projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: qgroup: Introduce a may_use to account space_info->bytes_may_use.
[deliverable/linux.git] / fs / btrfs / inode.c
CommitLineData
6cbd5570
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
8f18cf13 19#include <linux/kernel.h>
065631f6 20#include <linux/bio.h>
39279cc3 21#include <linux/buffer_head.h>
f2eb0a24 22#include <linux/file.h>
39279cc3
CM		 23#include <linux/fs.h>
24#include <linux/pagemap.h>
25#include <linux/highmem.h>
26#include <linux/time.h>
27#include <linux/init.h>
28#include <linux/string.h>
39279cc3
CM		 29#include <linux/backing-dev.h>
30#include <linux/mpage.h>
31#include <linux/swap.h>
32#include <linux/writeback.h>
33#include <linux/statfs.h>
34#include <linux/compat.h>
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>
4a3d1caf 46#include <linux/uio.h>
39279cc3
CM		 47#include "ctree.h"
48#include "disk-io.h"
49#include "transaction.h"
50#include "btrfs_inode.h"
39279cc3 51#include "print-tree.h"
e6dcd2dc 52#include "ordered-data.h"
95819c05 53#include "xattr.h"
e02119d5 54#include "tree-log.h"
4a54c8c1 55#include "volumes.h"
c8b97818 56#include "compression.h"
b4ce94de 57#include "locking.h"
dc89e982 58#include "free-space-cache.h"
581bb050 59#include "inode-map.h"
38c227d8 60#include "backref.h"
f23b5a59 61#include "hash.h"
63541927 62#include "props.h"
31193213 63#include "qgroup.h"
39279cc3
CM		 64
65struct btrfs_iget_args {
90d3e592 66 struct btrfs_key *location;
39279cc3
CM		 67 struct btrfs_root *root;
68};
69
6e1d5dcc
AD		 70static const struct inode_operations btrfs_dir_inode_operations;
71static const struct inode_operations btrfs_symlink_inode_operations;
72static const struct inode_operations btrfs_dir_ro_inode_operations;
73static const struct inode_operations btrfs_special_inode_operations;
74static const struct inode_operations btrfs_file_inode_operations;
7f09410b
AD		 75static const struct address_space_operations btrfs_aops;
76static const struct address_space_operations btrfs_symlink_aops;
828c0950 77static const struct file_operations btrfs_dir_file_operations;
d1310b2e 78static struct extent_io_ops btrfs_extent_io_ops;
39279cc3
CM		 79
80static struct kmem_cache *btrfs_inode_cachep;
8ccf6f19 81static struct kmem_cache *btrfs_delalloc_work_cachep;
39279cc3
CM		 82struct kmem_cache *btrfs_trans_handle_cachep;
83struct kmem_cache *btrfs_transaction_cachep;
39279cc3 84struct kmem_cache *btrfs_path_cachep;
dc89e982 85struct kmem_cache *btrfs_free_space_cachep;
39279cc3
CM		 86
87#define S_SHIFT 12
88static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
89 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
90 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
91 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
92 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
93 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
94 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
95 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
96};
97
3972f260 98static int btrfs_setsize(struct inode *inode, struct iattr *attr);
a41ad394 99static int btrfs_truncate(struct inode *inode);
5fd02043 100static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
771ed689
CM		 101static noinline int cow_file_range(struct inode *inode,
102 struct page *locked_page,
103 u64 start, u64 end, int *page_started,
104 unsigned long *nr_written, int unlock);
70c8a91c
JB		 105static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
106 u64 len, u64 orig_start,
107 u64 block_start, u64 block_len,
cc95bef6
JB		 108 u64 orig_block_len, u64 ram_bytes,
109 int type);
7b128766 110
48a3b636 111static int btrfs_dirty_inode(struct inode *inode);
7b128766 112
6a3891c5
JB		 113#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
114void btrfs_test_inode_set_ops(struct inode *inode)
115{
116 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
117}
118#endif
119
f34f57a3 120static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
2a7dba39
EP		 121 struct inode *inode, struct inode *dir,
122 const struct qstr *qstr)
0279b4cd
JO		 123{
124 int err;
125
f34f57a3 126 err = btrfs_init_acl(trans, inode, dir);
0279b4cd 127 if (!err)
2a7dba39 128 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
0279b4cd
JO		 129 return err;
130}
131
c8b97818
CM		 132/*
133 * this does all the hard work for inserting an inline extent into
134 * the btree. The caller should have done a btrfs_drop_extents so that
135 * no overlapping inline items exist in the btree
136 */
40f76580 137static int insert_inline_extent(struct btrfs_trans_handle *trans,
1acae57b 138 struct btrfs_path *path, int extent_inserted,
c8b97818
CM		 139 struct btrfs_root *root, struct inode *inode,
140 u64 start, size_t size, size_t compressed_size,
fe3f566c 141 int compress_type,
c8b97818
CM		 142 struct page **compressed_pages)
143{
c8b97818
CM		 144 struct extent_buffer *leaf;
145 struct page *page = NULL;
146 char *kaddr;
147 unsigned long ptr;
148 struct btrfs_file_extent_item *ei;
149 int err = 0;
150 int ret;
151 size_t cur_size = size;
c8b97818 152 unsigned long offset;
c8b97818 153
fe3f566c 154 if (compressed_size && compressed_pages)
c8b97818 155 cur_size = compressed_size;
c8b97818 156
1acae57b 157 inode_add_bytes(inode, size);
c8b97818 158
1acae57b
FDBM		 159 if (!extent_inserted) {
160 struct btrfs_key key;
161 size_t datasize;
c8b97818 162
1acae57b
FDBM		 163 key.objectid = btrfs_ino(inode);
164 key.offset = start;
962a298f 165 key.type = BTRFS_EXTENT_DATA_KEY;
c8b97818 166
1acae57b
FDBM		 167 datasize = btrfs_file_extent_calc_inline_size(cur_size);
168 path->leave_spinning = 1;
169 ret = btrfs_insert_empty_item(trans, root, path, &key,
170 datasize);
171 if (ret) {
172 err = ret;
173 goto fail;
174 }
c8b97818
CM		 175 }
176 leaf = path->nodes[0];
177 ei = btrfs_item_ptr(leaf, path->slots[0],
178 struct btrfs_file_extent_item);
179 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
180 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
181 btrfs_set_file_extent_encryption(leaf, ei, 0);
182 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
183 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
184 ptr = btrfs_file_extent_inline_start(ei);
185
261507a0 186 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818
CM		 187 struct page *cpage;
188 int i = 0;
d397712b 189 while (compressed_size > 0) {
c8b97818 190 cpage = compressed_pages[i];
5b050f04 191 cur_size = min_t(unsigned long, compressed_size,
c8b97818
CM		 192 PAGE_CACHE_SIZE);
193
7ac687d9 194 kaddr = kmap_atomic(cpage);
c8b97818 195 write_extent_buffer(leaf, kaddr, ptr, cur_size);
7ac687d9 196 kunmap_atomic(kaddr);
c8b97818
CM		 197
198 i++;
199 ptr += cur_size;
200 compressed_size -= cur_size;
201 }
202 btrfs_set_file_extent_compression(leaf, ei,
261507a0 203 compress_type);
c8b97818
CM		 204 } else {
205 page = find_get_page(inode->i_mapping,
206 start >> PAGE_CACHE_SHIFT);
207 btrfs_set_file_extent_compression(leaf, ei, 0);
7ac687d9 208 kaddr = kmap_atomic(page);
c8b97818
CM		 209 offset = start & (PAGE_CACHE_SIZE - 1);
210 write_extent_buffer(leaf, kaddr + offset, ptr, size);
7ac687d9 211 kunmap_atomic(kaddr);
c8b97818
CM		 212 page_cache_release(page);
213 }
214 btrfs_mark_buffer_dirty(leaf);
1acae57b 215 btrfs_release_path(path);
c8b97818 216
c2167754
YZ		 217 /*
218 * we're an inline extent, so nobody can
219 * extend the file past i_size without locking
220 * a page we already have locked.
221 *
222 * We must do any isize and inode updates
223 * before we unlock the pages. Otherwise we
224 * could end up racing with unlink.
225 */
c8b97818 226 BTRFS_I(inode)->disk_i_size = inode->i_size;
79787eaa 227 ret = btrfs_update_inode(trans, root, inode);
c2167754 228
79787eaa 229 return ret;
c8b97818 230fail:
c8b97818
CM		 231 return err;
232}
233
234
235/*
236 * conditionally insert an inline extent into the file. This
237 * does the checks required to make sure the data is small enough
238 * to fit as an inline extent.
239 */
00361589
JB		 240static noinline int cow_file_range_inline(struct btrfs_root *root,
241 struct inode *inode, u64 start,
242 u64 end, size_t compressed_size,
243 int compress_type,
244 struct page **compressed_pages)
c8b97818 245{
00361589 246 struct btrfs_trans_handle *trans;
c8b97818
CM		 247 u64 isize = i_size_read(inode);
248 u64 actual_end = min(end + 1, isize);
249 u64 inline_len = actual_end - start;
fda2832f 250 u64 aligned_end = ALIGN(end, root->sectorsize);
c8b97818
CM		 251 u64 data_len = inline_len;
252 int ret;
1acae57b
FDBM		 253 struct btrfs_path *path;
254 int extent_inserted = 0;
255 u32 extent_item_size;
c8b97818
CM		 256
257 if (compressed_size)
258 data_len = compressed_size;
259
260 if (start > 0 ||
354877be
WS		 261 actual_end > PAGE_CACHE_SIZE ||
262 data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
c8b97818
CM		 263 (!compressed_size &&
264 (actual_end & (root->sectorsize - 1)) == 0) ||
265 end + 1 < isize ||
266 data_len > root->fs_info->max_inline) {
267 return 1;
268 }
269
1acae57b
FDBM		 270 path = btrfs_alloc_path();
271 if (!path)
272 return -ENOMEM;
273
00361589 274 trans = btrfs_join_transaction(root);
1acae57b
FDBM		 275 if (IS_ERR(trans)) {
276 btrfs_free_path(path);
00361589 277 return PTR_ERR(trans);
1acae57b 278 }
00361589
JB		 279 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
280
1acae57b
FDBM		 281 if (compressed_size && compressed_pages)
282 extent_item_size = btrfs_file_extent_calc_inline_size(
283 compressed_size);
284 else
285 extent_item_size = btrfs_file_extent_calc_inline_size(
286 inline_len);
287
288 ret = __btrfs_drop_extents(trans, root, inode, path,
289 start, aligned_end, NULL,
290 1, 1, extent_item_size, &extent_inserted);
00361589
JB		 291 if (ret) {
292 btrfs_abort_transaction(trans, root, ret);
293 goto out;
294 }
c8b97818
CM		 295
296 if (isize > actual_end)
297 inline_len = min_t(u64, isize, actual_end);
1acae57b
FDBM		 298 ret = insert_inline_extent(trans, path, extent_inserted,
299 root, inode, start,
c8b97818 300 inline_len, compressed_size,
fe3f566c 301 compress_type, compressed_pages);
2adcac1a 302 if (ret && ret != -ENOSPC) {
79787eaa 303 btrfs_abort_transaction(trans, root, ret);
00361589 304 goto out;
2adcac1a 305 } else if (ret == -ENOSPC) {
00361589
JB		 306 ret = 1;
307 goto out;
79787eaa 308 }
2adcac1a 309
bdc20e67 310 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
0ca1f7ce 311 btrfs_delalloc_release_metadata(inode, end + 1 - start);
a1ed835e 312 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
00361589 313out:
1acae57b 314 btrfs_free_path(path);
00361589
JB		 315 btrfs_end_transaction(trans, root);
316 return ret;
c8b97818
CM		 317}
318
771ed689
CM		 319struct async_extent {
320 u64 start;
321 u64 ram_size;
322 u64 compressed_size;
323 struct page **pages;
324 unsigned long nr_pages;
261507a0 325 int compress_type;
771ed689
CM		 326 struct list_head list;
327};
328
329struct async_cow {
330 struct inode *inode;
331 struct btrfs_root *root;
332 struct page *locked_page;
333 u64 start;
334 u64 end;
335 struct list_head extents;
336 struct btrfs_work work;
337};
338
339static noinline int add_async_extent(struct async_cow *cow,
340 u64 start, u64 ram_size,
341 u64 compressed_size,
342 struct page **pages,
261507a0
LZ		 343 unsigned long nr_pages,
344 int compress_type)
771ed689
CM		 345{
346 struct async_extent *async_extent;
347
348 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
79787eaa 349 BUG_ON(!async_extent); /* -ENOMEM */
771ed689
CM		 350 async_extent->start = start;
351 async_extent->ram_size = ram_size;
352 async_extent->compressed_size = compressed_size;
353 async_extent->pages = pages;
354 async_extent->nr_pages = nr_pages;
261507a0 355 async_extent->compress_type = compress_type;
771ed689
CM		 356 list_add_tail(&async_extent->list, &cow->extents);
357 return 0;
358}
359
f79707b0
WS		 360static inline int inode_need_compress(struct inode *inode)
361{
362 struct btrfs_root *root = BTRFS_I(inode)->root;
363
364 /* force compress */
365 if (btrfs_test_opt(root, FORCE_COMPRESS))
366 return 1;
367 /* bad compression ratios */
368 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
369 return 0;
370 if (btrfs_test_opt(root, COMPRESS) ||
371 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
372 BTRFS_I(inode)->force_compress)
373 return 1;
374 return 0;
375}
376
d352ac68 377/*
771ed689
CM		 378 * we create compressed extents in two phases. The first
379 * phase compresses a range of pages that have already been
380 * locked (both pages and state bits are locked).
c8b97818 381 *
771ed689
CM		 382 * This is done inside an ordered work queue, and the compression
383 * is spread across many cpus. The actual IO submission is step
384 * two, and the ordered work queue takes care of making sure that
385 * happens in the same order things were put onto the queue by
386 * writepages and friends.
c8b97818 387 *
771ed689
CM		 388 * If this code finds it can't get good compression, it puts an
389 * entry onto the work queue to write the uncompressed bytes. This
390 * makes sure that both compressed inodes and uncompressed inodes
b2570314
AB		 391 * are written in the same order that the flusher thread sent them
392 * down.
d352ac68 393 */
c44f649e 394static noinline void compress_file_range(struct inode *inode,
771ed689
CM		 395 struct page *locked_page,
396 u64 start, u64 end,
397 struct async_cow *async_cow,
398 int *num_added)
b888db2b
CM		 399{
400 struct btrfs_root *root = BTRFS_I(inode)->root;
db94535d 401 u64 num_bytes;
db94535d 402 u64 blocksize = root->sectorsize;
c8b97818 403 u64 actual_end;
42dc7bab 404 u64 isize = i_size_read(inode);
e6dcd2dc 405 int ret = 0;
c8b97818
CM		 406 struct page **pages = NULL;
407 unsigned long nr_pages;
408 unsigned long nr_pages_ret = 0;
409 unsigned long total_compressed = 0;
410 unsigned long total_in = 0;
411 unsigned long max_compressed = 128 * 1024;
771ed689 412 unsigned long max_uncompressed = 128 * 1024;
c8b97818
CM		 413 int i;
414 int will_compress;
261507a0 415 int compress_type = root->fs_info->compress_type;
4adaa611 416 int redirty = 0;
b888db2b 417
4cb13e5d
LB		 418 /* if this is a small write inside eof, kick off a defrag */
419 if ((end - start + 1) < 16 * 1024 &&
420 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
4cb5300b
CM		 421 btrfs_add_inode_defrag(NULL, inode);
422
42dc7bab 423 actual_end = min_t(u64, isize, end + 1);
c8b97818
CM		 424again:
425 will_compress = 0;
426 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
427 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
be20aa9d 428
f03d9301
CM		 429 /*
430 * we don't want to send crud past the end of i_size through
431 * compression, that's just a waste of CPU time. So, if the
432 * end of the file is before the start of our current
433 * requested range of bytes, we bail out to the uncompressed
434 * cleanup code that can deal with all of this.
435 *
436 * It isn't really the fastest way to fix things, but this is a
437 * very uncommon corner.
438 */
439 if (actual_end <= start)
440 goto cleanup_and_bail_uncompressed;
441
c8b97818
CM		 442 total_compressed = actual_end - start;
443
4bcbb332
SW		 444 /*
445 * skip compression for a small file range(<=blocksize) that
446 * isn't an inline extent, since it dosen't save disk space at all.
447 */
448 if (total_compressed <= blocksize &&
449 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
450 goto cleanup_and_bail_uncompressed;
451
c8b97818
CM		 452 /* we want to make sure that amount of ram required to uncompress
453 * an extent is reasonable, so we limit the total size in ram
771ed689
CM		 454 * of a compressed extent to 128k. This is a crucial number
455 * because it also controls how easily we can spread reads across
456 * cpus for decompression.
457 *
458 * We also want to make sure the amount of IO required to do
459 * a random read is reasonably small, so we limit the size of
460 * a compressed extent to 128k.
c8b97818
CM		 461 */
462 total_compressed = min(total_compressed, max_uncompressed);
fda2832f 463 num_bytes = ALIGN(end - start + 1, blocksize);
be20aa9d 464 num_bytes = max(blocksize, num_bytes);
c8b97818
CM		 465 total_in = 0;
466 ret = 0;
db94535d 467
771ed689
CM		 468 /*
469 * we do compression for mount -o compress and when the
470 * inode has not been flagged as nocompress. This flag can
471 * change at any time if we discover bad compression ratios.
c8b97818 472 */
f79707b0 473 if (inode_need_compress(inode)) {
c8b97818 474 WARN_ON(pages);
31e818fe 475 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
560f7d75
LZ		 476 if (!pages) {
477 /* just bail out to the uncompressed code */
478 goto cont;
479 }
c8b97818 480
261507a0
LZ		 481 if (BTRFS_I(inode)->force_compress)
482 compress_type = BTRFS_I(inode)->force_compress;
483
4adaa611
CM		 484 /*
485 * we need to call clear_page_dirty_for_io on each
486 * page in the range. Otherwise applications with the file
487 * mmap'd can wander in and change the page contents while
488 * we are compressing them.
489 *
490 * If the compression fails for any reason, we set the pages
491 * dirty again later on.
492 */
493 extent_range_clear_dirty_for_io(inode, start, end);
494 redirty = 1;
261507a0
LZ		 495 ret = btrfs_compress_pages(compress_type,
496 inode->i_mapping, start,
497 total_compressed, pages,
498 nr_pages, &nr_pages_ret,
499 &total_in,
500 &total_compressed,
501 max_compressed);
c8b97818
CM		 502
503 if (!ret) {
504 unsigned long offset = total_compressed &
505 (PAGE_CACHE_SIZE - 1);
506 struct page *page = pages[nr_pages_ret - 1];
507 char *kaddr;
508
509 /* zero the tail end of the last page, we might be
510 * sending it down to disk
511 */
512 if (offset) {
7ac687d9 513 kaddr = kmap_atomic(page);
c8b97818
CM		 514 memset(kaddr + offset, 0,
515 PAGE_CACHE_SIZE - offset);
7ac687d9 516 kunmap_atomic(kaddr);
c8b97818
CM		 517 }
518 will_compress = 1;
519 }
520 }
560f7d75 521cont:
c8b97818
CM		 522 if (start == 0) {
523 /* lets try to make an inline extent */
771ed689 524 if (ret || total_in < (actual_end - start)) {
c8b97818 525 /* we didn't compress the entire range, try
771ed689 526 * to make an uncompressed inline extent.
c8b97818 527 */
00361589
JB		 528 ret = cow_file_range_inline(root, inode, start, end,
529 0, 0, NULL);
c8b97818 530 } else {
771ed689 531 /* try making a compressed inline extent */
00361589 532 ret = cow_file_range_inline(root, inode, start, end,
fe3f566c
LZ		 533 total_compressed,
534 compress_type, pages);
c8b97818 535 }
79787eaa 536 if (ret <= 0) {
151a41bc
JB		 537 unsigned long clear_flags = EXTENT_DELALLOC |
538 EXTENT_DEFRAG;
e6eb4314
FM		 539 unsigned long page_error_op;
540
151a41bc 541 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
e6eb4314 542 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
151a41bc 543
771ed689 544 /*
79787eaa
JM		 545 * inline extent creation worked or returned error,
546 * we don't need to create any more async work items.
547 * Unlock and free up our temp pages.
771ed689 548 */
c2790a2e 549 extent_clear_unlock_delalloc(inode, start, end, NULL,
151a41bc 550 clear_flags, PAGE_UNLOCK |
c2790a2e
JB		 551 PAGE_CLEAR_DIRTY |
552 PAGE_SET_WRITEBACK |
e6eb4314 553 page_error_op |
c2790a2e 554 PAGE_END_WRITEBACK);
c8b97818
CM		 555 goto free_pages_out;
556 }
557 }
558
559 if (will_compress) {
560 /*
561 * we aren't doing an inline extent round the compressed size
562 * up to a block size boundary so the allocator does sane
563 * things
564 */
fda2832f 565 total_compressed = ALIGN(total_compressed, blocksize);
c8b97818
CM		 566
567 /*
568 * one last check to make sure the compression is really a
569 * win, compare the page count read with the blocks on disk
570 */
fda2832f 571 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
c8b97818
CM		 572 if (total_compressed >= total_in) {
573 will_compress = 0;
574 } else {
c8b97818
CM		 575 num_bytes = total_in;
576 }
577 }
578 if (!will_compress && pages) {
579 /*
580 * the compression code ran but failed to make things smaller,
581 * free any pages it allocated and our page pointer array
582 */
583 for (i = 0; i < nr_pages_ret; i++) {
70b99e69 584 WARN_ON(pages[i]->mapping);
c8b97818
CM		 585 page_cache_release(pages[i]);
586 }
587 kfree(pages);
588 pages = NULL;
589 total_compressed = 0;
590 nr_pages_ret = 0;
591
592 /* flag the file so we don't compress in the future */
1e701a32
CM		 593 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
594 !(BTRFS_I(inode)->force_compress)) {
a555f810 595 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
1e701a32 596 }
c8b97818 597 }
771ed689
CM		 598 if (will_compress) {
599 *num_added += 1;
c8b97818 600
771ed689
CM		 601 /* the async work queues will take care of doing actual
602 * allocation on disk for these compressed pages,
603 * and will submit them to the elevator.
604 */
605 add_async_extent(async_cow, start, num_bytes,
261507a0
LZ		 606 total_compressed, pages, nr_pages_ret,
607 compress_type);
179e29e4 608
24ae6365 609 if (start + num_bytes < end) {
771ed689
CM		 610 start += num_bytes;
611 pages = NULL;
612 cond_resched();
613 goto again;
614 }
615 } else {
f03d9301 616cleanup_and_bail_uncompressed:
771ed689
CM		 617 /*
618 * No compression, but we still need to write the pages in
619 * the file we've been given so far. redirty the locked
620 * page if it corresponds to our extent and set things up
621 * for the async work queue to run cow_file_range to do
622 * the normal delalloc dance
623 */
624 if (page_offset(locked_page) >= start &&
625 page_offset(locked_page) <= end) {
626 __set_page_dirty_nobuffers(locked_page);
627 /* unlocked later on in the async handlers */
628 }
4adaa611
CM		 629 if (redirty)
630 extent_range_redirty_for_io(inode, start, end);
261507a0
LZ		 631 add_async_extent(async_cow, start, end - start + 1,
632 0, NULL, 0, BTRFS_COMPRESS_NONE);
771ed689
CM		 633 *num_added += 1;
634 }
3b951516 635
c44f649e 636 return;
771ed689
CM		 637
638free_pages_out:
639 for (i = 0; i < nr_pages_ret; i++) {
640 WARN_ON(pages[i]->mapping);
641 page_cache_release(pages[i]);
642 }
d397712b 643 kfree(pages);
771ed689 644}
771ed689 645
40ae837b
FM		 646static void free_async_extent_pages(struct async_extent *async_extent)
647{
648 int i;
649
650 if (!async_extent->pages)
651 return;
652
653 for (i = 0; i < async_extent->nr_pages; i++) {
654 WARN_ON(async_extent->pages[i]->mapping);
655 page_cache_release(async_extent->pages[i]);
656 }
657 kfree(async_extent->pages);
658 async_extent->nr_pages = 0;
659 async_extent->pages = NULL;
771ed689
CM		 660}
661
662/*
663 * phase two of compressed writeback. This is the ordered portion
664 * of the code, which only gets called in the order the work was
665 * queued. We walk all the async extents created by compress_file_range
666 * and send them down to the disk.
667 */
dec8f175 668static noinline void submit_compressed_extents(struct inode *inode,
771ed689
CM		 669 struct async_cow *async_cow)
670{
671 struct async_extent *async_extent;
672 u64 alloc_hint = 0;
771ed689
CM		 673 struct btrfs_key ins;
674 struct extent_map *em;
675 struct btrfs_root *root = BTRFS_I(inode)->root;
676 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
677 struct extent_io_tree *io_tree;
f5a84ee3 678 int ret = 0;
771ed689 679
3e04e7f1 680again:
d397712b 681 while (!list_empty(&async_cow->extents)) {
771ed689
CM		 682 async_extent = list_entry(async_cow->extents.next,
683 struct async_extent, list);
684 list_del(&async_extent->list);
c8b97818 685
771ed689
CM		 686 io_tree = &BTRFS_I(inode)->io_tree;
687
f5a84ee3 688retry:
771ed689
CM		 689 /* did the compression code fall back to uncompressed IO? */
690 if (!async_extent->pages) {
691 int page_started = 0;
692 unsigned long nr_written = 0;
693
694 lock_extent(io_tree, async_extent->start,
2ac55d41 695 async_extent->start +
d0082371 696 async_extent->ram_size - 1);
771ed689
CM		 697
698 /* allocate blocks */
f5a84ee3
JB		 699 ret = cow_file_range(inode, async_cow->locked_page,
700 async_extent->start,
701 async_extent->start +
702 async_extent->ram_size - 1,
703 &page_started, &nr_written, 0);
771ed689 704
79787eaa
JM		 705 /* JDM XXX */
706
771ed689
CM		 707 /*
708 * if page_started, cow_file_range inserted an
709 * inline extent and took care of all the unlocking
710 * and IO for us. Otherwise, we need to submit
711 * all those pages down to the drive.
712 */
f5a84ee3 713 if (!page_started && !ret)
771ed689
CM		 714 extent_write_locked_range(io_tree,
715 inode, async_extent->start,
d397712b 716 async_extent->start +
771ed689
CM		 717 async_extent->ram_size - 1,
718 btrfs_get_extent,
719 WB_SYNC_ALL);
3e04e7f1
JB		 720 else if (ret)
721 unlock_page(async_cow->locked_page);
771ed689
CM		 722 kfree(async_extent);
723 cond_resched();
724 continue;
725 }
726
727 lock_extent(io_tree, async_extent->start,
d0082371 728 async_extent->start + async_extent->ram_size - 1);
771ed689 729
00361589 730 ret = btrfs_reserve_extent(root,
771ed689
CM		 731 async_extent->compressed_size,
732 async_extent->compressed_size,
e570fd27 733 0, alloc_hint, &ins, 1, 1);
f5a84ee3 734 if (ret) {
40ae837b 735 free_async_extent_pages(async_extent);
3e04e7f1 736
fdf8e2ea
JB		 737 if (ret == -ENOSPC) {
738 unlock_extent(io_tree, async_extent->start,
739 async_extent->start +
740 async_extent->ram_size - 1);
ce62003f
LB		 741
742 /*
743 * we need to redirty the pages if we decide to
744 * fallback to uncompressed IO, otherwise we
745 * will not submit these pages down to lower
746 * layers.
747 */
748 extent_range_redirty_for_io(inode,
749 async_extent->start,
750 async_extent->start +
751 async_extent->ram_size - 1);
752
79787eaa 753 goto retry;
fdf8e2ea 754 }
3e04e7f1 755 goto out_free;
f5a84ee3 756 }
31193213
DY		 757 btrfs_qgroup_update_reserved_bytes(root->fs_info,
758 root->root_key.objectid,
759 ins.offset, 1);
c2167754
YZ		 760 /*
761 * here we're doing allocation and writeback of the
762 * compressed pages
763 */
764 btrfs_drop_extent_cache(inode, async_extent->start,
765 async_extent->start +
766 async_extent->ram_size - 1, 0);
767
172ddd60 768 em = alloc_extent_map();
b9aa55be
LB		 769 if (!em) {
770 ret = -ENOMEM;
3e04e7f1 771 goto out_free_reserve;
b9aa55be 772 }
771ed689
CM		 773 em->start = async_extent->start;
774 em->len = async_extent->ram_size;
445a6944 775 em->orig_start = em->start;
2ab28f32
JB		 776 em->mod_start = em->start;
777 em->mod_len = em->len;
c8b97818 778
771ed689
CM		 779 em->block_start = ins.objectid;
780 em->block_len = ins.offset;
b4939680 781 em->orig_block_len = ins.offset;
cc95bef6 782 em->ram_bytes = async_extent->ram_size;
771ed689 783 em->bdev = root->fs_info->fs_devices->latest_bdev;
261507a0 784 em->compress_type = async_extent->compress_type;
771ed689
CM		 785 set_bit(EXTENT_FLAG_PINNED, &em->flags);
786 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
70c8a91c 787 em->generation = -1;
771ed689 788
d397712b 789 while (1) {
890871be 790 write_lock(&em_tree->lock);
09a2a8f9 791 ret = add_extent_mapping(em_tree, em, 1);
890871be 792 write_unlock(&em_tree->lock);
771ed689
CM		 793 if (ret != -EEXIST) {
794 free_extent_map(em);
795 break;
796 }
797 btrfs_drop_extent_cache(inode, async_extent->start,
798 async_extent->start +
799 async_extent->ram_size - 1, 0);
800 }
801
3e04e7f1
JB		 802 if (ret)
803 goto out_free_reserve;
804
261507a0
LZ		 805 ret = btrfs_add_ordered_extent_compress(inode,
806 async_extent->start,
807 ins.objectid,
808 async_extent->ram_size,
809 ins.offset,
810 BTRFS_ORDERED_COMPRESSED,
811 async_extent->compress_type);
d9f85963
FM		 812 if (ret) {
813 btrfs_drop_extent_cache(inode, async_extent->start,
814 async_extent->start +
815 async_extent->ram_size - 1, 0);
3e04e7f1 816 goto out_free_reserve;
d9f85963 817 }
771ed689 818
771ed689
CM		 819 /*
820 * clear dirty, set writeback and unlock the pages.
821 */
c2790a2e 822 extent_clear_unlock_delalloc(inode, async_extent->start,
a791e35e
CM		 823 async_extent->start +
824 async_extent->ram_size - 1,
151a41bc
JB		 825 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
826 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
c2790a2e 827 PAGE_SET_WRITEBACK);
771ed689 828 ret = btrfs_submit_compressed_write(inode,
d397712b
CM		 829 async_extent->start,
830 async_extent->ram_size,
831 ins.objectid,
832 ins.offset, async_extent->pages,
833 async_extent->nr_pages);
fce2a4e6
FM		 834 if (ret) {
835 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
836 struct page *p = async_extent->pages[0];
837 const u64 start = async_extent->start;
838 const u64 end = start + async_extent->ram_size - 1;
839
840 p->mapping = inode->i_mapping;
841 tree->ops->writepage_end_io_hook(p, start, end,
842 NULL, 0);
843 p->mapping = NULL;
844 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
845 PAGE_END_WRITEBACK |
846 PAGE_SET_ERROR);
40ae837b 847 free_async_extent_pages(async_extent);
fce2a4e6 848 }
771ed689
CM		 849 alloc_hint = ins.objectid + ins.offset;
850 kfree(async_extent);
851 cond_resched();
852 }
dec8f175 853 return;
3e04e7f1 854out_free_reserve:
e570fd27 855 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
79787eaa 856out_free:
c2790a2e 857 extent_clear_unlock_delalloc(inode, async_extent->start,
3e04e7f1
JB		 858 async_extent->start +
859 async_extent->ram_size - 1,
c2790a2e 860 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc
JB		 861 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
862 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
704de49d
FM		 863 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
864 PAGE_SET_ERROR);
40ae837b 865 free_async_extent_pages(async_extent);
79787eaa 866 kfree(async_extent);
3e04e7f1 867 goto again;
771ed689
CM		 868}
869
4b46fce2
JB		 870static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
871 u64 num_bytes)
872{
873 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
874 struct extent_map *em;
875 u64 alloc_hint = 0;
876
877 read_lock(&em_tree->lock);
878 em = search_extent_mapping(em_tree, start, num_bytes);
879 if (em) {
880 /*
881 * if block start isn't an actual block number then find the
882 * first block in this inode and use that as a hint. If that
883 * block is also bogus then just don't worry about it.
884 */
885 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
886 free_extent_map(em);
887 em = search_extent_mapping(em_tree, 0, 0);
888 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
889 alloc_hint = em->block_start;
890 if (em)
891 free_extent_map(em);
892 } else {
893 alloc_hint = em->block_start;
894 free_extent_map(em);
895 }
896 }
897 read_unlock(&em_tree->lock);
898
899 return alloc_hint;
900}
901
771ed689
CM		 902/*
903 * when extent_io.c finds a delayed allocation range in the file,
904 * the call backs end up in this code. The basic idea is to
905 * allocate extents on disk for the range, and create ordered data structs
906 * in ram to track those extents.
907 *
908 * locked_page is the page that writepage had locked already. We use
909 * it to make sure we don't do extra locks or unlocks.
910 *
911 * *page_started is set to one if we unlock locked_page and do everything
912 * required to start IO on it. It may be clean and already done with
913 * IO when we return.
914 */
00361589
JB		 915static noinline int cow_file_range(struct inode *inode,
916 struct page *locked_page,
917 u64 start, u64 end, int *page_started,
918 unsigned long *nr_written,
919 int unlock)
771ed689 920{
00361589 921 struct btrfs_root *root = BTRFS_I(inode)->root;
771ed689
CM		 922 u64 alloc_hint = 0;
923 u64 num_bytes;
924 unsigned long ram_size;
925 u64 disk_num_bytes;
926 u64 cur_alloc_size;
927 u64 blocksize = root->sectorsize;
771ed689
CM		 928 struct btrfs_key ins;
929 struct extent_map *em;
930 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
931 int ret = 0;
932
02ecd2c2
JB		 933 if (btrfs_is_free_space_inode(inode)) {
934 WARN_ON_ONCE(1);
29bce2f3
JB		 935 ret = -EINVAL;
936 goto out_unlock;
02ecd2c2 937 }
771ed689 938
fda2832f 939 num_bytes = ALIGN(end - start + 1, blocksize);
771ed689
CM		 940 num_bytes = max(blocksize, num_bytes);
941 disk_num_bytes = num_bytes;
771ed689 942
4cb5300b 943 /* if this is a small write inside eof, kick off defrag */
4cb13e5d
LB		 944 if (num_bytes < 64 * 1024 &&
945 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
00361589 946 btrfs_add_inode_defrag(NULL, inode);
4cb5300b 947
771ed689
CM		 948 if (start == 0) {
949 /* lets try to make an inline extent */
00361589
JB		 950 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
951 NULL);
771ed689 952 if (ret == 0) {
c2790a2e
JB		 953 extent_clear_unlock_delalloc(inode, start, end, NULL,
954 EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc 955 EXTENT_DEFRAG, PAGE_UNLOCK |
c2790a2e
JB		 956 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
957 PAGE_END_WRITEBACK);
c2167754 958
771ed689
CM		 959 *nr_written = *nr_written +
960 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
961 *page_started = 1;
771ed689 962 goto out;
79787eaa 963 } else if (ret < 0) {
79787eaa 964 goto out_unlock;
771ed689
CM		 965 }
966 }
967
968 BUG_ON(disk_num_bytes >
6c41761f 969 btrfs_super_total_bytes(root->fs_info->super_copy));
771ed689 970
4b46fce2 971 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
771ed689
CM		 972 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
973
d397712b 974 while (disk_num_bytes > 0) {
a791e35e
CM		 975 unsigned long op;
976
287a0ab9 977 cur_alloc_size = disk_num_bytes;
00361589 978 ret = btrfs_reserve_extent(root, cur_alloc_size,
771ed689 979 root->sectorsize, 0, alloc_hint,
e570fd27 980 &ins, 1, 1);
00361589 981 if (ret < 0)
79787eaa 982 goto out_unlock;
d397712b 983
31193213
DY		 984 btrfs_qgroup_update_reserved_bytes(root->fs_info,
985 root->root_key.objectid,
986 ins.offset, 1);
987
172ddd60 988 em = alloc_extent_map();
b9aa55be
LB		 989 if (!em) {
990 ret = -ENOMEM;
ace68bac 991 goto out_reserve;
b9aa55be 992 }
e6dcd2dc 993 em->start = start;
445a6944 994 em->orig_start = em->start;
771ed689
CM		 995 ram_size = ins.offset;
996 em->len = ins.offset;
2ab28f32
JB		 997 em->mod_start = em->start;
998 em->mod_len = em->len;
c8b97818 999
e6dcd2dc 1000 em->block_start = ins.objectid;
c8b97818 1001 em->block_len = ins.offset;
b4939680 1002 em->orig_block_len = ins.offset;
cc95bef6 1003 em->ram_bytes = ram_size;
e6dcd2dc 1004 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 1005 set_bit(EXTENT_FLAG_PINNED, &em->flags);
70c8a91c 1006 em->generation = -1;
c8b97818 1007
d397712b 1008 while (1) {
890871be 1009 write_lock(&em_tree->lock);
09a2a8f9 1010 ret = add_extent_mapping(em_tree, em, 1);
890871be 1011 write_unlock(&em_tree->lock);
e6dcd2dc
CM		 1012 if (ret != -EEXIST) {
1013 free_extent_map(em);
1014 break;
1015 }
1016 btrfs_drop_extent_cache(inode, start,
c8b97818 1017 start + ram_size - 1, 0);
e6dcd2dc 1018 }
ace68bac
LB		 1019 if (ret)
1020 goto out_reserve;
e6dcd2dc 1021
98d20f67 1022 cur_alloc_size = ins.offset;
e6dcd2dc 1023 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771ed689 1024 ram_size, cur_alloc_size, 0);
ace68bac 1025 if (ret)
d9f85963 1026 goto out_drop_extent_cache;
c8b97818 1027
17d217fe
YZ		 1028 if (root->root_key.objectid ==
1029 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1030 ret = btrfs_reloc_clone_csums(inode, start,
1031 cur_alloc_size);
00361589 1032 if (ret)
d9f85963 1033 goto out_drop_extent_cache;
17d217fe
YZ		 1034 }
1035
d397712b 1036 if (disk_num_bytes < cur_alloc_size)
3b951516 1037 break;
d397712b 1038
c8b97818
CM		 1039 /* we're not doing compressed IO, don't unlock the first
1040 * page (which the caller expects to stay locked), don't
1041 * clear any dirty bits and don't set any writeback bits
8b62b72b
CM		 1042 *
1043 * Do set the Private2 bit so we know this page was properly
1044 * setup for writepage
c8b97818 1045 */
c2790a2e
JB		 1046 op = unlock ? PAGE_UNLOCK : 0;
1047 op |= PAGE_SET_PRIVATE2;
a791e35e 1048
c2790a2e
JB		 1049 extent_clear_unlock_delalloc(inode, start,
1050 start + ram_size - 1, locked_page,
1051 EXTENT_LOCKED | EXTENT_DELALLOC,
1052 op);
c8b97818 1053 disk_num_bytes -= cur_alloc_size;
c59f8951
CM		 1054 num_bytes -= cur_alloc_size;
1055 alloc_hint = ins.objectid + ins.offset;
1056 start += cur_alloc_size;
b888db2b 1057 }
79787eaa 1058out:
be20aa9d 1059 return ret;
b7d5b0a8 1060
d9f85963
FM		 1061out_drop_extent_cache:
1062 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
ace68bac 1063out_reserve:
e570fd27 1064 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
79787eaa 1065out_unlock:
c2790a2e 1066 extent_clear_unlock_delalloc(inode, start, end, locked_page,
151a41bc
JB		 1067 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1068 EXTENT_DELALLOC | EXTENT_DEFRAG,
1069 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1070 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
79787eaa 1071 goto out;
771ed689 1072}
c8b97818 1073
771ed689
CM		 1074/*
1075 * work queue call back to started compression on a file and pages
1076 */
1077static noinline void async_cow_start(struct btrfs_work *work)
1078{
1079 struct async_cow *async_cow;
1080 int num_added = 0;
1081 async_cow = container_of(work, struct async_cow, work);
1082
1083 compress_file_range(async_cow->inode, async_cow->locked_page,
1084 async_cow->start, async_cow->end, async_cow,
1085 &num_added);
8180ef88 1086 if (num_added == 0) {
cb77fcd8 1087 btrfs_add_delayed_iput(async_cow->inode);
771ed689 1088 async_cow->inode = NULL;
8180ef88 1089 }
771ed689
CM		 1090}
1091
1092/*
1093 * work queue call back to submit previously compressed pages
1094 */
1095static noinline void async_cow_submit(struct btrfs_work *work)
1096{
1097 struct async_cow *async_cow;
1098 struct btrfs_root *root;
1099 unsigned long nr_pages;
1100
1101 async_cow = container_of(work, struct async_cow, work);
1102
1103 root = async_cow->root;
1104 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1105 PAGE_CACHE_SHIFT;
1106
66657b31 1107 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
287082b0 1108 5 * 1024 * 1024 &&
771ed689
CM		 1109 waitqueue_active(&root->fs_info->async_submit_wait))
1110 wake_up(&root->fs_info->async_submit_wait);
1111
d397712b 1112 if (async_cow->inode)
771ed689 1113 submit_compressed_extents(async_cow->inode, async_cow);
771ed689 1114}
c8b97818 1115
771ed689
CM		 1116static noinline void async_cow_free(struct btrfs_work *work)
1117{
1118 struct async_cow *async_cow;
1119 async_cow = container_of(work, struct async_cow, work);
8180ef88 1120 if (async_cow->inode)
cb77fcd8 1121 btrfs_add_delayed_iput(async_cow->inode);
771ed689
CM		 1122 kfree(async_cow);
1123}
1124
1125static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1126 u64 start, u64 end, int *page_started,
1127 unsigned long *nr_written)
1128{
1129 struct async_cow *async_cow;
1130 struct btrfs_root *root = BTRFS_I(inode)->root;
1131 unsigned long nr_pages;
1132 u64 cur_end;
287082b0 1133 int limit = 10 * 1024 * 1024;
771ed689 1134
a3429ab7
CM		 1135 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1136 1, 0, NULL, GFP_NOFS);
d397712b 1137 while (start < end) {
771ed689 1138 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
79787eaa 1139 BUG_ON(!async_cow); /* -ENOMEM */
8180ef88 1140 async_cow->inode = igrab(inode);
771ed689
CM		 1141 async_cow->root = root;
1142 async_cow->locked_page = locked_page;
1143 async_cow->start = start;
1144
f79707b0
WS		 1145 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1146 !btrfs_test_opt(root, FORCE_COMPRESS))
771ed689
CM		 1147 cur_end = end;
1148 else
1149 cur_end = min(end, start + 512 * 1024 - 1);
1150
1151 async_cow->end = cur_end;
1152 INIT_LIST_HEAD(&async_cow->extents);
1153
9e0af237
LB		 1154 btrfs_init_work(&async_cow->work,
1155 btrfs_delalloc_helper,
1156 async_cow_start, async_cow_submit,
1157 async_cow_free);
771ed689 1158
771ed689
CM		 1159 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1160 PAGE_CACHE_SHIFT;
1161 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1162
afe3d242
QW		 1163 btrfs_queue_work(root->fs_info->delalloc_workers,
1164 &async_cow->work);
771ed689
CM		 1165
1166 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1167 wait_event(root->fs_info->async_submit_wait,
1168 (atomic_read(&root->fs_info->async_delalloc_pages) <
1169 limit));
1170 }
1171
d397712b 1172 while (atomic_read(&root->fs_info->async_submit_draining) &&
771ed689
CM		 1173 atomic_read(&root->fs_info->async_delalloc_pages)) {
1174 wait_event(root->fs_info->async_submit_wait,
1175 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1176 0));
1177 }
1178
1179 *nr_written += nr_pages;
1180 start = cur_end + 1;
1181 }
1182 *page_started = 1;
1183 return 0;
be20aa9d
CM		 1184}
1185
d397712b 1186static noinline int csum_exist_in_range(struct btrfs_root *root,
17d217fe
YZ		 1187 u64 bytenr, u64 num_bytes)
1188{
1189 int ret;
1190 struct btrfs_ordered_sum *sums;
1191 LIST_HEAD(list);
1192
07d400a6 1193 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
a2de733c 1194 bytenr + num_bytes - 1, &list, 0);
17d217fe
YZ		 1195 if (ret == 0 && list_empty(&list))
1196 return 0;
1197
1198 while (!list_empty(&list)) {
1199 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1200 list_del(&sums->list);
1201 kfree(sums);
1202 }
1203 return 1;
1204}
1205
d352ac68
CM		 1206/*
1207 * when nowcow writeback call back. This checks for snapshots or COW copies
1208 * of the extents that exist in the file, and COWs the file as required.
1209 *
1210 * If no cow copies or snapshots exist, we write directly to the existing
1211 * blocks on disk
1212 */
7f366cfe
CM		 1213static noinline int run_delalloc_nocow(struct inode *inode,
1214 struct page *locked_page,
771ed689
CM		 1215 u64 start, u64 end, int *page_started, int force,
1216 unsigned long *nr_written)
be20aa9d 1217{
be20aa9d 1218 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 1219 struct btrfs_trans_handle *trans;
be20aa9d 1220 struct extent_buffer *leaf;
be20aa9d 1221 struct btrfs_path *path;
80ff3856 1222 struct btrfs_file_extent_item *fi;
be20aa9d 1223 struct btrfs_key found_key;
80ff3856
YZ		 1224 u64 cow_start;
1225 u64 cur_offset;
1226 u64 extent_end;
5d4f98a2 1227 u64 extent_offset;
80ff3856
YZ		 1228 u64 disk_bytenr;
1229 u64 num_bytes;
b4939680 1230 u64 disk_num_bytes;
cc95bef6 1231 u64 ram_bytes;
80ff3856 1232 int extent_type;
79787eaa 1233 int ret, err;
d899e052 1234 int type;
80ff3856
YZ		 1235 int nocow;
1236 int check_prev = 1;
82d5902d 1237 bool nolock;
33345d01 1238 u64 ino = btrfs_ino(inode);
be20aa9d
CM		 1239
1240 path = btrfs_alloc_path();
17ca04af 1241 if (!path) {
c2790a2e
JB		 1242 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1243 EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc
JB		 1244 EXTENT_DO_ACCOUNTING |
1245 EXTENT_DEFRAG, PAGE_UNLOCK |
c2790a2e
JB		 1246 PAGE_CLEAR_DIRTY |
1247 PAGE_SET_WRITEBACK |
1248 PAGE_END_WRITEBACK);
d8926bb3 1249 return -ENOMEM;
17ca04af 1250 }
82d5902d 1251
83eea1f1 1252 nolock = btrfs_is_free_space_inode(inode);
82d5902d
LZ		 1253
1254 if (nolock)
7a7eaa40 1255 trans = btrfs_join_transaction_nolock(root);
82d5902d 1256 else
7a7eaa40 1257 trans = btrfs_join_transaction(root);
ff5714cc 1258
79787eaa 1259 if (IS_ERR(trans)) {
c2790a2e
JB		 1260 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1261 EXTENT_LOCKED | EXTENT_DELALLOC |
151a41bc
JB		 1262 EXTENT_DO_ACCOUNTING |
1263 EXTENT_DEFRAG, PAGE_UNLOCK |
c2790a2e
JB		 1264 PAGE_CLEAR_DIRTY |
1265 PAGE_SET_WRITEBACK |
1266 PAGE_END_WRITEBACK);
79787eaa
JM		 1267 btrfs_free_path(path);
1268 return PTR_ERR(trans);
1269 }
1270
74b21075 1271 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
be20aa9d 1272
80ff3856
YZ		 1273 cow_start = (u64)-1;
1274 cur_offset = start;
1275 while (1) {
33345d01 1276 ret = btrfs_lookup_file_extent(trans, root, path, ino,
80ff3856 1277 cur_offset, 0);
d788a349 1278 if (ret < 0)
79787eaa 1279 goto error;
80ff3856
YZ		 1280 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1281 leaf = path->nodes[0];
1282 btrfs_item_key_to_cpu(leaf, &found_key,
1283 path->slots[0] - 1);
33345d01 1284 if (found_key.objectid == ino &&
80ff3856
YZ		 1285 found_key.type == BTRFS_EXTENT_DATA_KEY)
1286 path->slots[0]--;
1287 }
1288 check_prev = 0;
1289next_slot:
1290 leaf = path->nodes[0];
1291 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1292 ret = btrfs_next_leaf(root, path);
d788a349 1293 if (ret < 0)
79787eaa 1294 goto error;
80ff3856
YZ		 1295 if (ret > 0)
1296 break;
1297 leaf = path->nodes[0];
1298 }
be20aa9d 1299
80ff3856
YZ		 1300 nocow = 0;
1301 disk_bytenr = 0;
17d217fe 1302 num_bytes = 0;
80ff3856
YZ		 1303 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1304
33345d01 1305 if (found_key.objectid > ino ||
80ff3856
YZ		 1306 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1307 found_key.offset > end)
1308 break;
1309
1310 if (found_key.offset > cur_offset) {
1311 extent_end = found_key.offset;
e9061e21 1312 extent_type = 0;
80ff3856
YZ		 1313 goto out_check;
1314 }
1315
1316 fi = btrfs_item_ptr(leaf, path->slots[0],
1317 struct btrfs_file_extent_item);
1318 extent_type = btrfs_file_extent_type(leaf, fi);
1319
cc95bef6 1320 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
d899e052
YZ		 1321 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1322 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856 1323 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5d4f98a2 1324 extent_offset = btrfs_file_extent_offset(leaf, fi);
80ff3856
YZ		 1325 extent_end = found_key.offset +
1326 btrfs_file_extent_num_bytes(leaf, fi);
b4939680
JB		 1327 disk_num_bytes =
1328 btrfs_file_extent_disk_num_bytes(leaf, fi);
80ff3856
YZ		 1329 if (extent_end <= start) {
1330 path->slots[0]++;
1331 goto next_slot;
1332 }
17d217fe
YZ		 1333 if (disk_bytenr == 0)
1334 goto out_check;
80ff3856
YZ		 1335 if (btrfs_file_extent_compression(leaf, fi) ||
1336 btrfs_file_extent_encryption(leaf, fi) ||
1337 btrfs_file_extent_other_encoding(leaf, fi))
1338 goto out_check;
d899e052
YZ		 1339 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1340 goto out_check;
d2fb3437 1341 if (btrfs_extent_readonly(root, disk_bytenr))
80ff3856 1342 goto out_check;
33345d01 1343 if (btrfs_cross_ref_exist(trans, root, ino,
5d4f98a2
YZ		 1344 found_key.offset -
1345 extent_offset, disk_bytenr))
17d217fe 1346 goto out_check;
5d4f98a2 1347 disk_bytenr += extent_offset;
17d217fe
YZ		 1348 disk_bytenr += cur_offset - found_key.offset;
1349 num_bytes = min(end + 1, extent_end) - cur_offset;
e9894fd3
WS		 1350 /*
1351 * if there are pending snapshots for this root,
1352 * we fall into common COW way.
1353 */
1354 if (!nolock) {
9ea24bbe 1355 err = btrfs_start_write_no_snapshoting(root);
e9894fd3
WS		 1356 if (!err)
1357 goto out_check;
1358 }
17d217fe
YZ		 1359 /*
1360 * force cow if csum exists in the range.
1361 * this ensure that csum for a given extent are
1362 * either valid or do not exist.
1363 */
1364 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1365 goto out_check;
80ff3856
YZ		 1366 nocow = 1;
1367 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1368 extent_end = found_key.offset +
514ac8ad
CM		 1369 btrfs_file_extent_inline_len(leaf,
1370 path->slots[0], fi);
80ff3856
YZ		 1371 extent_end = ALIGN(extent_end, root->sectorsize);
1372 } else {
1373 BUG_ON(1);
1374 }
1375out_check:
1376 if (extent_end <= start) {
1377 path->slots[0]++;
e9894fd3 1378 if (!nolock && nocow)
9ea24bbe 1379 btrfs_end_write_no_snapshoting(root);
80ff3856
YZ		 1380 goto next_slot;
1381 }
1382 if (!nocow) {
1383 if (cow_start == (u64)-1)
1384 cow_start = cur_offset;
1385 cur_offset = extent_end;
1386 if (cur_offset > end)
1387 break;
1388 path->slots[0]++;
1389 goto next_slot;
7ea394f1
YZ		 1390 }
1391
b3b4aa74 1392 btrfs_release_path(path);
80ff3856 1393 if (cow_start != (u64)-1) {
00361589
JB		 1394 ret = cow_file_range(inode, locked_page,
1395 cow_start, found_key.offset - 1,
1396 page_started, nr_written, 1);
e9894fd3
WS		 1397 if (ret) {
1398 if (!nolock && nocow)
9ea24bbe 1399 btrfs_end_write_no_snapshoting(root);
79787eaa 1400 goto error;
e9894fd3 1401 }
80ff3856 1402 cow_start = (u64)-1;
7ea394f1 1403 }
80ff3856 1404
d899e052
YZ		 1405 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1406 struct extent_map *em;
1407 struct extent_map_tree *em_tree;
1408 em_tree = &BTRFS_I(inode)->extent_tree;
172ddd60 1409 em = alloc_extent_map();
79787eaa 1410 BUG_ON(!em); /* -ENOMEM */
d899e052 1411 em->start = cur_offset;
70c8a91c 1412 em->orig_start = found_key.offset - extent_offset;
d899e052
YZ		 1413 em->len = num_bytes;
1414 em->block_len = num_bytes;
1415 em->block_start = disk_bytenr;
b4939680 1416 em->orig_block_len = disk_num_bytes;
cc95bef6 1417 em->ram_bytes = ram_bytes;
d899e052 1418 em->bdev = root->fs_info->fs_devices->latest_bdev;
2ab28f32
JB		 1419 em->mod_start = em->start;
1420 em->mod_len = em->len;
d899e052 1421 set_bit(EXTENT_FLAG_PINNED, &em->flags);
b11e234d 1422 set_bit(EXTENT_FLAG_FILLING, &em->flags);
70c8a91c 1423 em->generation = -1;
d899e052 1424 while (1) {
890871be 1425 write_lock(&em_tree->lock);
09a2a8f9 1426 ret = add_extent_mapping(em_tree, em, 1);
890871be 1427 write_unlock(&em_tree->lock);
d899e052
YZ		 1428 if (ret != -EEXIST) {
1429 free_extent_map(em);
1430 break;
1431 }
1432 btrfs_drop_extent_cache(inode, em->start,
1433 em->start + em->len - 1, 0);
1434 }
1435 type = BTRFS_ORDERED_PREALLOC;
1436 } else {
1437 type = BTRFS_ORDERED_NOCOW;
1438 }
80ff3856
YZ		 1439
1440 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052 1441 num_bytes, num_bytes, type);
79787eaa 1442 BUG_ON(ret); /* -ENOMEM */
771ed689 1443
efa56464
YZ		 1444 if (root->root_key.objectid ==
1445 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1446 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1447 num_bytes);
e9894fd3
WS		 1448 if (ret) {
1449 if (!nolock && nocow)
9ea24bbe 1450 btrfs_end_write_no_snapshoting(root);
79787eaa 1451 goto error;
e9894fd3 1452 }
efa56464
YZ		 1453 }
1454
c2790a2e
JB		 1455 extent_clear_unlock_delalloc(inode, cur_offset,
1456 cur_offset + num_bytes - 1,
1457 locked_page, EXTENT_LOCKED |
1458 EXTENT_DELALLOC, PAGE_UNLOCK |
1459 PAGE_SET_PRIVATE2);
e9894fd3 1460 if (!nolock && nocow)
9ea24bbe 1461 btrfs_end_write_no_snapshoting(root);
80ff3856
YZ		 1462 cur_offset = extent_end;
1463 if (cur_offset > end)
1464 break;
be20aa9d 1465 }
b3b4aa74 1466 btrfs_release_path(path);
80ff3856 1467
17ca04af 1468 if (cur_offset <= end && cow_start == (u64)-1) {
80ff3856 1469 cow_start = cur_offset;
17ca04af
JB		 1470 cur_offset = end;
1471 }
1472
80ff3856 1473 if (cow_start != (u64)-1) {
00361589
JB		 1474 ret = cow_file_range(inode, locked_page, cow_start, end,
1475 page_started, nr_written, 1);
d788a349 1476 if (ret)
79787eaa 1477 goto error;
80ff3856
YZ		 1478 }
1479
79787eaa 1480error:
a698d075 1481 err = btrfs_end_transaction(trans, root);
79787eaa
JM		 1482 if (!ret)
1483 ret = err;
1484
17ca04af 1485 if (ret && cur_offset < end)
c2790a2e
JB		 1486 extent_clear_unlock_delalloc(inode, cur_offset, end,
1487 locked_page, EXTENT_LOCKED |
151a41bc
JB		 1488 EXTENT_DELALLOC | EXTENT_DEFRAG |
1489 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1490 PAGE_CLEAR_DIRTY |
c2790a2e
JB		 1491 PAGE_SET_WRITEBACK |
1492 PAGE_END_WRITEBACK);
7ea394f1 1493 btrfs_free_path(path);
79787eaa 1494 return ret;
be20aa9d
CM		 1495}
1496
47059d93
WS		 1497static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1498{
1499
1500 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1501 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1502 return 0;
1503
1504 /*
1505 * @defrag_bytes is a hint value, no spinlock held here,
1506 * if is not zero, it means the file is defragging.
1507 * Force cow if given extent needs to be defragged.
1508 */
1509 if (BTRFS_I(inode)->defrag_bytes &&
1510 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1511 EXTENT_DEFRAG, 0, NULL))
1512 return 1;
1513
1514 return 0;
1515}
1516
d352ac68
CM		 1517/*
1518 * extent_io.c call back to do delayed allocation processing
1519 */
c8b97818 1520static int run_delalloc_range(struct inode *inode, struct page *locked_page,
771ed689
CM		 1521 u64 start, u64 end, int *page_started,
1522 unsigned long *nr_written)
be20aa9d 1523{
be20aa9d 1524 int ret;
47059d93 1525 int force_cow = need_force_cow(inode, start, end);
a2135011 1526
47059d93 1527 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
c8b97818 1528 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1529 page_started, 1, nr_written);
47059d93 1530 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
d899e052 1531 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1532 page_started, 0, nr_written);
7816030e 1533 } else if (!inode_need_compress(inode)) {
7f366cfe
CM		 1534 ret = cow_file_range(inode, locked_page, start, end,
1535 page_started, nr_written, 1);
7ddf5a42
JB		 1536 } else {
1537 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1538 &BTRFS_I(inode)->runtime_flags);
771ed689 1539 ret = cow_file_range_async(inode, locked_page, start, end,
d397712b 1540 page_started, nr_written);
7ddf5a42 1541 }
b888db2b
CM		 1542 return ret;
1543}
1544
1bf85046
JM		 1545static void btrfs_split_extent_hook(struct inode *inode,
1546 struct extent_state *orig, u64 split)
9ed74f2d 1547{
dcab6a3b
JB		 1548 u64 size;
1549
0ca1f7ce 1550 /* not delalloc, ignore it */
9ed74f2d 1551 if (!(orig->state & EXTENT_DELALLOC))
1bf85046 1552 return;
9ed74f2d 1553
dcab6a3b
JB		 1554 size = orig->end - orig->start + 1;
1555 if (size > BTRFS_MAX_EXTENT_SIZE) {
1556 u64 num_extents;
1557 u64 new_size;
1558
1559 /*
ba117213
JB		 1560 * See the explanation in btrfs_merge_extent_hook, the same
1561 * applies here, just in reverse.
dcab6a3b
JB		 1562 */
1563 new_size = orig->end - split + 1;
ba117213 1564 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
dcab6a3b 1565 BTRFS_MAX_EXTENT_SIZE);
ba117213
JB		 1566 new_size = split - orig->start;
1567 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1568 BTRFS_MAX_EXTENT_SIZE);
1569 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1570 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
dcab6a3b
JB		 1571 return;
1572 }
1573
9e0baf60
JB		 1574 spin_lock(&BTRFS_I(inode)->lock);
1575 BTRFS_I(inode)->outstanding_extents++;
1576 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1577}
1578
1579/*
1580 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1581 * extents so we can keep track of new extents that are just merged onto old
1582 * extents, such as when we are doing sequential writes, so we can properly
1583 * account for the metadata space we'll need.
1584 */
1bf85046
JM		 1585static void btrfs_merge_extent_hook(struct inode *inode,
1586 struct extent_state *new,
1587 struct extent_state *other)
9ed74f2d 1588{
dcab6a3b
JB		 1589 u64 new_size, old_size;
1590 u64 num_extents;
1591
9ed74f2d
JB		 1592 /* not delalloc, ignore it */
1593 if (!(other->state & EXTENT_DELALLOC))
1bf85046 1594 return;
9ed74f2d 1595
8461a3de
JB		 1596 if (new->start > other->start)
1597 new_size = new->end - other->start + 1;
1598 else
1599 new_size = other->end - new->start + 1;
dcab6a3b
JB		 1600
1601 /* we're not bigger than the max, unreserve the space and go */
1602 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1603 spin_lock(&BTRFS_I(inode)->lock);
1604 BTRFS_I(inode)->outstanding_extents--;
1605 spin_unlock(&BTRFS_I(inode)->lock);
1606 return;
1607 }
1608
1609 /*
ba117213
JB		 1610 * We have to add up either side to figure out how many extents were
1611 * accounted for before we merged into one big extent. If the number of
1612 * extents we accounted for is <= the amount we need for the new range
1613 * then we can return, otherwise drop. Think of it like this
1614 *
1615 * [ 4k][MAX_SIZE]
1616 *
1617 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1618 * need 2 outstanding extents, on one side we have 1 and the other side
1619 * we have 1 so they are == and we can return. But in this case
1620 *
1621 * [MAX_SIZE+4k][MAX_SIZE+4k]
1622 *
1623 * Each range on their own accounts for 2 extents, but merged together
1624 * they are only 3 extents worth of accounting, so we need to drop in
1625 * this case.
dcab6a3b 1626 */
ba117213 1627 old_size = other->end - other->start + 1;
dcab6a3b
JB		 1628 num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1629 BTRFS_MAX_EXTENT_SIZE);
ba117213
JB		 1630 old_size = new->end - new->start + 1;
1631 num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1632 BTRFS_MAX_EXTENT_SIZE);
1633
dcab6a3b 1634 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
ba117213 1635 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
dcab6a3b
JB		 1636 return;
1637
9e0baf60
JB		 1638 spin_lock(&BTRFS_I(inode)->lock);
1639 BTRFS_I(inode)->outstanding_extents--;
1640 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1641}
1642
eb73c1b7
MX		 1643static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1644 struct inode *inode)
1645{
1646 spin_lock(&root->delalloc_lock);
1647 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1648 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1649 &root->delalloc_inodes);
1650 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1651 &BTRFS_I(inode)->runtime_flags);
1652 root->nr_delalloc_inodes++;
1653 if (root->nr_delalloc_inodes == 1) {
1654 spin_lock(&root->fs_info->delalloc_root_lock);
1655 BUG_ON(!list_empty(&root->delalloc_root));
1656 list_add_tail(&root->delalloc_root,
1657 &root->fs_info->delalloc_roots);
1658 spin_unlock(&root->fs_info->delalloc_root_lock);
1659 }
1660 }
1661 spin_unlock(&root->delalloc_lock);
1662}
1663
1664static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1665 struct inode *inode)
1666{
1667 spin_lock(&root->delalloc_lock);
1668 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1669 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1670 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1671 &BTRFS_I(inode)->runtime_flags);
1672 root->nr_delalloc_inodes--;
1673 if (!root->nr_delalloc_inodes) {
1674 spin_lock(&root->fs_info->delalloc_root_lock);
1675 BUG_ON(list_empty(&root->delalloc_root));
1676 list_del_init(&root->delalloc_root);
1677 spin_unlock(&root->fs_info->delalloc_root_lock);
1678 }
1679 }
1680 spin_unlock(&root->delalloc_lock);
1681}
1682
d352ac68
CM		 1683/*
1684 * extent_io.c set_bit_hook, used to track delayed allocation
1685 * bytes in this file, and to maintain the list of inodes that
1686 * have pending delalloc work to be done.
1687 */
1bf85046 1688static void btrfs_set_bit_hook(struct inode *inode,
9ee49a04 1689 struct extent_state *state, unsigned *bits)
291d673e 1690{
9ed74f2d 1691
47059d93
WS		 1692 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1693 WARN_ON(1);
75eff68e
CM		 1694 /*
1695 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1696 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1697 * bit, which is only set or cleared with irqs on
1698 */
0ca1f7ce 1699 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1700 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1701 u64 len = state->end + 1 - state->start;
83eea1f1 1702 bool do_list = !btrfs_is_free_space_inode(inode);
9ed74f2d 1703
9e0baf60 1704 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1705 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1706 } else {
1707 spin_lock(&BTRFS_I(inode)->lock);
1708 BTRFS_I(inode)->outstanding_extents++;
1709 spin_unlock(&BTRFS_I(inode)->lock);
1710 }
287a0ab9 1711
6a3891c5
JB		 1712 /* For sanity tests */
1713 if (btrfs_test_is_dummy_root(root))
1714 return;
1715
963d678b
MX		 1716 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1717 root->fs_info->delalloc_batch);
df0af1a5 1718 spin_lock(&BTRFS_I(inode)->lock);
0ca1f7ce 1719 BTRFS_I(inode)->delalloc_bytes += len;
47059d93
WS		 1720 if (*bits & EXTENT_DEFRAG)
1721 BTRFS_I(inode)->defrag_bytes += len;
df0af1a5 1722 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
eb73c1b7
MX		 1723 &BTRFS_I(inode)->runtime_flags))
1724 btrfs_add_delalloc_inodes(root, inode);
df0af1a5 1725 spin_unlock(&BTRFS_I(inode)->lock);
291d673e 1726 }
291d673e
CM		 1727}
1728
d352ac68
CM		 1729/*
1730 * extent_io.c clear_bit_hook, see set_bit_hook for why
1731 */
1bf85046 1732static void btrfs_clear_bit_hook(struct inode *inode,
41074888 1733 struct extent_state *state,
9ee49a04 1734 unsigned *bits)
291d673e 1735{
47059d93 1736 u64 len = state->end + 1 - state->start;
dcab6a3b
JB		 1737 u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1738 BTRFS_MAX_EXTENT_SIZE);
47059d93
WS		 1739
1740 spin_lock(&BTRFS_I(inode)->lock);
1741 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1742 BTRFS_I(inode)->defrag_bytes -= len;
1743 spin_unlock(&BTRFS_I(inode)->lock);
1744
75eff68e
CM		 1745 /*
1746 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1747 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1748 * bit, which is only set or cleared with irqs on
1749 */
0ca1f7ce 1750 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1751 struct btrfs_root *root = BTRFS_I(inode)->root;
83eea1f1 1752 bool do_list = !btrfs_is_free_space_inode(inode);
bcbfce8a 1753
9e0baf60 1754 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1755 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1756 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1757 spin_lock(&BTRFS_I(inode)->lock);
dcab6a3b 1758 BTRFS_I(inode)->outstanding_extents -= num_extents;
9e0baf60
JB		 1759 spin_unlock(&BTRFS_I(inode)->lock);
1760 }
0ca1f7ce 1761
b6d08f06
JB		 1762 /*
1763 * We don't reserve metadata space for space cache inodes so we
1764 * don't need to call dellalloc_release_metadata if there is an
1765 * error.
1766 */
1767 if (*bits & EXTENT_DO_ACCOUNTING &&
1768 root != root->fs_info->tree_root)
0ca1f7ce
YZ		 1769 btrfs_delalloc_release_metadata(inode, len);
1770
6a3891c5
JB		 1771 /* For sanity tests. */
1772 if (btrfs_test_is_dummy_root(root))
1773 return;
1774
0cb59c99 1775 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
7ee9e440 1776 && do_list && !(state->state & EXTENT_NORESERVE))
0ca1f7ce 1777 btrfs_free_reserved_data_space(inode, len);
9ed74f2d 1778
963d678b
MX		 1779 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1780 root->fs_info->delalloc_batch);
df0af1a5 1781 spin_lock(&BTRFS_I(inode)->lock);
0ca1f7ce 1782 BTRFS_I(inode)->delalloc_bytes -= len;
0cb59c99 1783 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
df0af1a5 1784 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
eb73c1b7
MX		 1785 &BTRFS_I(inode)->runtime_flags))
1786 btrfs_del_delalloc_inode(root, inode);
df0af1a5 1787 spin_unlock(&BTRFS_I(inode)->lock);
291d673e 1788 }
291d673e
CM		 1789}
1790
d352ac68
CM		 1791/*
1792 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1793 * we don't create bios that span stripes or chunks
1794 */
64a16701 1795int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
c8b97818
CM		 1796 size_t size, struct bio *bio,
1797 unsigned long bio_flags)
239b14b3
CM		 1798{
1799 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
4f024f37 1800 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
239b14b3
CM		 1801 u64 length = 0;
1802 u64 map_length;
239b14b3
CM		 1803 int ret;
1804
771ed689
CM		 1805 if (bio_flags & EXTENT_BIO_COMPRESSED)
1806 return 0;
1807
4f024f37 1808 length = bio->bi_iter.bi_size;
239b14b3 1809 map_length = length;
64a16701 1810 ret = btrfs_map_block(root->fs_info, rw, logical,
f188591e 1811 &map_length, NULL, 0);
3ec706c8 1812 /* Will always return 0 with map_multi == NULL */
3444a972 1813 BUG_ON(ret < 0);
d397712b 1814 if (map_length < length + size)
239b14b3 1815 return 1;
3444a972 1816 return 0;
239b14b3
CM		 1817}
1818
d352ac68
CM		 1819/*
1820 * in order to insert checksums into the metadata in large chunks,
1821 * we wait until bio submission time. All the pages in the bio are
1822 * checksummed and sums are attached onto the ordered extent record.
1823 *
1824 * At IO completion time the cums attached on the ordered extent record
1825 * are inserted into the btree
1826 */
d397712b
CM		 1827static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1828 struct bio *bio, int mirror_num,
eaf25d93
CM		 1829 unsigned long bio_flags,
1830 u64 bio_offset)
065631f6 1831{
065631f6 1832 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 1833 int ret = 0;
e015640f 1834
d20f7043 1835 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
79787eaa 1836 BUG_ON(ret); /* -ENOMEM */
4a69a410
CM		 1837 return 0;
1838}
e015640f 1839
4a69a410
CM		 1840/*
1841 * in order to insert checksums into the metadata in large chunks,
1842 * we wait until bio submission time. All the pages in the bio are
1843 * checksummed and sums are attached onto the ordered extent record.
1844 *
1845 * At IO completion time the cums attached on the ordered extent record
1846 * are inserted into the btree
1847 */
b2950863 1848static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1849 int mirror_num, unsigned long bio_flags,
1850 u64 bio_offset)
4a69a410
CM		 1851{
1852 struct btrfs_root *root = BTRFS_I(inode)->root;
61891923
SB		 1853 int ret;
1854
1855 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1856 if (ret)
1857 bio_endio(bio, ret);
1858 return ret;
44b8bd7e
CM		 1859}
1860
d352ac68 1861/*
cad321ad
CM		 1862 * extent_io.c submission hook. This does the right thing for csum calculation
1863 * on write, or reading the csums from the tree before a read
d352ac68 1864 */
b2950863 1865static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1866 int mirror_num, unsigned long bio_flags,
1867 u64 bio_offset)
44b8bd7e
CM		 1868{
1869 struct btrfs_root *root = BTRFS_I(inode)->root;
1870 int ret = 0;
19b9bdb0 1871 int skip_sum;
0417341e 1872 int metadata = 0;
b812ce28 1873 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
44b8bd7e 1874
6cbff00f 1875 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
cad321ad 1876
83eea1f1 1877 if (btrfs_is_free_space_inode(inode))
0417341e
JM		 1878 metadata = 2;
1879
7b6d91da 1880 if (!(rw & REQ_WRITE)) {
5fd02043
JB		 1881 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1882 if (ret)
61891923 1883 goto out;
5fd02043 1884
d20f7043 1885 if (bio_flags & EXTENT_BIO_COMPRESSED) {
61891923
SB		 1886 ret = btrfs_submit_compressed_read(inode, bio,
1887 mirror_num,
1888 bio_flags);
1889 goto out;
c2db1073
TI		 1890 } else if (!skip_sum) {
1891 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1892 if (ret)
61891923 1893 goto out;
c2db1073 1894 }
4d1b5fb4 1895 goto mapit;
b812ce28 1896 } else if (async && !skip_sum) {
17d217fe
YZ		 1897 /* csum items have already been cloned */
1898 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1899 goto mapit;
19b9bdb0 1900 /* we're doing a write, do the async checksumming */
61891923 1901 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 1902 inode, rw, bio, mirror_num,
eaf25d93
CM		 1903 bio_flags, bio_offset,
1904 __btrfs_submit_bio_start,
4a69a410 1905 __btrfs_submit_bio_done);
61891923 1906 goto out;
b812ce28
JB		 1907 } else if (!skip_sum) {
1908 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1909 if (ret)
1910 goto out;
19b9bdb0
CM		 1911 }
1912
0b86a832 1913mapit:
61891923
SB		 1914 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1915
1916out:
1917 if (ret < 0)
1918 bio_endio(bio, ret);
1919 return ret;
065631f6 1920}
6885f308 1921
d352ac68
CM		 1922/*
1923 * given a list of ordered sums record them in the inode. This happens
1924 * at IO completion time based on sums calculated at bio submission time.
1925 */
ba1da2f4 1926static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM		 1927 struct inode *inode, u64 file_offset,
1928 struct list_head *list)
1929{
e6dcd2dc
CM		 1930 struct btrfs_ordered_sum *sum;
1931
c6e30871 1932 list_for_each_entry(sum, list, list) {
39847c4d 1933 trans->adding_csums = 1;
d20f7043
CM		 1934 btrfs_csum_file_blocks(trans,
1935 BTRFS_I(inode)->root->fs_info->csum_root, sum);
39847c4d 1936 trans->adding_csums = 0;
e6dcd2dc
CM		 1937 }
1938 return 0;
1939}
1940
2ac55d41
JB		 1941int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1942 struct extent_state **cached_state)
ea8c2819 1943{
6c1500f2 1944 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
ea8c2819 1945 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2ac55d41 1946 cached_state, GFP_NOFS);
ea8c2819
CM		 1947}
1948
d352ac68 1949/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM		 1950struct btrfs_writepage_fixup {
1951 struct page *page;
1952 struct btrfs_work work;
1953};
1954
b2950863 1955static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
247e743c
CM		 1956{
1957 struct btrfs_writepage_fixup *fixup;
1958 struct btrfs_ordered_extent *ordered;
2ac55d41 1959 struct extent_state *cached_state = NULL;
247e743c
CM		 1960 struct page *page;
1961 struct inode *inode;
1962 u64 page_start;
1963 u64 page_end;
87826df0 1964 int ret;
247e743c
CM		 1965
1966 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1967 page = fixup->page;
4a096752 1968again:
247e743c
CM		 1969 lock_page(page);
1970 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1971 ClearPageChecked(page);
1972 goto out_page;
1973 }
1974
1975 inode = page->mapping->host;
1976 page_start = page_offset(page);
1977 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1978
2ac55d41 1979 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
d0082371 1980 &cached_state);
4a096752
CM		 1981
1982 /* already ordered? We're done */
8b62b72b 1983 if (PagePrivate2(page))
247e743c 1984 goto out;
4a096752
CM		 1985
1986 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1987 if (ordered) {
2ac55d41
JB		 1988 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1989 page_end, &cached_state, GFP_NOFS);
4a096752
CM		 1990 unlock_page(page);
1991 btrfs_start_ordered_extent(inode, ordered, 1);
87826df0 1992 btrfs_put_ordered_extent(ordered);
4a096752
CM		 1993 goto again;
1994 }
247e743c 1995
87826df0
JM		 1996 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1997 if (ret) {
1998 mapping_set_error(page->mapping, ret);
1999 end_extent_writepage(page, ret, page_start, page_end);
2000 ClearPageChecked(page);
2001 goto out;
2002 }
2003
2ac55d41 2004 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
247e743c 2005 ClearPageChecked(page);
87826df0 2006 set_page_dirty(page);
247e743c 2007out:
2ac55d41
JB		 2008 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2009 &cached_state, GFP_NOFS);
247e743c
CM		 2010out_page:
2011 unlock_page(page);
2012 page_cache_release(page);
b897abec 2013 kfree(fixup);
247e743c
CM		 2014}
2015
2016/*
2017 * There are a few paths in the higher layers of the kernel that directly
2018 * set the page dirty bit without asking the filesystem if it is a
2019 * good idea. This causes problems because we want to make sure COW
2020 * properly happens and the data=ordered rules are followed.
2021 *
c8b97818 2022 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 2023 * hasn't been properly setup for IO. We kick off an async process
2024 * to fix it up. The async helper will wait for ordered extents, set
2025 * the delalloc bit and make it safe to write the page.
2026 */
b2950863 2027static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
247e743c
CM		 2028{
2029 struct inode *inode = page->mapping->host;
2030 struct btrfs_writepage_fixup *fixup;
2031 struct btrfs_root *root = BTRFS_I(inode)->root;
247e743c 2032
8b62b72b
CM		 2033 /* this page is properly in the ordered list */
2034 if (TestClearPagePrivate2(page))
247e743c
CM		 2035 return 0;
2036
2037 if (PageChecked(page))
2038 return -EAGAIN;
2039
2040 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2041 if (!fixup)
2042 return -EAGAIN;
f421950f 2043
247e743c
CM		 2044 SetPageChecked(page);
2045 page_cache_get(page);
9e0af237
LB		 2046 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2047 btrfs_writepage_fixup_worker, NULL, NULL);
247e743c 2048 fixup->page = page;
dc6e3209 2049 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
87826df0 2050 return -EBUSY;
247e743c
CM		 2051}
2052
d899e052
YZ		 2053static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2054 struct inode *inode, u64 file_pos,
2055 u64 disk_bytenr, u64 disk_num_bytes,
2056 u64 num_bytes, u64 ram_bytes,
2057 u8 compression, u8 encryption,
2058 u16 other_encoding, int extent_type)
2059{
2060 struct btrfs_root *root = BTRFS_I(inode)->root;
2061 struct btrfs_file_extent_item *fi;
2062 struct btrfs_path *path;
2063 struct extent_buffer *leaf;
2064 struct btrfs_key ins;
1acae57b 2065 int extent_inserted = 0;
d899e052
YZ		 2066 int ret;
2067
2068 path = btrfs_alloc_path();
d8926bb3
MF		 2069 if (!path)
2070 return -ENOMEM;
d899e052 2071
a1ed835e
CM		 2072 /*
2073 * we may be replacing one extent in the tree with another.
2074 * The new extent is pinned in the extent map, and we don't want
2075 * to drop it from the cache until it is completely in the btree.
2076 *
2077 * So, tell btrfs_drop_extents to leave this extent in the cache.
2078 * the caller is expected to unpin it and allow it to be merged
2079 * with the others.
2080 */
1acae57b
FDBM		 2081 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2082 file_pos + num_bytes, NULL, 0,
2083 1, sizeof(*fi), &extent_inserted);
79787eaa
JM		 2084 if (ret)
2085 goto out;
d899e052 2086
1acae57b
FDBM		 2087 if (!extent_inserted) {
2088 ins.objectid = btrfs_ino(inode);
2089 ins.offset = file_pos;
2090 ins.type = BTRFS_EXTENT_DATA_KEY;
2091
2092 path->leave_spinning = 1;
2093 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2094 sizeof(*fi));
2095 if (ret)
2096 goto out;
2097 }
d899e052
YZ		 2098 leaf = path->nodes[0];
2099 fi = btrfs_item_ptr(leaf, path->slots[0],
2100 struct btrfs_file_extent_item);
2101 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2102 btrfs_set_file_extent_type(leaf, fi, extent_type);
2103 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2104 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2105 btrfs_set_file_extent_offset(leaf, fi, 0);
2106 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2107 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2108 btrfs_set_file_extent_compression(leaf, fi, compression);
2109 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2110 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
b9473439 2111
d899e052 2112 btrfs_mark_buffer_dirty(leaf);
ce195332 2113 btrfs_release_path(path);
d899e052
YZ		 2114
2115 inode_add_bytes(inode, num_bytes);
d899e052
YZ		 2116
2117 ins.objectid = disk_bytenr;
2118 ins.offset = disk_num_bytes;
2119 ins.type = BTRFS_EXTENT_ITEM_KEY;
5d4f98a2
YZ		 2120 ret = btrfs_alloc_reserved_file_extent(trans, root,
2121 root->root_key.objectid,
33345d01 2122 btrfs_ino(inode), file_pos, &ins);
79787eaa 2123out:
d899e052 2124 btrfs_free_path(path);
b9473439 2125
79787eaa 2126 return ret;
d899e052
YZ		 2127}
2128
38c227d8
LB		 2129/* snapshot-aware defrag */
2130struct sa_defrag_extent_backref {
2131 struct rb_node node;
2132 struct old_sa_defrag_extent *old;
2133 u64 root_id;
2134 u64 inum;
2135 u64 file_pos;
2136 u64 extent_offset;
2137 u64 num_bytes;
2138 u64 generation;
2139};
2140
2141struct old_sa_defrag_extent {
2142 struct list_head list;
2143 struct new_sa_defrag_extent *new;
2144
2145 u64 extent_offset;
2146 u64 bytenr;
2147 u64 offset;
2148 u64 len;
2149 int count;
2150};
2151
2152struct new_sa_defrag_extent {
2153 struct rb_root root;
2154 struct list_head head;
2155 struct btrfs_path *path;
2156 struct inode *inode;
2157 u64 file_pos;
2158 u64 len;
2159 u64 bytenr;
2160 u64 disk_len;
2161 u8 compress_type;
2162};
2163
2164static int backref_comp(struct sa_defrag_extent_backref *b1,
2165 struct sa_defrag_extent_backref *b2)
2166{
2167 if (b1->root_id < b2->root_id)
2168 return -1;
2169 else if (b1->root_id > b2->root_id)
2170 return 1;
2171
2172 if (b1->inum < b2->inum)
2173 return -1;
2174 else if (b1->inum > b2->inum)
2175 return 1;
2176
2177 if (b1->file_pos < b2->file_pos)
2178 return -1;
2179 else if (b1->file_pos > b2->file_pos)
2180 return 1;
2181
2182 /*
2183 * [------------------------------] ===> (a range of space)
2184 * |<--->| |<---->| =============> (fs/file tree A)
2185 * |<---------------------------->| ===> (fs/file tree B)
2186 *
2187 * A range of space can refer to two file extents in one tree while
2188 * refer to only one file extent in another tree.
2189 *
2190 * So we may process a disk offset more than one time(two extents in A)
2191 * and locate at the same extent(one extent in B), then insert two same
2192 * backrefs(both refer to the extent in B).
2193 */
2194 return 0;
2195}
2196
2197static void backref_insert(struct rb_root *root,
2198 struct sa_defrag_extent_backref *backref)
2199{
2200 struct rb_node **p = &root->rb_node;
2201 struct rb_node *parent = NULL;
2202 struct sa_defrag_extent_backref *entry;
2203 int ret;
2204
2205 while (*p) {
2206 parent = *p;
2207 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2208
2209 ret = backref_comp(backref, entry);
2210 if (ret < 0)
2211 p = &(*p)->rb_left;
2212 else
2213 p = &(*p)->rb_right;
2214 }
2215
2216 rb_link_node(&backref->node, parent, p);
2217 rb_insert_color(&backref->node, root);
2218}
2219
2220/*
2221 * Note the backref might has changed, and in this case we just return 0.
2222 */
2223static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2224 void *ctx)
2225{
2226 struct btrfs_file_extent_item *extent;
2227 struct btrfs_fs_info *fs_info;
2228 struct old_sa_defrag_extent *old = ctx;
2229 struct new_sa_defrag_extent *new = old->new;
2230 struct btrfs_path *path = new->path;
2231 struct btrfs_key key;
2232 struct btrfs_root *root;
2233 struct sa_defrag_extent_backref *backref;
2234 struct extent_buffer *leaf;
2235 struct inode *inode = new->inode;
2236 int slot;
2237 int ret;
2238 u64 extent_offset;
2239 u64 num_bytes;
2240
2241 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2242 inum == btrfs_ino(inode))
2243 return 0;
2244
2245 key.objectid = root_id;
2246 key.type = BTRFS_ROOT_ITEM_KEY;
2247 key.offset = (u64)-1;
2248
2249 fs_info = BTRFS_I(inode)->root->fs_info;
2250 root = btrfs_read_fs_root_no_name(fs_info, &key);
2251 if (IS_ERR(root)) {
2252 if (PTR_ERR(root) == -ENOENT)
2253 return 0;
2254 WARN_ON(1);
2255 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2256 inum, offset, root_id);
2257 return PTR_ERR(root);
2258 }
2259
2260 key.objectid = inum;
2261 key.type = BTRFS_EXTENT_DATA_KEY;
2262 if (offset > (u64)-1 << 32)
2263 key.offset = 0;
2264 else
2265 key.offset = offset;
2266
2267 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
fae7f21c 2268 if (WARN_ON(ret < 0))
38c227d8 2269 return ret;
50f1319c 2270 ret = 0;
38c227d8
LB		 2271
2272 while (1) {
2273 cond_resched();
2274
2275 leaf = path->nodes[0];
2276 slot = path->slots[0];
2277
2278 if (slot >= btrfs_header_nritems(leaf)) {
2279 ret = btrfs_next_leaf(root, path);
2280 if (ret < 0) {
2281 goto out;
2282 } else if (ret > 0) {
2283 ret = 0;
2284 goto out;
2285 }
2286 continue;
2287 }
2288
2289 path->slots[0]++;
2290
2291 btrfs_item_key_to_cpu(leaf, &key, slot);
2292
2293 if (key.objectid > inum)
2294 goto out;
2295
2296 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2297 continue;
2298
2299 extent = btrfs_item_ptr(leaf, slot,
2300 struct btrfs_file_extent_item);
2301
2302 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2303 continue;
2304
e68afa49
LB		 2305 /*
2306 * 'offset' refers to the exact key.offset,
2307 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2308 * (key.offset - extent_offset).
2309 */
2310 if (key.offset != offset)
38c227d8
LB		 2311 continue;
2312
e68afa49 2313 extent_offset = btrfs_file_extent_offset(leaf, extent);
38c227d8 2314 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
e68afa49 2315
38c227d8
LB		 2316 if (extent_offset >= old->extent_offset + old->offset +
2317 old->len || extent_offset + num_bytes <=
2318 old->extent_offset + old->offset)
2319 continue;
38c227d8
LB		 2320 break;
2321 }
2322
2323 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2324 if (!backref) {
2325 ret = -ENOENT;
2326 goto out;
2327 }
2328
2329 backref->root_id = root_id;
2330 backref->inum = inum;
e68afa49 2331 backref->file_pos = offset;
38c227d8
LB		 2332 backref->num_bytes = num_bytes;
2333 backref->extent_offset = extent_offset;
2334 backref->generation = btrfs_file_extent_generation(leaf, extent);
2335 backref->old = old;
2336 backref_insert(&new->root, backref);
2337 old->count++;
2338out:
2339 btrfs_release_path(path);
2340 WARN_ON(ret);
2341 return ret;
2342}
2343
2344static noinline bool record_extent_backrefs(struct btrfs_path *path,
2345 struct new_sa_defrag_extent *new)
2346{
2347 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2348 struct old_sa_defrag_extent *old, *tmp;
2349 int ret;
2350
2351 new->path = path;
2352
2353 list_for_each_entry_safe(old, tmp, &new->head, list) {
e68afa49
LB		 2354 ret = iterate_inodes_from_logical(old->bytenr +
2355 old->extent_offset, fs_info,
38c227d8
LB		 2356 path, record_one_backref,
2357 old);
4724b106
JB		 2358 if (ret < 0 && ret != -ENOENT)
2359 return false;
38c227d8
LB		 2360
2361 /* no backref to be processed for this extent */
2362 if (!old->count) {
2363 list_del(&old->list);
2364 kfree(old);
2365 }
2366 }
2367
2368 if (list_empty(&new->head))
2369 return false;
2370
2371 return true;
2372}
2373
2374static int relink_is_mergable(struct extent_buffer *leaf,
2375 struct btrfs_file_extent_item *fi,
116e0024 2376 struct new_sa_defrag_extent *new)
38c227d8 2377{
116e0024 2378 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
38c227d8
LB		 2379 return 0;
2380
2381 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2382 return 0;
2383
116e0024
LB		 2384 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2385 return 0;
2386
2387 if (btrfs_file_extent_encryption(leaf, fi) ||
38c227d8
LB		 2388 btrfs_file_extent_other_encoding(leaf, fi))
2389 return 0;
2390
2391 return 1;
2392}
2393
2394/*
2395 * Note the backref might has changed, and in this case we just return 0.
2396 */
2397static noinline int relink_extent_backref(struct btrfs_path *path,
2398 struct sa_defrag_extent_backref *prev,
2399 struct sa_defrag_extent_backref *backref)
2400{
2401 struct btrfs_file_extent_item *extent;
2402 struct btrfs_file_extent_item *item;
2403 struct btrfs_ordered_extent *ordered;
2404 struct btrfs_trans_handle *trans;
2405 struct btrfs_fs_info *fs_info;
2406 struct btrfs_root *root;
2407 struct btrfs_key key;
2408 struct extent_buffer *leaf;
2409 struct old_sa_defrag_extent *old = backref->old;
2410 struct new_sa_defrag_extent *new = old->new;
2411 struct inode *src_inode = new->inode;
2412 struct inode *inode;
2413 struct extent_state *cached = NULL;
2414 int ret = 0;
2415 u64 start;
2416 u64 len;
2417 u64 lock_start;
2418 u64 lock_end;
2419 bool merge = false;
2420 int index;
2421
2422 if (prev && prev->root_id == backref->root_id &&
2423 prev->inum == backref->inum &&
2424 prev->file_pos + prev->num_bytes == backref->file_pos)
2425 merge = true;
2426
2427 /* step 1: get root */
2428 key.objectid = backref->root_id;
2429 key.type = BTRFS_ROOT_ITEM_KEY;
2430 key.offset = (u64)-1;
2431
2432 fs_info = BTRFS_I(src_inode)->root->fs_info;
2433 index = srcu_read_lock(&fs_info->subvol_srcu);
2434
2435 root = btrfs_read_fs_root_no_name(fs_info, &key);
2436 if (IS_ERR(root)) {
2437 srcu_read_unlock(&fs_info->subvol_srcu, index);
2438 if (PTR_ERR(root) == -ENOENT)
2439 return 0;
2440 return PTR_ERR(root);
2441 }
38c227d8 2442
bcbba5e6
WS		 2443 if (btrfs_root_readonly(root)) {
2444 srcu_read_unlock(&fs_info->subvol_srcu, index);
2445 return 0;
2446 }
2447
38c227d8
LB		 2448 /* step 2: get inode */
2449 key.objectid = backref->inum;
2450 key.type = BTRFS_INODE_ITEM_KEY;
2451 key.offset = 0;
2452
2453 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2454 if (IS_ERR(inode)) {
2455 srcu_read_unlock(&fs_info->subvol_srcu, index);
2456 return 0;
2457 }
2458
2459 srcu_read_unlock(&fs_info->subvol_srcu, index);
2460
2461 /* step 3: relink backref */
2462 lock_start = backref->file_pos;
2463 lock_end = backref->file_pos + backref->num_bytes - 1;
2464 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2465 0, &cached);
2466
2467 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2468 if (ordered) {
2469 btrfs_put_ordered_extent(ordered);
2470 goto out_unlock;
2471 }
2472
2473 trans = btrfs_join_transaction(root);
2474 if (IS_ERR(trans)) {
2475 ret = PTR_ERR(trans);
2476 goto out_unlock;
2477 }
2478
2479 key.objectid = backref->inum;
2480 key.type = BTRFS_EXTENT_DATA_KEY;
2481 key.offset = backref->file_pos;
2482
2483 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2484 if (ret < 0) {
2485 goto out_free_path;
2486 } else if (ret > 0) {
2487 ret = 0;
2488 goto out_free_path;
2489 }
2490
2491 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2492 struct btrfs_file_extent_item);
2493
2494 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2495 backref->generation)
2496 goto out_free_path;
2497
2498 btrfs_release_path(path);
2499
2500 start = backref->file_pos;
2501 if (backref->extent_offset < old->extent_offset + old->offset)
2502 start += old->extent_offset + old->offset -
2503 backref->extent_offset;
2504
2505 len = min(backref->extent_offset + backref->num_bytes,
2506 old->extent_offset + old->offset + old->len);
2507 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2508
2509 ret = btrfs_drop_extents(trans, root, inode, start,
2510 start + len, 1);
2511 if (ret)
2512 goto out_free_path;
2513again:
2514 key.objectid = btrfs_ino(inode);
2515 key.type = BTRFS_EXTENT_DATA_KEY;
2516 key.offset = start;
2517
a09a0a70 2518 path->leave_spinning = 1;
38c227d8
LB		 2519 if (merge) {
2520 struct btrfs_file_extent_item *fi;
2521 u64 extent_len;
2522 struct btrfs_key found_key;
2523
3c9665df 2524 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
38c227d8
LB		 2525 if (ret < 0)
2526 goto out_free_path;
2527
2528 path->slots[0]--;
2529 leaf = path->nodes[0];
2530 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2531
2532 fi = btrfs_item_ptr(leaf, path->slots[0],
2533 struct btrfs_file_extent_item);
2534 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2535
116e0024
LB		 2536 if (extent_len + found_key.offset == start &&
2537 relink_is_mergable(leaf, fi, new)) {
38c227d8
LB		 2538 btrfs_set_file_extent_num_bytes(leaf, fi,
2539 extent_len + len);
2540 btrfs_mark_buffer_dirty(leaf);
2541 inode_add_bytes(inode, len);
2542
2543 ret = 1;
2544 goto out_free_path;
2545 } else {
2546 merge = false;
2547 btrfs_release_path(path);
2548 goto again;
2549 }
2550 }
2551
2552 ret = btrfs_insert_empty_item(trans, root, path, &key,
2553 sizeof(*extent));
2554 if (ret) {
2555 btrfs_abort_transaction(trans, root, ret);
2556 goto out_free_path;
2557 }
2558
2559 leaf = path->nodes[0];
2560 item = btrfs_item_ptr(leaf, path->slots[0],
2561 struct btrfs_file_extent_item);
2562 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2563 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2564 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2565 btrfs_set_file_extent_num_bytes(leaf, item, len);
2566 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2567 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2568 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2569 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2570 btrfs_set_file_extent_encryption(leaf, item, 0);
2571 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2572
2573 btrfs_mark_buffer_dirty(leaf);
2574 inode_add_bytes(inode, len);
a09a0a70 2575 btrfs_release_path(path);
38c227d8
LB		 2576
2577 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2578 new->disk_len, 0,
2579 backref->root_id, backref->inum,
2580 new->file_pos, 0); /* start - extent_offset */
2581 if (ret) {
2582 btrfs_abort_transaction(trans, root, ret);
2583 goto out_free_path;
2584 }
2585
2586 ret = 1;
2587out_free_path:
2588 btrfs_release_path(path);
a09a0a70 2589 path->leave_spinning = 0;
38c227d8
LB		 2590 btrfs_end_transaction(trans, root);
2591out_unlock:
2592 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2593 &cached, GFP_NOFS);
2594 iput(inode);
2595 return ret;
2596}
2597
6f519564
LB		 2598static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2599{
2600 struct old_sa_defrag_extent *old, *tmp;
2601
2602 if (!new)
2603 return;
2604
2605 list_for_each_entry_safe(old, tmp, &new->head, list) {
2606 list_del(&old->list);
2607 kfree(old);
2608 }
2609 kfree(new);
2610}
2611
38c227d8
LB		 2612static void relink_file_extents(struct new_sa_defrag_extent *new)
2613{
2614 struct btrfs_path *path;
38c227d8
LB		 2615 struct sa_defrag_extent_backref *backref;
2616 struct sa_defrag_extent_backref *prev = NULL;
2617 struct inode *inode;
2618 struct btrfs_root *root;
2619 struct rb_node *node;
2620 int ret;
2621
2622 inode = new->inode;
2623 root = BTRFS_I(inode)->root;
2624
2625 path = btrfs_alloc_path();
2626 if (!path)
2627 return;
2628
2629 if (!record_extent_backrefs(path, new)) {
2630 btrfs_free_path(path);
2631 goto out;
2632 }
2633 btrfs_release_path(path);
2634
2635 while (1) {
2636 node = rb_first(&new->root);
2637 if (!node)
2638 break;
2639 rb_erase(node, &new->root);
2640
2641 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2642
2643 ret = relink_extent_backref(path, prev, backref);
2644 WARN_ON(ret < 0);
2645
2646 kfree(prev);
2647
2648 if (ret == 1)
2649 prev = backref;
2650 else
2651 prev = NULL;
2652 cond_resched();
2653 }
2654 kfree(prev);
2655
2656 btrfs_free_path(path);
38c227d8 2657out:
6f519564
LB		 2658 free_sa_defrag_extent(new);
2659
38c227d8
LB		 2660 atomic_dec(&root->fs_info->defrag_running);
2661 wake_up(&root->fs_info->transaction_wait);
38c227d8
LB		 2662}
2663
2664static struct new_sa_defrag_extent *
2665record_old_file_extents(struct inode *inode,
2666 struct btrfs_ordered_extent *ordered)
2667{
2668 struct btrfs_root *root = BTRFS_I(inode)->root;
2669 struct btrfs_path *path;
2670 struct btrfs_key key;
6f519564 2671 struct old_sa_defrag_extent *old;
38c227d8
LB		 2672 struct new_sa_defrag_extent *new;
2673 int ret;
2674
2675 new = kmalloc(sizeof(*new), GFP_NOFS);
2676 if (!new)
2677 return NULL;
2678
2679 new->inode = inode;
2680 new->file_pos = ordered->file_offset;
2681 new->len = ordered->len;
2682 new->bytenr = ordered->start;
2683 new->disk_len = ordered->disk_len;
2684 new->compress_type = ordered->compress_type;
2685 new->root = RB_ROOT;
2686 INIT_LIST_HEAD(&new->head);
2687
2688 path = btrfs_alloc_path();
2689 if (!path)
2690 goto out_kfree;
2691
2692 key.objectid = btrfs_ino(inode);
2693 key.type = BTRFS_EXTENT_DATA_KEY;
2694 key.offset = new->file_pos;
2695
2696 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2697 if (ret < 0)
2698 goto out_free_path;
2699 if (ret > 0 && path->slots[0] > 0)
2700 path->slots[0]--;
2701
2702 /* find out all the old extents for the file range */
2703 while (1) {
2704 struct btrfs_file_extent_item *extent;
2705 struct extent_buffer *l;
2706 int slot;
2707 u64 num_bytes;
2708 u64 offset;
2709 u64 end;
2710 u64 disk_bytenr;
2711 u64 extent_offset;
2712
2713 l = path->nodes[0];
2714 slot = path->slots[0];
2715
2716 if (slot >= btrfs_header_nritems(l)) {
2717 ret = btrfs_next_leaf(root, path);
2718 if (ret < 0)
6f519564 2719 goto out_free_path;
38c227d8
LB		 2720 else if (ret > 0)
2721 break;
2722 continue;
2723 }
2724
2725 btrfs_item_key_to_cpu(l, &key, slot);
2726
2727 if (key.objectid != btrfs_ino(inode))
2728 break;
2729 if (key.type != BTRFS_EXTENT_DATA_KEY)
2730 break;
2731 if (key.offset >= new->file_pos + new->len)
2732 break;
2733
2734 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2735
2736 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2737 if (key.offset + num_bytes < new->file_pos)
2738 goto next;
2739
2740 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2741 if (!disk_bytenr)
2742 goto next;
2743
2744 extent_offset = btrfs_file_extent_offset(l, extent);
2745
2746 old = kmalloc(sizeof(*old), GFP_NOFS);
2747 if (!old)
6f519564 2748 goto out_free_path;
38c227d8
LB		 2749
2750 offset = max(new->file_pos, key.offset);
2751 end = min(new->file_pos + new->len, key.offset + num_bytes);
2752
2753 old->bytenr = disk_bytenr;
2754 old->extent_offset = extent_offset;
2755 old->offset = offset - key.offset;
2756 old->len = end - offset;
2757 old->new = new;
2758 old->count = 0;
2759 list_add_tail(&old->list, &new->head);
2760next:
2761 path->slots[0]++;
2762 cond_resched();
2763 }
2764
2765 btrfs_free_path(path);
2766 atomic_inc(&root->fs_info->defrag_running);
2767
2768 return new;
2769
38c227d8
LB		 2770out_free_path:
2771 btrfs_free_path(path);
2772out_kfree:
6f519564 2773 free_sa_defrag_extent(new);
38c227d8
LB		 2774 return NULL;
2775}
2776
e570fd27
MX		 2777static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2778 u64 start, u64 len)
2779{
2780 struct btrfs_block_group_cache *cache;
2781
2782 cache = btrfs_lookup_block_group(root->fs_info, start);
2783 ASSERT(cache);
2784
2785 spin_lock(&cache->lock);
2786 cache->delalloc_bytes -= len;
2787 spin_unlock(&cache->lock);
2788
2789 btrfs_put_block_group(cache);
2790}
2791
d352ac68
CM		 2792/* as ordered data IO finishes, this gets called so we can finish
2793 * an ordered extent if the range of bytes in the file it covers are
2794 * fully written.
2795 */
5fd02043 2796static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
e6dcd2dc 2797{
5fd02043 2798 struct inode *inode = ordered_extent->inode;
e6dcd2dc 2799 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 2800 struct btrfs_trans_handle *trans = NULL;
e6dcd2dc 2801 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2ac55d41 2802 struct extent_state *cached_state = NULL;
38c227d8 2803 struct new_sa_defrag_extent *new = NULL;
261507a0 2804 int compress_type = 0;
77cef2ec
JB		 2805 int ret = 0;
2806 u64 logical_len = ordered_extent->len;
82d5902d 2807 bool nolock;
77cef2ec 2808 bool truncated = false;
e6dcd2dc 2809
83eea1f1 2810 nolock = btrfs_is_free_space_inode(inode);
0cb59c99 2811
5fd02043
JB		 2812 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2813 ret = -EIO;
2814 goto out;
2815 }
2816
f612496b
MX		 2817 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2818 ordered_extent->file_offset +
2819 ordered_extent->len - 1);
2820
77cef2ec
JB		 2821 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2822 truncated = true;
2823 logical_len = ordered_extent->truncated_len;
2824 /* Truncated the entire extent, don't bother adding */
2825 if (!logical_len)
2826 goto out;
2827 }
2828
c2167754 2829 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
79787eaa 2830 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
6c760c07
JB		 2831 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2832 if (nolock)
2833 trans = btrfs_join_transaction_nolock(root);
2834 else
2835 trans = btrfs_join_transaction(root);
2836 if (IS_ERR(trans)) {
2837 ret = PTR_ERR(trans);
2838 trans = NULL;
2839 goto out;
c2167754 2840 }
6c760c07
JB		 2841 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2842 ret = btrfs_update_inode_fallback(trans, root, inode);
2843 if (ret) /* -ENOMEM or corruption */
2844 btrfs_abort_transaction(trans, root, ret);
c2167754
YZ		 2845 goto out;
2846 }
e6dcd2dc 2847
2ac55d41
JB		 2848 lock_extent_bits(io_tree, ordered_extent->file_offset,
2849 ordered_extent->file_offset + ordered_extent->len - 1,
d0082371 2850 0, &cached_state);
e6dcd2dc 2851
38c227d8
LB		 2852 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2853 ordered_extent->file_offset + ordered_extent->len - 1,
2854 EXTENT_DEFRAG, 1, cached_state);
2855 if (ret) {
2856 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
8101c8db 2857 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
38c227d8
LB		 2858 /* the inode is shared */
2859 new = record_old_file_extents(inode, ordered_extent);
2860
2861 clear_extent_bit(io_tree, ordered_extent->file_offset,
2862 ordered_extent->file_offset + ordered_extent->len - 1,
2863 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2864 }
2865
0cb59c99 2866 if (nolock)
7a7eaa40 2867 trans = btrfs_join_transaction_nolock(root);
0cb59c99 2868 else
7a7eaa40 2869 trans = btrfs_join_transaction(root);
79787eaa
JM		 2870 if (IS_ERR(trans)) {
2871 ret = PTR_ERR(trans);
2872 trans = NULL;
2873 goto out_unlock;
2874 }
a79b7d4b 2875
0ca1f7ce 2876 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
c2167754 2877
c8b97818 2878 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
261507a0 2879 compress_type = ordered_extent->compress_type;
d899e052 2880 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
261507a0 2881 BUG_ON(compress_type);
920bbbfb 2882 ret = btrfs_mark_extent_written(trans, inode,
d899e052
YZ		 2883 ordered_extent->file_offset,
2884 ordered_extent->file_offset +
77cef2ec 2885 logical_len);
d899e052 2886 } else {
0af3d00b 2887 BUG_ON(root == root->fs_info->tree_root);
d899e052
YZ		 2888 ret = insert_reserved_file_extent(trans, inode,
2889 ordered_extent->file_offset,
2890 ordered_extent->start,
2891 ordered_extent->disk_len,
77cef2ec 2892 logical_len, logical_len,
261507a0 2893 compress_type, 0, 0,
d899e052 2894 BTRFS_FILE_EXTENT_REG);
e570fd27
MX		 2895 if (!ret)
2896 btrfs_release_delalloc_bytes(root,
2897 ordered_extent->start,
2898 ordered_extent->disk_len);
d899e052 2899 }
5dc562c5
JB		 2900 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2901 ordered_extent->file_offset, ordered_extent->len,
2902 trans->transid);
79787eaa
JM		 2903 if (ret < 0) {
2904 btrfs_abort_transaction(trans, root, ret);
5fd02043 2905 goto out_unlock;
79787eaa 2906 }
2ac55d41 2907
e6dcd2dc
CM		 2908 add_pending_csums(trans, inode, ordered_extent->file_offset,
2909 &ordered_extent->list);
2910
6c760c07
JB		 2911 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2912 ret = btrfs_update_inode_fallback(trans, root, inode);
2913 if (ret) { /* -ENOMEM or corruption */
2914 btrfs_abort_transaction(trans, root, ret);
2915 goto out_unlock;
1ef30be1
JB		 2916 }
2917 ret = 0;
5fd02043
JB		 2918out_unlock:
2919 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2920 ordered_extent->file_offset +
2921 ordered_extent->len - 1, &cached_state, GFP_NOFS);
c2167754 2922out:
5b0e95bf 2923 if (root != root->fs_info->tree_root)
0cb59c99 2924 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
a698d075
MX		 2925 if (trans)
2926 btrfs_end_transaction(trans, root);
0cb59c99 2927
77cef2ec
JB		 2928 if (ret || truncated) {
2929 u64 start, end;
2930
2931 if (truncated)
2932 start = ordered_extent->file_offset + logical_len;
2933 else
2934 start = ordered_extent->file_offset;
2935 end = ordered_extent->file_offset + ordered_extent->len - 1;
2936 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2937
2938 /* Drop the cache for the part of the extent we didn't write. */
2939 btrfs_drop_extent_cache(inode, start, end, 0);
5fd02043 2940
0bec9ef5
JB		 2941 /*
2942 * If the ordered extent had an IOERR or something else went
2943 * wrong we need to return the space for this ordered extent
77cef2ec
JB		 2944 * back to the allocator. We only free the extent in the
2945 * truncated case if we didn't write out the extent at all.
0bec9ef5 2946 */
77cef2ec
JB		 2947 if ((ret || !logical_len) &&
2948 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
0bec9ef5
JB		 2949 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2950 btrfs_free_reserved_extent(root, ordered_extent->start,
e570fd27 2951 ordered_extent->disk_len, 1);
0bec9ef5
JB		 2952 }
2953
2954
5fd02043 2955 /*
8bad3c02
LB		 2956 * This needs to be done to make sure anybody waiting knows we are done
2957 * updating everything for this ordered extent.
5fd02043
JB		 2958 */
2959 btrfs_remove_ordered_extent(inode, ordered_extent);
2960
38c227d8 2961 /* for snapshot-aware defrag */
6f519564
LB		 2962 if (new) {
2963 if (ret) {
2964 free_sa_defrag_extent(new);
2965 atomic_dec(&root->fs_info->defrag_running);
2966 } else {
2967 relink_file_extents(new);
2968 }
2969 }
38c227d8 2970
e6dcd2dc
CM		 2971 /* once for us */
2972 btrfs_put_ordered_extent(ordered_extent);
2973 /* once for the tree */
2974 btrfs_put_ordered_extent(ordered_extent);
2975
5fd02043
JB		 2976 return ret;
2977}
2978
2979static void finish_ordered_fn(struct btrfs_work *work)
2980{
2981 struct btrfs_ordered_extent *ordered_extent;
2982 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2983 btrfs_finish_ordered_io(ordered_extent);
e6dcd2dc
CM		 2984}
2985
b2950863 2986static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM		 2987 struct extent_state *state, int uptodate)
2988{
5fd02043
JB		 2989 struct inode *inode = page->mapping->host;
2990 struct btrfs_root *root = BTRFS_I(inode)->root;
2991 struct btrfs_ordered_extent *ordered_extent = NULL;
9e0af237
LB		 2992 struct btrfs_workqueue *wq;
2993 btrfs_work_func_t func;
5fd02043 2994
1abe9b8a 2995 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2996
8b62b72b 2997 ClearPagePrivate2(page);
5fd02043
JB		 2998 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2999 end - start + 1, uptodate))
3000 return 0;
3001
9e0af237
LB		 3002 if (btrfs_is_free_space_inode(inode)) {
3003 wq = root->fs_info->endio_freespace_worker;
3004 func = btrfs_freespace_write_helper;
3005 } else {
3006 wq = root->fs_info->endio_write_workers;
3007 func = btrfs_endio_write_helper;
3008 }
5fd02043 3009
9e0af237
LB		 3010 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3011 NULL);
3012 btrfs_queue_work(wq, &ordered_extent->work);
5fd02043
JB		 3013
3014 return 0;
211f90e6
CM		 3015}
3016
dc380aea
MX		 3017static int __readpage_endio_check(struct inode *inode,
3018 struct btrfs_io_bio *io_bio,
3019 int icsum, struct page *page,
3020 int pgoff, u64 start, size_t len)
3021{
3022 char *kaddr;
3023 u32 csum_expected;
3024 u32 csum = ~(u32)0;
3025 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
3026 DEFAULT_RATELIMIT_BURST);
3027
3028 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3029
3030 kaddr = kmap_atomic(page);
3031 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3032 btrfs_csum_final(csum, (char *)&csum);
3033 if (csum != csum_expected)
3034 goto zeroit;
3035
3036 kunmap_atomic(kaddr);
3037 return 0;
3038zeroit:
3039 if (__ratelimit(&_rs))
f0954c66 3040 btrfs_warn(BTRFS_I(inode)->root->fs_info,
dc380aea
MX		 3041 "csum failed ino %llu off %llu csum %u expected csum %u",
3042 btrfs_ino(inode), start, csum, csum_expected);
3043 memset(kaddr + pgoff, 1, len);
3044 flush_dcache_page(page);
3045 kunmap_atomic(kaddr);
3046 if (csum_expected == 0)
3047 return 0;
3048 return -EIO;
3049}
3050
d352ac68
CM		 3051/*
3052 * when reads are done, we need to check csums to verify the data is correct
4a54c8c1
JS		 3053 * if there's a match, we allow the bio to finish. If not, the code in
3054 * extent_io.c will try to find good copies for us.
d352ac68 3055 */
facc8a22
MX		 3056static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3057 u64 phy_offset, struct page *page,
3058 u64 start, u64 end, int mirror)
07157aac 3059{
4eee4fa4 3060 size_t offset = start - page_offset(page);
07157aac 3061 struct inode *inode = page->mapping->host;
d1310b2e 3062 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
ff79f819 3063 struct btrfs_root *root = BTRFS_I(inode)->root;
d1310b2e 3064
d20f7043
CM		 3065 if (PageChecked(page)) {
3066 ClearPageChecked(page);
dc380aea 3067 return 0;
d20f7043 3068 }
6cbff00f
CH		 3069
3070 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
dc380aea 3071 return 0;
17d217fe
YZ		 3072
3073 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
9655d298 3074 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
17d217fe
YZ		 3075 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
3076 GFP_NOFS);
b6cda9bc 3077 return 0;
17d217fe 3078 }
d20f7043 3079
facc8a22 3080 phy_offset >>= inode->i_sb->s_blocksize_bits;
dc380aea
MX		 3081 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3082 start, (size_t)(end - start + 1));
07157aac 3083}
b888db2b 3084
24bbcf04
YZ		 3085struct delayed_iput {
3086 struct list_head list;
3087 struct inode *inode;
3088};
3089
79787eaa
JM		 3090/* JDM: If this is fs-wide, why can't we add a pointer to
3091 * btrfs_inode instead and avoid the allocation? */
24bbcf04
YZ		 3092void btrfs_add_delayed_iput(struct inode *inode)
3093{
3094 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3095 struct delayed_iput *delayed;
3096
3097 if (atomic_add_unless(&inode->i_count, -1, 1))
3098 return;
3099
3100 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
3101 delayed->inode = inode;
3102
3103 spin_lock(&fs_info->delayed_iput_lock);
3104 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
3105 spin_unlock(&fs_info->delayed_iput_lock);
3106}
3107
3108void btrfs_run_delayed_iputs(struct btrfs_root *root)
3109{
3110 LIST_HEAD(list);
3111 struct btrfs_fs_info *fs_info = root->fs_info;
3112 struct delayed_iput *delayed;
3113 int empty;
3114
3115 spin_lock(&fs_info->delayed_iput_lock);
3116 empty = list_empty(&fs_info->delayed_iputs);
3117 spin_unlock(&fs_info->delayed_iput_lock);
3118 if (empty)
3119 return;
3120
d7c15171
ZL		 3121 down_read(&fs_info->delayed_iput_sem);
3122
24bbcf04
YZ		 3123 spin_lock(&fs_info->delayed_iput_lock);
3124 list_splice_init(&fs_info->delayed_iputs, &list);
3125 spin_unlock(&fs_info->delayed_iput_lock);
3126
3127 while (!list_empty(&list)) {
3128 delayed = list_entry(list.next, struct delayed_iput, list);
3129 list_del(&delayed->list);
3130 iput(delayed->inode);
3131 kfree(delayed);
3132 }
d7c15171
ZL		 3133
3134 up_read(&root->fs_info->delayed_iput_sem);
24bbcf04
YZ		 3135}
3136
d68fc57b 3137/*
42b2aa86 3138 * This is called in transaction commit time. If there are no orphan
d68fc57b
YZ		 3139 * files in the subvolume, it removes orphan item and frees block_rsv
3140 * structure.
3141 */
3142void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3143 struct btrfs_root *root)
3144{
90290e19 3145 struct btrfs_block_rsv *block_rsv;
d68fc57b
YZ		 3146 int ret;
3147
8a35d95f 3148 if (atomic_read(&root->orphan_inodes) ||
d68fc57b
YZ		 3149 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3150 return;
3151
90290e19 3152 spin_lock(&root->orphan_lock);
8a35d95f 3153 if (atomic_read(&root->orphan_inodes)) {
90290e19
JB		 3154 spin_unlock(&root->orphan_lock);
3155 return;
3156 }
3157
3158 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3159 spin_unlock(&root->orphan_lock);
3160 return;
3161 }
3162
3163 block_rsv = root->orphan_block_rsv;
3164 root->orphan_block_rsv = NULL;
3165 spin_unlock(&root->orphan_lock);
3166
27cdeb70 3167 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
d68fc57b
YZ		 3168 btrfs_root_refs(&root->root_item) > 0) {
3169 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3170 root->root_key.objectid);
4ef31a45
JB		 3171 if (ret)
3172 btrfs_abort_transaction(trans, root, ret);
3173 else
27cdeb70
MX		 3174 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3175 &root->state);
d68fc57b
YZ		 3176 }
3177
90290e19
JB		 3178 if (block_rsv) {
3179 WARN_ON(block_rsv->size > 0);
3180 btrfs_free_block_rsv(root, block_rsv);
d68fc57b
YZ		 3181 }
3182}
3183
7b128766
JB		 3184/*
3185 * This creates an orphan entry for the given inode in case something goes
3186 * wrong in the middle of an unlink/truncate.
d68fc57b
YZ		 3187 *
3188 * NOTE: caller of this function should reserve 5 units of metadata for
3189 * this function.
7b128766
JB		 3190 */
3191int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3192{
3193 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 3194 struct btrfs_block_rsv *block_rsv = NULL;
3195 int reserve = 0;
3196 int insert = 0;
3197 int ret;
7b128766 3198
d68fc57b 3199 if (!root->orphan_block_rsv) {
66d8f3dd 3200 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
b532402e
TI		 3201 if (!block_rsv)
3202 return -ENOMEM;
d68fc57b 3203 }
7b128766 3204
d68fc57b
YZ		 3205 spin_lock(&root->orphan_lock);
3206 if (!root->orphan_block_rsv) {
3207 root->orphan_block_rsv = block_rsv;
3208 } else if (block_rsv) {
3209 btrfs_free_block_rsv(root, block_rsv);
3210 block_rsv = NULL;
7b128766 3211 }
7b128766 3212
8a35d95f
JB		 3213 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3214 &BTRFS_I(inode)->runtime_flags)) {
d68fc57b
YZ		 3215#if 0
3216 /*
3217 * For proper ENOSPC handling, we should do orphan
3218 * cleanup when mounting. But this introduces backward
3219 * compatibility issue.
3220 */
3221 if (!xchg(&root->orphan_item_inserted, 1))
3222 insert = 2;
3223 else
3224 insert = 1;
3225#endif
3226 insert = 1;
321f0e70 3227 atomic_inc(&root->orphan_inodes);
7b128766
JB		 3228 }
3229
72ac3c0d
JB		 3230 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3231 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3232 reserve = 1;
d68fc57b 3233 spin_unlock(&root->orphan_lock);
7b128766 3234
d68fc57b
YZ		 3235 /* grab metadata reservation from transaction handle */
3236 if (reserve) {
3237 ret = btrfs_orphan_reserve_metadata(trans, inode);
79787eaa 3238 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
d68fc57b 3239 }
7b128766 3240
d68fc57b
YZ		 3241 /* insert an orphan item to track this unlinked/truncated file */
3242 if (insert >= 1) {
33345d01 3243 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
4ef31a45 3244 if (ret) {
703c88e0 3245 atomic_dec(&root->orphan_inodes);
4ef31a45
JB		 3246 if (reserve) {
3247 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3248 &BTRFS_I(inode)->runtime_flags);
3249 btrfs_orphan_release_metadata(inode);
3250 }
3251 if (ret != -EEXIST) {
e8e7cff6
JB		 3252 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3253 &BTRFS_I(inode)->runtime_flags);
4ef31a45
JB		 3254 btrfs_abort_transaction(trans, root, ret);
3255 return ret;
3256 }
79787eaa
JM		 3257 }
3258 ret = 0;
d68fc57b
YZ		 3259 }
3260
3261 /* insert an orphan item to track subvolume contains orphan files */
3262 if (insert >= 2) {
3263 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3264 root->root_key.objectid);
79787eaa
JM		 3265 if (ret && ret != -EEXIST) {
3266 btrfs_abort_transaction(trans, root, ret);
3267 return ret;
3268 }
d68fc57b
YZ		 3269 }
3270 return 0;
7b128766
JB		 3271}
3272
3273/*
3274 * We have done the truncate/delete so we can go ahead and remove the orphan
3275 * item for this particular inode.
3276 */
48a3b636
ES		 3277static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3278 struct inode *inode)
7b128766
JB		 3279{
3280 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 3281 int delete_item = 0;
3282 int release_rsv = 0;
7b128766
JB		 3283 int ret = 0;
3284
d68fc57b 3285 spin_lock(&root->orphan_lock);
8a35d95f
JB		 3286 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3287 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3288 delete_item = 1;
7b128766 3289
72ac3c0d
JB		 3290 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3291 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3292 release_rsv = 1;
d68fc57b 3293 spin_unlock(&root->orphan_lock);
7b128766 3294
703c88e0 3295 if (delete_item) {
8a35d95f 3296 atomic_dec(&root->orphan_inodes);
703c88e0
FDBM		 3297 if (trans)
3298 ret = btrfs_del_orphan_item(trans, root,
3299 btrfs_ino(inode));
8a35d95f 3300 }
7b128766 3301
703c88e0
FDBM		 3302 if (release_rsv)
3303 btrfs_orphan_release_metadata(inode);
3304
4ef31a45 3305 return ret;
7b128766
JB		 3306}
3307
3308/*
3309 * this cleans up any orphans that may be left on the list from the last use
3310 * of this root.
3311 */
66b4ffd1 3312int btrfs_orphan_cleanup(struct btrfs_root *root)
7b128766
JB		 3313{
3314 struct btrfs_path *path;
3315 struct extent_buffer *leaf;
7b128766
JB		 3316 struct btrfs_key key, found_key;
3317 struct btrfs_trans_handle *trans;
3318 struct inode *inode;
8f6d7f4f 3319 u64 last_objectid = 0;
7b128766
JB		 3320 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3321
d68fc57b 3322 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
66b4ffd1 3323 return 0;
c71bf099
YZ		 3324
3325 path = btrfs_alloc_path();
66b4ffd1
JB		 3326 if (!path) {
3327 ret = -ENOMEM;
3328 goto out;
3329 }
7b128766
JB		 3330 path->reada = -1;
3331
3332 key.objectid = BTRFS_ORPHAN_OBJECTID;
962a298f 3333 key.type = BTRFS_ORPHAN_ITEM_KEY;
7b128766
JB		 3334 key.offset = (u64)-1;
3335
7b128766
JB		 3336 while (1) {
3337 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
66b4ffd1
JB		 3338 if (ret < 0)
3339 goto out;
7b128766
JB		 3340
3341 /*
3342 * if ret == 0 means we found what we were searching for, which
25985edc 3343 * is weird, but possible, so only screw with path if we didn't
7b128766
JB		 3344 * find the key and see if we have stuff that matches
3345 */
3346 if (ret > 0) {
66b4ffd1 3347 ret = 0;
7b128766
JB		 3348 if (path->slots[0] == 0)
3349 break;
3350 path->slots[0]--;
3351 }
3352
3353 /* pull out the item */
3354 leaf = path->nodes[0];
7b128766
JB		 3355 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3356
3357 /* make sure the item matches what we want */
3358 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3359 break;
962a298f 3360 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
7b128766
JB		 3361 break;
3362
3363 /* release the path since we're done with it */
b3b4aa74 3364 btrfs_release_path(path);
7b128766
JB		 3365
3366 /*
3367 * this is where we are basically btrfs_lookup, without the
3368 * crossing root thing. we store the inode number in the
3369 * offset of the orphan item.
3370 */
8f6d7f4f
JB		 3371
3372 if (found_key.offset == last_objectid) {
c2cf52eb
SK		 3373 btrfs_err(root->fs_info,
3374 "Error removing orphan entry, stopping orphan cleanup");
8f6d7f4f
JB		 3375 ret = -EINVAL;
3376 goto out;
3377 }
3378
3379 last_objectid = found_key.offset;
3380
5d4f98a2
YZ		 3381 found_key.objectid = found_key.offset;
3382 found_key.type = BTRFS_INODE_ITEM_KEY;
3383 found_key.offset = 0;
73f73415 3384 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
8c6ffba0 3385 ret = PTR_ERR_OR_ZERO(inode);
a8c9e576 3386 if (ret && ret != -ESTALE)
66b4ffd1 3387 goto out;
7b128766 3388
f8e9e0b0
AJ		 3389 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3390 struct btrfs_root *dead_root;
3391 struct btrfs_fs_info *fs_info = root->fs_info;
3392 int is_dead_root = 0;
3393
3394 /*
3395 * this is an orphan in the tree root. Currently these
3396 * could come from 2 sources:
3397 * a) a snapshot deletion in progress
3398 * b) a free space cache inode
3399 * We need to distinguish those two, as the snapshot
3400 * orphan must not get deleted.
3401 * find_dead_roots already ran before us, so if this
3402 * is a snapshot deletion, we should find the root
3403 * in the dead_roots list
3404 */
3405 spin_lock(&fs_info->trans_lock);
3406 list_for_each_entry(dead_root, &fs_info->dead_roots,
3407 root_list) {
3408 if (dead_root->root_key.objectid ==
3409 found_key.objectid) {
3410 is_dead_root = 1;
3411 break;
3412 }
3413 }
3414 spin_unlock(&fs_info->trans_lock);
3415 if (is_dead_root) {
3416 /* prevent this orphan from being found again */
3417 key.offset = found_key.objectid - 1;
3418 continue;
3419 }
3420 }
7b128766 3421 /*
a8c9e576
JB		 3422 * Inode is already gone but the orphan item is still there,
3423 * kill the orphan item.
7b128766 3424 */
a8c9e576
JB		 3425 if (ret == -ESTALE) {
3426 trans = btrfs_start_transaction(root, 1);
66b4ffd1
JB		 3427 if (IS_ERR(trans)) {
3428 ret = PTR_ERR(trans);
3429 goto out;
3430 }
c2cf52eb
SK		 3431 btrfs_debug(root->fs_info, "auto deleting %Lu",
3432 found_key.objectid);
a8c9e576
JB		 3433 ret = btrfs_del_orphan_item(trans, root,
3434 found_key.objectid);
5b21f2ed 3435 btrfs_end_transaction(trans, root);
4ef31a45
JB		 3436 if (ret)
3437 goto out;
7b128766
JB		 3438 continue;
3439 }
3440
a8c9e576
JB		 3441 /*
3442 * add this inode to the orphan list so btrfs_orphan_del does
3443 * the proper thing when we hit it
3444 */
8a35d95f
JB		 3445 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3446 &BTRFS_I(inode)->runtime_flags);
925396ec 3447 atomic_inc(&root->orphan_inodes);
a8c9e576 3448
7b128766
JB		 3449 /* if we have links, this was a truncate, lets do that */
3450 if (inode->i_nlink) {
fae7f21c 3451 if (WARN_ON(!S_ISREG(inode->i_mode))) {
a41ad394
JB		 3452 iput(inode);
3453 continue;
3454 }
7b128766 3455 nr_truncate++;
f3fe820c
JB		 3456
3457 /* 1 for the orphan item deletion. */
3458 trans = btrfs_start_transaction(root, 1);
3459 if (IS_ERR(trans)) {
c69b26b0 3460 iput(inode);
f3fe820c
JB		 3461 ret = PTR_ERR(trans);
3462 goto out;
3463 }
3464 ret = btrfs_orphan_add(trans, inode);
3465 btrfs_end_transaction(trans, root);
c69b26b0
JB		 3466 if (ret) {
3467 iput(inode);
f3fe820c 3468 goto out;
c69b26b0 3469 }
f3fe820c 3470
66b4ffd1 3471 ret = btrfs_truncate(inode);
4a7d0f68
JB		 3472 if (ret)
3473 btrfs_orphan_del(NULL, inode);
7b128766
JB		 3474 } else {
3475 nr_unlink++;
3476 }
3477
3478 /* this will do delete_inode and everything for us */
3479 iput(inode);
66b4ffd1
JB		 3480 if (ret)
3481 goto out;
7b128766 3482 }
3254c876
MX		 3483 /* release the path since we're done with it */
3484 btrfs_release_path(path);
3485
d68fc57b
YZ		 3486 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3487
3488 if (root->orphan_block_rsv)
3489 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3490 (u64)-1);
3491
27cdeb70
MX		 3492 if (root->orphan_block_rsv ||
3493 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
7a7eaa40 3494 trans = btrfs_join_transaction(root);
66b4ffd1
JB		 3495 if (!IS_ERR(trans))
3496 btrfs_end_transaction(trans, root);
d68fc57b 3497 }
7b128766
JB		 3498
3499 if (nr_unlink)
4884b476 3500 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
7b128766 3501 if (nr_truncate)
4884b476 3502 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
66b4ffd1
JB		 3503
3504out:
3505 if (ret)
68b663d1 3506 btrfs_err(root->fs_info,
c2cf52eb 3507 "could not do orphan cleanup %d", ret);
66b4ffd1
JB		 3508 btrfs_free_path(path);
3509 return ret;
7b128766
JB		 3510}
3511
46a53cca
CM		 3512/*
3513 * very simple check to peek ahead in the leaf looking for xattrs. If we
3514 * don't find any xattrs, we know there can't be any acls.
3515 *
3516 * slot is the slot the inode is in, objectid is the objectid of the inode
3517 */
3518static noinline int acls_after_inode_item(struct extent_buffer *leaf,
63541927
FDBM		 3519 int slot, u64 objectid,
3520 int *first_xattr_slot)
46a53cca
CM		 3521{
3522 u32 nritems = btrfs_header_nritems(leaf);
3523 struct btrfs_key found_key;
f23b5a59
JB		 3524 static u64 xattr_access = 0;
3525 static u64 xattr_default = 0;
46a53cca
CM		 3526 int scanned = 0;
3527
f23b5a59
JB		 3528 if (!xattr_access) {
3529 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3530 strlen(POSIX_ACL_XATTR_ACCESS));
3531 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3532 strlen(POSIX_ACL_XATTR_DEFAULT));
3533 }
3534
46a53cca 3535 slot++;
63541927 3536 *first_xattr_slot = -1;
46a53cca
CM		 3537 while (slot < nritems) {
3538 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3539
3540 /* we found a different objectid, there must not be acls */
3541 if (found_key.objectid != objectid)
3542 return 0;
3543
3544 /* we found an xattr, assume we've got an acl */
f23b5a59 3545 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
63541927
FDBM		 3546 if (*first_xattr_slot == -1)
3547 *first_xattr_slot = slot;
f23b5a59
JB		 3548 if (found_key.offset == xattr_access ||
3549 found_key.offset == xattr_default)
3550 return 1;
3551 }
46a53cca
CM		 3552
3553 /*
3554 * we found a key greater than an xattr key, there can't
3555 * be any acls later on
3556 */
3557 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3558 return 0;
3559
3560 slot++;
3561 scanned++;
3562
3563 /*
3564 * it goes inode, inode backrefs, xattrs, extents,
3565 * so if there are a ton of hard links to an inode there can
3566 * be a lot of backrefs. Don't waste time searching too hard,
3567 * this is just an optimization
3568 */
3569 if (scanned >= 8)
3570 break;
3571 }
3572 /* we hit the end of the leaf before we found an xattr or
3573 * something larger than an xattr. We have to assume the inode
3574 * has acls
3575 */
63541927
FDBM		 3576 if (*first_xattr_slot == -1)
3577 *first_xattr_slot = slot;
46a53cca
CM		 3578 return 1;
3579}
3580
d352ac68
CM		 3581/*
3582 * read an inode from the btree into the in-memory inode
3583 */
5d4f98a2 3584static void btrfs_read_locked_inode(struct inode *inode)
39279cc3
CM		 3585{
3586 struct btrfs_path *path;
5f39d397 3587 struct extent_buffer *leaf;
39279cc3
CM		 3588 struct btrfs_inode_item *inode_item;
3589 struct btrfs_root *root = BTRFS_I(inode)->root;
3590 struct btrfs_key location;
67de1176 3591 unsigned long ptr;
46a53cca 3592 int maybe_acls;
618e21d5 3593 u32 rdev;
39279cc3 3594 int ret;
2f7e33d4 3595 bool filled = false;
63541927 3596 int first_xattr_slot;
2f7e33d4
MX		 3597
3598 ret = btrfs_fill_inode(inode, &rdev);
3599 if (!ret)
3600 filled = true;
39279cc3
CM		 3601
3602 path = btrfs_alloc_path();
1748f843
MF		 3603 if (!path)
3604 goto make_bad;
3605
39279cc3 3606 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 3607
39279cc3 3608 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 3609 if (ret)
39279cc3 3610 goto make_bad;
39279cc3 3611
5f39d397 3612 leaf = path->nodes[0];
2f7e33d4
MX		 3613
3614 if (filled)
67de1176 3615 goto cache_index;
2f7e33d4 3616
5f39d397
CM		 3617 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3618 struct btrfs_inode_item);
5f39d397 3619 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
bfe86848 3620 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2f2f43d3
EB		 3621 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3622 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
dbe674a9 3623 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397 3624
a937b979
DS		 3625 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3626 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
5f39d397 3627
a937b979
DS		 3628 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3629 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
5f39d397 3630
a937b979
DS		 3631 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3632 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
5f39d397 3633
9cc97d64 3634 BTRFS_I(inode)->i_otime.tv_sec =
3635 btrfs_timespec_sec(leaf, &inode_item->otime);
3636 BTRFS_I(inode)->i_otime.tv_nsec =
3637 btrfs_timespec_nsec(leaf, &inode_item->otime);
5f39d397 3638
a76a3cd4 3639 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 3640 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
5dc562c5
JB		 3641 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3642
3643 /*
3644 * If we were modified in the current generation and evicted from memory
3645 * and then re-read we need to do a full sync since we don't have any
3646 * idea about which extents were modified before we were evicted from
3647 * cache.
3648 */
3649 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3650 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3651 &BTRFS_I(inode)->runtime_flags);
3652
0c4d2d95 3653 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
e02119d5 3654 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 3655 inode->i_rdev = 0;
5f39d397
CM		 3656 rdev = btrfs_inode_rdev(leaf, inode_item);
3657
aec7477b 3658 BTRFS_I(inode)->index_cnt = (u64)-1;
d2fb3437 3659 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
67de1176
MX		 3660
3661cache_index:
3662 path->slots[0]++;
3663 if (inode->i_nlink != 1 ||
3664 path->slots[0] >= btrfs_header_nritems(leaf))
3665 goto cache_acl;
3666
3667 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3668 if (location.objectid != btrfs_ino(inode))
3669 goto cache_acl;
3670
3671 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3672 if (location.type == BTRFS_INODE_REF_KEY) {
3673 struct btrfs_inode_ref *ref;
3674
3675 ref = (struct btrfs_inode_ref *)ptr;
3676 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3677 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3678 struct btrfs_inode_extref *extref;
3679
3680 extref = (struct btrfs_inode_extref *)ptr;
3681 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3682 extref);
3683 }
2f7e33d4 3684cache_acl:
46a53cca
CM		 3685 /*
3686 * try to precache a NULL acl entry for files that don't have
3687 * any xattrs or acls
3688 */
33345d01 3689 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
63541927
FDBM		 3690 btrfs_ino(inode), &first_xattr_slot);
3691 if (first_xattr_slot != -1) {
3692 path->slots[0] = first_xattr_slot;
3693 ret = btrfs_load_inode_props(inode, path);
3694 if (ret)
3695 btrfs_err(root->fs_info,
351fd353 3696 "error loading props for ino %llu (root %llu): %d",
63541927
FDBM		 3697 btrfs_ino(inode),
3698 root->root_key.objectid, ret);
3699 }
3700 btrfs_free_path(path);
3701
72c04902
AV		 3702 if (!maybe_acls)
3703 cache_no_acl(inode);
46a53cca 3704
39279cc3 3705 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 3706 case S_IFREG:
3707 inode->i_mapping->a_ops = &btrfs_aops;
d1310b2e 3708 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 3709 inode->i_fop = &btrfs_file_operations;
3710 inode->i_op = &btrfs_file_inode_operations;
3711 break;
3712 case S_IFDIR:
3713 inode->i_fop = &btrfs_dir_file_operations;
3714 if (root == root->fs_info->tree_root)
3715 inode->i_op = &btrfs_dir_ro_inode_operations;
3716 else
3717 inode->i_op = &btrfs_dir_inode_operations;
3718 break;
3719 case S_IFLNK:
3720 inode->i_op = &btrfs_symlink_inode_operations;
3721 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3722 break;
618e21d5 3723 default:
0279b4cd 3724 inode->i_op = &btrfs_special_inode_operations;
618e21d5
JB		 3725 init_special_inode(inode, inode->i_mode, rdev);
3726 break;
39279cc3 3727 }
6cbff00f
CH		 3728
3729 btrfs_update_iflags(inode);
39279cc3
CM		 3730 return;
3731
3732make_bad:
39279cc3 3733 btrfs_free_path(path);
39279cc3
CM		 3734 make_bad_inode(inode);
3735}
3736
d352ac68
CM		 3737/*
3738 * given a leaf and an inode, copy the inode fields into the leaf
3739 */
e02119d5
CM		 3740static void fill_inode_item(struct btrfs_trans_handle *trans,
3741 struct extent_buffer *leaf,
5f39d397 3742 struct btrfs_inode_item *item,
39279cc3
CM		 3743 struct inode *inode)
3744{
51fab693
LB		 3745 struct btrfs_map_token token;
3746
3747 btrfs_init_map_token(&token);
5f39d397 3748
51fab693
LB		 3749 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3750 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3751 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3752 &token);
3753 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3754 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
5f39d397 3755
a937b979 3756 btrfs_set_token_timespec_sec(leaf, &item->atime,
51fab693 3757 inode->i_atime.tv_sec, &token);
a937b979 3758 btrfs_set_token_timespec_nsec(leaf, &item->atime,
51fab693 3759 inode->i_atime.tv_nsec, &token);
5f39d397 3760
a937b979 3761 btrfs_set_token_timespec_sec(leaf, &item->mtime,
51fab693 3762 inode->i_mtime.tv_sec, &token);
a937b979 3763 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
51fab693 3764 inode->i_mtime.tv_nsec, &token);
5f39d397 3765
a937b979 3766 btrfs_set_token_timespec_sec(leaf, &item->ctime,
51fab693 3767 inode->i_ctime.tv_sec, &token);
a937b979 3768 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
51fab693 3769 inode->i_ctime.tv_nsec, &token);
5f39d397 3770
9cc97d64 3771 btrfs_set_token_timespec_sec(leaf, &item->otime,
3772 BTRFS_I(inode)->i_otime.tv_sec, &token);
3773 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3774 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3775
51fab693
LB		 3776 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3777 &token);
3778 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3779 &token);
3780 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3781 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3782 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3783 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3784 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
39279cc3
CM		 3785}
3786
d352ac68
CM		 3787/*
3788 * copy everything in the in-memory inode into the btree.
3789 */
2115133f 3790static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
d397712b 3791 struct btrfs_root *root, struct inode *inode)
39279cc3
CM		 3792{
3793 struct btrfs_inode_item *inode_item;
3794 struct btrfs_path *path;
5f39d397 3795 struct extent_buffer *leaf;
39279cc3
CM		 3796 int ret;
3797
3798 path = btrfs_alloc_path();
16cdcec7
MX		 3799 if (!path)
3800 return -ENOMEM;
3801
b9473439 3802 path->leave_spinning = 1;
16cdcec7
MX		 3803 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3804 1);
39279cc3
CM		 3805 if (ret) {
3806 if (ret > 0)
3807 ret = -ENOENT;
3808 goto failed;
3809 }
3810
5f39d397
CM		 3811 leaf = path->nodes[0];
3812 inode_item = btrfs_item_ptr(leaf, path->slots[0],
16cdcec7 3813 struct btrfs_inode_item);
39279cc3 3814
e02119d5 3815 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 3816 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 3817 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 3818 ret = 0;
3819failed:
39279cc3
CM		 3820 btrfs_free_path(path);
3821 return ret;
3822}
3823
2115133f
CM		 3824/*
3825 * copy everything in the in-memory inode into the btree.
3826 */
3827noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3828 struct btrfs_root *root, struct inode *inode)
3829{
3830 int ret;
3831
3832 /*
3833 * If the inode is a free space inode, we can deadlock during commit
3834 * if we put it into the delayed code.
3835 *
3836 * The data relocation inode should also be directly updated
3837 * without delay
3838 */
83eea1f1 3839 if (!btrfs_is_free_space_inode(inode)
1d52c78a
JB		 3840 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3841 && !root->fs_info->log_root_recovering) {
8ea05e3a
AB		 3842 btrfs_update_root_times(trans, root);
3843
2115133f
CM		 3844 ret = btrfs_delayed_update_inode(trans, root, inode);
3845 if (!ret)
3846 btrfs_set_inode_last_trans(trans, inode);
3847 return ret;
3848 }
3849
3850 return btrfs_update_inode_item(trans, root, inode);
3851}
3852
be6aef60
JB		 3853noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3854 struct btrfs_root *root,
3855 struct inode *inode)
2115133f
CM		 3856{
3857 int ret;
3858
3859 ret = btrfs_update_inode(trans, root, inode);
3860 if (ret == -ENOSPC)
3861 return btrfs_update_inode_item(trans, root, inode);
3862 return ret;
3863}
3864
d352ac68
CM		 3865/*
3866 * unlink helper that gets used here in inode.c and in the tree logging
3867 * recovery code. It remove a link in a directory with a given name, and
3868 * also drops the back refs in the inode to the directory
3869 */
92986796
AV		 3870static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3871 struct btrfs_root *root,
3872 struct inode *dir, struct inode *inode,
3873 const char *name, int name_len)
39279cc3
CM		 3874{
3875 struct btrfs_path *path;
39279cc3 3876 int ret = 0;
5f39d397 3877 struct extent_buffer *leaf;
39279cc3 3878 struct btrfs_dir_item *di;
5f39d397 3879 struct btrfs_key key;
aec7477b 3880 u64 index;
33345d01
LZ		 3881 u64 ino = btrfs_ino(inode);
3882 u64 dir_ino = btrfs_ino(dir);
39279cc3
CM		 3883
3884 path = btrfs_alloc_path();
54aa1f4d
CM		 3885 if (!path) {
3886 ret = -ENOMEM;
554233a6 3887 goto out;
54aa1f4d
CM		 3888 }
3889
b9473439 3890 path->leave_spinning = 1;
33345d01 3891 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
39279cc3
CM		 3892 name, name_len, -1);
3893 if (IS_ERR(di)) {
3894 ret = PTR_ERR(di);
3895 goto err;
3896 }
3897 if (!di) {
3898 ret = -ENOENT;
3899 goto err;
3900 }
5f39d397
CM		 3901 leaf = path->nodes[0];
3902 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 3903 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 3904 if (ret)
3905 goto err;
b3b4aa74 3906 btrfs_release_path(path);
39279cc3 3907
67de1176
MX		 3908 /*
3909 * If we don't have dir index, we have to get it by looking up
3910 * the inode ref, since we get the inode ref, remove it directly,
3911 * it is unnecessary to do delayed deletion.
3912 *
3913 * But if we have dir index, needn't search inode ref to get it.
3914 * Since the inode ref is close to the inode item, it is better
3915 * that we delay to delete it, and just do this deletion when
3916 * we update the inode item.
3917 */
3918 if (BTRFS_I(inode)->dir_index) {
3919 ret = btrfs_delayed_delete_inode_ref(inode);
3920 if (!ret) {
3921 index = BTRFS_I(inode)->dir_index;
3922 goto skip_backref;
3923 }
3924 }
3925
33345d01
LZ		 3926 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3927 dir_ino, &index);
aec7477b 3928 if (ret) {
c2cf52eb
SK		 3929 btrfs_info(root->fs_info,
3930 "failed to delete reference to %.*s, inode %llu parent %llu",
c1c9ff7c 3931 name_len, name, ino, dir_ino);
79787eaa 3932 btrfs_abort_transaction(trans, root, ret);
aec7477b
JB		 3933 goto err;
3934 }
67de1176 3935skip_backref:
16cdcec7 3936 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 3937 if (ret) {
3938 btrfs_abort_transaction(trans, root, ret);
39279cc3 3939 goto err;
79787eaa 3940 }
39279cc3 3941
e02119d5 3942 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
33345d01 3943 inode, dir_ino);
79787eaa
JM		 3944 if (ret != 0 && ret != -ENOENT) {
3945 btrfs_abort_transaction(trans, root, ret);
3946 goto err;
3947 }
e02119d5
CM		 3948
3949 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3950 dir, index);
6418c961
CM		 3951 if (ret == -ENOENT)
3952 ret = 0;
d4e3991b
ZB		 3953 else if (ret)
3954 btrfs_abort_transaction(trans, root, ret);
39279cc3
CM		 3955err:
3956 btrfs_free_path(path);
e02119d5
CM		 3957 if (ret)
3958 goto out;
3959
3960 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95
JB		 3961 inode_inc_iversion(inode);
3962 inode_inc_iversion(dir);
e02119d5 3963 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
b9959295 3964 ret = btrfs_update_inode(trans, root, dir);
e02119d5 3965out:
39279cc3
CM		 3966 return ret;
3967}
3968
92986796
AV		 3969int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3970 struct btrfs_root *root,
3971 struct inode *dir, struct inode *inode,
3972 const char *name, int name_len)
3973{
3974 int ret;
3975 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3976 if (!ret) {
8b558c5f 3977 drop_nlink(inode);
92986796
AV		 3978 ret = btrfs_update_inode(trans, root, inode);
3979 }
3980 return ret;
3981}
39279cc3 3982
a22285a6
YZ		 3983/*
3984 * helper to start transaction for unlink and rmdir.
3985 *
d52be818
JB		 3986 * unlink and rmdir are special in btrfs, they do not always free space, so
3987 * if we cannot make our reservations the normal way try and see if there is
3988 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3989 * allow the unlink to occur.
a22285a6 3990 */
d52be818 3991static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4df27c4d 3992{
39279cc3 3993 struct btrfs_trans_handle *trans;
a22285a6 3994 struct btrfs_root *root = BTRFS_I(dir)->root;
4df27c4d
YZ		 3995 int ret;
3996
e70bea5f
JB		 3997 /*
3998 * 1 for the possible orphan item
3999 * 1 for the dir item
4000 * 1 for the dir index
4001 * 1 for the inode ref
e70bea5f
JB		 4002 * 1 for the inode
4003 */
6e137ed3 4004 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 4005 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
4006 return trans;
4df27c4d 4007
d52be818
JB		 4008 if (PTR_ERR(trans) == -ENOSPC) {
4009 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4df27c4d 4010
d52be818
JB		 4011 trans = btrfs_start_transaction(root, 0);
4012 if (IS_ERR(trans))
4013 return trans;
4014 ret = btrfs_cond_migrate_bytes(root->fs_info,
4015 &root->fs_info->trans_block_rsv,
4016 num_bytes, 5);
4017 if (ret) {
4018 btrfs_end_transaction(trans, root);
4019 return ERR_PTR(ret);
a22285a6 4020 }
5a77d76c 4021 trans->block_rsv = &root->fs_info->trans_block_rsv;
d52be818 4022 trans->bytes_reserved = num_bytes;
a22285a6 4023 }
d52be818 4024 return trans;
a22285a6
YZ		 4025}
4026
4027static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4028{
4029 struct btrfs_root *root = BTRFS_I(dir)->root;
4030 struct btrfs_trans_handle *trans;
4031 struct inode *inode = dentry->d_inode;
4032 int ret;
a22285a6 4033
d52be818 4034 trans = __unlink_start_trans(dir);
a22285a6
YZ		 4035 if (IS_ERR(trans))
4036 return PTR_ERR(trans);
5f39d397 4037
12fcfd22
CM		 4038 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
4039
e02119d5
CM		 4040 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
4041 dentry->d_name.name, dentry->d_name.len);
b532402e
TI		 4042 if (ret)
4043 goto out;
7b128766 4044
a22285a6 4045 if (inode->i_nlink == 0) {
7b128766 4046 ret = btrfs_orphan_add(trans, inode);
b532402e
TI		 4047 if (ret)
4048 goto out;
a22285a6 4049 }
7b128766 4050
b532402e 4051out:
d52be818 4052 btrfs_end_transaction(trans, root);
b53d3f5d 4053 btrfs_btree_balance_dirty(root);
39279cc3
CM		 4054 return ret;
4055}
4056
4df27c4d
YZ		 4057int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4058 struct btrfs_root *root,
4059 struct inode *dir, u64 objectid,
4060 const char *name, int name_len)
4061{
4062 struct btrfs_path *path;
4063 struct extent_buffer *leaf;
4064 struct btrfs_dir_item *di;
4065 struct btrfs_key key;
4066 u64 index;
4067 int ret;
33345d01 4068 u64 dir_ino = btrfs_ino(dir);
4df27c4d
YZ		 4069
4070 path = btrfs_alloc_path();
4071 if (!path)
4072 return -ENOMEM;
4073
33345d01 4074 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4df27c4d 4075 name, name_len, -1);
79787eaa
JM		 4076 if (IS_ERR_OR_NULL(di)) {
4077 if (!di)
4078 ret = -ENOENT;
4079 else
4080 ret = PTR_ERR(di);
4081 goto out;
4082 }
4df27c4d
YZ		 4083
4084 leaf = path->nodes[0];
4085 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4086 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4087 ret = btrfs_delete_one_dir_name(trans, root, path, di);
79787eaa
JM		 4088 if (ret) {
4089 btrfs_abort_transaction(trans, root, ret);
4090 goto out;
4091 }
b3b4aa74 4092 btrfs_release_path(path);
4df27c4d
YZ		 4093
4094 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4095 objectid, root->root_key.objectid,
33345d01 4096 dir_ino, &index, name, name_len);
4df27c4d 4097 if (ret < 0) {
79787eaa
JM		 4098 if (ret != -ENOENT) {
4099 btrfs_abort_transaction(trans, root, ret);
4100 goto out;
4101 }
33345d01 4102 di = btrfs_search_dir_index_item(root, path, dir_ino,
4df27c4d 4103 name, name_len);
79787eaa
JM		 4104 if (IS_ERR_OR_NULL(di)) {
4105 if (!di)
4106 ret = -ENOENT;
4107 else
4108 ret = PTR_ERR(di);
4109 btrfs_abort_transaction(trans, root, ret);
4110 goto out;
4111 }
4df27c4d
YZ		 4112
4113 leaf = path->nodes[0];
4114 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
b3b4aa74 4115 btrfs_release_path(path);
4df27c4d
YZ		 4116 index = key.offset;
4117 }
945d8962 4118 btrfs_release_path(path);
4df27c4d 4119
16cdcec7 4120 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 4121 if (ret) {
4122 btrfs_abort_transaction(trans, root, ret);
4123 goto out;
4124 }
4df27c4d
YZ		 4125
4126 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95 4127 inode_inc_iversion(dir);
4df27c4d 4128 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
5a24e84c 4129 ret = btrfs_update_inode_fallback(trans, root, dir);
79787eaa
JM		 4130 if (ret)
4131 btrfs_abort_transaction(trans, root, ret);
4132out:
71d7aed0 4133 btrfs_free_path(path);
79787eaa 4134 return ret;
4df27c4d
YZ		 4135}
4136
39279cc3
CM		 4137static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4138{
4139 struct inode *inode = dentry->d_inode;
1832a6d5 4140 int err = 0;
39279cc3 4141 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 4142 struct btrfs_trans_handle *trans;
39279cc3 4143
b3ae244e 4144 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
134d4512 4145 return -ENOTEMPTY;
b3ae244e
DS		 4146 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4147 return -EPERM;
134d4512 4148
d52be818 4149 trans = __unlink_start_trans(dir);
a22285a6 4150 if (IS_ERR(trans))
5df6a9f6 4151 return PTR_ERR(trans);
5df6a9f6 4152
33345d01 4153 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4df27c4d
YZ		 4154 err = btrfs_unlink_subvol(trans, root, dir,
4155 BTRFS_I(inode)->location.objectid,
4156 dentry->d_name.name,
4157 dentry->d_name.len);
4158 goto out;
4159 }
4160
7b128766
JB		 4161 err = btrfs_orphan_add(trans, inode);
4162 if (err)
4df27c4d 4163 goto out;
7b128766 4164
39279cc3 4165 /* now the directory is empty */
e02119d5
CM		 4166 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
4167 dentry->d_name.name, dentry->d_name.len);
d397712b 4168 if (!err)
dbe674a9 4169 btrfs_i_size_write(inode, 0);
4df27c4d 4170out:
d52be818 4171 btrfs_end_transaction(trans, root);
b53d3f5d 4172 btrfs_btree_balance_dirty(root);
3954401f 4173
39279cc3
CM		 4174 return err;
4175}
4176
28f75a0e
CM		 4177static int truncate_space_check(struct btrfs_trans_handle *trans,
4178 struct btrfs_root *root,
4179 u64 bytes_deleted)
4180{
4181 int ret;
4182
4183 bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4184 ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4185 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4186 if (!ret)
4187 trans->bytes_reserved += bytes_deleted;
4188 return ret;
4189
4190}
4191
39279cc3
CM		 4192/*
4193 * this can truncate away extent items, csum items and directory items.
4194 * It starts at a high offset and removes keys until it can't find
d352ac68 4195 * any higher than new_size
39279cc3
CM		 4196 *
4197 * csum items that cross the new i_size are truncated to the new size
4198 * as well.
7b128766
JB		 4199 *
4200 * min_type is the minimum key type to truncate down to. If set to 0, this
4201 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 4202 */
8082510e
YZ		 4203int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4204 struct btrfs_root *root,
4205 struct inode *inode,
4206 u64 new_size, u32 min_type)
39279cc3 4207{
39279cc3 4208 struct btrfs_path *path;
5f39d397 4209 struct extent_buffer *leaf;
39279cc3 4210 struct btrfs_file_extent_item *fi;
8082510e
YZ		 4211 struct btrfs_key key;
4212 struct btrfs_key found_key;
39279cc3 4213 u64 extent_start = 0;
db94535d 4214 u64 extent_num_bytes = 0;
5d4f98a2 4215 u64 extent_offset = 0;
39279cc3 4216 u64 item_end = 0;
7f4f6e0a 4217 u64 last_size = (u64)-1;
8082510e 4218 u32 found_type = (u8)-1;
39279cc3
CM		 4219 int found_extent;
4220 int del_item;
85e21bac
CM		 4221 int pending_del_nr = 0;
4222 int pending_del_slot = 0;
179e29e4 4223 int extent_type = -1;
8082510e
YZ		 4224 int ret;
4225 int err = 0;
33345d01 4226 u64 ino = btrfs_ino(inode);
28ed1345 4227 u64 bytes_deleted = 0;
1262133b
JB		 4228 bool be_nice = 0;
4229 bool should_throttle = 0;
28f75a0e 4230 bool should_end = 0;
8082510e
YZ		 4231
4232 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
39279cc3 4233
28ed1345
CM		 4234 /*
4235 * for non-free space inodes and ref cows, we want to back off from
4236 * time to time
4237 */
4238 if (!btrfs_is_free_space_inode(inode) &&
4239 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4240 be_nice = 1;
4241
0eb0e19c
MF		 4242 path = btrfs_alloc_path();
4243 if (!path)
4244 return -ENOMEM;
4245 path->reada = -1;
4246
5dc562c5
JB		 4247 /*
4248 * We want to drop from the next block forward in case this new size is
4249 * not block aligned since we will be keeping the last block of the
4250 * extent just the way it is.
4251 */
27cdeb70
MX		 4252 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4253 root == root->fs_info->tree_root)
fda2832f
QW		 4254 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4255 root->sectorsize), (u64)-1, 0);
8082510e 4256
16cdcec7
MX		 4257 /*
4258 * This function is also used to drop the items in the log tree before
4259 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4260 * it is used to drop the loged items. So we shouldn't kill the delayed
4261 * items.
4262 */
4263 if (min_type == 0 && root == BTRFS_I(inode)->root)
4264 btrfs_kill_delayed_inode_items(inode);
4265
33345d01 4266 key.objectid = ino;
39279cc3 4267 key.offset = (u64)-1;
5f39d397
CM		 4268 key.type = (u8)-1;
4269
85e21bac 4270search_again:
28ed1345
CM		 4271 /*
4272 * with a 16K leaf size and 128MB extents, you can actually queue
4273 * up a huge file in a single leaf. Most of the time that
4274 * bytes_deleted is > 0, it will be huge by the time we get here
4275 */
4276 if (be_nice && bytes_deleted > 32 * 1024 * 1024) {
4277 if (btrfs_should_end_transaction(trans, root)) {
4278 err = -EAGAIN;
4279 goto error;
4280 }
4281 }
4282
4283
b9473439 4284 path->leave_spinning = 1;
85e21bac 4285 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8082510e
YZ		 4286 if (ret < 0) {
4287 err = ret;
4288 goto out;
4289 }
d397712b 4290
85e21bac 4291 if (ret > 0) {
e02119d5
CM		 4292 /* there are no items in the tree for us to truncate, we're
4293 * done
4294 */
8082510e
YZ		 4295 if (path->slots[0] == 0)
4296 goto out;
85e21bac
CM		 4297 path->slots[0]--;
4298 }
4299
d397712b 4300 while (1) {
39279cc3 4301 fi = NULL;
5f39d397
CM		 4302 leaf = path->nodes[0];
4303 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
962a298f 4304 found_type = found_key.type;
39279cc3 4305
33345d01 4306 if (found_key.objectid != ino)
39279cc3 4307 break;
5f39d397 4308
85e21bac 4309 if (found_type < min_type)
39279cc3
CM		 4310 break;
4311
5f39d397 4312 item_end = found_key.offset;
39279cc3 4313 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 4314 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 4315 struct btrfs_file_extent_item);
179e29e4
CM		 4316 extent_type = btrfs_file_extent_type(leaf, fi);
4317 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 4318 item_end +=
db94535d 4319 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 4320 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 4321 item_end += btrfs_file_extent_inline_len(leaf,
514ac8ad 4322 path->slots[0], fi);
39279cc3 4323 }
008630c1 4324 item_end--;
39279cc3 4325 }
8082510e
YZ		 4326 if (found_type > min_type) {
4327 del_item = 1;
4328 } else {
4329 if (item_end < new_size)
b888db2b 4330 break;
8082510e
YZ		 4331 if (found_key.offset >= new_size)
4332 del_item = 1;
4333 else
4334 del_item = 0;
39279cc3 4335 }
39279cc3 4336 found_extent = 0;
39279cc3 4337 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 4338 if (found_type != BTRFS_EXTENT_DATA_KEY)
4339 goto delete;
4340
7f4f6e0a
JB		 4341 if (del_item)
4342 last_size = found_key.offset;
4343 else
4344 last_size = new_size;
4345
179e29e4 4346 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 4347 u64 num_dec;
db94535d 4348 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
f70a9a6b 4349 if (!del_item) {
db94535d
CM		 4350 u64 orig_num_bytes =
4351 btrfs_file_extent_num_bytes(leaf, fi);
fda2832f
QW		 4352 extent_num_bytes = ALIGN(new_size -
4353 found_key.offset,
4354 root->sectorsize);
db94535d
CM		 4355 btrfs_set_file_extent_num_bytes(leaf, fi,
4356 extent_num_bytes);
4357 num_dec = (orig_num_bytes -
9069218d 4358 extent_num_bytes);
27cdeb70
MX		 4359 if (test_bit(BTRFS_ROOT_REF_COWS,
4360 &root->state) &&
4361 extent_start != 0)
a76a3cd4 4362 inode_sub_bytes(inode, num_dec);
5f39d397 4363 btrfs_mark_buffer_dirty(leaf);
39279cc3 4364 } else {
db94535d
CM		 4365 extent_num_bytes =
4366 btrfs_file_extent_disk_num_bytes(leaf,
4367 fi);
5d4f98a2
YZ		 4368 extent_offset = found_key.offset -
4369 btrfs_file_extent_offset(leaf, fi);
4370
39279cc3 4371 /* FIXME blocksize != 4096 */
9069218d 4372 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 4373 if (extent_start != 0) {
4374 found_extent = 1;
27cdeb70
MX		 4375 if (test_bit(BTRFS_ROOT_REF_COWS,
4376 &root->state))
a76a3cd4 4377 inode_sub_bytes(inode, num_dec);
e02119d5 4378 }
39279cc3 4379 }
9069218d 4380 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 4381 /*
4382 * we can't truncate inline items that have had
4383 * special encodings
4384 */
4385 if (!del_item &&
4386 btrfs_file_extent_compression(leaf, fi) == 0 &&
4387 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4388 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM		 4389 u32 size = new_size - found_key.offset;
4390
27cdeb70 4391 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
a76a3cd4
YZ		 4392 inode_sub_bytes(inode, item_end + 1 -
4393 new_size);
514ac8ad
CM		 4394
4395 /*
4396 * update the ram bytes to properly reflect
4397 * the new size of our item
4398 */
4399 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
e02119d5
CM		 4400 size =
4401 btrfs_file_extent_calc_inline_size(size);
afe5fea7 4402 btrfs_truncate_item(root, path, size, 1);
27cdeb70
MX		 4403 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4404 &root->state)) {
a76a3cd4
YZ		 4405 inode_sub_bytes(inode, item_end + 1 -
4406 found_key.offset);
9069218d 4407 }
39279cc3 4408 }
179e29e4 4409delete:
39279cc3 4410 if (del_item) {
85e21bac
CM		 4411 if (!pending_del_nr) {
4412 /* no pending yet, add ourselves */
4413 pending_del_slot = path->slots[0];
4414 pending_del_nr = 1;
4415 } else if (pending_del_nr &&
4416 path->slots[0] + 1 == pending_del_slot) {
4417 /* hop on the pending chunk */
4418 pending_del_nr++;
4419 pending_del_slot = path->slots[0];
4420 } else {
d397712b 4421 BUG();
85e21bac 4422 }
39279cc3
CM		 4423 } else {
4424 break;
4425 }
28f75a0e
CM		 4426 should_throttle = 0;
4427
27cdeb70
MX		 4428 if (found_extent &&
4429 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4430 root == root->fs_info->tree_root)) {
b9473439 4431 btrfs_set_path_blocking(path);
28ed1345 4432 bytes_deleted += extent_num_bytes;
39279cc3 4433 ret = btrfs_free_extent(trans, root, extent_start,
5d4f98a2
YZ		 4434 extent_num_bytes, 0,
4435 btrfs_header_owner(leaf),
66d7e7f0 4436 ino, extent_offset, 0);
39279cc3 4437 BUG_ON(ret);
1262133b 4438 if (btrfs_should_throttle_delayed_refs(trans, root))
28ed1345
CM		 4439 btrfs_async_run_delayed_refs(root,
4440 trans->delayed_ref_updates * 2, 0);
28f75a0e
CM		 4441 if (be_nice) {
4442 if (truncate_space_check(trans, root,
4443 extent_num_bytes)) {
4444 should_end = 1;
4445 }
4446 if (btrfs_should_throttle_delayed_refs(trans,
4447 root)) {
4448 should_throttle = 1;
4449 }
4450 }
39279cc3 4451 }
85e21bac 4452
8082510e
YZ		 4453 if (found_type == BTRFS_INODE_ITEM_KEY)
4454 break;
4455
4456 if (path->slots[0] == 0 ||
1262133b 4457 path->slots[0] != pending_del_slot ||
28f75a0e 4458 should_throttle || should_end) {
8082510e
YZ		 4459 if (pending_del_nr) {
4460 ret = btrfs_del_items(trans, root, path,
4461 pending_del_slot,
4462 pending_del_nr);
79787eaa
JM		 4463 if (ret) {
4464 btrfs_abort_transaction(trans,
4465 root, ret);
4466 goto error;
4467 }
8082510e
YZ		 4468 pending_del_nr = 0;
4469 }
b3b4aa74 4470 btrfs_release_path(path);
28f75a0e 4471 if (should_throttle) {
1262133b
JB		 4472 unsigned long updates = trans->delayed_ref_updates;
4473 if (updates) {
4474 trans->delayed_ref_updates = 0;
4475 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4476 if (ret && !err)
4477 err = ret;
4478 }
4479 }
28f75a0e
CM		 4480 /*
4481 * if we failed to refill our space rsv, bail out
4482 * and let the transaction restart
4483 */
4484 if (should_end) {
4485 err = -EAGAIN;
4486 goto error;
4487 }
85e21bac 4488 goto search_again;
8082510e
YZ		 4489 } else {
4490 path->slots[0]--;
85e21bac 4491 }
39279cc3 4492 }
8082510e 4493out:
85e21bac
CM		 4494 if (pending_del_nr) {
4495 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4496 pending_del_nr);
79787eaa
JM		 4497 if (ret)
4498 btrfs_abort_transaction(trans, root, ret);
85e21bac 4499 }
79787eaa 4500error:
dac5705c
FM		 4501 if (last_size != (u64)-1 &&
4502 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
7f4f6e0a 4503 btrfs_ordered_update_i_size(inode, last_size, NULL);
28ed1345 4504
39279cc3 4505 btrfs_free_path(path);
28ed1345 4506
28f75a0e 4507 if (be_nice && bytes_deleted > 32 * 1024 * 1024) {
28ed1345
CM		 4508 unsigned long updates = trans->delayed_ref_updates;
4509 if (updates) {
4510 trans->delayed_ref_updates = 0;
4511 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4512 if (ret && !err)
4513 err = ret;
4514 }
4515 }
8082510e 4516 return err;
39279cc3
CM		 4517}
4518
4519/*
2aaa6655
JB		 4520 * btrfs_truncate_page - read, zero a chunk and write a page
4521 * @inode - inode that we're zeroing
4522 * @from - the offset to start zeroing
4523 * @len - the length to zero, 0 to zero the entire range respective to the
4524 * offset
4525 * @front - zero up to the offset instead of from the offset on
4526 *
4527 * This will find the page for the "from" offset and cow the page and zero the
4528 * part we want to zero. This is used with truncate and hole punching.
39279cc3 4529 */
2aaa6655
JB		 4530int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4531 int front)
39279cc3 4532{
2aaa6655 4533 struct address_space *mapping = inode->i_mapping;
db94535d 4534 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 4535 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4536 struct btrfs_ordered_extent *ordered;
2ac55d41 4537 struct extent_state *cached_state = NULL;
e6dcd2dc 4538 char *kaddr;
db94535d 4539 u32 blocksize = root->sectorsize;
39279cc3
CM		 4540 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4541 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4542 struct page *page;
3b16a4e3 4543 gfp_t mask = btrfs_alloc_write_mask(mapping);
39279cc3 4544 int ret = 0;
a52d9a80 4545 u64 page_start;
e6dcd2dc 4546 u64 page_end;
39279cc3 4547
2aaa6655
JB		 4548 if ((offset & (blocksize - 1)) == 0 &&
4549 (!len || ((len & (blocksize - 1)) == 0)))
39279cc3 4550 goto out;
0ca1f7ce 4551 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
5d5e103a
JB		 4552 if (ret)
4553 goto out;
39279cc3 4554
211c17f5 4555again:
3b16a4e3 4556 page = find_or_create_page(mapping, index, mask);
5d5e103a 4557 if (!page) {
0ca1f7ce 4558 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
ac6a2b36 4559 ret = -ENOMEM;
39279cc3 4560 goto out;
5d5e103a 4561 }
e6dcd2dc
CM		 4562
4563 page_start = page_offset(page);
4564 page_end = page_start + PAGE_CACHE_SIZE - 1;
4565
39279cc3 4566 if (!PageUptodate(page)) {
9ebefb18 4567 ret = btrfs_readpage(NULL, page);
39279cc3 4568 lock_page(page);
211c17f5
CM		 4569 if (page->mapping != mapping) {
4570 unlock_page(page);
4571 page_cache_release(page);
4572 goto again;
4573 }
39279cc3
CM		 4574 if (!PageUptodate(page)) {
4575 ret = -EIO;
89642229 4576 goto out_unlock;
39279cc3
CM		 4577 }
4578 }
211c17f5 4579 wait_on_page_writeback(page);
e6dcd2dc 4580
d0082371 4581 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 4582 set_page_extent_mapped(page);
4583
4584 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4585 if (ordered) {
2ac55d41
JB		 4586 unlock_extent_cached(io_tree, page_start, page_end,
4587 &cached_state, GFP_NOFS);
e6dcd2dc
CM		 4588 unlock_page(page);
4589 page_cache_release(page);
eb84ae03 4590 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 4591 btrfs_put_ordered_extent(ordered);
4592 goto again;
4593 }
4594
2ac55d41 4595 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 4596 EXTENT_DIRTY | EXTENT_DELALLOC |
4597 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 4598 0, 0, &cached_state, GFP_NOFS);
5d5e103a 4599
2ac55d41
JB		 4600 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4601 &cached_state);
9ed74f2d 4602 if (ret) {
2ac55d41
JB		 4603 unlock_extent_cached(io_tree, page_start, page_end,
4604 &cached_state, GFP_NOFS);
9ed74f2d
JB		 4605 goto out_unlock;
4606 }
4607
e6dcd2dc 4608 if (offset != PAGE_CACHE_SIZE) {
2aaa6655
JB		 4609 if (!len)
4610 len = PAGE_CACHE_SIZE - offset;
e6dcd2dc 4611 kaddr = kmap(page);
2aaa6655
JB		 4612 if (front)
4613 memset(kaddr, 0, offset);
4614 else
4615 memset(kaddr + offset, 0, len);
e6dcd2dc
CM		 4616 flush_dcache_page(page);
4617 kunmap(page);
4618 }
247e743c 4619 ClearPageChecked(page);
e6dcd2dc 4620 set_page_dirty(page);
2ac55d41
JB		 4621 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4622 GFP_NOFS);
39279cc3 4623
89642229 4624out_unlock:
5d5e103a 4625 if (ret)
0ca1f7ce 4626 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
39279cc3
CM		 4627 unlock_page(page);
4628 page_cache_release(page);
4629out:
4630 return ret;
4631}
4632
16e7549f
JB		 4633static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4634 u64 offset, u64 len)
4635{
4636 struct btrfs_trans_handle *trans;
4637 int ret;
4638
4639 /*
4640 * Still need to make sure the inode looks like it's been updated so
4641 * that any holes get logged if we fsync.
4642 */
4643 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4644 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4645 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4646 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4647 return 0;
4648 }
4649
4650 /*
4651 * 1 - for the one we're dropping
4652 * 1 - for the one we're adding
4653 * 1 - for updating the inode.
4654 */
4655 trans = btrfs_start_transaction(root, 3);
4656 if (IS_ERR(trans))
4657 return PTR_ERR(trans);
4658
4659 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4660 if (ret) {
4661 btrfs_abort_transaction(trans, root, ret);
4662 btrfs_end_transaction(trans, root);
4663 return ret;
4664 }
4665
4666 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4667 0, 0, len, 0, len, 0, 0, 0);
4668 if (ret)
4669 btrfs_abort_transaction(trans, root, ret);
4670 else
4671 btrfs_update_inode(trans, root, inode);
4672 btrfs_end_transaction(trans, root);
4673 return ret;
4674}
4675
695a0d0d
JB		 4676/*
4677 * This function puts in dummy file extents for the area we're creating a hole
4678 * for. So if we are truncating this file to a larger size we need to insert
4679 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4680 * the range between oldsize and size
4681 */
a41ad394 4682int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
39279cc3 4683{
9036c102
YZ		 4684 struct btrfs_root *root = BTRFS_I(inode)->root;
4685 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a22285a6 4686 struct extent_map *em = NULL;
2ac55d41 4687 struct extent_state *cached_state = NULL;
5dc562c5 4688 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
fda2832f
QW		 4689 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4690 u64 block_end = ALIGN(size, root->sectorsize);
9036c102
YZ		 4691 u64 last_byte;
4692 u64 cur_offset;
4693 u64 hole_size;
9ed74f2d 4694 int err = 0;
39279cc3 4695
a71754fc
JB		 4696 /*
4697 * If our size started in the middle of a page we need to zero out the
4698 * rest of the page before we expand the i_size, otherwise we could
4699 * expose stale data.
4700 */
4701 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4702 if (err)
4703 return err;
4704
9036c102
YZ		 4705 if (size <= hole_start)
4706 return 0;
4707
9036c102
YZ		 4708 while (1) {
4709 struct btrfs_ordered_extent *ordered;
fa7c1494 4710
2ac55d41 4711 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
d0082371 4712 &cached_state);
fa7c1494
MX		 4713 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4714 block_end - hole_start);
9036c102
YZ		 4715 if (!ordered)
4716 break;
2ac55d41
JB		 4717 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4718 &cached_state, GFP_NOFS);
fa7c1494 4719 btrfs_start_ordered_extent(inode, ordered, 1);
9036c102
YZ		 4720 btrfs_put_ordered_extent(ordered);
4721 }
39279cc3 4722
9036c102
YZ		 4723 cur_offset = hole_start;
4724 while (1) {
4725 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4726 block_end - cur_offset, 0);
79787eaa
JM		 4727 if (IS_ERR(em)) {
4728 err = PTR_ERR(em);
f2767956 4729 em = NULL;
79787eaa
JM		 4730 break;
4731 }
9036c102 4732 last_byte = min(extent_map_end(em), block_end);
fda2832f 4733 last_byte = ALIGN(last_byte , root->sectorsize);
8082510e 4734 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5dc562c5 4735 struct extent_map *hole_em;
9036c102 4736 hole_size = last_byte - cur_offset;
9ed74f2d 4737
16e7549f
JB		 4738 err = maybe_insert_hole(root, inode, cur_offset,
4739 hole_size);
4740 if (err)
3893e33b 4741 break;
5dc562c5
JB		 4742 btrfs_drop_extent_cache(inode, cur_offset,
4743 cur_offset + hole_size - 1, 0);
4744 hole_em = alloc_extent_map();
4745 if (!hole_em) {
4746 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4747 &BTRFS_I(inode)->runtime_flags);
4748 goto next;
4749 }
4750 hole_em->start = cur_offset;
4751 hole_em->len = hole_size;
4752 hole_em->orig_start = cur_offset;
8082510e 4753
5dc562c5
JB		 4754 hole_em->block_start = EXTENT_MAP_HOLE;
4755 hole_em->block_len = 0;
b4939680 4756 hole_em->orig_block_len = 0;
cc95bef6 4757 hole_em->ram_bytes = hole_size;
5dc562c5
JB		 4758 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4759 hole_em->compress_type = BTRFS_COMPRESS_NONE;
16e7549f 4760 hole_em->generation = root->fs_info->generation;
8082510e 4761
5dc562c5
JB		 4762 while (1) {
4763 write_lock(&em_tree->lock);
09a2a8f9 4764 err = add_extent_mapping(em_tree, hole_em, 1);
5dc562c5
JB		 4765 write_unlock(&em_tree->lock);
4766 if (err != -EEXIST)
4767 break;
4768 btrfs_drop_extent_cache(inode, cur_offset,
4769 cur_offset +
4770 hole_size - 1, 0);
4771 }
4772 free_extent_map(hole_em);
9036c102 4773 }
16e7549f 4774next:
9036c102 4775 free_extent_map(em);
a22285a6 4776 em = NULL;
9036c102 4777 cur_offset = last_byte;
8082510e 4778 if (cur_offset >= block_end)
9036c102
YZ		 4779 break;
4780 }
a22285a6 4781 free_extent_map(em);
2ac55d41
JB		 4782 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4783 GFP_NOFS);
9036c102
YZ		 4784 return err;
4785}
39279cc3 4786
9ea24bbe
FM		 4787static int wait_snapshoting_atomic_t(atomic_t *a)
4788{
4789 schedule();
4790 return 0;
4791}
4792
4793static void wait_for_snapshot_creation(struct btrfs_root *root)
4794{
4795 while (true) {
4796 int ret;
4797
4798 ret = btrfs_start_write_no_snapshoting(root);
4799 if (ret)
4800 break;
4801 wait_on_atomic_t(&root->will_be_snapshoted,
4802 wait_snapshoting_atomic_t,
4803 TASK_UNINTERRUPTIBLE);
4804 }
4805}
4806
3972f260 4807static int btrfs_setsize(struct inode *inode, struct iattr *attr)
8082510e 4808{
f4a2f4c5
MX		 4809 struct btrfs_root *root = BTRFS_I(inode)->root;
4810 struct btrfs_trans_handle *trans;
a41ad394 4811 loff_t oldsize = i_size_read(inode);
3972f260
ES		 4812 loff_t newsize = attr->ia_size;
4813 int mask = attr->ia_valid;
8082510e
YZ		 4814 int ret;
4815
3972f260
ES		 4816 /*
4817 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4818 * special case where we need to update the times despite not having
4819 * these flags set. For all other operations the VFS set these flags
4820 * explicitly if it wants a timestamp update.
4821 */
dff6efc3
CH		 4822 if (newsize != oldsize) {
4823 inode_inc_iversion(inode);
4824 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4825 inode->i_ctime = inode->i_mtime =
4826 current_fs_time(inode->i_sb);
4827 }
3972f260 4828
a41ad394 4829 if (newsize > oldsize) {
7caef267 4830 truncate_pagecache(inode, newsize);
9ea24bbe
FM		 4831 /*
4832 * Don't do an expanding truncate while snapshoting is ongoing.
4833 * This is to ensure the snapshot captures a fully consistent
4834 * state of this file - if the snapshot captures this expanding
4835 * truncation, it must capture all writes that happened before
4836 * this truncation.
4837 */
4838 wait_for_snapshot_creation(root);
a41ad394 4839 ret = btrfs_cont_expand(inode, oldsize, newsize);
9ea24bbe
FM		 4840 if (ret) {
4841 btrfs_end_write_no_snapshoting(root);
8082510e 4842 return ret;
9ea24bbe 4843 }
8082510e 4844
f4a2f4c5 4845 trans = btrfs_start_transaction(root, 1);
9ea24bbe
FM		 4846 if (IS_ERR(trans)) {
4847 btrfs_end_write_no_snapshoting(root);
f4a2f4c5 4848 return PTR_ERR(trans);
9ea24bbe 4849 }
f4a2f4c5
MX		 4850
4851 i_size_write(inode, newsize);
4852 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4853 ret = btrfs_update_inode(trans, root, inode);
9ea24bbe 4854 btrfs_end_write_no_snapshoting(root);
7ad85bb7 4855 btrfs_end_transaction(trans, root);
a41ad394 4856 } else {
8082510e 4857
a41ad394
JB		 4858 /*
4859 * We're truncating a file that used to have good data down to
4860 * zero. Make sure it gets into the ordered flush list so that
4861 * any new writes get down to disk quickly.
4862 */
4863 if (newsize == 0)
72ac3c0d
JB		 4864 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4865 &BTRFS_I(inode)->runtime_flags);
8082510e 4866
f3fe820c
JB		 4867 /*
4868 * 1 for the orphan item we're going to add
4869 * 1 for the orphan item deletion.
4870 */
4871 trans = btrfs_start_transaction(root, 2);
4872 if (IS_ERR(trans))
4873 return PTR_ERR(trans);
4874
4875 /*
4876 * We need to do this in case we fail at _any_ point during the
4877 * actual truncate. Once we do the truncate_setsize we could
4878 * invalidate pages which forces any outstanding ordered io to
4879 * be instantly completed which will give us extents that need
4880 * to be truncated. If we fail to get an orphan inode down we
4881 * could have left over extents that were never meant to live,
4882 * so we need to garuntee from this point on that everything
4883 * will be consistent.
4884 */
4885 ret = btrfs_orphan_add(trans, inode);
4886 btrfs_end_transaction(trans, root);
4887 if (ret)
4888 return ret;
4889
a41ad394
JB		 4890 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4891 truncate_setsize(inode, newsize);
2e60a51e
MX		 4892
4893 /* Disable nonlocked read DIO to avoid the end less truncate */
4894 btrfs_inode_block_unlocked_dio(inode);
4895 inode_dio_wait(inode);
4896 btrfs_inode_resume_unlocked_dio(inode);
4897
a41ad394 4898 ret = btrfs_truncate(inode);
7f4f6e0a
JB		 4899 if (ret && inode->i_nlink) {
4900 int err;
4901
4902 /*
4903 * failed to truncate, disk_i_size is only adjusted down
4904 * as we remove extents, so it should represent the true
4905 * size of the inode, so reset the in memory size and
4906 * delete our orphan entry.
4907 */
4908 trans = btrfs_join_transaction(root);
4909 if (IS_ERR(trans)) {
4910 btrfs_orphan_del(NULL, inode);
4911 return ret;
4912 }
4913 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4914 err = btrfs_orphan_del(trans, inode);
4915 if (err)
4916 btrfs_abort_transaction(trans, root, err);
4917 btrfs_end_transaction(trans, root);
4918 }
8082510e
YZ		 4919 }
4920
a41ad394 4921 return ret;
8082510e
YZ		 4922}
4923
9036c102
YZ		 4924static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4925{
4926 struct inode *inode = dentry->d_inode;
b83cc969 4927 struct btrfs_root *root = BTRFS_I(inode)->root;
9036c102 4928 int err;
39279cc3 4929
b83cc969
LZ		 4930 if (btrfs_root_readonly(root))
4931 return -EROFS;
4932
9036c102
YZ		 4933 err = inode_change_ok(inode, attr);
4934 if (err)
4935 return err;
2bf5a725 4936
5a3f23d5 4937 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3972f260 4938 err = btrfs_setsize(inode, attr);
8082510e
YZ		 4939 if (err)
4940 return err;
39279cc3 4941 }
9036c102 4942
1025774c
CH		 4943 if (attr->ia_valid) {
4944 setattr_copy(inode, attr);
0c4d2d95 4945 inode_inc_iversion(inode);
22c44fe6 4946 err = btrfs_dirty_inode(inode);
1025774c 4947
22c44fe6 4948 if (!err && attr->ia_valid & ATTR_MODE)
996a710d 4949 err = posix_acl_chmod(inode, inode->i_mode);
1025774c 4950 }
33268eaf 4951
39279cc3
CM		 4952 return err;
4953}
61295eb8 4954
131e404a
FDBM		 4955/*
4956 * While truncating the inode pages during eviction, we get the VFS calling
4957 * btrfs_invalidatepage() against each page of the inode. This is slow because
4958 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4959 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4960 * extent_state structures over and over, wasting lots of time.
4961 *
4962 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4963 * those expensive operations on a per page basis and do only the ordered io
4964 * finishing, while we release here the extent_map and extent_state structures,
4965 * without the excessive merging and splitting.
4966 */
4967static void evict_inode_truncate_pages(struct inode *inode)
4968{
4969 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4970 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4971 struct rb_node *node;
4972
4973 ASSERT(inode->i_state & I_FREEING);
91b0abe3 4974 truncate_inode_pages_final(&inode->i_data);
131e404a
FDBM		 4975
4976 write_lock(&map_tree->lock);
4977 while (!RB_EMPTY_ROOT(&map_tree->map)) {
4978 struct extent_map *em;
4979
4980 node = rb_first(&map_tree->map);
4981 em = rb_entry(node, struct extent_map, rb_node);
180589ef
WS		 4982 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4983 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
131e404a
FDBM		 4984 remove_extent_mapping(map_tree, em);
4985 free_extent_map(em);
7064dd5c
FM		 4986 if (need_resched()) {
4987 write_unlock(&map_tree->lock);
4988 cond_resched();
4989 write_lock(&map_tree->lock);
4990 }
131e404a
FDBM		 4991 }
4992 write_unlock(&map_tree->lock);
4993
4994 spin_lock(&io_tree->lock);
4995 while (!RB_EMPTY_ROOT(&io_tree->state)) {
4996 struct extent_state *state;
4997 struct extent_state *cached_state = NULL;
4998
4999 node = rb_first(&io_tree->state);
5000 state = rb_entry(node, struct extent_state, rb_node);
5001 atomic_inc(&state->refs);
5002 spin_unlock(&io_tree->lock);
5003
5004 lock_extent_bits(io_tree, state->start, state->end,
5005 0, &cached_state);
5006 clear_extent_bit(io_tree, state->start, state->end,
5007 EXTENT_LOCKED | EXTENT_DIRTY |
5008 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5009 EXTENT_DEFRAG, 1, 1,
5010 &cached_state, GFP_NOFS);
5011 free_extent_state(state);
5012
7064dd5c 5013 cond_resched();
131e404a
FDBM		 5014 spin_lock(&io_tree->lock);
5015 }
5016 spin_unlock(&io_tree->lock);
5017}
5018
bd555975 5019void btrfs_evict_inode(struct inode *inode)
39279cc3
CM		 5020{
5021 struct btrfs_trans_handle *trans;
5022 struct btrfs_root *root = BTRFS_I(inode)->root;
726c35fa 5023 struct btrfs_block_rsv *rsv, *global_rsv;
3bce876f 5024 int steal_from_global = 0;
07127184 5025 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3
CM		 5026 int ret;
5027
1abe9b8a 5028 trace_btrfs_inode_evict(inode);
5029
131e404a
FDBM		 5030 evict_inode_truncate_pages(inode);
5031
69e9c6c6
SB		 5032 if (inode->i_nlink &&
5033 ((btrfs_root_refs(&root->root_item) != 0 &&
5034 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5035 btrfs_is_free_space_inode(inode)))
bd555975
AV		 5036 goto no_delete;
5037
39279cc3 5038 if (is_bad_inode(inode)) {
7b128766 5039 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 5040 goto no_delete;
5041 }
bd555975 5042 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4a096752 5043 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 5044
f612496b
MX		 5045 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5046
c71bf099 5047 if (root->fs_info->log_root_recovering) {
6bf02314 5048 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8a35d95f 5049 &BTRFS_I(inode)->runtime_flags));
c71bf099
YZ		 5050 goto no_delete;
5051 }
5052
76dda93c 5053 if (inode->i_nlink > 0) {
69e9c6c6
SB		 5054 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5055 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
76dda93c
YZ		 5056 goto no_delete;
5057 }
5058
0e8c36a9
MX		 5059 ret = btrfs_commit_inode_delayed_inode(inode);
5060 if (ret) {
5061 btrfs_orphan_del(NULL, inode);
5062 goto no_delete;
5063 }
5064
66d8f3dd 5065 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4289a667
JB		 5066 if (!rsv) {
5067 btrfs_orphan_del(NULL, inode);
5068 goto no_delete;
5069 }
4a338542 5070 rsv->size = min_size;
ca7e70f5 5071 rsv->failfast = 1;
726c35fa 5072 global_rsv = &root->fs_info->global_block_rsv;
4289a667 5073
dbe674a9 5074 btrfs_i_size_write(inode, 0);
5f39d397 5075
4289a667 5076 /*
8407aa46
MX		 5077 * This is a bit simpler than btrfs_truncate since we've already
5078 * reserved our space for our orphan item in the unlink, so we just
5079 * need to reserve some slack space in case we add bytes and update
5080 * inode item when doing the truncate.
4289a667 5081 */
8082510e 5082 while (1) {
08e007d2
MX		 5083 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5084 BTRFS_RESERVE_FLUSH_LIMIT);
726c35fa
JB		 5085
5086 /*
5087 * Try and steal from the global reserve since we will
5088 * likely not use this space anyway, we want to try as
5089 * hard as possible to get this to work.
5090 */
5091 if (ret)
3bce876f
JB		 5092 steal_from_global++;
5093 else
5094 steal_from_global = 0;
5095 ret = 0;
d68fc57b 5096
3bce876f
JB		 5097 /*
5098 * steal_from_global == 0: we reserved stuff, hooray!
5099 * steal_from_global == 1: we didn't reserve stuff, boo!
5100 * steal_from_global == 2: we've committed, still not a lot of
5101 * room but maybe we'll have room in the global reserve this
5102 * time.
5103 * steal_from_global == 3: abandon all hope!
5104 */
5105 if (steal_from_global > 2) {
c2cf52eb
SK		 5106 btrfs_warn(root->fs_info,
5107 "Could not get space for a delete, will truncate on mount %d",
5108 ret);
4289a667
JB		 5109 btrfs_orphan_del(NULL, inode);
5110 btrfs_free_block_rsv(root, rsv);
5111 goto no_delete;
d68fc57b 5112 }
7b128766 5113
0e8c36a9 5114 trans = btrfs_join_transaction(root);
4289a667
JB		 5115 if (IS_ERR(trans)) {
5116 btrfs_orphan_del(NULL, inode);
5117 btrfs_free_block_rsv(root, rsv);
5118 goto no_delete;
d68fc57b 5119 }
7b128766 5120
3bce876f
JB		 5121 /*
5122 * We can't just steal from the global reserve, we need tomake
5123 * sure there is room to do it, if not we need to commit and try
5124 * again.
5125 */
5126 if (steal_from_global) {
5127 if (!btrfs_check_space_for_delayed_refs(trans, root))
5128 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5129 min_size);
5130 else
5131 ret = -ENOSPC;
5132 }
5133
5134 /*
5135 * Couldn't steal from the global reserve, we have too much
5136 * pending stuff built up, commit the transaction and try it
5137 * again.
5138 */
5139 if (ret) {
5140 ret = btrfs_commit_transaction(trans, root);
5141 if (ret) {
5142 btrfs_orphan_del(NULL, inode);
5143 btrfs_free_block_rsv(root, rsv);
5144 goto no_delete;
5145 }
5146 continue;
5147 } else {
5148 steal_from_global = 0;
5149 }
5150
4289a667
JB		 5151 trans->block_rsv = rsv;
5152
d68fc57b 5153 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
28ed1345 5154 if (ret != -ENOSPC && ret != -EAGAIN)
8082510e 5155 break;
85e21bac 5156
8407aa46 5157 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e
YZ		 5158 btrfs_end_transaction(trans, root);
5159 trans = NULL;
b53d3f5d 5160 btrfs_btree_balance_dirty(root);
8082510e 5161 }
5f39d397 5162
4289a667
JB		 5163 btrfs_free_block_rsv(root, rsv);
5164
4ef31a45
JB		 5165 /*
5166 * Errors here aren't a big deal, it just means we leave orphan items
5167 * in the tree. They will be cleaned up on the next mount.
5168 */
8082510e 5169 if (ret == 0) {
4289a667 5170 trans->block_rsv = root->orphan_block_rsv;
4ef31a45
JB		 5171 btrfs_orphan_del(trans, inode);
5172 } else {
5173 btrfs_orphan_del(NULL, inode);
8082510e 5174 }
54aa1f4d 5175
4289a667 5176 trans->block_rsv = &root->fs_info->trans_block_rsv;
581bb050
LZ		 5177 if (!(root == root->fs_info->tree_root ||
5178 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
33345d01 5179 btrfs_return_ino(root, btrfs_ino(inode));
581bb050 5180
54aa1f4d 5181 btrfs_end_transaction(trans, root);
b53d3f5d 5182 btrfs_btree_balance_dirty(root);
39279cc3 5183no_delete:
89042e5a 5184 btrfs_remove_delayed_node(inode);
dbd5768f 5185 clear_inode(inode);
8082510e 5186 return;
39279cc3
CM		 5187}
5188
5189/*
5190 * this returns the key found in the dir entry in the location pointer.
5191 * If no dir entries were found, location->objectid is 0.
5192 */
5193static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5194 struct btrfs_key *location)
5195{
5196 const char *name = dentry->d_name.name;
5197 int namelen = dentry->d_name.len;
5198 struct btrfs_dir_item *di;
5199 struct btrfs_path *path;
5200 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 5201 int ret = 0;
39279cc3
CM		 5202
5203 path = btrfs_alloc_path();
d8926bb3
MF		 5204 if (!path)
5205 return -ENOMEM;
3954401f 5206
33345d01 5207 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
39279cc3 5208 namelen, 0);
0d9f7f3e
Y		 5209 if (IS_ERR(di))
5210 ret = PTR_ERR(di);
d397712b 5211
c704005d 5212 if (IS_ERR_OR_NULL(di))
3954401f 5213 goto out_err;
d397712b 5214
5f39d397 5215 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 5216out:
39279cc3
CM		 5217 btrfs_free_path(path);
5218 return ret;
3954401f
CM		 5219out_err:
5220 location->objectid = 0;
5221 goto out;
39279cc3
CM		 5222}
5223
5224/*
5225 * when we hit a tree root in a directory, the btrfs part of the inode
5226 * needs to be changed to reflect the root directory of the tree root. This
5227 * is kind of like crossing a mount point.
5228 */
5229static int fixup_tree_root_location(struct btrfs_root *root,
4df27c4d
YZ		 5230 struct inode *dir,
5231 struct dentry *dentry,
5232 struct btrfs_key *location,
5233 struct btrfs_root **sub_root)
39279cc3 5234{
4df27c4d
YZ		 5235 struct btrfs_path *path;
5236 struct btrfs_root *new_root;
5237 struct btrfs_root_ref *ref;
5238 struct extent_buffer *leaf;
1d4c08e0 5239 struct btrfs_key key;
4df27c4d
YZ		 5240 int ret;
5241 int err = 0;
39279cc3 5242
4df27c4d
YZ		 5243 path = btrfs_alloc_path();
5244 if (!path) {
5245 err = -ENOMEM;
5246 goto out;
5247 }
39279cc3 5248
4df27c4d 5249 err = -ENOENT;
1d4c08e0
DS		 5250 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5251 key.type = BTRFS_ROOT_REF_KEY;
5252 key.offset = location->objectid;
5253
5254 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5255 0, 0);
4df27c4d
YZ		 5256 if (ret) {
5257 if (ret < 0)
5258 err = ret;
5259 goto out;
5260 }
39279cc3 5261
4df27c4d
YZ		 5262 leaf = path->nodes[0];
5263 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
33345d01 5264 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4df27c4d
YZ		 5265 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5266 goto out;
39279cc3 5267
4df27c4d
YZ		 5268 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5269 (unsigned long)(ref + 1),
5270 dentry->d_name.len);
5271 if (ret)
5272 goto out;
5273
b3b4aa74 5274 btrfs_release_path(path);
4df27c4d
YZ		 5275
5276 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5277 if (IS_ERR(new_root)) {
5278 err = PTR_ERR(new_root);
5279 goto out;
5280 }
5281
4df27c4d
YZ		 5282 *sub_root = new_root;
5283 location->objectid = btrfs_root_dirid(&new_root->root_item);
5284 location->type = BTRFS_INODE_ITEM_KEY;
5285 location->offset = 0;
5286 err = 0;
5287out:
5288 btrfs_free_path(path);
5289 return err;
39279cc3
CM		 5290}
5291
5d4f98a2
YZ		 5292static void inode_tree_add(struct inode *inode)
5293{
5294 struct btrfs_root *root = BTRFS_I(inode)->root;
5295 struct btrfs_inode *entry;
03e860bd
FNP		 5296 struct rb_node **p;
5297 struct rb_node *parent;
cef21937 5298 struct rb_node *new = &BTRFS_I(inode)->rb_node;
33345d01 5299 u64 ino = btrfs_ino(inode);
5d4f98a2 5300
1d3382cb 5301 if (inode_unhashed(inode))
76dda93c 5302 return;
e1409cef 5303 parent = NULL;
5d4f98a2 5304 spin_lock(&root->inode_lock);
e1409cef 5305 p = &root->inode_tree.rb_node;
5d4f98a2
YZ		 5306 while (*p) {
5307 parent = *p;
5308 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5309
33345d01 5310 if (ino < btrfs_ino(&entry->vfs_inode))
03e860bd 5311 p = &parent->rb_left;
33345d01 5312 else if (ino > btrfs_ino(&entry->vfs_inode))
03e860bd 5313 p = &parent->rb_right;
5d4f98a2
YZ		 5314 else {
5315 WARN_ON(!(entry->vfs_inode.i_state &
a4ffdde6 5316 (I_WILL_FREE | I_FREEING)));
cef21937 5317 rb_replace_node(parent, new, &root->inode_tree);
03e860bd
FNP		 5318 RB_CLEAR_NODE(parent);
5319 spin_unlock(&root->inode_lock);
cef21937 5320 return;
5d4f98a2
YZ		 5321 }
5322 }
cef21937
FDBM		 5323 rb_link_node(new, parent, p);
5324 rb_insert_color(new, &root->inode_tree);
5d4f98a2
YZ		 5325 spin_unlock(&root->inode_lock);
5326}
5327
5328static void inode_tree_del(struct inode *inode)
5329{
5330 struct btrfs_root *root = BTRFS_I(inode)->root;
76dda93c 5331 int empty = 0;
5d4f98a2 5332
03e860bd 5333 spin_lock(&root->inode_lock);
5d4f98a2 5334 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5d4f98a2 5335 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5d4f98a2 5336 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
76dda93c 5337 empty = RB_EMPTY_ROOT(&root->inode_tree);
5d4f98a2 5338 }
03e860bd 5339 spin_unlock(&root->inode_lock);
76dda93c 5340
69e9c6c6 5341 if (empty && btrfs_root_refs(&root->root_item) == 0) {
76dda93c
YZ		 5342 synchronize_srcu(&root->fs_info->subvol_srcu);
5343 spin_lock(&root->inode_lock);
5344 empty = RB_EMPTY_ROOT(&root->inode_tree);
5345 spin_unlock(&root->inode_lock);
5346 if (empty)
5347 btrfs_add_dead_root(root);
5348 }
5349}
5350
143bede5 5351void btrfs_invalidate_inodes(struct btrfs_root *root)
76dda93c
YZ		 5352{
5353 struct rb_node *node;
5354 struct rb_node *prev;
5355 struct btrfs_inode *entry;
5356 struct inode *inode;
5357 u64 objectid = 0;
5358
7813b3db
LB		 5359 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5360 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
76dda93c
YZ		 5361
5362 spin_lock(&root->inode_lock);
5363again:
5364 node = root->inode_tree.rb_node;
5365 prev = NULL;
5366 while (node) {
5367 prev = node;
5368 entry = rb_entry(node, struct btrfs_inode, rb_node);
5369
33345d01 5370 if (objectid < btrfs_ino(&entry->vfs_inode))
76dda93c 5371 node = node->rb_left;
33345d01 5372 else if (objectid > btrfs_ino(&entry->vfs_inode))
76dda93c
YZ		 5373 node = node->rb_right;
5374 else
5375 break;
5376 }
5377 if (!node) {
5378 while (prev) {
5379 entry = rb_entry(prev, struct btrfs_inode, rb_node);
33345d01 5380 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
76dda93c
YZ		 5381 node = prev;
5382 break;
5383 }
5384 prev = rb_next(prev);
5385 }
5386 }
5387 while (node) {
5388 entry = rb_entry(node, struct btrfs_inode, rb_node);
33345d01 5389 objectid = btrfs_ino(&entry->vfs_inode) + 1;
76dda93c
YZ		 5390 inode = igrab(&entry->vfs_inode);
5391 if (inode) {
5392 spin_unlock(&root->inode_lock);
5393 if (atomic_read(&inode->i_count) > 1)
5394 d_prune_aliases(inode);
5395 /*
45321ac5 5396 * btrfs_drop_inode will have it removed from
76dda93c
YZ		 5397 * the inode cache when its usage count
5398 * hits zero.
5399 */
5400 iput(inode);
5401 cond_resched();
5402 spin_lock(&root->inode_lock);
5403 goto again;
5404 }
5405
5406 if (cond_resched_lock(&root->inode_lock))
5407 goto again;
5408
5409 node = rb_next(node);
5410 }
5411 spin_unlock(&root->inode_lock);
5d4f98a2
YZ		 5412}
5413
e02119d5
CM		 5414static int btrfs_init_locked_inode(struct inode *inode, void *p)
5415{
5416 struct btrfs_iget_args *args = p;
90d3e592
CM		 5417 inode->i_ino = args->location->objectid;
5418 memcpy(&BTRFS_I(inode)->location, args->location,
5419 sizeof(*args->location));
e02119d5 5420 BTRFS_I(inode)->root = args->root;
39279cc3
CM		 5421 return 0;
5422}
5423
5424static int btrfs_find_actor(struct inode *inode, void *opaque)
5425{
5426 struct btrfs_iget_args *args = opaque;
90d3e592 5427 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
d397712b 5428 args->root == BTRFS_I(inode)->root;
39279cc3
CM		 5429}
5430
5d4f98a2 5431static struct inode *btrfs_iget_locked(struct super_block *s,
90d3e592 5432 struct btrfs_key *location,
5d4f98a2 5433 struct btrfs_root *root)
39279cc3
CM		 5434{
5435 struct inode *inode;
5436 struct btrfs_iget_args args;
90d3e592 5437 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
778ba82b 5438
90d3e592 5439 args.location = location;
39279cc3
CM		 5440 args.root = root;
5441
778ba82b 5442 inode = iget5_locked(s, hashval, btrfs_find_actor,
39279cc3
CM		 5443 btrfs_init_locked_inode,
5444 (void *)&args);
5445 return inode;
5446}
5447
1a54ef8c
BR		 5448/* Get an inode object given its location and corresponding root.
5449 * Returns in *is_new if the inode was read from disk
5450 */
5451struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
73f73415 5452 struct btrfs_root *root, int *new)
1a54ef8c
BR		 5453{
5454 struct inode *inode;
5455
90d3e592 5456 inode = btrfs_iget_locked(s, location, root);
1a54ef8c 5457 if (!inode)
5d4f98a2 5458 return ERR_PTR(-ENOMEM);
1a54ef8c
BR		 5459
5460 if (inode->i_state & I_NEW) {
1a54ef8c 5461 btrfs_read_locked_inode(inode);
1748f843
MF		 5462 if (!is_bad_inode(inode)) {
5463 inode_tree_add(inode);
5464 unlock_new_inode(inode);
5465 if (new)
5466 *new = 1;
5467 } else {
e0b6d65b
ST		 5468 unlock_new_inode(inode);
5469 iput(inode);
5470 inode = ERR_PTR(-ESTALE);
1748f843
MF		 5471 }
5472 }
5473
1a54ef8c
BR		 5474 return inode;
5475}
5476
4df27c4d
YZ		 5477static struct inode *new_simple_dir(struct super_block *s,
5478 struct btrfs_key *key,
5479 struct btrfs_root *root)
5480{
5481 struct inode *inode = new_inode(s);
5482
5483 if (!inode)
5484 return ERR_PTR(-ENOMEM);
5485
4df27c4d
YZ		 5486 BTRFS_I(inode)->root = root;
5487 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
72ac3c0d 5488 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4df27c4d
YZ		 5489
5490 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
848cce0d 5491 inode->i_op = &btrfs_dir_ro_inode_operations;
4df27c4d
YZ		 5492 inode->i_fop = &simple_dir_operations;
5493 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
9cc97d64 5494 inode->i_mtime = CURRENT_TIME;
5495 inode->i_atime = inode->i_mtime;
5496 inode->i_ctime = inode->i_mtime;
5497 BTRFS_I(inode)->i_otime = inode->i_mtime;
4df27c4d
YZ		 5498
5499 return inode;
5500}
5501
3de4586c 5502struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3 5503{
d397712b 5504 struct inode *inode;
4df27c4d 5505 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3
CM		 5506 struct btrfs_root *sub_root = root;
5507 struct btrfs_key location;
76dda93c 5508 int index;
b4aff1f8 5509 int ret = 0;
39279cc3
CM		 5510
5511 if (dentry->d_name.len > BTRFS_NAME_LEN)
5512 return ERR_PTR(-ENAMETOOLONG);
5f39d397 5513
39e3c955 5514 ret = btrfs_inode_by_name(dir, dentry, &location);
39279cc3
CM		 5515 if (ret < 0)
5516 return ERR_PTR(ret);
5f39d397 5517
4df27c4d 5518 if (location.objectid == 0)
5662344b 5519 return ERR_PTR(-ENOENT);
4df27c4d
YZ		 5520
5521 if (location.type == BTRFS_INODE_ITEM_KEY) {
73f73415 5522 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4df27c4d
YZ		 5523 return inode;
5524 }
5525
5526 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5527
76dda93c 5528 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4df27c4d
YZ		 5529 ret = fixup_tree_root_location(root, dir, dentry,
5530 &location, &sub_root);
5531 if (ret < 0) {
5532 if (ret != -ENOENT)
5533 inode = ERR_PTR(ret);
5534 else
5535 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5536 } else {
73f73415 5537 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
39279cc3 5538 }
76dda93c
YZ		 5539 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5540
34d19bad 5541 if (!IS_ERR(inode) && root != sub_root) {
c71bf099
YZ		 5542 down_read(&root->fs_info->cleanup_work_sem);
5543 if (!(inode->i_sb->s_flags & MS_RDONLY))
66b4ffd1 5544 ret = btrfs_orphan_cleanup(sub_root);
c71bf099 5545 up_read(&root->fs_info->cleanup_work_sem);
01cd3367
JB		 5546 if (ret) {
5547 iput(inode);
66b4ffd1 5548 inode = ERR_PTR(ret);
01cd3367 5549 }
c71bf099
YZ		 5550 }
5551
3de4586c
CM		 5552 return inode;
5553}
5554
fe15ce44 5555static int btrfs_dentry_delete(const struct dentry *dentry)
76dda93c
YZ		 5556{
5557 struct btrfs_root *root;
848cce0d 5558 struct inode *inode = dentry->d_inode;
76dda93c 5559
848cce0d
LZ		 5560 if (!inode && !IS_ROOT(dentry))
5561 inode = dentry->d_parent->d_inode;
76dda93c 5562
848cce0d
LZ		 5563 if (inode) {
5564 root = BTRFS_I(inode)->root;
efefb143
YZ		 5565 if (btrfs_root_refs(&root->root_item) == 0)
5566 return 1;
848cce0d
LZ		 5567
5568 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5569 return 1;
efefb143 5570 }
76dda93c
YZ		 5571 return 0;
5572}
5573
b4aff1f8
JB		 5574static void btrfs_dentry_release(struct dentry *dentry)
5575{
944a4515 5576 kfree(dentry->d_fsdata);
b4aff1f8
JB		 5577}
5578
3de4586c 5579static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
00cd8dd3 5580 unsigned int flags)
3de4586c 5581{
5662344b 5582 struct inode *inode;
a66e7cc6 5583
5662344b
TI		 5584 inode = btrfs_lookup_dentry(dir, dentry);
5585 if (IS_ERR(inode)) {
5586 if (PTR_ERR(inode) == -ENOENT)
5587 inode = NULL;
5588 else
5589 return ERR_CAST(inode);
5590 }
5591
41d28bca 5592 return d_splice_alias(inode, dentry);
39279cc3
CM		 5593}
5594
16cdcec7 5595unsigned char btrfs_filetype_table[] = {
39279cc3
CM		 5596 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5597};
5598
9cdda8d3 5599static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
39279cc3 5600{
9cdda8d3 5601 struct inode *inode = file_inode(file);
39279cc3
CM		 5602 struct btrfs_root *root = BTRFS_I(inode)->root;
5603 struct btrfs_item *item;
5604 struct btrfs_dir_item *di;
5605 struct btrfs_key key;
5f39d397 5606 struct btrfs_key found_key;
39279cc3 5607 struct btrfs_path *path;
16cdcec7
MX		 5608 struct list_head ins_list;
5609 struct list_head del_list;
39279cc3 5610 int ret;
5f39d397 5611 struct extent_buffer *leaf;
39279cc3 5612 int slot;
39279cc3
CM		 5613 unsigned char d_type;
5614 int over = 0;
5615 u32 di_cur;
5616 u32 di_total;
5617 u32 di_len;
5618 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 5619 char tmp_name[32];
5620 char *name_ptr;
5621 int name_len;
9cdda8d3 5622 int is_curr = 0; /* ctx->pos points to the current index? */
39279cc3
CM		 5623
5624 /* FIXME, use a real flag for deciding about the key type */
5625 if (root->fs_info->tree_root == root)
5626 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 5627
9cdda8d3
AV		 5628 if (!dir_emit_dots(file, ctx))
5629 return 0;
5630
49593bfa 5631 path = btrfs_alloc_path();
16cdcec7
MX		 5632 if (!path)
5633 return -ENOMEM;
ff5714cc 5634
026fd317 5635 path->reada = 1;
49593bfa 5636
16cdcec7
MX		 5637 if (key_type == BTRFS_DIR_INDEX_KEY) {
5638 INIT_LIST_HEAD(&ins_list);
5639 INIT_LIST_HEAD(&del_list);
5640 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5641 }
5642
962a298f 5643 key.type = key_type;
9cdda8d3 5644 key.offset = ctx->pos;
33345d01 5645 key.objectid = btrfs_ino(inode);
5f39d397 5646
39279cc3
CM		 5647 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5648 if (ret < 0)
5649 goto err;
49593bfa
DW		 5650
5651 while (1) {
5f39d397 5652 leaf = path->nodes[0];
39279cc3 5653 slot = path->slots[0];
b9e03af0
LZ		 5654 if (slot >= btrfs_header_nritems(leaf)) {
5655 ret = btrfs_next_leaf(root, path);
5656 if (ret < 0)
5657 goto err;
5658 else if (ret > 0)
5659 break;
5660 continue;
39279cc3 5661 }
3de4586c 5662
dd3cc16b 5663 item = btrfs_item_nr(slot);
5f39d397
CM		 5664 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5665
5666 if (found_key.objectid != key.objectid)
39279cc3 5667 break;
962a298f 5668 if (found_key.type != key_type)
39279cc3 5669 break;
9cdda8d3 5670 if (found_key.offset < ctx->pos)
b9e03af0 5671 goto next;
16cdcec7
MX		 5672 if (key_type == BTRFS_DIR_INDEX_KEY &&
5673 btrfs_should_delete_dir_index(&del_list,
5674 found_key.offset))
5675 goto next;
5f39d397 5676
9cdda8d3 5677 ctx->pos = found_key.offset;
16cdcec7 5678 is_curr = 1;
49593bfa 5679
39279cc3
CM		 5680 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5681 di_cur = 0;
5f39d397 5682 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 5683
5684 while (di_cur < di_total) {
5f39d397
CM		 5685 struct btrfs_key location;
5686
22a94d44
JB		 5687 if (verify_dir_item(root, leaf, di))
5688 break;
5689
5f39d397 5690 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 5691 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 5692 name_ptr = tmp_name;
5693 } else {
5694 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 5695 if (!name_ptr) {
5696 ret = -ENOMEM;
5697 goto err;
5698 }
5f39d397
CM		 5699 }
5700 read_extent_buffer(leaf, name_ptr,
5701 (unsigned long)(di + 1), name_len);
5702
5703 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5704 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c 5705
fede766f 5706
3de4586c 5707 /* is this a reference to our own snapshot? If so
8c9c2bf7
AJ		 5708 * skip it.
5709 *
5710 * In contrast to old kernels, we insert the snapshot's
5711 * dir item and dir index after it has been created, so
5712 * we won't find a reference to our own snapshot. We
5713 * still keep the following code for backward
5714 * compatibility.
3de4586c
CM		 5715 */
5716 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5717 location.objectid == root->root_key.objectid) {
5718 over = 0;
5719 goto skip;
5720 }
9cdda8d3
AV		 5721 over = !dir_emit(ctx, name_ptr, name_len,
5722 location.objectid, d_type);
5f39d397 5723
3de4586c 5724skip:
5f39d397
CM		 5725 if (name_ptr != tmp_name)
5726 kfree(name_ptr);
5727
39279cc3
CM		 5728 if (over)
5729 goto nopos;
5103e947 5730 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 5731 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 5732 di_cur += di_len;
5733 di = (struct btrfs_dir_item *)((char *)di + di_len);
5734 }
b9e03af0
LZ		 5735next:
5736 path->slots[0]++;
39279cc3 5737 }
49593bfa 5738
16cdcec7
MX		 5739 if (key_type == BTRFS_DIR_INDEX_KEY) {
5740 if (is_curr)
9cdda8d3
AV		 5741 ctx->pos++;
5742 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
16cdcec7
MX		 5743 if (ret)
5744 goto nopos;
5745 }
5746
49593bfa 5747 /* Reached end of directory/root. Bump pos past the last item. */
db62efbb
ZB		 5748 ctx->pos++;
5749
5750 /*
5751 * Stop new entries from being returned after we return the last
5752 * entry.
5753 *
5754 * New directory entries are assigned a strictly increasing
5755 * offset. This means that new entries created during readdir
5756 * are *guaranteed* to be seen in the future by that readdir.
5757 * This has broken buggy programs which operate on names as
5758 * they're returned by readdir. Until we re-use freed offsets
5759 * we have this hack to stop new entries from being returned
5760 * under the assumption that they'll never reach this huge
5761 * offset.
5762 *
5763 * This is being careful not to overflow 32bit loff_t unless the
5764 * last entry requires it because doing so has broken 32bit apps
5765 * in the past.
5766 */
5767 if (key_type == BTRFS_DIR_INDEX_KEY) {
5768 if (ctx->pos >= INT_MAX)
5769 ctx->pos = LLONG_MAX;
5770 else
5771 ctx->pos = INT_MAX;
5772 }
39279cc3
CM		 5773nopos:
5774 ret = 0;
5775err:
16cdcec7
MX		 5776 if (key_type == BTRFS_DIR_INDEX_KEY)
5777 btrfs_put_delayed_items(&ins_list, &del_list);
39279cc3 5778 btrfs_free_path(path);
39279cc3
CM		 5779 return ret;
5780}
5781
a9185b41 5782int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
39279cc3
CM		 5783{
5784 struct btrfs_root *root = BTRFS_I(inode)->root;
5785 struct btrfs_trans_handle *trans;
5786 int ret = 0;
0af3d00b 5787 bool nolock = false;
39279cc3 5788
72ac3c0d 5789 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4ca8b41e
CM		 5790 return 0;
5791
83eea1f1 5792 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
82d5902d 5793 nolock = true;
0af3d00b 5794
a9185b41 5795 if (wbc->sync_mode == WB_SYNC_ALL) {
0af3d00b 5796 if (nolock)
7a7eaa40 5797 trans = btrfs_join_transaction_nolock(root);
0af3d00b 5798 else
7a7eaa40 5799 trans = btrfs_join_transaction(root);
3612b495
TI		 5800 if (IS_ERR(trans))
5801 return PTR_ERR(trans);
a698d075 5802 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 5803 }
5804 return ret;
5805}
5806
5807/*
54aa1f4d 5808 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 5809 * inode changes. But, it is most likely to find the inode in cache.
5810 * FIXME, needs more benchmarking...there are no reasons other than performance
5811 * to keep or drop this code.
5812 */
48a3b636 5813static int btrfs_dirty_inode(struct inode *inode)
39279cc3
CM		 5814{
5815 struct btrfs_root *root = BTRFS_I(inode)->root;
5816 struct btrfs_trans_handle *trans;
8929ecfa
YZ		 5817 int ret;
5818
72ac3c0d 5819 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
22c44fe6 5820 return 0;
39279cc3 5821
7a7eaa40 5822 trans = btrfs_join_transaction(root);
22c44fe6
JB		 5823 if (IS_ERR(trans))
5824 return PTR_ERR(trans);
8929ecfa
YZ		 5825
5826 ret = btrfs_update_inode(trans, root, inode);
94b60442
CM		 5827 if (ret && ret == -ENOSPC) {
5828 /* whoops, lets try again with the full transaction */
5829 btrfs_end_transaction(trans, root);
5830 trans = btrfs_start_transaction(root, 1);
22c44fe6
JB		 5831 if (IS_ERR(trans))
5832 return PTR_ERR(trans);
8929ecfa 5833
94b60442 5834 ret = btrfs_update_inode(trans, root, inode);
94b60442 5835 }
39279cc3 5836 btrfs_end_transaction(trans, root);
16cdcec7
MX		 5837 if (BTRFS_I(inode)->delayed_node)
5838 btrfs_balance_delayed_items(root);
22c44fe6
JB		 5839
5840 return ret;
5841}
5842
5843/*
5844 * This is a copy of file_update_time. We need this so we can return error on
5845 * ENOSPC for updating the inode in the case of file write and mmap writes.
5846 */
e41f941a
JB		 5847static int btrfs_update_time(struct inode *inode, struct timespec *now,
5848 int flags)
22c44fe6 5849{
2bc55652
AB		 5850 struct btrfs_root *root = BTRFS_I(inode)->root;
5851
5852 if (btrfs_root_readonly(root))
5853 return -EROFS;
5854
e41f941a 5855 if (flags & S_VERSION)
22c44fe6 5856 inode_inc_iversion(inode);
e41f941a
JB		 5857 if (flags & S_CTIME)
5858 inode->i_ctime = *now;
5859 if (flags & S_MTIME)
5860 inode->i_mtime = *now;
5861 if (flags & S_ATIME)
5862 inode->i_atime = *now;
5863 return btrfs_dirty_inode(inode);
39279cc3
CM		 5864}
5865
d352ac68
CM		 5866/*
5867 * find the highest existing sequence number in a directory
5868 * and then set the in-memory index_cnt variable to reflect
5869 * free sequence numbers
5870 */
aec7477b
JB		 5871static int btrfs_set_inode_index_count(struct inode *inode)
5872{
5873 struct btrfs_root *root = BTRFS_I(inode)->root;
5874 struct btrfs_key key, found_key;
5875 struct btrfs_path *path;
5876 struct extent_buffer *leaf;
5877 int ret;
5878
33345d01 5879 key.objectid = btrfs_ino(inode);
962a298f 5880 key.type = BTRFS_DIR_INDEX_KEY;
aec7477b
JB		 5881 key.offset = (u64)-1;
5882
5883 path = btrfs_alloc_path();
5884 if (!path)
5885 return -ENOMEM;
5886
5887 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5888 if (ret < 0)
5889 goto out;
5890 /* FIXME: we should be able to handle this */
5891 if (ret == 0)
5892 goto out;
5893 ret = 0;
5894
5895 /*
5896 * MAGIC NUMBER EXPLANATION:
5897 * since we search a directory based on f_pos we have to start at 2
5898 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5899 * else has to start at 2
5900 */
5901 if (path->slots[0] == 0) {
5902 BTRFS_I(inode)->index_cnt = 2;
5903 goto out;
5904 }
5905
5906 path->slots[0]--;
5907
5908 leaf = path->nodes[0];
5909 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5910
33345d01 5911 if (found_key.objectid != btrfs_ino(inode) ||
962a298f 5912 found_key.type != BTRFS_DIR_INDEX_KEY) {
aec7477b
JB		 5913 BTRFS_I(inode)->index_cnt = 2;
5914 goto out;
5915 }
5916
5917 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5918out:
5919 btrfs_free_path(path);
5920 return ret;
5921}
5922
d352ac68
CM		 5923/*
5924 * helper to find a free sequence number in a given directory. This current
5925 * code is very simple, later versions will do smarter things in the btree
5926 */
3de4586c 5927int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 5928{
5929 int ret = 0;
5930
5931 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
16cdcec7
MX		 5932 ret = btrfs_inode_delayed_dir_index_count(dir);
5933 if (ret) {
5934 ret = btrfs_set_inode_index_count(dir);
5935 if (ret)
5936 return ret;
5937 }
aec7477b
JB		 5938 }
5939
00e4e6b3 5940 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 5941 BTRFS_I(dir)->index_cnt++;
5942
5943 return ret;
5944}
5945
b0d5d10f
CM		 5946static int btrfs_insert_inode_locked(struct inode *inode)
5947{
5948 struct btrfs_iget_args args;
5949 args.location = &BTRFS_I(inode)->location;
5950 args.root = BTRFS_I(inode)->root;
5951
5952 return insert_inode_locked4(inode,
5953 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
5954 btrfs_find_actor, &args);
5955}
5956
39279cc3
CM		 5957static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5958 struct btrfs_root *root,
aec7477b 5959 struct inode *dir,
9c58309d 5960 const char *name, int name_len,
175a4eb7
AV		 5961 u64 ref_objectid, u64 objectid,
5962 umode_t mode, u64 *index)
39279cc3
CM		 5963{
5964 struct inode *inode;
5f39d397 5965 struct btrfs_inode_item *inode_item;
39279cc3 5966 struct btrfs_key *location;
5f39d397 5967 struct btrfs_path *path;
9c58309d
CM		 5968 struct btrfs_inode_ref *ref;
5969 struct btrfs_key key[2];
5970 u32 sizes[2];
ef3b9af5 5971 int nitems = name ? 2 : 1;
9c58309d 5972 unsigned long ptr;
39279cc3 5973 int ret;
39279cc3 5974
5f39d397 5975 path = btrfs_alloc_path();
d8926bb3
MF		 5976 if (!path)
5977 return ERR_PTR(-ENOMEM);
5f39d397 5978
39279cc3 5979 inode = new_inode(root->fs_info->sb);
8fb27640
YS		 5980 if (!inode) {
5981 btrfs_free_path(path);
39279cc3 5982 return ERR_PTR(-ENOMEM);
8fb27640 5983 }
39279cc3 5984
5762b5c9
FM		 5985 /*
5986 * O_TMPFILE, set link count to 0, so that after this point,
5987 * we fill in an inode item with the correct link count.
5988 */
5989 if (!name)
5990 set_nlink(inode, 0);
5991
581bb050
LZ		 5992 /*
5993 * we have to initialize this early, so we can reclaim the inode
5994 * number if we fail afterwards in this function.
5995 */
5996 inode->i_ino = objectid;
5997
ef3b9af5 5998 if (dir && name) {
1abe9b8a 5999 trace_btrfs_inode_request(dir);
6000
3de4586c 6001 ret = btrfs_set_inode_index(dir, index);
09771430 6002 if (ret) {
8fb27640 6003 btrfs_free_path(path);
09771430 6004 iput(inode);
aec7477b 6005 return ERR_PTR(ret);
09771430 6006 }
ef3b9af5
FM		 6007 } else if (dir) {
6008 *index = 0;
aec7477b
JB		 6009 }
6010 /*
6011 * index_cnt is ignored for everything but a dir,
6012 * btrfs_get_inode_index_count has an explanation for the magic
6013 * number
6014 */
6015 BTRFS_I(inode)->index_cnt = 2;
67de1176 6016 BTRFS_I(inode)->dir_index = *index;
39279cc3 6017 BTRFS_I(inode)->root = root;
e02119d5 6018 BTRFS_I(inode)->generation = trans->transid;
76195853 6019 inode->i_generation = BTRFS_I(inode)->generation;
b888db2b 6020
5dc562c5
JB		 6021 /*
6022 * We could have gotten an inode number from somebody who was fsynced
6023 * and then removed in this same transaction, so let's just set full
6024 * sync since it will be a full sync anyway and this will blow away the
6025 * old info in the log.
6026 */
6027 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6028
9c58309d 6029 key[0].objectid = objectid;
962a298f 6030 key[0].type = BTRFS_INODE_ITEM_KEY;
9c58309d
CM		 6031 key[0].offset = 0;
6032
9c58309d 6033 sizes[0] = sizeof(struct btrfs_inode_item);
ef3b9af5
FM		 6034
6035 if (name) {
6036 /*
6037 * Start new inodes with an inode_ref. This is slightly more
6038 * efficient for small numbers of hard links since they will
6039 * be packed into one item. Extended refs will kick in if we
6040 * add more hard links than can fit in the ref item.
6041 */
6042 key[1].objectid = objectid;
962a298f 6043 key[1].type = BTRFS_INODE_REF_KEY;
ef3b9af5
FM		 6044 key[1].offset = ref_objectid;
6045
6046 sizes[1] = name_len + sizeof(*ref);
6047 }
9c58309d 6048
b0d5d10f
CM		 6049 location = &BTRFS_I(inode)->location;
6050 location->objectid = objectid;
6051 location->offset = 0;
962a298f 6052 location->type = BTRFS_INODE_ITEM_KEY;
b0d5d10f
CM		 6053
6054 ret = btrfs_insert_inode_locked(inode);
6055 if (ret < 0)
6056 goto fail;
6057
b9473439 6058 path->leave_spinning = 1;
ef3b9af5 6059 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
9c58309d 6060 if (ret != 0)
b0d5d10f 6061 goto fail_unlock;
5f39d397 6062
ecc11fab 6063 inode_init_owner(inode, dir, mode);
a76a3cd4 6064 inode_set_bytes(inode, 0);
9cc97d64 6065
6066 inode->i_mtime = CURRENT_TIME;
6067 inode->i_atime = inode->i_mtime;
6068 inode->i_ctime = inode->i_mtime;
6069 BTRFS_I(inode)->i_otime = inode->i_mtime;
6070
5f39d397
CM		 6071 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6072 struct btrfs_inode_item);
293f7e07
LZ		 6073 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6074 sizeof(*inode_item));
e02119d5 6075 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d 6076
ef3b9af5
FM		 6077 if (name) {
6078 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6079 struct btrfs_inode_ref);
6080 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6081 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6082 ptr = (unsigned long)(ref + 1);
6083 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6084 }
9c58309d 6085
5f39d397
CM		 6086 btrfs_mark_buffer_dirty(path->nodes[0]);
6087 btrfs_free_path(path);
6088
6cbff00f
CH		 6089 btrfs_inherit_iflags(inode, dir);
6090
569254b0 6091 if (S_ISREG(mode)) {
94272164
CM		 6092 if (btrfs_test_opt(root, NODATASUM))
6093 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
213490b3 6094 if (btrfs_test_opt(root, NODATACOW))
f2bdf9a8
JB		 6095 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6096 BTRFS_INODE_NODATASUM;
94272164
CM		 6097 }
6098
5d4f98a2 6099 inode_tree_add(inode);
1abe9b8a 6100
6101 trace_btrfs_inode_new(inode);
1973f0fa 6102 btrfs_set_inode_last_trans(trans, inode);
1abe9b8a 6103
8ea05e3a
AB		 6104 btrfs_update_root_times(trans, root);
6105
63541927
FDBM		 6106 ret = btrfs_inode_inherit_props(trans, inode, dir);
6107 if (ret)
6108 btrfs_err(root->fs_info,
6109 "error inheriting props for ino %llu (root %llu): %d",
6110 btrfs_ino(inode), root->root_key.objectid, ret);
6111
39279cc3 6112 return inode;
b0d5d10f
CM		 6113
6114fail_unlock:
6115 unlock_new_inode(inode);
5f39d397 6116fail:
ef3b9af5 6117 if (dir && name)
aec7477b 6118 BTRFS_I(dir)->index_cnt--;
5f39d397 6119 btrfs_free_path(path);
09771430 6120 iput(inode);
5f39d397 6121 return ERR_PTR(ret);
39279cc3
CM		 6122}
6123
6124static inline u8 btrfs_inode_type(struct inode *inode)
6125{
6126 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6127}
6128
d352ac68
CM		 6129/*
6130 * utility function to add 'inode' into 'parent_inode' with
6131 * a give name and a given sequence number.
6132 * if 'add_backref' is true, also insert a backref from the
6133 * inode to the parent directory.
6134 */
e02119d5
CM		 6135int btrfs_add_link(struct btrfs_trans_handle *trans,
6136 struct inode *parent_inode, struct inode *inode,
6137 const char *name, int name_len, int add_backref, u64 index)
39279cc3 6138{
4df27c4d 6139 int ret = 0;
39279cc3 6140 struct btrfs_key key;
e02119d5 6141 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
33345d01
LZ		 6142 u64 ino = btrfs_ino(inode);
6143 u64 parent_ino = btrfs_ino(parent_inode);
5f39d397 6144
33345d01 6145 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 6146 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6147 } else {
33345d01 6148 key.objectid = ino;
962a298f 6149 key.type = BTRFS_INODE_ITEM_KEY;
4df27c4d
YZ		 6150 key.offset = 0;
6151 }
6152
33345d01 6153 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 6154 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6155 key.objectid, root->root_key.objectid,
33345d01 6156 parent_ino, index, name, name_len);
4df27c4d 6157 } else if (add_backref) {
33345d01
LZ		 6158 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6159 parent_ino, index);
4df27c4d 6160 }
39279cc3 6161
79787eaa
JM		 6162 /* Nothing to clean up yet */
6163 if (ret)
6164 return ret;
4df27c4d 6165
79787eaa
JM		 6166 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6167 parent_inode, &key,
6168 btrfs_inode_type(inode), index);
9c52057c 6169 if (ret == -EEXIST || ret == -EOVERFLOW)
79787eaa
JM		 6170 goto fail_dir_item;
6171 else if (ret) {
6172 btrfs_abort_transaction(trans, root, ret);
6173 return ret;
39279cc3 6174 }
79787eaa
JM		 6175
6176 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6177 name_len * 2);
0c4d2d95 6178 inode_inc_iversion(parent_inode);
79787eaa
JM		 6179 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
6180 ret = btrfs_update_inode(trans, root, parent_inode);
6181 if (ret)
6182 btrfs_abort_transaction(trans, root, ret);
39279cc3 6183 return ret;
fe66a05a
CM		 6184
6185fail_dir_item:
6186 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6187 u64 local_index;
6188 int err;
6189 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6190 key.objectid, root->root_key.objectid,
6191 parent_ino, &local_index, name, name_len);
6192
6193 } else if (add_backref) {
6194 u64 local_index;
6195 int err;
6196
6197 err = btrfs_del_inode_ref(trans, root, name, name_len,
6198 ino, parent_ino, &local_index);
6199 }
6200 return ret;
39279cc3
CM		 6201}
6202
6203static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
a1b075d2
JB		 6204 struct inode *dir, struct dentry *dentry,
6205 struct inode *inode, int backref, u64 index)
39279cc3 6206{
a1b075d2
JB		 6207 int err = btrfs_add_link(trans, dir, inode,
6208 dentry->d_name.name, dentry->d_name.len,
6209 backref, index);
39279cc3
CM		 6210 if (err > 0)
6211 err = -EEXIST;
6212 return err;
6213}
6214
618e21d5 6215static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1a67aafb 6216 umode_t mode, dev_t rdev)
618e21d5
JB		 6217{
6218 struct btrfs_trans_handle *trans;
6219 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 6220 struct inode *inode = NULL;
618e21d5
JB		 6221 int err;
6222 int drop_inode = 0;
6223 u64 objectid;
00e4e6b3 6224 u64 index = 0;
618e21d5
JB		 6225
6226 if (!new_valid_dev(rdev))
6227 return -EINVAL;
6228
9ed74f2d
JB		 6229 /*
6230 * 2 for inode item and ref
6231 * 2 for dir items
6232 * 1 for xattr if selinux is on
6233 */
a22285a6
YZ		 6234 trans = btrfs_start_transaction(root, 5);
6235 if (IS_ERR(trans))
6236 return PTR_ERR(trans);
1832a6d5 6237
581bb050
LZ		 6238 err = btrfs_find_free_ino(root, &objectid);
6239 if (err)
6240 goto out_unlock;
6241
aec7477b 6242 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 6243 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 6244 mode, &index);
7cf96da3
TI		 6245 if (IS_ERR(inode)) {
6246 err = PTR_ERR(inode);
618e21d5 6247 goto out_unlock;
7cf96da3 6248 }
618e21d5 6249
ad19db71
CS		 6250 /*
6251 * If the active LSM wants to access the inode during
6252 * d_instantiate it needs these. Smack checks to see
6253 * if the filesystem supports xattrs by looking at the
6254 * ops vector.
6255 */
ad19db71 6256 inode->i_op = &btrfs_special_inode_operations;
b0d5d10f
CM		 6257 init_special_inode(inode, inode->i_mode, rdev);
6258
6259 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
618e21d5 6260 if (err)
b0d5d10f
CM		 6261 goto out_unlock_inode;
6262
6263 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6264 if (err) {
6265 goto out_unlock_inode;
6266 } else {
1b4ab1bb 6267 btrfs_update_inode(trans, root, inode);
b0d5d10f 6268 unlock_new_inode(inode);
08c422c2 6269 d_instantiate(dentry, inode);
618e21d5 6270 }
b0d5d10f 6271
618e21d5 6272out_unlock:
7ad85bb7 6273 btrfs_end_transaction(trans, root);
c581afc8 6274 btrfs_balance_delayed_items(root);
b53d3f5d 6275 btrfs_btree_balance_dirty(root);
618e21d5
JB		 6276 if (drop_inode) {
6277 inode_dec_link_count(inode);
6278 iput(inode);
6279 }
618e21d5 6280 return err;
b0d5d10f
CM		 6281
6282out_unlock_inode:
6283 drop_inode = 1;
6284 unlock_new_inode(inode);
6285 goto out_unlock;
6286
618e21d5
JB		 6287}
6288
39279cc3 6289static int btrfs_create(struct inode *dir, struct dentry *dentry,
ebfc3b49 6290 umode_t mode, bool excl)
39279cc3
CM		 6291{
6292 struct btrfs_trans_handle *trans;
6293 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 6294 struct inode *inode = NULL;
43baa579 6295 int drop_inode_on_err = 0;
a22285a6 6296 int err;
39279cc3 6297 u64 objectid;
00e4e6b3 6298 u64 index = 0;
39279cc3 6299
9ed74f2d
JB		 6300 /*
6301 * 2 for inode item and ref
6302 * 2 for dir items
6303 * 1 for xattr if selinux is on
6304 */
a22285a6
YZ		 6305 trans = btrfs_start_transaction(root, 5);
6306 if (IS_ERR(trans))
6307 return PTR_ERR(trans);
9ed74f2d 6308
581bb050
LZ		 6309 err = btrfs_find_free_ino(root, &objectid);
6310 if (err)
6311 goto out_unlock;
6312
aec7477b 6313 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 6314 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 6315 mode, &index);
7cf96da3
TI		 6316 if (IS_ERR(inode)) {
6317 err = PTR_ERR(inode);
39279cc3 6318 goto out_unlock;
7cf96da3 6319 }
43baa579 6320 drop_inode_on_err = 1;
ad19db71
CS		 6321 /*
6322 * If the active LSM wants to access the inode during
6323 * d_instantiate it needs these. Smack checks to see
6324 * if the filesystem supports xattrs by looking at the
6325 * ops vector.
6326 */
6327 inode->i_fop = &btrfs_file_operations;
6328 inode->i_op = &btrfs_file_inode_operations;
b0d5d10f 6329 inode->i_mapping->a_ops = &btrfs_aops;
b0d5d10f
CM		 6330
6331 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6332 if (err)
6333 goto out_unlock_inode;
6334
6335 err = btrfs_update_inode(trans, root, inode);
6336 if (err)
6337 goto out_unlock_inode;
ad19db71 6338
a1b075d2 6339 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3 6340 if (err)
b0d5d10f 6341 goto out_unlock_inode;
43baa579 6342
43baa579 6343 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
b0d5d10f 6344 unlock_new_inode(inode);
43baa579
FB		 6345 d_instantiate(dentry, inode);
6346
39279cc3 6347out_unlock:
7ad85bb7 6348 btrfs_end_transaction(trans, root);
43baa579 6349 if (err && drop_inode_on_err) {
39279cc3
CM		 6350 inode_dec_link_count(inode);
6351 iput(inode);
6352 }
c581afc8 6353 btrfs_balance_delayed_items(root);
b53d3f5d 6354 btrfs_btree_balance_dirty(root);
39279cc3 6355 return err;
b0d5d10f
CM		 6356
6357out_unlock_inode:
6358 unlock_new_inode(inode);
6359 goto out_unlock;
6360
39279cc3
CM		 6361}
6362
6363static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6364 struct dentry *dentry)
6365{
6366 struct btrfs_trans_handle *trans;
6367 struct btrfs_root *root = BTRFS_I(dir)->root;
6368 struct inode *inode = old_dentry->d_inode;
00e4e6b3 6369 u64 index;
39279cc3
CM		 6370 int err;
6371 int drop_inode = 0;
6372
4a8be425
TH		 6373 /* do not allow sys_link's with other subvols of the same device */
6374 if (root->objectid != BTRFS_I(inode)->root->objectid)
3ab3564f 6375 return -EXDEV;
4a8be425 6376
f186373f 6377 if (inode->i_nlink >= BTRFS_LINK_MAX)
c055e99e 6378 return -EMLINK;
4a8be425 6379
3de4586c 6380 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 6381 if (err)
6382 goto fail;
6383
a22285a6 6384 /*
7e6b6465 6385 * 2 items for inode and inode ref
a22285a6 6386 * 2 items for dir items
7e6b6465 6387 * 1 item for parent inode
a22285a6 6388 */
7e6b6465 6389 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 6390 if (IS_ERR(trans)) {
6391 err = PTR_ERR(trans);
6392 goto fail;
6393 }
5f39d397 6394
67de1176
MX		 6395 /* There are several dir indexes for this inode, clear the cache. */
6396 BTRFS_I(inode)->dir_index = 0ULL;
8b558c5f 6397 inc_nlink(inode);
0c4d2d95 6398 inode_inc_iversion(inode);
3153495d 6399 inode->i_ctime = CURRENT_TIME;
7de9c6ee 6400 ihold(inode);
e9976151 6401 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
aec7477b 6402
a1b075d2 6403 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5f39d397 6404
a5719521 6405 if (err) {
54aa1f4d 6406 drop_inode = 1;
a5719521 6407 } else {
10d9f309 6408 struct dentry *parent = dentry->d_parent;
a5719521 6409 err = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 6410 if (err)
6411 goto fail;
ef3b9af5
FM		 6412 if (inode->i_nlink == 1) {
6413 /*
6414 * If new hard link count is 1, it's a file created
6415 * with open(2) O_TMPFILE flag.
6416 */
6417 err = btrfs_orphan_del(trans, inode);
6418 if (err)
6419 goto fail;
6420 }
08c422c2 6421 d_instantiate(dentry, inode);
6a912213 6422 btrfs_log_new_name(trans, inode, NULL, parent);
a5719521 6423 }
39279cc3 6424
7ad85bb7 6425 btrfs_end_transaction(trans, root);
c581afc8 6426 btrfs_balance_delayed_items(root);
1832a6d5 6427fail:
39279cc3
CM		 6428 if (drop_inode) {
6429 inode_dec_link_count(inode);
6430 iput(inode);
6431 }
b53d3f5d 6432 btrfs_btree_balance_dirty(root);
39279cc3
CM		 6433 return err;
6434}
6435
18bb1db3 6436static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
39279cc3 6437{
b9d86667 6438 struct inode *inode = NULL;
39279cc3
CM		 6439 struct btrfs_trans_handle *trans;
6440 struct btrfs_root *root = BTRFS_I(dir)->root;
6441 int err = 0;
6442 int drop_on_err = 0;
b9d86667 6443 u64 objectid = 0;
00e4e6b3 6444 u64 index = 0;
39279cc3 6445
9ed74f2d
JB		 6446 /*
6447 * 2 items for inode and ref
6448 * 2 items for dir items
6449 * 1 for xattr if selinux is on
6450 */
a22285a6
YZ		 6451 trans = btrfs_start_transaction(root, 5);
6452 if (IS_ERR(trans))
6453 return PTR_ERR(trans);
39279cc3 6454
581bb050
LZ		 6455 err = btrfs_find_free_ino(root, &objectid);
6456 if (err)
6457 goto out_fail;
6458
aec7477b 6459 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 6460 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 6461 S_IFDIR | mode, &index);
39279cc3
CM		 6462 if (IS_ERR(inode)) {
6463 err = PTR_ERR(inode);
6464 goto out_fail;
6465 }
5f39d397 6466
39279cc3 6467 drop_on_err = 1;
b0d5d10f
CM		 6468 /* these must be set before we unlock the inode */
6469 inode->i_op = &btrfs_dir_inode_operations;
6470 inode->i_fop = &btrfs_dir_file_operations;
33268eaf 6471
2a7dba39 6472 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf 6473 if (err)
b0d5d10f 6474 goto out_fail_inode;
39279cc3 6475
dbe674a9 6476 btrfs_i_size_write(inode, 0);
39279cc3
CM		 6477 err = btrfs_update_inode(trans, root, inode);
6478 if (err)
b0d5d10f 6479 goto out_fail_inode;
5f39d397 6480
a1b075d2
JB		 6481 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6482 dentry->d_name.len, 0, index);
39279cc3 6483 if (err)
b0d5d10f 6484 goto out_fail_inode;
5f39d397 6485
39279cc3 6486 d_instantiate(dentry, inode);
b0d5d10f
CM		 6487 /*
6488 * mkdir is special. We're unlocking after we call d_instantiate
6489 * to avoid a race with nfsd calling d_instantiate.
6490 */
6491 unlock_new_inode(inode);
39279cc3 6492 drop_on_err = 0;
39279cc3
CM		 6493
6494out_fail:
7ad85bb7 6495 btrfs_end_transaction(trans, root);
c7cfb8a5
WS		 6496 if (drop_on_err) {
6497 inode_dec_link_count(inode);
39279cc3 6498 iput(inode);
c7cfb8a5 6499 }
c581afc8 6500 btrfs_balance_delayed_items(root);
b53d3f5d 6501 btrfs_btree_balance_dirty(root);
39279cc3 6502 return err;
b0d5d10f
CM		 6503
6504out_fail_inode:
6505 unlock_new_inode(inode);
6506 goto out_fail;
39279cc3
CM		 6507}
6508
e6c4efd8
QW		 6509/* Find next extent map of a given extent map, caller needs to ensure locks */
6510static struct extent_map *next_extent_map(struct extent_map *em)
6511{
6512 struct rb_node *next;
6513
6514 next = rb_next(&em->rb_node);
6515 if (!next)
6516 return NULL;
6517 return container_of(next, struct extent_map, rb_node);
6518}
6519
6520static struct extent_map *prev_extent_map(struct extent_map *em)
6521{
6522 struct rb_node *prev;
6523
6524 prev = rb_prev(&em->rb_node);
6525 if (!prev)
6526 return NULL;
6527 return container_of(prev, struct extent_map, rb_node);
6528}
6529
d352ac68 6530/* helper for btfs_get_extent. Given an existing extent in the tree,
e6c4efd8 6531 * the existing extent is the nearest extent to map_start,
d352ac68 6532 * and an extent that you want to insert, deal with overlap and insert
e6c4efd8 6533 * the best fitted new extent into the tree.
d352ac68 6534 */
3b951516
CM		 6535static int merge_extent_mapping(struct extent_map_tree *em_tree,
6536 struct extent_map *existing,
e6dcd2dc 6537 struct extent_map *em,
51f395ad 6538 u64 map_start)
3b951516 6539{
e6c4efd8
QW		 6540 struct extent_map *prev;
6541 struct extent_map *next;
6542 u64 start;
6543 u64 end;
3b951516 6544 u64 start_diff;
3b951516 6545
e6dcd2dc 6546 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
e6c4efd8
QW		 6547
6548 if (existing->start > map_start) {
6549 next = existing;
6550 prev = prev_extent_map(next);
6551 } else {
6552 prev = existing;
6553 next = next_extent_map(prev);
6554 }
6555
6556 start = prev ? extent_map_end(prev) : em->start;
6557 start = max_t(u64, start, em->start);
6558 end = next ? next->start : extent_map_end(em);
6559 end = min_t(u64, end, extent_map_end(em));
6560 start_diff = start - em->start;
6561 em->start = start;
6562 em->len = end - start;
c8b97818
CM		 6563 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6564 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 6565 em->block_start += start_diff;
c8b97818
CM		 6566 em->block_len -= start_diff;
6567 }
09a2a8f9 6568 return add_extent_mapping(em_tree, em, 0);
3b951516
CM		 6569}
6570
c8b97818
CM		 6571static noinline int uncompress_inline(struct btrfs_path *path,
6572 struct inode *inode, struct page *page,
6573 size_t pg_offset, u64 extent_offset,
6574 struct btrfs_file_extent_item *item)
6575{
6576 int ret;
6577 struct extent_buffer *leaf = path->nodes[0];
6578 char *tmp;
6579 size_t max_size;
6580 unsigned long inline_size;
6581 unsigned long ptr;
261507a0 6582 int compress_type;
c8b97818
CM		 6583
6584 WARN_ON(pg_offset != 0);
261507a0 6585 compress_type = btrfs_file_extent_compression(leaf, item);
c8b97818
CM		 6586 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6587 inline_size = btrfs_file_extent_inline_item_len(leaf,
dd3cc16b 6588 btrfs_item_nr(path->slots[0]));
c8b97818 6589 tmp = kmalloc(inline_size, GFP_NOFS);
8d413713
TI		 6590 if (!tmp)
6591 return -ENOMEM;
c8b97818
CM		 6592 ptr = btrfs_file_extent_inline_start(item);
6593
6594 read_extent_buffer(leaf, tmp, ptr, inline_size);
6595
5b050f04 6596 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
261507a0
LZ		 6597 ret = btrfs_decompress(compress_type, tmp, page,
6598 extent_offset, inline_size, max_size);
c8b97818 6599 kfree(tmp);
166ae5a4 6600 return ret;
c8b97818
CM		 6601}
6602
d352ac68
CM		 6603/*
6604 * a bit scary, this does extent mapping from logical file offset to the disk.
d397712b
CM		 6605 * the ugly parts come from merging extents from the disk with the in-ram
6606 * representation. This gets more complex because of the data=ordered code,
d352ac68
CM		 6607 * where the in-ram extents might be locked pending data=ordered completion.
6608 *
6609 * This also copies inline extents directly into the page.
6610 */
d397712b 6611
a52d9a80 6612struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 6613 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 6614 int create)
6615{
6616 int ret;
6617 int err = 0;
a52d9a80
CM		 6618 u64 extent_start = 0;
6619 u64 extent_end = 0;
33345d01 6620 u64 objectid = btrfs_ino(inode);
a52d9a80 6621 u32 found_type;
f421950f 6622 struct btrfs_path *path = NULL;
a52d9a80
CM		 6623 struct btrfs_root *root = BTRFS_I(inode)->root;
6624 struct btrfs_file_extent_item *item;
5f39d397
CM		 6625 struct extent_buffer *leaf;
6626 struct btrfs_key found_key;
a52d9a80
CM		 6627 struct extent_map *em = NULL;
6628 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 6629 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 6630 struct btrfs_trans_handle *trans = NULL;
7ffbb598 6631 const bool new_inline = !page || create;
a52d9a80 6632
a52d9a80 6633again:
890871be 6634 read_lock(&em_tree->lock);
d1310b2e 6635 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 6636 if (em)
6637 em->bdev = root->fs_info->fs_devices->latest_bdev;
890871be 6638 read_unlock(&em_tree->lock);
d1310b2e 6639
a52d9a80 6640 if (em) {
e1c4b745
CM		 6641 if (em->start > start || em->start + em->len <= start)
6642 free_extent_map(em);
6643 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 6644 free_extent_map(em);
6645 else
6646 goto out;
a52d9a80 6647 }
172ddd60 6648 em = alloc_extent_map();
a52d9a80 6649 if (!em) {
d1310b2e
CM		 6650 err = -ENOMEM;
6651 goto out;
a52d9a80 6652 }
e6dcd2dc 6653 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 6654 em->start = EXTENT_MAP_HOLE;
445a6944 6655 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 6656 em->len = (u64)-1;
c8b97818 6657 em->block_len = (u64)-1;
f421950f
CM		 6658
6659 if (!path) {
6660 path = btrfs_alloc_path();
026fd317
JB		 6661 if (!path) {
6662 err = -ENOMEM;
6663 goto out;
6664 }
6665 /*
6666 * Chances are we'll be called again, so go ahead and do
6667 * readahead
6668 */
6669 path->reada = 1;
f421950f
CM		 6670 }
6671
179e29e4
CM		 6672 ret = btrfs_lookup_file_extent(trans, root, path,
6673 objectid, start, trans != NULL);
a52d9a80
CM		 6674 if (ret < 0) {
6675 err = ret;
6676 goto out;
6677 }
6678
6679 if (ret != 0) {
6680 if (path->slots[0] == 0)
6681 goto not_found;
6682 path->slots[0]--;
6683 }
6684
5f39d397
CM		 6685 leaf = path->nodes[0];
6686 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 6687 struct btrfs_file_extent_item);
a52d9a80 6688 /* are we inside the extent that was found? */
5f39d397 6689 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
962a298f 6690 found_type = found_key.type;
5f39d397 6691 if (found_key.objectid != objectid ||
a52d9a80 6692 found_type != BTRFS_EXTENT_DATA_KEY) {
25a50341
JB		 6693 /*
6694 * If we backup past the first extent we want to move forward
6695 * and see if there is an extent in front of us, otherwise we'll
6696 * say there is a hole for our whole search range which can
6697 * cause problems.
6698 */
6699 extent_end = start;
6700 goto next;
a52d9a80
CM		 6701 }
6702
5f39d397
CM		 6703 found_type = btrfs_file_extent_type(leaf, item);
6704 extent_start = found_key.offset;
d899e052
YZ		 6705 if (found_type == BTRFS_FILE_EXTENT_REG ||
6706 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 6707 extent_end = extent_start +
db94535d 6708 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 6709 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6710 size_t size;
514ac8ad 6711 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
fda2832f 6712 extent_end = ALIGN(extent_start + size, root->sectorsize);
9036c102 6713 }
25a50341 6714next:
9036c102
YZ		 6715 if (start >= extent_end) {
6716 path->slots[0]++;
6717 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6718 ret = btrfs_next_leaf(root, path);
6719 if (ret < 0) {
6720 err = ret;
6721 goto out;
a52d9a80 6722 }
9036c102
YZ		 6723 if (ret > 0)
6724 goto not_found;
6725 leaf = path->nodes[0];
a52d9a80 6726 }
9036c102
YZ		 6727 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6728 if (found_key.objectid != objectid ||
6729 found_key.type != BTRFS_EXTENT_DATA_KEY)
6730 goto not_found;
6731 if (start + len <= found_key.offset)
6732 goto not_found;
e2eca69d
WS		 6733 if (start > found_key.offset)
6734 goto next;
9036c102 6735 em->start = start;
70c8a91c 6736 em->orig_start = start;
9036c102
YZ		 6737 em->len = found_key.offset - start;
6738 goto not_found_em;
6739 }
6740
7ffbb598
FM		 6741 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6742
d899e052
YZ		 6743 if (found_type == BTRFS_FILE_EXTENT_REG ||
6744 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80
CM		 6745 goto insert;
6746 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 6747 unsigned long ptr;
a52d9a80 6748 char *map;
3326d1b0
CM		 6749 size_t size;
6750 size_t extent_offset;
6751 size_t copy_size;
a52d9a80 6752
7ffbb598 6753 if (new_inline)
689f9346 6754 goto out;
5f39d397 6755
514ac8ad 6756 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
9036c102 6757 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 6758 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 6759 size - extent_offset);
3326d1b0 6760 em->start = extent_start + extent_offset;
fda2832f 6761 em->len = ALIGN(copy_size, root->sectorsize);
b4939680 6762 em->orig_block_len = em->len;
70c8a91c 6763 em->orig_start = em->start;
689f9346 6764 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 6765 if (create == 0 && !PageUptodate(page)) {
261507a0
LZ		 6766 if (btrfs_file_extent_compression(leaf, item) !=
6767 BTRFS_COMPRESS_NONE) {
c8b97818
CM		 6768 ret = uncompress_inline(path, inode, page,
6769 pg_offset,
6770 extent_offset, item);
166ae5a4
ZB		 6771 if (ret) {
6772 err = ret;
6773 goto out;
6774 }
c8b97818
CM		 6775 } else {
6776 map = kmap(page);
6777 read_extent_buffer(leaf, map + pg_offset, ptr,
6778 copy_size);
93c82d57
CM		 6779 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6780 memset(map + pg_offset + copy_size, 0,
6781 PAGE_CACHE_SIZE - pg_offset -
6782 copy_size);
6783 }
c8b97818
CM		 6784 kunmap(page);
6785 }
179e29e4
CM		 6786 flush_dcache_page(page);
6787 } else if (create && PageUptodate(page)) {
6bf7e080 6788 BUG();
179e29e4
CM		 6789 if (!trans) {
6790 kunmap(page);
6791 free_extent_map(em);
6792 em = NULL;
ff5714cc 6793
b3b4aa74 6794 btrfs_release_path(path);
7a7eaa40 6795 trans = btrfs_join_transaction(root);
ff5714cc 6796
3612b495
TI		 6797 if (IS_ERR(trans))
6798 return ERR_CAST(trans);
179e29e4
CM		 6799 goto again;
6800 }
c8b97818 6801 map = kmap(page);
70dec807 6802 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 6803 copy_size);
c8b97818 6804 kunmap(page);
179e29e4 6805 btrfs_mark_buffer_dirty(leaf);
a52d9a80 6806 }
d1310b2e 6807 set_extent_uptodate(io_tree, em->start,
507903b8 6808 extent_map_end(em) - 1, NULL, GFP_NOFS);
a52d9a80 6809 goto insert;
a52d9a80
CM		 6810 }
6811not_found:
6812 em->start = start;
70c8a91c 6813 em->orig_start = start;
d1310b2e 6814 em->len = len;
a52d9a80 6815not_found_em:
5f39d397 6816 em->block_start = EXTENT_MAP_HOLE;
9036c102 6817 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80 6818insert:
b3b4aa74 6819 btrfs_release_path(path);
d1310b2e 6820 if (em->start > start || extent_map_end(em) <= start) {
c2cf52eb 6821 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
c1c9ff7c 6822 em->start, em->len, start, len);
a52d9a80
CM		 6823 err = -EIO;
6824 goto out;
6825 }
d1310b2e
CM		 6826
6827 err = 0;
890871be 6828 write_lock(&em_tree->lock);
09a2a8f9 6829 ret = add_extent_mapping(em_tree, em, 0);
3b951516
CM		 6830 /* it is possible that someone inserted the extent into the tree
6831 * while we had the lock dropped. It is also possible that
6832 * an overlapping map exists in the tree
6833 */
a52d9a80 6834 if (ret == -EEXIST) {
3b951516 6835 struct extent_map *existing;
e6dcd2dc
CM		 6836
6837 ret = 0;
6838
e6c4efd8
QW		 6839 existing = search_extent_mapping(em_tree, start, len);
6840 /*
6841 * existing will always be non-NULL, since there must be
6842 * extent causing the -EEXIST.
6843 */
6844 if (start >= extent_map_end(existing) ||
32be3a1a 6845 start <= existing->start) {
e6c4efd8
QW		 6846 /*
6847 * The existing extent map is the one nearest to
6848 * the [start, start + len) range which overlaps
6849 */
6850 err = merge_extent_mapping(em_tree, existing,
6851 em, start);
e1c4b745 6852 free_extent_map(existing);
e6c4efd8 6853 if (err) {
3b951516
CM		 6854 free_extent_map(em);
6855 em = NULL;
6856 }
6857 } else {
6858 free_extent_map(em);
6859 em = existing;
e6dcd2dc 6860 err = 0;
a52d9a80 6861 }
a52d9a80 6862 }
890871be 6863 write_unlock(&em_tree->lock);
a52d9a80 6864out:
1abe9b8a 6865
4cd8587c 6866 trace_btrfs_get_extent(root, em);
1abe9b8a 6867
f421950f
CM		 6868 if (path)
6869 btrfs_free_path(path);
a52d9a80
CM		 6870 if (trans) {
6871 ret = btrfs_end_transaction(trans, root);
d397712b 6872 if (!err)
a52d9a80
CM		 6873 err = ret;
6874 }
a52d9a80
CM		 6875 if (err) {
6876 free_extent_map(em);
a52d9a80
CM		 6877 return ERR_PTR(err);
6878 }
79787eaa 6879 BUG_ON(!em); /* Error is always set */
a52d9a80
CM		 6880 return em;
6881}
6882
ec29ed5b
CM		 6883struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6884 size_t pg_offset, u64 start, u64 len,
6885 int create)
6886{
6887 struct extent_map *em;
6888 struct extent_map *hole_em = NULL;
6889 u64 range_start = start;
6890 u64 end;
6891 u64 found;
6892 u64 found_end;
6893 int err = 0;
6894
6895 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6896 if (IS_ERR(em))
6897 return em;
6898 if (em) {
6899 /*
f9e4fb53
LB		 6900 * if our em maps to
6901 * - a hole or
6902 * - a pre-alloc extent,
6903 * there might actually be delalloc bytes behind it.
ec29ed5b 6904 */
f9e4fb53
LB		 6905 if (em->block_start != EXTENT_MAP_HOLE &&
6906 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
ec29ed5b
CM		 6907 return em;
6908 else
6909 hole_em = em;
6910 }
6911
6912 /* check to see if we've wrapped (len == -1 or similar) */
6913 end = start + len;
6914 if (end < start)
6915 end = (u64)-1;
6916 else
6917 end -= 1;
6918
6919 em = NULL;
6920
6921 /* ok, we didn't find anything, lets look for delalloc */
6922 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6923 end, len, EXTENT_DELALLOC, 1);
6924 found_end = range_start + found;
6925 if (found_end < range_start)
6926 found_end = (u64)-1;
6927
6928 /*
6929 * we didn't find anything useful, return
6930 * the original results from get_extent()
6931 */
6932 if (range_start > end || found_end <= start) {
6933 em = hole_em;
6934 hole_em = NULL;
6935 goto out;
6936 }
6937
6938 /* adjust the range_start to make sure it doesn't
6939 * go backwards from the start they passed in
6940 */
67871254 6941 range_start = max(start, range_start);
ec29ed5b
CM		 6942 found = found_end - range_start;
6943
6944 if (found > 0) {
6945 u64 hole_start = start;
6946 u64 hole_len = len;
6947
172ddd60 6948 em = alloc_extent_map();
ec29ed5b
CM		 6949 if (!em) {
6950 err = -ENOMEM;
6951 goto out;
6952 }
6953 /*
6954 * when btrfs_get_extent can't find anything it
6955 * returns one huge hole
6956 *
6957 * make sure what it found really fits our range, and
6958 * adjust to make sure it is based on the start from
6959 * the caller
6960 */
6961 if (hole_em) {
6962 u64 calc_end = extent_map_end(hole_em);
6963
6964 if (calc_end <= start || (hole_em->start > end)) {
6965 free_extent_map(hole_em);
6966 hole_em = NULL;
6967 } else {
6968 hole_start = max(hole_em->start, start);
6969 hole_len = calc_end - hole_start;
6970 }
6971 }
6972 em->bdev = NULL;
6973 if (hole_em && range_start > hole_start) {
6974 /* our hole starts before our delalloc, so we
6975 * have to return just the parts of the hole
6976 * that go until the delalloc starts
6977 */
6978 em->len = min(hole_len,
6979 range_start - hole_start);
6980 em->start = hole_start;
6981 em->orig_start = hole_start;
6982 /*
6983 * don't adjust block start at all,
6984 * it is fixed at EXTENT_MAP_HOLE
6985 */
6986 em->block_start = hole_em->block_start;
6987 em->block_len = hole_len;
f9e4fb53
LB		 6988 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6989 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
ec29ed5b
CM		 6990 } else {
6991 em->start = range_start;
6992 em->len = found;
6993 em->orig_start = range_start;
6994 em->block_start = EXTENT_MAP_DELALLOC;
6995 em->block_len = found;
6996 }
6997 } else if (hole_em) {
6998 return hole_em;
6999 }
7000out:
7001
7002 free_extent_map(hole_em);
7003 if (err) {
7004 free_extent_map(em);
7005 return ERR_PTR(err);
7006 }
7007 return em;
7008}
7009
4b46fce2
JB		 7010static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7011 u64 start, u64 len)
7012{
7013 struct btrfs_root *root = BTRFS_I(inode)->root;
70c8a91c 7014 struct extent_map *em;
4b46fce2
JB		 7015 struct btrfs_key ins;
7016 u64 alloc_hint;
7017 int ret;
4b46fce2 7018
4b46fce2 7019 alloc_hint = get_extent_allocation_hint(inode, start, len);
00361589 7020 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
e570fd27 7021 alloc_hint, &ins, 1, 1);
00361589
JB		 7022 if (ret)
7023 return ERR_PTR(ret);
4b46fce2 7024
70c8a91c 7025 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
cc95bef6 7026 ins.offset, ins.offset, ins.offset, 0);
00361589 7027 if (IS_ERR(em)) {
e570fd27 7028 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
00361589
JB		 7029 return em;
7030 }
4b46fce2
JB		 7031
7032 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
7033 ins.offset, ins.offset, 0);
7034 if (ret) {
e570fd27 7035 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
00361589
JB		 7036 free_extent_map(em);
7037 return ERR_PTR(ret);
4b46fce2 7038 }
00361589 7039
31193213
DY		 7040 btrfs_qgroup_update_reserved_bytes(root->fs_info,
7041 root->root_key.objectid,
7042 ins.offset, 1);
7043
4b46fce2
JB		 7044 return em;
7045}
7046
46bfbb5c
CM		 7047/*
7048 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7049 * block must be cow'd
7050 */
00361589 7051noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7ee9e440
JB		 7052 u64 *orig_start, u64 *orig_block_len,
7053 u64 *ram_bytes)
46bfbb5c 7054{
00361589 7055 struct btrfs_trans_handle *trans;
46bfbb5c
CM		 7056 struct btrfs_path *path;
7057 int ret;
7058 struct extent_buffer *leaf;
7059 struct btrfs_root *root = BTRFS_I(inode)->root;
7b2b7085 7060 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
46bfbb5c
CM		 7061 struct btrfs_file_extent_item *fi;
7062 struct btrfs_key key;
7063 u64 disk_bytenr;
7064 u64 backref_offset;
7065 u64 extent_end;
7066 u64 num_bytes;
7067 int slot;
7068 int found_type;
7ee9e440 7069 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
e77751aa 7070
46bfbb5c
CM		 7071 path = btrfs_alloc_path();
7072 if (!path)
7073 return -ENOMEM;
7074
00361589 7075 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
46bfbb5c
CM		 7076 offset, 0);
7077 if (ret < 0)
7078 goto out;
7079
7080 slot = path->slots[0];
7081 if (ret == 1) {
7082 if (slot == 0) {
7083 /* can't find the item, must cow */
7084 ret = 0;
7085 goto out;
7086 }
7087 slot--;
7088 }
7089 ret = 0;
7090 leaf = path->nodes[0];
7091 btrfs_item_key_to_cpu(leaf, &key, slot);
33345d01 7092 if (key.objectid != btrfs_ino(inode) ||
46bfbb5c
CM		 7093 key.type != BTRFS_EXTENT_DATA_KEY) {
7094 /* not our file or wrong item type, must cow */
7095 goto out;
7096 }
7097
7098 if (key.offset > offset) {
7099 /* Wrong offset, must cow */
7100 goto out;
7101 }
7102
7103 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7104 found_type = btrfs_file_extent_type(leaf, fi);
7105 if (found_type != BTRFS_FILE_EXTENT_REG &&
7106 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7107 /* not a regular extent, must cow */
7108 goto out;
7109 }
7ee9e440
JB		 7110
7111 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7112 goto out;
7113
e77751aa
MX		 7114 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7115 if (extent_end <= offset)
7116 goto out;
7117
46bfbb5c 7118 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7ee9e440
JB		 7119 if (disk_bytenr == 0)
7120 goto out;
7121
7122 if (btrfs_file_extent_compression(leaf, fi) ||
7123 btrfs_file_extent_encryption(leaf, fi) ||
7124 btrfs_file_extent_other_encoding(leaf, fi))
7125 goto out;
7126
46bfbb5c
CM		 7127 backref_offset = btrfs_file_extent_offset(leaf, fi);
7128
7ee9e440
JB		 7129 if (orig_start) {
7130 *orig_start = key.offset - backref_offset;
7131 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7132 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7133 }
eb384b55 7134
46bfbb5c
CM		 7135 if (btrfs_extent_readonly(root, disk_bytenr))
7136 goto out;
7b2b7085
MX		 7137
7138 num_bytes = min(offset + *len, extent_end) - offset;
7139 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7140 u64 range_end;
7141
7142 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7143 ret = test_range_bit(io_tree, offset, range_end,
7144 EXTENT_DELALLOC, 0, NULL);
7145 if (ret) {
7146 ret = -EAGAIN;
7147 goto out;
7148 }
7149 }
7150
1bda19eb 7151 btrfs_release_path(path);
46bfbb5c
CM		 7152
7153 /*
7154 * look for other files referencing this extent, if we
7155 * find any we must cow
7156 */
00361589
JB		 7157 trans = btrfs_join_transaction(root);
7158 if (IS_ERR(trans)) {
7159 ret = 0;
46bfbb5c 7160 goto out;
00361589
JB		 7161 }
7162
7163 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7164 key.offset - backref_offset, disk_bytenr);
7165 btrfs_end_transaction(trans, root);
7166 if (ret) {
7167 ret = 0;
7168 goto out;
7169 }
46bfbb5c
CM		 7170
7171 /*
7172 * adjust disk_bytenr and num_bytes to cover just the bytes
7173 * in this extent we are about to write. If there
7174 * are any csums in that range we have to cow in order
7175 * to keep the csums correct
7176 */
7177 disk_bytenr += backref_offset;
7178 disk_bytenr += offset - key.offset;
46bfbb5c
CM		 7179 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7180 goto out;
7181 /*
7182 * all of the above have passed, it is safe to overwrite this extent
7183 * without cow
7184 */
eb384b55 7185 *len = num_bytes;
46bfbb5c
CM		 7186 ret = 1;
7187out:
7188 btrfs_free_path(path);
7189 return ret;
7190}
7191
fc4adbff
AG		 7192bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7193{
7194 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7195 int found = false;
7196 void **pagep = NULL;
7197 struct page *page = NULL;
7198 int start_idx;
7199 int end_idx;
7200
7201 start_idx = start >> PAGE_CACHE_SHIFT;
7202
7203 /*
7204 * end is the last byte in the last page. end == start is legal
7205 */
7206 end_idx = end >> PAGE_CACHE_SHIFT;
7207
7208 rcu_read_lock();
7209
7210 /* Most of the code in this while loop is lifted from
7211 * find_get_page. It's been modified to begin searching from a
7212 * page and return just the first page found in that range. If the
7213 * found idx is less than or equal to the end idx then we know that
7214 * a page exists. If no pages are found or if those pages are
7215 * outside of the range then we're fine (yay!) */
7216 while (page == NULL &&
7217 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7218 page = radix_tree_deref_slot(pagep);
7219 if (unlikely(!page))
7220 break;
7221
7222 if (radix_tree_exception(page)) {
809f9016
FM		 7223 if (radix_tree_deref_retry(page)) {
7224 page = NULL;
fc4adbff 7225 continue;
809f9016 7226 }
fc4adbff
AG		 7227 /*
7228 * Otherwise, shmem/tmpfs must be storing a swap entry
7229 * here as an exceptional entry: so return it without
7230 * attempting to raise page count.
7231 */
6fdef6d4 7232 page = NULL;
fc4adbff
AG		 7233 break; /* TODO: Is this relevant for this use case? */
7234 }
7235
91405151
FM		 7236 if (!page_cache_get_speculative(page)) {
7237 page = NULL;
fc4adbff 7238 continue;
91405151 7239 }
fc4adbff
AG		 7240
7241 /*
7242 * Has the page moved?
7243 * This is part of the lockless pagecache protocol. See
7244 * include/linux/pagemap.h for details.
7245 */
7246 if (unlikely(page != *pagep)) {
7247 page_cache_release(page);
7248 page = NULL;
7249 }
7250 }
7251
7252 if (page) {
7253 if (page->index <= end_idx)
7254 found = true;
7255 page_cache_release(page);
7256 }
7257
7258 rcu_read_unlock();
7259 return found;
7260}
7261
eb838e73
JB		 7262static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7263 struct extent_state **cached_state, int writing)
7264{
7265 struct btrfs_ordered_extent *ordered;
7266 int ret = 0;
7267
7268 while (1) {
7269 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7270 0, cached_state);
7271 /*
7272 * We're concerned with the entire range that we're going to be
7273 * doing DIO to, so we need to make sure theres no ordered
7274 * extents in this range.
7275 */
7276 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7277 lockend - lockstart + 1);
7278
7279 /*
7280 * We need to make sure there are no buffered pages in this
7281 * range either, we could have raced between the invalidate in
7282 * generic_file_direct_write and locking the extent. The
7283 * invalidate needs to happen so that reads after a write do not
7284 * get stale data.
7285 */
fc4adbff
AG		 7286 if (!ordered &&
7287 (!writing ||
7288 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
eb838e73
JB		 7289 break;
7290
7291 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7292 cached_state, GFP_NOFS);
7293
7294 if (ordered) {
7295 btrfs_start_ordered_extent(inode, ordered, 1);
7296 btrfs_put_ordered_extent(ordered);
7297 } else {
7298 /* Screw you mmap */
728404da 7299 ret = btrfs_fdatawrite_range(inode, lockstart, lockend);
075bdbdb
FM		 7300 if (ret)
7301 break;
7302 ret = filemap_fdatawait_range(inode->i_mapping,
7303 lockstart,
7304 lockend);
eb838e73
JB		 7305 if (ret)
7306 break;
7307
7308 /*
7309 * If we found a page that couldn't be invalidated just
7310 * fall back to buffered.
7311 */
7312 ret = invalidate_inode_pages2_range(inode->i_mapping,
7313 lockstart >> PAGE_CACHE_SHIFT,
7314 lockend >> PAGE_CACHE_SHIFT);
7315 if (ret)
7316 break;
7317 }
7318
7319 cond_resched();
7320 }
7321
7322 return ret;
7323}
7324
69ffb543
JB		 7325static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7326 u64 len, u64 orig_start,
7327 u64 block_start, u64 block_len,
cc95bef6
JB		 7328 u64 orig_block_len, u64 ram_bytes,
7329 int type)
69ffb543
JB		 7330{
7331 struct extent_map_tree *em_tree;
7332 struct extent_map *em;
7333 struct btrfs_root *root = BTRFS_I(inode)->root;
7334 int ret;
7335
7336 em_tree = &BTRFS_I(inode)->extent_tree;
7337 em = alloc_extent_map();
7338 if (!em)
7339 return ERR_PTR(-ENOMEM);
7340
7341 em->start = start;
7342 em->orig_start = orig_start;
2ab28f32
JB		 7343 em->mod_start = start;
7344 em->mod_len = len;
69ffb543
JB		 7345 em->len = len;
7346 em->block_len = block_len;
7347 em->block_start = block_start;
7348 em->bdev = root->fs_info->fs_devices->latest_bdev;
b4939680 7349 em->orig_block_len = orig_block_len;
cc95bef6 7350 em->ram_bytes = ram_bytes;
70c8a91c 7351 em->generation = -1;
69ffb543
JB		 7352 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7353 if (type == BTRFS_ORDERED_PREALLOC)
b11e234d 7354 set_bit(EXTENT_FLAG_FILLING, &em->flags);
69ffb543
JB		 7355
7356 do {
7357 btrfs_drop_extent_cache(inode, em->start,
7358 em->start + em->len - 1, 0);
7359 write_lock(&em_tree->lock);
09a2a8f9 7360 ret = add_extent_mapping(em_tree, em, 1);
69ffb543
JB		 7361 write_unlock(&em_tree->lock);
7362 } while (ret == -EEXIST);
7363
7364 if (ret) {
7365 free_extent_map(em);
7366 return ERR_PTR(ret);
7367 }
7368
7369 return em;
7370}
7371
7372
4b46fce2
JB		 7373static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7374 struct buffer_head *bh_result, int create)
7375{
7376 struct extent_map *em;
7377 struct btrfs_root *root = BTRFS_I(inode)->root;
eb838e73 7378 struct extent_state *cached_state = NULL;
4b46fce2 7379 u64 start = iblock << inode->i_blkbits;
eb838e73 7380 u64 lockstart, lockend;
4b46fce2 7381 u64 len = bh_result->b_size;
e1cbbfa5 7382 u64 *outstanding_extents = NULL;
eb838e73 7383 int unlock_bits = EXTENT_LOCKED;
0934856d 7384 int ret = 0;
eb838e73 7385
172a5049 7386 if (create)
3266789f 7387 unlock_bits |= EXTENT_DIRTY;
172a5049 7388 else
c329861d 7389 len = min_t(u64, len, root->sectorsize);
eb838e73 7390
c329861d
JB		 7391 lockstart = start;
7392 lockend = start + len - 1;
7393
e1cbbfa5
JB		 7394 if (current->journal_info) {
7395 /*
7396 * Need to pull our outstanding extents and set journal_info to NULL so
7397 * that anything that needs to check if there's a transction doesn't get
7398 * confused.
7399 */
7400 outstanding_extents = current->journal_info;
7401 current->journal_info = NULL;
7402 }
7403
eb838e73
JB		 7404 /*
7405 * If this errors out it's because we couldn't invalidate pagecache for
7406 * this range and we need to fallback to buffered.
7407 */
7408 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
7409 return -ENOTBLK;
7410
4b46fce2 7411 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
eb838e73
JB		 7412 if (IS_ERR(em)) {
7413 ret = PTR_ERR(em);
7414 goto unlock_err;
7415 }
4b46fce2
JB		 7416
7417 /*
7418 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7419 * io. INLINE is special, and we could probably kludge it in here, but
7420 * it's still buffered so for safety lets just fall back to the generic
7421 * buffered path.
7422 *
7423 * For COMPRESSED we _have_ to read the entire extent in so we can
7424 * decompress it, so there will be buffering required no matter what we
7425 * do, so go ahead and fallback to buffered.
7426 *
7427 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7428 * to buffered IO. Don't blame me, this is the price we pay for using
7429 * the generic code.
7430 */
7431 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7432 em->block_start == EXTENT_MAP_INLINE) {
7433 free_extent_map(em);
eb838e73
JB		 7434 ret = -ENOTBLK;
7435 goto unlock_err;
4b46fce2
JB		 7436 }
7437
7438 /* Just a good old fashioned hole, return */
7439 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7440 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7441 free_extent_map(em);
eb838e73 7442 goto unlock_err;
4b46fce2
JB		 7443 }
7444
7445 /*
7446 * We don't allocate a new extent in the following cases
7447 *
7448 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7449 * existing extent.
7450 * 2) The extent is marked as PREALLOC. We're good to go here and can
7451 * just use the extent.
7452 *
7453 */
46bfbb5c 7454 if (!create) {
eb838e73
JB		 7455 len = min(len, em->len - (start - em->start));
7456 lockstart = start + len;
7457 goto unlock;
46bfbb5c 7458 }
4b46fce2
JB		 7459
7460 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7461 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7462 em->block_start != EXTENT_MAP_HOLE)) {
4b46fce2 7463 int type;
eb384b55 7464 u64 block_start, orig_start, orig_block_len, ram_bytes;
4b46fce2
JB		 7465
7466 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7467 type = BTRFS_ORDERED_PREALLOC;
7468 else
7469 type = BTRFS_ORDERED_NOCOW;
46bfbb5c 7470 len = min(len, em->len - (start - em->start));
4b46fce2 7471 block_start = em->block_start + (start - em->start);
46bfbb5c 7472
00361589 7473 if (can_nocow_extent(inode, start, &len, &orig_start,
7ee9e440 7474 &orig_block_len, &ram_bytes) == 1) {
69ffb543
JB		 7475 if (type == BTRFS_ORDERED_PREALLOC) {
7476 free_extent_map(em);
7477 em = create_pinned_em(inode, start, len,
7478 orig_start,
b4939680 7479 block_start, len,
cc95bef6
JB		 7480 orig_block_len,
7481 ram_bytes, type);
555e1286
FM		 7482 if (IS_ERR(em)) {
7483 ret = PTR_ERR(em);
69ffb543 7484 goto unlock_err;
555e1286 7485 }
69ffb543
JB		 7486 }
7487
46bfbb5c
CM		 7488 ret = btrfs_add_ordered_extent_dio(inode, start,
7489 block_start, len, len, type);
46bfbb5c
CM		 7490 if (ret) {
7491 free_extent_map(em);
eb838e73 7492 goto unlock_err;
46bfbb5c
CM		 7493 }
7494 goto unlock;
4b46fce2 7495 }
4b46fce2 7496 }
00361589 7497
46bfbb5c
CM		 7498 /*
7499 * this will cow the extent, reset the len in case we changed
7500 * it above
7501 */
7502 len = bh_result->b_size;
70c8a91c
JB		 7503 free_extent_map(em);
7504 em = btrfs_new_extent_direct(inode, start, len);
eb838e73
JB		 7505 if (IS_ERR(em)) {
7506 ret = PTR_ERR(em);
7507 goto unlock_err;
7508 }
46bfbb5c
CM		 7509 len = min(len, em->len - (start - em->start));
7510unlock:
4b46fce2
JB		 7511 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7512 inode->i_blkbits;
46bfbb5c 7513 bh_result->b_size = len;
4b46fce2
JB		 7514 bh_result->b_bdev = em->bdev;
7515 set_buffer_mapped(bh_result);
c3473e83
JB		 7516 if (create) {
7517 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7518 set_buffer_new(bh_result);
7519
7520 /*
7521 * Need to update the i_size under the extent lock so buffered
7522 * readers will get the updated i_size when we unlock.
7523 */
7524 if (start + len > i_size_read(inode))
7525 i_size_write(inode, start + len);
0934856d 7526
e1cbbfa5
JB		 7527 /*
7528 * If we have an outstanding_extents count still set then we're
7529 * within our reservation, otherwise we need to adjust our inode
7530 * counter appropriately.
7531 */
7532 if (*outstanding_extents) {
7533 (*outstanding_extents)--;
7534 } else {
3e05bde8
JB		 7535 spin_lock(&BTRFS_I(inode)->lock);
7536 BTRFS_I(inode)->outstanding_extents++;
7537 spin_unlock(&BTRFS_I(inode)->lock);
7538 }
e1cbbfa5
JB		 7539
7540 current->journal_info = outstanding_extents;
3266789f 7541 btrfs_free_reserved_data_space(inode, len);
c3473e83 7542 }
4b46fce2 7543
eb838e73
JB		 7544 /*
7545 * In the case of write we need to clear and unlock the entire range,
7546 * in the case of read we need to unlock only the end area that we
7547 * aren't using if there is any left over space.
7548 */
24c03fa5 7549 if (lockstart < lockend) {
0934856d
MX		 7550 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7551 lockend, unlock_bits, 1, 0,
7552 &cached_state, GFP_NOFS);
24c03fa5 7553 } else {
eb838e73 7554 free_extent_state(cached_state);
24c03fa5 7555 }
eb838e73 7556
4b46fce2
JB		 7557 free_extent_map(em);
7558
7559 return 0;
eb838e73
JB		 7560
7561unlock_err:
eb838e73
JB		 7562 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7563 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
e1cbbfa5
JB		 7564 if (outstanding_extents)
7565 current->journal_info = outstanding_extents;
eb838e73 7566 return ret;
4b46fce2
JB		 7567}
7568
8b110e39
MX		 7569static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7570 int rw, int mirror_num)
7571{
7572 struct btrfs_root *root = BTRFS_I(inode)->root;
7573 int ret;
7574
7575 BUG_ON(rw & REQ_WRITE);
7576
7577 bio_get(bio);
7578
7579 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7580 BTRFS_WQ_ENDIO_DIO_REPAIR);
7581 if (ret)
7582 goto err;
7583
7584 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7585err:
7586 bio_put(bio);
7587 return ret;
7588}
7589
7590static int btrfs_check_dio_repairable(struct inode *inode,
7591 struct bio *failed_bio,
7592 struct io_failure_record *failrec,
7593 int failed_mirror)
7594{
7595 int num_copies;
7596
7597 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7598 failrec->logical, failrec->len);
7599 if (num_copies == 1) {
7600 /*
7601 * we only have a single copy of the data, so don't bother with
7602 * all the retry and error correction code that follows. no
7603 * matter what the error is, it is very likely to persist.
7604 */
7605 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7606 num_copies, failrec->this_mirror, failed_mirror);
7607 return 0;
7608 }
7609
7610 failrec->failed_mirror = failed_mirror;
7611 failrec->this_mirror++;
7612 if (failrec->this_mirror == failed_mirror)
7613 failrec->this_mirror++;
7614
7615 if (failrec->this_mirror > num_copies) {
7616 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7617 num_copies, failrec->this_mirror, failed_mirror);
7618 return 0;
7619 }
7620
7621 return 1;
7622}
7623
7624static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7625 struct page *page, u64 start, u64 end,
7626 int failed_mirror, bio_end_io_t *repair_endio,
7627 void *repair_arg)
7628{
7629 struct io_failure_record *failrec;
7630 struct bio *bio;
7631 int isector;
7632 int read_mode;
7633 int ret;
7634
7635 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7636
7637 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7638 if (ret)
7639 return ret;
7640
7641 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7642 failed_mirror);
7643 if (!ret) {
7644 free_io_failure(inode, failrec);
7645 return -EIO;
7646 }
7647
7648 if (failed_bio->bi_vcnt > 1)
7649 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7650 else
7651 read_mode = READ_SYNC;
7652
7653 isector = start - btrfs_io_bio(failed_bio)->logical;
7654 isector >>= inode->i_sb->s_blocksize_bits;
7655 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7656 0, isector, repair_endio, repair_arg);
7657 if (!bio) {
7658 free_io_failure(inode, failrec);
7659 return -EIO;
7660 }
7661
7662 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7663 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7664 read_mode, failrec->this_mirror, failrec->in_validation);
7665
7666 ret = submit_dio_repair_bio(inode, bio, read_mode,
7667 failrec->this_mirror);
7668 if (ret) {
7669 free_io_failure(inode, failrec);
7670 bio_put(bio);
7671 }
7672
7673 return ret;
7674}
7675
7676struct btrfs_retry_complete {
7677 struct completion done;
7678 struct inode *inode;
7679 u64 start;
7680 int uptodate;
7681};
7682
7683static void btrfs_retry_endio_nocsum(struct bio *bio, int err)
7684{
7685 struct btrfs_retry_complete *done = bio->bi_private;
7686 struct bio_vec *bvec;
7687 int i;
7688
7689 if (err)
7690 goto end;
7691
7692 done->uptodate = 1;
7693 bio_for_each_segment_all(bvec, bio, i)
7694 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7695end:
7696 complete(&done->done);
7697 bio_put(bio);
7698}
7699
7700static int __btrfs_correct_data_nocsum(struct inode *inode,
7701 struct btrfs_io_bio *io_bio)
4b46fce2 7702{
2c30c71b 7703 struct bio_vec *bvec;
8b110e39 7704 struct btrfs_retry_complete done;
4b46fce2 7705 u64 start;
2c30c71b 7706 int i;
c1dc0896 7707 int ret;
4b46fce2 7708
8b110e39
MX		 7709 start = io_bio->logical;
7710 done.inode = inode;
7711
7712 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7713try_again:
7714 done.uptodate = 0;
7715 done.start = start;
7716 init_completion(&done.done);
7717
7718 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7719 start + bvec->bv_len - 1,
7720 io_bio->mirror_num,
7721 btrfs_retry_endio_nocsum, &done);
7722 if (ret)
7723 return ret;
7724
7725 wait_for_completion(&done.done);
7726
7727 if (!done.uptodate) {
7728 /* We might have another mirror, so try again */
7729 goto try_again;
7730 }
7731
7732 start += bvec->bv_len;
7733 }
7734
7735 return 0;
7736}
7737
7738static void btrfs_retry_endio(struct bio *bio, int err)
7739{
7740 struct btrfs_retry_complete *done = bio->bi_private;
7741 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7742 struct bio_vec *bvec;
7743 int uptodate;
7744 int ret;
7745 int i;
7746
7747 if (err)
7748 goto end;
7749
7750 uptodate = 1;
7751 bio_for_each_segment_all(bvec, bio, i) {
7752 ret = __readpage_endio_check(done->inode, io_bio, i,
7753 bvec->bv_page, 0,
7754 done->start, bvec->bv_len);
7755 if (!ret)
7756 clean_io_failure(done->inode, done->start,
7757 bvec->bv_page, 0);
7758 else
7759 uptodate = 0;
7760 }
7761
7762 done->uptodate = uptodate;
7763end:
7764 complete(&done->done);
7765 bio_put(bio);
7766}
7767
7768static int __btrfs_subio_endio_read(struct inode *inode,
7769 struct btrfs_io_bio *io_bio, int err)
7770{
7771 struct bio_vec *bvec;
7772 struct btrfs_retry_complete done;
7773 u64 start;
7774 u64 offset = 0;
7775 int i;
7776 int ret;
dc380aea 7777
8b110e39 7778 err = 0;
c1dc0896 7779 start = io_bio->logical;
8b110e39
MX		 7780 done.inode = inode;
7781
c1dc0896 7782 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
dc380aea
MX		 7783 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7784 0, start, bvec->bv_len);
8b110e39
MX		 7785 if (likely(!ret))
7786 goto next;
7787try_again:
7788 done.uptodate = 0;
7789 done.start = start;
7790 init_completion(&done.done);
7791
7792 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7793 start + bvec->bv_len - 1,
7794 io_bio->mirror_num,
7795 btrfs_retry_endio, &done);
7796 if (ret) {
7797 err = ret;
7798 goto next;
7799 }
7800
7801 wait_for_completion(&done.done);
7802
7803 if (!done.uptodate) {
7804 /* We might have another mirror, so try again */
7805 goto try_again;
7806 }
7807next:
7808 offset += bvec->bv_len;
4b46fce2 7809 start += bvec->bv_len;
2c30c71b 7810 }
c1dc0896
MX		 7811
7812 return err;
7813}
7814
8b110e39
MX		 7815static int btrfs_subio_endio_read(struct inode *inode,
7816 struct btrfs_io_bio *io_bio, int err)
7817{
7818 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7819
7820 if (skip_csum) {
7821 if (unlikely(err))
7822 return __btrfs_correct_data_nocsum(inode, io_bio);
7823 else
7824 return 0;
7825 } else {
7826 return __btrfs_subio_endio_read(inode, io_bio, err);
7827 }
7828}
7829
c1dc0896
MX		 7830static void btrfs_endio_direct_read(struct bio *bio, int err)
7831{
7832 struct btrfs_dio_private *dip = bio->bi_private;
7833 struct inode *inode = dip->inode;
7834 struct bio *dio_bio;
7835 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7836
8b110e39
MX		 7837 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
7838 err = btrfs_subio_endio_read(inode, io_bio, err);
c1dc0896 7839
4b46fce2 7840 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
d0082371 7841 dip->logical_offset + dip->bytes - 1);
9be3395b 7842 dio_bio = dip->dio_bio;
4b46fce2 7843
4b46fce2 7844 kfree(dip);
c0da7aa1
JB		 7845
7846 /* If we had a csum failure make sure to clear the uptodate flag */
7847 if (err)
9be3395b
CM		 7848 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7849 dio_end_io(dio_bio, err);
23ea8e5a
MX		 7850
7851 if (io_bio->end_io)
7852 io_bio->end_io(io_bio, err);
9be3395b 7853 bio_put(bio);
4b46fce2
JB		 7854}
7855
7856static void btrfs_endio_direct_write(struct bio *bio, int err)
7857{
7858 struct btrfs_dio_private *dip = bio->bi_private;
7859 struct inode *inode = dip->inode;
7860 struct btrfs_root *root = BTRFS_I(inode)->root;
4b46fce2 7861 struct btrfs_ordered_extent *ordered = NULL;
163cf09c
CM		 7862 u64 ordered_offset = dip->logical_offset;
7863 u64 ordered_bytes = dip->bytes;
9be3395b 7864 struct bio *dio_bio;
4b46fce2
JB		 7865 int ret;
7866
7867 if (err)
7868 goto out_done;
163cf09c
CM		 7869again:
7870 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7871 &ordered_offset,
5fd02043 7872 ordered_bytes, !err);
4b46fce2 7873 if (!ret)
163cf09c 7874 goto out_test;
4b46fce2 7875
9e0af237
LB		 7876 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
7877 finish_ordered_fn, NULL, NULL);
fccb5d86
QW		 7878 btrfs_queue_work(root->fs_info->endio_write_workers,
7879 &ordered->work);
163cf09c
CM		 7880out_test:
7881 /*
7882 * our bio might span multiple ordered extents. If we haven't
7883 * completed the accounting for the whole dio, go back and try again
7884 */
7885 if (ordered_offset < dip->logical_offset + dip->bytes) {
7886 ordered_bytes = dip->logical_offset + dip->bytes -
7887 ordered_offset;
5fd02043 7888 ordered = NULL;
163cf09c
CM		 7889 goto again;
7890 }
4b46fce2 7891out_done:
9be3395b 7892 dio_bio = dip->dio_bio;
4b46fce2 7893
4b46fce2 7894 kfree(dip);
c0da7aa1
JB		 7895
7896 /* If we had an error make sure to clear the uptodate flag */
7897 if (err)
9be3395b
CM		 7898 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7899 dio_end_io(dio_bio, err);
7900 bio_put(bio);
4b46fce2
JB		 7901}
7902
eaf25d93
CM		 7903static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7904 struct bio *bio, int mirror_num,
7905 unsigned long bio_flags, u64 offset)
7906{
7907 int ret;
7908 struct btrfs_root *root = BTRFS_I(inode)->root;
7909 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
79787eaa 7910 BUG_ON(ret); /* -ENOMEM */
eaf25d93
CM		 7911 return 0;
7912}
7913
e65e1535
MX		 7914static void btrfs_end_dio_bio(struct bio *bio, int err)
7915{
7916 struct btrfs_dio_private *dip = bio->bi_private;
7917
8b110e39
MX		 7918 if (err)
7919 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7920 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7921 btrfs_ino(dip->inode), bio->bi_rw,
7922 (unsigned long long)bio->bi_iter.bi_sector,
7923 bio->bi_iter.bi_size, err);
7924
7925 if (dip->subio_endio)
7926 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
c1dc0896
MX		 7927
7928 if (err) {
e65e1535
MX		 7929 dip->errors = 1;
7930
7931 /*
7932 * before atomic variable goto zero, we must make sure
7933 * dip->errors is perceived to be set.
7934 */
4e857c58 7935 smp_mb__before_atomic();
e65e1535
MX		 7936 }
7937
7938 /* if there are more bios still pending for this dio, just exit */
7939 if (!atomic_dec_and_test(&dip->pending_bios))
7940 goto out;
7941
9be3395b 7942 if (dip->errors) {
e65e1535 7943 bio_io_error(dip->orig_bio);
9be3395b
CM		 7944 } else {
7945 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
e65e1535
MX		 7946 bio_endio(dip->orig_bio, 0);
7947 }
7948out:
7949 bio_put(bio);
7950}
7951
7952static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7953 u64 first_sector, gfp_t gfp_flags)
7954{
7955 int nr_vecs = bio_get_nr_vecs(bdev);
7956 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7957}
7958
c1dc0896
MX		 7959static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
7960 struct inode *inode,
7961 struct btrfs_dio_private *dip,
7962 struct bio *bio,
7963 u64 file_offset)
7964{
7965 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7966 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
7967 int ret;
7968
7969 /*
7970 * We load all the csum data we need when we submit
7971 * the first bio to reduce the csum tree search and
7972 * contention.
7973 */
7974 if (dip->logical_offset == file_offset) {
7975 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
7976 file_offset);
7977 if (ret)
7978 return ret;
7979 }
7980
7981 if (bio == dip->orig_bio)
7982 return 0;
7983
7984 file_offset -= dip->logical_offset;
7985 file_offset >>= inode->i_sb->s_blocksize_bits;
7986 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
7987
7988 return 0;
7989}
7990
e65e1535
MX		 7991static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7992 int rw, u64 file_offset, int skip_sum,
c329861d 7993 int async_submit)
e65e1535 7994{
facc8a22 7995 struct btrfs_dio_private *dip = bio->bi_private;
e65e1535
MX		 7996 int write = rw & REQ_WRITE;
7997 struct btrfs_root *root = BTRFS_I(inode)->root;
7998 int ret;
7999
b812ce28
JB		 8000 if (async_submit)
8001 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8002
e65e1535 8003 bio_get(bio);
5fd02043
JB		 8004
8005 if (!write) {
bfebd8b5
DS		 8006 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8007 BTRFS_WQ_ENDIO_DATA);
5fd02043
JB		 8008 if (ret)
8009 goto err;
8010 }
e65e1535 8011
1ae39938
JB		 8012 if (skip_sum)
8013 goto map;
8014
8015 if (write && async_submit) {
e65e1535
MX		 8016 ret = btrfs_wq_submit_bio(root->fs_info,
8017 inode, rw, bio, 0, 0,
8018 file_offset,
8019 __btrfs_submit_bio_start_direct_io,
8020 __btrfs_submit_bio_done);
8021 goto err;
1ae39938
JB		 8022 } else if (write) {
8023 /*
8024 * If we aren't doing async submit, calculate the csum of the
8025 * bio now.
8026 */
8027 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8028 if (ret)
8029 goto err;
23ea8e5a 8030 } else {
c1dc0896
MX		 8031 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8032 file_offset);
c2db1073
TI		 8033 if (ret)
8034 goto err;
8035 }
1ae39938
JB		 8036map:
8037 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
e65e1535
MX		 8038err:
8039 bio_put(bio);
8040 return ret;
8041}
8042
8043static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
8044 int skip_sum)
8045{
8046 struct inode *inode = dip->inode;
8047 struct btrfs_root *root = BTRFS_I(inode)->root;
e65e1535
MX		 8048 struct bio *bio;
8049 struct bio *orig_bio = dip->orig_bio;
8050 struct bio_vec *bvec = orig_bio->bi_io_vec;
4f024f37 8051 u64 start_sector = orig_bio->bi_iter.bi_sector;
e65e1535
MX		 8052 u64 file_offset = dip->logical_offset;
8053 u64 submit_len = 0;
8054 u64 map_length;
8055 int nr_pages = 0;
23ea8e5a 8056 int ret;
1ae39938 8057 int async_submit = 0;
e65e1535 8058
4f024f37 8059 map_length = orig_bio->bi_iter.bi_size;
53b381b3 8060 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
e65e1535 8061 &map_length, NULL, 0);
7a5c3c9b 8062 if (ret)
e65e1535 8063 return -EIO;
facc8a22 8064
4f024f37 8065 if (map_length >= orig_bio->bi_iter.bi_size) {
02f57c7a 8066 bio = orig_bio;
c1dc0896 8067 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
02f57c7a
JB		 8068 goto submit;
8069 }
8070
53b381b3 8071 /* async crcs make it difficult to collect full stripe writes. */
ffe2d203 8072 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
53b381b3
DW		 8073 async_submit = 0;
8074 else
8075 async_submit = 1;
8076
02f57c7a
JB		 8077 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8078 if (!bio)
8079 return -ENOMEM;
7a5c3c9b 8080
02f57c7a
JB		 8081 bio->bi_private = dip;
8082 bio->bi_end_io = btrfs_end_dio_bio;
c1dc0896 8083 btrfs_io_bio(bio)->logical = file_offset;
02f57c7a
JB		 8084 atomic_inc(&dip->pending_bios);
8085
e65e1535 8086 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
ee39b432 8087 if (map_length < submit_len + bvec->bv_len ||
e65e1535 8088 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
ee39b432 8089 bvec->bv_offset) < bvec->bv_len) {
e65e1535
MX		 8090 /*
8091 * inc the count before we submit the bio so
8092 * we know the end IO handler won't happen before
8093 * we inc the count. Otherwise, the dip might get freed
8094 * before we're done setting it up
8095 */
8096 atomic_inc(&dip->pending_bios);
8097 ret = __btrfs_submit_dio_bio(bio, inode, rw,
8098 file_offset, skip_sum,
c329861d 8099 async_submit);
e65e1535
MX		 8100 if (ret) {
8101 bio_put(bio);
8102 atomic_dec(&dip->pending_bios);
8103 goto out_err;
8104 }
8105
e65e1535
MX		 8106 start_sector += submit_len >> 9;
8107 file_offset += submit_len;
8108
8109 submit_len = 0;
8110 nr_pages = 0;
8111
8112 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8113 start_sector, GFP_NOFS);
8114 if (!bio)
8115 goto out_err;
8116 bio->bi_private = dip;
8117 bio->bi_end_io = btrfs_end_dio_bio;
c1dc0896 8118 btrfs_io_bio(bio)->logical = file_offset;
e65e1535 8119
4f024f37 8120 map_length = orig_bio->bi_iter.bi_size;
53b381b3 8121 ret = btrfs_map_block(root->fs_info, rw,
3ec706c8 8122 start_sector << 9,
e65e1535
MX		 8123 &map_length, NULL, 0);
8124 if (ret) {
8125 bio_put(bio);
8126 goto out_err;
8127 }
8128 } else {
8129 submit_len += bvec->bv_len;
67871254 8130 nr_pages++;
e65e1535
MX		 8131 bvec++;
8132 }
8133 }
8134
02f57c7a 8135submit:
e65e1535 8136 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
c329861d 8137 async_submit);
e65e1535
MX		 8138 if (!ret)
8139 return 0;
8140
8141 bio_put(bio);
8142out_err:
8143 dip->errors = 1;
8144 /*
8145 * before atomic variable goto zero, we must
8146 * make sure dip->errors is perceived to be set.
8147 */
4e857c58 8148 smp_mb__before_atomic();
e65e1535
MX		 8149 if (atomic_dec_and_test(&dip->pending_bios))
8150 bio_io_error(dip->orig_bio);
8151
8152 /* bio_end_io() will handle error, so we needn't return it */
8153 return 0;
8154}
8155
9be3395b
CM		 8156static void btrfs_submit_direct(int rw, struct bio *dio_bio,
8157 struct inode *inode, loff_t file_offset)
4b46fce2
JB		 8158{
8159 struct btrfs_root *root = BTRFS_I(inode)->root;
8160 struct btrfs_dio_private *dip;
9be3395b 8161 struct bio *io_bio;
23ea8e5a 8162 struct btrfs_io_bio *btrfs_bio;
4b46fce2 8163 int skip_sum;
7b6d91da 8164 int write = rw & REQ_WRITE;
4b46fce2
JB		 8165 int ret = 0;
8166
8167 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8168
9be3395b 8169 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
9be3395b
CM		 8170 if (!io_bio) {
8171 ret = -ENOMEM;
8172 goto free_ordered;
8173 }
8174
c1dc0896 8175 dip = kzalloc(sizeof(*dip), GFP_NOFS);
4b46fce2
JB		 8176 if (!dip) {
8177 ret = -ENOMEM;
9be3395b 8178 goto free_io_bio;
4b46fce2 8179 }
4b46fce2 8180
9be3395b 8181 dip->private = dio_bio->bi_private;
4b46fce2
JB		 8182 dip->inode = inode;
8183 dip->logical_offset = file_offset;
4f024f37
KO		 8184 dip->bytes = dio_bio->bi_iter.bi_size;
8185 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
9be3395b 8186 io_bio->bi_private = dip;
9be3395b
CM		 8187 dip->orig_bio = io_bio;
8188 dip->dio_bio = dio_bio;
e65e1535 8189 atomic_set(&dip->pending_bios, 0);
c1dc0896
MX		 8190 btrfs_bio = btrfs_io_bio(io_bio);
8191 btrfs_bio->logical = file_offset;
4b46fce2 8192
c1dc0896 8193 if (write) {
9be3395b 8194 io_bio->bi_end_io = btrfs_endio_direct_write;
c1dc0896 8195 } else {
9be3395b 8196 io_bio->bi_end_io = btrfs_endio_direct_read;
c1dc0896
MX		 8197 dip->subio_endio = btrfs_subio_endio_read;
8198 }
4b46fce2 8199
e65e1535
MX		 8200 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8201 if (!ret)
eaf25d93 8202 return;
9be3395b 8203
23ea8e5a
MX		 8204 if (btrfs_bio->end_io)
8205 btrfs_bio->end_io(btrfs_bio, ret);
9be3395b
CM		 8206free_io_bio:
8207 bio_put(io_bio);
8208
4b46fce2
JB		 8209free_ordered:
8210 /*
8211 * If this is a write, we need to clean up the reserved space and kill
8212 * the ordered extent.
8213 */
8214 if (write) {
8215 struct btrfs_ordered_extent *ordered;
955256f2 8216 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
4b46fce2
JB		 8217 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
8218 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
8219 btrfs_free_reserved_extent(root, ordered->start,
e570fd27 8220 ordered->disk_len, 1);
4b46fce2
JB		 8221 btrfs_put_ordered_extent(ordered);
8222 btrfs_put_ordered_extent(ordered);
8223 }
9be3395b 8224 bio_endio(dio_bio, ret);
4b46fce2
JB		 8225}
8226
5a5f79b5 8227static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
28060d5d 8228 const struct iov_iter *iter, loff_t offset)
5a5f79b5
CM		 8229{
8230 int seg;
a1b75f7d 8231 int i;
5a5f79b5
CM		 8232 unsigned blocksize_mask = root->sectorsize - 1;
8233 ssize_t retval = -EINVAL;
5a5f79b5
CM		 8234
8235 if (offset & blocksize_mask)
8236 goto out;
8237
28060d5d
AV		 8238 if (iov_iter_alignment(iter) & blocksize_mask)
8239 goto out;
a1b75f7d 8240
28060d5d
AV		 8241 /* If this is a write we don't need to check anymore */
8242 if (rw & WRITE)
8243 return 0;
8244 /*
8245 * Check to make sure we don't have duplicate iov_base's in this
8246 * iovec, if so return EINVAL, otherwise we'll get csum errors
8247 * when reading back.
8248 */
8249 for (seg = 0; seg < iter->nr_segs; seg++) {
8250 for (i = seg + 1; i < iter->nr_segs; i++) {
8251 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
a1b75f7d
JB		 8252 goto out;
8253 }
5a5f79b5
CM		 8254 }
8255 retval = 0;
8256out:
8257 return retval;
8258}
eb838e73 8259
16432985 8260static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
d8d3d94b 8261 struct iov_iter *iter, loff_t offset)
16432985 8262{
4b46fce2
JB		 8263 struct file *file = iocb->ki_filp;
8264 struct inode *inode = file->f_mapping->host;
e1cbbfa5 8265 u64 outstanding_extents = 0;
0934856d 8266 size_t count = 0;
2e60a51e 8267 int flags = 0;
38851cc1
MX		 8268 bool wakeup = true;
8269 bool relock = false;
0934856d 8270 ssize_t ret;
4b46fce2 8271
28060d5d 8272 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iter, offset))
5a5f79b5 8273 return 0;
3f7c579c 8274
38851cc1 8275 atomic_inc(&inode->i_dio_count);
4e857c58 8276 smp_mb__after_atomic();
38851cc1 8277
0e267c44 8278 /*
41bd9ca4
MX		 8279 * The generic stuff only does filemap_write_and_wait_range, which
8280 * isn't enough if we've written compressed pages to this area, so
8281 * we need to flush the dirty pages again to make absolutely sure
8282 * that any outstanding dirty pages are on disk.
0e267c44 8283 */
a6cbcd4a 8284 count = iov_iter_count(iter);
41bd9ca4
MX		 8285 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8286 &BTRFS_I(inode)->runtime_flags))
9a025a08
WS		 8287 filemap_fdatawrite_range(inode->i_mapping, offset,
8288 offset + count - 1);
0e267c44 8289
0934856d 8290 if (rw & WRITE) {
38851cc1
MX		 8291 /*
8292 * If the write DIO is beyond the EOF, we need update
8293 * the isize, but it is protected by i_mutex. So we can
8294 * not unlock the i_mutex at this case.
8295 */
8296 if (offset + count <= inode->i_size) {
8297 mutex_unlock(&inode->i_mutex);
8298 relock = true;
8299 }
0934856d
MX		 8300 ret = btrfs_delalloc_reserve_space(inode, count);
8301 if (ret)
38851cc1 8302 goto out;
e1cbbfa5
JB		 8303 outstanding_extents = div64_u64(count +
8304 BTRFS_MAX_EXTENT_SIZE - 1,
8305 BTRFS_MAX_EXTENT_SIZE);
8306
8307 /*
8308 * We need to know how many extents we reserved so that we can
8309 * do the accounting properly if we go over the number we
8310 * originally calculated. Abuse current->journal_info for this.
8311 */
8312 current->journal_info = &outstanding_extents;
ee39b432
DS		 8313 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8314 &BTRFS_I(inode)->runtime_flags)) {
38851cc1
MX		 8315 inode_dio_done(inode);
8316 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8317 wakeup = false;
0934856d
MX		 8318 }
8319
8320 ret = __blockdev_direct_IO(rw, iocb, inode,
8321 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
31b14039 8322 iter, offset, btrfs_get_blocks_direct, NULL,
2e60a51e 8323 btrfs_submit_direct, flags);
0934856d 8324 if (rw & WRITE) {
e1cbbfa5 8325 current->journal_info = NULL;
0934856d
MX		 8326 if (ret < 0 && ret != -EIOCBQUEUED)
8327 btrfs_delalloc_release_space(inode, count);
172a5049 8328 else if (ret >= 0 && (size_t)ret < count)
0934856d
MX		 8329 btrfs_delalloc_release_space(inode,
8330 count - (size_t)ret);
0934856d 8331 }
38851cc1 8332out:
2e60a51e
MX		 8333 if (wakeup)
8334 inode_dio_done(inode);
38851cc1
MX		 8335 if (relock)
8336 mutex_lock(&inode->i_mutex);
0934856d
MX		 8337
8338 return ret;
16432985
CM		 8339}
8340
05dadc09
TI		 8341#define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8342
1506fcc8
YS		 8343static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8344 __u64 start, __u64 len)
8345{
05dadc09
TI		 8346 int ret;
8347
8348 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8349 if (ret)
8350 return ret;
8351
ec29ed5b 8352 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
1506fcc8
YS		 8353}
8354
a52d9a80 8355int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 8356{
d1310b2e
CM		 8357 struct extent_io_tree *tree;
8358 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 8359 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
9ebefb18 8360}
1832a6d5 8361
a52d9a80 8362static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 8363{
d1310b2e 8364 struct extent_io_tree *tree;
b888db2b
CM		 8365
8366
8367 if (current->flags & PF_MEMALLOC) {
8368 redirty_page_for_writepage(wbc, page);
8369 unlock_page(page);
8370 return 0;
8371 }
d1310b2e 8372 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 8373 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM		 8374}
8375
48a3b636
ES		 8376static int btrfs_writepages(struct address_space *mapping,
8377 struct writeback_control *wbc)
b293f02e 8378{
d1310b2e 8379 struct extent_io_tree *tree;
771ed689 8380
d1310b2e 8381 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 8382 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8383}
8384
3ab2fb5a
CM		 8385static int
8386btrfs_readpages(struct file *file, struct address_space *mapping,
8387 struct list_head *pages, unsigned nr_pages)
8388{
d1310b2e
CM		 8389 struct extent_io_tree *tree;
8390 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 8391 return extent_readpages(tree, mapping, pages, nr_pages,
8392 btrfs_get_extent);
8393}
e6dcd2dc 8394static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 8395{
d1310b2e
CM		 8396 struct extent_io_tree *tree;
8397 struct extent_map_tree *map;
a52d9a80 8398 int ret;
8c2383c3 8399
d1310b2e
CM		 8400 tree = &BTRFS_I(page->mapping->host)->io_tree;
8401 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 8402 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 8403 if (ret == 1) {
8404 ClearPagePrivate(page);
8405 set_page_private(page, 0);
8406 page_cache_release(page);
39279cc3 8407 }
a52d9a80 8408 return ret;
39279cc3
CM		 8409}
8410
e6dcd2dc
CM		 8411static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8412{
98509cfc
CM		 8413 if (PageWriteback(page) || PageDirty(page))
8414 return 0;
b335b003 8415 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
e6dcd2dc
CM		 8416}
8417
d47992f8
LC		 8418static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8419 unsigned int length)
39279cc3 8420{
5fd02043 8421 struct inode *inode = page->mapping->host;
d1310b2e 8422 struct extent_io_tree *tree;
e6dcd2dc 8423 struct btrfs_ordered_extent *ordered;
2ac55d41 8424 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 8425 u64 page_start = page_offset(page);
8426 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
131e404a 8427 int inode_evicting = inode->i_state & I_FREEING;
39279cc3 8428
8b62b72b
CM		 8429 /*
8430 * we have the page locked, so new writeback can't start,
8431 * and the dirty bit won't be cleared while we are here.
8432 *
8433 * Wait for IO on this page so that we can safely clear
8434 * the PagePrivate2 bit and do ordered accounting
8435 */
e6dcd2dc 8436 wait_on_page_writeback(page);
8b62b72b 8437
5fd02043 8438 tree = &BTRFS_I(inode)->io_tree;
e6dcd2dc
CM		 8439 if (offset) {
8440 btrfs_releasepage(page, GFP_NOFS);
8441 return;
8442 }
131e404a
FDBM		 8443
8444 if (!inode_evicting)
8445 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
8446 ordered = btrfs_lookup_ordered_extent(inode, page_start);
e6dcd2dc 8447 if (ordered) {
eb84ae03
CM		 8448 /*
8449 * IO on this page will never be started, so we need
8450 * to account for any ordered extents now
8451 */
131e404a
FDBM		 8452 if (!inode_evicting)
8453 clear_extent_bit(tree, page_start, page_end,
8454 EXTENT_DIRTY | EXTENT_DELALLOC |
8455 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8456 EXTENT_DEFRAG, 1, 0, &cached_state,
8457 GFP_NOFS);
8b62b72b
CM		 8458 /*
8459 * whoever cleared the private bit is responsible
8460 * for the finish_ordered_io
8461 */
77cef2ec
JB		 8462 if (TestClearPagePrivate2(page)) {
8463 struct btrfs_ordered_inode_tree *tree;
8464 u64 new_len;
8465
8466 tree = &BTRFS_I(inode)->ordered_tree;
8467
8468 spin_lock_irq(&tree->lock);
8469 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8470 new_len = page_start - ordered->file_offset;
8471 if (new_len < ordered->truncated_len)
8472 ordered->truncated_len = new_len;
8473 spin_unlock_irq(&tree->lock);
8474
8475 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8476 page_start,
8477 PAGE_CACHE_SIZE, 1))
8478 btrfs_finish_ordered_io(ordered);
8b62b72b 8479 }
e6dcd2dc 8480 btrfs_put_ordered_extent(ordered);
131e404a
FDBM		 8481 if (!inode_evicting) {
8482 cached_state = NULL;
8483 lock_extent_bits(tree, page_start, page_end, 0,
8484 &cached_state);
8485 }
8486 }
8487
8488 if (!inode_evicting) {
8489 clear_extent_bit(tree, page_start, page_end,
8490 EXTENT_LOCKED | EXTENT_DIRTY |
8491 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8492 EXTENT_DEFRAG, 1, 1,
8493 &cached_state, GFP_NOFS);
8494
8495 __btrfs_releasepage(page, GFP_NOFS);
e6dcd2dc 8496 }
e6dcd2dc 8497
4a096752 8498 ClearPageChecked(page);
9ad6b7bc 8499 if (PagePrivate(page)) {
9ad6b7bc
CM		 8500 ClearPagePrivate(page);
8501 set_page_private(page, 0);
8502 page_cache_release(page);
8503 }
39279cc3
CM		 8504}
8505
9ebefb18
CM		 8506/*
8507 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8508 * called from a page fault handler when a page is first dirtied. Hence we must
8509 * be careful to check for EOF conditions here. We set the page up correctly
8510 * for a written page which means we get ENOSPC checking when writing into
8511 * holes and correct delalloc and unwritten extent mapping on filesystems that
8512 * support these features.
8513 *
8514 * We are not allowed to take the i_mutex here so we have to play games to
8515 * protect against truncate races as the page could now be beyond EOF. Because
8516 * vmtruncate() writes the inode size before removing pages, once we have the
8517 * page lock we can determine safely if the page is beyond EOF. If it is not
8518 * beyond EOF, then the page is guaranteed safe against truncation until we
8519 * unlock the page.
8520 */
c2ec175c 8521int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
9ebefb18 8522{
c2ec175c 8523 struct page *page = vmf->page;
496ad9aa 8524 struct inode *inode = file_inode(vma->vm_file);
1832a6d5 8525 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 8526 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8527 struct btrfs_ordered_extent *ordered;
2ac55d41 8528 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 8529 char *kaddr;
8530 unsigned long zero_start;
9ebefb18 8531 loff_t size;
1832a6d5 8532 int ret;
9998eb70 8533 int reserved = 0;
a52d9a80 8534 u64 page_start;
e6dcd2dc 8535 u64 page_end;
9ebefb18 8536
b2b5ef5c 8537 sb_start_pagefault(inode->i_sb);
0ca1f7ce 8538 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
9998eb70 8539 if (!ret) {
e41f941a 8540 ret = file_update_time(vma->vm_file);
9998eb70
CM		 8541 reserved = 1;
8542 }
56a76f82
NP		 8543 if (ret) {
8544 if (ret == -ENOMEM)
8545 ret = VM_FAULT_OOM;
8546 else /* -ENOSPC, -EIO, etc */
8547 ret = VM_FAULT_SIGBUS;
9998eb70
CM		 8548 if (reserved)
8549 goto out;
8550 goto out_noreserve;
56a76f82 8551 }
1832a6d5 8552
56a76f82 8553 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
e6dcd2dc 8554again:
9ebefb18 8555 lock_page(page);
9ebefb18 8556 size = i_size_read(inode);
e6dcd2dc
CM		 8557 page_start = page_offset(page);
8558 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 8559
9ebefb18 8560 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 8561 (page_start >= size)) {
9ebefb18
CM		 8562 /* page got truncated out from underneath us */
8563 goto out_unlock;
8564 }
e6dcd2dc
CM		 8565 wait_on_page_writeback(page);
8566
d0082371 8567 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 8568 set_page_extent_mapped(page);
8569
eb84ae03
CM		 8570 /*
8571 * we can't set the delalloc bits if there are pending ordered
8572 * extents. Drop our locks and wait for them to finish
8573 */
e6dcd2dc
CM		 8574 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8575 if (ordered) {
2ac55d41
JB		 8576 unlock_extent_cached(io_tree, page_start, page_end,
8577 &cached_state, GFP_NOFS);
e6dcd2dc 8578 unlock_page(page);
eb84ae03 8579 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 8580 btrfs_put_ordered_extent(ordered);
8581 goto again;
8582 }
8583
fbf19087
JB		 8584 /*
8585 * XXX - page_mkwrite gets called every time the page is dirtied, even
8586 * if it was already dirty, so for space accounting reasons we need to
8587 * clear any delalloc bits for the range we are fixing to save. There
8588 * is probably a better way to do this, but for now keep consistent with
8589 * prepare_pages in the normal write path.
8590 */
2ac55d41 8591 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 8592 EXTENT_DIRTY | EXTENT_DELALLOC |
8593 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 8594 0, 0, &cached_state, GFP_NOFS);
fbf19087 8595
2ac55d41
JB		 8596 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
8597 &cached_state);
9ed74f2d 8598 if (ret) {
2ac55d41
JB		 8599 unlock_extent_cached(io_tree, page_start, page_end,
8600 &cached_state, GFP_NOFS);
9ed74f2d
JB		 8601 ret = VM_FAULT_SIGBUS;
8602 goto out_unlock;
8603 }
e6dcd2dc 8604 ret = 0;
9ebefb18
CM		 8605
8606 /* page is wholly or partially inside EOF */
a52d9a80 8607 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 8608 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 8609 else
e6dcd2dc 8610 zero_start = PAGE_CACHE_SIZE;
9ebefb18 8611
e6dcd2dc
CM		 8612 if (zero_start != PAGE_CACHE_SIZE) {
8613 kaddr = kmap(page);
8614 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
8615 flush_dcache_page(page);
8616 kunmap(page);
8617 }
247e743c 8618 ClearPageChecked(page);
e6dcd2dc 8619 set_page_dirty(page);
50a9b214 8620 SetPageUptodate(page);
5a3f23d5 8621
257c62e1
CM		 8622 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8623 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
46d8bc34 8624 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
257c62e1 8625
2ac55d41 8626 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9ebefb18
CM		 8627
8628out_unlock:
b2b5ef5c
JK		 8629 if (!ret) {
8630 sb_end_pagefault(inode->i_sb);
50a9b214 8631 return VM_FAULT_LOCKED;
b2b5ef5c 8632 }
9ebefb18 8633 unlock_page(page);
1832a6d5 8634out:
ec39e180 8635 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
9998eb70 8636out_noreserve:
b2b5ef5c 8637 sb_end_pagefault(inode->i_sb);
9ebefb18
CM		 8638 return ret;
8639}
8640
a41ad394 8641static int btrfs_truncate(struct inode *inode)
39279cc3
CM		 8642{
8643 struct btrfs_root *root = BTRFS_I(inode)->root;
fcb80c2a 8644 struct btrfs_block_rsv *rsv;
a71754fc 8645 int ret = 0;
3893e33b 8646 int err = 0;
39279cc3 8647 struct btrfs_trans_handle *trans;
dbe674a9 8648 u64 mask = root->sectorsize - 1;
07127184 8649 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3 8650
0ef8b726
JB		 8651 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8652 (u64)-1);
8653 if (ret)
8654 return ret;
39279cc3 8655
fcb80c2a
JB		 8656 /*
8657 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8658 * 3 things going on here
8659 *
8660 * 1) We need to reserve space for our orphan item and the space to
8661 * delete our orphan item. Lord knows we don't want to have a dangling
8662 * orphan item because we didn't reserve space to remove it.
8663 *
8664 * 2) We need to reserve space to update our inode.
8665 *
8666 * 3) We need to have something to cache all the space that is going to
8667 * be free'd up by the truncate operation, but also have some slack
8668 * space reserved in case it uses space during the truncate (thank you
8669 * very much snapshotting).
8670 *
8671 * And we need these to all be seperate. The fact is we can use alot of
8672 * space doing the truncate, and we have no earthly idea how much space
8673 * we will use, so we need the truncate reservation to be seperate so it
8674 * doesn't end up using space reserved for updating the inode or
8675 * removing the orphan item. We also need to be able to stop the
8676 * transaction and start a new one, which means we need to be able to
8677 * update the inode several times, and we have no idea of knowing how
8678 * many times that will be, so we can't just reserve 1 item for the
8679 * entirety of the opration, so that has to be done seperately as well.
8680 * Then there is the orphan item, which does indeed need to be held on
8681 * to for the whole operation, and we need nobody to touch this reserved
8682 * space except the orphan code.
8683 *
8684 * So that leaves us with
8685 *
8686 * 1) root->orphan_block_rsv - for the orphan deletion.
8687 * 2) rsv - for the truncate reservation, which we will steal from the
8688 * transaction reservation.
8689 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8690 * updating the inode.
8691 */
66d8f3dd 8692 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
fcb80c2a
JB		 8693 if (!rsv)
8694 return -ENOMEM;
4a338542 8695 rsv->size = min_size;
ca7e70f5 8696 rsv->failfast = 1;
f0cd846e 8697
907cbceb 8698 /*
07127184 8699 * 1 for the truncate slack space
907cbceb
JB		 8700 * 1 for updating the inode.
8701 */
f3fe820c 8702 trans = btrfs_start_transaction(root, 2);
fcb80c2a
JB		 8703 if (IS_ERR(trans)) {
8704 err = PTR_ERR(trans);
8705 goto out;
8706 }
f0cd846e 8707
907cbceb
JB		 8708 /* Migrate the slack space for the truncate to our reserve */
8709 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8710 min_size);
fcb80c2a 8711 BUG_ON(ret);
f0cd846e 8712
5dc562c5
JB		 8713 /*
8714 * So if we truncate and then write and fsync we normally would just
8715 * write the extents that changed, which is a problem if we need to
8716 * first truncate that entire inode. So set this flag so we write out
8717 * all of the extents in the inode to the sync log so we're completely
8718 * safe.
8719 */
8720 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
ca7e70f5 8721 trans->block_rsv = rsv;
907cbceb 8722
8082510e
YZ		 8723 while (1) {
8724 ret = btrfs_truncate_inode_items(trans, root, inode,
8725 inode->i_size,
8726 BTRFS_EXTENT_DATA_KEY);
28ed1345 8727 if (ret != -ENOSPC && ret != -EAGAIN) {
3893e33b 8728 err = ret;
8082510e 8729 break;
3893e33b 8730 }
39279cc3 8731
fcb80c2a 8732 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e 8733 ret = btrfs_update_inode(trans, root, inode);
3893e33b
JB		 8734 if (ret) {
8735 err = ret;
8736 break;
8737 }
ca7e70f5 8738
8082510e 8739 btrfs_end_transaction(trans, root);
b53d3f5d 8740 btrfs_btree_balance_dirty(root);
ca7e70f5
JB		 8741
8742 trans = btrfs_start_transaction(root, 2);
8743 if (IS_ERR(trans)) {
8744 ret = err = PTR_ERR(trans);
8745 trans = NULL;
8746 break;
8747 }
8748
8749 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
8750 rsv, min_size);
8751 BUG_ON(ret); /* shouldn't happen */
8752 trans->block_rsv = rsv;
8082510e
YZ		 8753 }
8754
8755 if (ret == 0 && inode->i_nlink > 0) {
fcb80c2a 8756 trans->block_rsv = root->orphan_block_rsv;
8082510e 8757 ret = btrfs_orphan_del(trans, inode);
3893e33b
JB		 8758 if (ret)
8759 err = ret;
8082510e
YZ		 8760 }
8761
917c16b2
CM		 8762 if (trans) {
8763 trans->block_rsv = &root->fs_info->trans_block_rsv;
8764 ret = btrfs_update_inode(trans, root, inode);
8765 if (ret && !err)
8766 err = ret;
7b128766 8767
7ad85bb7 8768 ret = btrfs_end_transaction(trans, root);
b53d3f5d 8769 btrfs_btree_balance_dirty(root);
917c16b2 8770 }
fcb80c2a
JB		 8771
8772out:
8773 btrfs_free_block_rsv(root, rsv);
8774
3893e33b
JB		 8775 if (ret && !err)
8776 err = ret;
a41ad394 8777
3893e33b 8778 return err;
39279cc3
CM		 8779}
8780
d352ac68
CM		 8781/*
8782 * create a new subvolume directory/inode (helper for the ioctl).
8783 */
d2fb3437 8784int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
63541927
FDBM		 8785 struct btrfs_root *new_root,
8786 struct btrfs_root *parent_root,
8787 u64 new_dirid)
39279cc3 8788{
39279cc3 8789 struct inode *inode;
76dda93c 8790 int err;
00e4e6b3 8791 u64 index = 0;
39279cc3 8792
12fc9d09
FA		 8793 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8794 new_dirid, new_dirid,
8795 S_IFDIR | (~current_umask() & S_IRWXUGO),
8796 &index);
54aa1f4d 8797 if (IS_ERR(inode))
f46b5a66 8798 return PTR_ERR(inode);
39279cc3
CM		 8799 inode->i_op = &btrfs_dir_inode_operations;
8800 inode->i_fop = &btrfs_dir_file_operations;
8801
bfe86848 8802 set_nlink(inode, 1);
dbe674a9 8803 btrfs_i_size_write(inode, 0);
b0d5d10f 8804 unlock_new_inode(inode);
3b96362c 8805
63541927
FDBM		 8806 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8807 if (err)
8808 btrfs_err(new_root->fs_info,
351fd353 8809 "error inheriting subvolume %llu properties: %d",
63541927
FDBM		 8810 new_root->root_key.objectid, err);
8811
76dda93c 8812 err = btrfs_update_inode(trans, new_root, inode);
cb8e7090 8813
76dda93c 8814 iput(inode);
ce598979 8815 return err;
39279cc3
CM		 8816}
8817
39279cc3
CM		 8818struct inode *btrfs_alloc_inode(struct super_block *sb)
8819{
8820 struct btrfs_inode *ei;
2ead6ae7 8821 struct inode *inode;
39279cc3
CM		 8822
8823 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
8824 if (!ei)
8825 return NULL;
2ead6ae7
YZ		 8826
8827 ei->root = NULL;
2ead6ae7 8828 ei->generation = 0;
15ee9bc7 8829 ei->last_trans = 0;
257c62e1 8830 ei->last_sub_trans = 0;
e02119d5 8831 ei->logged_trans = 0;
2ead6ae7 8832 ei->delalloc_bytes = 0;
47059d93 8833 ei->defrag_bytes = 0;
2ead6ae7
YZ		 8834 ei->disk_i_size = 0;
8835 ei->flags = 0;
7709cde3 8836 ei->csum_bytes = 0;
2ead6ae7 8837 ei->index_cnt = (u64)-1;
67de1176 8838 ei->dir_index = 0;
2ead6ae7 8839 ei->last_unlink_trans = 0;
46d8bc34 8840 ei->last_log_commit = 0;
2ead6ae7 8841
9e0baf60
JB		 8842 spin_lock_init(&ei->lock);
8843 ei->outstanding_extents = 0;
8844 ei->reserved_extents = 0;
2ead6ae7 8845
72ac3c0d 8846 ei->runtime_flags = 0;
261507a0 8847 ei->force_compress = BTRFS_COMPRESS_NONE;
2ead6ae7 8848
16cdcec7
MX		 8849 ei->delayed_node = NULL;
8850
9cc97d64 8851 ei->i_otime.tv_sec = 0;
8852 ei->i_otime.tv_nsec = 0;
8853
2ead6ae7 8854 inode = &ei->vfs_inode;
a8067e02 8855 extent_map_tree_init(&ei->extent_tree);
f993c883
DS		 8856 extent_io_tree_init(&ei->io_tree, &inode->i_data);
8857 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
0b32f4bb
JB		 8858 ei->io_tree.track_uptodate = 1;
8859 ei->io_failure_tree.track_uptodate = 1;
b812ce28 8860 atomic_set(&ei->sync_writers, 0);
2ead6ae7 8861 mutex_init(&ei->log_mutex);
f248679e 8862 mutex_init(&ei->delalloc_mutex);
e6dcd2dc 8863 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
2ead6ae7 8864 INIT_LIST_HEAD(&ei->delalloc_inodes);
2ead6ae7
YZ		 8865 RB_CLEAR_NODE(&ei->rb_node);
8866
8867 return inode;
39279cc3
CM		 8868}
8869
aaedb55b
JB		 8870#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8871void btrfs_test_destroy_inode(struct inode *inode)
8872{
8873 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8874 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8875}
8876#endif
8877
fa0d7e3d
NP		 8878static void btrfs_i_callback(struct rcu_head *head)
8879{
8880 struct inode *inode = container_of(head, struct inode, i_rcu);
fa0d7e3d
NP		 8881 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8882}
8883
39279cc3
CM		 8884void btrfs_destroy_inode(struct inode *inode)
8885{
e6dcd2dc 8886 struct btrfs_ordered_extent *ordered;
5a3f23d5
CM		 8887 struct btrfs_root *root = BTRFS_I(inode)->root;
8888
b3d9b7a3 8889 WARN_ON(!hlist_empty(&inode->i_dentry));
39279cc3 8890 WARN_ON(inode->i_data.nrpages);
9e0baf60
JB		 8891 WARN_ON(BTRFS_I(inode)->outstanding_extents);
8892 WARN_ON(BTRFS_I(inode)->reserved_extents);
7709cde3
JB		 8893 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8894 WARN_ON(BTRFS_I(inode)->csum_bytes);
47059d93 8895 WARN_ON(BTRFS_I(inode)->defrag_bytes);
39279cc3 8896
a6dbd429
JB		 8897 /*
8898 * This can happen where we create an inode, but somebody else also
8899 * created the same inode and we need to destroy the one we already
8900 * created.
8901 */
8902 if (!root)
8903 goto free;
8904
8a35d95f
JB		 8905 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8906 &BTRFS_I(inode)->runtime_flags)) {
c2cf52eb 8907 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
c1c9ff7c 8908 btrfs_ino(inode));
8a35d95f 8909 atomic_dec(&root->orphan_inodes);
7b128766 8910 }
7b128766 8911
d397712b 8912 while (1) {
e6dcd2dc
CM		 8913 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8914 if (!ordered)
8915 break;
8916 else {
c2cf52eb 8917 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
c1c9ff7c 8918 ordered->file_offset, ordered->len);
e6dcd2dc
CM		 8919 btrfs_remove_ordered_extent(inode, ordered);
8920 btrfs_put_ordered_extent(ordered);
8921 btrfs_put_ordered_extent(ordered);
8922 }
8923 }
5d4f98a2 8924 inode_tree_del(inode);
5b21f2ed 8925 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
a6dbd429 8926free:
fa0d7e3d 8927 call_rcu(&inode->i_rcu, btrfs_i_callback);
39279cc3
CM		 8928}
8929
45321ac5 8930int btrfs_drop_inode(struct inode *inode)
76dda93c
YZ		 8931{
8932 struct btrfs_root *root = BTRFS_I(inode)->root;
45321ac5 8933
6379ef9f
NA		 8934 if (root == NULL)
8935 return 1;
8936
fa6ac876 8937 /* the snap/subvol tree is on deleting */
69e9c6c6 8938 if (btrfs_root_refs(&root->root_item) == 0)
45321ac5 8939 return 1;
76dda93c 8940 else
45321ac5 8941 return generic_drop_inode(inode);
76dda93c
YZ		 8942}
8943
0ee0fda0 8944static void init_once(void *foo)
39279cc3
CM		 8945{
8946 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8947
8948 inode_init_once(&ei->vfs_inode);
8949}
8950
8951void btrfs_destroy_cachep(void)
8952{
8c0a8537
KS		 8953 /*
8954 * Make sure all delayed rcu free inodes are flushed before we
8955 * destroy cache.
8956 */
8957 rcu_barrier();
39279cc3
CM		 8958 if (btrfs_inode_cachep)
8959 kmem_cache_destroy(btrfs_inode_cachep);
8960 if (btrfs_trans_handle_cachep)
8961 kmem_cache_destroy(btrfs_trans_handle_cachep);
8962 if (btrfs_transaction_cachep)
8963 kmem_cache_destroy(btrfs_transaction_cachep);
39279cc3
CM		 8964 if (btrfs_path_cachep)
8965 kmem_cache_destroy(btrfs_path_cachep);
dc89e982
JB		 8966 if (btrfs_free_space_cachep)
8967 kmem_cache_destroy(btrfs_free_space_cachep);
8ccf6f19
MX		 8968 if (btrfs_delalloc_work_cachep)
8969 kmem_cache_destroy(btrfs_delalloc_work_cachep);
39279cc3
CM		 8970}
8971
8972int btrfs_init_cachep(void)
8973{
837e1972 8974 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9601e3f6
CH		 8975 sizeof(struct btrfs_inode), 0,
8976 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
39279cc3
CM		 8977 if (!btrfs_inode_cachep)
8978 goto fail;
9601e3f6 8979
837e1972 8980 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9601e3f6
CH		 8981 sizeof(struct btrfs_trans_handle), 0,
8982 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 8983 if (!btrfs_trans_handle_cachep)
8984 goto fail;
9601e3f6 8985
837e1972 8986 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9601e3f6
CH		 8987 sizeof(struct btrfs_transaction), 0,
8988 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 8989 if (!btrfs_transaction_cachep)
8990 goto fail;
9601e3f6 8991
837e1972 8992 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9601e3f6
CH		 8993 sizeof(struct btrfs_path), 0,
8994 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 8995 if (!btrfs_path_cachep)
8996 goto fail;
9601e3f6 8997
837e1972 8998 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
dc89e982
JB		 8999 sizeof(struct btrfs_free_space), 0,
9000 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9001 if (!btrfs_free_space_cachep)
9002 goto fail;
9003
8ccf6f19
MX		 9004 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
9005 sizeof(struct btrfs_delalloc_work), 0,
9006 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
9007 NULL);
9008 if (!btrfs_delalloc_work_cachep)
9009 goto fail;
9010
39279cc3
CM		 9011 return 0;
9012fail:
9013 btrfs_destroy_cachep();
9014 return -ENOMEM;
9015}
9016
9017static int btrfs_getattr(struct vfsmount *mnt,
9018 struct dentry *dentry, struct kstat *stat)
9019{
df0af1a5 9020 u64 delalloc_bytes;
39279cc3 9021 struct inode *inode = dentry->d_inode;
fadc0d8b
DS		 9022 u32 blocksize = inode->i_sb->s_blocksize;
9023
39279cc3 9024 generic_fillattr(inode, stat);
0ee5dc67 9025 stat->dev = BTRFS_I(inode)->root->anon_dev;
d6667462 9026 stat->blksize = PAGE_CACHE_SIZE;
df0af1a5
MX		 9027
9028 spin_lock(&BTRFS_I(inode)->lock);
9029 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9030 spin_unlock(&BTRFS_I(inode)->lock);
fadc0d8b 9031 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
df0af1a5 9032 ALIGN(delalloc_bytes, blocksize)) >> 9;
39279cc3
CM		 9033 return 0;
9034}
9035
d397712b
CM		 9036static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9037 struct inode *new_dir, struct dentry *new_dentry)
39279cc3
CM		 9038{
9039 struct btrfs_trans_handle *trans;
9040 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4df27c4d 9041 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
39279cc3
CM		 9042 struct inode *new_inode = new_dentry->d_inode;
9043 struct inode *old_inode = old_dentry->d_inode;
9044 struct timespec ctime = CURRENT_TIME;
00e4e6b3 9045 u64 index = 0;
4df27c4d 9046 u64 root_objectid;
39279cc3 9047 int ret;
33345d01 9048 u64 old_ino = btrfs_ino(old_inode);
39279cc3 9049
33345d01 9050 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
f679a840
YZ		 9051 return -EPERM;
9052
4df27c4d 9053 /* we only allow rename subvolume link between subvolumes */
33345d01 9054 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
3394e160
CM		 9055 return -EXDEV;
9056
33345d01
LZ		 9057 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9058 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
39279cc3 9059 return -ENOTEMPTY;
5f39d397 9060
4df27c4d
YZ		 9061 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9062 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9063 return -ENOTEMPTY;
9c52057c
CM		 9064
9065
9066 /* check for collisions, even if the name isn't there */
4871c158 9067 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9c52057c
CM		 9068 new_dentry->d_name.name,
9069 new_dentry->d_name.len);
9070
9071 if (ret) {
9072 if (ret == -EEXIST) {
9073 /* we shouldn't get
9074 * eexist without a new_inode */
fae7f21c 9075 if (WARN_ON(!new_inode)) {
9c52057c
CM		 9076 return ret;
9077 }
9078 } else {
9079 /* maybe -EOVERFLOW */
9080 return ret;
9081 }
9082 }
9083 ret = 0;
9084
5a3f23d5 9085 /*
8d875f95
CM		 9086 * we're using rename to replace one file with another. Start IO on it
9087 * now so we don't add too much work to the end of the transaction
5a3f23d5 9088 */
8d875f95 9089 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
5a3f23d5
CM		 9090 filemap_flush(old_inode->i_mapping);
9091
76dda93c 9092 /* close the racy window with snapshot create/destroy ioctl */
33345d01 9093 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 9094 down_read(&root->fs_info->subvol_sem);
a22285a6
YZ		 9095 /*
9096 * We want to reserve the absolute worst case amount of items. So if
9097 * both inodes are subvols and we need to unlink them then that would
9098 * require 4 item modifications, but if they are both normal inodes it
9099 * would require 5 item modifications, so we'll assume their normal
9100 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9101 * should cover the worst case number of items we'll modify.
9102 */
6e137ed3 9103 trans = btrfs_start_transaction(root, 11);
b44c59a8
JL		 9104 if (IS_ERR(trans)) {
9105 ret = PTR_ERR(trans);
9106 goto out_notrans;
9107 }
76dda93c 9108
4df27c4d
YZ		 9109 if (dest != root)
9110 btrfs_record_root_in_trans(trans, dest);
5f39d397 9111
a5719521
YZ		 9112 ret = btrfs_set_inode_index(new_dir, &index);
9113 if (ret)
9114 goto out_fail;
5a3f23d5 9115
67de1176 9116 BTRFS_I(old_inode)->dir_index = 0ULL;
33345d01 9117 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d 9118 /* force full log commit if subvolume involved. */
995946dd 9119 btrfs_set_log_full_commit(root->fs_info, trans);
4df27c4d 9120 } else {
a5719521
YZ		 9121 ret = btrfs_insert_inode_ref(trans, dest,
9122 new_dentry->d_name.name,
9123 new_dentry->d_name.len,
33345d01
LZ		 9124 old_ino,
9125 btrfs_ino(new_dir), index);
a5719521
YZ		 9126 if (ret)
9127 goto out_fail;
4df27c4d
YZ		 9128 /*
9129 * this is an ugly little race, but the rename is required
9130 * to make sure that if we crash, the inode is either at the
9131 * old name or the new one. pinning the log transaction lets
9132 * us make sure we don't allow a log commit to come in after
9133 * we unlink the name but before we add the new name back in.
9134 */
9135 btrfs_pin_log_trans(root);
9136 }
5a3f23d5 9137
0c4d2d95
JB		 9138 inode_inc_iversion(old_dir);
9139 inode_inc_iversion(new_dir);
9140 inode_inc_iversion(old_inode);
39279cc3
CM		 9141 old_dir->i_ctime = old_dir->i_mtime = ctime;
9142 new_dir->i_ctime = new_dir->i_mtime = ctime;
9143 old_inode->i_ctime = ctime;
5f39d397 9144
12fcfd22
CM		 9145 if (old_dentry->d_parent != new_dentry->d_parent)
9146 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9147
33345d01 9148 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 9149 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9150 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9151 old_dentry->d_name.name,
9152 old_dentry->d_name.len);
9153 } else {
92986796
AV		 9154 ret = __btrfs_unlink_inode(trans, root, old_dir,
9155 old_dentry->d_inode,
9156 old_dentry->d_name.name,
9157 old_dentry->d_name.len);
9158 if (!ret)
9159 ret = btrfs_update_inode(trans, root, old_inode);
4df27c4d 9160 }
79787eaa
JM		 9161 if (ret) {
9162 btrfs_abort_transaction(trans, root, ret);
9163 goto out_fail;
9164 }
39279cc3
CM		 9165
9166 if (new_inode) {
0c4d2d95 9167 inode_inc_iversion(new_inode);
39279cc3 9168 new_inode->i_ctime = CURRENT_TIME;
33345d01 9169 if (unlikely(btrfs_ino(new_inode) ==
4df27c4d
YZ		 9170 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9171 root_objectid = BTRFS_I(new_inode)->location.objectid;
9172 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9173 root_objectid,
9174 new_dentry->d_name.name,
9175 new_dentry->d_name.len);
9176 BUG_ON(new_inode->i_nlink == 0);
9177 } else {
9178 ret = btrfs_unlink_inode(trans, dest, new_dir,
9179 new_dentry->d_inode,
9180 new_dentry->d_name.name,
9181 new_dentry->d_name.len);
9182 }
4ef31a45 9183 if (!ret && new_inode->i_nlink == 0)
e02119d5 9184 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
79787eaa
JM		 9185 if (ret) {
9186 btrfs_abort_transaction(trans, root, ret);
9187 goto out_fail;
9188 }
39279cc3 9189 }
aec7477b 9190
4df27c4d
YZ		 9191 ret = btrfs_add_link(trans, new_dir, old_inode,
9192 new_dentry->d_name.name,
a5719521 9193 new_dentry->d_name.len, 0, index);
79787eaa
JM		 9194 if (ret) {
9195 btrfs_abort_transaction(trans, root, ret);
9196 goto out_fail;
9197 }
39279cc3 9198
67de1176
MX		 9199 if (old_inode->i_nlink == 1)
9200 BTRFS_I(old_inode)->dir_index = index;
9201
33345d01 9202 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
10d9f309 9203 struct dentry *parent = new_dentry->d_parent;
6a912213 9204 btrfs_log_new_name(trans, old_inode, old_dir, parent);
4df27c4d
YZ		 9205 btrfs_end_log_trans(root);
9206 }
39279cc3 9207out_fail:
7ad85bb7 9208 btrfs_end_transaction(trans, root);
b44c59a8 9209out_notrans:
33345d01 9210 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 9211 up_read(&root->fs_info->subvol_sem);
9ed74f2d 9212
39279cc3
CM		 9213 return ret;
9214}
9215
80ace85c
MS		 9216static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9217 struct inode *new_dir, struct dentry *new_dentry,
9218 unsigned int flags)
9219{
9220 if (flags & ~RENAME_NOREPLACE)
9221 return -EINVAL;
9222
9223 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
9224}
9225
8ccf6f19
MX		 9226static void btrfs_run_delalloc_work(struct btrfs_work *work)
9227{
9228 struct btrfs_delalloc_work *delalloc_work;
9f23e289 9229 struct inode *inode;
8ccf6f19
MX		 9230
9231 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9232 work);
9f23e289
JB		 9233 inode = delalloc_work->inode;
9234 if (delalloc_work->wait) {
9235 btrfs_wait_ordered_range(inode, 0, (u64)-1);
9236 } else {
9237 filemap_flush(inode->i_mapping);
9238 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9239 &BTRFS_I(inode)->runtime_flags))
9240 filemap_flush(inode->i_mapping);
9241 }
8ccf6f19
MX		 9242
9243 if (delalloc_work->delay_iput)
9f23e289 9244 btrfs_add_delayed_iput(inode);
8ccf6f19 9245 else
9f23e289 9246 iput(inode);
8ccf6f19
MX		 9247 complete(&delalloc_work->completion);
9248}
9249
9250struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9251 int wait, int delay_iput)
9252{
9253 struct btrfs_delalloc_work *work;
9254
9255 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
9256 if (!work)
9257 return NULL;
9258
9259 init_completion(&work->completion);
9260 INIT_LIST_HEAD(&work->list);
9261 work->inode = inode;
9262 work->wait = wait;
9263 work->delay_iput = delay_iput;
9e0af237
LB		 9264 WARN_ON_ONCE(!inode);
9265 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9266 btrfs_run_delalloc_work, NULL, NULL);
8ccf6f19
MX		 9267
9268 return work;
9269}
9270
9271void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9272{
9273 wait_for_completion(&work->completion);
9274 kmem_cache_free(btrfs_delalloc_work_cachep, work);
9275}
9276
d352ac68
CM		 9277/*
9278 * some fairly slow code that needs optimization. This walks the list
9279 * of all the inodes with pending delalloc and forces them to disk.
9280 */
6c255e67
MX		 9281static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9282 int nr)
ea8c2819 9283{
ea8c2819 9284 struct btrfs_inode *binode;
5b21f2ed 9285 struct inode *inode;
8ccf6f19
MX		 9286 struct btrfs_delalloc_work *work, *next;
9287 struct list_head works;
1eafa6c7 9288 struct list_head splice;
8ccf6f19 9289 int ret = 0;
ea8c2819 9290
8ccf6f19 9291 INIT_LIST_HEAD(&works);
1eafa6c7 9292 INIT_LIST_HEAD(&splice);
63607cc8 9293
573bfb72 9294 mutex_lock(&root->delalloc_mutex);
eb73c1b7
MX		 9295 spin_lock(&root->delalloc_lock);
9296 list_splice_init(&root->delalloc_inodes, &splice);
1eafa6c7
MX		 9297 while (!list_empty(&splice)) {
9298 binode = list_entry(splice.next, struct btrfs_inode,
ea8c2819 9299 delalloc_inodes);
1eafa6c7 9300
eb73c1b7
MX		 9301 list_move_tail(&binode->delalloc_inodes,
9302 &root->delalloc_inodes);
5b21f2ed 9303 inode = igrab(&binode->vfs_inode);
df0af1a5 9304 if (!inode) {
eb73c1b7 9305 cond_resched_lock(&root->delalloc_lock);
1eafa6c7 9306 continue;
df0af1a5 9307 }
eb73c1b7 9308 spin_unlock(&root->delalloc_lock);
1eafa6c7
MX		 9309
9310 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
5d99a998 9311 if (!work) {
f4ab9ea7
JB		 9312 if (delay_iput)
9313 btrfs_add_delayed_iput(inode);
9314 else
9315 iput(inode);
1eafa6c7 9316 ret = -ENOMEM;
a1ecaabb 9317 goto out;
5b21f2ed 9318 }
1eafa6c7 9319 list_add_tail(&work->list, &works);
a44903ab
QW		 9320 btrfs_queue_work(root->fs_info->flush_workers,
9321 &work->work);
6c255e67
MX		 9322 ret++;
9323 if (nr != -1 && ret >= nr)
a1ecaabb 9324 goto out;
5b21f2ed 9325 cond_resched();
eb73c1b7 9326 spin_lock(&root->delalloc_lock);
ea8c2819 9327 }
eb73c1b7 9328 spin_unlock(&root->delalloc_lock);
8c8bee1d 9329
a1ecaabb 9330out:
eb73c1b7
MX		 9331 list_for_each_entry_safe(work, next, &works, list) {
9332 list_del_init(&work->list);
9333 btrfs_wait_and_free_delalloc_work(work);
9334 }
9335
9336 if (!list_empty_careful(&splice)) {
9337 spin_lock(&root->delalloc_lock);
9338 list_splice_tail(&splice, &root->delalloc_inodes);
9339 spin_unlock(&root->delalloc_lock);
9340 }
573bfb72 9341 mutex_unlock(&root->delalloc_mutex);
eb73c1b7
MX		 9342 return ret;
9343}
1eafa6c7 9344
eb73c1b7
MX		 9345int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9346{
9347 int ret;
1eafa6c7 9348
2c21b4d7 9349 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
eb73c1b7
MX		 9350 return -EROFS;
9351
6c255e67
MX		 9352 ret = __start_delalloc_inodes(root, delay_iput, -1);
9353 if (ret > 0)
9354 ret = 0;
eb73c1b7
MX		 9355 /*
9356 * the filemap_flush will queue IO into the worker threads, but
8c8bee1d
CM		 9357 * we have to make sure the IO is actually started and that
9358 * ordered extents get created before we return
9359 */
9360 atomic_inc(&root->fs_info->async_submit_draining);
d397712b 9361 while (atomic_read(&root->fs_info->nr_async_submits) ||
771ed689 9362 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 9363 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 9364 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9365 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 9366 }
9367 atomic_dec(&root->fs_info->async_submit_draining);
eb73c1b7
MX		 9368 return ret;
9369}
9370
6c255e67
MX		 9371int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9372 int nr)
eb73c1b7
MX		 9373{
9374 struct btrfs_root *root;
9375 struct list_head splice;
9376 int ret;
9377
2c21b4d7 9378 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
eb73c1b7
MX		 9379 return -EROFS;
9380
9381 INIT_LIST_HEAD(&splice);
9382
573bfb72 9383 mutex_lock(&fs_info->delalloc_root_mutex);
eb73c1b7
MX		 9384 spin_lock(&fs_info->delalloc_root_lock);
9385 list_splice_init(&fs_info->delalloc_roots, &splice);
6c255e67 9386 while (!list_empty(&splice) && nr) {
eb73c1b7
MX		 9387 root = list_first_entry(&splice, struct btrfs_root,
9388 delalloc_root);
9389 root = btrfs_grab_fs_root(root);
9390 BUG_ON(!root);
9391 list_move_tail(&root->delalloc_root,
9392 &fs_info->delalloc_roots);
9393 spin_unlock(&fs_info->delalloc_root_lock);
9394
6c255e67 9395 ret = __start_delalloc_inodes(root, delay_iput, nr);
eb73c1b7 9396 btrfs_put_fs_root(root);
6c255e67 9397 if (ret < 0)
eb73c1b7
MX		 9398 goto out;
9399
6c255e67
MX		 9400 if (nr != -1) {
9401 nr -= ret;
9402 WARN_ON(nr < 0);
9403 }
eb73c1b7 9404 spin_lock(&fs_info->delalloc_root_lock);
8ccf6f19 9405 }
eb73c1b7 9406 spin_unlock(&fs_info->delalloc_root_lock);
1eafa6c7 9407
6c255e67 9408 ret = 0;
eb73c1b7
MX		 9409 atomic_inc(&fs_info->async_submit_draining);
9410 while (atomic_read(&fs_info->nr_async_submits) ||
9411 atomic_read(&fs_info->async_delalloc_pages)) {
9412 wait_event(fs_info->async_submit_wait,
9413 (atomic_read(&fs_info->nr_async_submits) == 0 &&
9414 atomic_read(&fs_info->async_delalloc_pages) == 0));
9415 }
9416 atomic_dec(&fs_info->async_submit_draining);
eb73c1b7 9417out:
1eafa6c7 9418 if (!list_empty_careful(&splice)) {
eb73c1b7
MX		 9419 spin_lock(&fs_info->delalloc_root_lock);
9420 list_splice_tail(&splice, &fs_info->delalloc_roots);
9421 spin_unlock(&fs_info->delalloc_root_lock);
1eafa6c7 9422 }
573bfb72 9423 mutex_unlock(&fs_info->delalloc_root_mutex);
8ccf6f19 9424 return ret;
ea8c2819
CM		 9425}
9426
39279cc3
CM		 9427static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9428 const char *symname)
9429{
9430 struct btrfs_trans_handle *trans;
9431 struct btrfs_root *root = BTRFS_I(dir)->root;
9432 struct btrfs_path *path;
9433 struct btrfs_key key;
1832a6d5 9434 struct inode *inode = NULL;
39279cc3
CM		 9435 int err;
9436 int drop_inode = 0;
9437 u64 objectid;
67871254 9438 u64 index = 0;
39279cc3
CM		 9439 int name_len;
9440 int datasize;
5f39d397 9441 unsigned long ptr;
39279cc3 9442 struct btrfs_file_extent_item *ei;
5f39d397 9443 struct extent_buffer *leaf;
39279cc3 9444
f06becc4 9445 name_len = strlen(symname);
39279cc3
CM		 9446 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9447 return -ENAMETOOLONG;
1832a6d5 9448
9ed74f2d
JB		 9449 /*
9450 * 2 items for inode item and ref
9451 * 2 items for dir items
9452 * 1 item for xattr if selinux is on
9453 */
a22285a6
YZ		 9454 trans = btrfs_start_transaction(root, 5);
9455 if (IS_ERR(trans))
9456 return PTR_ERR(trans);
1832a6d5 9457
581bb050
LZ		 9458 err = btrfs_find_free_ino(root, &objectid);
9459 if (err)
9460 goto out_unlock;
9461
aec7477b 9462 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 9463 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 9464 S_IFLNK|S_IRWXUGO, &index);
7cf96da3
TI		 9465 if (IS_ERR(inode)) {
9466 err = PTR_ERR(inode);
39279cc3 9467 goto out_unlock;
7cf96da3 9468 }
39279cc3 9469
ad19db71
CS		 9470 /*
9471 * If the active LSM wants to access the inode during
9472 * d_instantiate it needs these. Smack checks to see
9473 * if the filesystem supports xattrs by looking at the
9474 * ops vector.
9475 */
9476 inode->i_fop = &btrfs_file_operations;
9477 inode->i_op = &btrfs_file_inode_operations;
b0d5d10f 9478 inode->i_mapping->a_ops = &btrfs_aops;
b0d5d10f
CM		 9479 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9480
9481 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9482 if (err)
9483 goto out_unlock_inode;
ad19db71 9484
a1b075d2 9485 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3 9486 if (err)
b0d5d10f 9487 goto out_unlock_inode;
39279cc3
CM		 9488
9489 path = btrfs_alloc_path();
d8926bb3
MF		 9490 if (!path) {
9491 err = -ENOMEM;
b0d5d10f 9492 goto out_unlock_inode;
d8926bb3 9493 }
33345d01 9494 key.objectid = btrfs_ino(inode);
39279cc3 9495 key.offset = 0;
962a298f 9496 key.type = BTRFS_EXTENT_DATA_KEY;
39279cc3
CM		 9497 datasize = btrfs_file_extent_calc_inline_size(name_len);
9498 err = btrfs_insert_empty_item(trans, root, path, &key,
9499 datasize);
54aa1f4d 9500 if (err) {
b0839166 9501 btrfs_free_path(path);
b0d5d10f 9502 goto out_unlock_inode;
54aa1f4d 9503 }
5f39d397
CM		 9504 leaf = path->nodes[0];
9505 ei = btrfs_item_ptr(leaf, path->slots[0],
9506 struct btrfs_file_extent_item);
9507 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9508 btrfs_set_file_extent_type(leaf, ei,
39279cc3 9509 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 9510 btrfs_set_file_extent_encryption(leaf, ei, 0);
9511 btrfs_set_file_extent_compression(leaf, ei, 0);
9512 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9513 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9514
39279cc3 9515 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 9516 write_extent_buffer(leaf, symname, ptr, name_len);
9517 btrfs_mark_buffer_dirty(leaf);
39279cc3 9518 btrfs_free_path(path);
5f39d397 9519
39279cc3
CM		 9520 inode->i_op = &btrfs_symlink_inode_operations;
9521 inode->i_mapping->a_ops = &btrfs_symlink_aops;
d899e052 9522 inode_set_bytes(inode, name_len);
f06becc4 9523 btrfs_i_size_write(inode, name_len);
54aa1f4d 9524 err = btrfs_update_inode(trans, root, inode);
b0d5d10f 9525 if (err) {
54aa1f4d 9526 drop_inode = 1;
b0d5d10f
CM		 9527 goto out_unlock_inode;
9528 }
9529
9530 unlock_new_inode(inode);
9531 d_instantiate(dentry, inode);
39279cc3
CM		 9532
9533out_unlock:
7ad85bb7 9534 btrfs_end_transaction(trans, root);
39279cc3
CM		 9535 if (drop_inode) {
9536 inode_dec_link_count(inode);
9537 iput(inode);
9538 }
b53d3f5d 9539 btrfs_btree_balance_dirty(root);
39279cc3 9540 return err;
b0d5d10f
CM		 9541
9542out_unlock_inode:
9543 drop_inode = 1;
9544 unlock_new_inode(inode);
9545 goto out_unlock;
39279cc3 9546}
16432985 9547
0af3d00b
JB		 9548static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9549 u64 start, u64 num_bytes, u64 min_size,
9550 loff_t actual_len, u64 *alloc_hint,
9551 struct btrfs_trans_handle *trans)
d899e052 9552{
5dc562c5
JB		 9553 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9554 struct extent_map *em;
d899e052
YZ		 9555 struct btrfs_root *root = BTRFS_I(inode)->root;
9556 struct btrfs_key ins;
d899e052 9557 u64 cur_offset = start;
55a61d1d 9558 u64 i_size;
154ea289 9559 u64 cur_bytes;
d899e052 9560 int ret = 0;
0af3d00b 9561 bool own_trans = true;
d899e052 9562
0af3d00b
JB		 9563 if (trans)
9564 own_trans = false;
d899e052 9565 while (num_bytes > 0) {
0af3d00b
JB		 9566 if (own_trans) {
9567 trans = btrfs_start_transaction(root, 3);
9568 if (IS_ERR(trans)) {
9569 ret = PTR_ERR(trans);
9570 break;
9571 }
5a303d5d
YZ		 9572 }
9573
154ea289
CM		 9574 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
9575 cur_bytes = max(cur_bytes, min_size);
00361589 9576 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
e570fd27 9577 *alloc_hint, &ins, 1, 0);
5a303d5d 9578 if (ret) {
0af3d00b
JB		 9579 if (own_trans)
9580 btrfs_end_transaction(trans, root);
a22285a6 9581 break;
d899e052 9582 }
5a303d5d 9583
d899e052
YZ		 9584 ret = insert_reserved_file_extent(trans, inode,
9585 cur_offset, ins.objectid,
9586 ins.offset, ins.offset,
920bbbfb 9587 ins.offset, 0, 0, 0,
d899e052 9588 BTRFS_FILE_EXTENT_PREALLOC);
79787eaa 9589 if (ret) {
857cc2fc 9590 btrfs_free_reserved_extent(root, ins.objectid,
e570fd27 9591 ins.offset, 0);
79787eaa
JM		 9592 btrfs_abort_transaction(trans, root, ret);
9593 if (own_trans)
9594 btrfs_end_transaction(trans, root);
9595 break;
9596 }
31193213
DY		 9597
9598 btrfs_qgroup_update_reserved_bytes(root->fs_info,
9599 root->root_key.objectid,
9600 ins.offset, 1);
9601
a1ed835e
CM		 9602 btrfs_drop_extent_cache(inode, cur_offset,
9603 cur_offset + ins.offset -1, 0);
5a303d5d 9604
5dc562c5
JB		 9605 em = alloc_extent_map();
9606 if (!em) {
9607 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9608 &BTRFS_I(inode)->runtime_flags);
9609 goto next;
9610 }
9611
9612 em->start = cur_offset;
9613 em->orig_start = cur_offset;
9614 em->len = ins.offset;
9615 em->block_start = ins.objectid;
9616 em->block_len = ins.offset;
b4939680 9617 em->orig_block_len = ins.offset;
cc95bef6 9618 em->ram_bytes = ins.offset;
5dc562c5
JB		 9619 em->bdev = root->fs_info->fs_devices->latest_bdev;
9620 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9621 em->generation = trans->transid;
9622
9623 while (1) {
9624 write_lock(&em_tree->lock);
09a2a8f9 9625 ret = add_extent_mapping(em_tree, em, 1);
5dc562c5
JB		 9626 write_unlock(&em_tree->lock);
9627 if (ret != -EEXIST)
9628 break;
9629 btrfs_drop_extent_cache(inode, cur_offset,
9630 cur_offset + ins.offset - 1,
9631 0);
9632 }
9633 free_extent_map(em);
9634next:
d899e052
YZ		 9635 num_bytes -= ins.offset;
9636 cur_offset += ins.offset;
efa56464 9637 *alloc_hint = ins.objectid + ins.offset;
5a303d5d 9638
0c4d2d95 9639 inode_inc_iversion(inode);
d899e052 9640 inode->i_ctime = CURRENT_TIME;
6cbff00f 9641 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
d899e052 9642 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
efa56464
YZ		 9643 (actual_len > inode->i_size) &&
9644 (cur_offset > inode->i_size)) {
d1ea6a61 9645 if (cur_offset > actual_len)
55a61d1d 9646 i_size = actual_len;
d1ea6a61 9647 else
55a61d1d
JB		 9648 i_size = cur_offset;
9649 i_size_write(inode, i_size);
9650 btrfs_ordered_update_i_size(inode, i_size, NULL);
5a303d5d
YZ		 9651 }
9652
d899e052 9653 ret = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 9654
9655 if (ret) {
9656 btrfs_abort_transaction(trans, root, ret);
9657 if (own_trans)
9658 btrfs_end_transaction(trans, root);
9659 break;
9660 }
d899e052 9661
0af3d00b
JB		 9662 if (own_trans)
9663 btrfs_end_transaction(trans, root);
5a303d5d 9664 }
d899e052
YZ		 9665 return ret;
9666}
9667
0af3d00b
JB		 9668int btrfs_prealloc_file_range(struct inode *inode, int mode,
9669 u64 start, u64 num_bytes, u64 min_size,
9670 loff_t actual_len, u64 *alloc_hint)
9671{
9672 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9673 min_size, actual_len, alloc_hint,
9674 NULL);
9675}
9676
9677int btrfs_prealloc_file_range_trans(struct inode *inode,
9678 struct btrfs_trans_handle *trans, int mode,
9679 u64 start, u64 num_bytes, u64 min_size,
9680 loff_t actual_len, u64 *alloc_hint)
9681{
9682 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9683 min_size, actual_len, alloc_hint, trans);
9684}
9685
e6dcd2dc
CM		 9686static int btrfs_set_page_dirty(struct page *page)
9687{
e6dcd2dc
CM		 9688 return __set_page_dirty_nobuffers(page);
9689}
9690
10556cb2 9691static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd 9692{
b83cc969 9693 struct btrfs_root *root = BTRFS_I(inode)->root;
cb6db4e5 9694 umode_t mode = inode->i_mode;
b83cc969 9695
cb6db4e5
JM		 9696 if (mask & MAY_WRITE &&
9697 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9698 if (btrfs_root_readonly(root))
9699 return -EROFS;
9700 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9701 return -EACCES;
9702 }
2830ba7f 9703 return generic_permission(inode, mask);
fdebe2bd 9704}
39279cc3 9705
ef3b9af5
FM		 9706static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9707{
9708 struct btrfs_trans_handle *trans;
9709 struct btrfs_root *root = BTRFS_I(dir)->root;
9710 struct inode *inode = NULL;
9711 u64 objectid;
9712 u64 index;
9713 int ret = 0;
9714
9715 /*
9716 * 5 units required for adding orphan entry
9717 */
9718 trans = btrfs_start_transaction(root, 5);
9719 if (IS_ERR(trans))
9720 return PTR_ERR(trans);
9721
9722 ret = btrfs_find_free_ino(root, &objectid);
9723 if (ret)
9724 goto out;
9725
9726 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9727 btrfs_ino(dir), objectid, mode, &index);
9728 if (IS_ERR(inode)) {
9729 ret = PTR_ERR(inode);
9730 inode = NULL;
9731 goto out;
9732 }
9733
ef3b9af5
FM		 9734 inode->i_fop = &btrfs_file_operations;
9735 inode->i_op = &btrfs_file_inode_operations;
9736
9737 inode->i_mapping->a_ops = &btrfs_aops;
ef3b9af5
FM		 9738 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9739
b0d5d10f
CM		 9740 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9741 if (ret)
9742 goto out_inode;
9743
9744 ret = btrfs_update_inode(trans, root, inode);
9745 if (ret)
9746 goto out_inode;
ef3b9af5
FM		 9747 ret = btrfs_orphan_add(trans, inode);
9748 if (ret)
b0d5d10f 9749 goto out_inode;
ef3b9af5 9750
5762b5c9
FM		 9751 /*
9752 * We set number of links to 0 in btrfs_new_inode(), and here we set
9753 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9754 * through:
9755 *
9756 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9757 */
9758 set_nlink(inode, 1);
b0d5d10f 9759 unlock_new_inode(inode);
ef3b9af5
FM		 9760 d_tmpfile(dentry, inode);
9761 mark_inode_dirty(inode);
9762
9763out:
9764 btrfs_end_transaction(trans, root);
9765 if (ret)
9766 iput(inode);
9767 btrfs_balance_delayed_items(root);
9768 btrfs_btree_balance_dirty(root);
ef3b9af5 9769 return ret;
b0d5d10f
CM		 9770
9771out_inode:
9772 unlock_new_inode(inode);
9773 goto out;
9774
ef3b9af5
FM		 9775}
9776
b38ef71c
FM		 9777/* Inspired by filemap_check_errors() */
9778int btrfs_inode_check_errors(struct inode *inode)
9779{
9780 int ret = 0;
9781
9782 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
9783 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
9784 ret = -ENOSPC;
9785 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
9786 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
9787 ret = -EIO;
9788
9789 return ret;
9790}
9791
6e1d5dcc 9792static const struct inode_operations btrfs_dir_inode_operations = {
3394e160 9793 .getattr = btrfs_getattr,
39279cc3
CM		 9794 .lookup = btrfs_lookup,
9795 .create = btrfs_create,
9796 .unlink = btrfs_unlink,
9797 .link = btrfs_link,
9798 .mkdir = btrfs_mkdir,
9799 .rmdir = btrfs_rmdir,
80ace85c 9800 .rename2 = btrfs_rename2,
39279cc3
CM		 9801 .symlink = btrfs_symlink,
9802 .setattr = btrfs_setattr,
618e21d5 9803 .mknod = btrfs_mknod,
95819c05
CH		 9804 .setxattr = btrfs_setxattr,
9805 .getxattr = btrfs_getxattr,
5103e947 9806 .listxattr = btrfs_listxattr,
95819c05 9807 .removexattr = btrfs_removexattr,
fdebe2bd 9808 .permission = btrfs_permission,
4e34e719 9809 .get_acl = btrfs_get_acl,
996a710d 9810 .set_acl = btrfs_set_acl,
93fd63c2 9811 .update_time = btrfs_update_time,
ef3b9af5 9812 .tmpfile = btrfs_tmpfile,
39279cc3 9813};
6e1d5dcc 9814static const struct inode_operations btrfs_dir_ro_inode_operations = {
39279cc3 9815 .lookup = btrfs_lookup,
fdebe2bd 9816 .permission = btrfs_permission,
4e34e719 9817 .get_acl = btrfs_get_acl,
996a710d 9818 .set_acl = btrfs_set_acl,
93fd63c2 9819 .update_time = btrfs_update_time,
39279cc3 9820};
76dda93c 9821
828c0950 9822static const struct file_operations btrfs_dir_file_operations = {
39279cc3
CM		 9823 .llseek = generic_file_llseek,
9824 .read = generic_read_dir,
9cdda8d3 9825 .iterate = btrfs_real_readdir,
34287aa3 9826 .unlocked_ioctl = btrfs_ioctl,
39279cc3 9827#ifdef CONFIG_COMPAT
34287aa3 9828 .compat_ioctl = btrfs_ioctl,
39279cc3 9829#endif
6bf13c0c 9830 .release = btrfs_release_file,
e02119d5 9831 .fsync = btrfs_sync_file,
39279cc3
CM		 9832};
9833
d1310b2e 9834static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 9835 .fill_delalloc = run_delalloc_range,
065631f6 9836 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 9837 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 9838 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 9839 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 9840 .writepage_start_hook = btrfs_writepage_start_hook,
b0c68f8b
CM		 9841 .set_bit_hook = btrfs_set_bit_hook,
9842 .clear_bit_hook = btrfs_clear_bit_hook,
9ed74f2d
JB		 9843 .merge_extent_hook = btrfs_merge_extent_hook,
9844 .split_extent_hook = btrfs_split_extent_hook,
07157aac
CM		 9845};
9846
35054394
CM		 9847/*
9848 * btrfs doesn't support the bmap operation because swapfiles
9849 * use bmap to make a mapping of extents in the file. They assume
9850 * these extents won't change over the life of the file and they
9851 * use the bmap result to do IO directly to the drive.
9852 *
9853 * the btrfs bmap call would return logical addresses that aren't
9854 * suitable for IO and they also will change frequently as COW
9855 * operations happen. So, swapfile + btrfs == corruption.
9856 *
9857 * For now we're avoiding this by dropping bmap.
9858 */
7f09410b 9859static const struct address_space_operations btrfs_aops = {
39279cc3
CM		 9860 .readpage = btrfs_readpage,
9861 .writepage = btrfs_writepage,
b293f02e 9862 .writepages = btrfs_writepages,
3ab2fb5a 9863 .readpages = btrfs_readpages,
16432985 9864 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 9865 .invalidatepage = btrfs_invalidatepage,
9866 .releasepage = btrfs_releasepage,
e6dcd2dc 9867 .set_page_dirty = btrfs_set_page_dirty,
465fdd97 9868 .error_remove_page = generic_error_remove_page,
39279cc3
CM		 9869};
9870
7f09410b 9871static const struct address_space_operations btrfs_symlink_aops = {
39279cc3
CM		 9872 .readpage = btrfs_readpage,
9873 .writepage = btrfs_writepage,
2bf5a725
CM		 9874 .invalidatepage = btrfs_invalidatepage,
9875 .releasepage = btrfs_releasepage,
39279cc3
CM		 9876};
9877
6e1d5dcc 9878static const struct inode_operations btrfs_file_inode_operations = {
39279cc3
CM		 9879 .getattr = btrfs_getattr,
9880 .setattr = btrfs_setattr,
95819c05
CH		 9881 .setxattr = btrfs_setxattr,
9882 .getxattr = btrfs_getxattr,
5103e947 9883 .listxattr = btrfs_listxattr,
95819c05 9884 .removexattr = btrfs_removexattr,
fdebe2bd 9885 .permission = btrfs_permission,
1506fcc8 9886 .fiemap = btrfs_fiemap,
4e34e719 9887 .get_acl = btrfs_get_acl,
996a710d 9888 .set_acl = btrfs_set_acl,
e41f941a 9889 .update_time = btrfs_update_time,
39279cc3 9890};
6e1d5dcc 9891static const struct inode_operations btrfs_special_inode_operations = {
618e21d5
JB		 9892 .getattr = btrfs_getattr,
9893 .setattr = btrfs_setattr,
fdebe2bd 9894 .permission = btrfs_permission,
95819c05
CH		 9895 .setxattr = btrfs_setxattr,
9896 .getxattr = btrfs_getxattr,
33268eaf 9897 .listxattr = btrfs_listxattr,
95819c05 9898 .removexattr = btrfs_removexattr,
4e34e719 9899 .get_acl = btrfs_get_acl,
996a710d 9900 .set_acl = btrfs_set_acl,
e41f941a 9901 .update_time = btrfs_update_time,
618e21d5 9902};
6e1d5dcc 9903static const struct inode_operations btrfs_symlink_inode_operations = {
39279cc3
CM		 9904 .readlink = generic_readlink,
9905 .follow_link = page_follow_link_light,
9906 .put_link = page_put_link,
f209561a 9907 .getattr = btrfs_getattr,
22c44fe6 9908 .setattr = btrfs_setattr,
fdebe2bd 9909 .permission = btrfs_permission,
0279b4cd
JO		 9910 .setxattr = btrfs_setxattr,
9911 .getxattr = btrfs_getxattr,
9912 .listxattr = btrfs_listxattr,
9913 .removexattr = btrfs_removexattr,
e41f941a 9914 .update_time = btrfs_update_time,
39279cc3 9915};
76dda93c 9916
82d339d9 9917const struct dentry_operations btrfs_dentry_operations = {
76dda93c 9918 .d_delete = btrfs_dentry_delete,
b4aff1f8 9919 .d_release = btrfs_dentry_release,
76dda93c 9920};
Linux kernel - Deliverables
This page took 2.154343 seconds and 5 git commands to generate.