTENT_REG ||
6593 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80
CM		 6594 goto insert;
6595 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 6596 unsigned long ptr;
a52d9a80 6597 char *map;
3326d1b0
CM		 6598 size_t size;
6599 size_t extent_offset;
6600 size_t copy_size;
a52d9a80 6601
7ffbb598 6602 if (new_inline)
689f9346 6603 goto out;
5f39d397 6604
514ac8ad 6605 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
9036c102 6606 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 6607 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 6608 size - extent_offset);
3326d1b0 6609 em->start = extent_start + extent_offset;
fda2832f 6610 em->len = ALIGN(copy_size, root->sectorsize);
b4939680 6611 em->orig_block_len = em->len;
70c8a91c 6612 em->orig_start = em->start;
689f9346 6613 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 6614 if (create == 0 && !PageUptodate(page)) {
261507a0
LZ		 6615 if (btrfs_file_extent_compression(leaf, item) !=
6616 BTRFS_COMPRESS_NONE) {
c8b97818
CM		 6617 ret = uncompress_inline(path, inode, page,
6618 pg_offset,
6619 extent_offset, item);
166ae5a4
ZB		 6620 if (ret) {
6621 err = ret;
6622 goto out;
6623 }
c8b97818
CM		 6624 } else {
6625 map = kmap(page);
6626 read_extent_buffer(leaf, map + pg_offset, ptr,
6627 copy_size);
93c82d57
CM		 6628 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6629 memset(map + pg_offset + copy_size, 0,
6630 PAGE_CACHE_SIZE - pg_offset -
6631 copy_size);
6632 }
c8b97818
CM		 6633 kunmap(page);
6634 }
179e29e4
CM		 6635 flush_dcache_page(page);
6636 } else if (create && PageUptodate(page)) {
6bf7e080 6637 BUG();
179e29e4
CM		 6638 if (!trans) {
6639 kunmap(page);
6640 free_extent_map(em);
6641 em = NULL;
ff5714cc 6642
b3b4aa74 6643 btrfs_release_path(path);
7a7eaa40 6644 trans = btrfs_join_transaction(root);
ff5714cc 6645
3612b495
TI		 6646 if (IS_ERR(trans))
6647 return ERR_CAST(trans);
179e29e4
CM		 6648 goto again;
6649 }
c8b97818 6650 map = kmap(page);
70dec807 6651 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 6652 copy_size);
c8b97818 6653 kunmap(page);
179e29e4 6654 btrfs_mark_buffer_dirty(leaf);
a52d9a80 6655 }
d1310b2e 6656 set_extent_uptodate(io_tree, em->start,
507903b8 6657 extent_map_end(em) - 1, NULL, GFP_NOFS);
a52d9a80 6658 goto insert;
a52d9a80
CM		 6659 }
6660not_found:
6661 em->start = start;
70c8a91c 6662 em->orig_start = start;
d1310b2e 6663 em->len = len;
a52d9a80 6664not_found_em:
5f39d397 6665 em->block_start = EXTENT_MAP_HOLE;
9036c102 6666 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80 6667insert:
b3b4aa74 6668 btrfs_release_path(path);
d1310b2e 6669 if (em->start > start || extent_map_end(em) <= start) {
c2cf52eb 6670 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
c1c9ff7c 6671 em->start, em->len, start, len);
a52d9a80
CM		 6672 err = -EIO;
6673 goto out;
6674 }
d1310b2e
CM		 6675
6676 err = 0;
890871be 6677 write_lock(&em_tree->lock);
09a2a8f9 6678 ret = add_extent_mapping(em_tree, em, 0);
3b951516
CM		 6679 /* it is possible that someone inserted the extent into the tree
6680 * while we had the lock dropped. It is also possible that
6681 * an overlapping map exists in the tree
6682 */
a52d9a80 6683 if (ret == -EEXIST) {
3b951516 6684 struct extent_map *existing;
e6dcd2dc
CM		 6685
6686 ret = 0;
6687
e6c4efd8
QW		 6688 existing = search_extent_mapping(em_tree, start, len);
6689 /*
6690 * existing will always be non-NULL, since there must be
6691 * extent causing the -EEXIST.
6692 */
6693 if (start >= extent_map_end(existing) ||
32be3a1a 6694 start <= existing->start) {
e6c4efd8
QW		 6695 /*
6696 * The existing extent map is the one nearest to
6697 * the [start, start + len) range which overlaps
6698 */
6699 err = merge_extent_mapping(em_tree, existing,
6700 em, start);
e1c4b745 6701 free_extent_map(existing);
e6c4efd8 6702 if (err) {
3b951516
CM		 6703 free_extent_map(em);
6704 em = NULL;
6705 }
6706 } else {
6707 free_extent_map(em);
6708 em = existing;
e6dcd2dc 6709 err = 0;
a52d9a80 6710 }
a52d9a80 6711 }
890871be 6712 write_unlock(&em_tree->lock);
a52d9a80 6713out:
1abe9b8a 6714
4cd8587c 6715 trace_btrfs_get_extent(root, em);
1abe9b8a 6716
f421950f
CM		 6717 if (path)
6718 btrfs_free_path(path);
a52d9a80
CM		 6719 if (trans) {
6720 ret = btrfs_end_transaction(trans, root);
d397712b 6721 if (!err)
a52d9a80
CM		 6722 err = ret;
6723 }
a52d9a80
CM		 6724 if (err) {
6725 free_extent_map(em);
a52d9a80
CM		 6726 return ERR_PTR(err);
6727 }
79787eaa 6728 BUG_ON(!em); /* Error is always set */
a52d9a80
CM		 6729 return em;
6730}
6731
ec29ed5b
CM		 6732struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6733 size_t pg_offset, u64 start, u64 len,
6734 int create)
6735{
6736 struct extent_map *em;
6737 struct extent_map *hole_em = NULL;
6738 u64 range_start = start;
6739 u64 end;
6740 u64 found;
6741 u64 found_end;
6742 int err = 0;
6743
6744 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6745 if (IS_ERR(em))
6746 return em;
6747 if (em) {
6748 /*
f9e4fb53
LB		 6749 * if our em maps to
6750 * - a hole or
6751 * - a pre-alloc extent,
6752 * there might actually be delalloc bytes behind it.
ec29ed5b 6753 */
f9e4fb53
LB		 6754 if (em->block_start != EXTENT_MAP_HOLE &&
6755 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
ec29ed5b
CM		 6756 return em;
6757 else
6758 hole_em = em;
6759 }
6760
6761 /* check to see if we've wrapped (len == -1 or similar) */
6762 end = start + len;
6763 if (end < start)
6764 end = (u64)-1;
6765 else
6766 end -= 1;
6767
6768 em = NULL;
6769
6770 /* ok, we didn't find anything, lets look for delalloc */
6771 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6772 end, len, EXTENT_DELALLOC, 1);
6773 found_end = range_start + found;
6774 if (found_end < range_start)
6775 found_end = (u64)-1;
6776
6777 /*
6778 * we didn't find anything useful, return
6779 * the original results from get_extent()
6780 */
6781 if (range_start > end || found_end <= start) {
6782 em = hole_em;
6783 hole_em = NULL;
6784 goto out;
6785 }
6786
6787 /* adjust the range_start to make sure it doesn't
6788 * go backwards from the start they passed in
6789 */
67871254 6790 range_start = max(start, range_start);
ec29ed5b
CM		 6791 found = found_end - range_start;
6792
6793 if (found > 0) {
6794 u64 hole_start = start;
6795 u64 hole_len = len;
6796
172ddd60 6797 em = alloc_extent_map();
ec29ed5b
CM		 6798 if (!em) {
6799 err = -ENOMEM;
6800 goto out;
6801 }
6802 /*
6803 * when btrfs_get_extent can't find anything it
6804 * returns one huge hole
6805 *
6806 * make sure what it found really fits our range, and
6807 * adjust to make sure it is based on the start from
6808 * the caller
6809 */
6810 if (hole_em) {
6811 u64 calc_end = extent_map_end(hole_em);
6812
6813 if (calc_end <= start || (hole_em->start > end)) {
6814 free_extent_map(hole_em);
6815 hole_em = NULL;
6816 } else {
6817 hole_start = max(hole_em->start, start);
6818 hole_len = calc_end - hole_start;
6819 }
6820 }
6821 em->bdev = NULL;
6822 if (hole_em && range_start > hole_start) {
6823 /* our hole starts before our delalloc, so we
6824 * have to return just the parts of the hole
6825 * that go until the delalloc starts
6826 */
6827 em->len = min(hole_len,
6828 range_start - hole_start);
6829 em->start = hole_start;
6830 em->orig_start = hole_start;
6831 /*
6832 * don't adjust block start at all,
6833 * it is fixed at EXTENT_MAP_HOLE
6834 */
6835 em->block_start = hole_em->block_start;
6836 em->block_len = hole_len;
f9e4fb53
LB		 6837 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6838 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
ec29ed5b
CM		 6839 } else {
6840 em->start = range_start;
6841 em->len = found;
6842 em->orig_start = range_start;
6843 em->block_start = EXTENT_MAP_DELALLOC;
6844 em->block_len = found;
6845 }
6846 } else if (hole_em) {
6847 return hole_em;
6848 }
6849out:
6850
6851 free_extent_map(hole_em);
6852 if (err) {
6853 free_extent_map(em);
6854 return ERR_PTR(err);
6855 }
6856 return em;
6857}
6858
4b46fce2
JB		 6859static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6860 u64 start, u64 len)
6861{
6862 struct btrfs_root *root = BTRFS_I(inode)->root;
70c8a91c 6863 struct extent_map *em;
4b46fce2
JB		 6864 struct btrfs_key ins;
6865 u64 alloc_hint;
6866 int ret;
4b46fce2 6867
4b46fce2 6868 alloc_hint = get_extent_allocation_hint(inode, start, len);
00361589 6869 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
e570fd27 6870 alloc_hint, &ins, 1, 1);
00361589
JB		 6871 if (ret)
6872 return ERR_PTR(ret);
4b46fce2 6873
70c8a91c 6874 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
cc95bef6 6875 ins.offset, ins.offset, ins.offset, 0);
00361589 6876 if (IS_ERR(em)) {
e570fd27 6877 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
00361589
JB		 6878 return em;
6879 }
4b46fce2
JB		 6880
6881 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6882 ins.offset, ins.offset, 0);
6883 if (ret) {
e570fd27 6884 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
00361589
JB		 6885 free_extent_map(em);
6886 return ERR_PTR(ret);
4b46fce2 6887 }
00361589 6888
4b46fce2
JB		 6889 return em;
6890}
6891
46bfbb5c
CM		 6892/*
6893 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6894 * block must be cow'd
6895 */
00361589 6896noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7ee9e440
JB		 6897 u64 *orig_start, u64 *orig_block_len,
6898 u64 *ram_bytes)
46bfbb5c 6899{
00361589 6900 struct btrfs_trans_handle *trans;
46bfbb5c
CM		 6901 struct btrfs_path *path;
6902 int ret;
6903 struct extent_buffer *leaf;
6904 struct btrfs_root *root = BTRFS_I(inode)->root;
7b2b7085 6905 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
46bfbb5c
CM		 6906 struct btrfs_file_extent_item *fi;
6907 struct btrfs_key key;
6908 u64 disk_bytenr;
6909 u64 backref_offset;
6910 u64 extent_end;
6911 u64 num_bytes;
6912 int slot;
6913 int found_type;
7ee9e440 6914 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
e77751aa 6915
46bfbb5c
CM		 6916 path = btrfs_alloc_path();
6917 if (!path)
6918 return -ENOMEM;
6919
00361589 6920 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
46bfbb5c
CM		 6921 offset, 0);
6922 if (ret < 0)
6923 goto out;
6924
6925 slot = path->slots[0];
6926 if (ret == 1) {
6927 if (slot == 0) {
6928 /* can't find the item, must cow */
6929 ret = 0;
6930 goto out;
6931 }
6932 slot--;
6933 }
6934 ret = 0;
6935 leaf = path->nodes[0];
6936 btrfs_item_key_to_cpu(leaf, &key, slot);
33345d01 6937 if (key.objectid != btrfs_ino(inode) ||
46bfbb5c
CM		 6938 key.type != BTRFS_EXTENT_DATA_KEY) {
6939 /* not our file or wrong item type, must cow */
6940 goto out;
6941 }
6942
6943 if (key.offset > offset) {
6944 /* Wrong offset, must cow */
6945 goto out;
6946 }
6947
6948 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6949 found_type = btrfs_file_extent_type(leaf, fi);
6950 if (found_type != BTRFS_FILE_EXTENT_REG &&
6951 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6952 /* not a regular extent, must cow */
6953 goto out;
6954 }
7ee9e440
JB		 6955
6956 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6957 goto out;
6958
e77751aa
MX		 6959 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6960 if (extent_end <= offset)
6961 goto out;
6962
46bfbb5c 6963 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7ee9e440
JB		 6964 if (disk_bytenr == 0)
6965 goto out;
6966
6967 if (btrfs_file_extent_compression(leaf, fi) ||
6968 btrfs_file_extent_encryption(leaf, fi) ||
6969 btrfs_file_extent_other_encoding(leaf, fi))
6970 goto out;
6971
46bfbb5c
CM		 6972 backref_offset = btrfs_file_extent_offset(leaf, fi);
6973
7ee9e440
JB		 6974 if (orig_start) {
6975 *orig_start = key.offset - backref_offset;
6976 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6977 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6978 }
eb384b55 6979
46bfbb5c
CM		 6980 if (btrfs_extent_readonly(root, disk_bytenr))
6981 goto out;
7b2b7085
MX		 6982
6983 num_bytes = min(offset + *len, extent_end) - offset;
6984 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6985 u64 range_end;
6986
6987 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
6988 ret = test_range_bit(io_tree, offset, range_end,
6989 EXTENT_DELALLOC, 0, NULL);
6990 if (ret) {
6991 ret = -EAGAIN;
6992 goto out;
6993 }
6994 }
6995
1bda19eb 6996 btrfs_release_path(path);
46bfbb5c
CM		 6997
6998 /*
6999 * look for other files referencing this extent, if we
7000 * find any we must cow
7001 */
00361589
JB		 7002 trans = btrfs_join_transaction(root);
7003 if (IS_ERR(trans)) {
7004 ret = 0;
46bfbb5c 7005 goto out;
00361589
JB		 7006 }
7007
7008 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7009 key.offset - backref_offset, disk_bytenr);
7010 btrfs_end_transaction(trans, root);
7011 if (ret) {
7012 ret = 0;
7013 goto out;
7014 }
46bfbb5c
CM		 7015
7016 /*
7017 * adjust disk_bytenr and num_bytes to cover just the bytes
7018 * in this extent we are about to write. If there
7019 * are any csums in that range we have to cow in order
7020 * to keep the csums correct
7021 */
7022 disk_bytenr += backref_offset;
7023 disk_bytenr += offset - key.offset;
46bfbb5c
CM		 7024 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7025 goto out;
7026 /*
7027 * all of the above have passed, it is safe to overwrite this extent
7028 * without cow
7029 */
eb384b55 7030 *len = num_bytes;
46bfbb5c
CM		 7031 ret = 1;
7032out:
7033 btrfs_free_path(path);
7034 return ret;
7035}
7036
fc4adbff
AG		 7037bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7038{
7039 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7040 int found = false;
7041 void **pagep = NULL;
7042 struct page *page = NULL;
7043 int start_idx;
7044 int end_idx;
7045
7046 start_idx = start >> PAGE_CACHE_SHIFT;
7047
7048 /*
7049 * end is the last byte in the last page. end == start is legal
7050 */
7051 end_idx = end >> PAGE_CACHE_SHIFT;
7052
7053 rcu_read_lock();
7054
7055 /* Most of the code in this while loop is lifted from
7056 * find_get_page. It's been modified to begin searching from a
7057 * page and return just the first page found in that range. If the
7058 * found idx is less than or equal to the end idx then we know that
7059 * a page exists. If no pages are found or if those pages are
7060 * outside of the range then we're fine (yay!) */
7061 while (page == NULL &&
7062 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7063 page = radix_tree_deref_slot(pagep);
7064 if (unlikely(!page))
7065 break;
7066
7067 if (radix_tree_exception(page)) {
809f9016
FM		 7068 if (radix_tree_deref_retry(page)) {
7069 page = NULL;
fc4adbff 7070 continue;
809f9016 7071 }
fc4adbff
AG		 7072 /*
7073 * Otherwise, shmem/tmpfs must be storing a swap entry
7074 * here as an exceptional entry: so return it without
7075 * attempting to raise page count.
7076 */
6fdef6d4 7077 page = NULL;
fc4adbff
AG		 7078 break; /* TODO: Is this relevant for this use case? */
7079 }
7080
91405151
FM		 7081 if (!page_cache_get_speculative(page)) {
7082 page = NULL;
fc4adbff 7083 continue;
91405151 7084 }
fc4adbff
AG		 7085
7086 /*
7087 * Has the page moved?
7088 * This is part of the lockless pagecache protocol. See
7089 * include/linux/pagemap.h for details.
7090 */
7091 if (unlikely(page != *pagep)) {
7092 page_cache_release(page);
7093 page = NULL;
7094 }
7095 }
7096
7097 if (page) {
7098 if (page->index <= end_idx)
7099 found = true;
7100 page_cache_release(page);
7101 }
7102
7103 rcu_read_unlock();
7104 return found;
7105}
7106
eb838e73
JB		 7107static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7108 struct extent_state **cached_state, int writing)
7109{
7110 struct btrfs_ordered_extent *ordered;
7111 int ret = 0;
7112
7113 while (1) {
7114 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7115 0, cached_state);
7116 /*
7117 * We're concerned with the entire range that we're going to be
7118 * doing DIO to, so we need to make sure theres no ordered
7119 * extents in this range.
7120 */
7121 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7122 lockend - lockstart + 1);
7123
7124 /*
7125 * We need to make sure there are no buffered pages in this
7126 * range either, we could have raced between the invalidate in
7127 * generic_file_direct_write and locking the extent. The
7128 * invalidate needs to happen so that reads after a write do not
7129 * get stale data.
7130 */
fc4adbff
AG		 7131 if (!ordered &&
7132 (!writing ||
7133 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
eb838e73
JB		 7134 break;
7135
7136 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7137 cached_state, GFP_NOFS);
7138
7139 if (ordered) {
7140 btrfs_start_ordered_extent(inode, ordered, 1);
7141 btrfs_put_ordered_extent(ordered);
7142 } else {
7143 /* Screw you mmap */
728404da 7144 ret = btrfs_fdatawrite_range(inode, lockstart, lockend);
075bdbdb
FM		 7145 if (ret)
7146 break;
7147 ret = filemap_fdatawait_range(inode->i_mapping,
7148 lockstart,
7149 lockend);
eb838e73
JB		 7150 if (ret)
7151 break;
7152
7153 /*
7154 * If we found a page that couldn't be invalidated just
7155 * fall back to buffered.
7156 */
7157 ret = invalidate_inode_pages2_range(inode->i_mapping,
7158 lockstart >> PAGE_CACHE_SHIFT,
7159 lockend >> PAGE_CACHE_SHIFT);
7160 if (ret)
7161 break;
7162 }
7163
7164 cond_resched();
7165 }
7166
7167 return ret;
7168}
7169
69ffb543
JB		 7170static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7171 u64 len, u64 orig_start,
7172 u64 block_start, u64 block_len,
cc95bef6
JB		 7173 u64 orig_block_len, u64 ram_bytes,
7174 int type)
69ffb543
JB		 7175{
7176 struct extent_map_tree *em_tree;
7177 struct extent_map *em;
7178 struct btrfs_root *root = BTRFS_I(inode)->root;
7179 int ret;
7180
7181 em_tree = &BTRFS_I(inode)->extent_tree;
7182 em = alloc_extent_map();
7183 if (!em)
7184 return ERR_PTR(-ENOMEM);
7185
7186 em->start = start;
7187 em->orig_start = orig_start;
2ab28f32
JB		 7188 em->mod_start = start;
7189 em->mod_len = len;
69ffb543
JB		 7190 em->len = len;
7191 em->block_len = block_len;
7192 em->block_start = block_start;
7193 em->bdev = root->fs_info->fs_devices->latest_bdev;
b4939680 7194 em->orig_block_len = orig_block_len;
cc95bef6 7195 em->ram_bytes = ram_bytes;
70c8a91c 7196 em->generation = -1;
69ffb543
JB		 7197 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7198 if (type == BTRFS_ORDERED_PREALLOC)
b11e234d 7199 set_bit(EXTENT_FLAG_FILLING, &em->flags);
69ffb543
JB		 7200
7201 do {
7202 btrfs_drop_extent_cache(inode, em->start,
7203 em->start + em->len - 1, 0);
7204 write_lock(&em_tree->lock);
09a2a8f9 7205 ret = add_extent_mapping(em_tree, em, 1);
69ffb543
JB		 7206 write_unlock(&em_tree->lock);
7207 } while (ret == -EEXIST);
7208
7209 if (ret) {
7210 free_extent_map(em);
7211 return ERR_PTR(ret);
7212 }
7213
7214 return em;
7215}
7216
7217
4b46fce2
JB		 7218static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7219 struct buffer_head *bh_result, int create)
7220{
7221 struct extent_map *em;
7222 struct btrfs_root *root = BTRFS_I(inode)->root;
eb838e73 7223 struct extent_state *cached_state = NULL;
4b46fce2 7224 u64 start = iblock << inode->i_blkbits;
eb838e73 7225 u64 lockstart, lockend;
4b46fce2 7226 u64 len = bh_result->b_size;
3e05bde8 7227 u64 orig_len = len;
eb838e73 7228 int unlock_bits = EXTENT_LOCKED;
0934856d 7229 int ret = 0;
eb838e73 7230
172a5049 7231 if (create)
3266789f 7232 unlock_bits |= EXTENT_DIRTY;
172a5049 7233 else
c329861d 7234 len = min_t(u64, len, root->sectorsize);
eb838e73 7235
c329861d
JB		 7236 lockstart = start;
7237 lockend = start + len - 1;
7238
eb838e73
JB		 7239 /*
7240 * If this errors out it's because we couldn't invalidate pagecache for
7241 * this range and we need to fallback to buffered.
7242 */
7243 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
7244 return -ENOTBLK;
7245
4b46fce2 7246 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
eb838e73
JB		 7247 if (IS_ERR(em)) {
7248 ret = PTR_ERR(em);
7249 goto unlock_err;
7250 }
4b46fce2
JB		 7251
7252 /*
7253 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7254 * io. INLINE is special, and we could probably kludge it in here, but
7255 * it's still buffered so for safety lets just fall back to the generic
7256 * buffered path.
7257 *
7258 * For COMPRESSED we _have_ to read the entire extent in so we can
7259 * decompress it, so there will be buffering required no matter what we
7260 * do, so go ahead and fallback to buffered.
7261 *
7262 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7263 * to buffered IO. Don't blame me, this is the price we pay for using
7264 * the generic code.
7265 */
7266 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7267 em->block_start == EXTENT_MAP_INLINE) {
7268 free_extent_map(em);
eb838e73
JB		 7269 ret = -ENOTBLK;
7270 goto unlock_err;
4b46fce2
JB		 7271 }
7272
7273 /* Just a good old fashioned hole, return */
7274 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7275 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7276 free_extent_map(em);
eb838e73 7277 goto unlock_err;
4b46fce2
JB		 7278 }
7279
7280 /*
7281 * We don't allocate a new extent in the following cases
7282 *
7283 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7284 * existing extent.
7285 * 2) The extent is marked as PREALLOC. We're good to go here and can
7286 * just use the extent.
7287 *
7288 */
46bfbb5c 7289 if (!create) {
eb838e73
JB		 7290 len = min(len, em->len - (start - em->start));
7291 lockstart = start + len;
7292 goto unlock;
46bfbb5c 7293 }
4b46fce2
JB		 7294
7295 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7296 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7297 em->block_start != EXTENT_MAP_HOLE)) {
4b46fce2 7298 int type;
eb384b55 7299 u64 block_start, orig_start, orig_block_len, ram_bytes;
4b46fce2
JB		 7300
7301 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7302 type = BTRFS_ORDERED_PREALLOC;
7303 else
7304 type = BTRFS_ORDERED_NOCOW;
46bfbb5c 7305 len = min(len, em->len - (start - em->start));
4b46fce2 7306 block_start = em->block_start + (start - em->start);
46bfbb5c 7307
00361589 7308 if (can_nocow_extent(inode, start, &len, &orig_start,
7ee9e440 7309 &orig_block_len, &ram_bytes) == 1) {
69ffb543
JB		 7310 if (type == BTRFS_ORDERED_PREALLOC) {
7311 free_extent_map(em);
7312 em = create_pinned_em(inode, start, len,
7313 orig_start,
b4939680 7314 block_start, len,
cc95bef6
JB		 7315 orig_block_len,
7316 ram_bytes, type);
555e1286
FM		 7317 if (IS_ERR(em)) {
7318 ret = PTR_ERR(em);
69ffb543 7319 goto unlock_err;
555e1286 7320 }
69ffb543
JB		 7321 }
7322
46bfbb5c
CM		 7323 ret = btrfs_add_ordered_extent_dio(inode, start,
7324 block_start, len, len, type);
46bfbb5c
CM		 7325 if (ret) {
7326 free_extent_map(em);
eb838e73 7327 goto unlock_err;
46bfbb5c
CM		 7328 }
7329 goto unlock;
4b46fce2 7330 }
4b46fce2 7331 }
00361589 7332
46bfbb5c
CM		 7333 /*
7334 * this will cow the extent, reset the len in case we changed
7335 * it above
7336 */
7337 len = bh_result->b_size;
70c8a91c
JB		 7338 free_extent_map(em);
7339 em = btrfs_new_extent_direct(inode, start, len);
eb838e73
JB		 7340 if (IS_ERR(em)) {
7341 ret = PTR_ERR(em);
7342 goto unlock_err;
7343 }
46bfbb5c
CM		 7344 len = min(len, em->len - (start - em->start));
7345unlock:
4b46fce2
JB		 7346 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7347 inode->i_blkbits;
46bfbb5c 7348 bh_result->b_size = len;
4b46fce2
JB		 7349 bh_result->b_bdev = em->bdev;
7350 set_buffer_mapped(bh_result);
c3473e83
JB		 7351 if (create) {
7352 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7353 set_buffer_new(bh_result);
7354
7355 /*
7356 * Need to update the i_size under the extent lock so buffered
7357 * readers will get the updated i_size when we unlock.
7358 */
7359 if (start + len > i_size_read(inode))
7360 i_size_write(inode, start + len);
0934856d 7361
3e05bde8
JB		 7362 if (len < orig_len) {
7363 spin_lock(&BTRFS_I(inode)->lock);
7364 BTRFS_I(inode)->outstanding_extents++;
7365 spin_unlock(&BTRFS_I(inode)->lock);
7366 }
3266789f 7367 btrfs_free_reserved_data_space(inode, len);
c3473e83 7368 }
4b46fce2 7369
eb838e73
JB		 7370 /*
7371 * In the case of write we need to clear and unlock the entire range,
7372 * in the case of read we need to unlock only the end area that we
7373 * aren't using if there is any left over space.
7374 */
24c03fa5 7375 if (lockstart < lockend) {
0934856d
MX		 7376 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7377 lockend, unlock_bits, 1, 0,
7378 &cached_state, GFP_NOFS);
24c03fa5 7379 } else {
eb838e73 7380 free_extent_state(cached_state);
24c03fa5 7381 }
eb838e73 7382
4b46fce2
JB		 7383 free_extent_map(em);
7384
7385 return 0;
eb838e73
JB		 7386
7387unlock_err:
eb838e73
JB		 7388 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7389 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7390 return ret;
4b46fce2
JB		 7391}
7392
8b110e39
MX		 7393static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7394 int rw, int mirror_num)
7395{
7396 struct btrfs_root *root = BTRFS_I(inode)->root;
7397 int ret;
7398
7399 BUG_ON(rw & REQ_WRITE);
7400
7401 bio_get(bio);
7402
7403 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7404 BTRFS_WQ_ENDIO_DIO_REPAIR);
7405 if (ret)
7406 goto err;
7407
7408 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7409err:
7410 bio_put(bio);
7411 return ret;
7412}
7413
7414static int btrfs_check_dio_repairable(struct inode *inode,
7415 struct bio *failed_bio,
7416 struct io_failure_record *failrec,
7417 int failed_mirror)
7418{
7419 int num_copies;
7420
7421 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7422 failrec->logical, failrec->len);
7423 if (num_copies == 1) {
7424 /*
7425 * we only have a single copy of the data, so don't bother with
7426 * all the retry and error correction code that follows. no
7427 * matter what the error is, it is very likely to persist.
7428 */
7429 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7430 num_copies, failrec->this_mirror, failed_mirror);
7431 return 0;
7432 }
7433
7434 failrec->failed_mirror = failed_mirror;
7435 failrec->this_mirror++;
7436 if (failrec->this_mirror == failed_mirror)
7437 failrec->this_mirror++;
7438
7439 if (failrec->this_mirror > num_copies) {
7440 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7441 num_copies, failrec->this_mirror, failed_mirror);
7442 return 0;
7443 }
7444
7445 return 1;
7446}
7447
7448static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7449 struct page *page, u64 start, u64 end,
7450 int failed_mirror, bio_end_io_t *repair_endio,
7451 void *repair_arg)
7452{
7453 struct io_failure_record *failrec;
7454 struct bio *bio;
7455 int isector;
7456 int read_mode;
7457 int ret;
7458
7459 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7460
7461 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7462 if (ret)
7463 return ret;
7464
7465 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7466 failed_mirror);
7467 if (!ret) {
7468 free_io_failure(inode, failrec);
7469 return -EIO;
7470 }
7471
7472 if (failed_bio->bi_vcnt > 1)
7473 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7474 else
7475 read_mode = READ_SYNC;
7476
7477 isector = start - btrfs_io_bio(failed_bio)->logical;
7478 isector >>= inode->i_sb->s_blocksize_bits;
7479 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7480 0, isector, repair_endio, repair_arg);
7481 if (!bio) {
7482 free_io_failure(inode, failrec);
7483 return -EIO;
7484 }
7485
7486 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7487 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7488 read_mode, failrec->this_mirror, failrec->in_validation);
7489
7490 ret = submit_dio_repair_bio(inode, bio, read_mode,
7491 failrec->this_mirror);
7492 if (ret) {
7493 free_io_failure(inode, failrec);
7494 bio_put(bio);
7495 }
7496
7497 return ret;
7498}
7499
7500struct btrfs_retry_complete {
7501 struct completion done;
7502 struct inode *inode;
7503 u64 start;
7504 int uptodate;
7505};
7506
7507static void btrfs_retry_endio_nocsum(struct bio *bio, int err)
7508{
7509 struct btrfs_retry_complete *done = bio->bi_private;
7510 struct bio_vec *bvec;
7511 int i;
7512
7513 if (err)
7514 goto end;
7515
7516 done->uptodate = 1;
7517 bio_for_each_segment_all(bvec, bio, i)
7518 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7519end:
7520 complete(&done->done);
7521 bio_put(bio);
7522}
7523
7524static int __btrfs_correct_data_nocsum(struct inode *inode,
7525 struct btrfs_io_bio *io_bio)
4b46fce2 7526{
2c30c71b 7527 struct bio_vec *bvec;
8b110e39 7528 struct btrfs_retry_complete done;
4b46fce2 7529 u64 start;
2c30c71b 7530 int i;
c1dc0896 7531 int ret;
4b46fce2 7532
8b110e39
MX		 7533 start = io_bio->logical;
7534 done.inode = inode;
7535
7536 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7537try_again:
7538 done.uptodate = 0;
7539 done.start = start;
7540 init_completion(&done.done);
7541
7542 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7543 start + bvec->bv_len - 1,
7544 io_bio->mirror_num,
7545 btrfs_retry_endio_nocsum, &done);
7546 if (ret)
7547 return ret;
7548
7549 wait_for_completion(&done.done);
7550
7551 if (!done.uptodate) {
7552 /* We might have another mirror, so try again */
7553 goto try_again;
7554 }
7555
7556 start += bvec->bv_len;
7557 }
7558
7559 return 0;
7560}
7561
7562static void btrfs_retry_endio(struct bio *bio, int err)
7563{
7564 struct btrfs_retry_complete *done = bio->bi_private;
7565 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7566 struct bio_vec *bvec;
7567 int uptodate;
7568 int ret;
7569 int i;
7570
7571 if (err)
7572 goto end;
7573
7574 uptodate = 1;
7575 bio_for_each_segment_all(bvec, bio, i) {
7576 ret = __readpage_endio_check(done->inode, io_bio, i,
7577 bvec->bv_page, 0,
7578 done->start, bvec->bv_len);
7579 if (!ret)
7580 clean_io_failure(done->inode, done->start,
7581 bvec->bv_page, 0);
7582 else
7583 uptodate = 0;
7584 }
7585
7586 done->uptodate = uptodate;
7587end:
7588 complete(&done->done);
7589 bio_put(bio);
7590}
7591
7592static int __btrfs_subio_endio_read(struct inode *inode,
7593 struct btrfs_io_bio *io_bio, int err)
7594{
7595 struct bio_vec *bvec;
7596 struct btrfs_retry_complete done;
7597 u64 start;
7598 u64 offset = 0;
7599 int i;
7600 int ret;
dc380aea 7601
8b110e39 7602 err = 0;
c1dc0896 7603 start = io_bio->logical;
8b110e39
MX		 7604 done.inode = inode;
7605
c1dc0896 7606 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
dc380aea
MX		 7607 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7608 0, start, bvec->bv_len);
8b110e39
MX		 7609 if (likely(!ret))
7610 goto next;
7611try_again:
7612 done.uptodate = 0;
7613 done.start = start;
7614 init_completion(&done.done);
7615
7616 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7617 start + bvec->bv_len - 1,
7618 io_bio->mirror_num,
7619 btrfs_retry_endio, &done);
7620 if (ret) {
7621 err = ret;
7622 goto next;
7623 }
7624
7625 wait_for_completion(&done.done);
7626
7627 if (!done.uptodate) {
7628 /* We might have another mirror, so try again */
7629 goto try_again;
7630 }
7631next:
7632 offset += bvec->bv_len;
4b46fce2 7633 start += bvec->bv_len;
2c30c71b 7634 }
c1dc0896
MX		 7635
7636 return err;
7637}
7638
8b110e39
MX		 7639static int btrfs_subio_endio_read(struct inode *inode,
7640 struct btrfs_io_bio *io_bio, int err)
7641{
7642 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7643
7644 if (skip_csum) {
7645 if (unlikely(err))
7646 return __btrfs_correct_data_nocsum(inode, io_bio);
7647 else
7648 return 0;
7649 } else {
7650 return __btrfs_subio_endio_read(inode, io_bio, err);
7651 }
7652}
7653
c1dc0896
MX		 7654static void btrfs_endio_direct_read(struct bio *bio, int err)
7655{
7656 struct btrfs_dio_private *dip = bio->bi_private;
7657 struct inode *inode = dip->inode;
7658 struct bio *dio_bio;
7659 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7660
8b110e39
MX		 7661 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
7662 err = btrfs_subio_endio_read(inode, io_bio, err);
c1dc0896 7663
4b46fce2 7664 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
d0082371 7665 dip->logical_offset + dip->bytes - 1);
9be3395b 7666 dio_bio = dip->dio_bio;
4b46fce2 7667
4b46fce2 7668 kfree(dip);
c0da7aa1
JB		 7669
7670 /* If we had a csum failure make sure to clear the uptodate flag */
7671 if (err)
9be3395b
CM		 7672 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7673 dio_end_io(dio_bio, err);
23ea8e5a
MX		 7674
7675 if (io_bio->end_io)
7676 io_bio->end_io(io_bio, err);
9be3395b 7677 bio_put(bio);
4b46fce2
JB		 7678}
7679
7680static void btrfs_endio_direct_write(struct bio *bio, int err)
7681{
7682 struct btrfs_dio_private *dip = bio->bi_private;
7683 struct inode *inode = dip->inode;
7684 struct btrfs_root *root = BTRFS_I(inode)->root;
4b46fce2 7685 struct btrfs_ordered_extent *ordered = NULL;
163cf09c
CM		 7686 u64 ordered_offset = dip->logical_offset;
7687 u64 ordered_bytes = dip->bytes;
9be3395b 7688 struct bio *dio_bio;
4b46fce2
JB		 7689 int ret;
7690
7691 if (err)
7692 goto out_done;
163cf09c
CM		 7693again:
7694 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7695 &ordered_offset,
5fd02043 7696 ordered_bytes, !err);
4b46fce2 7697 if (!ret)
163cf09c 7698 goto out_test;
4b46fce2 7699
9e0af237
LB		 7700 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
7701 finish_ordered_fn, NULL, NULL);
fccb5d86
QW		 7702 btrfs_queue_work(root->fs_info->endio_write_workers,
7703 &ordered->work);
163cf09c
CM		 7704out_test:
7705 /*
7706 * our bio might span multiple ordered extents. If we haven't
7707 * completed the accounting for the whole dio, go back and try again
7708 */
7709 if (ordered_offset < dip->logical_offset + dip->bytes) {
7710 ordered_bytes = dip->logical_offset + dip->bytes -
7711 ordered_offset;
5fd02043 7712 ordered = NULL;
163cf09c
CM		 7713 goto again;
7714 }
4b46fce2 7715out_done:
9be3395b 7716 dio_bio = dip->dio_bio;
4b46fce2 7717
4b46fce2 7718 kfree(dip);
c0da7aa1
JB		 7719
7720 /* If we had an error make sure to clear the uptodate flag */
7721 if (err)
9be3395b
CM		 7722 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7723 dio_end_io(dio_bio, err);
7724 bio_put(bio);
4b46fce2
JB		 7725}
7726
eaf25d93
CM		 7727static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7728 struct bio *bio, int mirror_num,
7729 unsigned long bio_flags, u64 offset)
7730{
7731 int ret;
7732 struct btrfs_root *root = BTRFS_I(inode)->root;
7733 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
79787eaa 7734 BUG_ON(ret); /* -ENOMEM */
eaf25d93
CM		 7735 return 0;
7736}
7737
e65e1535
MX		 7738static void btrfs_end_dio_bio(struct bio *bio, int err)
7739{
7740 struct btrfs_dio_private *dip = bio->bi_private;
7741
8b110e39
MX		 7742 if (err)
7743 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7744 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7745 btrfs_ino(dip->inode), bio->bi_rw,
7746 (unsigned long long)bio->bi_iter.bi_sector,
7747 bio->bi_iter.bi_size, err);
7748
7749 if (dip->subio_endio)
7750 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
c1dc0896
MX		 7751
7752 if (err) {
e65e1535
MX		 7753 dip->errors = 1;
7754
7755 /*
7756 * before atomic variable goto zero, we must make sure
7757 * dip->errors is perceived to be set.
7758 */
4e857c58 7759 smp_mb__before_atomic();
e65e1535
MX		 7760 }
7761
7762 /* if there are more bios still pending for this dio, just exit */
7763 if (!atomic_dec_and_test(&dip->pending_bios))
7764 goto out;
7765
9be3395b 7766 if (dip->errors) {
e65e1535 7767 bio_io_error(dip->orig_bio);
9be3395b
CM		 7768 } else {
7769 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
e65e1535
MX		 7770 bio_endio(dip->orig_bio, 0);
7771 }
7772out:
7773 bio_put(bio);
7774}
7775
7776static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7777 u64 first_sector, gfp_t gfp_flags)
7778{
7779 int nr_vecs = bio_get_nr_vecs(bdev);
7780 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7781}
7782
c1dc0896
MX		 7783static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
7784 struct inode *inode,
7785 struct btrfs_dio_private *dip,
7786 struct bio *bio,
7787 u64 file_offset)
7788{
7789 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7790 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
7791 int ret;
7792
7793 /*
7794 * We load all the csum data we need when we submit
7795 * the first bio to reduce the csum tree search and
7796 * contention.
7797 */
7798 if (dip->logical_offset == file_offset) {
7799 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
7800 file_offset);
7801 if (ret)
7802 return ret;
7803 }
7804
7805 if (bio == dip->orig_bio)
7806 return 0;
7807
7808 file_offset -= dip->logical_offset;
7809 file_offset >>= inode->i_sb->s_blocksize_bits;
7810 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
7811
7812 return 0;
7813}
7814
e65e1535
MX		 7815static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7816 int rw, u64 file_offset, int skip_sum,
c329861d 7817 int async_submit)
e65e1535 7818{
facc8a22 7819 struct btrfs_dio_private *dip = bio->bi_private;
e65e1535
MX		 7820 int write = rw & REQ_WRITE;
7821 struct btrfs_root *root = BTRFS_I(inode)->root;
7822 int ret;
7823
b812ce28
JB		 7824 if (async_submit)
7825 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7826
e65e1535 7827 bio_get(bio);
5fd02043
JB		 7828
7829 if (!write) {
bfebd8b5
DS		 7830 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7831 BTRFS_WQ_ENDIO_DATA);
5fd02043
JB		 7832 if (ret)
7833 goto err;
7834 }
e65e1535 7835
1ae39938
JB		 7836 if (skip_sum)
7837 goto map;
7838
7839 if (write && async_submit) {
e65e1535
MX		 7840 ret = btrfs_wq_submit_bio(root->fs_info,
7841 inode, rw, bio, 0, 0,
7842 file_offset,
7843 __btrfs_submit_bio_start_direct_io,
7844 __btrfs_submit_bio_done);
7845 goto err;
1ae39938
JB		 7846 } else if (write) {
7847 /*
7848 * If we aren't doing async submit, calculate the csum of the
7849 * bio now.
7850 */
7851 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7852 if (ret)
7853 goto err;
23ea8e5a 7854 } else {
c1dc0896
MX		 7855 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
7856 file_offset);
c2db1073
TI		 7857 if (ret)
7858 goto err;
7859 }
1ae39938
JB		 7860map:
7861 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
e65e1535
MX		 7862err:
7863 bio_put(bio);
7864 return ret;
7865}
7866
7867static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7868 int skip_sum)
7869{
7870 struct inode *inode = dip->inode;
7871 struct btrfs_root *root = BTRFS_I(inode)->root;
e65e1535
MX		 7872 struct bio *bio;
7873 struct bio *orig_bio = dip->orig_bio;
7874 struct bio_vec *bvec = orig_bio->bi_io_vec;
4f024f37 7875 u64 start_sector = orig_bio->bi_iter.bi_sector;
e65e1535
MX		 7876 u64 file_offset = dip->logical_offset;
7877 u64 submit_len = 0;
7878 u64 map_length;
7879 int nr_pages = 0;
23ea8e5a 7880 int ret;
1ae39938 7881 int async_submit = 0;
e65e1535 7882
4f024f37 7883 map_length = orig_bio->bi_iter.bi_size;
53b381b3 7884 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
e65e1535 7885 &map_length, NULL, 0);
7a5c3c9b 7886 if (ret)
e65e1535 7887 return -EIO;
facc8a22 7888
4f024f37 7889 if (map_length >= orig_bio->bi_iter.bi_size) {
02f57c7a 7890 bio = orig_bio;
c1dc0896 7891 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
02f57c7a
JB		 7892 goto submit;
7893 }
7894
53b381b3 7895 /* async crcs make it difficult to collect full stripe writes. */
ffe2d203 7896 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
53b381b3
DW		 7897 async_submit = 0;
7898 else
7899 async_submit = 1;
7900
02f57c7a
JB		 7901 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7902 if (!bio)
7903 return -ENOMEM;
7a5c3c9b 7904
02f57c7a
JB		 7905 bio->bi_private = dip;
7906 bio->bi_end_io = btrfs_end_dio_bio;
c1dc0896 7907 btrfs_io_bio(bio)->logical = file_offset;
02f57c7a
JB		 7908 atomic_inc(&dip->pending_bios);
7909
e65e1535 7910 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
ee39b432 7911 if (map_length < submit_len + bvec->bv_len ||
e65e1535 7912 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
ee39b432 7913 bvec->bv_offset) < bvec->bv_len) {
e65e1535
MX		 7914 /*
7915 * inc the count before we submit the bio so
7916 * we know the end IO handler won't happen before
7917 * we inc the count. Otherwise, the dip might get freed
7918 * before we're done setting it up
7919 */
7920 atomic_inc(&dip->pending_bios);
7921 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7922 file_offset, skip_sum,
c329861d 7923 async_submit);
e65e1535
MX		 7924 if (ret) {
7925 bio_put(bio);
7926 atomic_dec(&dip->pending_bios);
7927 goto out_err;
7928 }
7929
e65e1535
MX		 7930 start_sector += submit_len >> 9;
7931 file_offset += submit_len;
7932
7933 submit_len = 0;
7934 nr_pages = 0;
7935
7936 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7937 start_sector, GFP_NOFS);
7938 if (!bio)
7939 goto out_err;
7940 bio->bi_private = dip;
7941 bio->bi_end_io = btrfs_end_dio_bio;
c1dc0896 7942 btrfs_io_bio(bio)->logical = file_offset;
e65e1535 7943
4f024f37 7944 map_length = orig_bio->bi_iter.bi_size;
53b381b3 7945 ret = btrfs_map_block(root->fs_info, rw,
3ec706c8 7946 start_sector << 9,
e65e1535
MX		 7947 &map_length, NULL, 0);
7948 if (ret) {
7949 bio_put(bio);
7950 goto out_err;
7951 }
7952 } else {
7953 submit_len += bvec->bv_len;
67871254 7954 nr_pages++;
e65e1535
MX		 7955 bvec++;
7956 }
7957 }
7958
02f57c7a 7959submit:
e65e1535 7960 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
c329861d 7961 async_submit);
e65e1535
MX		 7962 if (!ret)
7963 return 0;
7964
7965 bio_put(bio);
7966out_err:
7967 dip->errors = 1;
7968 /*
7969 * before atomic variable goto zero, we must
7970 * make sure dip->errors is perceived to be set.
7971 */
4e857c58 7972 smp_mb__before_atomic();
e65e1535
MX		 7973 if (atomic_dec_and_test(&dip->pending_bios))
7974 bio_io_error(dip->orig_bio);
7975
7976 /* bio_end_io() will handle error, so we needn't return it */
7977 return 0;
7978}
7979
9be3395b
CM		 7980static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7981 struct inode *inode, loff_t file_offset)
4b46fce2
JB		 7982{
7983 struct btrfs_root *root = BTRFS_I(inode)->root;
7984 struct btrfs_dio_private *dip;
9be3395b 7985 struct bio *io_bio;
23ea8e5a 7986 struct btrfs_io_bio *btrfs_bio;
4b46fce2 7987 int skip_sum;
7b6d91da 7988 int write = rw & REQ_WRITE;
4b46fce2
JB		 7989 int ret = 0;
7990
7991 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7992
9be3395b 7993 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
9be3395b
CM		 7994 if (!io_bio) {
7995 ret = -ENOMEM;
7996 goto free_ordered;
7997 }
7998
c1dc0896 7999 dip = kzalloc(sizeof(*dip), GFP_NOFS);
4b46fce2
JB		 8000 if (!dip) {
8001 ret = -ENOMEM;
9be3395b 8002 goto free_io_bio;
4b46fce2 8003 }
4b46fce2 8004
9be3395b 8005 dip->private = dio_bio->bi_private;
4b46fce2
JB		 8006 dip->inode = inode;
8007 dip->logical_offset = file_offset;
4f024f37
KO		 8008 dip->bytes = dio_bio->bi_iter.bi_size;
8009 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
9be3395b 8010 io_bio->bi_private = dip;
9be3395b
CM		 8011 dip->orig_bio = io_bio;
8012 dip->dio_bio = dio_bio;
e65e1535 8013 atomic_set(&dip->pending_bios, 0);
c1dc0896
MX		 8014 btrfs_bio = btrfs_io_bio(io_bio);
8015 btrfs_bio->logical = file_offset;
4b46fce2 8016
c1dc0896 8017 if (write) {
9be3395b 8018 io_bio->bi_end_io = btrfs_endio_direct_write;
c1dc0896 8019 } else {
9be3395b 8020 io_bio->bi_end_io = btrfs_endio_direct_read;
c1dc0896
MX		 8021 dip->subio_endio = btrfs_subio_endio_read;
8022 }
4b46fce2 8023
e65e1535
MX		 8024 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8025 if (!ret)
eaf25d93 8026 return;
9be3395b 8027
23ea8e5a
MX		 8028 if (btrfs_bio->end_io)
8029 btrfs_bio->end_io(btrfs_bio, ret);
9be3395b
CM		 8030free_io_bio:
8031 bio_put(io_bio);
8032
4b46fce2
JB		 8033free_ordered:
8034 /*
8035 * If this is a write, we need to clean up the reserved space and kill
8036 * the ordered extent.
8037 */
8038 if (write) {
8039 struct btrfs_ordered_extent *ordered;
955256f2 8040 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
4b46fce2
JB		 8041 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
8042 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
8043 btrfs_free_reserved_extent(root, ordered->start,
e570fd27 8044 ordered->disk_len, 1);
4b46fce2
JB		 8045 btrfs_put_ordered_extent(ordered);
8046 btrfs_put_ordered_extent(ordered);
8047 }
9be3395b 8048 bio_endio(dio_bio, ret);
4b46fce2
JB		 8049}
8050
5a5f79b5 8051static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
28060d5d 8052 const struct iov_iter *iter, loff_t offset)
5a5f79b5
CM		 8053{
8054 int seg;
a1b75f7d 8055 int i;
5a5f79b5
CM		 8056 unsigned blocksize_mask = root->sectorsize - 1;
8057 ssize_t retval = -EINVAL;
5a5f79b5
CM		 8058
8059 if (offset & blocksize_mask)
8060 goto out;
8061
28060d5d
AV		 8062 if (iov_iter_alignment(iter) & blocksize_mask)
8063 goto out;
a1b75f7d 8064
28060d5d
AV		 8065 /* If this is a write we don't need to check anymore */
8066 if (rw & WRITE)
8067 return 0;
8068 /*
8069 * Check to make sure we don't have duplicate iov_base's in this
8070 * iovec, if so return EINVAL, otherwise we'll get csum errors
8071 * when reading back.
8072 */
8073 for (seg = 0; seg < iter->nr_segs; seg++) {
8074 for (i = seg + 1; i < iter->nr_segs; i++) {
8075 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
a1b75f7d
JB		 8076 goto out;
8077 }
5a5f79b5
CM		 8078 }
8079 retval = 0;
8080out:
8081 return retval;
8082}
eb838e73 8083
16432985 8084static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
d8d3d94b 8085 struct iov_iter *iter, loff_t offset)
16432985 8086{
4b46fce2
JB		 8087 struct file *file = iocb->ki_filp;
8088 struct inode *inode = file->f_mapping->host;
0934856d 8089 size_t count = 0;
2e60a51e 8090 int flags = 0;
38851cc1
MX		 8091 bool wakeup = true;
8092 bool relock = false;
0934856d 8093 ssize_t ret;
4b46fce2 8094
28060d5d 8095 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iter, offset))
5a5f79b5 8096 return 0;
3f7c579c 8097
38851cc1 8098 atomic_inc(&inode->i_dio_count);
4e857c58 8099 smp_mb__after_atomic();
38851cc1 8100
0e267c44 8101 /*
41bd9ca4
MX		 8102 * The generic stuff only does filemap_write_and_wait_range, which
8103 * isn't enough if we've written compressed pages to this area, so
8104 * we need to flush the dirty pages again to make absolutely sure
8105 * that any outstanding dirty pages are on disk.
0e267c44 8106 */
a6cbcd4a 8107 count = iov_iter_count(iter);
41bd9ca4
MX		 8108 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8109 &BTRFS_I(inode)->runtime_flags))
9a025a08
WS		 8110 filemap_fdatawrite_range(inode->i_mapping, offset,
8111 offset + count - 1);
0e267c44 8112
0934856d 8113 if (rw & WRITE) {
38851cc1
MX		 8114 /*
8115 * If the write DIO is beyond the EOF, we need update
8116 * the isize, but it is protected by i_mutex. So we can
8117 * not unlock the i_mutex at this case.
8118 */
8119 if (offset + count <= inode->i_size) {
8120 mutex_unlock(&inode->i_mutex);
8121 relock = true;
8122 }
0934856d
MX		 8123 ret = btrfs_delalloc_reserve_space(inode, count);
8124 if (ret)
38851cc1 8125 goto out;
ee39b432
DS		 8126 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8127 &BTRFS_I(inode)->runtime_flags)) {
38851cc1
MX		 8128 inode_dio_done(inode);
8129 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8130 wakeup = false;
0934856d
MX		 8131 }
8132
8133 ret = __blockdev_direct_IO(rw, iocb, inode,
8134 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
31b14039 8135 iter, offset, btrfs_get_blocks_direct, NULL,
2e60a51e 8136 btrfs_submit_direct, flags);
0934856d
MX		 8137 if (rw & WRITE) {
8138 if (ret < 0 && ret != -EIOCBQUEUED)
8139 btrfs_delalloc_release_space(inode, count);
172a5049 8140 else if (ret >= 0 && (size_t)ret < count)
0934856d
MX		 8141 btrfs_delalloc_release_space(inode,
8142 count - (size_t)ret);
0934856d 8143 }
38851cc1 8144out:
2e60a51e
MX		 8145 if (wakeup)
8146 inode_dio_done(inode);
38851cc1
MX		 8147 if (relock)
8148 mutex_lock(&inode->i_mutex);
0934856d
MX		 8149
8150 return ret;
16432985
CM		 8151}
8152
05dadc09
TI		 8153#define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8154
1506fcc8
YS		 8155static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8156 __u64 start, __u64 len)
8157{
05dadc09
TI		 8158 int ret;
8159
8160 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8161 if (ret)
8162 return ret;
8163
ec29ed5b 8164 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
1506fcc8
YS		 8165}
8166
a52d9a80 8167int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 8168{
d1310b2e
CM		 8169 struct extent_io_tree *tree;
8170 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 8171 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
9ebefb18 8172}
1832a6d5 8173
a52d9a80 8174static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 8175{
d1310b2e 8176 struct extent_io_tree *tree;
b888db2b
CM		 8177
8178
8179 if (current->flags & PF_MEMALLOC) {
8180 redirty_page_for_writepage(wbc, page);
8181 unlock_page(page);
8182 return 0;
8183 }
d1310b2e 8184 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 8185 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM		 8186}
8187
48a3b636
ES		 8188static int btrfs_writepages(struct address_space *mapping,
8189 struct writeback_control *wbc)
b293f02e 8190{
d1310b2e 8191 struct extent_io_tree *tree;
771ed689 8192
d1310b2e 8193 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 8194 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8195}
8196
3ab2fb5a
CM		 8197static int
8198btrfs_readpages(struct file *file, struct address_space *mapping,
8199 struct list_head *pages, unsigned nr_pages)
8200{
d1310b2e
CM		 8201 struct extent_io_tree *tree;
8202 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 8203 return extent_readpages(tree, mapping, pages, nr_pages,
8204 btrfs_get_extent);
8205}
e6dcd2dc 8206static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 8207{
d1310b2e
CM		 8208 struct extent_io_tree *tree;
8209 struct extent_map_tree *map;
a52d9a80 8210 int ret;
8c2383c3 8211
d1310b2e
CM		 8212 tree = &BTRFS_I(page->mapping->host)->io_tree;
8213 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 8214 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 8215 if (ret == 1) {
8216 ClearPagePrivate(page);
8217 set_page_private(page, 0);
8218 page_cache_release(page);
39279cc3 8219 }
a52d9a80 8220 return ret;
39279cc3
CM		 8221}
8222
e6dcd2dc
CM		 8223static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8224{
98509cfc
CM		 8225 if (PageWriteback(page) || PageDirty(page))
8226 return 0;
b335b003 8227 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
e6dcd2dc
CM		 8228}
8229
d47992f8
LC		 8230static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8231 unsigned int length)
39279cc3 8232{
5fd02043 8233 struct inode *inode = page->mapping->host;
d1310b2e 8234 struct extent_io_tree *tree;
e6dcd2dc 8235 struct btrfs_ordered_extent *ordered;
2ac55d41 8236 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 8237 u64 page_start = page_offset(page);
8238 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
131e404a 8239 int inode_evicting = inode->i_state & I_FREEING;
39279cc3 8240
8b62b72b
CM		 8241 /*
8242 * we have the page locked, so new writeback can't start,
8243 * and the dirty bit won't be cleared while we are here.
8244 *
8245 * Wait for IO on this page so that we can safely clear
8246 * the PagePrivate2 bit and do ordered accounting
8247 */
e6dcd2dc 8248 wait_on_page_writeback(page);
8b62b72b 8249
5fd02043 8250 tree = &BTRFS_I(inode)->io_tree;
e6dcd2dc
CM		 8251 if (offset) {
8252 btrfs_releasepage(page, GFP_NOFS);
8253 return;
8254 }
131e404a
FDBM		 8255
8256 if (!inode_evicting)
8257 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
8258 ordered = btrfs_lookup_ordered_extent(inode, page_start);
e6dcd2dc 8259 if (ordered) {
eb84ae03
CM		 8260 /*
8261 * IO on this page will never be started, so we need
8262 * to account for any ordered extents now
8263 */
131e404a
FDBM		 8264 if (!inode_evicting)
8265 clear_extent_bit(tree, page_start, page_end,
8266 EXTENT_DIRTY | EXTENT_DELALLOC |
8267 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8268 EXTENT_DEFRAG, 1, 0, &cached_state,
8269 GFP_NOFS);
8b62b72b
CM		 8270 /*
8271 * whoever cleared the private bit is responsible
8272 * for the finish_ordered_io
8273 */
77cef2ec
JB		 8274 if (TestClearPagePrivate2(page)) {
8275 struct btrfs_ordered_inode_tree *tree;
8276 u64 new_len;
8277
8278 tree = &BTRFS_I(inode)->ordered_tree;
8279
8280 spin_lock_irq(&tree->lock);
8281 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8282 new_len = page_start - ordered->file_offset;
8283 if (new_len < ordered->truncated_len)
8284 ordered->truncated_len = new_len;
8285 spin_unlock_irq(&tree->lock);
8286
8287 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8288 page_start,
8289 PAGE_CACHE_SIZE, 1))
8290 btrfs_finish_ordered_io(ordered);
8b62b72b 8291 }
e6dcd2dc 8292 btrfs_put_ordered_extent(ordered);
131e404a
FDBM		 8293 if (!inode_evicting) {
8294 cached_state = NULL;
8295 lock_extent_bits(tree, page_start, page_end, 0,
8296 &cached_state);
8297 }
8298 }
8299
8300 if (!inode_evicting) {
8301 clear_extent_bit(tree, page_start, page_end,
8302 EXTENT_LOCKED | EXTENT_DIRTY |
8303 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8304 EXTENT_DEFRAG, 1, 1,
8305 &cached_state, GFP_NOFS);
8306
8307 __btrfs_releasepage(page, GFP_NOFS);
e6dcd2dc 8308 }
e6dcd2dc 8309
4a096752 8310 ClearPageChecked(page);
9ad6b7bc 8311 if (PagePrivate(page)) {
9ad6b7bc
CM		 8312 ClearPagePrivate(page);
8313 set_page_private(page, 0);
8314 page_cache_release(page);
8315 }
39279cc3
CM		 8316}
8317
9ebefb18
CM		 8318/*
8319 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8320 * called from a page fault handler when a page is first dirtied. Hence we must
8321 * be careful to check for EOF conditions here. We set the page up correctly
8322 * for a written page which means we get ENOSPC checking when writing into
8323 * holes and correct delalloc and unwritten extent mapping on filesystems that
8324 * support these features.
8325 *
8326 * We are not allowed to take the i_mutex here so we have to play games to
8327 * protect against truncate races as the page could now be beyond EOF. Because
8328 * vmtruncate() writes the inode size before removing pages, once we have the
8329 * page lock we can determine safely if the page is beyond EOF. If it is not
8330 * beyond EOF, then the page is guaranteed safe against truncation until we
8331 * unlock the page.
8332 */
c2ec175c 8333int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
9ebefb18 8334{
c2ec175c 8335 struct page *page = vmf->page;
496ad9aa 8336 struct inode *inode = file_inode(vma->vm_file);
1832a6d5 8337 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 8338 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8339 struct btrfs_ordered_extent *ordered;
2ac55d41 8340 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 8341 char *kaddr;
8342 unsigned long zero_start;
9ebefb18 8343 loff_t size;
1832a6d5 8344 int ret;
9998eb70 8345 int reserved = 0;
a52d9a80 8346 u64 page_start;
e6dcd2dc 8347 u64 page_end;
9ebefb18 8348
b2b5ef5c 8349 sb_start_pagefault(inode->i_sb);
0ca1f7ce 8350 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
9998eb70 8351 if (!ret) {
e41f941a 8352 ret = file_update_time(vma->vm_file);
9998eb70
CM		 8353 reserved = 1;
8354 }
56a76f82
NP		 8355 if (ret) {
8356 if (ret == -ENOMEM)
8357 ret = VM_FAULT_OOM;
8358 else /* -ENOSPC, -EIO, etc */
8359 ret = VM_FAULT_SIGBUS;
9998eb70
CM		 8360 if (reserved)
8361 goto out;
8362 goto out_noreserve;
56a76f82 8363 }
1832a6d5 8364
56a76f82 8365 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
e6dcd2dc 8366again:
9ebefb18 8367 lock_page(page);
9ebefb18 8368 size = i_size_read(inode);
e6dcd2dc
CM		 8369 page_start = page_offset(page);
8370 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 8371
9ebefb18 8372 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 8373 (page_start >= size)) {
9ebefb18
CM		 8374 /* page got truncated out from underneath us */
8375 goto out_unlock;
8376 }
e6dcd2dc
CM		 8377 wait_on_page_writeback(page);
8378
d0082371 8379 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 8380 set_page_extent_mapped(page);
8381
eb84ae03
CM		 8382 /*
8383 * we can't set the delalloc bits if there are pending ordered
8384 * extents. Drop our locks and wait for them to finish
8385 */
e6dcd2dc
CM		 8386 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8387 if (ordered) {
2ac55d41
JB		 8388 unlock_extent_cached(io_tree, page_start, page_end,
8389 &cached_state, GFP_NOFS);
e6dcd2dc 8390 unlock_page(page);
eb84ae03 8391 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 8392 btrfs_put_ordered_extent(ordered);
8393 goto again;
8394 }
8395
fbf19087
JB		 8396 /*
8397 * XXX - page_mkwrite gets called every time the page is dirtied, even
8398 * if it was already dirty, so for space accounting reasons we need to
8399 * clear any delalloc bits for the range we are fixing to save. There
8400 * is probably a better way to do this, but for now keep consistent with
8401 * prepare_pages in the normal write path.
8402 */
2ac55d41 8403 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 8404 EXTENT_DIRTY | EXTENT_DELALLOC |
8405 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 8406 0, 0, &cached_state, GFP_NOFS);
fbf19087 8407
2ac55d41
JB		 8408 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
8409 &cached_state);
9ed74f2d 8410 if (ret) {
2ac55d41
JB		 8411 unlock_extent_cached(io_tree, page_start, page_end,
8412 &cached_state, GFP_NOFS);
9ed74f2d
JB		 8413 ret = VM_FAULT_SIGBUS;
8414 goto out_unlock;
8415 }
e6dcd2dc 8416 ret = 0;
9ebefb18
CM		 8417
8418 /* page is wholly or partially inside EOF */
a52d9a80 8419 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 8420 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 8421 else
e6dcd2dc 8422 zero_start = PAGE_CACHE_SIZE;
9ebefb18 8423
e6dcd2dc
CM		 8424 if (zero_start != PAGE_CACHE_SIZE) {
8425 kaddr = kmap(page);
8426 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
8427 flush_dcache_page(page);
8428 kunmap(page);
8429 }
247e743c 8430 ClearPageChecked(page);
e6dcd2dc 8431 set_page_dirty(page);
50a9b214 8432 SetPageUptodate(page);
5a3f23d5 8433
257c62e1
CM		 8434 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8435 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
46d8bc34 8436 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
257c62e1 8437
2ac55d41 8438 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9ebefb18
CM		 8439
8440out_unlock:
b2b5ef5c
JK		 8441 if (!ret) {
8442 sb_end_pagefault(inode->i_sb);
50a9b214 8443 return VM_FAULT_LOCKED;
b2b5ef5c 8444 }
9ebefb18 8445 unlock_page(page);
1832a6d5 8446out:
ec39e180 8447 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
9998eb70 8448out_noreserve:
b2b5ef5c 8449 sb_end_pagefault(inode->i_sb);
9ebefb18
CM		 8450 return ret;
8451}
8452
a41ad394 8453static int btrfs_truncate(struct inode *inode)
39279cc3
CM		 8454{
8455 struct btrfs_root *root = BTRFS_I(inode)->root;
fcb80c2a 8456 struct btrfs_block_rsv *rsv;
a71754fc 8457 int ret = 0;
3893e33b 8458 int err = 0;
39279cc3 8459 struct btrfs_trans_handle *trans;
dbe674a9 8460 u64 mask = root->sectorsize - 1;
07127184 8461 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3 8462
0ef8b726
JB		 8463 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8464 (u64)-1);
8465 if (ret)
8466 return ret;
39279cc3 8467
fcb80c2a
JB		 8468 /*
8469 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8470 * 3 things going on here
8471 *
8472 * 1) We need to reserve space for our orphan item and the space to
8473 * delete our orphan item. Lord knows we don't want to have a dangling
8474 * orphan item because we didn't reserve space to remove it.
8475 *
8476 * 2) We need to reserve space to update our inode.
8477 *
8478 * 3) We need to have something to cache all the space that is going to
8479 * be free'd up by the truncate operation, but also have some slack
8480 * space reserved in case it uses space during the truncate (thank you
8481 * very much snapshotting).
8482 *
8483 * And we need these to all be seperate. The fact is we can use alot of
8484 * space doing the truncate, and we have no earthly idea how much space
8485 * we will use, so we need the truncate reservation to be seperate so it
8486 * doesn't end up using space reserved for updating the inode or
8487 * removing the orphan item. We also need to be able to stop the
8488 * transaction and start a new one, which means we need to be able to
8489 * update the inode several times, and we have no idea of knowing how
8490 * many times that will be, so we can't just reserve 1 item for the
8491 * entirety of the opration, so that has to be done seperately as well.
8492 * Then there is the orphan item, which does indeed need to be held on
8493 * to for the whole operation, and we need nobody to touch this reserved
8494 * space except the orphan code.
8495 *
8496 * So that leaves us with
8497 *
8498 * 1) root->orphan_block_rsv - for the orphan deletion.
8499 * 2) rsv - for the truncate reservation, which we will steal from the
8500 * transaction reservation.
8501 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8502 * updating the inode.
8503 */
66d8f3dd 8504 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
fcb80c2a
JB		 8505 if (!rsv)
8506 return -ENOMEM;
4a338542 8507 rsv->size = min_size;
ca7e70f5 8508 rsv->failfast = 1;
f0cd846e 8509
907cbceb 8510 /*
07127184 8511 * 1 for the truncate slack space
907cbceb
JB		 8512 * 1 for updating the inode.
8513 */
f3fe820c 8514 trans = btrfs_start_transaction(root, 2);
fcb80c2a
JB		 8515 if (IS_ERR(trans)) {
8516 err = PTR_ERR(trans);
8517 goto out;
8518 }
f0cd846e 8519
907cbceb
JB		 8520 /* Migrate the slack space for the truncate to our reserve */
8521 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8522 min_size);
fcb80c2a 8523 BUG_ON(ret);
f0cd846e 8524
5dc562c5
JB		 8525 /*
8526 * So if we truncate and then write and fsync we normally would just
8527 * write the extents that changed, which is a problem if we need to
8528 * first truncate that entire inode. So set this flag so we write out
8529 * all of the extents in the inode to the sync log so we're completely
8530 * safe.
8531 */
8532 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
ca7e70f5 8533 trans->block_rsv = rsv;
907cbceb 8534
8082510e
YZ		 8535 while (1) {
8536 ret = btrfs_truncate_inode_items(trans, root, inode,
8537 inode->i_size,
8538 BTRFS_EXTENT_DATA_KEY);
ca7e70f5 8539 if (ret != -ENOSPC) {
3893e33b 8540 err = ret;
8082510e 8541 break;
3893e33b 8542 }
39279cc3 8543
fcb80c2a 8544 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e 8545 ret = btrfs_update_inode(trans, root, inode);
3893e33b
JB		 8546 if (ret) {
8547 err = ret;
8548 break;
8549 }
ca7e70f5 8550
8082510e 8551 btrfs_end_transaction(trans, root);
b53d3f5d 8552 btrfs_btree_balance_dirty(root);
ca7e70f5
JB		 8553
8554 trans = btrfs_start_transaction(root, 2);
8555 if (IS_ERR(trans)) {
8556 ret = err = PTR_ERR(trans);
8557 trans = NULL;
8558 break;
8559 }
8560
8561 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
8562 rsv, min_size);
8563 BUG_ON(ret); /* shouldn't happen */
8564 trans->block_rsv = rsv;
8082510e
YZ		 8565 }
8566
8567 if (ret == 0 && inode->i_nlink > 0) {
fcb80c2a 8568 trans->block_rsv = root->orphan_block_rsv;
8082510e 8569 ret = btrfs_orphan_del(trans, inode);
3893e33b
JB		 8570 if (ret)
8571 err = ret;
8082510e
YZ		 8572 }
8573
917c16b2
CM		 8574 if (trans) {
8575 trans->block_rsv = &root->fs_info->trans_block_rsv;
8576 ret = btrfs_update_inode(trans, root, inode);
8577 if (ret && !err)
8578 err = ret;
7b128766 8579
7ad85bb7 8580 ret = btrfs_end_transaction(trans, root);
b53d3f5d 8581 btrfs_btree_balance_dirty(root);
917c16b2 8582 }
fcb80c2a
JB		 8583
8584out:
8585 btrfs_free_block_rsv(root, rsv);
8586
3893e33b
JB		 8587 if (ret && !err)
8588 err = ret;
a41ad394 8589
3893e33b 8590 return err;
39279cc3
CM		 8591}
8592
d352ac68
CM		 8593/*
8594 * create a new subvolume directory/inode (helper for the ioctl).
8595 */
d2fb3437 8596int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
63541927
FDBM		 8597 struct btrfs_root *new_root,
8598 struct btrfs_root *parent_root,
8599 u64 new_dirid)
39279cc3 8600{
39279cc3 8601 struct inode *inode;
76dda93c 8602 int err;
00e4e6b3 8603 u64 index = 0;
39279cc3 8604
12fc9d09
FA		 8605 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8606 new_dirid, new_dirid,
8607 S_IFDIR | (~current_umask() & S_IRWXUGO),
8608 &index);
54aa1f4d 8609 if (IS_ERR(inode))
f46b5a66 8610 return PTR_ERR(inode);
39279cc3
CM		 8611 inode->i_op = &btrfs_dir_inode_operations;
8612 inode->i_fop = &btrfs_dir_file_operations;
8613
bfe86848 8614 set_nlink(inode, 1);
dbe674a9 8615 btrfs_i_size_write(inode, 0);
b0d5d10f 8616 unlock_new_inode(inode);
3b96362c 8617
63541927
FDBM		 8618 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8619 if (err)
8620 btrfs_err(new_root->fs_info,
351fd353 8621 "error inheriting subvolume %llu properties: %d",
63541927
FDBM		 8622 new_root->root_key.objectid, err);
8623
76dda93c 8624 err = btrfs_update_inode(trans, new_root, inode);
cb8e7090 8625
76dda93c 8626 iput(inode);
ce598979 8627 return err;
39279cc3
CM		 8628}
8629
39279cc3
CM		 8630struct inode *btrfs_alloc_inode(struct super_block *sb)
8631{
8632 struct btrfs_inode *ei;
2ead6ae7 8633 struct inode *inode;
39279cc3
CM		 8634
8635 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
8636 if (!ei)
8637 return NULL;
2ead6ae7
YZ		 8638
8639 ei->root = NULL;
2ead6ae7 8640 ei->generation = 0;
15ee9bc7 8641 ei->last_trans = 0;
257c62e1 8642 ei->last_sub_trans = 0;
e02119d5 8643 ei->logged_trans = 0;
2ead6ae7 8644 ei->delalloc_bytes = 0;
47059d93 8645 ei->defrag_bytes = 0;
2ead6ae7
YZ		 8646 ei->disk_i_size = 0;
8647 ei->flags = 0;
7709cde3 8648 ei->csum_bytes = 0;
2ead6ae7 8649 ei->index_cnt = (u64)-1;
67de1176 8650 ei->dir_index = 0;
2ead6ae7 8651 ei->last_unlink_trans = 0;
46d8bc34 8652 ei->last_log_commit = 0;
2ead6ae7 8653
9e0baf60
JB		 8654 spin_lock_init(&ei->lock);
8655 ei->outstanding_extents = 0;
8656 ei->reserved_extents = 0;
2ead6ae7 8657
72ac3c0d 8658 ei->runtime_flags = 0;
261507a0 8659 ei->force_compress = BTRFS_COMPRESS_NONE;
2ead6ae7 8660
16cdcec7
MX		 8661 ei->delayed_node = NULL;
8662
9cc97d64 8663 ei->i_otime.tv_sec = 0;
8664 ei->i_otime.tv_nsec = 0;
8665
2ead6ae7 8666 inode = &ei->vfs_inode;
a8067e02 8667 extent_map_tree_init(&ei->extent_tree);
f993c883
DS		 8668 extent_io_tree_init(&ei->io_tree, &inode->i_data);
8669 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
0b32f4bb
JB		 8670 ei->io_tree.track_uptodate = 1;
8671 ei->io_failure_tree.track_uptodate = 1;
b812ce28 8672 atomic_set(&ei->sync_writers, 0);
2ead6ae7 8673 mutex_init(&ei->log_mutex);
f248679e 8674 mutex_init(&ei->delalloc_mutex);
e6dcd2dc 8675 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
2ead6ae7 8676 INIT_LIST_HEAD(&ei->delalloc_inodes);
2ead6ae7
YZ		 8677 RB_CLEAR_NODE(&ei->rb_node);
8678
8679 return inode;
39279cc3
CM		 8680}
8681
aaedb55b
JB		 8682#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8683void btrfs_test_destroy_inode(struct inode *inode)
8684{
8685 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8686 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8687}
8688#endif
8689
fa0d7e3d
NP		 8690static void btrfs_i_callback(struct rcu_head *head)
8691{
8692 struct inode *inode = container_of(head, struct inode, i_rcu);
fa0d7e3d
NP		 8693 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8694}
8695
39279cc3
CM		 8696void btrfs_destroy_inode(struct inode *inode)
8697{
e6dcd2dc 8698 struct btrfs_ordered_extent *ordered;
5a3f23d5
CM		 8699 struct btrfs_root *root = BTRFS_I(inode)->root;
8700
b3d9b7a3 8701 WARN_ON(!hlist_empty(&inode->i_dentry));
39279cc3 8702 WARN_ON(inode->i_data.nrpages);
9e0baf60
JB		 8703 WARN_ON(BTRFS_I(inode)->outstanding_extents);
8704 WARN_ON(BTRFS_I(inode)->reserved_extents);
7709cde3
JB		 8705 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8706 WARN_ON(BTRFS_I(inode)->csum_bytes);
47059d93 8707 WARN_ON(BTRFS_I(inode)->defrag_bytes);
39279cc3 8708
a6dbd429
JB		 8709 /*
8710 * This can happen where we create an inode, but somebody else also
8711 * created the same inode and we need to destroy the one we already
8712 * created.
8713 */
8714 if (!root)
8715 goto free;
8716
8a35d95f
JB		 8717 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8718 &BTRFS_I(inode)->runtime_flags)) {
c2cf52eb 8719 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
c1c9ff7c 8720 btrfs_ino(inode));
8a35d95f 8721 atomic_dec(&root->orphan_inodes);
7b128766 8722 }
7b128766 8723
d397712b 8724 while (1) {
e6dcd2dc
CM		 8725 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8726 if (!ordered)
8727 break;
8728 else {
c2cf52eb 8729 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
c1c9ff7c 8730 ordered->file_offset, ordered->len);
e6dcd2dc
CM		 8731 btrfs_remove_ordered_extent(inode, ordered);
8732 btrfs_put_ordered_extent(ordered);
8733 btrfs_put_ordered_extent(ordered);
8734 }
8735 }
5d4f98a2 8736 inode_tree_del(inode);
5b21f2ed 8737 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
a6dbd429 8738free:
fa0d7e3d 8739 call_rcu(&inode->i_rcu, btrfs_i_callback);
39279cc3
CM		 8740}
8741
45321ac5 8742int btrfs_drop_inode(struct inode *inode)
76dda93c
YZ		 8743{
8744 struct btrfs_root *root = BTRFS_I(inode)->root;
45321ac5 8745
6379ef9f
NA		 8746 if (root == NULL)
8747 return 1;
8748
fa6ac876 8749 /* the snap/subvol tree is on deleting */
69e9c6c6 8750 if (btrfs_root_refs(&root->root_item) == 0)
45321ac5 8751 return 1;
76dda93c 8752 else
45321ac5 8753 return generic_drop_inode(inode);
76dda93c
YZ		 8754}
8755
0ee0fda0 8756static void init_once(void *foo)
39279cc3
CM		 8757{
8758 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8759
8760 inode_init_once(&ei->vfs_inode);
8761}
8762
8763void btrfs_destroy_cachep(void)
8764{
8c0a8537
KS		 8765 /*
8766 * Make sure all delayed rcu free inodes are flushed before we
8767 * destroy cache.
8768 */
8769 rcu_barrier();
39279cc3
CM		 8770 if (btrfs_inode_cachep)
8771 kmem_cache_destroy(btrfs_inode_cachep);
8772 if (btrfs_trans_handle_cachep)
8773 kmem_cache_destroy(btrfs_trans_handle_cachep);
8774 if (btrfs_transaction_cachep)
8775 kmem_cache_destroy(btrfs_transaction_cachep);
39279cc3
CM		 8776 if (btrfs_path_cachep)
8777 kmem_cache_destroy(btrfs_path_cachep);
dc89e982
JB		 8778 if (btrfs_free_space_cachep)
8779 kmem_cache_destroy(btrfs_free_space_cachep);
8ccf6f19
MX		 8780 if (btrfs_delalloc_work_cachep)
8781 kmem_cache_destroy(btrfs_delalloc_work_cachep);
39279cc3
CM		 8782}
8783
8784int btrfs_init_cachep(void)
8785{
837e1972 8786 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9601e3f6
CH		 8787 sizeof(struct btrfs_inode), 0,
8788 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
39279cc3
CM		 8789 if (!btrfs_inode_cachep)
8790 goto fail;
9601e3f6 8791
837e1972 8792 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9601e3f6
CH		 8793 sizeof(struct btrfs_trans_handle), 0,
8794 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 8795 if (!btrfs_trans_handle_cachep)
8796 goto fail;
9601e3f6 8797
837e1972 8798 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9601e3f6
CH		 8799 sizeof(struct btrfs_transaction), 0,
8800 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 8801 if (!btrfs_transaction_cachep)
8802 goto fail;
9601e3f6 8803
837e1972 8804 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9601e3f6
CH		 8805 sizeof(struct btrfs_path), 0,
8806 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 8807 if (!btrfs_path_cachep)
8808 goto fail;
9601e3f6 8809
837e1972 8810 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
dc89e982
JB		 8811 sizeof(struct btrfs_free_space), 0,
8812 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8813 if (!btrfs_free_space_cachep)
8814 goto fail;
8815
8ccf6f19
MX		 8816 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8817 sizeof(struct btrfs_delalloc_work), 0,
8818 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8819 NULL);
8820 if (!btrfs_delalloc_work_cachep)
8821 goto fail;
8822
39279cc3
CM		 8823 return 0;
8824fail:
8825 btrfs_destroy_cachep();
8826 return -ENOMEM;
8827}
8828
8829static int btrfs_getattr(struct vfsmount *mnt,
8830 struct dentry *dentry, struct kstat *stat)
8831{
df0af1a5 8832 u64 delalloc_bytes;
39279cc3 8833 struct inode *inode = dentry->d_inode;
fadc0d8b
DS		 8834 u32 blocksize = inode->i_sb->s_blocksize;
8835
39279cc3 8836 generic_fillattr(inode, stat);
0ee5dc67 8837 stat->dev = BTRFS_I(inode)->root->anon_dev;
d6667462 8838 stat->blksize = PAGE_CACHE_SIZE;
df0af1a5
MX		 8839
8840 spin_lock(&BTRFS_I(inode)->lock);
8841 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8842 spin_unlock(&BTRFS_I(inode)->lock);
fadc0d8b 8843 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
df0af1a5 8844 ALIGN(delalloc_bytes, blocksize)) >> 9;
39279cc3
CM		 8845 return 0;
8846}
8847
d397712b
CM		 8848static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8849 struct inode *new_dir, struct dentry *new_dentry)
39279cc3
CM		 8850{
8851 struct btrfs_trans_handle *trans;
8852 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4df27c4d 8853 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
39279cc3
CM		 8854 struct inode *new_inode = new_dentry->d_inode;
8855 struct inode *old_inode = old_dentry->d_inode;
8856 struct timespec ctime = CURRENT_TIME;
00e4e6b3 8857 u64 index = 0;
4df27c4d 8858 u64 root_objectid;
39279cc3 8859 int ret;
33345d01 8860 u64 old_ino = btrfs_ino(old_inode);
39279cc3 8861
33345d01 8862 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
f679a840
YZ		 8863 return -EPERM;
8864
4df27c4d 8865 /* we only allow rename subvolume link between subvolumes */
33345d01 8866 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
3394e160
CM		 8867 return -EXDEV;
8868
33345d01
LZ		 8869 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8870 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
39279cc3 8871 return -ENOTEMPTY;
5f39d397 8872
4df27c4d
YZ		 8873 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8874 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8875 return -ENOTEMPTY;
9c52057c
CM		 8876
8877
8878 /* check for collisions, even if the name isn't there */
4871c158 8879 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9c52057c
CM		 8880 new_dentry->d_name.name,
8881 new_dentry->d_name.len);
8882
8883 if (ret) {
8884 if (ret == -EEXIST) {
8885 /* we shouldn't get
8886 * eexist without a new_inode */
fae7f21c 8887 if (WARN_ON(!new_inode)) {
9c52057c
CM		 8888 return ret;
8889 }
8890 } else {
8891 /* maybe -EOVERFLOW */
8892 return ret;
8893 }
8894 }
8895 ret = 0;
8896
5a3f23d5 8897 /*
8d875f95
CM		 8898 * we're using rename to replace one file with another. Start IO on it
8899 * now so we don't add too much work to the end of the transaction
5a3f23d5 8900 */
8d875f95 8901 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
5a3f23d5
CM		 8902 filemap_flush(old_inode->i_mapping);
8903
76dda93c 8904 /* close the racy window with snapshot create/destroy ioctl */
33345d01 8905 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 8906 down_read(&root->fs_info->subvol_sem);
a22285a6
YZ		 8907 /*
8908 * We want to reserve the absolute worst case amount of items. So if
8909 * both inodes are subvols and we need to unlink them then that would
8910 * require 4 item modifications, but if they are both normal inodes it
8911 * would require 5 item modifications, so we'll assume their normal
8912 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8913 * should cover the worst case number of items we'll modify.
8914 */
6e137ed3 8915 trans = btrfs_start_transaction(root, 11);
b44c59a8
JL		 8916 if (IS_ERR(trans)) {
8917 ret = PTR_ERR(trans);
8918 goto out_notrans;
8919 }
76dda93c 8920
4df27c4d
YZ		 8921 if (dest != root)
8922 btrfs_record_root_in_trans(trans, dest);
5f39d397 8923
a5719521
YZ		 8924 ret = btrfs_set_inode_index(new_dir, &index);
8925 if (ret)
8926 goto out_fail;
5a3f23d5 8927
67de1176 8928 BTRFS_I(old_inode)->dir_index = 0ULL;
33345d01 8929 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d 8930 /* force full log commit if subvolume involved. */
995946dd 8931 btrfs_set_log_full_commit(root->fs_info, trans);
4df27c4d 8932 } else {
a5719521
YZ		 8933 ret = btrfs_insert_inode_ref(trans, dest,
8934 new_dentry->d_name.name,
8935 new_dentry->d_name.len,
33345d01
LZ		 8936 old_ino,
8937 btrfs_ino(new_dir), index);
a5719521
YZ		 8938 if (ret)
8939 goto out_fail;
4df27c4d
YZ		 8940 /*
8941 * this is an ugly little race, but the rename is required
8942 * to make sure that if we crash, the inode is either at the
8943 * old name or the new one. pinning the log transaction lets
8944 * us make sure we don't allow a log commit to come in after
8945 * we unlink the name but before we add the new name back in.
8946 */
8947 btrfs_pin_log_trans(root);
8948 }
5a3f23d5 8949
0c4d2d95
JB		 8950 inode_inc_iversion(old_dir);
8951 inode_inc_iversion(new_dir);
8952 inode_inc_iversion(old_inode);
39279cc3
CM		 8953 old_dir->i_ctime = old_dir->i_mtime = ctime;
8954 new_dir->i_ctime = new_dir->i_mtime = ctime;
8955 old_inode->i_ctime = ctime;
5f39d397 8956
12fcfd22
CM		 8957 if (old_dentry->d_parent != new_dentry->d_parent)
8958 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8959
33345d01 8960 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 8961 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8962 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8963 old_dentry->d_name.name,
8964 old_dentry->d_name.len);
8965 } else {
92986796
AV		 8966 ret = __btrfs_unlink_inode(trans, root, old_dir,
8967 old_dentry->d_inode,
8968 old_dentry->d_name.name,
8969 old_dentry->d_name.len);
8970 if (!ret)
8971 ret = btrfs_update_inode(trans, root, old_inode);
4df27c4d 8972 }
79787eaa
JM		 8973 if (ret) {
8974 btrfs_abort_transaction(trans, root, ret);
8975 goto out_fail;
8976 }
39279cc3
CM		 8977
8978 if (new_inode) {
0c4d2d95 8979 inode_inc_iversion(new_inode);
39279cc3 8980 new_inode->i_ctime = CURRENT_TIME;
33345d01 8981 if (unlikely(btrfs_ino(new_inode) ==
4df27c4d
YZ		 8982 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8983 root_objectid = BTRFS_I(new_inode)->location.objectid;
8984 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8985 root_objectid,
8986 new_dentry->d_name.name,
8987 new_dentry->d_name.len);
8988 BUG_ON(new_inode->i_nlink == 0);
8989 } else {
8990 ret = btrfs_unlink_inode(trans, dest, new_dir,
8991 new_dentry->d_inode,
8992 new_dentry->d_name.name,
8993 new_dentry->d_name.len);
8994 }
4ef31a45 8995 if (!ret && new_inode->i_nlink == 0)
e02119d5 8996 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
79787eaa
JM		 8997 if (ret) {
8998 btrfs_abort_transaction(trans, root, ret);
8999 goto out_fail;
9000 }
39279cc3 9001 }
aec7477b 9002
4df27c4d
YZ		 9003 ret = btrfs_add_link(trans, new_dir, old_inode,
9004 new_dentry->d_name.name,
a5719521 9005 new_dentry->d_name.len, 0, index);
79787eaa
JM		 9006 if (ret) {
9007 btrfs_abort_transaction(trans, root, ret);
9008 goto out_fail;
9009 }
39279cc3 9010
67de1176
MX		 9011 if (old_inode->i_nlink == 1)
9012 BTRFS_I(old_inode)->dir_index = index;
9013
33345d01 9014 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
10d9f309 9015 struct dentry *parent = new_dentry->d_parent;
6a912213 9016 btrfs_log_new_name(trans, old_inode, old_dir, parent);
4df27c4d
YZ		 9017 btrfs_end_log_trans(root);
9018 }
39279cc3 9019out_fail:
7ad85bb7 9020 btrfs_end_transaction(trans, root);
b44c59a8 9021out_notrans:
33345d01 9022 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 9023 up_read(&root->fs_info->subvol_sem);
9ed74f2d 9024
39279cc3
CM		 9025 return ret;
9026}
9027
80ace85c
MS		 9028static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9029 struct inode *new_dir, struct dentry *new_dentry,
9030 unsigned int flags)
9031{
9032 if (flags & ~RENAME_NOREPLACE)
9033 return -EINVAL;
9034
9035 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
9036}
9037
8ccf6f19
MX		 9038static void btrfs_run_delalloc_work(struct btrfs_work *work)
9039{
9040 struct btrfs_delalloc_work *delalloc_work;
9f23e289 9041 struct inode *inode;
8ccf6f19
MX		 9042
9043 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9044 work);
9f23e289
JB		 9045 inode = delalloc_work->inode;
9046 if (delalloc_work->wait) {
9047 btrfs_wait_ordered_range(inode, 0, (u64)-1);
9048 } else {
9049 filemap_flush(inode->i_mapping);
9050 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9051 &BTRFS_I(inode)->runtime_flags))
9052 filemap_flush(inode->i_mapping);
9053 }
8ccf6f19
MX		 9054
9055 if (delalloc_work->delay_iput)
9f23e289 9056 btrfs_add_delayed_iput(inode);
8ccf6f19 9057 else
9f23e289 9058 iput(inode);
8ccf6f19
MX		 9059 complete(&delalloc_work->completion);
9060}
9061
9062struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9063 int wait, int delay_iput)
9064{
9065 struct btrfs_delalloc_work *work;
9066
9067 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
9068 if (!work)
9069 return NULL;
9070
9071 init_completion(&work->completion);
9072 INIT_LIST_HEAD(&work->list);
9073 work->inode = inode;
9074 work->wait = wait;
9075 work->delay_iput = delay_iput;
9e0af237
LB		 9076 WARN_ON_ONCE(!inode);
9077 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9078 btrfs_run_delalloc_work, NULL, NULL);
8ccf6f19
MX		 9079
9080 return work;
9081}
9082
9083void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9084{
9085 wait_for_completion(&work->completion);
9086 kmem_cache_free(btrfs_delalloc_work_cachep, work);
9087}
9088
d352ac68
CM		 9089/*
9090 * some fairly slow code that needs optimization. This walks the list
9091 * of all the inodes with pending delalloc and forces them to disk.
9092 */
6c255e67
MX		 9093static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9094 int nr)
ea8c2819 9095{
ea8c2819 9096 struct btrfs_inode *binode;
5b21f2ed 9097 struct inode *inode;
8ccf6f19
MX		 9098 struct btrfs_delalloc_work *work, *next;
9099 struct list_head works;
1eafa6c7 9100 struct list_head splice;
8ccf6f19 9101 int ret = 0;
ea8c2819 9102
8ccf6f19 9103 INIT_LIST_HEAD(&works);
1eafa6c7 9104 INIT_LIST_HEAD(&splice);
63607cc8 9105
573bfb72 9106 mutex_lock(&root->delalloc_mutex);
eb73c1b7
MX		 9107 spin_lock(&root->delalloc_lock);
9108 list_splice_init(&root->delalloc_inodes, &splice);
1eafa6c7
MX		 9109 while (!list_empty(&splice)) {
9110 binode = list_entry(splice.next, struct btrfs_inode,
ea8c2819 9111 delalloc_inodes);
1eafa6c7 9112
eb73c1b7
MX		 9113 list_move_tail(&binode->delalloc_inodes,
9114 &root->delalloc_inodes);
5b21f2ed 9115 inode = igrab(&binode->vfs_inode);
df0af1a5 9116 if (!inode) {
eb73c1b7 9117 cond_resched_lock(&root->delalloc_lock);
1eafa6c7 9118 continue;
df0af1a5 9119 }
eb73c1b7 9120 spin_unlock(&root->delalloc_lock);
1eafa6c7
MX		 9121
9122 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
5d99a998 9123 if (!work) {
f4ab9ea7
JB		 9124 if (delay_iput)
9125 btrfs_add_delayed_iput(inode);
9126 else
9127 iput(inode);
1eafa6c7 9128 ret = -ENOMEM;
a1ecaabb 9129 goto out;
5b21f2ed 9130 }
1eafa6c7 9131 list_add_tail(&work->list, &works);
a44903ab
QW		 9132 btrfs_queue_work(root->fs_info->flush_workers,
9133 &work->work);
6c255e67
MX		 9134 ret++;
9135 if (nr != -1 && ret >= nr)
a1ecaabb 9136 goto out;
5b21f2ed 9137 cond_resched();
eb73c1b7 9138 spin_lock(&root->delalloc_lock);
ea8c2819 9139 }
eb73c1b7 9140 spin_unlock(&root->delalloc_lock);
8c8bee1d 9141
a1ecaabb 9142out:
eb73c1b7
MX		 9143 list_for_each_entry_safe(work, next, &works, list) {
9144 list_del_init(&work->list);
9145 btrfs_wait_and_free_delalloc_work(work);
9146 }
9147
9148 if (!list_empty_careful(&splice)) {
9149 spin_lock(&root->delalloc_lock);
9150 list_splice_tail(&splice, &root->delalloc_inodes);
9151 spin_unlock(&root->delalloc_lock);
9152 }
573bfb72 9153 mutex_unlock(&root->delalloc_mutex);
eb73c1b7
MX		 9154 return ret;
9155}
1eafa6c7 9156
eb73c1b7
MX		 9157int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9158{
9159 int ret;
1eafa6c7 9160
2c21b4d7 9161 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
eb73c1b7
MX		 9162 return -EROFS;
9163
6c255e67
MX		 9164 ret = __start_delalloc_inodes(root, delay_iput, -1);
9165 if (ret > 0)
9166 ret = 0;
eb73c1b7
MX		 9167 /*
9168 * the filemap_flush will queue IO into the worker threads, but
8c8bee1d
CM		 9169 * we have to make sure the IO is actually started and that
9170 * ordered extents get created before we return
9171 */
9172 atomic_inc(&root->fs_info->async_submit_draining);
d397712b 9173 while (atomic_read(&root->fs_info->nr_async_submits) ||
771ed689 9174 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 9175 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 9176 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9177 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 9178 }
9179 atomic_dec(&root->fs_info->async_submit_draining);
eb73c1b7
MX		 9180 return ret;
9181}
9182
6c255e67
MX		 9183int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9184 int nr)
eb73c1b7
MX		 9185{
9186 struct btrfs_root *root;
9187 struct list_head splice;
9188 int ret;
9189
2c21b4d7 9190 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
eb73c1b7
MX		 9191 return -EROFS;
9192
9193 INIT_LIST_HEAD(&splice);
9194
573bfb72 9195 mutex_lock(&fs_info->delalloc_root_mutex);
eb73c1b7
MX		 9196 spin_lock(&fs_info->delalloc_root_lock);
9197 list_splice_init(&fs_info->delalloc_roots, &splice);
6c255e67 9198 while (!list_empty(&splice) && nr) {
eb73c1b7
MX		 9199 root = list_first_entry(&splice, struct btrfs_root,
9200 delalloc_root);
9201 root = btrfs_grab_fs_root(root);
9202 BUG_ON(!root);
9203 list_move_tail(&root->delalloc_root,
9204 &fs_info->delalloc_roots);
9205 spin_unlock(&fs_info->delalloc_root_lock);
9206
6c255e67 9207 ret = __start_delalloc_inodes(root, delay_iput, nr);
eb73c1b7 9208 btrfs_put_fs_root(root);
6c255e67 9209 if (ret < 0)
eb73c1b7
MX		 9210 goto out;
9211
6c255e67
MX		 9212 if (nr != -1) {
9213 nr -= ret;
9214 WARN_ON(nr < 0);
9215 }
eb73c1b7 9216 spin_lock(&fs_info->delalloc_root_lock);
8ccf6f19 9217 }
eb73c1b7 9218 spin_unlock(&fs_info->delalloc_root_lock);
1eafa6c7 9219
6c255e67 9220 ret = 0;
eb73c1b7
MX		 9221 atomic_inc(&fs_info->async_submit_draining);
9222 while (atomic_read(&fs_info->nr_async_submits) ||
9223 atomic_read(&fs_info->async_delalloc_pages)) {
9224 wait_event(fs_info->async_submit_wait,
9225 (atomic_read(&fs_info->nr_async_submits) == 0 &&
9226 atomic_read(&fs_info->async_delalloc_pages) == 0));
9227 }
9228 atomic_dec(&fs_info->async_submit_draining);
eb73c1b7 9229out:
1eafa6c7 9230 if (!list_empty_careful(&splice)) {
eb73c1b7
MX		 9231 spin_lock(&fs_info->delalloc_root_lock);
9232 list_splice_tail(&splice, &fs_info->delalloc_roots);
9233 spin_unlock(&fs_info->delalloc_root_lock);
1eafa6c7 9234 }
573bfb72 9235 mutex_unlock(&fs_info->delalloc_root_mutex);
8ccf6f19 9236 return ret;
ea8c2819
CM		 9237}
9238
39279cc3
CM		 9239static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9240 const char *symname)
9241{
9242 struct btrfs_trans_handle *trans;
9243 struct btrfs_root *root = BTRFS_I(dir)->root;
9244 struct btrfs_path *path;
9245 struct btrfs_key key;
1832a6d5 9246 struct inode *inode = NULL;
39279cc3
CM		 9247 int err;
9248 int drop_inode = 0;
9249 u64 objectid;
67871254 9250 u64 index = 0;
39279cc3
CM		 9251 int name_len;
9252 int datasize;
5f39d397 9253 unsigned long ptr;
39279cc3 9254 struct btrfs_file_extent_item *ei;
5f39d397 9255 struct extent_buffer *leaf;
39279cc3 9256
f06becc4 9257 name_len = strlen(symname);
39279cc3
CM		 9258 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9259 return -ENAMETOOLONG;
1832a6d5 9260
9ed74f2d
JB		 9261 /*
9262 * 2 items for inode item and ref
9263 * 2 items for dir items
9264 * 1 item for xattr if selinux is on
9265 */
a22285a6
YZ		 9266 trans = btrfs_start_transaction(root, 5);
9267 if (IS_ERR(trans))
9268 return PTR_ERR(trans);
1832a6d5 9269
581bb050
LZ		 9270 err = btrfs_find_free_ino(root, &objectid);
9271 if (err)
9272 goto out_unlock;
9273
aec7477b 9274 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 9275 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 9276 S_IFLNK|S_IRWXUGO, &index);
7cf96da3
TI		 9277 if (IS_ERR(inode)) {
9278 err = PTR_ERR(inode);
39279cc3 9279 goto out_unlock;
7cf96da3 9280 }
39279cc3 9281
ad19db71
CS		 9282 /*
9283 * If the active LSM wants to access the inode during
9284 * d_instantiate it needs these. Smack checks to see
9285 * if the filesystem supports xattrs by looking at the
9286 * ops vector.
9287 */
9288 inode->i_fop = &btrfs_file_operations;
9289 inode->i_op = &btrfs_file_inode_operations;
b0d5d10f
CM		 9290 inode->i_mapping->a_ops = &btrfs_aops;
9291 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
9292 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9293
9294 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9295 if (err)
9296 goto out_unlock_inode;
ad19db71 9297
a1b075d2 9298 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3 9299 if (err)
b0d5d10f 9300 goto out_unlock_inode;
39279cc3
CM		 9301
9302 path = btrfs_alloc_path();
d8926bb3
MF		 9303 if (!path) {
9304 err = -ENOMEM;
b0d5d10f 9305 goto out_unlock_inode;
d8926bb3 9306 }
33345d01 9307 key.objectid = btrfs_ino(inode);
39279cc3 9308 key.offset = 0;
962a298f 9309 key.type = BTRFS_EXTENT_DATA_KEY;
39279cc3
CM		 9310 datasize = btrfs_file_extent_calc_inline_size(name_len);
9311 err = btrfs_insert_empty_item(trans, root, path, &key,
9312 datasize);
54aa1f4d 9313 if (err) {
b0839166 9314 btrfs_free_path(path);
b0d5d10f 9315 goto out_unlock_inode;
54aa1f4d 9316 }
5f39d397
CM		 9317 leaf = path->nodes[0];
9318 ei = btrfs_item_ptr(leaf, path->slots[0],
9319 struct btrfs_file_extent_item);
9320 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9321 btrfs_set_file_extent_type(leaf, ei,
39279cc3 9322 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 9323 btrfs_set_file_extent_encryption(leaf, ei, 0);
9324 btrfs_set_file_extent_compression(leaf, ei, 0);
9325 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9326 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9327
39279cc3 9328 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 9329 write_extent_buffer(leaf, symname, ptr, name_len);
9330 btrfs_mark_buffer_dirty(leaf);
39279cc3 9331 btrfs_free_path(path);
5f39d397 9332
39279cc3
CM		 9333 inode->i_op = &btrfs_symlink_inode_operations;
9334 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 9335 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d899e052 9336 inode_set_bytes(inode, name_len);
f06becc4 9337 btrfs_i_size_write(inode, name_len);
54aa1f4d 9338 err = btrfs_update_inode(trans, root, inode);
b0d5d10f 9339 if (err) {
54aa1f4d 9340 drop_inode = 1;
b0d5d10f
CM		 9341 goto out_unlock_inode;
9342 }
9343
9344 unlock_new_inode(inode);
9345 d_instantiate(dentry, inode);
39279cc3
CM		 9346
9347out_unlock:
7ad85bb7 9348 btrfs_end_transaction(trans, root);
39279cc3
CM		 9349 if (drop_inode) {
9350 inode_dec_link_count(inode);
9351 iput(inode);
9352 }
b53d3f5d 9353 btrfs_btree_balance_dirty(root);
39279cc3 9354 return err;
b0d5d10f
CM		 9355
9356out_unlock_inode:
9357 drop_inode = 1;
9358 unlock_new_inode(inode);
9359 goto out_unlock;
39279cc3 9360}
16432985 9361
0af3d00b
JB		 9362static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9363 u64 start, u64 num_bytes, u64 min_size,
9364 loff_t actual_len, u64 *alloc_hint,
9365 struct btrfs_trans_handle *trans)
d899e052 9366{
5dc562c5
JB		 9367 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9368 struct extent_map *em;
d899e052
YZ		 9369 struct btrfs_root *root = BTRFS_I(inode)->root;
9370 struct btrfs_key ins;
d899e052 9371 u64 cur_offset = start;
55a61d1d 9372 u64 i_size;
154ea289 9373 u64 cur_bytes;
d899e052 9374 int ret = 0;
0af3d00b 9375 bool own_trans = true;
d899e052 9376
0af3d00b
JB		 9377 if (trans)
9378 own_trans = false;
d899e052 9379 while (num_bytes > 0) {
0af3d00b
JB		 9380 if (own_trans) {
9381 trans = btrfs_start_transaction(root, 3);
9382 if (IS_ERR(trans)) {
9383 ret = PTR_ERR(trans);
9384 break;
9385 }
5a303d5d
YZ		 9386 }
9387
154ea289
CM		 9388 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
9389 cur_bytes = max(cur_bytes, min_size);
00361589 9390 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
e570fd27 9391 *alloc_hint, &ins, 1, 0);
5a303d5d 9392 if (ret) {
0af3d00b
JB		 9393 if (own_trans)
9394 btrfs_end_transaction(trans, root);
a22285a6 9395 break;
d899e052 9396 }
5a303d5d 9397
d899e052
YZ		 9398 ret = insert_reserved_file_extent(trans, inode,
9399 cur_offset, ins.objectid,
9400 ins.offset, ins.offset,
920bbbfb 9401 ins.offset, 0, 0, 0,
d899e052 9402 BTRFS_FILE_EXTENT_PREALLOC);
79787eaa 9403 if (ret) {
857cc2fc 9404 btrfs_free_reserved_extent(root, ins.objectid,
e570fd27 9405 ins.offset, 0);
79787eaa
JM		 9406 btrfs_abort_transaction(trans, root, ret);
9407 if (own_trans)
9408 btrfs_end_transaction(trans, root);
9409 break;
9410 }
a1ed835e
CM		 9411 btrfs_drop_extent_cache(inode, cur_offset,
9412 cur_offset + ins.offset -1, 0);
5a303d5d 9413
5dc562c5
JB		 9414 em = alloc_extent_map();
9415 if (!em) {
9416 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9417 &BTRFS_I(inode)->runtime_flags);
9418 goto next;
9419 }
9420
9421 em->start = cur_offset;
9422 em->orig_start = cur_offset;
9423 em->len = ins.offset;
9424 em->block_start = ins.objectid;
9425 em->block_len = ins.offset;
b4939680 9426 em->orig_block_len = ins.offset;
cc95bef6 9427 em->ram_bytes = ins.offset;
5dc562c5
JB		 9428 em->bdev = root->fs_info->fs_devices->latest_bdev;
9429 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9430 em->generation = trans->transid;
9431
9432 while (1) {
9433 write_lock(&em_tree->lock);
09a2a8f9 9434 ret = add_extent_mapping(em_tree, em, 1);
5dc562c5
JB		 9435 write_unlock(&em_tree->lock);
9436 if (ret != -EEXIST)
9437 break;
9438 btrfs_drop_extent_cache(inode, cur_offset,
9439 cur_offset + ins.offset - 1,
9440 0);
9441 }
9442 free_extent_map(em);
9443next:
d899e052
YZ		 9444 num_bytes -= ins.offset;
9445 cur_offset += ins.offset;
efa56464 9446 *alloc_hint = ins.objectid + ins.offset;
5a303d5d 9447
0c4d2d95 9448 inode_inc_iversion(inode);
d899e052 9449 inode->i_ctime = CURRENT_TIME;
6cbff00f 9450 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
d899e052 9451 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
efa56464
YZ		 9452 (actual_len > inode->i_size) &&
9453 (cur_offset > inode->i_size)) {
d1ea6a61 9454 if (cur_offset > actual_len)
55a61d1d 9455 i_size = actual_len;
d1ea6a61 9456 else
55a61d1d
JB		 9457 i_size = cur_offset;
9458 i_size_write(inode, i_size);
9459 btrfs_ordered_update_i_size(inode, i_size, NULL);
5a303d5d
YZ		 9460 }
9461
d899e052 9462 ret = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 9463
9464 if (ret) {
9465 btrfs_abort_transaction(trans, root, ret);
9466 if (own_trans)
9467 btrfs_end_transaction(trans, root);
9468 break;
9469 }
d899e052 9470
0af3d00b
JB		 9471 if (own_trans)
9472 btrfs_end_transaction(trans, root);
5a303d5d 9473 }
d899e052
YZ		 9474 return ret;
9475}
9476
0af3d00b
JB		 9477int btrfs_prealloc_file_range(struct inode *inode, int mode,
9478 u64 start, u64 num_bytes, u64 min_size,
9479 loff_t actual_len, u64 *alloc_hint)
9480{
9481 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9482 min_size, actual_len, alloc_hint,
9483 NULL);
9484}
9485
9486int btrfs_prealloc_file_range_trans(struct inode *inode,
9487 struct btrfs_trans_handle *trans, int mode,
9488 u64 start, u64 num_bytes, u64 min_size,
9489 loff_t actual_len, u64 *alloc_hint)
9490{
9491 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9492 min_size, actual_len, alloc_hint, trans);
9493}
9494
e6dcd2dc
CM		 9495static int btrfs_set_page_dirty(struct page *page)
9496{
e6dcd2dc
CM		 9497 return __set_page_dirty_nobuffers(page);
9498}
9499
10556cb2 9500static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd 9501{
b83cc969 9502 struct btrfs_root *root = BTRFS_I(inode)->root;
cb6db4e5 9503 umode_t mode = inode->i_mode;
b83cc969 9504
cb6db4e5
JM		 9505 if (mask & MAY_WRITE &&
9506 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9507 if (btrfs_root_readonly(root))
9508 return -EROFS;
9509 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9510 return -EACCES;
9511 }
2830ba7f 9512 return generic_permission(inode, mask);
fdebe2bd 9513}
39279cc3 9514
ef3b9af5
FM		 9515static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9516{
9517 struct btrfs_trans_handle *trans;
9518 struct btrfs_root *root = BTRFS_I(dir)->root;
9519 struct inode *inode = NULL;
9520 u64 objectid;
9521 u64 index;
9522 int ret = 0;
9523
9524 /*
9525 * 5 units required for adding orphan entry
9526 */
9527 trans = btrfs_start_transaction(root, 5);
9528 if (IS_ERR(trans))
9529 return PTR_ERR(trans);
9530
9531 ret = btrfs_find_free_ino(root, &objectid);
9532 if (ret)
9533 goto out;
9534
9535 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9536 btrfs_ino(dir), objectid, mode, &index);
9537 if (IS_ERR(inode)) {
9538 ret = PTR_ERR(inode);
9539 inode = NULL;
9540 goto out;
9541 }
9542
ef3b9af5
FM		 9543 inode->i_fop = &btrfs_file_operations;
9544 inode->i_op = &btrfs_file_inode_operations;
9545
9546 inode->i_mapping->a_ops = &btrfs_aops;
9547 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
9548 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9549
b0d5d10f
CM		 9550 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9551 if (ret)
9552 goto out_inode;
9553
9554 ret = btrfs_update_inode(trans, root, inode);
9555 if (ret)
9556 goto out_inode;
ef3b9af5
FM		 9557 ret = btrfs_orphan_add(trans, inode);
9558 if (ret)
b0d5d10f 9559 goto out_inode;
ef3b9af5 9560
5762b5c9
FM		 9561 /*
9562 * We set number of links to 0 in btrfs_new_inode(), and here we set
9563 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9564 * through:
9565 *
9566 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9567 */
9568 set_nlink(inode, 1);
b0d5d10f 9569 unlock_new_inode(inode);
ef3b9af5
FM		 9570 d_tmpfile(dentry, inode);
9571 mark_inode_dirty(inode);
9572
9573out:
9574 btrfs_end_transaction(trans, root);
9575 if (ret)
9576 iput(inode);
9577 btrfs_balance_delayed_items(root);
9578 btrfs_btree_balance_dirty(root);
ef3b9af5 9579 return ret;
b0d5d10f
CM		 9580
9581out_inode:
9582 unlock_new_inode(inode);
9583 goto out;
9584
ef3b9af5
FM		 9585}
9586
b38ef71c
FM		 9587/* Inspired by filemap_check_errors() */
9588int btrfs_inode_check_errors(struct inode *inode)
9589{
9590 int ret = 0;
9591
9592 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
9593 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
9594 ret = -ENOSPC;
9595 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
9596 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
9597 ret = -EIO;
9598
9599 return ret;
9600}
9601
6e1d5dcc 9602static const struct inode_operations btrfs_dir_inode_operations = {
3394e160 9603 .getattr = btrfs_getattr,
39279cc3
CM		 9604 .lookup = btrfs_lookup,
9605 .create = btrfs_create,
9606 .unlink = btrfs_unlink,
9607 .link = btrfs_link,
9608 .mkdir = btrfs_mkdir,
9609 .rmdir = btrfs_rmdir,
80ace85c 9610 .rename2 = btrfs_rename2,
39279cc3
CM		 9611 .symlink = btrfs_symlink,
9612 .setattr = btrfs_setattr,
618e21d5 9613 .mknod = btrfs_mknod,
95819c05
CH		 9614 .setxattr = btrfs_setxattr,
9615 .getxattr = btrfs_getxattr,
5103e947 9616 .listxattr = btrfs_listxattr,
95819c05 9617 .removexattr = btrfs_removexattr,
fdebe2bd 9618 .permission = btrfs_permission,
4e34e719 9619 .get_acl = btrfs_get_acl,
996a710d 9620 .set_acl = btrfs_set_acl,
93fd63c2 9621 .update_time = btrfs_update_time,
ef3b9af5 9622 .tmpfile = btrfs_tmpfile,
39279cc3 9623};
6e1d5dcc 9624static const struct inode_operations btrfs_dir_ro_inode_operations = {
39279cc3 9625 .lookup = btrfs_lookup,
fdebe2bd 9626 .permission = btrfs_permission,
4e34e719 9627 .get_acl = btrfs_get_acl,
996a710d 9628 .set_acl = btrfs_set_acl,
93fd63c2 9629 .update_time = btrfs_update_time,
39279cc3 9630};
76dda93c 9631
828c0950 9632static const struct file_operations btrfs_dir_file_operations = {
39279cc3
CM		 9633 .llseek = generic_file_llseek,
9634 .read = generic_read_dir,
9cdda8d3 9635 .iterate = btrfs_real_readdir,
34287aa3 9636 .unlocked_ioctl = btrfs_ioctl,
39279cc3 9637#ifdef CONFIG_COMPAT
34287aa3 9638 .compat_ioctl = btrfs_ioctl,
39279cc3 9639#endif
6bf13c0c 9640 .release = btrfs_release_file,
e02119d5 9641 .fsync = btrfs_sync_file,
39279cc3
CM		 9642};
9643
d1310b2e 9644static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 9645 .fill_delalloc = run_delalloc_range,
065631f6 9646 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 9647 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 9648 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 9649 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 9650 .writepage_start_hook = btrfs_writepage_start_hook,
b0c68f8b
CM		 9651 .set_bit_hook = btrfs_set_bit_hook,
9652 .clear_bit_hook = btrfs_clear_bit_hook,
9ed74f2d
JB		 9653 .merge_extent_hook = btrfs_merge_extent_hook,
9654 .split_extent_hook = btrfs_split_extent_hook,
07157aac
CM		 9655};
9656
35054394
CM		 9657/*
9658 * btrfs doesn't support the bmap operation because swapfiles
9659 * use bmap to make a mapping of extents in the file. They assume
9660 * these extents won't change over the life of the file and they
9661 * use the bmap result to do IO directly to the drive.
9662 *
9663 * the btrfs bmap call would return logical addresses that aren't
9664 * suitable for IO and they also will change frequently as COW
9665 * operations happen. So, swapfile + btrfs == corruption.
9666 *
9667 * For now we're avoiding this by dropping bmap.
9668 */
7f09410b 9669static const struct address_space_operations btrfs_aops = {
39279cc3
CM		 9670 .readpage = btrfs_readpage,
9671 .writepage = btrfs_writepage,
b293f02e 9672 .writepages = btrfs_writepages,
3ab2fb5a 9673 .readpages = btrfs_readpages,
16432985 9674 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 9675 .invalidatepage = btrfs_invalidatepage,
9676 .releasepage = btrfs_releasepage,
e6dcd2dc 9677 .set_page_dirty = btrfs_set_page_dirty,
465fdd97 9678 .error_remove_page = generic_error_remove_page,
39279cc3
CM		 9679};
9680
7f09410b 9681static const struct address_space_operations btrfs_symlink_aops = {
39279cc3
CM		 9682 .readpage = btrfs_readpage,
9683 .writepage = btrfs_writepage,
2bf5a725
CM		 9684 .invalidatepage = btrfs_invalidatepage,
9685 .releasepage = btrfs_releasepage,
39279cc3
CM		 9686};
9687
6e1d5dcc 9688static const struct inode_operations btrfs_file_inode_operations = {
39279cc3
CM		 9689 .getattr = btrfs_getattr,
9690 .setattr = btrfs_setattr,
95819c05
CH		 9691 .setxattr = btrfs_setxattr,
9692 .getxattr = btrfs_getxattr,
5103e947 9693 .listxattr = btrfs_listxattr,
95819c05 9694 .removexattr = btrfs_removexattr,
fdebe2bd 9695 .permission = btrfs_permission,
1506fcc8 9696 .fiemap = btrfs_fiemap,
4e34e719 9697 .get_acl = btrfs_get_acl,
996a710d 9698 .set_acl = btrfs_set_acl,
e41f941a 9699 .update_time = btrfs_update_time,
39279cc3 9700};
6e1d5dcc 9701static const struct inode_operations btrfs_special_inode_operations = {
618e21d5
JB		 9702 .getattr = btrfs_getattr,
9703 .setattr = btrfs_setattr,
fdebe2bd 9704 .permission = btrfs_permission,
95819c05
CH		 9705 .setxattr = btrfs_setxattr,
9706 .getxattr = btrfs_getxattr,
33268eaf 9707 .listxattr = btrfs_listxattr,
95819c05 9708 .removexattr = btrfs_removexattr,
4e34e719 9709 .get_acl = btrfs_get_acl,
996a710d 9710 .set_acl = btrfs_set_acl,
e41f941a 9711 .update_time = btrfs_update_time,
618e21d5 9712};
6e1d5dcc 9713static const struct inode_operations btrfs_symlink_inode_operations = {
39279cc3
CM		 9714 .readlink = generic_readlink,
9715 .follow_link = page_follow_link_light,
9716 .put_link = page_put_link,
f209561a 9717 .getattr = btrfs_getattr,
22c44fe6 9718 .setattr = btrfs_setattr,
fdebe2bd 9719 .permission = btrfs_permission,
0279b4cd
JO		 9720 .setxattr = btrfs_setxattr,
9721 .getxattr = btrfs_getxattr,
9722 .listxattr = btrfs_listxattr,
9723 .removexattr = btrfs_removexattr,
e41f941a 9724 .update_time = btrfs_update_time,
39279cc3 9725};
76dda93c 9726
82d339d9 9727const struct dentry_operations btrfs_dentry_operations = {
76dda93c 9728 .d_delete = btrfs_dentry_delete,
b4aff1f8 9729 .d_release = btrfs_dentry_release,
76dda93c 9730};
Linux 6.x block layer and io_uring tree(s)