projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: fall into nocompression codes quickly if possible
[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 ||
70b99e69 252 actual_end >= PAGE_CACHE_SIZE ||
c8b97818
CM		 253 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
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 */
771ed689
CM		 385static noinline int compress_file_range(struct inode *inode,
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
68bb462d
WS		 414 /*
415 * skip compression for a small file range(<=blocksize) that
416 * isn't an inline extent, since it dosen't save disk space at all.
417 */
418 if ((end - start + 1) <= blocksize &&
419 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
420 goto cleanup_and_bail_uncompressed;
421
42dc7bab 422 actual_end = min_t(u64, isize, end + 1);
c8b97818
CM		 423again:
424 will_compress = 0;
425 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
426 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
be20aa9d 427
f03d9301
CM		 428 /*
429 * we don't want to send crud past the end of i_size through
430 * compression, that's just a waste of CPU time. So, if the
431 * end of the file is before the start of our current
432 * requested range of bytes, we bail out to the uncompressed
433 * cleanup code that can deal with all of this.
434 *
435 * It isn't really the fastest way to fix things, but this is a
436 * very uncommon corner.
437 */
438 if (actual_end <= start)
439 goto cleanup_and_bail_uncompressed;
440
c8b97818
CM		 441 total_compressed = actual_end - start;
442
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;
530 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
531
771ed689 532 /*
79787eaa
JM		 533 * inline extent creation worked or returned error,
534 * we don't need to create any more async work items.
535 * Unlock and free up our temp pages.
771ed689 536 */
c2790a2e 537 extent_clear_unlock_delalloc(inode, start, end, NULL,
151a41bc 538 clear_flags, PAGE_UNLOCK |
c2790a2e
JB		 539 PAGE_CLEAR_DIRTY |
540 PAGE_SET_WRITEBACK |
541 PAGE_END_WRITEBACK);
c8b97818
CM		 542 goto free_pages_out;
543 }
544 }
545
546 if (will_compress) {
547 /*
548 * we aren't doing an inline extent round the compressed size
549 * up to a block size boundary so the allocator does sane
550 * things
551 */
fda2832f 552 total_compressed = ALIGN(total_compressed, blocksize);
c8b97818
CM		 553
554 /*
555 * one last check to make sure the compression is really a
556 * win, compare the page count read with the blocks on disk
557 */
fda2832f 558 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
c8b97818
CM		 559 if (total_compressed >= total_in) {
560 will_compress = 0;
561 } else {
c8b97818
CM		 562 num_bytes = total_in;
563 }
564 }
565 if (!will_compress && pages) {
566 /*
567 * the compression code ran but failed to make things smaller,
568 * free any pages it allocated and our page pointer array
569 */
570 for (i = 0; i < nr_pages_ret; i++) {
70b99e69 571 WARN_ON(pages[i]->mapping);
c8b97818
CM		 572 page_cache_release(pages[i]);
573 }
574 kfree(pages);
575 pages = NULL;
576 total_compressed = 0;
577 nr_pages_ret = 0;
578
579 /* flag the file so we don't compress in the future */
1e701a32
CM		 580 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
581 !(BTRFS_I(inode)->force_compress)) {
a555f810 582 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
1e701a32 583 }
c8b97818 584 }
771ed689
CM		 585 if (will_compress) {
586 *num_added += 1;
c8b97818 587
771ed689
CM		 588 /* the async work queues will take care of doing actual
589 * allocation on disk for these compressed pages,
590 * and will submit them to the elevator.
591 */
592 add_async_extent(async_cow, start, num_bytes,
261507a0
LZ		 593 total_compressed, pages, nr_pages_ret,
594 compress_type);
179e29e4 595
24ae6365 596 if (start + num_bytes < end) {
771ed689
CM		 597 start += num_bytes;
598 pages = NULL;
599 cond_resched();
600 goto again;
601 }
602 } else {
f03d9301 603cleanup_and_bail_uncompressed:
771ed689
CM		 604 /*
605 * No compression, but we still need to write the pages in
606 * the file we've been given so far. redirty the locked
607 * page if it corresponds to our extent and set things up
608 * for the async work queue to run cow_file_range to do
609 * the normal delalloc dance
610 */
611 if (page_offset(locked_page) >= start &&
612 page_offset(locked_page) <= end) {
613 __set_page_dirty_nobuffers(locked_page);
614 /* unlocked later on in the async handlers */
615 }
4adaa611
CM		 616 if (redirty)
617 extent_range_redirty_for_io(inode, start, end);
261507a0
LZ		 618 add_async_extent(async_cow, start, end - start + 1,
619 0, NULL, 0, BTRFS_COMPRESS_NONE);
771ed689
CM		 620 *num_added += 1;
621 }
3b951516 622
771ed689 623out:
79787eaa 624 return ret;
771ed689
CM		 625
626free_pages_out:
627 for (i = 0; i < nr_pages_ret; i++) {
628 WARN_ON(pages[i]->mapping);
629 page_cache_release(pages[i]);
630 }
d397712b 631 kfree(pages);
771ed689
CM		 632
633 goto out;
634}
635
636/*
637 * phase two of compressed writeback. This is the ordered portion
638 * of the code, which only gets called in the order the work was
639 * queued. We walk all the async extents created by compress_file_range
640 * and send them down to the disk.
641 */
642static noinline int submit_compressed_extents(struct inode *inode,
643 struct async_cow *async_cow)
644{
645 struct async_extent *async_extent;
646 u64 alloc_hint = 0;
771ed689
CM		 647 struct btrfs_key ins;
648 struct extent_map *em;
649 struct btrfs_root *root = BTRFS_I(inode)->root;
650 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
651 struct extent_io_tree *io_tree;
f5a84ee3 652 int ret = 0;
771ed689
CM		 653
654 if (list_empty(&async_cow->extents))
655 return 0;
656
3e04e7f1 657again:
d397712b 658 while (!list_empty(&async_cow->extents)) {
771ed689
CM		 659 async_extent = list_entry(async_cow->extents.next,
660 struct async_extent, list);
661 list_del(&async_extent->list);
c8b97818 662
771ed689
CM		 663 io_tree = &BTRFS_I(inode)->io_tree;
664
f5a84ee3 665retry:
771ed689
CM		 666 /* did the compression code fall back to uncompressed IO? */
667 if (!async_extent->pages) {
668 int page_started = 0;
669 unsigned long nr_written = 0;
670
671 lock_extent(io_tree, async_extent->start,
2ac55d41 672 async_extent->start +
d0082371 673 async_extent->ram_size - 1);
771ed689
CM		 674
675 /* allocate blocks */
f5a84ee3
JB		 676 ret = cow_file_range(inode, async_cow->locked_page,
677 async_extent->start,
678 async_extent->start +
679 async_extent->ram_size - 1,
680 &page_started, &nr_written, 0);
771ed689 681
79787eaa
JM		 682 /* JDM XXX */
683
771ed689
CM		 684 /*
685 * if page_started, cow_file_range inserted an
686 * inline extent and took care of all the unlocking
687 * and IO for us. Otherwise, we need to submit
688 * all those pages down to the drive.
689 */
f5a84ee3 690 if (!page_started && !ret)
771ed689
CM		 691 extent_write_locked_range(io_tree,
692 inode, async_extent->start,
d397712b 693 async_extent->start +
771ed689
CM		 694 async_extent->ram_size - 1,
695 btrfs_get_extent,
696 WB_SYNC_ALL);
3e04e7f1
JB		 697 else if (ret)
698 unlock_page(async_cow->locked_page);
771ed689
CM		 699 kfree(async_extent);
700 cond_resched();
701 continue;
702 }
703
704 lock_extent(io_tree, async_extent->start,
d0082371 705 async_extent->start + async_extent->ram_size - 1);
771ed689 706
00361589 707 ret = btrfs_reserve_extent(root,
771ed689
CM		 708 async_extent->compressed_size,
709 async_extent->compressed_size,
e570fd27 710 0, alloc_hint, &ins, 1, 1);
f5a84ee3
JB		 711 if (ret) {
712 int i;
3e04e7f1 713
f5a84ee3
JB		 714 for (i = 0; i < async_extent->nr_pages; i++) {
715 WARN_ON(async_extent->pages[i]->mapping);
716 page_cache_release(async_extent->pages[i]);
717 }
718 kfree(async_extent->pages);
719 async_extent->nr_pages = 0;
720 async_extent->pages = NULL;
3e04e7f1 721
fdf8e2ea
JB		 722 if (ret == -ENOSPC) {
723 unlock_extent(io_tree, async_extent->start,
724 async_extent->start +
725 async_extent->ram_size - 1);
ce62003f
LB		 726
727 /*
728 * we need to redirty the pages if we decide to
729 * fallback to uncompressed IO, otherwise we
730 * will not submit these pages down to lower
731 * layers.
732 */
733 extent_range_redirty_for_io(inode,
734 async_extent->start,
735 async_extent->start +
736 async_extent->ram_size - 1);
737
79787eaa 738 goto retry;
fdf8e2ea 739 }
3e04e7f1 740 goto out_free;
f5a84ee3
JB		 741 }
742
c2167754
YZ		 743 /*
744 * here we're doing allocation and writeback of the
745 * compressed pages
746 */
747 btrfs_drop_extent_cache(inode, async_extent->start,
748 async_extent->start +
749 async_extent->ram_size - 1, 0);
750
172ddd60 751 em = alloc_extent_map();
b9aa55be
LB		 752 if (!em) {
753 ret = -ENOMEM;
3e04e7f1 754 goto out_free_reserve;
b9aa55be 755 }
771ed689
CM		 756 em->start = async_extent->start;
757 em->len = async_extent->ram_size;
445a6944 758 em->orig_start = em->start;
2ab28f32
JB		 759 em->mod_start = em->start;
760 em->mod_len = em->len;
c8b97818 761
771ed689
CM		 762 em->block_start = ins.objectid;
763 em->block_len = ins.offset;
b4939680 764 em->orig_block_len = ins.offset;
cc95bef6 765 em->ram_bytes = async_extent->ram_size;
771ed689 766 em->bdev = root->fs_info->fs_devices->latest_bdev;
261507a0 767 em->compress_type = async_extent->compress_type;
771ed689
CM		 768 set_bit(EXTENT_FLAG_PINNED, &em->flags);
769 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
70c8a91c 770 em->generation = -1;
771ed689 771
d397712b 772 while (1) {
890871be 773 write_lock(&em_tree->lock);
09a2a8f9 774 ret = add_extent_mapping(em_tree, em, 1);
890871be 775 write_unlock(&em_tree->lock);
771ed689
CM		 776 if (ret != -EEXIST) {
777 free_extent_map(em);
778 break;
779 }
780 btrfs_drop_extent_cache(inode, async_extent->start,
781 async_extent->start +
782 async_extent->ram_size - 1, 0);
783 }
784
3e04e7f1
JB		 785 if (ret)
786 goto out_free_reserve;
787
261507a0
LZ		 788 ret = btrfs_add_ordered_extent_compress(inode,
789 async_extent->start,
790 ins.objectid,
791 async_extent->ram_size,
792 ins.offset,
793 BTRFS_ORDERED_COMPRESSED,
794 async_extent->compress_type);
d9f85963
FM		 795 if (ret) {
796 btrfs_drop_extent_cache(inode, async_extent->start,
797 async_extent->start +
798 async_extent->ram_size - 1, 0);
3e04e7f1 799 goto out_free_reserve;
d9f85963 800 }
771ed689 801
771ed689
CM		 802 /*
803 * clear dirty, set writeback and unlock the pages.
804 */
c2790a2e 805 extent_clear_unlock_delalloc(inode, async_extent->start,
a791e35e
CM		 806 async_extent->start +
807 async_extent->ram_size - 1,
151a41bc
JB		 808 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
809 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
c2790a2e 810 PAGE_SET_WRITEBACK);
771ed689 811 ret = btrfs_submit_compressed_write(inode,
d397712b
CM		 812 async_extent->start,
813 async_extent->ram_size,
814 ins.objectid,
815 ins.offset, async_extent->pages,
816 async_extent->nr_pages);
771ed689
CM		 817 alloc_hint = ins.objectid + ins.offset;
818 kfree(async_extent);
3e04e7f1
JB		 819 if (ret)
820 goto out;
771ed689
CM		 821 cond_resched();
822 }
79787eaa
JM		 823 ret = 0;
824out:
825 return ret;
3e04e7f1 826out_free_reserve:
e570fd27 827 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
79787eaa 828out_free:
c2790a2e 829 extent_clear_unlock_delalloc(inode, async_extent->start,
3e04e7f1
JB		 830 async_extent->start +
831 async_extent->ram_size - 1,
c2790a2e 832 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc
JB		 833 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
834 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
835 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
79787eaa 836 kfree(async_extent);
3e04e7f1 837 goto again;
771ed689
CM		 838}
839
4b46fce2
JB		 840static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
841 u64 num_bytes)
842{
843 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
844 struct extent_map *em;
845 u64 alloc_hint = 0;
846
847 read_lock(&em_tree->lock);
848 em = search_extent_mapping(em_tree, start, num_bytes);
849 if (em) {
850 /*
851 * if block start isn't an actual block number then find the
852 * first block in this inode and use that as a hint. If that
853 * block is also bogus then just don't worry about it.
854 */
855 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
856 free_extent_map(em);
857 em = search_extent_mapping(em_tree, 0, 0);
858 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
859 alloc_hint = em->block_start;
860 if (em)
861 free_extent_map(em);
862 } else {
863 alloc_hint = em->block_start;
864 free_extent_map(em);
865 }
866 }
867 read_unlock(&em_tree->lock);
868
869 return alloc_hint;
870}
871
771ed689
CM		 872/*
873 * when extent_io.c finds a delayed allocation range in the file,
874 * the call backs end up in this code. The basic idea is to
875 * allocate extents on disk for the range, and create ordered data structs
876 * in ram to track those extents.
877 *
878 * locked_page is the page that writepage had locked already. We use
879 * it to make sure we don't do extra locks or unlocks.
880 *
881 * *page_started is set to one if we unlock locked_page and do everything
882 * required to start IO on it. It may be clean and already done with
883 * IO when we return.
884 */
00361589
JB		 885static noinline int cow_file_range(struct inode *inode,
886 struct page *locked_page,
887 u64 start, u64 end, int *page_started,
888 unsigned long *nr_written,
889 int unlock)
771ed689 890{
00361589 891 struct btrfs_root *root = BTRFS_I(inode)->root;
771ed689
CM		 892 u64 alloc_hint = 0;
893 u64 num_bytes;
894 unsigned long ram_size;
895 u64 disk_num_bytes;
896 u64 cur_alloc_size;
897 u64 blocksize = root->sectorsize;
771ed689
CM		 898 struct btrfs_key ins;
899 struct extent_map *em;
900 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
901 int ret = 0;
902
02ecd2c2
JB		 903 if (btrfs_is_free_space_inode(inode)) {
904 WARN_ON_ONCE(1);
29bce2f3
JB		 905 ret = -EINVAL;
906 goto out_unlock;
02ecd2c2 907 }
771ed689 908
fda2832f 909 num_bytes = ALIGN(end - start + 1, blocksize);
771ed689
CM		 910 num_bytes = max(blocksize, num_bytes);
911 disk_num_bytes = num_bytes;
771ed689 912
4cb5300b 913 /* if this is a small write inside eof, kick off defrag */
4cb13e5d
LB		 914 if (num_bytes < 64 * 1024 &&
915 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
00361589 916 btrfs_add_inode_defrag(NULL, inode);
4cb5300b 917
771ed689
CM		 918 if (start == 0) {
919 /* lets try to make an inline extent */
00361589
JB		 920 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
921 NULL);
771ed689 922 if (ret == 0) {
c2790a2e
JB		 923 extent_clear_unlock_delalloc(inode, start, end, NULL,
924 EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc 925 EXTENT_DEFRAG, PAGE_UNLOCK |
c2790a2e
JB		 926 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
927 PAGE_END_WRITEBACK);
c2167754 928
771ed689
CM		 929 *nr_written = *nr_written +
930 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
931 *page_started = 1;
771ed689 932 goto out;
79787eaa 933 } else if (ret < 0) {
79787eaa 934 goto out_unlock;
771ed689
CM		 935 }
936 }
937
938 BUG_ON(disk_num_bytes >
6c41761f 939 btrfs_super_total_bytes(root->fs_info->super_copy));
771ed689 940
4b46fce2 941 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
771ed689
CM		 942 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
943
d397712b 944 while (disk_num_bytes > 0) {
a791e35e
CM		 945 unsigned long op;
946
287a0ab9 947 cur_alloc_size = disk_num_bytes;
00361589 948 ret = btrfs_reserve_extent(root, cur_alloc_size,
771ed689 949 root->sectorsize, 0, alloc_hint,
e570fd27 950 &ins, 1, 1);
00361589 951 if (ret < 0)
79787eaa 952 goto out_unlock;
d397712b 953
172ddd60 954 em = alloc_extent_map();
b9aa55be
LB		 955 if (!em) {
956 ret = -ENOMEM;
ace68bac 957 goto out_reserve;
b9aa55be 958 }
e6dcd2dc 959 em->start = start;
445a6944 960 em->orig_start = em->start;
771ed689
CM		 961 ram_size = ins.offset;
962 em->len = ins.offset;
2ab28f32
JB		 963 em->mod_start = em->start;
964 em->mod_len = em->len;
c8b97818 965
e6dcd2dc 966 em->block_start = ins.objectid;
c8b97818 967 em->block_len = ins.offset;
b4939680 968 em->orig_block_len = ins.offset;
cc95bef6 969 em->ram_bytes = ram_size;
e6dcd2dc 970 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 971 set_bit(EXTENT_FLAG_PINNED, &em->flags);
70c8a91c 972 em->generation = -1;
c8b97818 973
d397712b 974 while (1) {
890871be 975 write_lock(&em_tree->lock);
09a2a8f9 976 ret = add_extent_mapping(em_tree, em, 1);
890871be 977 write_unlock(&em_tree->lock);
e6dcd2dc
CM		 978 if (ret != -EEXIST) {
979 free_extent_map(em);
980 break;
981 }
982 btrfs_drop_extent_cache(inode, start,
c8b97818 983 start + ram_size - 1, 0);
e6dcd2dc 984 }
ace68bac
LB		 985 if (ret)
986 goto out_reserve;
e6dcd2dc 987
98d20f67 988 cur_alloc_size = ins.offset;
e6dcd2dc 989 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771ed689 990 ram_size, cur_alloc_size, 0);
ace68bac 991 if (ret)
d9f85963 992 goto out_drop_extent_cache;
c8b97818 993
17d217fe
YZ		 994 if (root->root_key.objectid ==
995 BTRFS_DATA_RELOC_TREE_OBJECTID) {
996 ret = btrfs_reloc_clone_csums(inode, start,
997 cur_alloc_size);
00361589 998 if (ret)
d9f85963 999 goto out_drop_extent_cache;
17d217fe
YZ		 1000 }
1001
d397712b 1002 if (disk_num_bytes < cur_alloc_size)
3b951516 1003 break;
d397712b 1004
c8b97818
CM		 1005 /* we're not doing compressed IO, don't unlock the first
1006 * page (which the caller expects to stay locked), don't
1007 * clear any dirty bits and don't set any writeback bits
8b62b72b
CM		 1008 *
1009 * Do set the Private2 bit so we know this page was properly
1010 * setup for writepage
c8b97818 1011 */
c2790a2e
JB		 1012 op = unlock ? PAGE_UNLOCK : 0;
1013 op |= PAGE_SET_PRIVATE2;
a791e35e 1014
c2790a2e
JB		 1015 extent_clear_unlock_delalloc(inode, start,
1016 start + ram_size - 1, locked_page,
1017 EXTENT_LOCKED | EXTENT_DELALLOC,
1018 op);
c8b97818 1019 disk_num_bytes -= cur_alloc_size;
c59f8951
CM		 1020 num_bytes -= cur_alloc_size;
1021 alloc_hint = ins.objectid + ins.offset;
1022 start += cur_alloc_size;
b888db2b 1023 }
79787eaa 1024out:
be20aa9d 1025 return ret;
b7d5b0a8 1026
d9f85963
FM		 1027out_drop_extent_cache:
1028 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
ace68bac 1029out_reserve:
e570fd27 1030 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
79787eaa 1031out_unlock:
c2790a2e 1032 extent_clear_unlock_delalloc(inode, start, end, locked_page,
151a41bc
JB		 1033 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1034 EXTENT_DELALLOC | EXTENT_DEFRAG,
1035 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1036 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
79787eaa 1037 goto out;
771ed689 1038}
c8b97818 1039
771ed689
CM		 1040/*
1041 * work queue call back to started compression on a file and pages
1042 */
1043static noinline void async_cow_start(struct btrfs_work *work)
1044{
1045 struct async_cow *async_cow;
1046 int num_added = 0;
1047 async_cow = container_of(work, struct async_cow, work);
1048
1049 compress_file_range(async_cow->inode, async_cow->locked_page,
1050 async_cow->start, async_cow->end, async_cow,
1051 &num_added);
8180ef88 1052 if (num_added == 0) {
cb77fcd8 1053 btrfs_add_delayed_iput(async_cow->inode);
771ed689 1054 async_cow->inode = NULL;
8180ef88 1055 }
771ed689
CM		 1056}
1057
1058/*
1059 * work queue call back to submit previously compressed pages
1060 */
1061static noinline void async_cow_submit(struct btrfs_work *work)
1062{
1063 struct async_cow *async_cow;
1064 struct btrfs_root *root;
1065 unsigned long nr_pages;
1066
1067 async_cow = container_of(work, struct async_cow, work);
1068
1069 root = async_cow->root;
1070 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1071 PAGE_CACHE_SHIFT;
1072
66657b31 1073 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
287082b0 1074 5 * 1024 * 1024 &&
771ed689
CM		 1075 waitqueue_active(&root->fs_info->async_submit_wait))
1076 wake_up(&root->fs_info->async_submit_wait);
1077
d397712b 1078 if (async_cow->inode)
771ed689 1079 submit_compressed_extents(async_cow->inode, async_cow);
771ed689 1080}
c8b97818 1081
771ed689
CM		 1082static noinline void async_cow_free(struct btrfs_work *work)
1083{
1084 struct async_cow *async_cow;
1085 async_cow = container_of(work, struct async_cow, work);
8180ef88 1086 if (async_cow->inode)
cb77fcd8 1087 btrfs_add_delayed_iput(async_cow->inode);
771ed689
CM		 1088 kfree(async_cow);
1089}
1090
1091static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1092 u64 start, u64 end, int *page_started,
1093 unsigned long *nr_written)
1094{
1095 struct async_cow *async_cow;
1096 struct btrfs_root *root = BTRFS_I(inode)->root;
1097 unsigned long nr_pages;
1098 u64 cur_end;
287082b0 1099 int limit = 10 * 1024 * 1024;
771ed689 1100
a3429ab7
CM		 1101 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1102 1, 0, NULL, GFP_NOFS);
d397712b 1103 while (start < end) {
771ed689 1104 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
79787eaa 1105 BUG_ON(!async_cow); /* -ENOMEM */
8180ef88 1106 async_cow->inode = igrab(inode);
771ed689
CM		 1107 async_cow->root = root;
1108 async_cow->locked_page = locked_page;
1109 async_cow->start = start;
1110
f79707b0
WS		 1111 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1112 !btrfs_test_opt(root, FORCE_COMPRESS))
771ed689
CM		 1113 cur_end = end;
1114 else
1115 cur_end = min(end, start + 512 * 1024 - 1);
1116
1117 async_cow->end = cur_end;
1118 INIT_LIST_HEAD(&async_cow->extents);
1119
9e0af237
LB		 1120 btrfs_init_work(&async_cow->work,
1121 btrfs_delalloc_helper,
1122 async_cow_start, async_cow_submit,
1123 async_cow_free);
771ed689 1124
771ed689
CM		 1125 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1126 PAGE_CACHE_SHIFT;
1127 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1128
afe3d242
QW		 1129 btrfs_queue_work(root->fs_info->delalloc_workers,
1130 &async_cow->work);
771ed689
CM		 1131
1132 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1133 wait_event(root->fs_info->async_submit_wait,
1134 (atomic_read(&root->fs_info->async_delalloc_pages) <
1135 limit));
1136 }
1137
d397712b 1138 while (atomic_read(&root->fs_info->async_submit_draining) &&
771ed689
CM		 1139 atomic_read(&root->fs_info->async_delalloc_pages)) {
1140 wait_event(root->fs_info->async_submit_wait,
1141 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1142 0));
1143 }
1144
1145 *nr_written += nr_pages;
1146 start = cur_end + 1;
1147 }
1148 *page_started = 1;
1149 return 0;
be20aa9d
CM		 1150}
1151
d397712b 1152static noinline int csum_exist_in_range(struct btrfs_root *root,
17d217fe
YZ		 1153 u64 bytenr, u64 num_bytes)
1154{
1155 int ret;
1156 struct btrfs_ordered_sum *sums;
1157 LIST_HEAD(list);
1158
07d400a6 1159 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
a2de733c 1160 bytenr + num_bytes - 1, &list, 0);
17d217fe
YZ		 1161 if (ret == 0 && list_empty(&list))
1162 return 0;
1163
1164 while (!list_empty(&list)) {
1165 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1166 list_del(&sums->list);
1167 kfree(sums);
1168 }
1169 return 1;
1170}
1171
d352ac68
CM		 1172/*
1173 * when nowcow writeback call back. This checks for snapshots or COW copies
1174 * of the extents that exist in the file, and COWs the file as required.
1175 *
1176 * If no cow copies or snapshots exist, we write directly to the existing
1177 * blocks on disk
1178 */
7f366cfe
CM		 1179static noinline int run_delalloc_nocow(struct inode *inode,
1180 struct page *locked_page,
771ed689
CM		 1181 u64 start, u64 end, int *page_started, int force,
1182 unsigned long *nr_written)
be20aa9d 1183{
be20aa9d 1184 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 1185 struct btrfs_trans_handle *trans;
be20aa9d 1186 struct extent_buffer *leaf;
be20aa9d 1187 struct btrfs_path *path;
80ff3856 1188 struct btrfs_file_extent_item *fi;
be20aa9d 1189 struct btrfs_key found_key;
80ff3856
YZ		 1190 u64 cow_start;
1191 u64 cur_offset;
1192 u64 extent_end;
5d4f98a2 1193 u64 extent_offset;
80ff3856
YZ		 1194 u64 disk_bytenr;
1195 u64 num_bytes;
b4939680 1196 u64 disk_num_bytes;
cc95bef6 1197 u64 ram_bytes;
80ff3856 1198 int extent_type;
79787eaa 1199 int ret, err;
d899e052 1200 int type;
80ff3856
YZ		 1201 int nocow;
1202 int check_prev = 1;
82d5902d 1203 bool nolock;
33345d01 1204 u64 ino = btrfs_ino(inode);
be20aa9d
CM		 1205
1206 path = btrfs_alloc_path();
17ca04af 1207 if (!path) {
c2790a2e
JB		 1208 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1209 EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc
JB		 1210 EXTENT_DO_ACCOUNTING |
1211 EXTENT_DEFRAG, PAGE_UNLOCK |
c2790a2e
JB		 1212 PAGE_CLEAR_DIRTY |
1213 PAGE_SET_WRITEBACK |
1214 PAGE_END_WRITEBACK);
d8926bb3 1215 return -ENOMEM;
17ca04af 1216 }
82d5902d 1217
83eea1f1 1218 nolock = btrfs_is_free_space_inode(inode);
82d5902d
LZ		 1219
1220 if (nolock)
7a7eaa40 1221 trans = btrfs_join_transaction_nolock(root);
82d5902d 1222 else
7a7eaa40 1223 trans = btrfs_join_transaction(root);
ff5714cc 1224
79787eaa 1225 if (IS_ERR(trans)) {
c2790a2e
JB		 1226 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1227 EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc
JB		 1228 EXTENT_DO_ACCOUNTING |
1229 EXTENT_DEFRAG, PAGE_UNLOCK |
c2790a2e
JB		 1230 PAGE_CLEAR_DIRTY |
1231 PAGE_SET_WRITEBACK |
1232 PAGE_END_WRITEBACK);
79787eaa
JM		 1233 btrfs_free_path(path);
1234 return PTR_ERR(trans);
1235 }
1236
74b21075 1237 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
be20aa9d 1238
80ff3856
YZ		 1239 cow_start = (u64)-1;
1240 cur_offset = start;
1241 while (1) {
33345d01 1242 ret = btrfs_lookup_file_extent(trans, root, path, ino,
80ff3856 1243 cur_offset, 0);
d788a349 1244 if (ret < 0)
79787eaa 1245 goto error;
80ff3856
YZ		 1246 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1247 leaf = path->nodes[0];
1248 btrfs_item_key_to_cpu(leaf, &found_key,
1249 path->slots[0] - 1);
33345d01 1250 if (found_key.objectid == ino &&
80ff3856
YZ		 1251 found_key.type == BTRFS_EXTENT_DATA_KEY)
1252 path->slots[0]--;
1253 }
1254 check_prev = 0;
1255next_slot:
1256 leaf = path->nodes[0];
1257 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1258 ret = btrfs_next_leaf(root, path);
d788a349 1259 if (ret < 0)
79787eaa 1260 goto error;
80ff3856
YZ		 1261 if (ret > 0)
1262 break;
1263 leaf = path->nodes[0];
1264 }
be20aa9d 1265
80ff3856
YZ		 1266 nocow = 0;
1267 disk_bytenr = 0;
17d217fe 1268 num_bytes = 0;
80ff3856
YZ		 1269 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1270
33345d01 1271 if (found_key.objectid > ino ||
80ff3856
YZ		 1272 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1273 found_key.offset > end)
1274 break;
1275
1276 if (found_key.offset > cur_offset) {
1277 extent_end = found_key.offset;
e9061e21 1278 extent_type = 0;
80ff3856
YZ		 1279 goto out_check;
1280 }
1281
1282 fi = btrfs_item_ptr(leaf, path->slots[0],
1283 struct btrfs_file_extent_item);
1284 extent_type = btrfs_file_extent_type(leaf, fi);
1285
cc95bef6 1286 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
d899e052
YZ		 1287 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1288 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856 1289 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5d4f98a2 1290 extent_offset = btrfs_file_extent_offset(leaf, fi);
80ff3856
YZ		 1291 extent_end = found_key.offset +
1292 btrfs_file_extent_num_bytes(leaf, fi);
b4939680
JB		 1293 disk_num_bytes =
1294 btrfs_file_extent_disk_num_bytes(leaf, fi);
80ff3856
YZ		 1295 if (extent_end <= start) {
1296 path->slots[0]++;
1297 goto next_slot;
1298 }
17d217fe
YZ		 1299 if (disk_bytenr == 0)
1300 goto out_check;
80ff3856
YZ		 1301 if (btrfs_file_extent_compression(leaf, fi) ||
1302 btrfs_file_extent_encryption(leaf, fi) ||
1303 btrfs_file_extent_other_encoding(leaf, fi))
1304 goto out_check;
d899e052
YZ		 1305 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1306 goto out_check;
d2fb3437 1307 if (btrfs_extent_readonly(root, disk_bytenr))
80ff3856 1308 goto out_check;
33345d01 1309 if (btrfs_cross_ref_exist(trans, root, ino,
5d4f98a2
YZ		 1310 found_key.offset -
1311 extent_offset, disk_bytenr))
17d217fe 1312 goto out_check;
5d4f98a2 1313 disk_bytenr += extent_offset;
17d217fe
YZ		 1314 disk_bytenr += cur_offset - found_key.offset;
1315 num_bytes = min(end + 1, extent_end) - cur_offset;
e9894fd3
WS		 1316 /*
1317 * if there are pending snapshots for this root,
1318 * we fall into common COW way.
1319 */
1320 if (!nolock) {
1321 err = btrfs_start_nocow_write(root);
1322 if (!err)
1323 goto out_check;
1324 }
17d217fe
YZ		 1325 /*
1326 * force cow if csum exists in the range.
1327 * this ensure that csum for a given extent are
1328 * either valid or do not exist.
1329 */
1330 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1331 goto out_check;
80ff3856
YZ		 1332 nocow = 1;
1333 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1334 extent_end = found_key.offset +
514ac8ad
CM		 1335 btrfs_file_extent_inline_len(leaf,
1336 path->slots[0], fi);
80ff3856
YZ		 1337 extent_end = ALIGN(extent_end, root->sectorsize);
1338 } else {
1339 BUG_ON(1);
1340 }
1341out_check:
1342 if (extent_end <= start) {
1343 path->slots[0]++;
e9894fd3
WS		 1344 if (!nolock && nocow)
1345 btrfs_end_nocow_write(root);
80ff3856
YZ		 1346 goto next_slot;
1347 }
1348 if (!nocow) {
1349 if (cow_start == (u64)-1)
1350 cow_start = cur_offset;
1351 cur_offset = extent_end;
1352 if (cur_offset > end)
1353 break;
1354 path->slots[0]++;
1355 goto next_slot;
7ea394f1
YZ		 1356 }
1357
b3b4aa74 1358 btrfs_release_path(path);
80ff3856 1359 if (cow_start != (u64)-1) {
00361589
JB		 1360 ret = cow_file_range(inode, locked_page,
1361 cow_start, found_key.offset - 1,
1362 page_started, nr_written, 1);
e9894fd3
WS		 1363 if (ret) {
1364 if (!nolock && nocow)
1365 btrfs_end_nocow_write(root);
79787eaa 1366 goto error;
e9894fd3 1367 }
80ff3856 1368 cow_start = (u64)-1;
7ea394f1 1369 }
80ff3856 1370
d899e052
YZ		 1371 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1372 struct extent_map *em;
1373 struct extent_map_tree *em_tree;
1374 em_tree = &BTRFS_I(inode)->extent_tree;
172ddd60 1375 em = alloc_extent_map();
79787eaa 1376 BUG_ON(!em); /* -ENOMEM */
d899e052 1377 em->start = cur_offset;
70c8a91c 1378 em->orig_start = found_key.offset - extent_offset;
d899e052
YZ		 1379 em->len = num_bytes;
1380 em->block_len = num_bytes;
1381 em->block_start = disk_bytenr;
b4939680 1382 em->orig_block_len = disk_num_bytes;
cc95bef6 1383 em->ram_bytes = ram_bytes;
d899e052 1384 em->bdev = root->fs_info->fs_devices->latest_bdev;
2ab28f32
JB		 1385 em->mod_start = em->start;
1386 em->mod_len = em->len;
d899e052 1387 set_bit(EXTENT_FLAG_PINNED, &em->flags);
b11e234d 1388 set_bit(EXTENT_FLAG_FILLING, &em->flags);
70c8a91c 1389 em->generation = -1;
d899e052 1390 while (1) {
890871be 1391 write_lock(&em_tree->lock);
09a2a8f9 1392 ret = add_extent_mapping(em_tree, em, 1);
890871be 1393 write_unlock(&em_tree->lock);
d899e052
YZ		 1394 if (ret != -EEXIST) {
1395 free_extent_map(em);
1396 break;
1397 }
1398 btrfs_drop_extent_cache(inode, em->start,
1399 em->start + em->len - 1, 0);
1400 }
1401 type = BTRFS_ORDERED_PREALLOC;
1402 } else {
1403 type = BTRFS_ORDERED_NOCOW;
1404 }
80ff3856
YZ		 1405
1406 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052 1407 num_bytes, num_bytes, type);
79787eaa 1408 BUG_ON(ret); /* -ENOMEM */
771ed689 1409
efa56464
YZ		 1410 if (root->root_key.objectid ==
1411 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1412 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1413 num_bytes);
e9894fd3
WS		 1414 if (ret) {
1415 if (!nolock && nocow)
1416 btrfs_end_nocow_write(root);
79787eaa 1417 goto error;
e9894fd3 1418 }
efa56464
YZ		 1419 }
1420
c2790a2e
JB		 1421 extent_clear_unlock_delalloc(inode, cur_offset,
1422 cur_offset + num_bytes - 1,
1423 locked_page, EXTENT_LOCKED |
1424 EXTENT_DELALLOC, PAGE_UNLOCK |
1425 PAGE_SET_PRIVATE2);
e9894fd3
WS		 1426 if (!nolock && nocow)
1427 btrfs_end_nocow_write(root);
80ff3856
YZ		 1428 cur_offset = extent_end;
1429 if (cur_offset > end)
1430 break;
be20aa9d 1431 }
b3b4aa74 1432 btrfs_release_path(path);
80ff3856 1433
17ca04af 1434 if (cur_offset <= end && cow_start == (u64)-1) {
80ff3856 1435 cow_start = cur_offset;
17ca04af
JB		 1436 cur_offset = end;
1437 }
1438
80ff3856 1439 if (cow_start != (u64)-1) {
00361589
JB		 1440 ret = cow_file_range(inode, locked_page, cow_start, end,
1441 page_started, nr_written, 1);
d788a349 1442 if (ret)
79787eaa 1443 goto error;
80ff3856
YZ		 1444 }
1445
79787eaa 1446error:
a698d075 1447 err = btrfs_end_transaction(trans, root);
79787eaa
JM		 1448 if (!ret)
1449 ret = err;
1450
17ca04af 1451 if (ret && cur_offset < end)
c2790a2e
JB		 1452 extent_clear_unlock_delalloc(inode, cur_offset, end,
1453 locked_page, EXTENT_LOCKED |
151a41bc
JB		 1454 EXTENT_DELALLOC | EXTENT_DEFRAG |
1455 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1456 PAGE_CLEAR_DIRTY |
c2790a2e
JB		 1457 PAGE_SET_WRITEBACK |
1458 PAGE_END_WRITEBACK);
7ea394f1 1459 btrfs_free_path(path);
79787eaa 1460 return ret;
be20aa9d
CM		 1461}
1462
47059d93
WS		 1463static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1464{
1465
1466 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1467 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1468 return 0;
1469
1470 /*
1471 * @defrag_bytes is a hint value, no spinlock held here,
1472 * if is not zero, it means the file is defragging.
1473 * Force cow if given extent needs to be defragged.
1474 */
1475 if (BTRFS_I(inode)->defrag_bytes &&
1476 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1477 EXTENT_DEFRAG, 0, NULL))
1478 return 1;
1479
1480 return 0;
1481}
1482
d352ac68
CM		 1483/*
1484 * extent_io.c call back to do delayed allocation processing
1485 */
c8b97818 1486static int run_delalloc_range(struct inode *inode, struct page *locked_page,
771ed689
CM		 1487 u64 start, u64 end, int *page_started,
1488 unsigned long *nr_written)
be20aa9d 1489{
be20aa9d 1490 int ret;
47059d93 1491 int force_cow = need_force_cow(inode, start, end);
a2135011 1492
47059d93 1493 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
c8b97818 1494 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1495 page_started, 1, nr_written);
47059d93 1496 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
d899e052 1497 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1498 page_started, 0, nr_written);
7816030e 1499 } else if (!inode_need_compress(inode)) {
7f366cfe
CM		 1500 ret = cow_file_range(inode, locked_page, start, end,
1501 page_started, nr_written, 1);
7ddf5a42
JB		 1502 } else {
1503 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1504 &BTRFS_I(inode)->runtime_flags);
771ed689 1505 ret = cow_file_range_async(inode, locked_page, start, end,
d397712b 1506 page_started, nr_written);
7ddf5a42 1507 }
b888db2b
CM		 1508 return ret;
1509}
1510
1bf85046
JM		 1511static void btrfs_split_extent_hook(struct inode *inode,
1512 struct extent_state *orig, u64 split)
9ed74f2d 1513{
0ca1f7ce 1514 /* not delalloc, ignore it */
9ed74f2d 1515 if (!(orig->state & EXTENT_DELALLOC))
1bf85046 1516 return;
9ed74f2d 1517
9e0baf60
JB		 1518 spin_lock(&BTRFS_I(inode)->lock);
1519 BTRFS_I(inode)->outstanding_extents++;
1520 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1521}
1522
1523/*
1524 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1525 * extents so we can keep track of new extents that are just merged onto old
1526 * extents, such as when we are doing sequential writes, so we can properly
1527 * account for the metadata space we'll need.
1528 */
1bf85046
JM		 1529static void btrfs_merge_extent_hook(struct inode *inode,
1530 struct extent_state *new,
1531 struct extent_state *other)
9ed74f2d 1532{
9ed74f2d
JB		 1533 /* not delalloc, ignore it */
1534 if (!(other->state & EXTENT_DELALLOC))
1bf85046 1535 return;
9ed74f2d 1536
9e0baf60
JB		 1537 spin_lock(&BTRFS_I(inode)->lock);
1538 BTRFS_I(inode)->outstanding_extents--;
1539 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1540}
1541
eb73c1b7
MX		 1542static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1543 struct inode *inode)
1544{
1545 spin_lock(&root->delalloc_lock);
1546 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1547 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1548 &root->delalloc_inodes);
1549 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1550 &BTRFS_I(inode)->runtime_flags);
1551 root->nr_delalloc_inodes++;
1552 if (root->nr_delalloc_inodes == 1) {
1553 spin_lock(&root->fs_info->delalloc_root_lock);
1554 BUG_ON(!list_empty(&root->delalloc_root));
1555 list_add_tail(&root->delalloc_root,
1556 &root->fs_info->delalloc_roots);
1557 spin_unlock(&root->fs_info->delalloc_root_lock);
1558 }
1559 }
1560 spin_unlock(&root->delalloc_lock);
1561}
1562
1563static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1564 struct inode *inode)
1565{
1566 spin_lock(&root->delalloc_lock);
1567 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1568 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1569 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1570 &BTRFS_I(inode)->runtime_flags);
1571 root->nr_delalloc_inodes--;
1572 if (!root->nr_delalloc_inodes) {
1573 spin_lock(&root->fs_info->delalloc_root_lock);
1574 BUG_ON(list_empty(&root->delalloc_root));
1575 list_del_init(&root->delalloc_root);
1576 spin_unlock(&root->fs_info->delalloc_root_lock);
1577 }
1578 }
1579 spin_unlock(&root->delalloc_lock);
1580}
1581
d352ac68
CM		 1582/*
1583 * extent_io.c set_bit_hook, used to track delayed allocation
1584 * bytes in this file, and to maintain the list of inodes that
1585 * have pending delalloc work to be done.
1586 */
1bf85046 1587static void btrfs_set_bit_hook(struct inode *inode,
41074888 1588 struct extent_state *state, unsigned long *bits)
291d673e 1589{
9ed74f2d 1590
47059d93
WS		 1591 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1592 WARN_ON(1);
75eff68e
CM		 1593 /*
1594 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1595 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1596 * bit, which is only set or cleared with irqs on
1597 */
0ca1f7ce 1598 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1599 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1600 u64 len = state->end + 1 - state->start;
83eea1f1 1601 bool do_list = !btrfs_is_free_space_inode(inode);
9ed74f2d 1602
9e0baf60 1603 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1604 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1605 } else {
1606 spin_lock(&BTRFS_I(inode)->lock);
1607 BTRFS_I(inode)->outstanding_extents++;
1608 spin_unlock(&BTRFS_I(inode)->lock);
1609 }
287a0ab9 1610
963d678b
MX		 1611 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1612 root->fs_info->delalloc_batch);
df0af1a5 1613 spin_lock(&BTRFS_I(inode)->lock);
0ca1f7ce 1614 BTRFS_I(inode)->delalloc_bytes += len;
47059d93
WS		 1615 if (*bits & EXTENT_DEFRAG)
1616 BTRFS_I(inode)->defrag_bytes += len;
df0af1a5 1617 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
eb73c1b7
MX		 1618 &BTRFS_I(inode)->runtime_flags))
1619 btrfs_add_delalloc_inodes(root, inode);
df0af1a5 1620 spin_unlock(&BTRFS_I(inode)->lock);
291d673e 1621 }
291d673e
CM		 1622}
1623
d352ac68
CM		 1624/*
1625 * extent_io.c clear_bit_hook, see set_bit_hook for why
1626 */
1bf85046 1627static void btrfs_clear_bit_hook(struct inode *inode,
41074888
DS		 1628 struct extent_state *state,
1629 unsigned long *bits)
291d673e 1630{
47059d93
WS		 1631 u64 len = state->end + 1 - state->start;
1632
1633 spin_lock(&BTRFS_I(inode)->lock);
1634 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1635 BTRFS_I(inode)->defrag_bytes -= len;
1636 spin_unlock(&BTRFS_I(inode)->lock);
1637
75eff68e
CM		 1638 /*
1639 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1640 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1641 * bit, which is only set or cleared with irqs on
1642 */
0ca1f7ce 1643 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1644 struct btrfs_root *root = BTRFS_I(inode)->root;
83eea1f1 1645 bool do_list = !btrfs_is_free_space_inode(inode);
bcbfce8a 1646
9e0baf60 1647 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1648 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1649 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1650 spin_lock(&BTRFS_I(inode)->lock);
1651 BTRFS_I(inode)->outstanding_extents--;
1652 spin_unlock(&BTRFS_I(inode)->lock);
1653 }
0ca1f7ce 1654
b6d08f06
JB		 1655 /*
1656 * We don't reserve metadata space for space cache inodes so we
1657 * don't need to call dellalloc_release_metadata if there is an
1658 * error.
1659 */
1660 if (*bits & EXTENT_DO_ACCOUNTING &&
1661 root != root->fs_info->tree_root)
0ca1f7ce
YZ		 1662 btrfs_delalloc_release_metadata(inode, len);
1663
0cb59c99 1664 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
7ee9e440 1665 && do_list && !(state->state & EXTENT_NORESERVE))
0ca1f7ce 1666 btrfs_free_reserved_data_space(inode, len);
9ed74f2d 1667
963d678b
MX		 1668 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1669 root->fs_info->delalloc_batch);
df0af1a5 1670 spin_lock(&BTRFS_I(inode)->lock);
0ca1f7ce 1671 BTRFS_I(inode)->delalloc_bytes -= len;
0cb59c99 1672 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
df0af1a5 1673 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
eb73c1b7
MX		 1674 &BTRFS_I(inode)->runtime_flags))
1675 btrfs_del_delalloc_inode(root, inode);
df0af1a5 1676 spin_unlock(&BTRFS_I(inode)->lock);
291d673e 1677 }
291d673e
CM		 1678}
1679
d352ac68
CM		 1680/*
1681 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1682 * we don't create bios that span stripes or chunks
1683 */
64a16701 1684int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
c8b97818
CM		 1685 size_t size, struct bio *bio,
1686 unsigned long bio_flags)
239b14b3
CM		 1687{
1688 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
4f024f37 1689 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
239b14b3
CM		 1690 u64 length = 0;
1691 u64 map_length;
239b14b3
CM		 1692 int ret;
1693
771ed689
CM		 1694 if (bio_flags & EXTENT_BIO_COMPRESSED)
1695 return 0;
1696
4f024f37 1697 length = bio->bi_iter.bi_size;
239b14b3 1698 map_length = length;
64a16701 1699 ret = btrfs_map_block(root->fs_info, rw, logical,
f188591e 1700 &map_length, NULL, 0);
3ec706c8 1701 /* Will always return 0 with map_multi == NULL */
3444a972 1702 BUG_ON(ret < 0);
d397712b 1703 if (map_length < length + size)
239b14b3 1704 return 1;
3444a972 1705 return 0;
239b14b3
CM		 1706}
1707
d352ac68
CM		 1708/*
1709 * in order to insert checksums into the metadata in large chunks,
1710 * we wait until bio submission time. All the pages in the bio are
1711 * checksummed and sums are attached onto the ordered extent record.
1712 *
1713 * At IO completion time the cums attached on the ordered extent record
1714 * are inserted into the btree
1715 */
d397712b
CM		 1716static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1717 struct bio *bio, int mirror_num,
eaf25d93
CM		 1718 unsigned long bio_flags,
1719 u64 bio_offset)
065631f6 1720{
065631f6 1721 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 1722 int ret = 0;
e015640f 1723
d20f7043 1724 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
79787eaa 1725 BUG_ON(ret); /* -ENOMEM */
4a69a410
CM		 1726 return 0;
1727}
e015640f 1728
4a69a410
CM		 1729/*
1730 * in order to insert checksums into the metadata in large chunks,
1731 * we wait until bio submission time. All the pages in the bio are
1732 * checksummed and sums are attached onto the ordered extent record.
1733 *
1734 * At IO completion time the cums attached on the ordered extent record
1735 * are inserted into the btree
1736 */
b2950863 1737static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1738 int mirror_num, unsigned long bio_flags,
1739 u64 bio_offset)
4a69a410
CM		 1740{
1741 struct btrfs_root *root = BTRFS_I(inode)->root;
61891923
SB		 1742 int ret;
1743
1744 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1745 if (ret)
1746 bio_endio(bio, ret);
1747 return ret;
44b8bd7e
CM		 1748}
1749
d352ac68 1750/*
cad321ad
CM		 1751 * extent_io.c submission hook. This does the right thing for csum calculation
1752 * on write, or reading the csums from the tree before a read
d352ac68 1753 */
b2950863 1754static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1755 int mirror_num, unsigned long bio_flags,
1756 u64 bio_offset)
44b8bd7e
CM		 1757{
1758 struct btrfs_root *root = BTRFS_I(inode)->root;
1759 int ret = 0;
19b9bdb0 1760 int skip_sum;
0417341e 1761 int metadata = 0;
b812ce28 1762 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
44b8bd7e 1763
6cbff00f 1764 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
cad321ad 1765
83eea1f1 1766 if (btrfs_is_free_space_inode(inode))
0417341e
JM		 1767 metadata = 2;
1768
7b6d91da 1769 if (!(rw & REQ_WRITE)) {
5fd02043
JB		 1770 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1771 if (ret)
61891923 1772 goto out;
5fd02043 1773
d20f7043 1774 if (bio_flags & EXTENT_BIO_COMPRESSED) {
61891923
SB		 1775 ret = btrfs_submit_compressed_read(inode, bio,
1776 mirror_num,
1777 bio_flags);
1778 goto out;
c2db1073
TI		 1779 } else if (!skip_sum) {
1780 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1781 if (ret)
61891923 1782 goto out;
c2db1073 1783 }
4d1b5fb4 1784 goto mapit;
b812ce28 1785 } else if (async && !skip_sum) {
17d217fe
YZ		 1786 /* csum items have already been cloned */
1787 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1788 goto mapit;
19b9bdb0 1789 /* we're doing a write, do the async checksumming */
61891923 1790 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 1791 inode, rw, bio, mirror_num,
eaf25d93
CM		 1792 bio_flags, bio_offset,
1793 __btrfs_submit_bio_start,
4a69a410 1794 __btrfs_submit_bio_done);
61891923 1795 goto out;
b812ce28
JB		 1796 } else if (!skip_sum) {
1797 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1798 if (ret)
1799 goto out;
19b9bdb0
CM		 1800 }
1801
0b86a832 1802mapit:
61891923
SB		 1803 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1804
1805out:
1806 if (ret < 0)
1807 bio_endio(bio, ret);
1808 return ret;
065631f6 1809}
6885f308 1810
d352ac68
CM		 1811/*
1812 * given a list of ordered sums record them in the inode. This happens
1813 * at IO completion time based on sums calculated at bio submission time.
1814 */
ba1da2f4 1815static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM		 1816 struct inode *inode, u64 file_offset,
1817 struct list_head *list)
1818{
e6dcd2dc
CM		 1819 struct btrfs_ordered_sum *sum;
1820
c6e30871 1821 list_for_each_entry(sum, list, list) {
39847c4d 1822 trans->adding_csums = 1;
d20f7043
CM		 1823 btrfs_csum_file_blocks(trans,
1824 BTRFS_I(inode)->root->fs_info->csum_root, sum);
39847c4d 1825 trans->adding_csums = 0;
e6dcd2dc
CM		 1826 }
1827 return 0;
1828}
1829
2ac55d41
JB		 1830int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1831 struct extent_state **cached_state)
ea8c2819 1832{
6c1500f2 1833 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
ea8c2819 1834 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2ac55d41 1835 cached_state, GFP_NOFS);
ea8c2819
CM		 1836}
1837
d352ac68 1838/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM		 1839struct btrfs_writepage_fixup {
1840 struct page *page;
1841 struct btrfs_work work;
1842};
1843
b2950863 1844static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
247e743c
CM		 1845{
1846 struct btrfs_writepage_fixup *fixup;
1847 struct btrfs_ordered_extent *ordered;
2ac55d41 1848 struct extent_state *cached_state = NULL;
247e743c
CM		 1849 struct page *page;
1850 struct inode *inode;
1851 u64 page_start;
1852 u64 page_end;
87826df0 1853 int ret;
247e743c
CM		 1854
1855 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1856 page = fixup->page;
4a096752 1857again:
247e743c
CM		 1858 lock_page(page);
1859 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1860 ClearPageChecked(page);
1861 goto out_page;
1862 }
1863
1864 inode = page->mapping->host;
1865 page_start = page_offset(page);
1866 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1867
2ac55d41 1868 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
d0082371 1869 &cached_state);
4a096752
CM		 1870
1871 /* already ordered? We're done */
8b62b72b 1872 if (PagePrivate2(page))
247e743c 1873 goto out;
4a096752
CM		 1874
1875 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1876 if (ordered) {
2ac55d41
JB		 1877 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1878 page_end, &cached_state, GFP_NOFS);
4a096752
CM		 1879 unlock_page(page);
1880 btrfs_start_ordered_extent(inode, ordered, 1);
87826df0 1881 btrfs_put_ordered_extent(ordered);
4a096752
CM		 1882 goto again;
1883 }
247e743c 1884
87826df0
JM		 1885 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1886 if (ret) {
1887 mapping_set_error(page->mapping, ret);
1888 end_extent_writepage(page, ret, page_start, page_end);
1889 ClearPageChecked(page);
1890 goto out;
1891 }
1892
2ac55d41 1893 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
247e743c 1894 ClearPageChecked(page);
87826df0 1895 set_page_dirty(page);
247e743c 1896out:
2ac55d41
JB		 1897 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1898 &cached_state, GFP_NOFS);
247e743c
CM		 1899out_page:
1900 unlock_page(page);
1901 page_cache_release(page);
b897abec 1902 kfree(fixup);
247e743c
CM		 1903}
1904
1905/*
1906 * There are a few paths in the higher layers of the kernel that directly
1907 * set the page dirty bit without asking the filesystem if it is a
1908 * good idea. This causes problems because we want to make sure COW
1909 * properly happens and the data=ordered rules are followed.
1910 *
c8b97818 1911 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 1912 * hasn't been properly setup for IO. We kick off an async process
1913 * to fix it up. The async helper will wait for ordered extents, set
1914 * the delalloc bit and make it safe to write the page.
1915 */
b2950863 1916static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
247e743c
CM		 1917{
1918 struct inode *inode = page->mapping->host;
1919 struct btrfs_writepage_fixup *fixup;
1920 struct btrfs_root *root = BTRFS_I(inode)->root;
247e743c 1921
8b62b72b
CM		 1922 /* this page is properly in the ordered list */
1923 if (TestClearPagePrivate2(page))
247e743c
CM		 1924 return 0;
1925
1926 if (PageChecked(page))
1927 return -EAGAIN;
1928
1929 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1930 if (!fixup)
1931 return -EAGAIN;
f421950f 1932
247e743c
CM		 1933 SetPageChecked(page);
1934 page_cache_get(page);
9e0af237
LB		 1935 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
1936 btrfs_writepage_fixup_worker, NULL, NULL);
247e743c 1937 fixup->page = page;
dc6e3209 1938 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
87826df0 1939 return -EBUSY;
247e743c
CM		 1940}
1941
d899e052
YZ		 1942static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1943 struct inode *inode, u64 file_pos,
1944 u64 disk_bytenr, u64 disk_num_bytes,
1945 u64 num_bytes, u64 ram_bytes,
1946 u8 compression, u8 encryption,
1947 u16 other_encoding, int extent_type)
1948{
1949 struct btrfs_root *root = BTRFS_I(inode)->root;
1950 struct btrfs_file_extent_item *fi;
1951 struct btrfs_path *path;
1952 struct extent_buffer *leaf;
1953 struct btrfs_key ins;
1acae57b 1954 int extent_inserted = 0;
d899e052
YZ		 1955 int ret;
1956
1957 path = btrfs_alloc_path();
d8926bb3
MF		 1958 if (!path)
1959 return -ENOMEM;
d899e052 1960
a1ed835e
CM		 1961 /*
1962 * we may be replacing one extent in the tree with another.
1963 * The new extent is pinned in the extent map, and we don't want
1964 * to drop it from the cache until it is completely in the btree.
1965 *
1966 * So, tell btrfs_drop_extents to leave this extent in the cache.
1967 * the caller is expected to unpin it and allow it to be merged
1968 * with the others.
1969 */
1acae57b
FDBM		 1970 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
1971 file_pos + num_bytes, NULL, 0,
1972 1, sizeof(*fi), &extent_inserted);
79787eaa
JM		 1973 if (ret)
1974 goto out;
d899e052 1975
1acae57b
FDBM		 1976 if (!extent_inserted) {
1977 ins.objectid = btrfs_ino(inode);
1978 ins.offset = file_pos;
1979 ins.type = BTRFS_EXTENT_DATA_KEY;
1980
1981 path->leave_spinning = 1;
1982 ret = btrfs_insert_empty_item(trans, root, path, &ins,
1983 sizeof(*fi));
1984 if (ret)
1985 goto out;
1986 }
d899e052
YZ		 1987 leaf = path->nodes[0];
1988 fi = btrfs_item_ptr(leaf, path->slots[0],
1989 struct btrfs_file_extent_item);
1990 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1991 btrfs_set_file_extent_type(leaf, fi, extent_type);
1992 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1993 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1994 btrfs_set_file_extent_offset(leaf, fi, 0);
1995 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1996 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1997 btrfs_set_file_extent_compression(leaf, fi, compression);
1998 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1999 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
b9473439 2000
d899e052 2001 btrfs_mark_buffer_dirty(leaf);
ce195332 2002 btrfs_release_path(path);
d899e052
YZ		 2003
2004 inode_add_bytes(inode, num_bytes);
d899e052
YZ		 2005
2006 ins.objectid = disk_bytenr;
2007 ins.offset = disk_num_bytes;
2008 ins.type = BTRFS_EXTENT_ITEM_KEY;
5d4f98a2
YZ		 2009 ret = btrfs_alloc_reserved_file_extent(trans, root,
2010 root->root_key.objectid,
33345d01 2011 btrfs_ino(inode), file_pos, &ins);
79787eaa 2012out:
d899e052 2013 btrfs_free_path(path);
b9473439 2014
79787eaa 2015 return ret;
d899e052
YZ		 2016}
2017
38c227d8
LB		 2018/* snapshot-aware defrag */
2019struct sa_defrag_extent_backref {
2020 struct rb_node node;
2021 struct old_sa_defrag_extent *old;
2022 u64 root_id;
2023 u64 inum;
2024 u64 file_pos;
2025 u64 extent_offset;
2026 u64 num_bytes;
2027 u64 generation;
2028};
2029
2030struct old_sa_defrag_extent {
2031 struct list_head list;
2032 struct new_sa_defrag_extent *new;
2033
2034 u64 extent_offset;
2035 u64 bytenr;
2036 u64 offset;
2037 u64 len;
2038 int count;
2039};
2040
2041struct new_sa_defrag_extent {
2042 struct rb_root root;
2043 struct list_head head;
2044 struct btrfs_path *path;
2045 struct inode *inode;
2046 u64 file_pos;
2047 u64 len;
2048 u64 bytenr;
2049 u64 disk_len;
2050 u8 compress_type;
2051};
2052
2053static int backref_comp(struct sa_defrag_extent_backref *b1,
2054 struct sa_defrag_extent_backref *b2)
2055{
2056 if (b1->root_id < b2->root_id)
2057 return -1;
2058 else if (b1->root_id > b2->root_id)
2059 return 1;
2060
2061 if (b1->inum < b2->inum)
2062 return -1;
2063 else if (b1->inum > b2->inum)
2064 return 1;
2065
2066 if (b1->file_pos < b2->file_pos)
2067 return -1;
2068 else if (b1->file_pos > b2->file_pos)
2069 return 1;
2070
2071 /*
2072 * [------------------------------] ===> (a range of space)
2073 * |<--->| |<---->| =============> (fs/file tree A)
2074 * |<---------------------------->| ===> (fs/file tree B)
2075 *
2076 * A range of space can refer to two file extents in one tree while
2077 * refer to only one file extent in another tree.
2078 *
2079 * So we may process a disk offset more than one time(two extents in A)
2080 * and locate at the same extent(one extent in B), then insert two same
2081 * backrefs(both refer to the extent in B).
2082 */
2083 return 0;
2084}
2085
2086static void backref_insert(struct rb_root *root,
2087 struct sa_defrag_extent_backref *backref)
2088{
2089 struct rb_node **p = &root->rb_node;
2090 struct rb_node *parent = NULL;
2091 struct sa_defrag_extent_backref *entry;
2092 int ret;
2093
2094 while (*p) {
2095 parent = *p;
2096 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2097
2098 ret = backref_comp(backref, entry);
2099 if (ret < 0)
2100 p = &(*p)->rb_left;
2101 else
2102 p = &(*p)->rb_right;
2103 }
2104
2105 rb_link_node(&backref->node, parent, p);
2106 rb_insert_color(&backref->node, root);
2107}
2108
2109/*
2110 * Note the backref might has changed, and in this case we just return 0.
2111 */
2112static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2113 void *ctx)
2114{
2115 struct btrfs_file_extent_item *extent;
2116 struct btrfs_fs_info *fs_info;
2117 struct old_sa_defrag_extent *old = ctx;
2118 struct new_sa_defrag_extent *new = old->new;
2119 struct btrfs_path *path = new->path;
2120 struct btrfs_key key;
2121 struct btrfs_root *root;
2122 struct sa_defrag_extent_backref *backref;
2123 struct extent_buffer *leaf;
2124 struct inode *inode = new->inode;
2125 int slot;
2126 int ret;
2127 u64 extent_offset;
2128 u64 num_bytes;
2129
2130 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2131 inum == btrfs_ino(inode))
2132 return 0;
2133
2134 key.objectid = root_id;
2135 key.type = BTRFS_ROOT_ITEM_KEY;
2136 key.offset = (u64)-1;
2137
2138 fs_info = BTRFS_I(inode)->root->fs_info;
2139 root = btrfs_read_fs_root_no_name(fs_info, &key);
2140 if (IS_ERR(root)) {
2141 if (PTR_ERR(root) == -ENOENT)
2142 return 0;
2143 WARN_ON(1);
2144 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2145 inum, offset, root_id);
2146 return PTR_ERR(root);
2147 }
2148
2149 key.objectid = inum;
2150 key.type = BTRFS_EXTENT_DATA_KEY;
2151 if (offset > (u64)-1 << 32)
2152 key.offset = 0;
2153 else
2154 key.offset = offset;
2155
2156 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
fae7f21c 2157 if (WARN_ON(ret < 0))
38c227d8 2158 return ret;
50f1319c 2159 ret = 0;
38c227d8
LB		 2160
2161 while (1) {
2162 cond_resched();
2163
2164 leaf = path->nodes[0];
2165 slot = path->slots[0];
2166
2167 if (slot >= btrfs_header_nritems(leaf)) {
2168 ret = btrfs_next_leaf(root, path);
2169 if (ret < 0) {
2170 goto out;
2171 } else if (ret > 0) {
2172 ret = 0;
2173 goto out;
2174 }
2175 continue;
2176 }
2177
2178 path->slots[0]++;
2179
2180 btrfs_item_key_to_cpu(leaf, &key, slot);
2181
2182 if (key.objectid > inum)
2183 goto out;
2184
2185 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2186 continue;
2187
2188 extent = btrfs_item_ptr(leaf, slot,
2189 struct btrfs_file_extent_item);
2190
2191 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2192 continue;
2193
e68afa49
LB		 2194 /*
2195 * 'offset' refers to the exact key.offset,
2196 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2197 * (key.offset - extent_offset).
2198 */
2199 if (key.offset != offset)
38c227d8
LB		 2200 continue;
2201
e68afa49 2202 extent_offset = btrfs_file_extent_offset(leaf, extent);
38c227d8 2203 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
e68afa49 2204
38c227d8
LB		 2205 if (extent_offset >= old->extent_offset + old->offset +
2206 old->len || extent_offset + num_bytes <=
2207 old->extent_offset + old->offset)
2208 continue;
38c227d8
LB		 2209 break;
2210 }
2211
2212 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2213 if (!backref) {
2214 ret = -ENOENT;
2215 goto out;
2216 }
2217
2218 backref->root_id = root_id;
2219 backref->inum = inum;
e68afa49 2220 backref->file_pos = offset;
38c227d8
LB		 2221 backref->num_bytes = num_bytes;
2222 backref->extent_offset = extent_offset;
2223 backref->generation = btrfs_file_extent_generation(leaf, extent);
2224 backref->old = old;
2225 backref_insert(&new->root, backref);
2226 old->count++;
2227out:
2228 btrfs_release_path(path);
2229 WARN_ON(ret);
2230 return ret;
2231}
2232
2233static noinline bool record_extent_backrefs(struct btrfs_path *path,
2234 struct new_sa_defrag_extent *new)
2235{
2236 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2237 struct old_sa_defrag_extent *old, *tmp;
2238 int ret;
2239
2240 new->path = path;
2241
2242 list_for_each_entry_safe(old, tmp, &new->head, list) {
e68afa49
LB		 2243 ret = iterate_inodes_from_logical(old->bytenr +
2244 old->extent_offset, fs_info,
38c227d8
LB		 2245 path, record_one_backref,
2246 old);
4724b106
JB		 2247 if (ret < 0 && ret != -ENOENT)
2248 return false;
38c227d8
LB		 2249
2250 /* no backref to be processed for this extent */
2251 if (!old->count) {
2252 list_del(&old->list);
2253 kfree(old);
2254 }
2255 }
2256
2257 if (list_empty(&new->head))
2258 return false;
2259
2260 return true;
2261}
2262
2263static int relink_is_mergable(struct extent_buffer *leaf,
2264 struct btrfs_file_extent_item *fi,
116e0024 2265 struct new_sa_defrag_extent *new)
38c227d8 2266{
116e0024 2267 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
38c227d8
LB		 2268 return 0;
2269
2270 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2271 return 0;
2272
116e0024
LB		 2273 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2274 return 0;
2275
2276 if (btrfs_file_extent_encryption(leaf, fi) ||
38c227d8
LB		 2277 btrfs_file_extent_other_encoding(leaf, fi))
2278 return 0;
2279
2280 return 1;
2281}
2282
2283/*
2284 * Note the backref might has changed, and in this case we just return 0.
2285 */
2286static noinline int relink_extent_backref(struct btrfs_path *path,
2287 struct sa_defrag_extent_backref *prev,
2288 struct sa_defrag_extent_backref *backref)
2289{
2290 struct btrfs_file_extent_item *extent;
2291 struct btrfs_file_extent_item *item;
2292 struct btrfs_ordered_extent *ordered;
2293 struct btrfs_trans_handle *trans;
2294 struct btrfs_fs_info *fs_info;
2295 struct btrfs_root *root;
2296 struct btrfs_key key;
2297 struct extent_buffer *leaf;
2298 struct old_sa_defrag_extent *old = backref->old;
2299 struct new_sa_defrag_extent *new = old->new;
2300 struct inode *src_inode = new->inode;
2301 struct inode *inode;
2302 struct extent_state *cached = NULL;
2303 int ret = 0;
2304 u64 start;
2305 u64 len;
2306 u64 lock_start;
2307 u64 lock_end;
2308 bool merge = false;
2309 int index;
2310
2311 if (prev && prev->root_id == backref->root_id &&
2312 prev->inum == backref->inum &&
2313 prev->file_pos + prev->num_bytes == backref->file_pos)
2314 merge = true;
2315
2316 /* step 1: get root */
2317 key.objectid = backref->root_id;
2318 key.type = BTRFS_ROOT_ITEM_KEY;
2319 key.offset = (u64)-1;
2320
2321 fs_info = BTRFS_I(src_inode)->root->fs_info;
2322 index = srcu_read_lock(&fs_info->subvol_srcu);
2323
2324 root = btrfs_read_fs_root_no_name(fs_info, &key);
2325 if (IS_ERR(root)) {
2326 srcu_read_unlock(&fs_info->subvol_srcu, index);
2327 if (PTR_ERR(root) == -ENOENT)
2328 return 0;
2329 return PTR_ERR(root);
2330 }
38c227d8 2331
bcbba5e6
WS		 2332 if (btrfs_root_readonly(root)) {
2333 srcu_read_unlock(&fs_info->subvol_srcu, index);
2334 return 0;
2335 }
2336
38c227d8
LB		 2337 /* step 2: get inode */
2338 key.objectid = backref->inum;
2339 key.type = BTRFS_INODE_ITEM_KEY;
2340 key.offset = 0;
2341
2342 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2343 if (IS_ERR(inode)) {
2344 srcu_read_unlock(&fs_info->subvol_srcu, index);
2345 return 0;
2346 }
2347
2348 srcu_read_unlock(&fs_info->subvol_srcu, index);
2349
2350 /* step 3: relink backref */
2351 lock_start = backref->file_pos;
2352 lock_end = backref->file_pos + backref->num_bytes - 1;
2353 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2354 0, &cached);
2355
2356 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2357 if (ordered) {
2358 btrfs_put_ordered_extent(ordered);
2359 goto out_unlock;
2360 }
2361
2362 trans = btrfs_join_transaction(root);
2363 if (IS_ERR(trans)) {
2364 ret = PTR_ERR(trans);
2365 goto out_unlock;
2366 }
2367
2368 key.objectid = backref->inum;
2369 key.type = BTRFS_EXTENT_DATA_KEY;
2370 key.offset = backref->file_pos;
2371
2372 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2373 if (ret < 0) {
2374 goto out_free_path;
2375 } else if (ret > 0) {
2376 ret = 0;
2377 goto out_free_path;
2378 }
2379
2380 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2381 struct btrfs_file_extent_item);
2382
2383 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2384 backref->generation)
2385 goto out_free_path;
2386
2387 btrfs_release_path(path);
2388
2389 start = backref->file_pos;
2390 if (backref->extent_offset < old->extent_offset + old->offset)
2391 start += old->extent_offset + old->offset -
2392 backref->extent_offset;
2393
2394 len = min(backref->extent_offset + backref->num_bytes,
2395 old->extent_offset + old->offset + old->len);
2396 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2397
2398 ret = btrfs_drop_extents(trans, root, inode, start,
2399 start + len, 1);
2400 if (ret)
2401 goto out_free_path;
2402again:
2403 key.objectid = btrfs_ino(inode);
2404 key.type = BTRFS_EXTENT_DATA_KEY;
2405 key.offset = start;
2406
a09a0a70 2407 path->leave_spinning = 1;
38c227d8
LB		 2408 if (merge) {
2409 struct btrfs_file_extent_item *fi;
2410 u64 extent_len;
2411 struct btrfs_key found_key;
2412
3c9665df 2413 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
38c227d8
LB		 2414 if (ret < 0)
2415 goto out_free_path;
2416
2417 path->slots[0]--;
2418 leaf = path->nodes[0];
2419 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2420
2421 fi = btrfs_item_ptr(leaf, path->slots[0],
2422 struct btrfs_file_extent_item);
2423 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2424
116e0024
LB		 2425 if (extent_len + found_key.offset == start &&
2426 relink_is_mergable(leaf, fi, new)) {
38c227d8
LB		 2427 btrfs_set_file_extent_num_bytes(leaf, fi,
2428 extent_len + len);
2429 btrfs_mark_buffer_dirty(leaf);
2430 inode_add_bytes(inode, len);
2431
2432 ret = 1;
2433 goto out_free_path;
2434 } else {
2435 merge = false;
2436 btrfs_release_path(path);
2437 goto again;
2438 }
2439 }
2440
2441 ret = btrfs_insert_empty_item(trans, root, path, &key,
2442 sizeof(*extent));
2443 if (ret) {
2444 btrfs_abort_transaction(trans, root, ret);
2445 goto out_free_path;
2446 }
2447
2448 leaf = path->nodes[0];
2449 item = btrfs_item_ptr(leaf, path->slots[0],
2450 struct btrfs_file_extent_item);
2451 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2452 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2453 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2454 btrfs_set_file_extent_num_bytes(leaf, item, len);
2455 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2456 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2457 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2458 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2459 btrfs_set_file_extent_encryption(leaf, item, 0);
2460 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2461
2462 btrfs_mark_buffer_dirty(leaf);
2463 inode_add_bytes(inode, len);
a09a0a70 2464 btrfs_release_path(path);
38c227d8
LB		 2465
2466 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2467 new->disk_len, 0,
2468 backref->root_id, backref->inum,
2469 new->file_pos, 0); /* start - extent_offset */
2470 if (ret) {
2471 btrfs_abort_transaction(trans, root, ret);
2472 goto out_free_path;
2473 }
2474
2475 ret = 1;
2476out_free_path:
2477 btrfs_release_path(path);
a09a0a70 2478 path->leave_spinning = 0;
38c227d8
LB		 2479 btrfs_end_transaction(trans, root);
2480out_unlock:
2481 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2482 &cached, GFP_NOFS);
2483 iput(inode);
2484 return ret;
2485}
2486
6f519564
LB		 2487static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2488{
2489 struct old_sa_defrag_extent *old, *tmp;
2490
2491 if (!new)
2492 return;
2493
2494 list_for_each_entry_safe(old, tmp, &new->head, list) {
2495 list_del(&old->list);
2496 kfree(old);
2497 }
2498 kfree(new);
2499}
2500
38c227d8
LB		 2501static void relink_file_extents(struct new_sa_defrag_extent *new)
2502{
2503 struct btrfs_path *path;
38c227d8
LB		 2504 struct sa_defrag_extent_backref *backref;
2505 struct sa_defrag_extent_backref *prev = NULL;
2506 struct inode *inode;
2507 struct btrfs_root *root;
2508 struct rb_node *node;
2509 int ret;
2510
2511 inode = new->inode;
2512 root = BTRFS_I(inode)->root;
2513
2514 path = btrfs_alloc_path();
2515 if (!path)
2516 return;
2517
2518 if (!record_extent_backrefs(path, new)) {
2519 btrfs_free_path(path);
2520 goto out;
2521 }
2522 btrfs_release_path(path);
2523
2524 while (1) {
2525 node = rb_first(&new->root);
2526 if (!node)
2527 break;
2528 rb_erase(node, &new->root);
2529
2530 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2531
2532 ret = relink_extent_backref(path, prev, backref);
2533 WARN_ON(ret < 0);
2534
2535 kfree(prev);
2536
2537 if (ret == 1)
2538 prev = backref;
2539 else
2540 prev = NULL;
2541 cond_resched();
2542 }
2543 kfree(prev);
2544
2545 btrfs_free_path(path);
38c227d8 2546out:
6f519564
LB		 2547 free_sa_defrag_extent(new);
2548
38c227d8
LB		 2549 atomic_dec(&root->fs_info->defrag_running);
2550 wake_up(&root->fs_info->transaction_wait);
38c227d8
LB		 2551}
2552
2553static struct new_sa_defrag_extent *
2554record_old_file_extents(struct inode *inode,
2555 struct btrfs_ordered_extent *ordered)
2556{
2557 struct btrfs_root *root = BTRFS_I(inode)->root;
2558 struct btrfs_path *path;
2559 struct btrfs_key key;
6f519564 2560 struct old_sa_defrag_extent *old;
38c227d8
LB		 2561 struct new_sa_defrag_extent *new;
2562 int ret;
2563
2564 new = kmalloc(sizeof(*new), GFP_NOFS);
2565 if (!new)
2566 return NULL;
2567
2568 new->inode = inode;
2569 new->file_pos = ordered->file_offset;
2570 new->len = ordered->len;
2571 new->bytenr = ordered->start;
2572 new->disk_len = ordered->disk_len;
2573 new->compress_type = ordered->compress_type;
2574 new->root = RB_ROOT;
2575 INIT_LIST_HEAD(&new->head);
2576
2577 path = btrfs_alloc_path();
2578 if (!path)
2579 goto out_kfree;
2580
2581 key.objectid = btrfs_ino(inode);
2582 key.type = BTRFS_EXTENT_DATA_KEY;
2583 key.offset = new->file_pos;
2584
2585 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2586 if (ret < 0)
2587 goto out_free_path;
2588 if (ret > 0 && path->slots[0] > 0)
2589 path->slots[0]--;
2590
2591 /* find out all the old extents for the file range */
2592 while (1) {
2593 struct btrfs_file_extent_item *extent;
2594 struct extent_buffer *l;
2595 int slot;
2596 u64 num_bytes;
2597 u64 offset;
2598 u64 end;
2599 u64 disk_bytenr;
2600 u64 extent_offset;
2601
2602 l = path->nodes[0];
2603 slot = path->slots[0];
2604
2605 if (slot >= btrfs_header_nritems(l)) {
2606 ret = btrfs_next_leaf(root, path);
2607 if (ret < 0)
6f519564 2608 goto out_free_path;
38c227d8
LB		 2609 else if (ret > 0)
2610 break;
2611 continue;
2612 }
2613
2614 btrfs_item_key_to_cpu(l, &key, slot);
2615
2616 if (key.objectid != btrfs_ino(inode))
2617 break;
2618 if (key.type != BTRFS_EXTENT_DATA_KEY)
2619 break;
2620 if (key.offset >= new->file_pos + new->len)
2621 break;
2622
2623 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2624
2625 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2626 if (key.offset + num_bytes < new->file_pos)
2627 goto next;
2628
2629 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2630 if (!disk_bytenr)
2631 goto next;
2632
2633 extent_offset = btrfs_file_extent_offset(l, extent);
2634
2635 old = kmalloc(sizeof(*old), GFP_NOFS);
2636 if (!old)
6f519564 2637 goto out_free_path;
38c227d8
LB		 2638
2639 offset = max(new->file_pos, key.offset);
2640 end = min(new->file_pos + new->len, key.offset + num_bytes);
2641
2642 old->bytenr = disk_bytenr;
2643 old->extent_offset = extent_offset;
2644 old->offset = offset - key.offset;
2645 old->len = end - offset;
2646 old->new = new;
2647 old->count = 0;
2648 list_add_tail(&old->list, &new->head);
2649next:
2650 path->slots[0]++;
2651 cond_resched();
2652 }
2653
2654 btrfs_free_path(path);
2655 atomic_inc(&root->fs_info->defrag_running);
2656
2657 return new;
2658
38c227d8
LB		 2659out_free_path:
2660 btrfs_free_path(path);
2661out_kfree:
6f519564 2662 free_sa_defrag_extent(new);
38c227d8
LB		 2663 return NULL;
2664}
2665
e570fd27
MX		 2666static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2667 u64 start, u64 len)
2668{
2669 struct btrfs_block_group_cache *cache;
2670
2671 cache = btrfs_lookup_block_group(root->fs_info, start);
2672 ASSERT(cache);
2673
2674 spin_lock(&cache->lock);
2675 cache->delalloc_bytes -= len;
2676 spin_unlock(&cache->lock);
2677
2678 btrfs_put_block_group(cache);
2679}
2680
d352ac68
CM		 2681/* as ordered data IO finishes, this gets called so we can finish
2682 * an ordered extent if the range of bytes in the file it covers are
2683 * fully written.
2684 */
5fd02043 2685static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
e6dcd2dc 2686{
5fd02043 2687 struct inode *inode = ordered_extent->inode;
e6dcd2dc 2688 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 2689 struct btrfs_trans_handle *trans = NULL;
e6dcd2dc 2690 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2ac55d41 2691 struct extent_state *cached_state = NULL;
38c227d8 2692 struct new_sa_defrag_extent *new = NULL;
261507a0 2693 int compress_type = 0;
77cef2ec
JB		 2694 int ret = 0;
2695 u64 logical_len = ordered_extent->len;
82d5902d 2696 bool nolock;
77cef2ec 2697 bool truncated = false;
e6dcd2dc 2698
83eea1f1 2699 nolock = btrfs_is_free_space_inode(inode);
0cb59c99 2700
5fd02043
JB		 2701 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2702 ret = -EIO;
2703 goto out;
2704 }
2705
77cef2ec
JB		 2706 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2707 truncated = true;
2708 logical_len = ordered_extent->truncated_len;
2709 /* Truncated the entire extent, don't bother adding */
2710 if (!logical_len)
2711 goto out;
2712 }
2713
c2167754 2714 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
79787eaa 2715 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
6c760c07
JB		 2716 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2717 if (nolock)
2718 trans = btrfs_join_transaction_nolock(root);
2719 else
2720 trans = btrfs_join_transaction(root);
2721 if (IS_ERR(trans)) {
2722 ret = PTR_ERR(trans);
2723 trans = NULL;
2724 goto out;
c2167754 2725 }
6c760c07
JB		 2726 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2727 ret = btrfs_update_inode_fallback(trans, root, inode);
2728 if (ret) /* -ENOMEM or corruption */
2729 btrfs_abort_transaction(trans, root, ret);
c2167754
YZ		 2730 goto out;
2731 }
e6dcd2dc 2732
2ac55d41
JB		 2733 lock_extent_bits(io_tree, ordered_extent->file_offset,
2734 ordered_extent->file_offset + ordered_extent->len - 1,
d0082371 2735 0, &cached_state);
e6dcd2dc 2736
38c227d8
LB		 2737 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2738 ordered_extent->file_offset + ordered_extent->len - 1,
2739 EXTENT_DEFRAG, 1, cached_state);
2740 if (ret) {
2741 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
8101c8db 2742 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
38c227d8
LB		 2743 /* the inode is shared */
2744 new = record_old_file_extents(inode, ordered_extent);
2745
2746 clear_extent_bit(io_tree, ordered_extent->file_offset,
2747 ordered_extent->file_offset + ordered_extent->len - 1,
2748 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2749 }
2750
0cb59c99 2751 if (nolock)
7a7eaa40 2752 trans = btrfs_join_transaction_nolock(root);
0cb59c99 2753 else
7a7eaa40 2754 trans = btrfs_join_transaction(root);
79787eaa
JM		 2755 if (IS_ERR(trans)) {
2756 ret = PTR_ERR(trans);
2757 trans = NULL;
2758 goto out_unlock;
2759 }
a79b7d4b 2760
0ca1f7ce 2761 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
c2167754 2762
c8b97818 2763 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
261507a0 2764 compress_type = ordered_extent->compress_type;
d899e052 2765 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
261507a0 2766 BUG_ON(compress_type);
920bbbfb 2767 ret = btrfs_mark_extent_written(trans, inode,
d899e052
YZ		 2768 ordered_extent->file_offset,
2769 ordered_extent->file_offset +
77cef2ec 2770 logical_len);
d899e052 2771 } else {
0af3d00b 2772 BUG_ON(root == root->fs_info->tree_root);
d899e052
YZ		 2773 ret = insert_reserved_file_extent(trans, inode,
2774 ordered_extent->file_offset,
2775 ordered_extent->start,
2776 ordered_extent->disk_len,
77cef2ec 2777 logical_len, logical_len,
261507a0 2778 compress_type, 0, 0,
d899e052 2779 BTRFS_FILE_EXTENT_REG);
e570fd27
MX		 2780 if (!ret)
2781 btrfs_release_delalloc_bytes(root,
2782 ordered_extent->start,
2783 ordered_extent->disk_len);
d899e052 2784 }
5dc562c5
JB		 2785 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2786 ordered_extent->file_offset, ordered_extent->len,
2787 trans->transid);
79787eaa
JM		 2788 if (ret < 0) {
2789 btrfs_abort_transaction(trans, root, ret);
5fd02043 2790 goto out_unlock;
79787eaa 2791 }
2ac55d41 2792
e6dcd2dc
CM		 2793 add_pending_csums(trans, inode, ordered_extent->file_offset,
2794 &ordered_extent->list);
2795
6c760c07
JB		 2796 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2797 ret = btrfs_update_inode_fallback(trans, root, inode);
2798 if (ret) { /* -ENOMEM or corruption */
2799 btrfs_abort_transaction(trans, root, ret);
2800 goto out_unlock;
1ef30be1
JB		 2801 }
2802 ret = 0;
5fd02043
JB		 2803out_unlock:
2804 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2805 ordered_extent->file_offset +
2806 ordered_extent->len - 1, &cached_state, GFP_NOFS);
c2167754 2807out:
5b0e95bf 2808 if (root != root->fs_info->tree_root)
0cb59c99 2809 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
a698d075
MX		 2810 if (trans)
2811 btrfs_end_transaction(trans, root);
0cb59c99 2812
77cef2ec
JB		 2813 if (ret || truncated) {
2814 u64 start, end;
2815
2816 if (truncated)
2817 start = ordered_extent->file_offset + logical_len;
2818 else
2819 start = ordered_extent->file_offset;
2820 end = ordered_extent->file_offset + ordered_extent->len - 1;
2821 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2822
2823 /* Drop the cache for the part of the extent we didn't write. */
2824 btrfs_drop_extent_cache(inode, start, end, 0);
5fd02043 2825
0bec9ef5
JB		 2826 /*
2827 * If the ordered extent had an IOERR or something else went
2828 * wrong we need to return the space for this ordered extent
77cef2ec
JB		 2829 * back to the allocator. We only free the extent in the
2830 * truncated case if we didn't write out the extent at all.
0bec9ef5 2831 */
77cef2ec
JB		 2832 if ((ret || !logical_len) &&
2833 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
0bec9ef5
JB		 2834 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2835 btrfs_free_reserved_extent(root, ordered_extent->start,
e570fd27 2836 ordered_extent->disk_len, 1);
0bec9ef5
JB		 2837 }
2838
2839
5fd02043 2840 /*
8bad3c02
LB		 2841 * This needs to be done to make sure anybody waiting knows we are done
2842 * updating everything for this ordered extent.
5fd02043
JB		 2843 */
2844 btrfs_remove_ordered_extent(inode, ordered_extent);
2845
38c227d8 2846 /* for snapshot-aware defrag */
6f519564
LB		 2847 if (new) {
2848 if (ret) {
2849 free_sa_defrag_extent(new);
2850 atomic_dec(&root->fs_info->defrag_running);
2851 } else {
2852 relink_file_extents(new);
2853 }
2854 }
38c227d8 2855
e6dcd2dc
CM		 2856 /* once for us */
2857 btrfs_put_ordered_extent(ordered_extent);
2858 /* once for the tree */
2859 btrfs_put_ordered_extent(ordered_extent);
2860
5fd02043
JB		 2861 return ret;
2862}
2863
2864static void finish_ordered_fn(struct btrfs_work *work)
2865{
2866 struct btrfs_ordered_extent *ordered_extent;
2867 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2868 btrfs_finish_ordered_io(ordered_extent);
e6dcd2dc
CM		 2869}
2870
b2950863 2871static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM		 2872 struct extent_state *state, int uptodate)
2873{
5fd02043
JB		 2874 struct inode *inode = page->mapping->host;
2875 struct btrfs_root *root = BTRFS_I(inode)->root;
2876 struct btrfs_ordered_extent *ordered_extent = NULL;
9e0af237
LB		 2877 struct btrfs_workqueue *wq;
2878 btrfs_work_func_t func;
5fd02043 2879
1abe9b8a 2880 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2881
8b62b72b 2882 ClearPagePrivate2(page);
5fd02043
JB		 2883 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2884 end - start + 1, uptodate))
2885 return 0;
2886
9e0af237
LB		 2887 if (btrfs_is_free_space_inode(inode)) {
2888 wq = root->fs_info->endio_freespace_worker;
2889 func = btrfs_freespace_write_helper;
2890 } else {
2891 wq = root->fs_info->endio_write_workers;
2892 func = btrfs_endio_write_helper;
2893 }
5fd02043 2894
9e0af237
LB		 2895 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
2896 NULL);
2897 btrfs_queue_work(wq, &ordered_extent->work);
5fd02043
JB		 2898
2899 return 0;
211f90e6
CM		 2900}
2901
d352ac68
CM		 2902/*
2903 * when reads are done, we need to check csums to verify the data is correct
4a54c8c1
JS		 2904 * if there's a match, we allow the bio to finish. If not, the code in
2905 * extent_io.c will try to find good copies for us.
d352ac68 2906 */
facc8a22
MX		 2907static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2908 u64 phy_offset, struct page *page,
2909 u64 start, u64 end, int mirror)
07157aac 2910{
4eee4fa4 2911 size_t offset = start - page_offset(page);
07157aac 2912 struct inode *inode = page->mapping->host;
d1310b2e 2913 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
07157aac 2914 char *kaddr;
ff79f819 2915 struct btrfs_root *root = BTRFS_I(inode)->root;
facc8a22 2916 u32 csum_expected;
ff79f819 2917 u32 csum = ~(u32)0;
c2cf52eb
SK		 2918 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2919 DEFAULT_RATELIMIT_BURST);
d1310b2e 2920
d20f7043
CM		 2921 if (PageChecked(page)) {
2922 ClearPageChecked(page);
2923 goto good;
2924 }
6cbff00f
CH		 2925
2926 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
08d2f347 2927 goto good;
17d217fe
YZ		 2928
2929 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
9655d298 2930 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
17d217fe
YZ		 2931 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2932 GFP_NOFS);
b6cda9bc 2933 return 0;
17d217fe 2934 }
d20f7043 2935
facc8a22
MX		 2936 phy_offset >>= inode->i_sb->s_blocksize_bits;
2937 csum_expected = *(((u32 *)io_bio->csum) + phy_offset);
d397712b 2938
facc8a22 2939 kaddr = kmap_atomic(page);
b0496686 2940 csum = btrfs_csum_data(kaddr + offset, csum, end - start + 1);
ff79f819 2941 btrfs_csum_final(csum, (char *)&csum);
facc8a22 2942 if (csum != csum_expected)
07157aac 2943 goto zeroit;
d397712b 2944
7ac687d9 2945 kunmap_atomic(kaddr);
d20f7043 2946good:
07157aac
CM		 2947 return 0;
2948
2949zeroit:
c2cf52eb 2950 if (__ratelimit(&_rs))
facc8a22 2951 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
c1c9ff7c 2952 btrfs_ino(page->mapping->host), start, csum, csum_expected);
db94535d
CM		 2953 memset(kaddr + offset, 1, end - start + 1);
2954 flush_dcache_page(page);
7ac687d9 2955 kunmap_atomic(kaddr);
facc8a22 2956 if (csum_expected == 0)
3b951516 2957 return 0;
7e38326f 2958 return -EIO;
07157aac 2959}
b888db2b 2960
24bbcf04
YZ		 2961struct delayed_iput {
2962 struct list_head list;
2963 struct inode *inode;
2964};
2965
79787eaa
JM		 2966/* JDM: If this is fs-wide, why can't we add a pointer to
2967 * btrfs_inode instead and avoid the allocation? */
24bbcf04
YZ		 2968void btrfs_add_delayed_iput(struct inode *inode)
2969{
2970 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2971 struct delayed_iput *delayed;
2972
2973 if (atomic_add_unless(&inode->i_count, -1, 1))
2974 return;
2975
2976 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2977 delayed->inode = inode;
2978
2979 spin_lock(&fs_info->delayed_iput_lock);
2980 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2981 spin_unlock(&fs_info->delayed_iput_lock);
2982}
2983
2984void btrfs_run_delayed_iputs(struct btrfs_root *root)
2985{
2986 LIST_HEAD(list);
2987 struct btrfs_fs_info *fs_info = root->fs_info;
2988 struct delayed_iput *delayed;
2989 int empty;
2990
2991 spin_lock(&fs_info->delayed_iput_lock);
2992 empty = list_empty(&fs_info->delayed_iputs);
2993 spin_unlock(&fs_info->delayed_iput_lock);
2994 if (empty)
2995 return;
2996
24bbcf04
YZ		 2997 spin_lock(&fs_info->delayed_iput_lock);
2998 list_splice_init(&fs_info->delayed_iputs, &list);
2999 spin_unlock(&fs_info->delayed_iput_lock);
3000
3001 while (!list_empty(&list)) {
3002 delayed = list_entry(list.next, struct delayed_iput, list);
3003 list_del(&delayed->list);
3004 iput(delayed->inode);
3005 kfree(delayed);
3006 }
24bbcf04
YZ		 3007}
3008
d68fc57b 3009/*
42b2aa86 3010 * This is called in transaction commit time. If there are no orphan
d68fc57b
YZ		 3011 * files in the subvolume, it removes orphan item and frees block_rsv
3012 * structure.
3013 */
3014void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3015 struct btrfs_root *root)
3016{
90290e19 3017 struct btrfs_block_rsv *block_rsv;
d68fc57b
YZ		 3018 int ret;
3019
8a35d95f 3020 if (atomic_read(&root->orphan_inodes) ||
d68fc57b
YZ		 3021 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3022 return;
3023
90290e19 3024 spin_lock(&root->orphan_lock);
8a35d95f 3025 if (atomic_read(&root->orphan_inodes)) {
90290e19
JB		 3026 spin_unlock(&root->orphan_lock);
3027 return;
3028 }
3029
3030 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3031 spin_unlock(&root->orphan_lock);
3032 return;
3033 }
3034
3035 block_rsv = root->orphan_block_rsv;
3036 root->orphan_block_rsv = NULL;
3037 spin_unlock(&root->orphan_lock);
3038
27cdeb70 3039 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
d68fc57b
YZ		 3040 btrfs_root_refs(&root->root_item) > 0) {
3041 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3042 root->root_key.objectid);
4ef31a45
JB		 3043 if (ret)
3044 btrfs_abort_transaction(trans, root, ret);
3045 else
27cdeb70
MX		 3046 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3047 &root->state);
d68fc57b
YZ		 3048 }
3049
90290e19
JB		 3050 if (block_rsv) {
3051 WARN_ON(block_rsv->size > 0);
3052 btrfs_free_block_rsv(root, block_rsv);
d68fc57b
YZ		 3053 }
3054}
3055
7b128766
JB		 3056/*
3057 * This creates an orphan entry for the given inode in case something goes
3058 * wrong in the middle of an unlink/truncate.
d68fc57b
YZ		 3059 *
3060 * NOTE: caller of this function should reserve 5 units of metadata for
3061 * this function.
7b128766
JB		 3062 */
3063int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3064{
3065 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 3066 struct btrfs_block_rsv *block_rsv = NULL;
3067 int reserve = 0;
3068 int insert = 0;
3069 int ret;
7b128766 3070
d68fc57b 3071 if (!root->orphan_block_rsv) {
66d8f3dd 3072 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
b532402e
TI		 3073 if (!block_rsv)
3074 return -ENOMEM;
d68fc57b 3075 }
7b128766 3076
d68fc57b
YZ		 3077 spin_lock(&root->orphan_lock);
3078 if (!root->orphan_block_rsv) {
3079 root->orphan_block_rsv = block_rsv;
3080 } else if (block_rsv) {
3081 btrfs_free_block_rsv(root, block_rsv);
3082 block_rsv = NULL;
7b128766 3083 }
7b128766 3084
8a35d95f
JB		 3085 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3086 &BTRFS_I(inode)->runtime_flags)) {
d68fc57b
YZ		 3087#if 0
3088 /*
3089 * For proper ENOSPC handling, we should do orphan
3090 * cleanup when mounting. But this introduces backward
3091 * compatibility issue.
3092 */
3093 if (!xchg(&root->orphan_item_inserted, 1))
3094 insert = 2;
3095 else
3096 insert = 1;
3097#endif
3098 insert = 1;
321f0e70 3099 atomic_inc(&root->orphan_inodes);
7b128766
JB		 3100 }
3101
72ac3c0d
JB		 3102 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3103 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3104 reserve = 1;
d68fc57b 3105 spin_unlock(&root->orphan_lock);
7b128766 3106
d68fc57b
YZ		 3107 /* grab metadata reservation from transaction handle */
3108 if (reserve) {
3109 ret = btrfs_orphan_reserve_metadata(trans, inode);
79787eaa 3110 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
d68fc57b 3111 }
7b128766 3112
d68fc57b
YZ		 3113 /* insert an orphan item to track this unlinked/truncated file */
3114 if (insert >= 1) {
33345d01 3115 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
4ef31a45 3116 if (ret) {
703c88e0 3117 atomic_dec(&root->orphan_inodes);
4ef31a45
JB		 3118 if (reserve) {
3119 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3120 &BTRFS_I(inode)->runtime_flags);
3121 btrfs_orphan_release_metadata(inode);
3122 }
3123 if (ret != -EEXIST) {
e8e7cff6
JB		 3124 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3125 &BTRFS_I(inode)->runtime_flags);
4ef31a45
JB		 3126 btrfs_abort_transaction(trans, root, ret);
3127 return ret;
3128 }
79787eaa
JM		 3129 }
3130 ret = 0;
d68fc57b
YZ		 3131 }
3132
3133 /* insert an orphan item to track subvolume contains orphan files */
3134 if (insert >= 2) {
3135 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3136 root->root_key.objectid);
79787eaa
JM		 3137 if (ret && ret != -EEXIST) {
3138 btrfs_abort_transaction(trans, root, ret);
3139 return ret;
3140 }
d68fc57b
YZ		 3141 }
3142 return 0;
7b128766
JB		 3143}
3144
3145/*
3146 * We have done the truncate/delete so we can go ahead and remove the orphan
3147 * item for this particular inode.
3148 */
48a3b636
ES		 3149static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3150 struct inode *inode)
7b128766
JB		 3151{
3152 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 3153 int delete_item = 0;
3154 int release_rsv = 0;
7b128766
JB		 3155 int ret = 0;
3156
d68fc57b 3157 spin_lock(&root->orphan_lock);
8a35d95f
JB		 3158 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3159 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3160 delete_item = 1;
7b128766 3161
72ac3c0d
JB		 3162 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3163 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3164 release_rsv = 1;
d68fc57b 3165 spin_unlock(&root->orphan_lock);
7b128766 3166
703c88e0 3167 if (delete_item) {
8a35d95f 3168 atomic_dec(&root->orphan_inodes);
703c88e0
FDBM		 3169 if (trans)
3170 ret = btrfs_del_orphan_item(trans, root,
3171 btrfs_ino(inode));
8a35d95f 3172 }
7b128766 3173
703c88e0
FDBM		 3174 if (release_rsv)
3175 btrfs_orphan_release_metadata(inode);
3176
4ef31a45 3177 return ret;
7b128766
JB		 3178}
3179
3180/*
3181 * this cleans up any orphans that may be left on the list from the last use
3182 * of this root.
3183 */
66b4ffd1 3184int btrfs_orphan_cleanup(struct btrfs_root *root)
7b128766
JB		 3185{
3186 struct btrfs_path *path;
3187 struct extent_buffer *leaf;
7b128766
JB		 3188 struct btrfs_key key, found_key;
3189 struct btrfs_trans_handle *trans;
3190 struct inode *inode;
8f6d7f4f 3191 u64 last_objectid = 0;
7b128766
JB		 3192 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3193
d68fc57b 3194 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
66b4ffd1 3195 return 0;
c71bf099
YZ		 3196
3197 path = btrfs_alloc_path();
66b4ffd1
JB		 3198 if (!path) {
3199 ret = -ENOMEM;
3200 goto out;
3201 }
7b128766
JB		 3202 path->reada = -1;
3203
3204 key.objectid = BTRFS_ORPHAN_OBJECTID;
962a298f 3205 key.type = BTRFS_ORPHAN_ITEM_KEY;
7b128766
JB		 3206 key.offset = (u64)-1;
3207
7b128766
JB		 3208 while (1) {
3209 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
66b4ffd1
JB		 3210 if (ret < 0)
3211 goto out;
7b128766
JB		 3212
3213 /*
3214 * if ret == 0 means we found what we were searching for, which
25985edc 3215 * is weird, but possible, so only screw with path if we didn't
7b128766
JB		 3216 * find the key and see if we have stuff that matches
3217 */
3218 if (ret > 0) {
66b4ffd1 3219 ret = 0;
7b128766
JB		 3220 if (path->slots[0] == 0)
3221 break;
3222 path->slots[0]--;
3223 }
3224
3225 /* pull out the item */
3226 leaf = path->nodes[0];
7b128766
JB		 3227 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3228
3229 /* make sure the item matches what we want */
3230 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3231 break;
962a298f 3232 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
7b128766
JB		 3233 break;
3234
3235 /* release the path since we're done with it */
b3b4aa74 3236 btrfs_release_path(path);
7b128766
JB		 3237
3238 /*
3239 * this is where we are basically btrfs_lookup, without the
3240 * crossing root thing. we store the inode number in the
3241 * offset of the orphan item.
3242 */
8f6d7f4f
JB		 3243
3244 if (found_key.offset == last_objectid) {
c2cf52eb
SK		 3245 btrfs_err(root->fs_info,
3246 "Error removing orphan entry, stopping orphan cleanup");
8f6d7f4f
JB		 3247 ret = -EINVAL;
3248 goto out;
3249 }
3250
3251 last_objectid = found_key.offset;
3252
5d4f98a2
YZ		 3253 found_key.objectid = found_key.offset;
3254 found_key.type = BTRFS_INODE_ITEM_KEY;
3255 found_key.offset = 0;
73f73415 3256 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
8c6ffba0 3257 ret = PTR_ERR_OR_ZERO(inode);
a8c9e576 3258 if (ret && ret != -ESTALE)
66b4ffd1 3259 goto out;
7b128766 3260
f8e9e0b0
AJ		 3261 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3262 struct btrfs_root *dead_root;
3263 struct btrfs_fs_info *fs_info = root->fs_info;
3264 int is_dead_root = 0;
3265
3266 /*
3267 * this is an orphan in the tree root. Currently these
3268 * could come from 2 sources:
3269 * a) a snapshot deletion in progress
3270 * b) a free space cache inode
3271 * We need to distinguish those two, as the snapshot
3272 * orphan must not get deleted.
3273 * find_dead_roots already ran before us, so if this
3274 * is a snapshot deletion, we should find the root
3275 * in the dead_roots list
3276 */
3277 spin_lock(&fs_info->trans_lock);
3278 list_for_each_entry(dead_root, &fs_info->dead_roots,
3279 root_list) {
3280 if (dead_root->root_key.objectid ==
3281 found_key.objectid) {
3282 is_dead_root = 1;
3283 break;
3284 }
3285 }
3286 spin_unlock(&fs_info->trans_lock);
3287 if (is_dead_root) {
3288 /* prevent this orphan from being found again */
3289 key.offset = found_key.objectid - 1;
3290 continue;
3291 }
3292 }
7b128766 3293 /*
a8c9e576
JB		 3294 * Inode is already gone but the orphan item is still there,
3295 * kill the orphan item.
7b128766 3296 */
a8c9e576
JB		 3297 if (ret == -ESTALE) {
3298 trans = btrfs_start_transaction(root, 1);
66b4ffd1
JB		 3299 if (IS_ERR(trans)) {
3300 ret = PTR_ERR(trans);
3301 goto out;
3302 }
c2cf52eb
SK		 3303 btrfs_debug(root->fs_info, "auto deleting %Lu",
3304 found_key.objectid);
a8c9e576
JB		 3305 ret = btrfs_del_orphan_item(trans, root,
3306 found_key.objectid);
5b21f2ed 3307 btrfs_end_transaction(trans, root);
4ef31a45
JB		 3308 if (ret)
3309 goto out;
7b128766
JB		 3310 continue;
3311 }
3312
a8c9e576
JB		 3313 /*
3314 * add this inode to the orphan list so btrfs_orphan_del does
3315 * the proper thing when we hit it
3316 */
8a35d95f
JB		 3317 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3318 &BTRFS_I(inode)->runtime_flags);
925396ec 3319 atomic_inc(&root->orphan_inodes);
a8c9e576 3320
7b128766
JB		 3321 /* if we have links, this was a truncate, lets do that */
3322 if (inode->i_nlink) {
fae7f21c 3323 if (WARN_ON(!S_ISREG(inode->i_mode))) {
a41ad394
JB		 3324 iput(inode);
3325 continue;
3326 }
7b128766 3327 nr_truncate++;
f3fe820c
JB		 3328
3329 /* 1 for the orphan item deletion. */
3330 trans = btrfs_start_transaction(root, 1);
3331 if (IS_ERR(trans)) {
c69b26b0 3332 iput(inode);
f3fe820c
JB		 3333 ret = PTR_ERR(trans);
3334 goto out;
3335 }
3336 ret = btrfs_orphan_add(trans, inode);
3337 btrfs_end_transaction(trans, root);
c69b26b0
JB		 3338 if (ret) {
3339 iput(inode);
f3fe820c 3340 goto out;
c69b26b0 3341 }
f3fe820c 3342
66b4ffd1 3343 ret = btrfs_truncate(inode);
4a7d0f68
JB		 3344 if (ret)
3345 btrfs_orphan_del(NULL, inode);
7b128766
JB		 3346 } else {
3347 nr_unlink++;
3348 }
3349
3350 /* this will do delete_inode and everything for us */
3351 iput(inode);
66b4ffd1
JB		 3352 if (ret)
3353 goto out;
7b128766 3354 }
3254c876
MX		 3355 /* release the path since we're done with it */
3356 btrfs_release_path(path);
3357
d68fc57b
YZ		 3358 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3359
3360 if (root->orphan_block_rsv)
3361 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3362 (u64)-1);
3363
27cdeb70
MX		 3364 if (root->orphan_block_rsv ||
3365 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
7a7eaa40 3366 trans = btrfs_join_transaction(root);
66b4ffd1
JB		 3367 if (!IS_ERR(trans))
3368 btrfs_end_transaction(trans, root);
d68fc57b 3369 }
7b128766
JB		 3370
3371 if (nr_unlink)
4884b476 3372 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
7b128766 3373 if (nr_truncate)
4884b476 3374 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
66b4ffd1
JB		 3375
3376out:
3377 if (ret)
c2cf52eb
SK		 3378 btrfs_crit(root->fs_info,
3379 "could not do orphan cleanup %d", ret);
66b4ffd1
JB		 3380 btrfs_free_path(path);
3381 return ret;
7b128766
JB		 3382}
3383
46a53cca
CM		 3384/*
3385 * very simple check to peek ahead in the leaf looking for xattrs. If we
3386 * don't find any xattrs, we know there can't be any acls.
3387 *
3388 * slot is the slot the inode is in, objectid is the objectid of the inode
3389 */
3390static noinline int acls_after_inode_item(struct extent_buffer *leaf,
63541927
FDBM		 3391 int slot, u64 objectid,
3392 int *first_xattr_slot)
46a53cca
CM		 3393{
3394 u32 nritems = btrfs_header_nritems(leaf);
3395 struct btrfs_key found_key;
f23b5a59
JB		 3396 static u64 xattr_access = 0;
3397 static u64 xattr_default = 0;
46a53cca
CM		 3398 int scanned = 0;
3399
f23b5a59
JB		 3400 if (!xattr_access) {
3401 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3402 strlen(POSIX_ACL_XATTR_ACCESS));
3403 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3404 strlen(POSIX_ACL_XATTR_DEFAULT));
3405 }
3406
46a53cca 3407 slot++;
63541927 3408 *first_xattr_slot = -1;
46a53cca
CM		 3409 while (slot < nritems) {
3410 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3411
3412 /* we found a different objectid, there must not be acls */
3413 if (found_key.objectid != objectid)
3414 return 0;
3415
3416 /* we found an xattr, assume we've got an acl */
f23b5a59 3417 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
63541927
FDBM		 3418 if (*first_xattr_slot == -1)
3419 *first_xattr_slot = slot;
f23b5a59
JB		 3420 if (found_key.offset == xattr_access ||
3421 found_key.offset == xattr_default)
3422 return 1;
3423 }
46a53cca
CM		 3424
3425 /*
3426 * we found a key greater than an xattr key, there can't
3427 * be any acls later on
3428 */
3429 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3430 return 0;
3431
3432 slot++;
3433 scanned++;
3434
3435 /*
3436 * it goes inode, inode backrefs, xattrs, extents,
3437 * so if there are a ton of hard links to an inode there can
3438 * be a lot of backrefs. Don't waste time searching too hard,
3439 * this is just an optimization
3440 */
3441 if (scanned >= 8)
3442 break;
3443 }
3444 /* we hit the end of the leaf before we found an xattr or
3445 * something larger than an xattr. We have to assume the inode
3446 * has acls
3447 */
63541927
FDBM		 3448 if (*first_xattr_slot == -1)
3449 *first_xattr_slot = slot;
46a53cca
CM		 3450 return 1;
3451}
3452
d352ac68
CM		 3453/*
3454 * read an inode from the btree into the in-memory inode
3455 */
5d4f98a2 3456static void btrfs_read_locked_inode(struct inode *inode)
39279cc3
CM		 3457{
3458 struct btrfs_path *path;
5f39d397 3459 struct extent_buffer *leaf;
39279cc3 3460 struct btrfs_inode_item *inode_item;
0b86a832 3461 struct btrfs_timespec *tspec;
39279cc3
CM		 3462 struct btrfs_root *root = BTRFS_I(inode)->root;
3463 struct btrfs_key location;
67de1176 3464 unsigned long ptr;
46a53cca 3465 int maybe_acls;
618e21d5 3466 u32 rdev;
39279cc3 3467 int ret;
2f7e33d4 3468 bool filled = false;
63541927 3469 int first_xattr_slot;
2f7e33d4
MX		 3470
3471 ret = btrfs_fill_inode(inode, &rdev);
3472 if (!ret)
3473 filled = true;
39279cc3
CM		 3474
3475 path = btrfs_alloc_path();
1748f843
MF		 3476 if (!path)
3477 goto make_bad;
3478
39279cc3 3479 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 3480
39279cc3 3481 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 3482 if (ret)
39279cc3 3483 goto make_bad;
39279cc3 3484
5f39d397 3485 leaf = path->nodes[0];
2f7e33d4
MX		 3486
3487 if (filled)
67de1176 3488 goto cache_index;
2f7e33d4 3489
5f39d397
CM		 3490 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3491 struct btrfs_inode_item);
5f39d397 3492 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
bfe86848 3493 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2f2f43d3
EB		 3494 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3495 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
dbe674a9 3496 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397
CM		 3497
3498 tspec = btrfs_inode_atime(inode_item);
3499 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3500 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3501
3502 tspec = btrfs_inode_mtime(inode_item);
3503 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3504 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3505
3506 tspec = btrfs_inode_ctime(inode_item);
3507 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3508 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3509
a76a3cd4 3510 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 3511 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
5dc562c5
JB		 3512 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3513
3514 /*
3515 * If we were modified in the current generation and evicted from memory
3516 * and then re-read we need to do a full sync since we don't have any
3517 * idea about which extents were modified before we were evicted from
3518 * cache.
3519 */
3520 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3521 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3522 &BTRFS_I(inode)->runtime_flags);
3523
0c4d2d95 3524 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
e02119d5 3525 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 3526 inode->i_rdev = 0;
5f39d397
CM		 3527 rdev = btrfs_inode_rdev(leaf, inode_item);
3528
aec7477b 3529 BTRFS_I(inode)->index_cnt = (u64)-1;
d2fb3437 3530 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
67de1176
MX		 3531
3532cache_index:
3533 path->slots[0]++;
3534 if (inode->i_nlink != 1 ||
3535 path->slots[0] >= btrfs_header_nritems(leaf))
3536 goto cache_acl;
3537
3538 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3539 if (location.objectid != btrfs_ino(inode))
3540 goto cache_acl;
3541
3542 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3543 if (location.type == BTRFS_INODE_REF_KEY) {
3544 struct btrfs_inode_ref *ref;
3545
3546 ref = (struct btrfs_inode_ref *)ptr;
3547 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3548 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3549 struct btrfs_inode_extref *extref;
3550
3551 extref = (struct btrfs_inode_extref *)ptr;
3552 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3553 extref);
3554 }
2f7e33d4 3555cache_acl:
46a53cca
CM		 3556 /*
3557 * try to precache a NULL acl entry for files that don't have
3558 * any xattrs or acls
3559 */
33345d01 3560 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
63541927
FDBM		 3561 btrfs_ino(inode), &first_xattr_slot);
3562 if (first_xattr_slot != -1) {
3563 path->slots[0] = first_xattr_slot;
3564 ret = btrfs_load_inode_props(inode, path);
3565 if (ret)
3566 btrfs_err(root->fs_info,
351fd353 3567 "error loading props for ino %llu (root %llu): %d",
63541927
FDBM		 3568 btrfs_ino(inode),
3569 root->root_key.objectid, ret);
3570 }
3571 btrfs_free_path(path);
3572
72c04902
AV		 3573 if (!maybe_acls)
3574 cache_no_acl(inode);
46a53cca 3575
39279cc3 3576 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 3577 case S_IFREG:
3578 inode->i_mapping->a_ops = &btrfs_aops;
04160088 3579 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 3580 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 3581 inode->i_fop = &btrfs_file_operations;
3582 inode->i_op = &btrfs_file_inode_operations;
3583 break;
3584 case S_IFDIR:
3585 inode->i_fop = &btrfs_dir_file_operations;
3586 if (root == root->fs_info->tree_root)
3587 inode->i_op = &btrfs_dir_ro_inode_operations;
3588 else
3589 inode->i_op = &btrfs_dir_inode_operations;
3590 break;
3591 case S_IFLNK:
3592 inode->i_op = &btrfs_symlink_inode_operations;
3593 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 3594 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3 3595 break;
618e21d5 3596 default:
0279b4cd 3597 inode->i_op = &btrfs_special_inode_operations;
618e21d5
JB		 3598 init_special_inode(inode, inode->i_mode, rdev);
3599 break;
39279cc3 3600 }
6cbff00f
CH		 3601
3602 btrfs_update_iflags(inode);
39279cc3
CM		 3603 return;
3604
3605make_bad:
39279cc3 3606 btrfs_free_path(path);
39279cc3
CM		 3607 make_bad_inode(inode);
3608}
3609
d352ac68
CM		 3610/*
3611 * given a leaf and an inode, copy the inode fields into the leaf
3612 */
e02119d5
CM		 3613static void fill_inode_item(struct btrfs_trans_handle *trans,
3614 struct extent_buffer *leaf,
5f39d397 3615 struct btrfs_inode_item *item,
39279cc3
CM		 3616 struct inode *inode)
3617{
51fab693
LB		 3618 struct btrfs_map_token token;
3619
3620 btrfs_init_map_token(&token);
5f39d397 3621
51fab693
LB		 3622 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3623 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3624 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3625 &token);
3626 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3627 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
5f39d397 3628
51fab693
LB		 3629 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3630 inode->i_atime.tv_sec, &token);
3631 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3632 inode->i_atime.tv_nsec, &token);
5f39d397 3633
51fab693
LB		 3634 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3635 inode->i_mtime.tv_sec, &token);
3636 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3637 inode->i_mtime.tv_nsec, &token);
5f39d397 3638
51fab693
LB		 3639 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3640 inode->i_ctime.tv_sec, &token);
3641 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3642 inode->i_ctime.tv_nsec, &token);
5f39d397 3643
51fab693
LB		 3644 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3645 &token);
3646 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3647 &token);
3648 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3649 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3650 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3651 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3652 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
39279cc3
CM		 3653}
3654
d352ac68
CM		 3655/*
3656 * copy everything in the in-memory inode into the btree.
3657 */
2115133f 3658static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
d397712b 3659 struct btrfs_root *root, struct inode *inode)
39279cc3
CM		 3660{
3661 struct btrfs_inode_item *inode_item;
3662 struct btrfs_path *path;
5f39d397 3663 struct extent_buffer *leaf;
39279cc3
CM		 3664 int ret;
3665
3666 path = btrfs_alloc_path();
16cdcec7
MX		 3667 if (!path)
3668 return -ENOMEM;
3669
b9473439 3670 path->leave_spinning = 1;
16cdcec7
MX		 3671 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3672 1);
39279cc3
CM		 3673 if (ret) {
3674 if (ret > 0)
3675 ret = -ENOENT;
3676 goto failed;
3677 }
3678
5f39d397
CM		 3679 leaf = path->nodes[0];
3680 inode_item = btrfs_item_ptr(leaf, path->slots[0],
16cdcec7 3681 struct btrfs_inode_item);
39279cc3 3682
e02119d5 3683 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 3684 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 3685 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 3686 ret = 0;
3687failed:
39279cc3
CM		 3688 btrfs_free_path(path);
3689 return ret;
3690}
3691
2115133f
CM		 3692/*
3693 * copy everything in the in-memory inode into the btree.
3694 */
3695noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3696 struct btrfs_root *root, struct inode *inode)
3697{
3698 int ret;
3699
3700 /*
3701 * If the inode is a free space inode, we can deadlock during commit
3702 * if we put it into the delayed code.
3703 *
3704 * The data relocation inode should also be directly updated
3705 * without delay
3706 */
83eea1f1 3707 if (!btrfs_is_free_space_inode(inode)
2115133f 3708 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
8ea05e3a
AB		 3709 btrfs_update_root_times(trans, root);
3710
2115133f
CM		 3711 ret = btrfs_delayed_update_inode(trans, root, inode);
3712 if (!ret)
3713 btrfs_set_inode_last_trans(trans, inode);
3714 return ret;
3715 }
3716
3717 return btrfs_update_inode_item(trans, root, inode);
3718}
3719
be6aef60
JB		 3720noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3721 struct btrfs_root *root,
3722 struct inode *inode)
2115133f
CM		 3723{
3724 int ret;
3725
3726 ret = btrfs_update_inode(trans, root, inode);
3727 if (ret == -ENOSPC)
3728 return btrfs_update_inode_item(trans, root, inode);
3729 return ret;
3730}
3731
d352ac68
CM		 3732/*
3733 * unlink helper that gets used here in inode.c and in the tree logging
3734 * recovery code. It remove a link in a directory with a given name, and
3735 * also drops the back refs in the inode to the directory
3736 */
92986796
AV		 3737static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3738 struct btrfs_root *root,
3739 struct inode *dir, struct inode *inode,
3740 const char *name, int name_len)
39279cc3
CM		 3741{
3742 struct btrfs_path *path;
39279cc3 3743 int ret = 0;
5f39d397 3744 struct extent_buffer *leaf;
39279cc3 3745 struct btrfs_dir_item *di;
5f39d397 3746 struct btrfs_key key;
aec7477b 3747 u64 index;
33345d01
LZ		 3748 u64 ino = btrfs_ino(inode);
3749 u64 dir_ino = btrfs_ino(dir);
39279cc3
CM		 3750
3751 path = btrfs_alloc_path();
54aa1f4d
CM		 3752 if (!path) {
3753 ret = -ENOMEM;
554233a6 3754 goto out;
54aa1f4d
CM		 3755 }
3756
b9473439 3757 path->leave_spinning = 1;
33345d01 3758 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
39279cc3
CM		 3759 name, name_len, -1);
3760 if (IS_ERR(di)) {
3761 ret = PTR_ERR(di);
3762 goto err;
3763 }
3764 if (!di) {
3765 ret = -ENOENT;
3766 goto err;
3767 }
5f39d397
CM		 3768 leaf = path->nodes[0];
3769 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 3770 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 3771 if (ret)
3772 goto err;
b3b4aa74 3773 btrfs_release_path(path);
39279cc3 3774
67de1176
MX		 3775 /*
3776 * If we don't have dir index, we have to get it by looking up
3777 * the inode ref, since we get the inode ref, remove it directly,
3778 * it is unnecessary to do delayed deletion.
3779 *
3780 * But if we have dir index, needn't search inode ref to get it.
3781 * Since the inode ref is close to the inode item, it is better
3782 * that we delay to delete it, and just do this deletion when
3783 * we update the inode item.
3784 */
3785 if (BTRFS_I(inode)->dir_index) {
3786 ret = btrfs_delayed_delete_inode_ref(inode);
3787 if (!ret) {
3788 index = BTRFS_I(inode)->dir_index;
3789 goto skip_backref;
3790 }
3791 }
3792
33345d01
LZ		 3793 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3794 dir_ino, &index);
aec7477b 3795 if (ret) {
c2cf52eb
SK		 3796 btrfs_info(root->fs_info,
3797 "failed to delete reference to %.*s, inode %llu parent %llu",
c1c9ff7c 3798 name_len, name, ino, dir_ino);
79787eaa 3799 btrfs_abort_transaction(trans, root, ret);
aec7477b
JB		 3800 goto err;
3801 }
67de1176 3802skip_backref:
16cdcec7 3803 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 3804 if (ret) {
3805 btrfs_abort_transaction(trans, root, ret);
39279cc3 3806 goto err;
79787eaa 3807 }
39279cc3 3808
e02119d5 3809 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
33345d01 3810 inode, dir_ino);
79787eaa
JM		 3811 if (ret != 0 && ret != -ENOENT) {
3812 btrfs_abort_transaction(trans, root, ret);
3813 goto err;
3814 }
e02119d5
CM		 3815
3816 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3817 dir, index);
6418c961
CM		 3818 if (ret == -ENOENT)
3819 ret = 0;
d4e3991b
ZB		 3820 else if (ret)
3821 btrfs_abort_transaction(trans, root, ret);
39279cc3
CM		 3822err:
3823 btrfs_free_path(path);
e02119d5
CM		 3824 if (ret)
3825 goto out;
3826
3827 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95
JB		 3828 inode_inc_iversion(inode);
3829 inode_inc_iversion(dir);
e02119d5 3830 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
b9959295 3831 ret = btrfs_update_inode(trans, root, dir);
e02119d5 3832out:
39279cc3
CM		 3833 return ret;
3834}
3835
92986796
AV		 3836int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3837 struct btrfs_root *root,
3838 struct inode *dir, struct inode *inode,
3839 const char *name, int name_len)
3840{
3841 int ret;
3842 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3843 if (!ret) {
8b558c5f 3844 drop_nlink(inode);
92986796
AV		 3845 ret = btrfs_update_inode(trans, root, inode);
3846 }
3847 return ret;
3848}
39279cc3 3849
a22285a6
YZ		 3850/*
3851 * helper to start transaction for unlink and rmdir.
3852 *
d52be818
JB		 3853 * unlink and rmdir are special in btrfs, they do not always free space, so
3854 * if we cannot make our reservations the normal way try and see if there is
3855 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3856 * allow the unlink to occur.
a22285a6 3857 */
d52be818 3858static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4df27c4d 3859{
39279cc3 3860 struct btrfs_trans_handle *trans;
a22285a6 3861 struct btrfs_root *root = BTRFS_I(dir)->root;
4df27c4d
YZ		 3862 int ret;
3863
e70bea5f
JB		 3864 /*
3865 * 1 for the possible orphan item
3866 * 1 for the dir item
3867 * 1 for the dir index
3868 * 1 for the inode ref
e70bea5f
JB		 3869 * 1 for the inode
3870 */
6e137ed3 3871 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 3872 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3873 return trans;
4df27c4d 3874
d52be818
JB		 3875 if (PTR_ERR(trans) == -ENOSPC) {
3876 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4df27c4d 3877
d52be818
JB		 3878 trans = btrfs_start_transaction(root, 0);
3879 if (IS_ERR(trans))
3880 return trans;
3881 ret = btrfs_cond_migrate_bytes(root->fs_info,
3882 &root->fs_info->trans_block_rsv,
3883 num_bytes, 5);
3884 if (ret) {
3885 btrfs_end_transaction(trans, root);
3886 return ERR_PTR(ret);
a22285a6 3887 }
5a77d76c 3888 trans->block_rsv = &root->fs_info->trans_block_rsv;
d52be818 3889 trans->bytes_reserved = num_bytes;
a22285a6 3890 }
d52be818 3891 return trans;
a22285a6
YZ		 3892}
3893
3894static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3895{
3896 struct btrfs_root *root = BTRFS_I(dir)->root;
3897 struct btrfs_trans_handle *trans;
3898 struct inode *inode = dentry->d_inode;
3899 int ret;
a22285a6 3900
d52be818 3901 trans = __unlink_start_trans(dir);
a22285a6
YZ		 3902 if (IS_ERR(trans))
3903 return PTR_ERR(trans);
5f39d397 3904
12fcfd22
CM		 3905 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3906
e02119d5
CM		 3907 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3908 dentry->d_name.name, dentry->d_name.len);
b532402e
TI		 3909 if (ret)
3910 goto out;
7b128766 3911
a22285a6 3912 if (inode->i_nlink == 0) {
7b128766 3913 ret = btrfs_orphan_add(trans, inode);
b532402e
TI		 3914 if (ret)
3915 goto out;
a22285a6 3916 }
7b128766 3917
b532402e 3918out:
d52be818 3919 btrfs_end_transaction(trans, root);
b53d3f5d 3920 btrfs_btree_balance_dirty(root);
39279cc3
CM		 3921 return ret;
3922}
3923
4df27c4d
YZ		 3924int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3925 struct btrfs_root *root,
3926 struct inode *dir, u64 objectid,
3927 const char *name, int name_len)
3928{
3929 struct btrfs_path *path;
3930 struct extent_buffer *leaf;
3931 struct btrfs_dir_item *di;
3932 struct btrfs_key key;
3933 u64 index;
3934 int ret;
33345d01 3935 u64 dir_ino = btrfs_ino(dir);
4df27c4d
YZ		 3936
3937 path = btrfs_alloc_path();
3938 if (!path)
3939 return -ENOMEM;
3940
33345d01 3941 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4df27c4d 3942 name, name_len, -1);
79787eaa
JM		 3943 if (IS_ERR_OR_NULL(di)) {
3944 if (!di)
3945 ret = -ENOENT;
3946 else
3947 ret = PTR_ERR(di);
3948 goto out;
3949 }
4df27c4d
YZ		 3950
3951 leaf = path->nodes[0];
3952 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3953 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3954 ret = btrfs_delete_one_dir_name(trans, root, path, di);
79787eaa
JM		 3955 if (ret) {
3956 btrfs_abort_transaction(trans, root, ret);
3957 goto out;
3958 }
b3b4aa74 3959 btrfs_release_path(path);
4df27c4d
YZ		 3960
3961 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3962 objectid, root->root_key.objectid,
33345d01 3963 dir_ino, &index, name, name_len);
4df27c4d 3964 if (ret < 0) {
79787eaa
JM		 3965 if (ret != -ENOENT) {
3966 btrfs_abort_transaction(trans, root, ret);
3967 goto out;
3968 }
33345d01 3969 di = btrfs_search_dir_index_item(root, path, dir_ino,
4df27c4d 3970 name, name_len);
79787eaa
JM		 3971 if (IS_ERR_OR_NULL(di)) {
3972 if (!di)
3973 ret = -ENOENT;
3974 else
3975 ret = PTR_ERR(di);
3976 btrfs_abort_transaction(trans, root, ret);
3977 goto out;
3978 }
4df27c4d
YZ		 3979
3980 leaf = path->nodes[0];
3981 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
b3b4aa74 3982 btrfs_release_path(path);
4df27c4d
YZ		 3983 index = key.offset;
3984 }
945d8962 3985 btrfs_release_path(path);
4df27c4d 3986
16cdcec7 3987 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 3988 if (ret) {
3989 btrfs_abort_transaction(trans, root, ret);
3990 goto out;
3991 }
4df27c4d
YZ		 3992
3993 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95 3994 inode_inc_iversion(dir);
4df27c4d 3995 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
5a24e84c 3996 ret = btrfs_update_inode_fallback(trans, root, dir);
79787eaa
JM		 3997 if (ret)
3998 btrfs_abort_transaction(trans, root, ret);
3999out:
71d7aed0 4000 btrfs_free_path(path);
79787eaa 4001 return ret;
4df27c4d
YZ		 4002}
4003
39279cc3
CM		 4004static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4005{
4006 struct inode *inode = dentry->d_inode;
1832a6d5 4007 int err = 0;
39279cc3 4008 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 4009 struct btrfs_trans_handle *trans;
39279cc3 4010
b3ae244e 4011 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
134d4512 4012 return -ENOTEMPTY;
b3ae244e
DS		 4013 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4014 return -EPERM;
134d4512 4015
d52be818 4016 trans = __unlink_start_trans(dir);
a22285a6 4017 if (IS_ERR(trans))
5df6a9f6 4018 return PTR_ERR(trans);
5df6a9f6 4019
33345d01 4020 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4df27c4d
YZ		 4021 err = btrfs_unlink_subvol(trans, root, dir,
4022 BTRFS_I(inode)->location.objectid,
4023 dentry->d_name.name,
4024 dentry->d_name.len);
4025 goto out;
4026 }
4027
7b128766
JB		 4028 err = btrfs_orphan_add(trans, inode);
4029 if (err)
4df27c4d 4030 goto out;
7b128766 4031
39279cc3 4032 /* now the directory is empty */
e02119d5
CM		 4033 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
4034 dentry->d_name.name, dentry->d_name.len);
d397712b 4035 if (!err)
dbe674a9 4036 btrfs_i_size_write(inode, 0);
4df27c4d 4037out:
d52be818 4038 btrfs_end_transaction(trans, root);
b53d3f5d 4039 btrfs_btree_balance_dirty(root);
3954401f 4040
39279cc3
CM		 4041 return err;
4042}
4043
39279cc3
CM		 4044/*
4045 * this can truncate away extent items, csum items and directory items.
4046 * It starts at a high offset and removes keys until it can't find
d352ac68 4047 * any higher than new_size
39279cc3
CM		 4048 *
4049 * csum items that cross the new i_size are truncated to the new size
4050 * as well.
7b128766
JB		 4051 *
4052 * min_type is the minimum key type to truncate down to. If set to 0, this
4053 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 4054 */
8082510e
YZ		 4055int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4056 struct btrfs_root *root,
4057 struct inode *inode,
4058 u64 new_size, u32 min_type)
39279cc3 4059{
39279cc3 4060 struct btrfs_path *path;
5f39d397 4061 struct extent_buffer *leaf;
39279cc3 4062 struct btrfs_file_extent_item *fi;
8082510e
YZ		 4063 struct btrfs_key key;
4064 struct btrfs_key found_key;
39279cc3 4065 u64 extent_start = 0;
db94535d 4066 u64 extent_num_bytes = 0;
5d4f98a2 4067 u64 extent_offset = 0;
39279cc3 4068 u64 item_end = 0;
7f4f6e0a 4069 u64 last_size = (u64)-1;
8082510e 4070 u32 found_type = (u8)-1;
39279cc3
CM		 4071 int found_extent;
4072 int del_item;
85e21bac
CM		 4073 int pending_del_nr = 0;
4074 int pending_del_slot = 0;
179e29e4 4075 int extent_type = -1;
8082510e
YZ		 4076 int ret;
4077 int err = 0;
33345d01 4078 u64 ino = btrfs_ino(inode);
8082510e
YZ		 4079
4080 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
39279cc3 4081
0eb0e19c
MF		 4082 path = btrfs_alloc_path();
4083 if (!path)
4084 return -ENOMEM;
4085 path->reada = -1;
4086
5dc562c5
JB		 4087 /*
4088 * We want to drop from the next block forward in case this new size is
4089 * not block aligned since we will be keeping the last block of the
4090 * extent just the way it is.
4091 */
27cdeb70
MX		 4092 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4093 root == root->fs_info->tree_root)
fda2832f
QW		 4094 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4095 root->sectorsize), (u64)-1, 0);
8082510e 4096
16cdcec7
MX		 4097 /*
4098 * This function is also used to drop the items in the log tree before
4099 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4100 * it is used to drop the loged items. So we shouldn't kill the delayed
4101 * items.
4102 */
4103 if (min_type == 0 && root == BTRFS_I(inode)->root)
4104 btrfs_kill_delayed_inode_items(inode);
4105
33345d01 4106 key.objectid = ino;
39279cc3 4107 key.offset = (u64)-1;
5f39d397
CM		 4108 key.type = (u8)-1;
4109
85e21bac 4110search_again:
b9473439 4111 path->leave_spinning = 1;
85e21bac 4112 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8082510e
YZ		 4113 if (ret < 0) {
4114 err = ret;
4115 goto out;
4116 }
d397712b 4117
85e21bac 4118 if (ret > 0) {
e02119d5
CM		 4119 /* there are no items in the tree for us to truncate, we're
4120 * done
4121 */
8082510e
YZ		 4122 if (path->slots[0] == 0)
4123 goto out;
85e21bac
CM		 4124 path->slots[0]--;
4125 }
4126
d397712b 4127 while (1) {
39279cc3 4128 fi = NULL;
5f39d397
CM		 4129 leaf = path->nodes[0];
4130 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
962a298f 4131 found_type = found_key.type;
39279cc3 4132
33345d01 4133 if (found_key.objectid != ino)
39279cc3 4134 break;
5f39d397 4135
85e21bac 4136 if (found_type < min_type)
39279cc3
CM		 4137 break;
4138
5f39d397 4139 item_end = found_key.offset;
39279cc3 4140 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 4141 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 4142 struct btrfs_file_extent_item);
179e29e4
CM		 4143 extent_type = btrfs_file_extent_type(leaf, fi);
4144 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 4145 item_end +=
db94535d 4146 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 4147 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 4148 item_end += btrfs_file_extent_inline_len(leaf,
514ac8ad 4149 path->slots[0], fi);
39279cc3 4150 }
008630c1 4151 item_end--;
39279cc3 4152 }
8082510e
YZ		 4153 if (found_type > min_type) {
4154 del_item = 1;
4155 } else {
4156 if (item_end < new_size)
b888db2b 4157 break;
8082510e
YZ		 4158 if (found_key.offset >= new_size)
4159 del_item = 1;
4160 else
4161 del_item = 0;
39279cc3 4162 }
39279cc3 4163 found_extent = 0;
39279cc3 4164 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 4165 if (found_type != BTRFS_EXTENT_DATA_KEY)
4166 goto delete;
4167
7f4f6e0a
JB		 4168 if (del_item)
4169 last_size = found_key.offset;
4170 else
4171 last_size = new_size;
4172
179e29e4 4173 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 4174 u64 num_dec;
db94535d 4175 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
f70a9a6b 4176 if (!del_item) {
db94535d
CM		 4177 u64 orig_num_bytes =
4178 btrfs_file_extent_num_bytes(leaf, fi);
fda2832f
QW		 4179 extent_num_bytes = ALIGN(new_size -
4180 found_key.offset,
4181 root->sectorsize);
db94535d
CM		 4182 btrfs_set_file_extent_num_bytes(leaf, fi,
4183 extent_num_bytes);
4184 num_dec = (orig_num_bytes -
9069218d 4185 extent_num_bytes);
27cdeb70
MX		 4186 if (test_bit(BTRFS_ROOT_REF_COWS,
4187 &root->state) &&
4188 extent_start != 0)
a76a3cd4 4189 inode_sub_bytes(inode, num_dec);
5f39d397 4190 btrfs_mark_buffer_dirty(leaf);
39279cc3 4191 } else {
db94535d
CM		 4192 extent_num_bytes =
4193 btrfs_file_extent_disk_num_bytes(leaf,
4194 fi);
5d4f98a2
YZ		 4195 extent_offset = found_key.offset -
4196 btrfs_file_extent_offset(leaf, fi);
4197
39279cc3 4198 /* FIXME blocksize != 4096 */
9069218d 4199 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 4200 if (extent_start != 0) {
4201 found_extent = 1;
27cdeb70
MX		 4202 if (test_bit(BTRFS_ROOT_REF_COWS,
4203 &root->state))
a76a3cd4 4204 inode_sub_bytes(inode, num_dec);
e02119d5 4205 }
39279cc3 4206 }
9069218d 4207 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 4208 /*
4209 * we can't truncate inline items that have had
4210 * special encodings
4211 */
4212 if (!del_item &&
4213 btrfs_file_extent_compression(leaf, fi) == 0 &&
4214 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4215 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM		 4216 u32 size = new_size - found_key.offset;
4217
27cdeb70 4218 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
a76a3cd4
YZ		 4219 inode_sub_bytes(inode, item_end + 1 -
4220 new_size);
514ac8ad
CM		 4221
4222 /*
4223 * update the ram bytes to properly reflect
4224 * the new size of our item
4225 */
4226 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
e02119d5
CM		 4227 size =
4228 btrfs_file_extent_calc_inline_size(size);
afe5fea7 4229 btrfs_truncate_item(root, path, size, 1);
27cdeb70
MX		 4230 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4231 &root->state)) {
a76a3cd4
YZ		 4232 inode_sub_bytes(inode, item_end + 1 -
4233 found_key.offset);
9069218d 4234 }
39279cc3 4235 }
179e29e4 4236delete:
39279cc3 4237 if (del_item) {
85e21bac
CM		 4238 if (!pending_del_nr) {
4239 /* no pending yet, add ourselves */
4240 pending_del_slot = path->slots[0];
4241 pending_del_nr = 1;
4242 } else if (pending_del_nr &&
4243 path->slots[0] + 1 == pending_del_slot) {
4244 /* hop on the pending chunk */
4245 pending_del_nr++;
4246 pending_del_slot = path->slots[0];
4247 } else {
d397712b 4248 BUG();
85e21bac 4249 }
39279cc3
CM		 4250 } else {
4251 break;
4252 }
27cdeb70
MX		 4253 if (found_extent &&
4254 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4255 root == root->fs_info->tree_root)) {
b9473439 4256 btrfs_set_path_blocking(path);
39279cc3 4257 ret = btrfs_free_extent(trans, root, extent_start,
5d4f98a2
YZ		 4258 extent_num_bytes, 0,
4259 btrfs_header_owner(leaf),
66d7e7f0 4260 ino, extent_offset, 0);
39279cc3
CM		 4261 BUG_ON(ret);
4262 }
85e21bac 4263
8082510e
YZ		 4264 if (found_type == BTRFS_INODE_ITEM_KEY)
4265 break;
4266
4267 if (path->slots[0] == 0 ||
4268 path->slots[0] != pending_del_slot) {
8082510e
YZ		 4269 if (pending_del_nr) {
4270 ret = btrfs_del_items(trans, root, path,
4271 pending_del_slot,
4272 pending_del_nr);
79787eaa
JM		 4273 if (ret) {
4274 btrfs_abort_transaction(trans,
4275 root, ret);
4276 goto error;
4277 }
8082510e
YZ		 4278 pending_del_nr = 0;
4279 }
b3b4aa74 4280 btrfs_release_path(path);
85e21bac 4281 goto search_again;
8082510e
YZ		 4282 } else {
4283 path->slots[0]--;
85e21bac 4284 }
39279cc3 4285 }
8082510e 4286out:
85e21bac
CM		 4287 if (pending_del_nr) {
4288 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4289 pending_del_nr);
79787eaa
JM		 4290 if (ret)
4291 btrfs_abort_transaction(trans, root, ret);
85e21bac 4292 }
79787eaa 4293error:
dac5705c
FM		 4294 if (last_size != (u64)-1 &&
4295 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
7f4f6e0a 4296 btrfs_ordered_update_i_size(inode, last_size, NULL);
39279cc3 4297 btrfs_free_path(path);
8082510e 4298 return err;
39279cc3
CM		 4299}
4300
4301/*
2aaa6655
JB		 4302 * btrfs_truncate_page - read, zero a chunk and write a page
4303 * @inode - inode that we're zeroing
4304 * @from - the offset to start zeroing
4305 * @len - the length to zero, 0 to zero the entire range respective to the
4306 * offset
4307 * @front - zero up to the offset instead of from the offset on
4308 *
4309 * This will find the page for the "from" offset and cow the page and zero the
4310 * part we want to zero. This is used with truncate and hole punching.
39279cc3 4311 */
2aaa6655
JB		 4312int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4313 int front)
39279cc3 4314{
2aaa6655 4315 struct address_space *mapping = inode->i_mapping;
db94535d 4316 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 4317 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4318 struct btrfs_ordered_extent *ordered;
2ac55d41 4319 struct extent_state *cached_state = NULL;
e6dcd2dc 4320 char *kaddr;
db94535d 4321 u32 blocksize = root->sectorsize;
39279cc3
CM		 4322 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4323 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4324 struct page *page;
3b16a4e3 4325 gfp_t mask = btrfs_alloc_write_mask(mapping);
39279cc3 4326 int ret = 0;
a52d9a80 4327 u64 page_start;
e6dcd2dc 4328 u64 page_end;
39279cc3 4329
2aaa6655
JB		 4330 if ((offset & (blocksize - 1)) == 0 &&
4331 (!len || ((len & (blocksize - 1)) == 0)))
39279cc3 4332 goto out;
0ca1f7ce 4333 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
5d5e103a
JB		 4334 if (ret)
4335 goto out;
39279cc3 4336
211c17f5 4337again:
3b16a4e3 4338 page = find_or_create_page(mapping, index, mask);
5d5e103a 4339 if (!page) {
0ca1f7ce 4340 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
ac6a2b36 4341 ret = -ENOMEM;
39279cc3 4342 goto out;
5d5e103a 4343 }
e6dcd2dc
CM		 4344
4345 page_start = page_offset(page);
4346 page_end = page_start + PAGE_CACHE_SIZE - 1;
4347
39279cc3 4348 if (!PageUptodate(page)) {
9ebefb18 4349 ret = btrfs_readpage(NULL, page);
39279cc3 4350 lock_page(page);
211c17f5
CM		 4351 if (page->mapping != mapping) {
4352 unlock_page(page);
4353 page_cache_release(page);
4354 goto again;
4355 }
39279cc3
CM		 4356 if (!PageUptodate(page)) {
4357 ret = -EIO;
89642229 4358 goto out_unlock;
39279cc3
CM		 4359 }
4360 }
211c17f5 4361 wait_on_page_writeback(page);
e6dcd2dc 4362
d0082371 4363 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 4364 set_page_extent_mapped(page);
4365
4366 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4367 if (ordered) {
2ac55d41
JB		 4368 unlock_extent_cached(io_tree, page_start, page_end,
4369 &cached_state, GFP_NOFS);
e6dcd2dc
CM		 4370 unlock_page(page);
4371 page_cache_release(page);
eb84ae03 4372 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 4373 btrfs_put_ordered_extent(ordered);
4374 goto again;
4375 }
4376
2ac55d41 4377 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 4378 EXTENT_DIRTY | EXTENT_DELALLOC |
4379 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 4380 0, 0, &cached_state, GFP_NOFS);
5d5e103a 4381
2ac55d41
JB		 4382 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4383 &cached_state);
9ed74f2d 4384 if (ret) {
2ac55d41
JB		 4385 unlock_extent_cached(io_tree, page_start, page_end,
4386 &cached_state, GFP_NOFS);
9ed74f2d
JB		 4387 goto out_unlock;
4388 }
4389
e6dcd2dc 4390 if (offset != PAGE_CACHE_SIZE) {
2aaa6655
JB		 4391 if (!len)
4392 len = PAGE_CACHE_SIZE - offset;
e6dcd2dc 4393 kaddr = kmap(page);
2aaa6655
JB		 4394 if (front)
4395 memset(kaddr, 0, offset);
4396 else
4397 memset(kaddr + offset, 0, len);
e6dcd2dc
CM		 4398 flush_dcache_page(page);
4399 kunmap(page);
4400 }
247e743c 4401 ClearPageChecked(page);
e6dcd2dc 4402 set_page_dirty(page);
2ac55d41
JB		 4403 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4404 GFP_NOFS);
39279cc3 4405
89642229 4406out_unlock:
5d5e103a 4407 if (ret)
0ca1f7ce 4408 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
39279cc3
CM		 4409 unlock_page(page);
4410 page_cache_release(page);
4411out:
4412 return ret;
4413}
4414
16e7549f
JB		 4415static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4416 u64 offset, u64 len)
4417{
4418 struct btrfs_trans_handle *trans;
4419 int ret;
4420
4421 /*
4422 * Still need to make sure the inode looks like it's been updated so
4423 * that any holes get logged if we fsync.
4424 */
4425 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4426 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4427 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4428 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4429 return 0;
4430 }
4431
4432 /*
4433 * 1 - for the one we're dropping
4434 * 1 - for the one we're adding
4435 * 1 - for updating the inode.
4436 */
4437 trans = btrfs_start_transaction(root, 3);
4438 if (IS_ERR(trans))
4439 return PTR_ERR(trans);
4440
4441 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4442 if (ret) {
4443 btrfs_abort_transaction(trans, root, ret);
4444 btrfs_end_transaction(trans, root);
4445 return ret;
4446 }
4447
4448 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4449 0, 0, len, 0, len, 0, 0, 0);
4450 if (ret)
4451 btrfs_abort_transaction(trans, root, ret);
4452 else
4453 btrfs_update_inode(trans, root, inode);
4454 btrfs_end_transaction(trans, root);
4455 return ret;
4456}
4457
695a0d0d
JB		 4458/*
4459 * This function puts in dummy file extents for the area we're creating a hole
4460 * for. So if we are truncating this file to a larger size we need to insert
4461 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4462 * the range between oldsize and size
4463 */
a41ad394 4464int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
39279cc3 4465{
9036c102
YZ		 4466 struct btrfs_root *root = BTRFS_I(inode)->root;
4467 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a22285a6 4468 struct extent_map *em = NULL;
2ac55d41 4469 struct extent_state *cached_state = NULL;
5dc562c5 4470 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
fda2832f
QW		 4471 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4472 u64 block_end = ALIGN(size, root->sectorsize);
9036c102
YZ		 4473 u64 last_byte;
4474 u64 cur_offset;
4475 u64 hole_size;
9ed74f2d 4476 int err = 0;
39279cc3 4477
a71754fc
JB		 4478 /*
4479 * If our size started in the middle of a page we need to zero out the
4480 * rest of the page before we expand the i_size, otherwise we could
4481 * expose stale data.
4482 */
4483 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4484 if (err)
4485 return err;
4486
9036c102
YZ		 4487 if (size <= hole_start)
4488 return 0;
4489
9036c102
YZ		 4490 while (1) {
4491 struct btrfs_ordered_extent *ordered;
fa7c1494 4492
2ac55d41 4493 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
d0082371 4494 &cached_state);
fa7c1494
MX		 4495 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4496 block_end - hole_start);
9036c102
YZ		 4497 if (!ordered)
4498 break;
2ac55d41
JB		 4499 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4500 &cached_state, GFP_NOFS);
fa7c1494 4501 btrfs_start_ordered_extent(inode, ordered, 1);
9036c102
YZ		 4502 btrfs_put_ordered_extent(ordered);
4503 }
39279cc3 4504
9036c102
YZ		 4505 cur_offset = hole_start;
4506 while (1) {
4507 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4508 block_end - cur_offset, 0);
79787eaa
JM		 4509 if (IS_ERR(em)) {
4510 err = PTR_ERR(em);
f2767956 4511 em = NULL;
79787eaa
JM		 4512 break;
4513 }
9036c102 4514 last_byte = min(extent_map_end(em), block_end);
fda2832f 4515 last_byte = ALIGN(last_byte , root->sectorsize);
8082510e 4516 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5dc562c5 4517 struct extent_map *hole_em;
9036c102 4518 hole_size = last_byte - cur_offset;
9ed74f2d 4519
16e7549f
JB		 4520 err = maybe_insert_hole(root, inode, cur_offset,
4521 hole_size);
4522 if (err)
3893e33b 4523 break;
5dc562c5
JB		 4524 btrfs_drop_extent_cache(inode, cur_offset,
4525 cur_offset + hole_size - 1, 0);
4526 hole_em = alloc_extent_map();
4527 if (!hole_em) {
4528 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4529 &BTRFS_I(inode)->runtime_flags);
4530 goto next;
4531 }
4532 hole_em->start = cur_offset;
4533 hole_em->len = hole_size;
4534 hole_em->orig_start = cur_offset;
8082510e 4535
5dc562c5
JB		 4536 hole_em->block_start = EXTENT_MAP_HOLE;
4537 hole_em->block_len = 0;
b4939680 4538 hole_em->orig_block_len = 0;
cc95bef6 4539 hole_em->ram_bytes = hole_size;
5dc562c5
JB		 4540 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4541 hole_em->compress_type = BTRFS_COMPRESS_NONE;
16e7549f 4542 hole_em->generation = root->fs_info->generation;
8082510e 4543
5dc562c5
JB		 4544 while (1) {
4545 write_lock(&em_tree->lock);
09a2a8f9 4546 err = add_extent_mapping(em_tree, hole_em, 1);
5dc562c5
JB		 4547 write_unlock(&em_tree->lock);
4548 if (err != -EEXIST)
4549 break;
4550 btrfs_drop_extent_cache(inode, cur_offset,
4551 cur_offset +
4552 hole_size - 1, 0);
4553 }
4554 free_extent_map(hole_em);
9036c102 4555 }
16e7549f 4556next:
9036c102 4557 free_extent_map(em);
a22285a6 4558 em = NULL;
9036c102 4559 cur_offset = last_byte;
8082510e 4560 if (cur_offset >= block_end)
9036c102
YZ		 4561 break;
4562 }
a22285a6 4563 free_extent_map(em);
2ac55d41
JB		 4564 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4565 GFP_NOFS);
9036c102
YZ		 4566 return err;
4567}
39279cc3 4568
3972f260 4569static int btrfs_setsize(struct inode *inode, struct iattr *attr)
8082510e 4570{
f4a2f4c5
MX		 4571 struct btrfs_root *root = BTRFS_I(inode)->root;
4572 struct btrfs_trans_handle *trans;
a41ad394 4573 loff_t oldsize = i_size_read(inode);
3972f260
ES		 4574 loff_t newsize = attr->ia_size;
4575 int mask = attr->ia_valid;
8082510e
YZ		 4576 int ret;
4577
3972f260
ES		 4578 /*
4579 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4580 * special case where we need to update the times despite not having
4581 * these flags set. For all other operations the VFS set these flags
4582 * explicitly if it wants a timestamp update.
4583 */
dff6efc3
CH		 4584 if (newsize != oldsize) {
4585 inode_inc_iversion(inode);
4586 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4587 inode->i_ctime = inode->i_mtime =
4588 current_fs_time(inode->i_sb);
4589 }
3972f260 4590
a41ad394 4591 if (newsize > oldsize) {
7caef267 4592 truncate_pagecache(inode, newsize);
a41ad394 4593 ret = btrfs_cont_expand(inode, oldsize, newsize);
f4a2f4c5 4594 if (ret)
8082510e 4595 return ret;
8082510e 4596
f4a2f4c5
MX		 4597 trans = btrfs_start_transaction(root, 1);
4598 if (IS_ERR(trans))
4599 return PTR_ERR(trans);
4600
4601 i_size_write(inode, newsize);
4602 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4603 ret = btrfs_update_inode(trans, root, inode);
7ad85bb7 4604 btrfs_end_transaction(trans, root);
a41ad394 4605 } else {
8082510e 4606
a41ad394
JB		 4607 /*
4608 * We're truncating a file that used to have good data down to
4609 * zero. Make sure it gets into the ordered flush list so that
4610 * any new writes get down to disk quickly.
4611 */
4612 if (newsize == 0)
72ac3c0d
JB		 4613 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4614 &BTRFS_I(inode)->runtime_flags);
8082510e 4615
f3fe820c
JB		 4616 /*
4617 * 1 for the orphan item we're going to add
4618 * 1 for the orphan item deletion.
4619 */
4620 trans = btrfs_start_transaction(root, 2);
4621 if (IS_ERR(trans))
4622 return PTR_ERR(trans);
4623
4624 /*
4625 * We need to do this in case we fail at _any_ point during the
4626 * actual truncate. Once we do the truncate_setsize we could
4627 * invalidate pages which forces any outstanding ordered io to
4628 * be instantly completed which will give us extents that need
4629 * to be truncated. If we fail to get an orphan inode down we
4630 * could have left over extents that were never meant to live,
4631 * so we need to garuntee from this point on that everything
4632 * will be consistent.
4633 */
4634 ret = btrfs_orphan_add(trans, inode);
4635 btrfs_end_transaction(trans, root);
4636 if (ret)
4637 return ret;
4638
a41ad394
JB		 4639 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4640 truncate_setsize(inode, newsize);
2e60a51e
MX		 4641
4642 /* Disable nonlocked read DIO to avoid the end less truncate */
4643 btrfs_inode_block_unlocked_dio(inode);
4644 inode_dio_wait(inode);
4645 btrfs_inode_resume_unlocked_dio(inode);
4646
a41ad394 4647 ret = btrfs_truncate(inode);
7f4f6e0a
JB		 4648 if (ret && inode->i_nlink) {
4649 int err;
4650
4651 /*
4652 * failed to truncate, disk_i_size is only adjusted down
4653 * as we remove extents, so it should represent the true
4654 * size of the inode, so reset the in memory size and
4655 * delete our orphan entry.
4656 */
4657 trans = btrfs_join_transaction(root);
4658 if (IS_ERR(trans)) {
4659 btrfs_orphan_del(NULL, inode);
4660 return ret;
4661 }
4662 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4663 err = btrfs_orphan_del(trans, inode);
4664 if (err)
4665 btrfs_abort_transaction(trans, root, err);
4666 btrfs_end_transaction(trans, root);
4667 }
8082510e
YZ		 4668 }
4669
a41ad394 4670 return ret;
8082510e
YZ		 4671}
4672
9036c102
YZ		 4673static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4674{
4675 struct inode *inode = dentry->d_inode;
b83cc969 4676 struct btrfs_root *root = BTRFS_I(inode)->root;
9036c102 4677 int err;
39279cc3 4678
b83cc969
LZ		 4679 if (btrfs_root_readonly(root))
4680 return -EROFS;
4681
9036c102
YZ		 4682 err = inode_change_ok(inode, attr);
4683 if (err)
4684 return err;
2bf5a725 4685
5a3f23d5 4686 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3972f260 4687 err = btrfs_setsize(inode, attr);
8082510e
YZ		 4688 if (err)
4689 return err;
39279cc3 4690 }
9036c102 4691
1025774c
CH		 4692 if (attr->ia_valid) {
4693 setattr_copy(inode, attr);
0c4d2d95 4694 inode_inc_iversion(inode);
22c44fe6 4695 err = btrfs_dirty_inode(inode);
1025774c 4696
22c44fe6 4697 if (!err && attr->ia_valid & ATTR_MODE)
996a710d 4698 err = posix_acl_chmod(inode, inode->i_mode);
1025774c 4699 }
33268eaf 4700
39279cc3
CM		 4701 return err;
4702}
61295eb8 4703
131e404a
FDBM		 4704/*
4705 * While truncating the inode pages during eviction, we get the VFS calling
4706 * btrfs_invalidatepage() against each page of the inode. This is slow because
4707 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4708 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4709 * extent_state structures over and over, wasting lots of time.
4710 *
4711 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4712 * those expensive operations on a per page basis and do only the ordered io
4713 * finishing, while we release here the extent_map and extent_state structures,
4714 * without the excessive merging and splitting.
4715 */
4716static void evict_inode_truncate_pages(struct inode *inode)
4717{
4718 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4719 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4720 struct rb_node *node;
4721
4722 ASSERT(inode->i_state & I_FREEING);
91b0abe3 4723 truncate_inode_pages_final(&inode->i_data);
131e404a
FDBM		 4724
4725 write_lock(&map_tree->lock);
4726 while (!RB_EMPTY_ROOT(&map_tree->map)) {
4727 struct extent_map *em;
4728
4729 node = rb_first(&map_tree->map);
4730 em = rb_entry(node, struct extent_map, rb_node);
180589ef
WS		 4731 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4732 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
131e404a
FDBM		 4733 remove_extent_mapping(map_tree, em);
4734 free_extent_map(em);
7064dd5c
FM		 4735 if (need_resched()) {
4736 write_unlock(&map_tree->lock);
4737 cond_resched();
4738 write_lock(&map_tree->lock);
4739 }
131e404a
FDBM		 4740 }
4741 write_unlock(&map_tree->lock);
4742
4743 spin_lock(&io_tree->lock);
4744 while (!RB_EMPTY_ROOT(&io_tree->state)) {
4745 struct extent_state *state;
4746 struct extent_state *cached_state = NULL;
4747
4748 node = rb_first(&io_tree->state);
4749 state = rb_entry(node, struct extent_state, rb_node);
4750 atomic_inc(&state->refs);
4751 spin_unlock(&io_tree->lock);
4752
4753 lock_extent_bits(io_tree, state->start, state->end,
4754 0, &cached_state);
4755 clear_extent_bit(io_tree, state->start, state->end,
4756 EXTENT_LOCKED | EXTENT_DIRTY |
4757 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
4758 EXTENT_DEFRAG, 1, 1,
4759 &cached_state, GFP_NOFS);
4760 free_extent_state(state);
4761
7064dd5c 4762 cond_resched();
131e404a
FDBM		 4763 spin_lock(&io_tree->lock);
4764 }
4765 spin_unlock(&io_tree->lock);
4766}
4767
bd555975 4768void btrfs_evict_inode(struct inode *inode)
39279cc3
CM		 4769{
4770 struct btrfs_trans_handle *trans;
4771 struct btrfs_root *root = BTRFS_I(inode)->root;
726c35fa 4772 struct btrfs_block_rsv *rsv, *global_rsv;
07127184 4773 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3
CM		 4774 int ret;
4775
1abe9b8a 4776 trace_btrfs_inode_evict(inode);
4777
131e404a
FDBM		 4778 evict_inode_truncate_pages(inode);
4779
69e9c6c6
SB		 4780 if (inode->i_nlink &&
4781 ((btrfs_root_refs(&root->root_item) != 0 &&
4782 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4783 btrfs_is_free_space_inode(inode)))
bd555975
AV		 4784 goto no_delete;
4785
39279cc3 4786 if (is_bad_inode(inode)) {
7b128766 4787 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 4788 goto no_delete;
4789 }
bd555975 4790 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4a096752 4791 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 4792
c71bf099 4793 if (root->fs_info->log_root_recovering) {
6bf02314 4794 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8a35d95f 4795 &BTRFS_I(inode)->runtime_flags));
c71bf099
YZ		 4796 goto no_delete;
4797 }
4798
76dda93c 4799 if (inode->i_nlink > 0) {
69e9c6c6
SB		 4800 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4801 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
76dda93c
YZ		 4802 goto no_delete;
4803 }
4804
0e8c36a9
MX		 4805 ret = btrfs_commit_inode_delayed_inode(inode);
4806 if (ret) {
4807 btrfs_orphan_del(NULL, inode);
4808 goto no_delete;
4809 }
4810
66d8f3dd 4811 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4289a667
JB		 4812 if (!rsv) {
4813 btrfs_orphan_del(NULL, inode);
4814 goto no_delete;
4815 }
4a338542 4816 rsv->size = min_size;
ca7e70f5 4817 rsv->failfast = 1;
726c35fa 4818 global_rsv = &root->fs_info->global_block_rsv;
4289a667 4819
dbe674a9 4820 btrfs_i_size_write(inode, 0);
5f39d397 4821
4289a667 4822 /*
8407aa46
MX		 4823 * This is a bit simpler than btrfs_truncate since we've already
4824 * reserved our space for our orphan item in the unlink, so we just
4825 * need to reserve some slack space in case we add bytes and update
4826 * inode item when doing the truncate.
4289a667 4827 */
8082510e 4828 while (1) {
08e007d2
MX		 4829 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4830 BTRFS_RESERVE_FLUSH_LIMIT);
726c35fa
JB		 4831
4832 /*
4833 * Try and steal from the global reserve since we will
4834 * likely not use this space anyway, we want to try as
4835 * hard as possible to get this to work.
4836 */
4837 if (ret)
4838 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
d68fc57b 4839
d68fc57b 4840 if (ret) {
c2cf52eb
SK		 4841 btrfs_warn(root->fs_info,
4842 "Could not get space for a delete, will truncate on mount %d",
4843 ret);
4289a667
JB		 4844 btrfs_orphan_del(NULL, inode);
4845 btrfs_free_block_rsv(root, rsv);
4846 goto no_delete;
d68fc57b 4847 }
7b128766 4848
0e8c36a9 4849 trans = btrfs_join_transaction(root);
4289a667
JB		 4850 if (IS_ERR(trans)) {
4851 btrfs_orphan_del(NULL, inode);
4852 btrfs_free_block_rsv(root, rsv);
4853 goto no_delete;
d68fc57b 4854 }
7b128766 4855
4289a667
JB		 4856 trans->block_rsv = rsv;
4857
d68fc57b 4858 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
ca7e70f5 4859 if (ret != -ENOSPC)
8082510e 4860 break;
85e21bac 4861
8407aa46 4862 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e
YZ		 4863 btrfs_end_transaction(trans, root);
4864 trans = NULL;
b53d3f5d 4865 btrfs_btree_balance_dirty(root);
8082510e 4866 }
5f39d397 4867
4289a667
JB		 4868 btrfs_free_block_rsv(root, rsv);
4869
4ef31a45
JB		 4870 /*
4871 * Errors here aren't a big deal, it just means we leave orphan items
4872 * in the tree. They will be cleaned up on the next mount.
4873 */
8082510e 4874 if (ret == 0) {
4289a667 4875 trans->block_rsv = root->orphan_block_rsv;
4ef31a45
JB		 4876 btrfs_orphan_del(trans, inode);
4877 } else {
4878 btrfs_orphan_del(NULL, inode);
8082510e 4879 }
54aa1f4d 4880
4289a667 4881 trans->block_rsv = &root->fs_info->trans_block_rsv;
581bb050
LZ		 4882 if (!(root == root->fs_info->tree_root ||
4883 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
33345d01 4884 btrfs_return_ino(root, btrfs_ino(inode));
581bb050 4885
54aa1f4d 4886 btrfs_end_transaction(trans, root);
b53d3f5d 4887 btrfs_btree_balance_dirty(root);
39279cc3 4888no_delete:
89042e5a 4889 btrfs_remove_delayed_node(inode);
dbd5768f 4890 clear_inode(inode);
8082510e 4891 return;
39279cc3
CM		 4892}
4893
4894/*
4895 * this returns the key found in the dir entry in the location pointer.
4896 * If no dir entries were found, location->objectid is 0.
4897 */
4898static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4899 struct btrfs_key *location)
4900{
4901 const char *name = dentry->d_name.name;
4902 int namelen = dentry->d_name.len;
4903 struct btrfs_dir_item *di;
4904 struct btrfs_path *path;
4905 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 4906 int ret = 0;
39279cc3
CM		 4907
4908 path = btrfs_alloc_path();
d8926bb3
MF		 4909 if (!path)
4910 return -ENOMEM;
3954401f 4911
33345d01 4912 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
39279cc3 4913 namelen, 0);
0d9f7f3e
Y		 4914 if (IS_ERR(di))
4915 ret = PTR_ERR(di);
d397712b 4916
c704005d 4917 if (IS_ERR_OR_NULL(di))
3954401f 4918 goto out_err;
d397712b 4919
5f39d397 4920 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 4921out:
39279cc3
CM		 4922 btrfs_free_path(path);
4923 return ret;
3954401f
CM		 4924out_err:
4925 location->objectid = 0;
4926 goto out;
39279cc3
CM		 4927}
4928
4929/*
4930 * when we hit a tree root in a directory, the btrfs part of the inode
4931 * needs to be changed to reflect the root directory of the tree root. This
4932 * is kind of like crossing a mount point.
4933 */
4934static int fixup_tree_root_location(struct btrfs_root *root,
4df27c4d
YZ		 4935 struct inode *dir,
4936 struct dentry *dentry,
4937 struct btrfs_key *location,
4938 struct btrfs_root **sub_root)
39279cc3 4939{
4df27c4d
YZ		 4940 struct btrfs_path *path;
4941 struct btrfs_root *new_root;
4942 struct btrfs_root_ref *ref;
4943 struct extent_buffer *leaf;
4944 int ret;
4945 int err = 0;
39279cc3 4946
4df27c4d
YZ		 4947 path = btrfs_alloc_path();
4948 if (!path) {
4949 err = -ENOMEM;
4950 goto out;
4951 }
39279cc3 4952
4df27c4d 4953 err = -ENOENT;
75ac2dd9
KN		 4954 ret = btrfs_find_item(root->fs_info->tree_root, path,
4955 BTRFS_I(dir)->root->root_key.objectid,
4956 location->objectid, BTRFS_ROOT_REF_KEY, NULL);
4df27c4d
YZ		 4957 if (ret) {
4958 if (ret < 0)
4959 err = ret;
4960 goto out;
4961 }
39279cc3 4962
4df27c4d
YZ		 4963 leaf = path->nodes[0];
4964 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
33345d01 4965 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4df27c4d
YZ		 4966 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4967 goto out;
39279cc3 4968
4df27c4d
YZ		 4969 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4970 (unsigned long)(ref + 1),
4971 dentry->d_name.len);
4972 if (ret)
4973 goto out;
4974
b3b4aa74 4975 btrfs_release_path(path);
4df27c4d
YZ		 4976
4977 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4978 if (IS_ERR(new_root)) {
4979 err = PTR_ERR(new_root);
4980 goto out;
4981 }
4982
4df27c4d
YZ		 4983 *sub_root = new_root;
4984 location->objectid = btrfs_root_dirid(&new_root->root_item);
4985 location->type = BTRFS_INODE_ITEM_KEY;
4986 location->offset = 0;
4987 err = 0;
4988out:
4989 btrfs_free_path(path);
4990 return err;
39279cc3
CM		 4991}
4992
5d4f98a2
YZ		 4993static void inode_tree_add(struct inode *inode)
4994{
4995 struct btrfs_root *root = BTRFS_I(inode)->root;
4996 struct btrfs_inode *entry;
03e860bd
FNP		 4997 struct rb_node **p;
4998 struct rb_node *parent;
cef21937 4999 struct rb_node *new = &BTRFS_I(inode)->rb_node;
33345d01 5000 u64 ino = btrfs_ino(inode);
5d4f98a2 5001
1d3382cb 5002 if (inode_unhashed(inode))
76dda93c 5003 return;
e1409cef 5004 parent = NULL;
5d4f98a2 5005 spin_lock(&root->inode_lock);
e1409cef 5006 p = &root->inode_tree.rb_node;
5d4f98a2
YZ		 5007 while (*p) {
5008 parent = *p;
5009 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5010
33345d01 5011 if (ino < btrfs_ino(&entry->vfs_inode))
03e860bd 5012 p = &parent->rb_left;
33345d01 5013 else if (ino > btrfs_ino(&entry->vfs_inode))
03e860bd 5014 p = &parent->rb_right;
5d4f98a2
YZ		 5015 else {
5016 WARN_ON(!(entry->vfs_inode.i_state &
a4ffdde6 5017 (I_WILL_FREE | I_FREEING)));
cef21937 5018 rb_replace_node(parent, new, &root->inode_tree);
03e860bd
FNP		 5019 RB_CLEAR_NODE(parent);
5020 spin_unlock(&root->inode_lock);
cef21937 5021 return;
5d4f98a2
YZ		 5022 }
5023 }
cef21937
FDBM		 5024 rb_link_node(new, parent, p);
5025 rb_insert_color(new, &root->inode_tree);
5d4f98a2
YZ		 5026 spin_unlock(&root->inode_lock);
5027}
5028
5029static void inode_tree_del(struct inode *inode)
5030{
5031 struct btrfs_root *root = BTRFS_I(inode)->root;
76dda93c 5032 int empty = 0;
5d4f98a2 5033
03e860bd 5034 spin_lock(&root->inode_lock);
5d4f98a2 5035 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5d4f98a2 5036 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5d4f98a2 5037 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
76dda93c 5038 empty = RB_EMPTY_ROOT(&root->inode_tree);
5d4f98a2 5039 }
03e860bd 5040 spin_unlock(&root->inode_lock);
76dda93c 5041
69e9c6c6 5042 if (empty && btrfs_root_refs(&root->root_item) == 0) {
76dda93c
YZ		 5043 synchronize_srcu(&root->fs_info->subvol_srcu);
5044 spin_lock(&root->inode_lock);
5045 empty = RB_EMPTY_ROOT(&root->inode_tree);
5046 spin_unlock(&root->inode_lock);
5047 if (empty)
5048 btrfs_add_dead_root(root);
5049 }
5050}
5051
143bede5 5052void btrfs_invalidate_inodes(struct btrfs_root *root)
76dda93c
YZ		 5053{
5054 struct rb_node *node;
5055 struct rb_node *prev;
5056 struct btrfs_inode *entry;
5057 struct inode *inode;
5058 u64 objectid = 0;
5059
7813b3db
LB		 5060 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5061 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
76dda93c
YZ		 5062
5063 spin_lock(&root->inode_lock);
5064again:
5065 node = root->inode_tree.rb_node;
5066 prev = NULL;
5067 while (node) {
5068 prev = node;
5069 entry = rb_entry(node, struct btrfs_inode, rb_node);
5070
33345d01 5071 if (objectid < btrfs_ino(&entry->vfs_inode))
76dda93c 5072 node = node->rb_left;
33345d01 5073 else if (objectid > btrfs_ino(&entry->vfs_inode))
76dda93c
YZ		 5074 node = node->rb_right;
5075 else
5076 break;
5077 }
5078 if (!node) {
5079 while (prev) {
5080 entry = rb_entry(prev, struct btrfs_inode, rb_node);
33345d01 5081 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
76dda93c
YZ		 5082 node = prev;
5083 break;
5084 }
5085 prev = rb_next(prev);
5086 }
5087 }
5088 while (node) {
5089 entry = rb_entry(node, struct btrfs_inode, rb_node);
33345d01 5090 objectid = btrfs_ino(&entry->vfs_inode) + 1;
76dda93c
YZ		 5091 inode = igrab(&entry->vfs_inode);
5092 if (inode) {
5093 spin_unlock(&root->inode_lock);
5094 if (atomic_read(&inode->i_count) > 1)
5095 d_prune_aliases(inode);
5096 /*
45321ac5 5097 * btrfs_drop_inode will have it removed from
76dda93c
YZ		 5098 * the inode cache when its usage count
5099 * hits zero.
5100 */
5101 iput(inode);
5102 cond_resched();
5103 spin_lock(&root->inode_lock);
5104 goto again;
5105 }
5106
5107 if (cond_resched_lock(&root->inode_lock))
5108 goto again;
5109
5110 node = rb_next(node);
5111 }
5112 spin_unlock(&root->inode_lock);
5d4f98a2
YZ		 5113}
5114
e02119d5
CM		 5115static int btrfs_init_locked_inode(struct inode *inode, void *p)
5116{
5117 struct btrfs_iget_args *args = p;
90d3e592
CM		 5118 inode->i_ino = args->location->objectid;
5119 memcpy(&BTRFS_I(inode)->location, args->location,
5120 sizeof(*args->location));
e02119d5 5121 BTRFS_I(inode)->root = args->root;
39279cc3
CM		 5122 return 0;
5123}
5124
5125static int btrfs_find_actor(struct inode *inode, void *opaque)
5126{
5127 struct btrfs_iget_args *args = opaque;
90d3e592 5128 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
d397712b 5129 args->root == BTRFS_I(inode)->root;
39279cc3
CM		 5130}
5131
5d4f98a2 5132static struct inode *btrfs_iget_locked(struct super_block *s,
90d3e592 5133 struct btrfs_key *location,
5d4f98a2 5134 struct btrfs_root *root)
39279cc3
CM		 5135{
5136 struct inode *inode;
5137 struct btrfs_iget_args args;
90d3e592 5138 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
778ba82b 5139
90d3e592 5140 args.location = location;
39279cc3
CM		 5141 args.root = root;
5142
778ba82b 5143 inode = iget5_locked(s, hashval, btrfs_find_actor,
39279cc3
CM		 5144 btrfs_init_locked_inode,
5145 (void *)&args);
5146 return inode;
5147}
5148
1a54ef8c
BR		 5149/* Get an inode object given its location and corresponding root.
5150 * Returns in *is_new if the inode was read from disk
5151 */
5152struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
73f73415 5153 struct btrfs_root *root, int *new)
1a54ef8c
BR		 5154{
5155 struct inode *inode;
5156
90d3e592 5157 inode = btrfs_iget_locked(s, location, root);
1a54ef8c 5158 if (!inode)
5d4f98a2 5159 return ERR_PTR(-ENOMEM);
1a54ef8c
BR		 5160
5161 if (inode->i_state & I_NEW) {
1a54ef8c 5162 btrfs_read_locked_inode(inode);
1748f843
MF		 5163 if (!is_bad_inode(inode)) {
5164 inode_tree_add(inode);
5165 unlock_new_inode(inode);
5166 if (new)
5167 *new = 1;
5168 } else {
e0b6d65b
ST		 5169 unlock_new_inode(inode);
5170 iput(inode);
5171 inode = ERR_PTR(-ESTALE);
1748f843
MF		 5172 }
5173 }
5174
1a54ef8c
BR		 5175 return inode;
5176}
5177
4df27c4d
YZ		 5178static struct inode *new_simple_dir(struct super_block *s,
5179 struct btrfs_key *key,
5180 struct btrfs_root *root)
5181{
5182 struct inode *inode = new_inode(s);
5183
5184 if (!inode)
5185 return ERR_PTR(-ENOMEM);
5186
4df27c4d
YZ		 5187 BTRFS_I(inode)->root = root;
5188 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
72ac3c0d 5189 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4df27c4d
YZ		 5190
5191 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
848cce0d 5192 inode->i_op = &btrfs_dir_ro_inode_operations;
4df27c4d
YZ		 5193 inode->i_fop = &simple_dir_operations;
5194 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5195 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5196
5197 return inode;
5198}
5199
3de4586c 5200struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3 5201{
d397712b 5202 struct inode *inode;
4df27c4d 5203 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3
CM		 5204 struct btrfs_root *sub_root = root;
5205 struct btrfs_key location;
76dda93c 5206 int index;
b4aff1f8 5207 int ret = 0;
39279cc3
CM		 5208
5209 if (dentry->d_name.len > BTRFS_NAME_LEN)
5210 return ERR_PTR(-ENAMETOOLONG);
5f39d397 5211
39e3c955 5212 ret = btrfs_inode_by_name(dir, dentry, &location);
39279cc3
CM		 5213 if (ret < 0)
5214 return ERR_PTR(ret);
5f39d397 5215
4df27c4d 5216 if (location.objectid == 0)
5662344b 5217 return ERR_PTR(-ENOENT);
4df27c4d
YZ		 5218
5219 if (location.type == BTRFS_INODE_ITEM_KEY) {
73f73415 5220 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4df27c4d
YZ		 5221 return inode;
5222 }
5223
5224 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5225
76dda93c 5226 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4df27c4d
YZ		 5227 ret = fixup_tree_root_location(root, dir, dentry,
5228 &location, &sub_root);
5229 if (ret < 0) {
5230 if (ret != -ENOENT)
5231 inode = ERR_PTR(ret);
5232 else
5233 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5234 } else {
73f73415 5235 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
39279cc3 5236 }
76dda93c
YZ		 5237 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5238
34d19bad 5239 if (!IS_ERR(inode) && root != sub_root) {
c71bf099
YZ		 5240 down_read(&root->fs_info->cleanup_work_sem);
5241 if (!(inode->i_sb->s_flags & MS_RDONLY))
66b4ffd1 5242 ret = btrfs_orphan_cleanup(sub_root);
c71bf099 5243 up_read(&root->fs_info->cleanup_work_sem);
01cd3367
JB		 5244 if (ret) {
5245 iput(inode);
66b4ffd1 5246 inode = ERR_PTR(ret);
01cd3367 5247 }
9c3b306e
FM		 5248 /*
5249 * If orphan cleanup did remove any orphans, it means the tree
5250 * was modified and therefore the commit root is not the same as
5251 * the current root anymore. This is a problem, because send
5252 * uses the commit root and therefore can see inode items that
5253 * don't exist in the current root anymore, and for example make
5254 * calls to btrfs_iget, which will do tree lookups based on the
5255 * current root and not on the commit root. Those lookups will
5256 * fail, returning a -ESTALE error, and making send fail with
5257 * that error. So make sure a send does not see any orphans we
5258 * have just removed, and that it will see the same inodes
5259 * regardless of whether a transaction commit happened before
5260 * it started (meaning that the commit root will be the same as
5261 * the current root) or not.
5262 */
5263 if (sub_root->node != sub_root->commit_root) {
5264 u64 sub_flags = btrfs_root_flags(&sub_root->root_item);
5265
5266 if (sub_flags & BTRFS_ROOT_SUBVOL_RDONLY) {
5267 struct extent_buffer *eb;
5268
5269 /*
5270 * Assert we can't have races between dentry
5271 * lookup called through the snapshot creation
5272 * ioctl and the VFS.
5273 */
5274 ASSERT(mutex_is_locked(&dir->i_mutex));
5275
5276 down_write(&root->fs_info->commit_root_sem);
5277 eb = sub_root->commit_root;
5278 sub_root->commit_root =
5279 btrfs_root_node(sub_root);
5280 up_write(&root->fs_info->commit_root_sem);
5281 free_extent_buffer(eb);
5282 }
5283 }
c71bf099
YZ		 5284 }
5285
3de4586c
CM		 5286 return inode;
5287}
5288
fe15ce44 5289static int btrfs_dentry_delete(const struct dentry *dentry)
76dda93c
YZ		 5290{
5291 struct btrfs_root *root;
848cce0d 5292 struct inode *inode = dentry->d_inode;
76dda93c 5293
848cce0d
LZ		 5294 if (!inode && !IS_ROOT(dentry))
5295 inode = dentry->d_parent->d_inode;
76dda93c 5296
848cce0d
LZ		 5297 if (inode) {
5298 root = BTRFS_I(inode)->root;
efefb143
YZ		 5299 if (btrfs_root_refs(&root->root_item) == 0)
5300 return 1;
848cce0d
LZ		 5301
5302 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5303 return 1;
efefb143 5304 }
76dda93c
YZ		 5305 return 0;
5306}
5307
b4aff1f8
JB		 5308static void btrfs_dentry_release(struct dentry *dentry)
5309{
944a4515 5310 kfree(dentry->d_fsdata);
b4aff1f8
JB		 5311}
5312
3de4586c 5313static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
00cd8dd3 5314 unsigned int flags)
3de4586c 5315{
5662344b 5316 struct inode *inode;
a66e7cc6 5317
5662344b
TI		 5318 inode = btrfs_lookup_dentry(dir, dentry);
5319 if (IS_ERR(inode)) {
5320 if (PTR_ERR(inode) == -ENOENT)
5321 inode = NULL;
5322 else
5323 return ERR_CAST(inode);
5324 }
5325
3a0dfa6a 5326 return d_materialise_unique(dentry, inode);
39279cc3
CM		 5327}
5328
16cdcec7 5329unsigned char btrfs_filetype_table[] = {
39279cc3
CM		 5330 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5331};
5332
9cdda8d3 5333static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
39279cc3 5334{
9cdda8d3 5335 struct inode *inode = file_inode(file);
39279cc3
CM		 5336 struct btrfs_root *root = BTRFS_I(inode)->root;
5337 struct btrfs_item *item;
5338 struct btrfs_dir_item *di;
5339 struct btrfs_key key;
5f39d397 5340 struct btrfs_key found_key;
39279cc3 5341 struct btrfs_path *path;
16cdcec7
MX		 5342 struct list_head ins_list;
5343 struct list_head del_list;
39279cc3 5344 int ret;
5f39d397 5345 struct extent_buffer *leaf;
39279cc3 5346 int slot;
39279cc3
CM		 5347 unsigned char d_type;
5348 int over = 0;
5349 u32 di_cur;
5350 u32 di_total;
5351 u32 di_len;
5352 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 5353 char tmp_name[32];
5354 char *name_ptr;
5355 int name_len;
9cdda8d3 5356 int is_curr = 0; /* ctx->pos points to the current index? */
39279cc3
CM		 5357
5358 /* FIXME, use a real flag for deciding about the key type */
5359 if (root->fs_info->tree_root == root)
5360 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 5361
9cdda8d3
AV		 5362 if (!dir_emit_dots(file, ctx))
5363 return 0;
5364
49593bfa 5365 path = btrfs_alloc_path();
16cdcec7
MX		 5366 if (!path)
5367 return -ENOMEM;
ff5714cc 5368
026fd317 5369 path->reada = 1;
49593bfa 5370
16cdcec7
MX		 5371 if (key_type == BTRFS_DIR_INDEX_KEY) {
5372 INIT_LIST_HEAD(&ins_list);
5373 INIT_LIST_HEAD(&del_list);
5374 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5375 }
5376
962a298f 5377 key.type = key_type;
9cdda8d3 5378 key.offset = ctx->pos;
33345d01 5379 key.objectid = btrfs_ino(inode);
5f39d397 5380
39279cc3
CM		 5381 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5382 if (ret < 0)
5383 goto err;
49593bfa
DW		 5384
5385 while (1) {
5f39d397 5386 leaf = path->nodes[0];
39279cc3 5387 slot = path->slots[0];
b9e03af0
LZ		 5388 if (slot >= btrfs_header_nritems(leaf)) {
5389 ret = btrfs_next_leaf(root, path);
5390 if (ret < 0)
5391 goto err;
5392 else if (ret > 0)
5393 break;
5394 continue;
39279cc3 5395 }
3de4586c 5396
dd3cc16b 5397 item = btrfs_item_nr(slot);
5f39d397
CM		 5398 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5399
5400 if (found_key.objectid != key.objectid)
39279cc3 5401 break;
962a298f 5402 if (found_key.type != key_type)
39279cc3 5403 break;
9cdda8d3 5404 if (found_key.offset < ctx->pos)
b9e03af0 5405 goto next;
16cdcec7
MX		 5406 if (key_type == BTRFS_DIR_INDEX_KEY &&
5407 btrfs_should_delete_dir_index(&del_list,
5408 found_key.offset))
5409 goto next;
5f39d397 5410
9cdda8d3 5411 ctx->pos = found_key.offset;
16cdcec7 5412 is_curr = 1;
49593bfa 5413
39279cc3
CM		 5414 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5415 di_cur = 0;
5f39d397 5416 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 5417
5418 while (di_cur < di_total) {
5f39d397
CM		 5419 struct btrfs_key location;
5420
22a94d44
JB		 5421 if (verify_dir_item(root, leaf, di))
5422 break;
5423
5f39d397 5424 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 5425 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 5426 name_ptr = tmp_name;
5427 } else {
5428 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 5429 if (!name_ptr) {
5430 ret = -ENOMEM;
5431 goto err;
5432 }
5f39d397
CM		 5433 }
5434 read_extent_buffer(leaf, name_ptr,
5435 (unsigned long)(di + 1), name_len);
5436
5437 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5438 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c 5439
fede766f 5440
3de4586c 5441 /* is this a reference to our own snapshot? If so
8c9c2bf7
AJ		 5442 * skip it.
5443 *
5444 * In contrast to old kernels, we insert the snapshot's
5445 * dir item and dir index after it has been created, so
5446 * we won't find a reference to our own snapshot. We
5447 * still keep the following code for backward
5448 * compatibility.
3de4586c
CM		 5449 */
5450 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5451 location.objectid == root->root_key.objectid) {
5452 over = 0;
5453 goto skip;
5454 }
9cdda8d3
AV		 5455 over = !dir_emit(ctx, name_ptr, name_len,
5456 location.objectid, d_type);
5f39d397 5457
3de4586c 5458skip:
5f39d397
CM		 5459 if (name_ptr != tmp_name)
5460 kfree(name_ptr);
5461
39279cc3
CM		 5462 if (over)
5463 goto nopos;
5103e947 5464 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 5465 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 5466 di_cur += di_len;
5467 di = (struct btrfs_dir_item *)((char *)di + di_len);
5468 }
b9e03af0
LZ		 5469next:
5470 path->slots[0]++;
39279cc3 5471 }
49593bfa 5472
16cdcec7
MX		 5473 if (key_type == BTRFS_DIR_INDEX_KEY) {
5474 if (is_curr)
9cdda8d3
AV		 5475 ctx->pos++;
5476 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
16cdcec7
MX		 5477 if (ret)
5478 goto nopos;
5479 }
5480
49593bfa 5481 /* Reached end of directory/root. Bump pos past the last item. */
db62efbb
ZB		 5482 ctx->pos++;
5483
5484 /*
5485 * Stop new entries from being returned after we return the last
5486 * entry.
5487 *
5488 * New directory entries are assigned a strictly increasing
5489 * offset. This means that new entries created during readdir
5490 * are *guaranteed* to be seen in the future by that readdir.
5491 * This has broken buggy programs which operate on names as
5492 * they're returned by readdir. Until we re-use freed offsets
5493 * we have this hack to stop new entries from being returned
5494 * under the assumption that they'll never reach this huge
5495 * offset.
5496 *
5497 * This is being careful not to overflow 32bit loff_t unless the
5498 * last entry requires it because doing so has broken 32bit apps
5499 * in the past.
5500 */
5501 if (key_type == BTRFS_DIR_INDEX_KEY) {
5502 if (ctx->pos >= INT_MAX)
5503 ctx->pos = LLONG_MAX;
5504 else
5505 ctx->pos = INT_MAX;
5506 }
39279cc3
CM		 5507nopos:
5508 ret = 0;
5509err:
16cdcec7
MX		 5510 if (key_type == BTRFS_DIR_INDEX_KEY)
5511 btrfs_put_delayed_items(&ins_list, &del_list);
39279cc3 5512 btrfs_free_path(path);
39279cc3
CM		 5513 return ret;
5514}
5515
a9185b41 5516int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
39279cc3
CM		 5517{
5518 struct btrfs_root *root = BTRFS_I(inode)->root;
5519 struct btrfs_trans_handle *trans;
5520 int ret = 0;
0af3d00b 5521 bool nolock = false;
39279cc3 5522
72ac3c0d 5523 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4ca8b41e
CM		 5524 return 0;
5525
83eea1f1 5526 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
82d5902d 5527 nolock = true;
0af3d00b 5528
a9185b41 5529 if (wbc->sync_mode == WB_SYNC_ALL) {
0af3d00b 5530 if (nolock)
7a7eaa40 5531 trans = btrfs_join_transaction_nolock(root);
0af3d00b 5532 else
7a7eaa40 5533 trans = btrfs_join_transaction(root);
3612b495
TI		 5534 if (IS_ERR(trans))
5535 return PTR_ERR(trans);
a698d075 5536 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 5537 }
5538 return ret;
5539}
5540
5541/*
54aa1f4d 5542 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 5543 * inode changes. But, it is most likely to find the inode in cache.
5544 * FIXME, needs more benchmarking...there are no reasons other than performance
5545 * to keep or drop this code.
5546 */
48a3b636 5547static int btrfs_dirty_inode(struct inode *inode)
39279cc3
CM		 5548{
5549 struct btrfs_root *root = BTRFS_I(inode)->root;
5550 struct btrfs_trans_handle *trans;
8929ecfa
YZ		 5551 int ret;
5552
72ac3c0d 5553 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
22c44fe6 5554 return 0;
39279cc3 5555
7a7eaa40 5556 trans = btrfs_join_transaction(root);
22c44fe6
JB		 5557 if (IS_ERR(trans))
5558 return PTR_ERR(trans);
8929ecfa
YZ		 5559
5560 ret = btrfs_update_inode(trans, root, inode);
94b60442
CM		 5561 if (ret && ret == -ENOSPC) {
5562 /* whoops, lets try again with the full transaction */
5563 btrfs_end_transaction(trans, root);
5564 trans = btrfs_start_transaction(root, 1);
22c44fe6
JB		 5565 if (IS_ERR(trans))
5566 return PTR_ERR(trans);
8929ecfa 5567
94b60442 5568 ret = btrfs_update_inode(trans, root, inode);
94b60442 5569 }
39279cc3 5570 btrfs_end_transaction(trans, root);
16cdcec7
MX		 5571 if (BTRFS_I(inode)->delayed_node)
5572 btrfs_balance_delayed_items(root);
22c44fe6
JB		 5573
5574 return ret;
5575}
5576
5577/*
5578 * This is a copy of file_update_time. We need this so we can return error on
5579 * ENOSPC for updating the inode in the case of file write and mmap writes.
5580 */
e41f941a
JB		 5581static int btrfs_update_time(struct inode *inode, struct timespec *now,
5582 int flags)
22c44fe6 5583{
2bc55652
AB		 5584 struct btrfs_root *root = BTRFS_I(inode)->root;
5585
5586 if (btrfs_root_readonly(root))
5587 return -EROFS;
5588
e41f941a 5589 if (flags & S_VERSION)
22c44fe6 5590 inode_inc_iversion(inode);
e41f941a
JB		 5591 if (flags & S_CTIME)
5592 inode->i_ctime = *now;
5593 if (flags & S_MTIME)
5594 inode->i_mtime = *now;
5595 if (flags & S_ATIME)
5596 inode->i_atime = *now;
5597 return btrfs_dirty_inode(inode);
39279cc3
CM		 5598}
5599
d352ac68
CM		 5600/*
5601 * find the highest existing sequence number in a directory
5602 * and then set the in-memory index_cnt variable to reflect
5603 * free sequence numbers
5604 */
aec7477b
JB		 5605static int btrfs_set_inode_index_count(struct inode *inode)
5606{
5607 struct btrfs_root *root = BTRFS_I(inode)->root;
5608 struct btrfs_key key, found_key;
5609 struct btrfs_path *path;
5610 struct extent_buffer *leaf;
5611 int ret;
5612
33345d01 5613 key.objectid = btrfs_ino(inode);
962a298f 5614 key.type = BTRFS_DIR_INDEX_KEY;
aec7477b
JB		 5615 key.offset = (u64)-1;
5616
5617 path = btrfs_alloc_path();
5618 if (!path)
5619 return -ENOMEM;
5620
5621 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5622 if (ret < 0)
5623 goto out;
5624 /* FIXME: we should be able to handle this */
5625 if (ret == 0)
5626 goto out;
5627 ret = 0;
5628
5629 /*
5630 * MAGIC NUMBER EXPLANATION:
5631 * since we search a directory based on f_pos we have to start at 2
5632 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5633 * else has to start at 2
5634 */
5635 if (path->slots[0] == 0) {
5636 BTRFS_I(inode)->index_cnt = 2;
5637 goto out;
5638 }
5639
5640 path->slots[0]--;
5641
5642 leaf = path->nodes[0];
5643 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5644
33345d01 5645 if (found_key.objectid != btrfs_ino(inode) ||
962a298f 5646 found_key.type != BTRFS_DIR_INDEX_KEY) {
aec7477b
JB		 5647 BTRFS_I(inode)->index_cnt = 2;
5648 goto out;
5649 }
5650
5651 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5652out:
5653 btrfs_free_path(path);
5654 return ret;
5655}
5656
d352ac68
CM		 5657/*
5658 * helper to find a free sequence number in a given directory. This current
5659 * code is very simple, later versions will do smarter things in the btree
5660 */
3de4586c 5661int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 5662{
5663 int ret = 0;
5664
5665 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
16cdcec7
MX		 5666 ret = btrfs_inode_delayed_dir_index_count(dir);
5667 if (ret) {
5668 ret = btrfs_set_inode_index_count(dir);
5669 if (ret)
5670 return ret;
5671 }
aec7477b
JB		 5672 }
5673
00e4e6b3 5674 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 5675 BTRFS_I(dir)->index_cnt++;
5676
5677 return ret;
5678}
5679
b0d5d10f
CM		 5680static int btrfs_insert_inode_locked(struct inode *inode)
5681{
5682 struct btrfs_iget_args args;
5683 args.location = &BTRFS_I(inode)->location;
5684 args.root = BTRFS_I(inode)->root;
5685
5686 return insert_inode_locked4(inode,
5687 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
5688 btrfs_find_actor, &args);
5689}
5690
39279cc3
CM		 5691static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5692 struct btrfs_root *root,
aec7477b 5693 struct inode *dir,
9c58309d 5694 const char *name, int name_len,
175a4eb7
AV		 5695 u64 ref_objectid, u64 objectid,
5696 umode_t mode, u64 *index)
39279cc3
CM		 5697{
5698 struct inode *inode;
5f39d397 5699 struct btrfs_inode_item *inode_item;
39279cc3 5700 struct btrfs_key *location;
5f39d397 5701 struct btrfs_path *path;
9c58309d
CM		 5702 struct btrfs_inode_ref *ref;
5703 struct btrfs_key key[2];
5704 u32 sizes[2];
ef3b9af5 5705 int nitems = name ? 2 : 1;
9c58309d 5706 unsigned long ptr;
39279cc3 5707 int ret;
39279cc3 5708
5f39d397 5709 path = btrfs_alloc_path();
d8926bb3
MF		 5710 if (!path)
5711 return ERR_PTR(-ENOMEM);
5f39d397 5712
39279cc3 5713 inode = new_inode(root->fs_info->sb);
8fb27640
YS		 5714 if (!inode) {
5715 btrfs_free_path(path);
39279cc3 5716 return ERR_PTR(-ENOMEM);
8fb27640 5717 }
39279cc3 5718
5762b5c9
FM		 5719 /*
5720 * O_TMPFILE, set link count to 0, so that after this point,
5721 * we fill in an inode item with the correct link count.
5722 */
5723 if (!name)
5724 set_nlink(inode, 0);
5725
581bb050
LZ		 5726 /*
5727 * we have to initialize this early, so we can reclaim the inode
5728 * number if we fail afterwards in this function.
5729 */
5730 inode->i_ino = objectid;
5731
ef3b9af5 5732 if (dir && name) {
1abe9b8a 5733 trace_btrfs_inode_request(dir);
5734
3de4586c 5735 ret = btrfs_set_inode_index(dir, index);
09771430 5736 if (ret) {
8fb27640 5737 btrfs_free_path(path);
09771430 5738 iput(inode);
aec7477b 5739 return ERR_PTR(ret);
09771430 5740 }
ef3b9af5
FM		 5741 } else if (dir) {
5742 *index = 0;
aec7477b
JB		 5743 }
5744 /*
5745 * index_cnt is ignored for everything but a dir,
5746 * btrfs_get_inode_index_count has an explanation for the magic
5747 * number
5748 */
5749 BTRFS_I(inode)->index_cnt = 2;
67de1176 5750 BTRFS_I(inode)->dir_index = *index;
39279cc3 5751 BTRFS_I(inode)->root = root;
e02119d5 5752 BTRFS_I(inode)->generation = trans->transid;
76195853 5753 inode->i_generation = BTRFS_I(inode)->generation;
b888db2b 5754
5dc562c5
JB		 5755 /*
5756 * We could have gotten an inode number from somebody who was fsynced
5757 * and then removed in this same transaction, so let's just set full
5758 * sync since it will be a full sync anyway and this will blow away the
5759 * old info in the log.
5760 */
5761 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5762
9c58309d 5763 key[0].objectid = objectid;
962a298f 5764 key[0].type = BTRFS_INODE_ITEM_KEY;
9c58309d
CM		 5765 key[0].offset = 0;
5766
9c58309d 5767 sizes[0] = sizeof(struct btrfs_inode_item);
ef3b9af5
FM		 5768
5769 if (name) {
5770 /*
5771 * Start new inodes with an inode_ref. This is slightly more
5772 * efficient for small numbers of hard links since they will
5773 * be packed into one item. Extended refs will kick in if we
5774 * add more hard links than can fit in the ref item.
5775 */
5776 key[1].objectid = objectid;
962a298f 5777 key[1].type = BTRFS_INODE_REF_KEY;
ef3b9af5
FM		 5778 key[1].offset = ref_objectid;
5779
5780 sizes[1] = name_len + sizeof(*ref);
5781 }
9c58309d 5782
b0d5d10f
CM		 5783 location = &BTRFS_I(inode)->location;
5784 location->objectid = objectid;
5785 location->offset = 0;
962a298f 5786 location->type = BTRFS_INODE_ITEM_KEY;
b0d5d10f
CM		 5787
5788 ret = btrfs_insert_inode_locked(inode);
5789 if (ret < 0)
5790 goto fail;
5791
b9473439 5792 path->leave_spinning = 1;
ef3b9af5 5793 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
9c58309d 5794 if (ret != 0)
b0d5d10f 5795 goto fail_unlock;
5f39d397 5796
ecc11fab 5797 inode_init_owner(inode, dir, mode);
a76a3cd4 5798 inode_set_bytes(inode, 0);
39279cc3 5799 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5f39d397
CM		 5800 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5801 struct btrfs_inode_item);
293f7e07
LZ		 5802 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5803 sizeof(*inode_item));
e02119d5 5804 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d 5805
ef3b9af5
FM		 5806 if (name) {
5807 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5808 struct btrfs_inode_ref);
5809 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5810 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5811 ptr = (unsigned long)(ref + 1);
5812 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5813 }
9c58309d 5814
5f39d397
CM		 5815 btrfs_mark_buffer_dirty(path->nodes[0]);
5816 btrfs_free_path(path);
5817
6cbff00f
CH		 5818 btrfs_inherit_iflags(inode, dir);
5819
569254b0 5820 if (S_ISREG(mode)) {
94272164
CM		 5821 if (btrfs_test_opt(root, NODATASUM))
5822 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
213490b3 5823 if (btrfs_test_opt(root, NODATACOW))
f2bdf9a8
JB		 5824 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5825 BTRFS_INODE_NODATASUM;
94272164
CM		 5826 }
5827
5d4f98a2 5828 inode_tree_add(inode);
1abe9b8a 5829
5830 trace_btrfs_inode_new(inode);
1973f0fa 5831 btrfs_set_inode_last_trans(trans, inode);
1abe9b8a 5832
8ea05e3a
AB		 5833 btrfs_update_root_times(trans, root);
5834
63541927
FDBM		 5835 ret = btrfs_inode_inherit_props(trans, inode, dir);
5836 if (ret)
5837 btrfs_err(root->fs_info,
5838 "error inheriting props for ino %llu (root %llu): %d",
5839 btrfs_ino(inode), root->root_key.objectid, ret);
5840
39279cc3 5841 return inode;
b0d5d10f
CM		 5842
5843fail_unlock:
5844 unlock_new_inode(inode);
5f39d397 5845fail:
ef3b9af5 5846 if (dir && name)
aec7477b 5847 BTRFS_I(dir)->index_cnt--;
5f39d397 5848 btrfs_free_path(path);
09771430 5849 iput(inode);
5f39d397 5850 return ERR_PTR(ret);
39279cc3
CM		 5851}
5852
5853static inline u8 btrfs_inode_type(struct inode *inode)
5854{
5855 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5856}
5857
d352ac68
CM		 5858/*
5859 * utility function to add 'inode' into 'parent_inode' with
5860 * a give name and a given sequence number.
5861 * if 'add_backref' is true, also insert a backref from the
5862 * inode to the parent directory.
5863 */
e02119d5
CM		 5864int btrfs_add_link(struct btrfs_trans_handle *trans,
5865 struct inode *parent_inode, struct inode *inode,
5866 const char *name, int name_len, int add_backref, u64 index)
39279cc3 5867{
4df27c4d 5868 int ret = 0;
39279cc3 5869 struct btrfs_key key;
e02119d5 5870 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
33345d01
LZ		 5871 u64 ino = btrfs_ino(inode);
5872 u64 parent_ino = btrfs_ino(parent_inode);
5f39d397 5873
33345d01 5874 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 5875 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5876 } else {
33345d01 5877 key.objectid = ino;
962a298f 5878 key.type = BTRFS_INODE_ITEM_KEY;
4df27c4d
YZ		 5879 key.offset = 0;
5880 }
5881
33345d01 5882 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 5883 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5884 key.objectid, root->root_key.objectid,
33345d01 5885 parent_ino, index, name, name_len);
4df27c4d 5886 } else if (add_backref) {
33345d01
LZ		 5887 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5888 parent_ino, index);
4df27c4d 5889 }
39279cc3 5890
79787eaa
JM		 5891 /* Nothing to clean up yet */
5892 if (ret)
5893 return ret;
4df27c4d 5894
79787eaa
JM		 5895 ret = btrfs_insert_dir_item(trans, root, name, name_len,
5896 parent_inode, &key,
5897 btrfs_inode_type(inode), index);
9c52057c 5898 if (ret == -EEXIST || ret == -EOVERFLOW)
79787eaa
JM		 5899 goto fail_dir_item;
5900 else if (ret) {
5901 btrfs_abort_transaction(trans, root, ret);
5902 return ret;
39279cc3 5903 }
79787eaa
JM		 5904
5905 btrfs_i_size_write(parent_inode, parent_inode->i_size +
5906 name_len * 2);
0c4d2d95 5907 inode_inc_iversion(parent_inode);
79787eaa
JM		 5908 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5909 ret = btrfs_update_inode(trans, root, parent_inode);
5910 if (ret)
5911 btrfs_abort_transaction(trans, root, ret);
39279cc3 5912 return ret;
fe66a05a
CM		 5913
5914fail_dir_item:
5915 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5916 u64 local_index;
5917 int err;
5918 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5919 key.objectid, root->root_key.objectid,
5920 parent_ino, &local_index, name, name_len);
5921
5922 } else if (add_backref) {
5923 u64 local_index;
5924 int err;
5925
5926 err = btrfs_del_inode_ref(trans, root, name, name_len,
5927 ino, parent_ino, &local_index);
5928 }
5929 return ret;
39279cc3
CM		 5930}
5931
5932static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
a1b075d2
JB		 5933 struct inode *dir, struct dentry *dentry,
5934 struct inode *inode, int backref, u64 index)
39279cc3 5935{
a1b075d2
JB		 5936 int err = btrfs_add_link(trans, dir, inode,
5937 dentry->d_name.name, dentry->d_name.len,
5938 backref, index);
39279cc3
CM		 5939 if (err > 0)
5940 err = -EEXIST;
5941 return err;
5942}
5943
618e21d5 5944static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1a67aafb 5945 umode_t mode, dev_t rdev)
618e21d5
JB		 5946{
5947 struct btrfs_trans_handle *trans;
5948 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 5949 struct inode *inode = NULL;
618e21d5
JB		 5950 int err;
5951 int drop_inode = 0;
5952 u64 objectid;
00e4e6b3 5953 u64 index = 0;
618e21d5
JB		 5954
5955 if (!new_valid_dev(rdev))
5956 return -EINVAL;
5957
9ed74f2d
JB		 5958 /*
5959 * 2 for inode item and ref
5960 * 2 for dir items
5961 * 1 for xattr if selinux is on
5962 */
a22285a6
YZ		 5963 trans = btrfs_start_transaction(root, 5);
5964 if (IS_ERR(trans))
5965 return PTR_ERR(trans);
1832a6d5 5966
581bb050
LZ		 5967 err = btrfs_find_free_ino(root, &objectid);
5968 if (err)
5969 goto out_unlock;
5970
aec7477b 5971 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 5972 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 5973 mode, &index);
7cf96da3
TI		 5974 if (IS_ERR(inode)) {
5975 err = PTR_ERR(inode);
618e21d5 5976 goto out_unlock;
7cf96da3 5977 }
618e21d5 5978
ad19db71
CS		 5979 /*
5980 * If the active LSM wants to access the inode during
5981 * d_instantiate it needs these. Smack checks to see
5982 * if the filesystem supports xattrs by looking at the
5983 * ops vector.
5984 */
ad19db71 5985 inode->i_op = &btrfs_special_inode_operations;
b0d5d10f
CM		 5986 init_special_inode(inode, inode->i_mode, rdev);
5987
5988 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
618e21d5 5989 if (err)
b0d5d10f
CM		 5990 goto out_unlock_inode;
5991
5992 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5993 if (err) {
5994 goto out_unlock_inode;
5995 } else {
1b4ab1bb 5996 btrfs_update_inode(trans, root, inode);
b0d5d10f 5997 unlock_new_inode(inode);
08c422c2 5998 d_instantiate(dentry, inode);
618e21d5 5999 }
b0d5d10f 6000
618e21d5 6001out_unlock:
7ad85bb7 6002 btrfs_end_transaction(trans, root);
c581afc8 6003 btrfs_balance_delayed_items(root);
b53d3f5d 6004 btrfs_btree_balance_dirty(root);
618e21d5
JB		 6005 if (drop_inode) {
6006 inode_dec_link_count(inode);
6007 iput(inode);
6008 }
618e21d5 6009 return err;
b0d5d10f
CM		 6010
6011out_unlock_inode:
6012 drop_inode = 1;
6013 unlock_new_inode(inode);
6014 goto out_unlock;
6015
618e21d5
JB		 6016}
6017
39279cc3 6018static int btrfs_create(struct inode *dir, struct dentry *dentry,
ebfc3b49 6019 umode_t mode, bool excl)
39279cc3
CM		 6020{
6021 struct btrfs_trans_handle *trans;
6022 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 6023 struct inode *inode = NULL;
43baa579 6024 int drop_inode_on_err = 0;
a22285a6 6025 int err;
39279cc3 6026 u64 objectid;
00e4e6b3 6027 u64 index = 0;
39279cc3 6028
9ed74f2d
JB		 6029 /*
6030 * 2 for inode item and ref
6031 * 2 for dir items
6032 * 1 for xattr if selinux is on
6033 */
a22285a6
YZ		 6034 trans = btrfs_start_transaction(root, 5);
6035 if (IS_ERR(trans))
6036 return PTR_ERR(trans);
9ed74f2d 6037
581bb050
LZ		 6038 err = btrfs_find_free_ino(root, &objectid);
6039 if (err)
6040 goto out_unlock;
6041
aec7477b 6042 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 6043 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 6044 mode, &index);
7cf96da3
TI		 6045 if (IS_ERR(inode)) {
6046 err = PTR_ERR(inode);
39279cc3 6047 goto out_unlock;
7cf96da3 6048 }
43baa579 6049 drop_inode_on_err = 1;
ad19db71
CS		 6050 /*
6051 * If the active LSM wants to access the inode during
6052 * d_instantiate it needs these. Smack checks to see
6053 * if the filesystem supports xattrs by looking at the
6054 * ops vector.
6055 */
6056 inode->i_fop = &btrfs_file_operations;
6057 inode->i_op = &btrfs_file_inode_operations;
b0d5d10f
CM		 6058 inode->i_mapping->a_ops = &btrfs_aops;
6059 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
6060
6061 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6062 if (err)
6063 goto out_unlock_inode;
6064
6065 err = btrfs_update_inode(trans, root, inode);
6066 if (err)
6067 goto out_unlock_inode;
ad19db71 6068
a1b075d2 6069 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3 6070 if (err)
b0d5d10f 6071 goto out_unlock_inode;
43baa579 6072
43baa579 6073 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
b0d5d10f 6074 unlock_new_inode(inode);
43baa579
FB		 6075 d_instantiate(dentry, inode);
6076
39279cc3 6077out_unlock:
7ad85bb7 6078 btrfs_end_transaction(trans, root);
43baa579 6079 if (err && drop_inode_on_err) {
39279cc3
CM		 6080 inode_dec_link_count(inode);
6081 iput(inode);
6082 }
c581afc8 6083 btrfs_balance_delayed_items(root);
b53d3f5d 6084 btrfs_btree_balance_dirty(root);
39279cc3 6085 return err;
b0d5d10f
CM		 6086
6087out_unlock_inode:
6088 unlock_new_inode(inode);
6089 goto out_unlock;
6090
39279cc3
CM		 6091}
6092
6093static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6094 struct dentry *dentry)
6095{
6096 struct btrfs_trans_handle *trans;
6097 struct btrfs_root *root = BTRFS_I(dir)->root;
6098 struct inode *inode = old_dentry->d_inode;
00e4e6b3 6099 u64 index;
39279cc3
CM		 6100 int err;
6101 int drop_inode = 0;
6102
4a8be425
TH		 6103 /* do not allow sys_link's with other subvols of the same device */
6104 if (root->objectid != BTRFS_I(inode)->root->objectid)
3ab3564f 6105 return -EXDEV;
4a8be425 6106
f186373f 6107 if (inode->i_nlink >= BTRFS_LINK_MAX)
c055e99e 6108 return -EMLINK;
4a8be425 6109
3de4586c 6110 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 6111 if (err)
6112 goto fail;
6113
a22285a6 6114 /*
7e6b6465 6115 * 2 items for inode and inode ref
a22285a6 6116 * 2 items for dir items
7e6b6465 6117 * 1 item for parent inode
a22285a6 6118 */
7e6b6465 6119 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 6120 if (IS_ERR(trans)) {
6121 err = PTR_ERR(trans);
6122 goto fail;
6123 }
5f39d397 6124
67de1176
MX		 6125 /* There are several dir indexes for this inode, clear the cache. */
6126 BTRFS_I(inode)->dir_index = 0ULL;
8b558c5f 6127 inc_nlink(inode);
0c4d2d95 6128 inode_inc_iversion(inode);
3153495d 6129 inode->i_ctime = CURRENT_TIME;
7de9c6ee 6130 ihold(inode);
e9976151 6131 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
aec7477b 6132
a1b075d2 6133 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5f39d397 6134
a5719521 6135 if (err) {
54aa1f4d 6136 drop_inode = 1;
a5719521 6137 } else {
10d9f309 6138 struct dentry *parent = dentry->d_parent;
a5719521 6139 err = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 6140 if (err)
6141 goto fail;
ef3b9af5
FM		 6142 if (inode->i_nlink == 1) {
6143 /*
6144 * If new hard link count is 1, it's a file created
6145 * with open(2) O_TMPFILE flag.
6146 */
6147 err = btrfs_orphan_del(trans, inode);
6148 if (err)
6149 goto fail;
6150 }
08c422c2 6151 d_instantiate(dentry, inode);
6a912213 6152 btrfs_log_new_name(trans, inode, NULL, parent);
a5719521 6153 }
39279cc3 6154
7ad85bb7 6155 btrfs_end_transaction(trans, root);
c581afc8 6156 btrfs_balance_delayed_items(root);
1832a6d5 6157fail:
39279cc3
CM		 6158 if (drop_inode) {
6159 inode_dec_link_count(inode);
6160 iput(inode);
6161 }
b53d3f5d 6162 btrfs_btree_balance_dirty(root);
39279cc3
CM		 6163 return err;
6164}
6165
18bb1db3 6166static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
39279cc3 6167{
b9d86667 6168 struct inode *inode = NULL;
39279cc3
CM		 6169 struct btrfs_trans_handle *trans;
6170 struct btrfs_root *root = BTRFS_I(dir)->root;
6171 int err = 0;
6172 int drop_on_err = 0;
b9d86667 6173 u64 objectid = 0;
00e4e6b3 6174 u64 index = 0;
39279cc3 6175
9ed74f2d
JB		 6176 /*
6177 * 2 items for inode and ref
6178 * 2 items for dir items
6179 * 1 for xattr if selinux is on
6180 */
a22285a6
YZ		 6181 trans = btrfs_start_transaction(root, 5);
6182 if (IS_ERR(trans))
6183 return PTR_ERR(trans);
39279cc3 6184
581bb050
LZ		 6185 err = btrfs_find_free_ino(root, &objectid);
6186 if (err)
6187 goto out_fail;
6188
aec7477b 6189 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 6190 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 6191 S_IFDIR | mode, &index);
39279cc3
CM		 6192 if (IS_ERR(inode)) {
6193 err = PTR_ERR(inode);
6194 goto out_fail;
6195 }
5f39d397 6196
39279cc3 6197 drop_on_err = 1;
b0d5d10f
CM		 6198 /* these must be set before we unlock the inode */
6199 inode->i_op = &btrfs_dir_inode_operations;
6200 inode->i_fop = &btrfs_dir_file_operations;
33268eaf 6201
2a7dba39 6202 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf 6203 if (err)
b0d5d10f 6204 goto out_fail_inode;
39279cc3 6205
dbe674a9 6206 btrfs_i_size_write(inode, 0);
39279cc3
CM		 6207 err = btrfs_update_inode(trans, root, inode);
6208 if (err)
b0d5d10f 6209 goto out_fail_inode;
5f39d397 6210
a1b075d2
JB		 6211 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6212 dentry->d_name.len, 0, index);
39279cc3 6213 if (err)
b0d5d10f 6214 goto out_fail_inode;
5f39d397 6215
39279cc3 6216 d_instantiate(dentry, inode);
b0d5d10f
CM		 6217 /*
6218 * mkdir is special. We're unlocking after we call d_instantiate
6219 * to avoid a race with nfsd calling d_instantiate.
6220 */
6221 unlock_new_inode(inode);
39279cc3 6222 drop_on_err = 0;
39279cc3
CM		 6223
6224out_fail:
7ad85bb7 6225 btrfs_end_transaction(trans, root);
39279cc3
CM		 6226 if (drop_on_err)
6227 iput(inode);
c581afc8 6228 btrfs_balance_delayed_items(root);
b53d3f5d 6229 btrfs_btree_balance_dirty(root);
39279cc3 6230 return err;
b0d5d10f
CM		 6231
6232out_fail_inode:
6233 unlock_new_inode(inode);
6234 goto out_fail;
39279cc3
CM		 6235}
6236
d352ac68
CM		 6237/* helper for btfs_get_extent. Given an existing extent in the tree,
6238 * and an extent that you want to insert, deal with overlap and insert
6239 * the new extent into the tree.
6240 */
3b951516
CM		 6241static int merge_extent_mapping(struct extent_map_tree *em_tree,
6242 struct extent_map *existing,
e6dcd2dc 6243 struct extent_map *em,
51f395ad 6244 u64 map_start)
3b951516
CM		 6245{
6246 u64 start_diff;
3b951516 6247
e6dcd2dc
CM		 6248 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6249 start_diff = map_start - em->start;
6250 em->start = map_start;
51f395ad 6251 em->len = existing->start - em->start;
c8b97818
CM		 6252 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6253 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 6254 em->block_start += start_diff;
c8b97818
CM		 6255 em->block_len -= start_diff;
6256 }
09a2a8f9 6257 return add_extent_mapping(em_tree, em, 0);
3b951516
CM		 6258}
6259
c8b97818
CM		 6260static noinline int uncompress_inline(struct btrfs_path *path,
6261 struct inode *inode, struct page *page,
6262 size_t pg_offset, u64 extent_offset,
6263 struct btrfs_file_extent_item *item)
6264{
6265 int ret;
6266 struct extent_buffer *leaf = path->nodes[0];
6267 char *tmp;
6268 size_t max_size;
6269 unsigned long inline_size;
6270 unsigned long ptr;
261507a0 6271 int compress_type;
c8b97818
CM		 6272
6273 WARN_ON(pg_offset != 0);
261507a0 6274 compress_type = btrfs_file_extent_compression(leaf, item);
c8b97818
CM		 6275 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6276 inline_size = btrfs_file_extent_inline_item_len(leaf,
dd3cc16b 6277 btrfs_item_nr(path->slots[0]));
c8b97818 6278 tmp = kmalloc(inline_size, GFP_NOFS);
8d413713
TI		 6279 if (!tmp)
6280 return -ENOMEM;
c8b97818
CM		 6281 ptr = btrfs_file_extent_inline_start(item);
6282
6283 read_extent_buffer(leaf, tmp, ptr, inline_size);
6284
5b050f04 6285 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
261507a0
LZ		 6286 ret = btrfs_decompress(compress_type, tmp, page,
6287 extent_offset, inline_size, max_size);
c8b97818 6288 kfree(tmp);
166ae5a4 6289 return ret;
c8b97818
CM		 6290}
6291
d352ac68
CM		 6292/*
6293 * a bit scary, this does extent mapping from logical file offset to the disk.
d397712b
CM		 6294 * the ugly parts come from merging extents from the disk with the in-ram
6295 * representation. This gets more complex because of the data=ordered code,
d352ac68
CM		 6296 * where the in-ram extents might be locked pending data=ordered completion.
6297 *
6298 * This also copies inline extents directly into the page.
6299 */
d397712b 6300
a52d9a80 6301struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 6302 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 6303 int create)
6304{
6305 int ret;
6306 int err = 0;
a52d9a80
CM		 6307 u64 extent_start = 0;
6308 u64 extent_end = 0;
33345d01 6309 u64 objectid = btrfs_ino(inode);
a52d9a80 6310 u32 found_type;
f421950f 6311 struct btrfs_path *path = NULL;
a52d9a80
CM		 6312 struct btrfs_root *root = BTRFS_I(inode)->root;
6313 struct btrfs_file_extent_item *item;
5f39d397
CM		 6314 struct extent_buffer *leaf;
6315 struct btrfs_key found_key;
a52d9a80
CM		 6316 struct extent_map *em = NULL;
6317 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 6318 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 6319 struct btrfs_trans_handle *trans = NULL;
7ffbb598 6320 const bool new_inline = !page || create;
a52d9a80 6321
a52d9a80 6322again:
890871be 6323 read_lock(&em_tree->lock);
d1310b2e 6324 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 6325 if (em)
6326 em->bdev = root->fs_info->fs_devices->latest_bdev;
890871be 6327 read_unlock(&em_tree->lock);
d1310b2e 6328
a52d9a80 6329 if (em) {
e1c4b745
CM		 6330 if (em->start > start || em->start + em->len <= start)
6331 free_extent_map(em);
6332 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 6333 free_extent_map(em);
6334 else
6335 goto out;
a52d9a80 6336 }
172ddd60 6337 em = alloc_extent_map();
a52d9a80 6338 if (!em) {
d1310b2e
CM		 6339 err = -ENOMEM;
6340 goto out;
a52d9a80 6341 }
e6dcd2dc 6342 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 6343 em->start = EXTENT_MAP_HOLE;
445a6944 6344 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 6345 em->len = (u64)-1;
c8b97818 6346 em->block_len = (u64)-1;
f421950f
CM		 6347
6348 if (!path) {
6349 path = btrfs_alloc_path();
026fd317
JB		 6350 if (!path) {
6351 err = -ENOMEM;
6352 goto out;
6353 }
6354 /*
6355 * Chances are we'll be called again, so go ahead and do
6356 * readahead
6357 */
6358 path->reada = 1;
f421950f
CM		 6359 }
6360
179e29e4
CM		 6361 ret = btrfs_lookup_file_extent(trans, root, path,
6362 objectid, start, trans != NULL);
a52d9a80
CM		 6363 if (ret < 0) {
6364 err = ret;
6365 goto out;
6366 }
6367
6368 if (ret != 0) {
6369 if (path->slots[0] == 0)
6370 goto not_found;
6371 path->slots[0]--;
6372 }
6373
5f39d397
CM		 6374 leaf = path->nodes[0];
6375 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 6376 struct btrfs_file_extent_item);
a52d9a80 6377 /* are we inside the extent that was found? */
5f39d397 6378 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
962a298f 6379 found_type = found_key.type;
5f39d397 6380 if (found_key.objectid != objectid ||
a52d9a80 6381 found_type != BTRFS_EXTENT_DATA_KEY) {
25a50341
JB		 6382 /*
6383 * If we backup past the first extent we want to move forward
6384 * and see if there is an extent in front of us, otherwise we'll
6385 * say there is a hole for our whole search range which can
6386 * cause problems.
6387 */
6388 extent_end = start;
6389 goto next;
a52d9a80
CM		 6390 }
6391
5f39d397
CM		 6392 found_type = btrfs_file_extent_type(leaf, item);
6393 extent_start = found_key.offset;
d899e052
YZ		 6394 if (found_type == BTRFS_FILE_EXTENT_REG ||
6395 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 6396 extent_end = extent_start +
db94535d 6397 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 6398 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6399 size_t size;
514ac8ad 6400 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
fda2832f 6401 extent_end = ALIGN(extent_start + size, root->sectorsize);
9036c102 6402 }
25a50341 6403next:
9036c102
YZ		 6404 if (start >= extent_end) {
6405 path->slots[0]++;
6406 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6407 ret = btrfs_next_leaf(root, path);
6408 if (ret < 0) {
6409 err = ret;
6410 goto out;
a52d9a80 6411 }
9036c102
YZ		 6412 if (ret > 0)
6413 goto not_found;
6414 leaf = path->nodes[0];
a52d9a80 6415 }
9036c102
YZ		 6416 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6417 if (found_key.objectid != objectid ||
6418 found_key.type != BTRFS_EXTENT_DATA_KEY)
6419 goto not_found;
6420 if (start + len <= found_key.offset)
6421 goto not_found;
e2eca69d
WS		 6422 if (start > found_key.offset)
6423 goto next;
9036c102 6424 em->start = start;
70c8a91c 6425 em->orig_start = start;
9036c102
YZ		 6426 em->len = found_key.offset - start;
6427 goto not_found_em;
6428 }
6429
7ffbb598
FM		 6430 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6431
d899e052
YZ		 6432 if (found_type == BTRFS_FILE_EXTENT_REG ||
6433 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80
CM		 6434 goto insert;
6435 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 6436 unsigned long ptr;
a52d9a80 6437 char *map;
3326d1b0
CM		 6438 size_t size;
6439 size_t extent_offset;
6440 size_t copy_size;
a52d9a80 6441
7ffbb598 6442 if (new_inline)
689f9346 6443 goto out;
5f39d397 6444
514ac8ad 6445 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
9036c102 6446 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 6447 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 6448 size - extent_offset);
3326d1b0 6449 em->start = extent_start + extent_offset;
fda2832f 6450 em->len = ALIGN(copy_size, root->sectorsize);
b4939680 6451 em->orig_block_len = em->len;
70c8a91c 6452 em->orig_start = em->start;
689f9346 6453 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 6454 if (create == 0 && !PageUptodate(page)) {
261507a0
LZ		 6455 if (btrfs_file_extent_compression(leaf, item) !=
6456 BTRFS_COMPRESS_NONE) {
c8b97818
CM		 6457 ret = uncompress_inline(path, inode, page,
6458 pg_offset,
6459 extent_offset, item);
166ae5a4
ZB		 6460 if (ret) {
6461 err = ret;
6462 goto out;
6463 }
c8b97818
CM		 6464 } else {
6465 map = kmap(page);
6466 read_extent_buffer(leaf, map + pg_offset, ptr,
6467 copy_size);
93c82d57
CM		 6468 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6469 memset(map + pg_offset + copy_size, 0,
6470 PAGE_CACHE_SIZE - pg_offset -
6471 copy_size);
6472 }
c8b97818
CM		 6473 kunmap(page);
6474 }
179e29e4
CM		 6475 flush_dcache_page(page);
6476 } else if (create && PageUptodate(page)) {
6bf7e080 6477 BUG();
179e29e4
CM		 6478 if (!trans) {
6479 kunmap(page);
6480 free_extent_map(em);
6481 em = NULL;
ff5714cc 6482
b3b4aa74 6483 btrfs_release_path(path);
7a7eaa40 6484 trans = btrfs_join_transaction(root);
ff5714cc 6485
3612b495
TI		 6486 if (IS_ERR(trans))
6487 return ERR_CAST(trans);
179e29e4
CM		 6488 goto again;
6489 }
c8b97818 6490 map = kmap(page);
70dec807 6491 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 6492 copy_size);
c8b97818 6493 kunmap(page);
179e29e4 6494 btrfs_mark_buffer_dirty(leaf);
a52d9a80 6495 }
d1310b2e 6496 set_extent_uptodate(io_tree, em->start,
507903b8 6497 extent_map_end(em) - 1, NULL, GFP_NOFS);
a52d9a80 6498 goto insert;
a52d9a80
CM		 6499 }
6500not_found:
6501 em->start = start;
70c8a91c 6502 em->orig_start = start;
d1310b2e 6503 em->len = len;
a52d9a80 6504not_found_em:
5f39d397 6505 em->block_start = EXTENT_MAP_HOLE;
9036c102 6506 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80 6507insert:
b3b4aa74 6508 btrfs_release_path(path);
d1310b2e 6509 if (em->start > start || extent_map_end(em) <= start) {
c2cf52eb 6510 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
c1c9ff7c 6511 em->start, em->len, start, len);
a52d9a80
CM		 6512 err = -EIO;
6513 goto out;
6514 }
d1310b2e
CM		 6515
6516 err = 0;
890871be 6517 write_lock(&em_tree->lock);
09a2a8f9 6518 ret = add_extent_mapping(em_tree, em, 0);
3b951516
CM		 6519 /* it is possible that someone inserted the extent into the tree
6520 * while we had the lock dropped. It is also possible that
6521 * an overlapping map exists in the tree
6522 */
a52d9a80 6523 if (ret == -EEXIST) {
3b951516 6524 struct extent_map *existing;
e6dcd2dc
CM		 6525
6526 ret = 0;
6527
3b951516 6528 existing = lookup_extent_mapping(em_tree, start, len);
e1c4b745
CM		 6529 if (existing && (existing->start > start ||
6530 existing->start + existing->len <= start)) {
6531 free_extent_map(existing);
6532 existing = NULL;
6533 }
3b951516
CM		 6534 if (!existing) {
6535 existing = lookup_extent_mapping(em_tree, em->start,
6536 em->len);
6537 if (existing) {
6538 err = merge_extent_mapping(em_tree, existing,
51f395ad 6539 em, start);
3b951516
CM		 6540 free_extent_map(existing);
6541 if (err) {
6542 free_extent_map(em);
6543 em = NULL;
6544 }
6545 } else {
6546 err = -EIO;
3b951516
CM		 6547 free_extent_map(em);
6548 em = NULL;
6549 }
6550 } else {
6551 free_extent_map(em);
6552 em = existing;
e6dcd2dc 6553 err = 0;
a52d9a80 6554 }
a52d9a80 6555 }
890871be 6556 write_unlock(&em_tree->lock);
a52d9a80 6557out:
1abe9b8a 6558
4cd8587c 6559 trace_btrfs_get_extent(root, em);
1abe9b8a 6560
f421950f
CM		 6561 if (path)
6562 btrfs_free_path(path);
a52d9a80
CM		 6563 if (trans) {
6564 ret = btrfs_end_transaction(trans, root);
d397712b 6565 if (!err)
a52d9a80
CM		 6566 err = ret;
6567 }
a52d9a80
CM		 6568 if (err) {
6569 free_extent_map(em);
a52d9a80
CM		 6570 return ERR_PTR(err);
6571 }
79787eaa 6572 BUG_ON(!em); /* Error is always set */
a52d9a80
CM		 6573 return em;
6574}
6575
ec29ed5b
CM		 6576struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6577 size_t pg_offset, u64 start, u64 len,
6578 int create)
6579{
6580 struct extent_map *em;
6581 struct extent_map *hole_em = NULL;
6582 u64 range_start = start;
6583 u64 end;
6584 u64 found;
6585 u64 found_end;
6586 int err = 0;
6587
6588 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6589 if (IS_ERR(em))
6590 return em;
6591 if (em) {
6592 /*
f9e4fb53
LB		 6593 * if our em maps to
6594 * - a hole or
6595 * - a pre-alloc extent,
6596 * there might actually be delalloc bytes behind it.
ec29ed5b 6597 */
f9e4fb53
LB		 6598 if (em->block_start != EXTENT_MAP_HOLE &&
6599 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
ec29ed5b
CM		 6600 return em;
6601 else
6602 hole_em = em;
6603 }
6604
6605 /* check to see if we've wrapped (len == -1 or similar) */
6606 end = start + len;
6607 if (end < start)
6608 end = (u64)-1;
6609 else
6610 end -= 1;
6611
6612 em = NULL;
6613
6614 /* ok, we didn't find anything, lets look for delalloc */
6615 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6616 end, len, EXTENT_DELALLOC, 1);
6617 found_end = range_start + found;
6618 if (found_end < range_start)
6619 found_end = (u64)-1;
6620
6621 /*
6622 * we didn't find anything useful, return
6623 * the original results from get_extent()
6624 */
6625 if (range_start > end || found_end <= start) {
6626 em = hole_em;
6627 hole_em = NULL;
6628 goto out;
6629 }
6630
6631 /* adjust the range_start to make sure it doesn't
6632 * go backwards from the start they passed in
6633 */
67871254 6634 range_start = max(start, range_start);
ec29ed5b
CM		 6635 found = found_end - range_start;
6636
6637 if (found > 0) {
6638 u64 hole_start = start;
6639 u64 hole_len = len;
6640
172ddd60 6641 em = alloc_extent_map();
ec29ed5b
CM		 6642 if (!em) {
6643 err = -ENOMEM;
6644 goto out;
6645 }
6646 /*
6647 * when btrfs_get_extent can't find anything it
6648 * returns one huge hole
6649 *
6650 * make sure what it found really fits our range, and
6651 * adjust to make sure it is based on the start from
6652 * the caller
6653 */
6654 if (hole_em) {
6655 u64 calc_end = extent_map_end(hole_em);
6656
6657 if (calc_end <= start || (hole_em->start > end)) {
6658 free_extent_map(hole_em);
6659 hole_em = NULL;
6660 } else {
6661 hole_start = max(hole_em->start, start);
6662 hole_len = calc_end - hole_start;
6663 }
6664 }
6665 em->bdev = NULL;
6666 if (hole_em && range_start > hole_start) {
6667 /* our hole starts before our delalloc, so we
6668 * have to return just the parts of the hole
6669 * that go until the delalloc starts
6670 */
6671 em->len = min(hole_len,
6672 range_start - hole_start);
6673 em->start = hole_start;
6674 em->orig_start = hole_start;
6675 /*
6676 * don't adjust block start at all,
6677 * it is fixed at EXTENT_MAP_HOLE
6678 */
6679 em->block_start = hole_em->block_start;
6680 em->block_len = hole_len;
f9e4fb53
LB		 6681 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6682 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
ec29ed5b
CM		 6683 } else {
6684 em->start = range_start;
6685 em->len = found;
6686 em->orig_start = range_start;
6687 em->block_start = EXTENT_MAP_DELALLOC;
6688 em->block_len = found;
6689 }
6690 } else if (hole_em) {
6691 return hole_em;
6692 }
6693out:
6694
6695 free_extent_map(hole_em);
6696 if (err) {
6697 free_extent_map(em);
6698 return ERR_PTR(err);
6699 }
6700 return em;
6701}
6702
4b46fce2
JB		 6703static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6704 u64 start, u64 len)
6705{
6706 struct btrfs_root *root = BTRFS_I(inode)->root;
70c8a91c 6707 struct extent_map *em;
4b46fce2
JB		 6708 struct btrfs_key ins;
6709 u64 alloc_hint;
6710 int ret;
4b46fce2 6711
4b46fce2 6712 alloc_hint = get_extent_allocation_hint(inode, start, len);
00361589 6713 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
e570fd27 6714 alloc_hint, &ins, 1, 1);
00361589
JB		 6715 if (ret)
6716 return ERR_PTR(ret);
4b46fce2 6717
70c8a91c 6718 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
cc95bef6 6719 ins.offset, ins.offset, ins.offset, 0);
00361589 6720 if (IS_ERR(em)) {
e570fd27 6721 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
00361589
JB		 6722 return em;
6723 }
4b46fce2
JB		 6724
6725 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6726 ins.offset, ins.offset, 0);
6727 if (ret) {
e570fd27 6728 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
00361589
JB		 6729 free_extent_map(em);
6730 return ERR_PTR(ret);
4b46fce2 6731 }
00361589 6732
4b46fce2
JB		 6733 return em;
6734}
6735
46bfbb5c
CM		 6736/*
6737 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6738 * block must be cow'd
6739 */
00361589 6740noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7ee9e440
JB		 6741 u64 *orig_start, u64 *orig_block_len,
6742 u64 *ram_bytes)
46bfbb5c 6743{
00361589 6744 struct btrfs_trans_handle *trans;
46bfbb5c
CM		 6745 struct btrfs_path *path;
6746 int ret;
6747 struct extent_buffer *leaf;
6748 struct btrfs_root *root = BTRFS_I(inode)->root;
7b2b7085 6749 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
46bfbb5c
CM		 6750 struct btrfs_file_extent_item *fi;
6751 struct btrfs_key key;
6752 u64 disk_bytenr;
6753 u64 backref_offset;
6754 u64 extent_end;
6755 u64 num_bytes;
6756 int slot;
6757 int found_type;
7ee9e440 6758 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
e77751aa 6759
46bfbb5c
CM		 6760 path = btrfs_alloc_path();
6761 if (!path)
6762 return -ENOMEM;
6763
00361589 6764 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
46bfbb5c
CM		 6765 offset, 0);
6766 if (ret < 0)
6767 goto out;
6768
6769 slot = path->slots[0];
6770 if (ret == 1) {
6771 if (slot == 0) {
6772 /* can't find the item, must cow */
6773 ret = 0;
6774 goto out;
6775 }
6776 slot--;
6777 }
6778 ret = 0;
6779 leaf = path->nodes[0];
6780 btrfs_item_key_to_cpu(leaf, &key, slot);
33345d01 6781 if (key.objectid != btrfs_ino(inode) ||
46bfbb5c
CM		 6782 key.type != BTRFS_EXTENT_DATA_KEY) {
6783 /* not our file or wrong item type, must cow */
6784 goto out;
6785 }
6786
6787 if (key.offset > offset) {
6788 /* Wrong offset, must cow */
6789 goto out;
6790 }
6791
6792 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6793 found_type = btrfs_file_extent_type(leaf, fi);
6794 if (found_type != BTRFS_FILE_EXTENT_REG &&
6795 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6796 /* not a regular extent, must cow */
6797 goto out;
6798 }
7ee9e440
JB		 6799
6800 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6801 goto out;
6802
e77751aa
MX		 6803 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6804 if (extent_end <= offset)
6805 goto out;
6806
46bfbb5c 6807 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7ee9e440
JB		 6808 if (disk_bytenr == 0)
6809 goto out;
6810
6811 if (btrfs_file_extent_compression(leaf, fi) ||
6812 btrfs_file_extent_encryption(leaf, fi) ||
6813 btrfs_file_extent_other_encoding(leaf, fi))
6814 goto out;
6815
46bfbb5c
CM		 6816 backref_offset = btrfs_file_extent_offset(leaf, fi);
6817
7ee9e440
JB		 6818 if (orig_start) {
6819 *orig_start = key.offset - backref_offset;
6820 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6821 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6822 }
eb384b55 6823
46bfbb5c
CM		 6824 if (btrfs_extent_readonly(root, disk_bytenr))
6825 goto out;
7b2b7085
MX		 6826
6827 num_bytes = min(offset + *len, extent_end) - offset;
6828 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6829 u64 range_end;
6830
6831 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
6832 ret = test_range_bit(io_tree, offset, range_end,
6833 EXTENT_DELALLOC, 0, NULL);
6834 if (ret) {
6835 ret = -EAGAIN;
6836 goto out;
6837 }
6838 }
6839
1bda19eb 6840 btrfs_release_path(path);
46bfbb5c
CM		 6841
6842 /*
6843 * look for other files referencing this extent, if we
6844 * find any we must cow
6845 */
00361589
JB		 6846 trans = btrfs_join_transaction(root);
6847 if (IS_ERR(trans)) {
6848 ret = 0;
46bfbb5c 6849 goto out;
00361589
JB		 6850 }
6851
6852 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6853 key.offset - backref_offset, disk_bytenr);
6854 btrfs_end_transaction(trans, root);
6855 if (ret) {
6856 ret = 0;
6857 goto out;
6858 }
46bfbb5c
CM		 6859
6860 /*
6861 * adjust disk_bytenr and num_bytes to cover just the bytes
6862 * in this extent we are about to write. If there
6863 * are any csums in that range we have to cow in order
6864 * to keep the csums correct
6865 */
6866 disk_bytenr += backref_offset;
6867 disk_bytenr += offset - key.offset;
46bfbb5c
CM		 6868 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6869 goto out;
6870 /*
6871 * all of the above have passed, it is safe to overwrite this extent
6872 * without cow
6873 */
eb384b55 6874 *len = num_bytes;
46bfbb5c
CM		 6875 ret = 1;
6876out:
6877 btrfs_free_path(path);
6878 return ret;
6879}
6880
fc4adbff
AG		 6881bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
6882{
6883 struct radix_tree_root *root = &inode->i_mapping->page_tree;
6884 int found = false;
6885 void **pagep = NULL;
6886 struct page *page = NULL;
6887 int start_idx;
6888 int end_idx;
6889
6890 start_idx = start >> PAGE_CACHE_SHIFT;
6891
6892 /*
6893 * end is the last byte in the last page. end == start is legal
6894 */
6895 end_idx = end >> PAGE_CACHE_SHIFT;
6896
6897 rcu_read_lock();
6898
6899 /* Most of the code in this while loop is lifted from
6900 * find_get_page. It's been modified to begin searching from a
6901 * page and return just the first page found in that range. If the
6902 * found idx is less than or equal to the end idx then we know that
6903 * a page exists. If no pages are found or if those pages are
6904 * outside of the range then we're fine (yay!) */
6905 while (page == NULL &&
6906 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
6907 page = radix_tree_deref_slot(pagep);
6908 if (unlikely(!page))
6909 break;
6910
6911 if (radix_tree_exception(page)) {
809f9016
FM		 6912 if (radix_tree_deref_retry(page)) {
6913 page = NULL;
fc4adbff 6914 continue;
809f9016 6915 }
fc4adbff
AG		 6916 /*
6917 * Otherwise, shmem/tmpfs must be storing a swap entry
6918 * here as an exceptional entry: so return it without
6919 * attempting to raise page count.
6920 */
6fdef6d4 6921 page = NULL;
fc4adbff
AG		 6922 break; /* TODO: Is this relevant for this use case? */
6923 }
6924
91405151
FM		 6925 if (!page_cache_get_speculative(page)) {
6926 page = NULL;
fc4adbff 6927 continue;
91405151 6928 }
fc4adbff
AG		 6929
6930 /*
6931 * Has the page moved?
6932 * This is part of the lockless pagecache protocol. See
6933 * include/linux/pagemap.h for details.
6934 */
6935 if (unlikely(page != *pagep)) {
6936 page_cache_release(page);
6937 page = NULL;
6938 }
6939 }
6940
6941 if (page) {
6942 if (page->index <= end_idx)
6943 found = true;
6944 page_cache_release(page);
6945 }
6946
6947 rcu_read_unlock();
6948 return found;
6949}
6950
eb838e73
JB		 6951static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6952 struct extent_state **cached_state, int writing)
6953{
6954 struct btrfs_ordered_extent *ordered;
6955 int ret = 0;
6956
6957 while (1) {
6958 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6959 0, cached_state);
6960 /*
6961 * We're concerned with the entire range that we're going to be
6962 * doing DIO to, so we need to make sure theres no ordered
6963 * extents in this range.
6964 */
6965 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6966 lockend - lockstart + 1);
6967
6968 /*
6969 * We need to make sure there are no buffered pages in this
6970 * range either, we could have raced between the invalidate in
6971 * generic_file_direct_write and locking the extent. The
6972 * invalidate needs to happen so that reads after a write do not
6973 * get stale data.
6974 */
fc4adbff
AG		 6975 if (!ordered &&
6976 (!writing ||
6977 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
eb838e73
JB		 6978 break;
6979
6980 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6981 cached_state, GFP_NOFS);
6982
6983 if (ordered) {
6984 btrfs_start_ordered_extent(inode, ordered, 1);
6985 btrfs_put_ordered_extent(ordered);
6986 } else {
6987 /* Screw you mmap */
6988 ret = filemap_write_and_wait_range(inode->i_mapping,
6989 lockstart,
6990 lockend);
6991 if (ret)
6992 break;
6993
6994 /*
6995 * If we found a page that couldn't be invalidated just
6996 * fall back to buffered.
6997 */
6998 ret = invalidate_inode_pages2_range(inode->i_mapping,
6999 lockstart >> PAGE_CACHE_SHIFT,
7000 lockend >> PAGE_CACHE_SHIFT);
7001 if (ret)
7002 break;
7003 }
7004
7005 cond_resched();
7006 }
7007
7008 return ret;
7009}
7010
69ffb543
JB		 7011static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7012 u64 len, u64 orig_start,
7013 u64 block_start, u64 block_len,
cc95bef6
JB		 7014 u64 orig_block_len, u64 ram_bytes,
7015 int type)
69ffb543
JB		 7016{
7017 struct extent_map_tree *em_tree;
7018 struct extent_map *em;
7019 struct btrfs_root *root = BTRFS_I(inode)->root;
7020 int ret;
7021
7022 em_tree = &BTRFS_I(inode)->extent_tree;
7023 em = alloc_extent_map();
7024 if (!em)
7025 return ERR_PTR(-ENOMEM);
7026
7027 em->start = start;
7028 em->orig_start = orig_start;
2ab28f32
JB		 7029 em->mod_start = start;
7030 em->mod_len = len;
69ffb543
JB		 7031 em->len = len;
7032 em->block_len = block_len;
7033 em->block_start = block_start;
7034 em->bdev = root->fs_info->fs_devices->latest_bdev;
b4939680 7035 em->orig_block_len = orig_block_len;
cc95bef6 7036 em->ram_bytes = ram_bytes;
70c8a91c 7037 em->generation = -1;
69ffb543
JB		 7038 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7039 if (type == BTRFS_ORDERED_PREALLOC)
b11e234d 7040 set_bit(EXTENT_FLAG_FILLING, &em->flags);
69ffb543
JB		 7041
7042 do {
7043 btrfs_drop_extent_cache(inode, em->start,
7044 em->start + em->len - 1, 0);
7045 write_lock(&em_tree->lock);
09a2a8f9 7046 ret = add_extent_mapping(em_tree, em, 1);
69ffb543
JB		 7047 write_unlock(&em_tree->lock);
7048 } while (ret == -EEXIST);
7049
7050 if (ret) {
7051 free_extent_map(em);
7052 return ERR_PTR(ret);
7053 }
7054
7055 return em;
7056}
7057
7058
4b46fce2
JB		 7059static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7060 struct buffer_head *bh_result, int create)
7061{
7062 struct extent_map *em;
7063 struct btrfs_root *root = BTRFS_I(inode)->root;
eb838e73 7064 struct extent_state *cached_state = NULL;
4b46fce2 7065 u64 start = iblock << inode->i_blkbits;
eb838e73 7066 u64 lockstart, lockend;
4b46fce2 7067 u64 len = bh_result->b_size;
eb838e73 7068 int unlock_bits = EXTENT_LOCKED;
0934856d 7069 int ret = 0;
eb838e73 7070
172a5049 7071 if (create)
eb838e73 7072 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
172a5049 7073 else
c329861d 7074 len = min_t(u64, len, root->sectorsize);
eb838e73 7075
c329861d
JB		 7076 lockstart = start;
7077 lockend = start + len - 1;
7078
eb838e73
JB		 7079 /*
7080 * If this errors out it's because we couldn't invalidate pagecache for
7081 * this range and we need to fallback to buffered.
7082 */
7083 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
7084 return -ENOTBLK;
7085
4b46fce2 7086 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
eb838e73
JB		 7087 if (IS_ERR(em)) {
7088 ret = PTR_ERR(em);
7089 goto unlock_err;
7090 }
4b46fce2
JB		 7091
7092 /*
7093 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7094 * io. INLINE is special, and we could probably kludge it in here, but
7095 * it's still buffered so for safety lets just fall back to the generic
7096 * buffered path.
7097 *
7098 * For COMPRESSED we _have_ to read the entire extent in so we can
7099 * decompress it, so there will be buffering required no matter what we
7100 * do, so go ahead and fallback to buffered.
7101 *
7102 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7103 * to buffered IO. Don't blame me, this is the price we pay for using
7104 * the generic code.
7105 */
7106 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7107 em->block_start == EXTENT_MAP_INLINE) {
7108 free_extent_map(em);
eb838e73
JB		 7109 ret = -ENOTBLK;
7110 goto unlock_err;
4b46fce2
JB		 7111 }
7112
7113 /* Just a good old fashioned hole, return */
7114 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7115 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7116 free_extent_map(em);
eb838e73 7117 goto unlock_err;
4b46fce2
JB		 7118 }
7119
7120 /*
7121 * We don't allocate a new extent in the following cases
7122 *
7123 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7124 * existing extent.
7125 * 2) The extent is marked as PREALLOC. We're good to go here and can
7126 * just use the extent.
7127 *
7128 */
46bfbb5c 7129 if (!create) {
eb838e73
JB		 7130 len = min(len, em->len - (start - em->start));
7131 lockstart = start + len;
7132 goto unlock;
46bfbb5c 7133 }
4b46fce2
JB		 7134
7135 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7136 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7137 em->block_start != EXTENT_MAP_HOLE)) {
4b46fce2
JB		 7138 int type;
7139 int ret;
eb384b55 7140 u64 block_start, orig_start, orig_block_len, ram_bytes;
4b46fce2
JB		 7141
7142 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7143 type = BTRFS_ORDERED_PREALLOC;
7144 else
7145 type = BTRFS_ORDERED_NOCOW;
46bfbb5c 7146 len = min(len, em->len - (start - em->start));
4b46fce2 7147 block_start = em->block_start + (start - em->start);
46bfbb5c 7148
00361589 7149 if (can_nocow_extent(inode, start, &len, &orig_start,
7ee9e440 7150 &orig_block_len, &ram_bytes) == 1) {
69ffb543
JB		 7151 if (type == BTRFS_ORDERED_PREALLOC) {
7152 free_extent_map(em);
7153 em = create_pinned_em(inode, start, len,
7154 orig_start,
b4939680 7155 block_start, len,
cc95bef6
JB		 7156 orig_block_len,
7157 ram_bytes, type);
555e1286
FM		 7158 if (IS_ERR(em)) {
7159 ret = PTR_ERR(em);
69ffb543 7160 goto unlock_err;
555e1286 7161 }
69ffb543
JB		 7162 }
7163
46bfbb5c
CM		 7164 ret = btrfs_add_ordered_extent_dio(inode, start,
7165 block_start, len, len, type);
46bfbb5c
CM		 7166 if (ret) {
7167 free_extent_map(em);
eb838e73 7168 goto unlock_err;
46bfbb5c
CM		 7169 }
7170 goto unlock;
4b46fce2 7171 }
4b46fce2 7172 }
00361589 7173
46bfbb5c
CM		 7174 /*
7175 * this will cow the extent, reset the len in case we changed
7176 * it above
7177 */
7178 len = bh_result->b_size;
70c8a91c
JB		 7179 free_extent_map(em);
7180 em = btrfs_new_extent_direct(inode, start, len);
eb838e73
JB		 7181 if (IS_ERR(em)) {
7182 ret = PTR_ERR(em);
7183 goto unlock_err;
7184 }
46bfbb5c
CM		 7185 len = min(len, em->len - (start - em->start));
7186unlock:
4b46fce2
JB		 7187 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7188 inode->i_blkbits;
46bfbb5c 7189 bh_result->b_size = len;
4b46fce2
JB		 7190 bh_result->b_bdev = em->bdev;
7191 set_buffer_mapped(bh_result);
c3473e83
JB		 7192 if (create) {
7193 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7194 set_buffer_new(bh_result);
7195
7196 /*
7197 * Need to update the i_size under the extent lock so buffered
7198 * readers will get the updated i_size when we unlock.
7199 */
7200 if (start + len > i_size_read(inode))
7201 i_size_write(inode, start + len);
0934856d 7202
172a5049
MX		 7203 spin_lock(&BTRFS_I(inode)->lock);
7204 BTRFS_I(inode)->outstanding_extents++;
7205 spin_unlock(&BTRFS_I(inode)->lock);
7206
0934856d
MX		 7207 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7208 lockstart + len - 1, EXTENT_DELALLOC, NULL,
7209 &cached_state, GFP_NOFS);
7210 BUG_ON(ret);
c3473e83 7211 }
4b46fce2 7212
eb838e73
JB		 7213 /*
7214 * In the case of write we need to clear and unlock the entire range,
7215 * in the case of read we need to unlock only the end area that we
7216 * aren't using if there is any left over space.
7217 */
24c03fa5 7218 if (lockstart < lockend) {
0934856d
MX		 7219 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7220 lockend, unlock_bits, 1, 0,
7221 &cached_state, GFP_NOFS);
24c03fa5 7222 } else {
eb838e73 7223 free_extent_state(cached_state);
24c03fa5 7224 }
eb838e73 7225
4b46fce2
JB		 7226 free_extent_map(em);
7227
7228 return 0;
eb838e73
JB		 7229
7230unlock_err:
eb838e73
JB		 7231 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7232 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7233 return ret;
4b46fce2
JB		 7234}
7235
4b46fce2
JB		 7236static void btrfs_endio_direct_read(struct bio *bio, int err)
7237{
e65e1535 7238 struct btrfs_dio_private *dip = bio->bi_private;
2c30c71b 7239 struct bio_vec *bvec;
4b46fce2
JB		 7240 struct inode *inode = dip->inode;
7241 struct btrfs_root *root = BTRFS_I(inode)->root;
9be3395b 7242 struct bio *dio_bio;
facc8a22 7243 u32 *csums = (u32 *)dip->csum;
4b46fce2 7244 u64 start;
2c30c71b 7245 int i;
4b46fce2
JB		 7246
7247 start = dip->logical_offset;
2c30c71b 7248 bio_for_each_segment_all(bvec, bio, i) {
4b46fce2
JB		 7249 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
7250 struct page *page = bvec->bv_page;
7251 char *kaddr;
7252 u32 csum = ~(u32)0;
7253 unsigned long flags;
7254
7255 local_irq_save(flags);
7ac687d9 7256 kaddr = kmap_atomic(page);
b0496686 7257 csum = btrfs_csum_data(kaddr + bvec->bv_offset,
4b46fce2
JB		 7258 csum, bvec->bv_len);
7259 btrfs_csum_final(csum, (char *)&csum);
7ac687d9 7260 kunmap_atomic(kaddr);
4b46fce2
JB		 7261 local_irq_restore(flags);
7262
7263 flush_dcache_page(bvec->bv_page);
2c30c71b 7264 if (csum != csums[i]) {
facc8a22 7265 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
c1c9ff7c 7266 btrfs_ino(inode), start, csum,
2c30c71b 7267 csums[i]);
4b46fce2
JB		 7268 err = -EIO;
7269 }
7270 }
7271
7272 start += bvec->bv_len;
2c30c71b 7273 }
4b46fce2
JB		 7274
7275 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
d0082371 7276 dip->logical_offset + dip->bytes - 1);
9be3395b 7277 dio_bio = dip->dio_bio;
4b46fce2 7278
4b46fce2 7279 kfree(dip);
c0da7aa1
JB		 7280
7281 /* If we had a csum failure make sure to clear the uptodate flag */
7282 if (err)
9be3395b
CM		 7283 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7284 dio_end_io(dio_bio, err);
7285 bio_put(bio);
4b46fce2
JB		 7286}
7287
7288static void btrfs_endio_direct_write(struct bio *bio, int err)
7289{
7290 struct btrfs_dio_private *dip = bio->bi_private;
7291 struct inode *inode = dip->inode;
7292 struct btrfs_root *root = BTRFS_I(inode)->root;
4b46fce2 7293 struct btrfs_ordered_extent *ordered = NULL;
163cf09c
CM		 7294 u64 ordered_offset = dip->logical_offset;
7295 u64 ordered_bytes = dip->bytes;
9be3395b 7296 struct bio *dio_bio;
4b46fce2
JB		 7297 int ret;
7298
7299 if (err)
7300 goto out_done;
163cf09c
CM		 7301again:
7302 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7303 &ordered_offset,
5fd02043 7304 ordered_bytes, !err);
4b46fce2 7305 if (!ret)
163cf09c 7306 goto out_test;
4b46fce2 7307
9e0af237
LB		 7308 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
7309 finish_ordered_fn, NULL, NULL);
fccb5d86
QW		 7310 btrfs_queue_work(root->fs_info->endio_write_workers,
7311 &ordered->work);
163cf09c
CM		 7312out_test:
7313 /*
7314 * our bio might span multiple ordered extents. If we haven't
7315 * completed the accounting for the whole dio, go back and try again
7316 */
7317 if (ordered_offset < dip->logical_offset + dip->bytes) {
7318 ordered_bytes = dip->logical_offset + dip->bytes -
7319 ordered_offset;
5fd02043 7320 ordered = NULL;
163cf09c
CM		 7321 goto again;
7322 }
4b46fce2 7323out_done:
9be3395b 7324 dio_bio = dip->dio_bio;
4b46fce2 7325
4b46fce2 7326 kfree(dip);
c0da7aa1
JB		 7327
7328 /* If we had an error make sure to clear the uptodate flag */
7329 if (err)
9be3395b
CM		 7330 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7331 dio_end_io(dio_bio, err);
7332 bio_put(bio);
4b46fce2
JB		 7333}
7334
eaf25d93
CM		 7335static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7336 struct bio *bio, int mirror_num,
7337 unsigned long bio_flags, u64 offset)
7338{
7339 int ret;
7340 struct btrfs_root *root = BTRFS_I(inode)->root;
7341 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
79787eaa 7342 BUG_ON(ret); /* -ENOMEM */
eaf25d93
CM		 7343 return 0;
7344}
7345
e65e1535
MX		 7346static void btrfs_end_dio_bio(struct bio *bio, int err)
7347{
7348 struct btrfs_dio_private *dip = bio->bi_private;
7349
7350 if (err) {
efe120a0
FH		 7351 btrfs_err(BTRFS_I(dip->inode)->root->fs_info,
7352 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
c1c9ff7c 7353 btrfs_ino(dip->inode), bio->bi_rw,
4f024f37
KO		 7354 (unsigned long long)bio->bi_iter.bi_sector,
7355 bio->bi_iter.bi_size, err);
e65e1535
MX		 7356 dip->errors = 1;
7357
7358 /*
7359 * before atomic variable goto zero, we must make sure
7360 * dip->errors is perceived to be set.
7361 */
4e857c58 7362 smp_mb__before_atomic();
e65e1535
MX		 7363 }
7364
7365 /* if there are more bios still pending for this dio, just exit */
7366 if (!atomic_dec_and_test(&dip->pending_bios))
7367 goto out;
7368
9be3395b 7369 if (dip->errors) {
e65e1535 7370 bio_io_error(dip->orig_bio);
9be3395b
CM		 7371 } else {
7372 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
e65e1535
MX		 7373 bio_endio(dip->orig_bio, 0);
7374 }
7375out:
7376 bio_put(bio);
7377}
7378
7379static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7380 u64 first_sector, gfp_t gfp_flags)
7381{
7382 int nr_vecs = bio_get_nr_vecs(bdev);
7383 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7384}
7385
7386static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7387 int rw, u64 file_offset, int skip_sum,
c329861d 7388 int async_submit)
e65e1535 7389{
facc8a22 7390 struct btrfs_dio_private *dip = bio->bi_private;
e65e1535
MX		 7391 int write = rw & REQ_WRITE;
7392 struct btrfs_root *root = BTRFS_I(inode)->root;
7393 int ret;
7394
b812ce28
JB		 7395 if (async_submit)
7396 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7397
e65e1535 7398 bio_get(bio);
5fd02043
JB		 7399
7400 if (!write) {
7401 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7402 if (ret)
7403 goto err;
7404 }
e65e1535 7405
1ae39938
JB		 7406 if (skip_sum)
7407 goto map;
7408
7409 if (write && async_submit) {
e65e1535
MX		 7410 ret = btrfs_wq_submit_bio(root->fs_info,
7411 inode, rw, bio, 0, 0,
7412 file_offset,
7413 __btrfs_submit_bio_start_direct_io,
7414 __btrfs_submit_bio_done);
7415 goto err;
1ae39938
JB		 7416 } else if (write) {
7417 /*
7418 * If we aren't doing async submit, calculate the csum of the
7419 * bio now.
7420 */
7421 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7422 if (ret)
7423 goto err;
c2db1073 7424 } else if (!skip_sum) {
facc8a22
MX		 7425 ret = btrfs_lookup_bio_sums_dio(root, inode, dip, bio,
7426 file_offset);
c2db1073
TI		 7427 if (ret)
7428 goto err;
7429 }
e65e1535 7430
1ae39938
JB		 7431map:
7432 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
e65e1535
MX		 7433err:
7434 bio_put(bio);
7435 return ret;
7436}
7437
7438static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7439 int skip_sum)
7440{
7441 struct inode *inode = dip->inode;
7442 struct btrfs_root *root = BTRFS_I(inode)->root;
e65e1535
MX		 7443 struct bio *bio;
7444 struct bio *orig_bio = dip->orig_bio;
7445 struct bio_vec *bvec = orig_bio->bi_io_vec;
4f024f37 7446 u64 start_sector = orig_bio->bi_iter.bi_sector;
e65e1535
MX		 7447 u64 file_offset = dip->logical_offset;
7448 u64 submit_len = 0;
7449 u64 map_length;
7450 int nr_pages = 0;
e65e1535 7451 int ret = 0;
1ae39938 7452 int async_submit = 0;
e65e1535 7453
4f024f37 7454 map_length = orig_bio->bi_iter.bi_size;
53b381b3 7455 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
e65e1535 7456 &map_length, NULL, 0);
7a5c3c9b 7457 if (ret)
e65e1535 7458 return -EIO;
facc8a22 7459
4f024f37 7460 if (map_length >= orig_bio->bi_iter.bi_size) {
02f57c7a
JB		 7461 bio = orig_bio;
7462 goto submit;
7463 }
7464
53b381b3
DW		 7465 /* async crcs make it difficult to collect full stripe writes. */
7466 if (btrfs_get_alloc_profile(root, 1) &
7467 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7468 async_submit = 0;
7469 else
7470 async_submit = 1;
7471
02f57c7a
JB		 7472 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7473 if (!bio)
7474 return -ENOMEM;
7a5c3c9b 7475
02f57c7a
JB		 7476 bio->bi_private = dip;
7477 bio->bi_end_io = btrfs_end_dio_bio;
7478 atomic_inc(&dip->pending_bios);
7479
e65e1535
MX		 7480 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7481 if (unlikely(map_length < submit_len + bvec->bv_len ||
7482 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7483 bvec->bv_offset) < bvec->bv_len)) {
7484 /*
7485 * inc the count before we submit the bio so
7486 * we know the end IO handler won't happen before
7487 * we inc the count. Otherwise, the dip might get freed
7488 * before we're done setting it up
7489 */
7490 atomic_inc(&dip->pending_bios);
7491 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7492 file_offset, skip_sum,
c329861d 7493 async_submit);
e65e1535
MX		 7494 if (ret) {
7495 bio_put(bio);
7496 atomic_dec(&dip->pending_bios);
7497 goto out_err;
7498 }
7499
e65e1535
MX		 7500 start_sector += submit_len >> 9;
7501 file_offset += submit_len;
7502
7503 submit_len = 0;
7504 nr_pages = 0;
7505
7506 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7507 start_sector, GFP_NOFS);
7508 if (!bio)
7509 goto out_err;
7510 bio->bi_private = dip;
7511 bio->bi_end_io = btrfs_end_dio_bio;
7512
4f024f37 7513 map_length = orig_bio->bi_iter.bi_size;
53b381b3 7514 ret = btrfs_map_block(root->fs_info, rw,
3ec706c8 7515 start_sector << 9,
e65e1535
MX		 7516 &map_length, NULL, 0);
7517 if (ret) {
7518 bio_put(bio);
7519 goto out_err;
7520 }
7521 } else {
7522 submit_len += bvec->bv_len;
67871254 7523 nr_pages++;
e65e1535
MX		 7524 bvec++;
7525 }
7526 }
7527
02f57c7a 7528submit:
e65e1535 7529 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
c329861d 7530 async_submit);
e65e1535
MX		 7531 if (!ret)
7532 return 0;
7533
7534 bio_put(bio);
7535out_err:
7536 dip->errors = 1;
7537 /*
7538 * before atomic variable goto zero, we must
7539 * make sure dip->errors is perceived to be set.
7540 */
4e857c58 7541 smp_mb__before_atomic();
e65e1535
MX		 7542 if (atomic_dec_and_test(&dip->pending_bios))
7543 bio_io_error(dip->orig_bio);
7544
7545 /* bio_end_io() will handle error, so we needn't return it */
7546 return 0;
7547}
7548
9be3395b
CM		 7549static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7550 struct inode *inode, loff_t file_offset)
4b46fce2
JB		 7551{
7552 struct btrfs_root *root = BTRFS_I(inode)->root;
7553 struct btrfs_dio_private *dip;
9be3395b 7554 struct bio *io_bio;
4b46fce2 7555 int skip_sum;
facc8a22 7556 int sum_len;
7b6d91da 7557 int write = rw & REQ_WRITE;
4b46fce2 7558 int ret = 0;
facc8a22 7559 u16 csum_size;
4b46fce2
JB		 7560
7561 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7562
9be3395b 7563 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
9be3395b
CM		 7564 if (!io_bio) {
7565 ret = -ENOMEM;
7566 goto free_ordered;
7567 }
7568
facc8a22
MX		 7569 if (!skip_sum && !write) {
7570 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
4f024f37
KO		 7571 sum_len = dio_bio->bi_iter.bi_size >>
7572 inode->i_sb->s_blocksize_bits;
facc8a22
MX		 7573 sum_len *= csum_size;
7574 } else {
7575 sum_len = 0;
7576 }
7577
7578 dip = kmalloc(sizeof(*dip) + sum_len, GFP_NOFS);
4b46fce2
JB		 7579 if (!dip) {
7580 ret = -ENOMEM;
9be3395b 7581 goto free_io_bio;
4b46fce2 7582 }
4b46fce2 7583
9be3395b 7584 dip->private = dio_bio->bi_private;
4b46fce2
JB		 7585 dip->inode = inode;
7586 dip->logical_offset = file_offset;
4f024f37
KO		 7587 dip->bytes = dio_bio->bi_iter.bi_size;
7588 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
9be3395b 7589 io_bio->bi_private = dip;
e65e1535 7590 dip->errors = 0;
9be3395b
CM		 7591 dip->orig_bio = io_bio;
7592 dip->dio_bio = dio_bio;
e65e1535 7593 atomic_set(&dip->pending_bios, 0);
4b46fce2
JB		 7594
7595 if (write)
9be3395b 7596 io_bio->bi_end_io = btrfs_endio_direct_write;
4b46fce2 7597 else
9be3395b 7598 io_bio->bi_end_io = btrfs_endio_direct_read;
4b46fce2 7599
e65e1535
MX		 7600 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7601 if (!ret)
eaf25d93 7602 return;
9be3395b
CM		 7603
7604free_io_bio:
7605 bio_put(io_bio);
7606
4b46fce2
JB		 7607free_ordered:
7608 /*
7609 * If this is a write, we need to clean up the reserved space and kill
7610 * the ordered extent.
7611 */
7612 if (write) {
7613 struct btrfs_ordered_extent *ordered;
955256f2 7614 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
4b46fce2
JB		 7615 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7616 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7617 btrfs_free_reserved_extent(root, ordered->start,
e570fd27 7618 ordered->disk_len, 1);
4b46fce2
JB		 7619 btrfs_put_ordered_extent(ordered);
7620 btrfs_put_ordered_extent(ordered);
7621 }
9be3395b 7622 bio_endio(dio_bio, ret);
4b46fce2
JB		 7623}
7624
5a5f79b5 7625static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
28060d5d 7626 const struct iov_iter *iter, loff_t offset)
5a5f79b5
CM		 7627{
7628 int seg;
a1b75f7d 7629 int i;
5a5f79b5
CM		 7630 unsigned blocksize_mask = root->sectorsize - 1;
7631 ssize_t retval = -EINVAL;
5a5f79b5
CM		 7632
7633 if (offset & blocksize_mask)
7634 goto out;
7635
28060d5d
AV		 7636 if (iov_iter_alignment(iter) & blocksize_mask)
7637 goto out;
a1b75f7d 7638
28060d5d
AV		 7639 /* If this is a write we don't need to check anymore */
7640 if (rw & WRITE)
7641 return 0;
7642 /*
7643 * Check to make sure we don't have duplicate iov_base's in this
7644 * iovec, if so return EINVAL, otherwise we'll get csum errors
7645 * when reading back.
7646 */
7647 for (seg = 0; seg < iter->nr_segs; seg++) {
7648 for (i = seg + 1; i < iter->nr_segs; i++) {
7649 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
a1b75f7d
JB		 7650 goto out;
7651 }
5a5f79b5
CM		 7652 }
7653 retval = 0;
7654out:
7655 return retval;
7656}
eb838e73 7657
16432985 7658static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
d8d3d94b 7659 struct iov_iter *iter, loff_t offset)
16432985 7660{
4b46fce2
JB		 7661 struct file *file = iocb->ki_filp;
7662 struct inode *inode = file->f_mapping->host;
0934856d 7663 size_t count = 0;
2e60a51e 7664 int flags = 0;
38851cc1
MX		 7665 bool wakeup = true;
7666 bool relock = false;
0934856d 7667 ssize_t ret;
4b46fce2 7668
28060d5d 7669 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iter, offset))
5a5f79b5 7670 return 0;
3f7c579c 7671
38851cc1 7672 atomic_inc(&inode->i_dio_count);
4e857c58 7673 smp_mb__after_atomic();
38851cc1 7674
0e267c44 7675 /*
41bd9ca4
MX		 7676 * The generic stuff only does filemap_write_and_wait_range, which
7677 * isn't enough if we've written compressed pages to this area, so
7678 * we need to flush the dirty pages again to make absolutely sure
7679 * that any outstanding dirty pages are on disk.
0e267c44 7680 */
a6cbcd4a 7681 count = iov_iter_count(iter);
41bd9ca4
MX		 7682 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7683 &BTRFS_I(inode)->runtime_flags))
9a025a08
WS		 7684 filemap_fdatawrite_range(inode->i_mapping, offset,
7685 offset + count - 1);
0e267c44 7686
0934856d 7687 if (rw & WRITE) {
38851cc1
MX		 7688 /*
7689 * If the write DIO is beyond the EOF, we need update
7690 * the isize, but it is protected by i_mutex. So we can
7691 * not unlock the i_mutex at this case.
7692 */
7693 if (offset + count <= inode->i_size) {
7694 mutex_unlock(&inode->i_mutex);
7695 relock = true;
7696 }
0934856d
MX		 7697 ret = btrfs_delalloc_reserve_space(inode, count);
7698 if (ret)
38851cc1
MX		 7699 goto out;
7700 } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7701 &BTRFS_I(inode)->runtime_flags))) {
7702 inode_dio_done(inode);
7703 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7704 wakeup = false;
0934856d
MX		 7705 }
7706
7707 ret = __blockdev_direct_IO(rw, iocb, inode,
7708 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
31b14039 7709 iter, offset, btrfs_get_blocks_direct, NULL,
2e60a51e 7710 btrfs_submit_direct, flags);
0934856d
MX		 7711 if (rw & WRITE) {
7712 if (ret < 0 && ret != -EIOCBQUEUED)
7713 btrfs_delalloc_release_space(inode, count);
172a5049 7714 else if (ret >= 0 && (size_t)ret < count)
0934856d
MX		 7715 btrfs_delalloc_release_space(inode,
7716 count - (size_t)ret);
172a5049
MX		 7717 else
7718 btrfs_delalloc_release_metadata(inode, 0);
0934856d 7719 }
38851cc1 7720out:
2e60a51e
MX		 7721 if (wakeup)
7722 inode_dio_done(inode);
38851cc1
MX		 7723 if (relock)
7724 mutex_lock(&inode->i_mutex);
0934856d
MX		 7725
7726 return ret;
16432985
CM		 7727}
7728
05dadc09
TI		 7729#define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
7730
1506fcc8
YS		 7731static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7732 __u64 start, __u64 len)
7733{
05dadc09
TI		 7734 int ret;
7735
7736 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7737 if (ret)
7738 return ret;
7739
ec29ed5b 7740 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
1506fcc8
YS		 7741}
7742
a52d9a80 7743int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 7744{
d1310b2e
CM		 7745 struct extent_io_tree *tree;
7746 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 7747 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
9ebefb18 7748}
1832a6d5 7749
a52d9a80 7750static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 7751{
d1310b2e 7752 struct extent_io_tree *tree;
b888db2b
CM		 7753
7754
7755 if (current->flags & PF_MEMALLOC) {
7756 redirty_page_for_writepage(wbc, page);
7757 unlock_page(page);
7758 return 0;
7759 }
d1310b2e 7760 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 7761 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM		 7762}
7763
48a3b636
ES		 7764static int btrfs_writepages(struct address_space *mapping,
7765 struct writeback_control *wbc)
b293f02e 7766{
d1310b2e 7767 struct extent_io_tree *tree;
771ed689 7768
d1310b2e 7769 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 7770 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7771}
7772
3ab2fb5a
CM		 7773static int
7774btrfs_readpages(struct file *file, struct address_space *mapping,
7775 struct list_head *pages, unsigned nr_pages)
7776{
d1310b2e
CM		 7777 struct extent_io_tree *tree;
7778 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 7779 return extent_readpages(tree, mapping, pages, nr_pages,
7780 btrfs_get_extent);
7781}
e6dcd2dc 7782static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 7783{
d1310b2e
CM		 7784 struct extent_io_tree *tree;
7785 struct extent_map_tree *map;
a52d9a80 7786 int ret;
8c2383c3 7787
d1310b2e
CM		 7788 tree = &BTRFS_I(page->mapping->host)->io_tree;
7789 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 7790 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 7791 if (ret == 1) {
7792 ClearPagePrivate(page);
7793 set_page_private(page, 0);
7794 page_cache_release(page);
39279cc3 7795 }
a52d9a80 7796 return ret;
39279cc3
CM		 7797}
7798
e6dcd2dc
CM		 7799static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7800{
98509cfc
CM		 7801 if (PageWriteback(page) || PageDirty(page))
7802 return 0;
b335b003 7803 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
e6dcd2dc
CM		 7804}
7805
d47992f8
LC		 7806static void btrfs_invalidatepage(struct page *page, unsigned int offset,
7807 unsigned int length)
39279cc3 7808{
5fd02043 7809 struct inode *inode = page->mapping->host;
d1310b2e 7810 struct extent_io_tree *tree;
e6dcd2dc 7811 struct btrfs_ordered_extent *ordered;
2ac55d41 7812 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 7813 u64 page_start = page_offset(page);
7814 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
131e404a 7815 int inode_evicting = inode->i_state & I_FREEING;
39279cc3 7816
8b62b72b
CM		 7817 /*
7818 * we have the page locked, so new writeback can't start,
7819 * and the dirty bit won't be cleared while we are here.
7820 *
7821 * Wait for IO on this page so that we can safely clear
7822 * the PagePrivate2 bit and do ordered accounting
7823 */
e6dcd2dc 7824 wait_on_page_writeback(page);
8b62b72b 7825
5fd02043 7826 tree = &BTRFS_I(inode)->io_tree;
e6dcd2dc
CM		 7827 if (offset) {
7828 btrfs_releasepage(page, GFP_NOFS);
7829 return;
7830 }
131e404a
FDBM		 7831
7832 if (!inode_evicting)
7833 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7834 ordered = btrfs_lookup_ordered_extent(inode, page_start);
e6dcd2dc 7835 if (ordered) {
eb84ae03
CM		 7836 /*
7837 * IO on this page will never be started, so we need
7838 * to account for any ordered extents now
7839 */
131e404a
FDBM		 7840 if (!inode_evicting)
7841 clear_extent_bit(tree, page_start, page_end,
7842 EXTENT_DIRTY | EXTENT_DELALLOC |
7843 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7844 EXTENT_DEFRAG, 1, 0, &cached_state,
7845 GFP_NOFS);
8b62b72b
CM		 7846 /*
7847 * whoever cleared the private bit is responsible
7848 * for the finish_ordered_io
7849 */
77cef2ec
JB		 7850 if (TestClearPagePrivate2(page)) {
7851 struct btrfs_ordered_inode_tree *tree;
7852 u64 new_len;
7853
7854 tree = &BTRFS_I(inode)->ordered_tree;
7855
7856 spin_lock_irq(&tree->lock);
7857 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
7858 new_len = page_start - ordered->file_offset;
7859 if (new_len < ordered->truncated_len)
7860 ordered->truncated_len = new_len;
7861 spin_unlock_irq(&tree->lock);
7862
7863 if (btrfs_dec_test_ordered_pending(inode, &ordered,
7864 page_start,
7865 PAGE_CACHE_SIZE, 1))
7866 btrfs_finish_ordered_io(ordered);
8b62b72b 7867 }
e6dcd2dc 7868 btrfs_put_ordered_extent(ordered);
131e404a
FDBM		 7869 if (!inode_evicting) {
7870 cached_state = NULL;
7871 lock_extent_bits(tree, page_start, page_end, 0,
7872 &cached_state);
7873 }
7874 }
7875
7876 if (!inode_evicting) {
7877 clear_extent_bit(tree, page_start, page_end,
7878 EXTENT_LOCKED | EXTENT_DIRTY |
7879 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
7880 EXTENT_DEFRAG, 1, 1,
7881 &cached_state, GFP_NOFS);
7882
7883 __btrfs_releasepage(page, GFP_NOFS);
e6dcd2dc 7884 }
e6dcd2dc 7885
4a096752 7886 ClearPageChecked(page);
9ad6b7bc 7887 if (PagePrivate(page)) {
9ad6b7bc
CM		 7888 ClearPagePrivate(page);
7889 set_page_private(page, 0);
7890 page_cache_release(page);
7891 }
39279cc3
CM		 7892}
7893
9ebefb18
CM		 7894/*
7895 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7896 * called from a page fault handler when a page is first dirtied. Hence we must
7897 * be careful to check for EOF conditions here. We set the page up correctly
7898 * for a written page which means we get ENOSPC checking when writing into
7899 * holes and correct delalloc and unwritten extent mapping on filesystems that
7900 * support these features.
7901 *
7902 * We are not allowed to take the i_mutex here so we have to play games to
7903 * protect against truncate races as the page could now be beyond EOF. Because
7904 * vmtruncate() writes the inode size before removing pages, once we have the
7905 * page lock we can determine safely if the page is beyond EOF. If it is not
7906 * beyond EOF, then the page is guaranteed safe against truncation until we
7907 * unlock the page.
7908 */
c2ec175c 7909int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
9ebefb18 7910{
c2ec175c 7911 struct page *page = vmf->page;
496ad9aa 7912 struct inode *inode = file_inode(vma->vm_file);
1832a6d5 7913 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 7914 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7915 struct btrfs_ordered_extent *ordered;
2ac55d41 7916 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 7917 char *kaddr;
7918 unsigned long zero_start;
9ebefb18 7919 loff_t size;
1832a6d5 7920 int ret;
9998eb70 7921 int reserved = 0;
a52d9a80 7922 u64 page_start;
e6dcd2dc 7923 u64 page_end;
9ebefb18 7924
b2b5ef5c 7925 sb_start_pagefault(inode->i_sb);
0ca1f7ce 7926 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
9998eb70 7927 if (!ret) {
e41f941a 7928 ret = file_update_time(vma->vm_file);
9998eb70
CM		 7929 reserved = 1;
7930 }
56a76f82
NP		 7931 if (ret) {
7932 if (ret == -ENOMEM)
7933 ret = VM_FAULT_OOM;
7934 else /* -ENOSPC, -EIO, etc */
7935 ret = VM_FAULT_SIGBUS;
9998eb70
CM		 7936 if (reserved)
7937 goto out;
7938 goto out_noreserve;
56a76f82 7939 }
1832a6d5 7940
56a76f82 7941 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
e6dcd2dc 7942again:
9ebefb18 7943 lock_page(page);
9ebefb18 7944 size = i_size_read(inode);
e6dcd2dc
CM		 7945 page_start = page_offset(page);
7946 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 7947
9ebefb18 7948 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 7949 (page_start >= size)) {
9ebefb18
CM		 7950 /* page got truncated out from underneath us */
7951 goto out_unlock;
7952 }
e6dcd2dc
CM		 7953 wait_on_page_writeback(page);
7954
d0082371 7955 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 7956 set_page_extent_mapped(page);
7957
eb84ae03
CM		 7958 /*
7959 * we can't set the delalloc bits if there are pending ordered
7960 * extents. Drop our locks and wait for them to finish
7961 */
e6dcd2dc
CM		 7962 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7963 if (ordered) {
2ac55d41
JB		 7964 unlock_extent_cached(io_tree, page_start, page_end,
7965 &cached_state, GFP_NOFS);
e6dcd2dc 7966 unlock_page(page);
eb84ae03 7967 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 7968 btrfs_put_ordered_extent(ordered);
7969 goto again;
7970 }
7971
fbf19087
JB		 7972 /*
7973 * XXX - page_mkwrite gets called every time the page is dirtied, even
7974 * if it was already dirty, so for space accounting reasons we need to
7975 * clear any delalloc bits for the range we are fixing to save. There
7976 * is probably a better way to do this, but for now keep consistent with
7977 * prepare_pages in the normal write path.
7978 */
2ac55d41 7979 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 7980 EXTENT_DIRTY | EXTENT_DELALLOC |
7981 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 7982 0, 0, &cached_state, GFP_NOFS);
fbf19087 7983
2ac55d41
JB		 7984 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7985 &cached_state);
9ed74f2d 7986 if (ret) {
2ac55d41
JB		 7987 unlock_extent_cached(io_tree, page_start, page_end,
7988 &cached_state, GFP_NOFS);
9ed74f2d
JB		 7989 ret = VM_FAULT_SIGBUS;
7990 goto out_unlock;
7991 }
e6dcd2dc 7992 ret = 0;
9ebefb18
CM		 7993
7994 /* page is wholly or partially inside EOF */
a52d9a80 7995 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 7996 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 7997 else
e6dcd2dc 7998 zero_start = PAGE_CACHE_SIZE;
9ebefb18 7999
e6dcd2dc
CM		 8000 if (zero_start != PAGE_CACHE_SIZE) {
8001 kaddr = kmap(page);
8002 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
8003 flush_dcache_page(page);
8004 kunmap(page);
8005 }
247e743c 8006 ClearPageChecked(page);
e6dcd2dc 8007 set_page_dirty(page);
50a9b214 8008 SetPageUptodate(page);
5a3f23d5 8009
257c62e1
CM		 8010 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8011 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
46d8bc34 8012 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
257c62e1 8013
2ac55d41 8014 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9ebefb18
CM		 8015
8016out_unlock:
b2b5ef5c
JK		 8017 if (!ret) {
8018 sb_end_pagefault(inode->i_sb);
50a9b214 8019 return VM_FAULT_LOCKED;
b2b5ef5c 8020 }
9ebefb18 8021 unlock_page(page);
1832a6d5 8022out:
ec39e180 8023 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
9998eb70 8024out_noreserve:
b2b5ef5c 8025 sb_end_pagefault(inode->i_sb);
9ebefb18
CM		 8026 return ret;
8027}
8028
a41ad394 8029static int btrfs_truncate(struct inode *inode)
39279cc3
CM		 8030{
8031 struct btrfs_root *root = BTRFS_I(inode)->root;
fcb80c2a 8032 struct btrfs_block_rsv *rsv;
a71754fc 8033 int ret = 0;
3893e33b 8034 int err = 0;
39279cc3 8035 struct btrfs_trans_handle *trans;
dbe674a9 8036 u64 mask = root->sectorsize - 1;
07127184 8037 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3 8038
0ef8b726
JB		 8039 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8040 (u64)-1);
8041 if (ret)
8042 return ret;
39279cc3 8043
fcb80c2a
JB		 8044 /*
8045 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8046 * 3 things going on here
8047 *
8048 * 1) We need to reserve space for our orphan item and the space to
8049 * delete our orphan item. Lord knows we don't want to have a dangling
8050 * orphan item because we didn't reserve space to remove it.
8051 *
8052 * 2) We need to reserve space to update our inode.
8053 *
8054 * 3) We need to have something to cache all the space that is going to
8055 * be free'd up by the truncate operation, but also have some slack
8056 * space reserved in case it uses space during the truncate (thank you
8057 * very much snapshotting).
8058 *
8059 * And we need these to all be seperate. The fact is we can use alot of
8060 * space doing the truncate, and we have no earthly idea how much space
8061 * we will use, so we need the truncate reservation to be seperate so it
8062 * doesn't end up using space reserved for updating the inode or
8063 * removing the orphan item. We also need to be able to stop the
8064 * transaction and start a new one, which means we need to be able to
8065 * update the inode several times, and we have no idea of knowing how
8066 * many times that will be, so we can't just reserve 1 item for the
8067 * entirety of the opration, so that has to be done seperately as well.
8068 * Then there is the orphan item, which does indeed need to be held on
8069 * to for the whole operation, and we need nobody to touch this reserved
8070 * space except the orphan code.
8071 *
8072 * So that leaves us with
8073 *
8074 * 1) root->orphan_block_rsv - for the orphan deletion.
8075 * 2) rsv - for the truncate reservation, which we will steal from the
8076 * transaction reservation.
8077 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8078 * updating the inode.
8079 */
66d8f3dd 8080 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
fcb80c2a
JB		 8081 if (!rsv)
8082 return -ENOMEM;
4a338542 8083 rsv->size = min_size;
ca7e70f5 8084 rsv->failfast = 1;
f0cd846e 8085
907cbceb 8086 /*
07127184 8087 * 1 for the truncate slack space
907cbceb
JB		 8088 * 1 for updating the inode.
8089 */
f3fe820c 8090 trans = btrfs_start_transaction(root, 2);
fcb80c2a
JB		 8091 if (IS_ERR(trans)) {
8092 err = PTR_ERR(trans);
8093 goto out;
8094 }
f0cd846e 8095
907cbceb
JB		 8096 /* Migrate the slack space for the truncate to our reserve */
8097 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8098 min_size);
fcb80c2a 8099 BUG_ON(ret);
f0cd846e 8100
5dc562c5
JB		 8101 /*
8102 * So if we truncate and then write and fsync we normally would just
8103 * write the extents that changed, which is a problem if we need to
8104 * first truncate that entire inode. So set this flag so we write out
8105 * all of the extents in the inode to the sync log so we're completely
8106 * safe.
8107 */
8108 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
ca7e70f5 8109 trans->block_rsv = rsv;
907cbceb 8110
8082510e
YZ		 8111 while (1) {
8112 ret = btrfs_truncate_inode_items(trans, root, inode,
8113 inode->i_size,
8114 BTRFS_EXTENT_DATA_KEY);
ca7e70f5 8115 if (ret != -ENOSPC) {
3893e33b 8116 err = ret;
8082510e 8117 break;
3893e33b 8118 }
39279cc3 8119
fcb80c2a 8120 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e 8121 ret = btrfs_update_inode(trans, root, inode);
3893e33b
JB		 8122 if (ret) {
8123 err = ret;
8124 break;
8125 }
ca7e70f5 8126
8082510e 8127 btrfs_end_transaction(trans, root);
b53d3f5d 8128 btrfs_btree_balance_dirty(root);
ca7e70f5
JB		 8129
8130 trans = btrfs_start_transaction(root, 2);
8131 if (IS_ERR(trans)) {
8132 ret = err = PTR_ERR(trans);
8133 trans = NULL;
8134 break;
8135 }
8136
8137 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
8138 rsv, min_size);
8139 BUG_ON(ret); /* shouldn't happen */
8140 trans->block_rsv = rsv;
8082510e
YZ		 8141 }
8142
8143 if (ret == 0 && inode->i_nlink > 0) {
fcb80c2a 8144 trans->block_rsv = root->orphan_block_rsv;
8082510e 8145 ret = btrfs_orphan_del(trans, inode);
3893e33b
JB		 8146 if (ret)
8147 err = ret;
8082510e
YZ		 8148 }
8149
917c16b2
CM		 8150 if (trans) {
8151 trans->block_rsv = &root->fs_info->trans_block_rsv;
8152 ret = btrfs_update_inode(trans, root, inode);
8153 if (ret && !err)
8154 err = ret;
7b128766 8155
7ad85bb7 8156 ret = btrfs_end_transaction(trans, root);
b53d3f5d 8157 btrfs_btree_balance_dirty(root);
917c16b2 8158 }
fcb80c2a
JB		 8159
8160out:
8161 btrfs_free_block_rsv(root, rsv);
8162
3893e33b
JB		 8163 if (ret && !err)
8164 err = ret;
a41ad394 8165
3893e33b 8166 return err;
39279cc3
CM		 8167}
8168
d352ac68
CM		 8169/*
8170 * create a new subvolume directory/inode (helper for the ioctl).
8171 */
d2fb3437 8172int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
63541927
FDBM		 8173 struct btrfs_root *new_root,
8174 struct btrfs_root *parent_root,
8175 u64 new_dirid)
39279cc3 8176{
39279cc3 8177 struct inode *inode;
76dda93c 8178 int err;
00e4e6b3 8179 u64 index = 0;
39279cc3 8180
12fc9d09
FA		 8181 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8182 new_dirid, new_dirid,
8183 S_IFDIR | (~current_umask() & S_IRWXUGO),
8184 &index);
54aa1f4d 8185 if (IS_ERR(inode))
f46b5a66 8186 return PTR_ERR(inode);
39279cc3
CM		 8187 inode->i_op = &btrfs_dir_inode_operations;
8188 inode->i_fop = &btrfs_dir_file_operations;
8189
bfe86848 8190 set_nlink(inode, 1);
dbe674a9 8191 btrfs_i_size_write(inode, 0);
b0d5d10f 8192 unlock_new_inode(inode);
3b96362c 8193
63541927
FDBM		 8194 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8195 if (err)
8196 btrfs_err(new_root->fs_info,
351fd353 8197 "error inheriting subvolume %llu properties: %d",
63541927
FDBM		 8198 new_root->root_key.objectid, err);
8199
76dda93c 8200 err = btrfs_update_inode(trans, new_root, inode);
cb8e7090 8201
76dda93c 8202 iput(inode);
ce598979 8203 return err;
39279cc3
CM		 8204}
8205
39279cc3
CM		 8206struct inode *btrfs_alloc_inode(struct super_block *sb)
8207{
8208 struct btrfs_inode *ei;
2ead6ae7 8209 struct inode *inode;
39279cc3
CM		 8210
8211 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
8212 if (!ei)
8213 return NULL;
2ead6ae7
YZ		 8214
8215 ei->root = NULL;
2ead6ae7 8216 ei->generation = 0;
15ee9bc7 8217 ei->last_trans = 0;
257c62e1 8218 ei->last_sub_trans = 0;
e02119d5 8219 ei->logged_trans = 0;
2ead6ae7 8220 ei->delalloc_bytes = 0;
47059d93 8221 ei->defrag_bytes = 0;
2ead6ae7
YZ		 8222 ei->disk_i_size = 0;
8223 ei->flags = 0;
7709cde3 8224 ei->csum_bytes = 0;
2ead6ae7 8225 ei->index_cnt = (u64)-1;
67de1176 8226 ei->dir_index = 0;
2ead6ae7 8227 ei->last_unlink_trans = 0;
46d8bc34 8228 ei->last_log_commit = 0;
2ead6ae7 8229
9e0baf60
JB		 8230 spin_lock_init(&ei->lock);
8231 ei->outstanding_extents = 0;
8232 ei->reserved_extents = 0;
2ead6ae7 8233
72ac3c0d 8234 ei->runtime_flags = 0;
261507a0 8235 ei->force_compress = BTRFS_COMPRESS_NONE;
2ead6ae7 8236
16cdcec7
MX		 8237 ei->delayed_node = NULL;
8238
2ead6ae7 8239 inode = &ei->vfs_inode;
a8067e02 8240 extent_map_tree_init(&ei->extent_tree);
f993c883
DS		 8241 extent_io_tree_init(&ei->io_tree, &inode->i_data);
8242 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
0b32f4bb
JB		 8243 ei->io_tree.track_uptodate = 1;
8244 ei->io_failure_tree.track_uptodate = 1;
b812ce28 8245 atomic_set(&ei->sync_writers, 0);
2ead6ae7 8246 mutex_init(&ei->log_mutex);
f248679e 8247 mutex_init(&ei->delalloc_mutex);
e6dcd2dc 8248 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
2ead6ae7 8249 INIT_LIST_HEAD(&ei->delalloc_inodes);
2ead6ae7
YZ		 8250 RB_CLEAR_NODE(&ei->rb_node);
8251
8252 return inode;
39279cc3
CM		 8253}
8254
aaedb55b
JB		 8255#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8256void btrfs_test_destroy_inode(struct inode *inode)
8257{
8258 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8259 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8260}
8261#endif
8262
fa0d7e3d
NP		 8263static void btrfs_i_callback(struct rcu_head *head)
8264{
8265 struct inode *inode = container_of(head, struct inode, i_rcu);
fa0d7e3d
NP		 8266 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8267}
8268
39279cc3
CM		 8269void btrfs_destroy_inode(struct inode *inode)
8270{
e6dcd2dc 8271 struct btrfs_ordered_extent *ordered;
5a3f23d5
CM		 8272 struct btrfs_root *root = BTRFS_I(inode)->root;
8273
b3d9b7a3 8274 WARN_ON(!hlist_empty(&inode->i_dentry));
39279cc3 8275 WARN_ON(inode->i_data.nrpages);
9e0baf60
JB		 8276 WARN_ON(BTRFS_I(inode)->outstanding_extents);
8277 WARN_ON(BTRFS_I(inode)->reserved_extents);
7709cde3
JB		 8278 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8279 WARN_ON(BTRFS_I(inode)->csum_bytes);
47059d93 8280 WARN_ON(BTRFS_I(inode)->defrag_bytes);
39279cc3 8281
a6dbd429
JB		 8282 /*
8283 * This can happen where we create an inode, but somebody else also
8284 * created the same inode and we need to destroy the one we already
8285 * created.
8286 */
8287 if (!root)
8288 goto free;
8289
8a35d95f
JB		 8290 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8291 &BTRFS_I(inode)->runtime_flags)) {
c2cf52eb 8292 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
c1c9ff7c 8293 btrfs_ino(inode));
8a35d95f 8294 atomic_dec(&root->orphan_inodes);
7b128766 8295 }
7b128766 8296
d397712b 8297 while (1) {
e6dcd2dc
CM		 8298 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8299 if (!ordered)
8300 break;
8301 else {
c2cf52eb 8302 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
c1c9ff7c 8303 ordered->file_offset, ordered->len);
e6dcd2dc
CM		 8304 btrfs_remove_ordered_extent(inode, ordered);
8305 btrfs_put_ordered_extent(ordered);
8306 btrfs_put_ordered_extent(ordered);
8307 }
8308 }
5d4f98a2 8309 inode_tree_del(inode);
5b21f2ed 8310 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
a6dbd429 8311free:
fa0d7e3d 8312 call_rcu(&inode->i_rcu, btrfs_i_callback);
39279cc3
CM		 8313}
8314
45321ac5 8315int btrfs_drop_inode(struct inode *inode)
76dda93c
YZ		 8316{
8317 struct btrfs_root *root = BTRFS_I(inode)->root;
45321ac5 8318
6379ef9f
NA		 8319 if (root == NULL)
8320 return 1;
8321
fa6ac876 8322 /* the snap/subvol tree is on deleting */
69e9c6c6 8323 if (btrfs_root_refs(&root->root_item) == 0)
45321ac5 8324 return 1;
76dda93c 8325 else
45321ac5 8326 return generic_drop_inode(inode);
76dda93c
YZ		 8327}
8328
0ee0fda0 8329static void init_once(void *foo)
39279cc3
CM		 8330{
8331 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8332
8333 inode_init_once(&ei->vfs_inode);
8334}
8335
8336void btrfs_destroy_cachep(void)
8337{
8c0a8537
KS		 8338 /*
8339 * Make sure all delayed rcu free inodes are flushed before we
8340 * destroy cache.
8341 */
8342 rcu_barrier();
39279cc3
CM		 8343 if (btrfs_inode_cachep)
8344 kmem_cache_destroy(btrfs_inode_cachep);
8345 if (btrfs_trans_handle_cachep)
8346 kmem_cache_destroy(btrfs_trans_handle_cachep);
8347 if (btrfs_transaction_cachep)
8348 kmem_cache_destroy(btrfs_transaction_cachep);
39279cc3
CM		 8349 if (btrfs_path_cachep)
8350 kmem_cache_destroy(btrfs_path_cachep);
dc89e982
JB		 8351 if (btrfs_free_space_cachep)
8352 kmem_cache_destroy(btrfs_free_space_cachep);
8ccf6f19
MX		 8353 if (btrfs_delalloc_work_cachep)
8354 kmem_cache_destroy(btrfs_delalloc_work_cachep);
39279cc3
CM		 8355}
8356
8357int btrfs_init_cachep(void)
8358{
837e1972 8359 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9601e3f6
CH		 8360 sizeof(struct btrfs_inode), 0,
8361 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
39279cc3
CM		 8362 if (!btrfs_inode_cachep)
8363 goto fail;
9601e3f6 8364
837e1972 8365 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9601e3f6
CH		 8366 sizeof(struct btrfs_trans_handle), 0,
8367 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 8368 if (!btrfs_trans_handle_cachep)
8369 goto fail;
9601e3f6 8370
837e1972 8371 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9601e3f6
CH		 8372 sizeof(struct btrfs_transaction), 0,
8373 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 8374 if (!btrfs_transaction_cachep)
8375 goto fail;
9601e3f6 8376
837e1972 8377 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9601e3f6
CH		 8378 sizeof(struct btrfs_path), 0,
8379 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 8380 if (!btrfs_path_cachep)
8381 goto fail;
9601e3f6 8382
837e1972 8383 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
dc89e982
JB		 8384 sizeof(struct btrfs_free_space), 0,
8385 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8386 if (!btrfs_free_space_cachep)
8387 goto fail;
8388
8ccf6f19
MX		 8389 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8390 sizeof(struct btrfs_delalloc_work), 0,
8391 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8392 NULL);
8393 if (!btrfs_delalloc_work_cachep)
8394 goto fail;
8395
39279cc3
CM		 8396 return 0;
8397fail:
8398 btrfs_destroy_cachep();
8399 return -ENOMEM;
8400}
8401
8402static int btrfs_getattr(struct vfsmount *mnt,
8403 struct dentry *dentry, struct kstat *stat)
8404{
df0af1a5 8405 u64 delalloc_bytes;
39279cc3 8406 struct inode *inode = dentry->d_inode;
fadc0d8b
DS		 8407 u32 blocksize = inode->i_sb->s_blocksize;
8408
39279cc3 8409 generic_fillattr(inode, stat);
0ee5dc67 8410 stat->dev = BTRFS_I(inode)->root->anon_dev;
d6667462 8411 stat->blksize = PAGE_CACHE_SIZE;
df0af1a5
MX		 8412
8413 spin_lock(&BTRFS_I(inode)->lock);
8414 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8415 spin_unlock(&BTRFS_I(inode)->lock);
fadc0d8b 8416 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
df0af1a5 8417 ALIGN(delalloc_bytes, blocksize)) >> 9;
39279cc3
CM		 8418 return 0;
8419}
8420
d397712b
CM		 8421static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8422 struct inode *new_dir, struct dentry *new_dentry)
39279cc3
CM		 8423{
8424 struct btrfs_trans_handle *trans;
8425 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4df27c4d 8426 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
39279cc3
CM		 8427 struct inode *new_inode = new_dentry->d_inode;
8428 struct inode *old_inode = old_dentry->d_inode;
8429 struct timespec ctime = CURRENT_TIME;
00e4e6b3 8430 u64 index = 0;
4df27c4d 8431 u64 root_objectid;
39279cc3 8432 int ret;
33345d01 8433 u64 old_ino = btrfs_ino(old_inode);
39279cc3 8434
33345d01 8435 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
f679a840
YZ		 8436 return -EPERM;
8437
4df27c4d 8438 /* we only allow rename subvolume link between subvolumes */
33345d01 8439 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
3394e160
CM		 8440 return -EXDEV;
8441
33345d01
LZ		 8442 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8443 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
39279cc3 8444 return -ENOTEMPTY;
5f39d397 8445
4df27c4d
YZ		 8446 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8447 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8448 return -ENOTEMPTY;
9c52057c
CM		 8449
8450
8451 /* check for collisions, even if the name isn't there */
4871c158 8452 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9c52057c
CM		 8453 new_dentry->d_name.name,
8454 new_dentry->d_name.len);
8455
8456 if (ret) {
8457 if (ret == -EEXIST) {
8458 /* we shouldn't get
8459 * eexist without a new_inode */
fae7f21c 8460 if (WARN_ON(!new_inode)) {
9c52057c
CM		 8461 return ret;
8462 }
8463 } else {
8464 /* maybe -EOVERFLOW */
8465 return ret;
8466 }
8467 }
8468 ret = 0;
8469
5a3f23d5 8470 /*
8d875f95
CM		 8471 * we're using rename to replace one file with another. Start IO on it
8472 * now so we don't add too much work to the end of the transaction
5a3f23d5 8473 */
8d875f95 8474 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
5a3f23d5
CM		 8475 filemap_flush(old_inode->i_mapping);
8476
76dda93c 8477 /* close the racy window with snapshot create/destroy ioctl */
33345d01 8478 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 8479 down_read(&root->fs_info->subvol_sem);
a22285a6
YZ		 8480 /*
8481 * We want to reserve the absolute worst case amount of items. So if
8482 * both inodes are subvols and we need to unlink them then that would
8483 * require 4 item modifications, but if they are both normal inodes it
8484 * would require 5 item modifications, so we'll assume their normal
8485 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8486 * should cover the worst case number of items we'll modify.
8487 */
6e137ed3 8488 trans = btrfs_start_transaction(root, 11);
b44c59a8
JL		 8489 if (IS_ERR(trans)) {
8490 ret = PTR_ERR(trans);
8491 goto out_notrans;
8492 }
76dda93c 8493
4df27c4d
YZ		 8494 if (dest != root)
8495 btrfs_record_root_in_trans(trans, dest);
5f39d397 8496
a5719521
YZ		 8497 ret = btrfs_set_inode_index(new_dir, &index);
8498 if (ret)
8499 goto out_fail;
5a3f23d5 8500
67de1176 8501 BTRFS_I(old_inode)->dir_index = 0ULL;
33345d01 8502 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d 8503 /* force full log commit if subvolume involved. */
995946dd 8504 btrfs_set_log_full_commit(root->fs_info, trans);
4df27c4d 8505 } else {
a5719521
YZ		 8506 ret = btrfs_insert_inode_ref(trans, dest,
8507 new_dentry->d_name.name,
8508 new_dentry->d_name.len,
33345d01
LZ		 8509 old_ino,
8510 btrfs_ino(new_dir), index);
a5719521
YZ		 8511 if (ret)
8512 goto out_fail;
4df27c4d
YZ		 8513 /*
8514 * this is an ugly little race, but the rename is required
8515 * to make sure that if we crash, the inode is either at the
8516 * old name or the new one. pinning the log transaction lets
8517 * us make sure we don't allow a log commit to come in after
8518 * we unlink the name but before we add the new name back in.
8519 */
8520 btrfs_pin_log_trans(root);
8521 }
5a3f23d5 8522
0c4d2d95
JB		 8523 inode_inc_iversion(old_dir);
8524 inode_inc_iversion(new_dir);
8525 inode_inc_iversion(old_inode);
39279cc3
CM		 8526 old_dir->i_ctime = old_dir->i_mtime = ctime;
8527 new_dir->i_ctime = new_dir->i_mtime = ctime;
8528 old_inode->i_ctime = ctime;
5f39d397 8529
12fcfd22
CM		 8530 if (old_dentry->d_parent != new_dentry->d_parent)
8531 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8532
33345d01 8533 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 8534 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8535 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8536 old_dentry->d_name.name,
8537 old_dentry->d_name.len);
8538 } else {
92986796
AV		 8539 ret = __btrfs_unlink_inode(trans, root, old_dir,
8540 old_dentry->d_inode,
8541 old_dentry->d_name.name,
8542 old_dentry->d_name.len);
8543 if (!ret)
8544 ret = btrfs_update_inode(trans, root, old_inode);
4df27c4d 8545 }
79787eaa
JM		 8546 if (ret) {
8547 btrfs_abort_transaction(trans, root, ret);
8548 goto out_fail;
8549 }
39279cc3
CM		 8550
8551 if (new_inode) {
0c4d2d95 8552 inode_inc_iversion(new_inode);
39279cc3 8553 new_inode->i_ctime = CURRENT_TIME;
33345d01 8554 if (unlikely(btrfs_ino(new_inode) ==
4df27c4d
YZ		 8555 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8556 root_objectid = BTRFS_I(new_inode)->location.objectid;
8557 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8558 root_objectid,
8559 new_dentry->d_name.name,
8560 new_dentry->d_name.len);
8561 BUG_ON(new_inode->i_nlink == 0);
8562 } else {
8563 ret = btrfs_unlink_inode(trans, dest, new_dir,
8564 new_dentry->d_inode,
8565 new_dentry->d_name.name,
8566 new_dentry->d_name.len);
8567 }
4ef31a45 8568 if (!ret && new_inode->i_nlink == 0)
e02119d5 8569 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
79787eaa
JM		 8570 if (ret) {
8571 btrfs_abort_transaction(trans, root, ret);
8572 goto out_fail;
8573 }
39279cc3 8574 }
aec7477b 8575
4df27c4d
YZ		 8576 ret = btrfs_add_link(trans, new_dir, old_inode,
8577 new_dentry->d_name.name,
a5719521 8578 new_dentry->d_name.len, 0, index);
79787eaa
JM		 8579 if (ret) {
8580 btrfs_abort_transaction(trans, root, ret);
8581 goto out_fail;
8582 }
39279cc3 8583
67de1176
MX		 8584 if (old_inode->i_nlink == 1)
8585 BTRFS_I(old_inode)->dir_index = index;
8586
33345d01 8587 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
10d9f309 8588 struct dentry *parent = new_dentry->d_parent;
6a912213 8589 btrfs_log_new_name(trans, old_inode, old_dir, parent);
4df27c4d
YZ		 8590 btrfs_end_log_trans(root);
8591 }
39279cc3 8592out_fail:
7ad85bb7 8593 btrfs_end_transaction(trans, root);
b44c59a8 8594out_notrans:
33345d01 8595 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 8596 up_read(&root->fs_info->subvol_sem);
9ed74f2d 8597
39279cc3
CM		 8598 return ret;
8599}
8600
80ace85c
MS		 8601static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
8602 struct inode *new_dir, struct dentry *new_dentry,
8603 unsigned int flags)
8604{
8605 if (flags & ~RENAME_NOREPLACE)
8606 return -EINVAL;
8607
8608 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
8609}
8610
8ccf6f19
MX		 8611static void btrfs_run_delalloc_work(struct btrfs_work *work)
8612{
8613 struct btrfs_delalloc_work *delalloc_work;
9f23e289 8614 struct inode *inode;
8ccf6f19
MX		 8615
8616 delalloc_work = container_of(work, struct btrfs_delalloc_work,
8617 work);
9f23e289
JB		 8618 inode = delalloc_work->inode;
8619 if (delalloc_work->wait) {
8620 btrfs_wait_ordered_range(inode, 0, (u64)-1);
8621 } else {
8622 filemap_flush(inode->i_mapping);
8623 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8624 &BTRFS_I(inode)->runtime_flags))
8625 filemap_flush(inode->i_mapping);
8626 }
8ccf6f19
MX		 8627
8628 if (delalloc_work->delay_iput)
9f23e289 8629 btrfs_add_delayed_iput(inode);
8ccf6f19 8630 else
9f23e289 8631 iput(inode);
8ccf6f19
MX		 8632 complete(&delalloc_work->completion);
8633}
8634
8635struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8636 int wait, int delay_iput)
8637{
8638 struct btrfs_delalloc_work *work;
8639
8640 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8641 if (!work)
8642 return NULL;
8643
8644 init_completion(&work->completion);
8645 INIT_LIST_HEAD(&work->list);
8646 work->inode = inode;
8647 work->wait = wait;
8648 work->delay_iput = delay_iput;
9e0af237
LB		 8649 WARN_ON_ONCE(!inode);
8650 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
8651 btrfs_run_delalloc_work, NULL, NULL);
8ccf6f19
MX		 8652
8653 return work;
8654}
8655
8656void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8657{
8658 wait_for_completion(&work->completion);
8659 kmem_cache_free(btrfs_delalloc_work_cachep, work);
8660}
8661
d352ac68
CM		 8662/*
8663 * some fairly slow code that needs optimization. This walks the list
8664 * of all the inodes with pending delalloc and forces them to disk.
8665 */
6c255e67
MX		 8666static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
8667 int nr)
ea8c2819 8668{
ea8c2819 8669 struct btrfs_inode *binode;
5b21f2ed 8670 struct inode *inode;
8ccf6f19
MX		 8671 struct btrfs_delalloc_work *work, *next;
8672 struct list_head works;
1eafa6c7 8673 struct list_head splice;
8ccf6f19 8674 int ret = 0;
ea8c2819 8675
8ccf6f19 8676 INIT_LIST_HEAD(&works);
1eafa6c7 8677 INIT_LIST_HEAD(&splice);
63607cc8 8678
573bfb72 8679 mutex_lock(&root->delalloc_mutex);
eb73c1b7
MX		 8680 spin_lock(&root->delalloc_lock);
8681 list_splice_init(&root->delalloc_inodes, &splice);
1eafa6c7
MX		 8682 while (!list_empty(&splice)) {
8683 binode = list_entry(splice.next, struct btrfs_inode,
ea8c2819 8684 delalloc_inodes);
1eafa6c7 8685
eb73c1b7
MX		 8686 list_move_tail(&binode->delalloc_inodes,
8687 &root->delalloc_inodes);
5b21f2ed 8688 inode = igrab(&binode->vfs_inode);
df0af1a5 8689 if (!inode) {
eb73c1b7 8690 cond_resched_lock(&root->delalloc_lock);
1eafa6c7 8691 continue;
df0af1a5 8692 }
eb73c1b7 8693 spin_unlock(&root->delalloc_lock);
1eafa6c7
MX		 8694
8695 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8696 if (unlikely(!work)) {
f4ab9ea7
JB		 8697 if (delay_iput)
8698 btrfs_add_delayed_iput(inode);
8699 else
8700 iput(inode);
1eafa6c7 8701 ret = -ENOMEM;
a1ecaabb 8702 goto out;
5b21f2ed 8703 }
1eafa6c7 8704 list_add_tail(&work->list, &works);
a44903ab
QW		 8705 btrfs_queue_work(root->fs_info->flush_workers,
8706 &work->work);
6c255e67
MX		 8707 ret++;
8708 if (nr != -1 && ret >= nr)
a1ecaabb 8709 goto out;
5b21f2ed 8710 cond_resched();
eb73c1b7 8711 spin_lock(&root->delalloc_lock);
ea8c2819 8712 }
eb73c1b7 8713 spin_unlock(&root->delalloc_lock);
8c8bee1d 8714
a1ecaabb 8715out:
eb73c1b7
MX		 8716 list_for_each_entry_safe(work, next, &works, list) {
8717 list_del_init(&work->list);
8718 btrfs_wait_and_free_delalloc_work(work);
8719 }
8720
8721 if (!list_empty_careful(&splice)) {
8722 spin_lock(&root->delalloc_lock);
8723 list_splice_tail(&splice, &root->delalloc_inodes);
8724 spin_unlock(&root->delalloc_lock);
8725 }
573bfb72 8726 mutex_unlock(&root->delalloc_mutex);
eb73c1b7
MX		 8727 return ret;
8728}
1eafa6c7 8729
eb73c1b7
MX		 8730int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8731{
8732 int ret;
1eafa6c7 8733
2c21b4d7 8734 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
eb73c1b7
MX		 8735 return -EROFS;
8736
6c255e67
MX		 8737 ret = __start_delalloc_inodes(root, delay_iput, -1);
8738 if (ret > 0)
8739 ret = 0;
eb73c1b7
MX		 8740 /*
8741 * the filemap_flush will queue IO into the worker threads, but
8c8bee1d
CM		 8742 * we have to make sure the IO is actually started and that
8743 * ordered extents get created before we return
8744 */
8745 atomic_inc(&root->fs_info->async_submit_draining);
d397712b 8746 while (atomic_read(&root->fs_info->nr_async_submits) ||
771ed689 8747 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 8748 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 8749 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8750 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 8751 }
8752 atomic_dec(&root->fs_info->async_submit_draining);
eb73c1b7
MX		 8753 return ret;
8754}
8755
6c255e67
MX		 8756int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
8757 int nr)
eb73c1b7
MX		 8758{
8759 struct btrfs_root *root;
8760 struct list_head splice;
8761 int ret;
8762
2c21b4d7 8763 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
eb73c1b7
MX		 8764 return -EROFS;
8765
8766 INIT_LIST_HEAD(&splice);
8767
573bfb72 8768 mutex_lock(&fs_info->delalloc_root_mutex);
eb73c1b7
MX		 8769 spin_lock(&fs_info->delalloc_root_lock);
8770 list_splice_init(&fs_info->delalloc_roots, &splice);
6c255e67 8771 while (!list_empty(&splice) && nr) {
eb73c1b7
MX		 8772 root = list_first_entry(&splice, struct btrfs_root,
8773 delalloc_root);
8774 root = btrfs_grab_fs_root(root);
8775 BUG_ON(!root);
8776 list_move_tail(&root->delalloc_root,
8777 &fs_info->delalloc_roots);
8778 spin_unlock(&fs_info->delalloc_root_lock);
8779
6c255e67 8780 ret = __start_delalloc_inodes(root, delay_iput, nr);
eb73c1b7 8781 btrfs_put_fs_root(root);
6c255e67 8782 if (ret < 0)
eb73c1b7
MX		 8783 goto out;
8784
6c255e67
MX		 8785 if (nr != -1) {
8786 nr -= ret;
8787 WARN_ON(nr < 0);
8788 }
eb73c1b7 8789 spin_lock(&fs_info->delalloc_root_lock);
8ccf6f19 8790 }
eb73c1b7 8791 spin_unlock(&fs_info->delalloc_root_lock);
1eafa6c7 8792
6c255e67 8793 ret = 0;
eb73c1b7
MX		 8794 atomic_inc(&fs_info->async_submit_draining);
8795 while (atomic_read(&fs_info->nr_async_submits) ||
8796 atomic_read(&fs_info->async_delalloc_pages)) {
8797 wait_event(fs_info->async_submit_wait,
8798 (atomic_read(&fs_info->nr_async_submits) == 0 &&
8799 atomic_read(&fs_info->async_delalloc_pages) == 0));
8800 }
8801 atomic_dec(&fs_info->async_submit_draining);
eb73c1b7 8802out:
1eafa6c7 8803 if (!list_empty_careful(&splice)) {
eb73c1b7
MX		 8804 spin_lock(&fs_info->delalloc_root_lock);
8805 list_splice_tail(&splice, &fs_info->delalloc_roots);
8806 spin_unlock(&fs_info->delalloc_root_lock);
1eafa6c7 8807 }
573bfb72 8808 mutex_unlock(&fs_info->delalloc_root_mutex);
8ccf6f19 8809 return ret;
ea8c2819
CM		 8810}
8811
39279cc3
CM		 8812static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8813 const char *symname)
8814{
8815 struct btrfs_trans_handle *trans;
8816 struct btrfs_root *root = BTRFS_I(dir)->root;
8817 struct btrfs_path *path;
8818 struct btrfs_key key;
1832a6d5 8819 struct inode *inode = NULL;
39279cc3
CM		 8820 int err;
8821 int drop_inode = 0;
8822 u64 objectid;
67871254 8823 u64 index = 0;
39279cc3
CM		 8824 int name_len;
8825 int datasize;
5f39d397 8826 unsigned long ptr;
39279cc3 8827 struct btrfs_file_extent_item *ei;
5f39d397 8828 struct extent_buffer *leaf;
39279cc3 8829
f06becc4 8830 name_len = strlen(symname);
39279cc3
CM		 8831 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8832 return -ENAMETOOLONG;
1832a6d5 8833
9ed74f2d
JB		 8834 /*
8835 * 2 items for inode item and ref
8836 * 2 items for dir items
8837 * 1 item for xattr if selinux is on
8838 */
a22285a6
YZ		 8839 trans = btrfs_start_transaction(root, 5);
8840 if (IS_ERR(trans))
8841 return PTR_ERR(trans);
1832a6d5 8842
581bb050
LZ		 8843 err = btrfs_find_free_ino(root, &objectid);
8844 if (err)
8845 goto out_unlock;
8846
aec7477b 8847 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 8848 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 8849 S_IFLNK|S_IRWXUGO, &index);
7cf96da3
TI		 8850 if (IS_ERR(inode)) {
8851 err = PTR_ERR(inode);
39279cc3 8852 goto out_unlock;
7cf96da3 8853 }
39279cc3 8854
ad19db71
CS		 8855 /*
8856 * If the active LSM wants to access the inode during
8857 * d_instantiate it needs these. Smack checks to see
8858 * if the filesystem supports xattrs by looking at the
8859 * ops vector.
8860 */
8861 inode->i_fop = &btrfs_file_operations;
8862 inode->i_op = &btrfs_file_inode_operations;
b0d5d10f
CM		 8863 inode->i_mapping->a_ops = &btrfs_aops;
8864 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8865 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8866
8867 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8868 if (err)
8869 goto out_unlock_inode;
ad19db71 8870
a1b075d2 8871 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3 8872 if (err)
b0d5d10f 8873 goto out_unlock_inode;
39279cc3
CM		 8874
8875 path = btrfs_alloc_path();
d8926bb3
MF		 8876 if (!path) {
8877 err = -ENOMEM;
b0d5d10f 8878 goto out_unlock_inode;
d8926bb3 8879 }
33345d01 8880 key.objectid = btrfs_ino(inode);
39279cc3 8881 key.offset = 0;
962a298f 8882 key.type = BTRFS_EXTENT_DATA_KEY;
39279cc3
CM		 8883 datasize = btrfs_file_extent_calc_inline_size(name_len);
8884 err = btrfs_insert_empty_item(trans, root, path, &key,
8885 datasize);
54aa1f4d 8886 if (err) {
b0839166 8887 btrfs_free_path(path);
b0d5d10f 8888 goto out_unlock_inode;
54aa1f4d 8889 }
5f39d397
CM		 8890 leaf = path->nodes[0];
8891 ei = btrfs_item_ptr(leaf, path->slots[0],
8892 struct btrfs_file_extent_item);
8893 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8894 btrfs_set_file_extent_type(leaf, ei,
39279cc3 8895 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 8896 btrfs_set_file_extent_encryption(leaf, ei, 0);
8897 btrfs_set_file_extent_compression(leaf, ei, 0);
8898 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8899 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8900
39279cc3 8901 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 8902 write_extent_buffer(leaf, symname, ptr, name_len);
8903 btrfs_mark_buffer_dirty(leaf);
39279cc3 8904 btrfs_free_path(path);
5f39d397 8905
39279cc3
CM		 8906 inode->i_op = &btrfs_symlink_inode_operations;
8907 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 8908 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d899e052 8909 inode_set_bytes(inode, name_len);
f06becc4 8910 btrfs_i_size_write(inode, name_len);
54aa1f4d 8911 err = btrfs_update_inode(trans, root, inode);
b0d5d10f 8912 if (err) {
54aa1f4d 8913 drop_inode = 1;
b0d5d10f
CM		 8914 goto out_unlock_inode;
8915 }
8916
8917 unlock_new_inode(inode);
8918 d_instantiate(dentry, inode);
39279cc3
CM		 8919
8920out_unlock:
7ad85bb7 8921 btrfs_end_transaction(trans, root);
39279cc3
CM		 8922 if (drop_inode) {
8923 inode_dec_link_count(inode);
8924 iput(inode);
8925 }
b53d3f5d 8926 btrfs_btree_balance_dirty(root);
39279cc3 8927 return err;
b0d5d10f
CM		 8928
8929out_unlock_inode:
8930 drop_inode = 1;
8931 unlock_new_inode(inode);
8932 goto out_unlock;
39279cc3 8933}
16432985 8934
0af3d00b
JB		 8935static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8936 u64 start, u64 num_bytes, u64 min_size,
8937 loff_t actual_len, u64 *alloc_hint,
8938 struct btrfs_trans_handle *trans)
d899e052 8939{
5dc562c5
JB		 8940 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8941 struct extent_map *em;
d899e052
YZ		 8942 struct btrfs_root *root = BTRFS_I(inode)->root;
8943 struct btrfs_key ins;
d899e052 8944 u64 cur_offset = start;
55a61d1d 8945 u64 i_size;
154ea289 8946 u64 cur_bytes;
d899e052 8947 int ret = 0;
0af3d00b 8948 bool own_trans = true;
d899e052 8949
0af3d00b
JB		 8950 if (trans)
8951 own_trans = false;
d899e052 8952 while (num_bytes > 0) {
0af3d00b
JB		 8953 if (own_trans) {
8954 trans = btrfs_start_transaction(root, 3);
8955 if (IS_ERR(trans)) {
8956 ret = PTR_ERR(trans);
8957 break;
8958 }
5a303d5d
YZ		 8959 }
8960
154ea289
CM		 8961 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8962 cur_bytes = max(cur_bytes, min_size);
00361589 8963 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
e570fd27 8964 *alloc_hint, &ins, 1, 0);
5a303d5d 8965 if (ret) {
0af3d00b
JB		 8966 if (own_trans)
8967 btrfs_end_transaction(trans, root);
a22285a6 8968 break;
d899e052 8969 }
5a303d5d 8970
d899e052
YZ		 8971 ret = insert_reserved_file_extent(trans, inode,
8972 cur_offset, ins.objectid,
8973 ins.offset, ins.offset,
920bbbfb 8974 ins.offset, 0, 0, 0,
d899e052 8975 BTRFS_FILE_EXTENT_PREALLOC);
79787eaa 8976 if (ret) {
857cc2fc 8977 btrfs_free_reserved_extent(root, ins.objectid,
e570fd27 8978 ins.offset, 0);
79787eaa
JM		 8979 btrfs_abort_transaction(trans, root, ret);
8980 if (own_trans)
8981 btrfs_end_transaction(trans, root);
8982 break;
8983 }
a1ed835e
CM		 8984 btrfs_drop_extent_cache(inode, cur_offset,
8985 cur_offset + ins.offset -1, 0);
5a303d5d 8986
5dc562c5
JB		 8987 em = alloc_extent_map();
8988 if (!em) {
8989 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8990 &BTRFS_I(inode)->runtime_flags);
8991 goto next;
8992 }
8993
8994 em->start = cur_offset;
8995 em->orig_start = cur_offset;
8996 em->len = ins.offset;
8997 em->block_start = ins.objectid;
8998 em->block_len = ins.offset;
b4939680 8999 em->orig_block_len = ins.offset;
cc95bef6 9000 em->ram_bytes = ins.offset;
5dc562c5
JB		 9001 em->bdev = root->fs_info->fs_devices->latest_bdev;
9002 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9003 em->generation = trans->transid;
9004
9005 while (1) {
9006 write_lock(&em_tree->lock);
09a2a8f9 9007 ret = add_extent_mapping(em_tree, em, 1);
5dc562c5
JB		 9008 write_unlock(&em_tree->lock);
9009 if (ret != -EEXIST)
9010 break;
9011 btrfs_drop_extent_cache(inode, cur_offset,
9012 cur_offset + ins.offset - 1,
9013 0);
9014 }
9015 free_extent_map(em);
9016next:
d899e052
YZ		 9017 num_bytes -= ins.offset;
9018 cur_offset += ins.offset;
efa56464 9019 *alloc_hint = ins.objectid + ins.offset;
5a303d5d 9020
0c4d2d95 9021 inode_inc_iversion(inode);
d899e052 9022 inode->i_ctime = CURRENT_TIME;
6cbff00f 9023 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
d899e052 9024 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
efa56464
YZ		 9025 (actual_len > inode->i_size) &&
9026 (cur_offset > inode->i_size)) {
d1ea6a61 9027 if (cur_offset > actual_len)
55a61d1d 9028 i_size = actual_len;
d1ea6a61 9029 else
55a61d1d
JB		 9030 i_size = cur_offset;
9031 i_size_write(inode, i_size);
9032 btrfs_ordered_update_i_size(inode, i_size, NULL);
5a303d5d
YZ		 9033 }
9034
d899e052 9035 ret = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 9036
9037 if (ret) {
9038 btrfs_abort_transaction(trans, root, ret);
9039 if (own_trans)
9040 btrfs_end_transaction(trans, root);
9041 break;
9042 }
d899e052 9043
0af3d00b
JB		 9044 if (own_trans)
9045 btrfs_end_transaction(trans, root);
5a303d5d 9046 }
d899e052
YZ		 9047 return ret;
9048}
9049
0af3d00b
JB		 9050int btrfs_prealloc_file_range(struct inode *inode, int mode,
9051 u64 start, u64 num_bytes, u64 min_size,
9052 loff_t actual_len, u64 *alloc_hint)
9053{
9054 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9055 min_size, actual_len, alloc_hint,
9056 NULL);
9057}
9058
9059int btrfs_prealloc_file_range_trans(struct inode *inode,
9060 struct btrfs_trans_handle *trans, int mode,
9061 u64 start, u64 num_bytes, u64 min_size,
9062 loff_t actual_len, u64 *alloc_hint)
9063{
9064 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9065 min_size, actual_len, alloc_hint, trans);
9066}
9067
e6dcd2dc
CM		 9068static int btrfs_set_page_dirty(struct page *page)
9069{
e6dcd2dc
CM		 9070 return __set_page_dirty_nobuffers(page);
9071}
9072
10556cb2 9073static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd 9074{
b83cc969 9075 struct btrfs_root *root = BTRFS_I(inode)->root;
cb6db4e5 9076 umode_t mode = inode->i_mode;
b83cc969 9077
cb6db4e5
JM		 9078 if (mask & MAY_WRITE &&
9079 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9080 if (btrfs_root_readonly(root))
9081 return -EROFS;
9082 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9083 return -EACCES;
9084 }
2830ba7f 9085 return generic_permission(inode, mask);
fdebe2bd 9086}
39279cc3 9087
ef3b9af5
FM		 9088static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9089{
9090 struct btrfs_trans_handle *trans;
9091 struct btrfs_root *root = BTRFS_I(dir)->root;
9092 struct inode *inode = NULL;
9093 u64 objectid;
9094 u64 index;
9095 int ret = 0;
9096
9097 /*
9098 * 5 units required for adding orphan entry
9099 */
9100 trans = btrfs_start_transaction(root, 5);
9101 if (IS_ERR(trans))
9102 return PTR_ERR(trans);
9103
9104 ret = btrfs_find_free_ino(root, &objectid);
9105 if (ret)
9106 goto out;
9107
9108 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9109 btrfs_ino(dir), objectid, mode, &index);
9110 if (IS_ERR(inode)) {
9111 ret = PTR_ERR(inode);
9112 inode = NULL;
9113 goto out;
9114 }
9115
ef3b9af5
FM		 9116 inode->i_fop = &btrfs_file_operations;
9117 inode->i_op = &btrfs_file_inode_operations;
9118
9119 inode->i_mapping->a_ops = &btrfs_aops;
9120 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
9121 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9122
b0d5d10f
CM		 9123 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9124 if (ret)
9125 goto out_inode;
9126
9127 ret = btrfs_update_inode(trans, root, inode);
9128 if (ret)
9129 goto out_inode;
ef3b9af5
FM		 9130 ret = btrfs_orphan_add(trans, inode);
9131 if (ret)
b0d5d10f 9132 goto out_inode;
ef3b9af5 9133
5762b5c9
FM		 9134 /*
9135 * We set number of links to 0 in btrfs_new_inode(), and here we set
9136 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9137 * through:
9138 *
9139 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9140 */
9141 set_nlink(inode, 1);
b0d5d10f 9142 unlock_new_inode(inode);
ef3b9af5
FM		 9143 d_tmpfile(dentry, inode);
9144 mark_inode_dirty(inode);
9145
9146out:
9147 btrfs_end_transaction(trans, root);
9148 if (ret)
9149 iput(inode);
9150 btrfs_balance_delayed_items(root);
9151 btrfs_btree_balance_dirty(root);
ef3b9af5 9152 return ret;
b0d5d10f
CM		 9153
9154out_inode:
9155 unlock_new_inode(inode);
9156 goto out;
9157
ef3b9af5
FM		 9158}
9159
6e1d5dcc 9160static const struct inode_operations btrfs_dir_inode_operations = {
3394e160 9161 .getattr = btrfs_getattr,
39279cc3
CM		 9162 .lookup = btrfs_lookup,
9163 .create = btrfs_create,
9164 .unlink = btrfs_unlink,
9165 .link = btrfs_link,
9166 .mkdir = btrfs_mkdir,
9167 .rmdir = btrfs_rmdir,
80ace85c 9168 .rename2 = btrfs_rename2,
39279cc3
CM		 9169 .symlink = btrfs_symlink,
9170 .setattr = btrfs_setattr,
618e21d5 9171 .mknod = btrfs_mknod,
95819c05
CH		 9172 .setxattr = btrfs_setxattr,
9173 .getxattr = btrfs_getxattr,
5103e947 9174 .listxattr = btrfs_listxattr,
95819c05 9175 .removexattr = btrfs_removexattr,
fdebe2bd 9176 .permission = btrfs_permission,
4e34e719 9177 .get_acl = btrfs_get_acl,
996a710d 9178 .set_acl = btrfs_set_acl,
93fd63c2 9179 .update_time = btrfs_update_time,
ef3b9af5 9180 .tmpfile = btrfs_tmpfile,
39279cc3 9181};
6e1d5dcc 9182static const struct inode_operations btrfs_dir_ro_inode_operations = {
39279cc3 9183 .lookup = btrfs_lookup,
fdebe2bd 9184 .permission = btrfs_permission,
4e34e719 9185 .get_acl = btrfs_get_acl,
996a710d 9186 .set_acl = btrfs_set_acl,
93fd63c2 9187 .update_time = btrfs_update_time,
39279cc3 9188};
76dda93c 9189
828c0950 9190static const struct file_operations btrfs_dir_file_operations = {
39279cc3
CM		 9191 .llseek = generic_file_llseek,
9192 .read = generic_read_dir,
9cdda8d3 9193 .iterate = btrfs_real_readdir,
34287aa3 9194 .unlocked_ioctl = btrfs_ioctl,
39279cc3 9195#ifdef CONFIG_COMPAT
34287aa3 9196 .compat_ioctl = btrfs_ioctl,
39279cc3 9197#endif
6bf13c0c 9198 .release = btrfs_release_file,
e02119d5 9199 .fsync = btrfs_sync_file,
39279cc3
CM		 9200};
9201
d1310b2e 9202static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 9203 .fill_delalloc = run_delalloc_range,
065631f6 9204 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 9205 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 9206 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 9207 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 9208 .writepage_start_hook = btrfs_writepage_start_hook,
b0c68f8b
CM		 9209 .set_bit_hook = btrfs_set_bit_hook,
9210 .clear_bit_hook = btrfs_clear_bit_hook,
9ed74f2d
JB		 9211 .merge_extent_hook = btrfs_merge_extent_hook,
9212 .split_extent_hook = btrfs_split_extent_hook,
07157aac
CM		 9213};
9214
35054394
CM		 9215/*
9216 * btrfs doesn't support the bmap operation because swapfiles
9217 * use bmap to make a mapping of extents in the file. They assume
9218 * these extents won't change over the life of the file and they
9219 * use the bmap result to do IO directly to the drive.
9220 *
9221 * the btrfs bmap call would return logical addresses that aren't
9222 * suitable for IO and they also will change frequently as COW
9223 * operations happen. So, swapfile + btrfs == corruption.
9224 *
9225 * For now we're avoiding this by dropping bmap.
9226 */
7f09410b 9227static const struct address_space_operations btrfs_aops = {
39279cc3
CM		 9228 .readpage = btrfs_readpage,
9229 .writepage = btrfs_writepage,
b293f02e 9230 .writepages = btrfs_writepages,
3ab2fb5a 9231 .readpages = btrfs_readpages,
16432985 9232 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 9233 .invalidatepage = btrfs_invalidatepage,
9234 .releasepage = btrfs_releasepage,
e6dcd2dc 9235 .set_page_dirty = btrfs_set_page_dirty,
465fdd97 9236 .error_remove_page = generic_error_remove_page,
39279cc3
CM		 9237};
9238
7f09410b 9239static const struct address_space_operations btrfs_symlink_aops = {
39279cc3
CM		 9240 .readpage = btrfs_readpage,
9241 .writepage = btrfs_writepage,
2bf5a725
CM		 9242 .invalidatepage = btrfs_invalidatepage,
9243 .releasepage = btrfs_releasepage,
39279cc3
CM		 9244};
9245
6e1d5dcc 9246static const struct inode_operations btrfs_file_inode_operations = {
39279cc3
CM		 9247 .getattr = btrfs_getattr,
9248 .setattr = btrfs_setattr,
95819c05
CH		 9249 .setxattr = btrfs_setxattr,
9250 .getxattr = btrfs_getxattr,
5103e947 9251 .listxattr = btrfs_listxattr,
95819c05 9252 .removexattr = btrfs_removexattr,
fdebe2bd 9253 .permission = btrfs_permission,
1506fcc8 9254 .fiemap = btrfs_fiemap,
4e34e719 9255 .get_acl = btrfs_get_acl,
996a710d 9256 .set_acl = btrfs_set_acl,
e41f941a 9257 .update_time = btrfs_update_time,
39279cc3 9258};
6e1d5dcc 9259static const struct inode_operations btrfs_special_inode_operations = {
618e21d5
JB		 9260 .getattr = btrfs_getattr,
9261 .setattr = btrfs_setattr,
fdebe2bd 9262 .permission = btrfs_permission,
95819c05
CH		 9263 .setxattr = btrfs_setxattr,
9264 .getxattr = btrfs_getxattr,
33268eaf 9265 .listxattr = btrfs_listxattr,
95819c05 9266 .removexattr = btrfs_removexattr,
4e34e719 9267 .get_acl = btrfs_get_acl,
996a710d 9268 .set_acl = btrfs_set_acl,
e41f941a 9269 .update_time = btrfs_update_time,
618e21d5 9270};
6e1d5dcc 9271static const struct inode_operations btrfs_symlink_inode_operations = {
39279cc3
CM		 9272 .readlink = generic_readlink,
9273 .follow_link = page_follow_link_light,
9274 .put_link = page_put_link,
f209561a 9275 .getattr = btrfs_getattr,
22c44fe6 9276 .setattr = btrfs_setattr,
fdebe2bd 9277 .permission = btrfs_permission,
0279b4cd
JO		 9278 .setxattr = btrfs_setxattr,
9279 .getxattr = btrfs_getxattr,
9280 .listxattr = btrfs_listxattr,
9281 .removexattr = btrfs_removexattr,
e41f941a 9282 .update_time = btrfs_update_time,
39279cc3 9283};
76dda93c 9284
82d339d9 9285const struct dentry_operations btrfs_dentry_operations = {
76dda93c 9286 .d_delete = btrfs_dentry_delete,
b4aff1f8 9287 .d_release = btrfs_dentry_release,
76dda93c 9288};
Linux 6.x block layer and io_uring tree(s)