projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: merge pending IO for tree log write back
[linux-2.6-block.git] / fs / btrfs / inode.c
CommitLineData
6cbd5570
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
8f18cf13 19#include <linux/kernel.h>
065631f6 20#include <linux/bio.h>
39279cc3 21#include <linux/buffer_head.h>
f2eb0a24 22#include <linux/file.h>
39279cc3
CM		 23#include <linux/fs.h>
24#include <linux/pagemap.h>
25#include <linux/highmem.h>
26#include <linux/time.h>
27#include <linux/init.h>
28#include <linux/string.h>
39279cc3
CM		 29#include <linux/backing-dev.h>
30#include <linux/mpage.h>
31#include <linux/swap.h>
32#include <linux/writeback.h>
33#include <linux/statfs.h>
34#include <linux/compat.h>
9ebefb18 35#include <linux/bit_spinlock.h>
5103e947 36#include <linux/xattr.h>
33268eaf 37#include <linux/posix_acl.h>
d899e052 38#include <linux/falloc.h>
5a0e3ad6 39#include <linux/slab.h>
7a36ddec 40#include <linux/ratelimit.h>
22c44fe6 41#include <linux/mount.h>
55e301fd 42#include <linux/btrfs.h>
53b381b3 43#include <linux/blkdev.h>
4b4e25f2 44#include "compat.h"
39279cc3
CM		 45#include "ctree.h"
46#include "disk-io.h"
47#include "transaction.h"
48#include "btrfs_inode.h"
39279cc3 49#include "print-tree.h"
e6dcd2dc 50#include "ordered-data.h"
95819c05 51#include "xattr.h"
e02119d5 52#include "tree-log.h"
4a54c8c1 53#include "volumes.h"
c8b97818 54#include "compression.h"
b4ce94de 55#include "locking.h"
dc89e982 56#include "free-space-cache.h"
581bb050 57#include "inode-map.h"
38c227d8 58#include "backref.h"
39279cc3
CM		 59
60struct btrfs_iget_args {
61 u64 ino;
62 struct btrfs_root *root;
63};
64
6e1d5dcc
AD		 65static const struct inode_operations btrfs_dir_inode_operations;
66static const struct inode_operations btrfs_symlink_inode_operations;
67static const struct inode_operations btrfs_dir_ro_inode_operations;
68static const struct inode_operations btrfs_special_inode_operations;
69static const struct inode_operations btrfs_file_inode_operations;
7f09410b
AD		 70static const struct address_space_operations btrfs_aops;
71static const struct address_space_operations btrfs_symlink_aops;
828c0950 72static const struct file_operations btrfs_dir_file_operations;
d1310b2e 73static struct extent_io_ops btrfs_extent_io_ops;
39279cc3
CM		 74
75static struct kmem_cache *btrfs_inode_cachep;
8ccf6f19 76static struct kmem_cache *btrfs_delalloc_work_cachep;
39279cc3
CM		 77struct kmem_cache *btrfs_trans_handle_cachep;
78struct kmem_cache *btrfs_transaction_cachep;
39279cc3 79struct kmem_cache *btrfs_path_cachep;
dc89e982 80struct kmem_cache *btrfs_free_space_cachep;
39279cc3
CM		 81
82#define S_SHIFT 12
83static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
84 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
85 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
86 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
87 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
88 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
89 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
90 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
91};
92
3972f260 93static int btrfs_setsize(struct inode *inode, struct iattr *attr);
a41ad394 94static int btrfs_truncate(struct inode *inode);
5fd02043 95static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
771ed689
CM		 96static noinline int cow_file_range(struct inode *inode,
97 struct page *locked_page,
98 u64 start, u64 end, int *page_started,
99 unsigned long *nr_written, int unlock);
70c8a91c
JB		 100static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
101 u64 len, u64 orig_start,
102 u64 block_start, u64 block_len,
cc95bef6
JB		 103 u64 orig_block_len, u64 ram_bytes,
104 int type);
7b128766 105
48a3b636
ES		 106static int btrfs_dirty_inode(struct inode *inode);
107
f34f57a3 108static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
2a7dba39
EP		 109 struct inode *inode, struct inode *dir,
110 const struct qstr *qstr)
0279b4cd
JO		 111{
112 int err;
113
f34f57a3 114 err = btrfs_init_acl(trans, inode, dir);
0279b4cd 115 if (!err)
2a7dba39 116 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
0279b4cd
JO		 117 return err;
118}
119
c8b97818
CM		 120/*
121 * this does all the hard work for inserting an inline extent into
122 * the btree. The caller should have done a btrfs_drop_extents so that
123 * no overlapping inline items exist in the btree
124 */
d397712b 125static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
c8b97818
CM		 126 struct btrfs_root *root, struct inode *inode,
127 u64 start, size_t size, size_t compressed_size,
fe3f566c 128 int compress_type,
c8b97818
CM		 129 struct page **compressed_pages)
130{
131 struct btrfs_key key;
132 struct btrfs_path *path;
133 struct extent_buffer *leaf;
134 struct page *page = NULL;
135 char *kaddr;
136 unsigned long ptr;
137 struct btrfs_file_extent_item *ei;
138 int err = 0;
139 int ret;
140 size_t cur_size = size;
141 size_t datasize;
142 unsigned long offset;
c8b97818 143
fe3f566c 144 if (compressed_size && compressed_pages)
c8b97818 145 cur_size = compressed_size;
c8b97818 146
d397712b
CM		 147 path = btrfs_alloc_path();
148 if (!path)
c8b97818
CM		 149 return -ENOMEM;
150
b9473439 151 path->leave_spinning = 1;
c8b97818 152
33345d01 153 key.objectid = btrfs_ino(inode);
c8b97818
CM		 154 key.offset = start;
155 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
c8b97818
CM		 156 datasize = btrfs_file_extent_calc_inline_size(cur_size);
157
158 inode_add_bytes(inode, size);
159 ret = btrfs_insert_empty_item(trans, root, path, &key,
160 datasize);
c8b97818
CM		 161 if (ret) {
162 err = ret;
c8b97818
CM		 163 goto fail;
164 }
165 leaf = path->nodes[0];
166 ei = btrfs_item_ptr(leaf, path->slots[0],
167 struct btrfs_file_extent_item);
168 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
169 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
170 btrfs_set_file_extent_encryption(leaf, ei, 0);
171 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
172 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
173 ptr = btrfs_file_extent_inline_start(ei);
174
261507a0 175 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818
CM		 176 struct page *cpage;
177 int i = 0;
d397712b 178 while (compressed_size > 0) {
c8b97818 179 cpage = compressed_pages[i];
5b050f04 180 cur_size = min_t(unsigned long, compressed_size,
c8b97818
CM		 181 PAGE_CACHE_SIZE);
182
7ac687d9 183 kaddr = kmap_atomic(cpage);
c8b97818 184 write_extent_buffer(leaf, kaddr, ptr, cur_size);
7ac687d9 185 kunmap_atomic(kaddr);
c8b97818
CM		 186
187 i++;
188 ptr += cur_size;
189 compressed_size -= cur_size;
190 }
191 btrfs_set_file_extent_compression(leaf, ei,
261507a0 192 compress_type);
c8b97818
CM		 193 } else {
194 page = find_get_page(inode->i_mapping,
195 start >> PAGE_CACHE_SHIFT);
196 btrfs_set_file_extent_compression(leaf, ei, 0);
7ac687d9 197 kaddr = kmap_atomic(page);
c8b97818
CM		 198 offset = start & (PAGE_CACHE_SIZE - 1);
199 write_extent_buffer(leaf, kaddr + offset, ptr, size);
7ac687d9 200 kunmap_atomic(kaddr);
c8b97818
CM		 201 page_cache_release(page);
202 }
203 btrfs_mark_buffer_dirty(leaf);
204 btrfs_free_path(path);
205
c2167754
YZ		 206 /*
207 * we're an inline extent, so nobody can
208 * extend the file past i_size without locking
209 * a page we already have locked.
210 *
211 * We must do any isize and inode updates
212 * before we unlock the pages. Otherwise we
213 * could end up racing with unlink.
214 */
c8b97818 215 BTRFS_I(inode)->disk_i_size = inode->i_size;
79787eaa 216 ret = btrfs_update_inode(trans, root, inode);
c2167754 217
79787eaa 218 return ret;
c8b97818
CM		 219fail:
220 btrfs_free_path(path);
221 return err;
222}
223
224
225/*
226 * conditionally insert an inline extent into the file. This
227 * does the checks required to make sure the data is small enough
228 * to fit as an inline extent.
229 */
7f366cfe 230static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
c8b97818
CM		 231 struct btrfs_root *root,
232 struct inode *inode, u64 start, u64 end,
fe3f566c 233 size_t compressed_size, int compress_type,
c8b97818
CM		 234 struct page **compressed_pages)
235{
236 u64 isize = i_size_read(inode);
237 u64 actual_end = min(end + 1, isize);
238 u64 inline_len = actual_end - start;
fda2832f 239 u64 aligned_end = ALIGN(end, root->sectorsize);
c8b97818
CM		 240 u64 data_len = inline_len;
241 int ret;
242
243 if (compressed_size)
244 data_len = compressed_size;
245
246 if (start > 0 ||
70b99e69 247 actual_end >= PAGE_CACHE_SIZE ||
c8b97818
CM		 248 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
249 (!compressed_size &&
250 (actual_end & (root->sectorsize - 1)) == 0) ||
251 end + 1 < isize ||
252 data_len > root->fs_info->max_inline) {
253 return 1;
254 }
255
2671485d 256 ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
79787eaa
JM		 257 if (ret)
258 return ret;
c8b97818
CM		 259
260 if (isize > actual_end)
261 inline_len = min_t(u64, isize, actual_end);
262 ret = insert_inline_extent(trans, root, inode, start,
263 inline_len, compressed_size,
fe3f566c 264 compress_type, compressed_pages);
2adcac1a 265 if (ret && ret != -ENOSPC) {
79787eaa
JM		 266 btrfs_abort_transaction(trans, root, ret);
267 return ret;
2adcac1a
JB		 268 } else if (ret == -ENOSPC) {
269 return 1;
79787eaa 270 }
2adcac1a 271
bdc20e67 272 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
0ca1f7ce 273 btrfs_delalloc_release_metadata(inode, end + 1 - start);
a1ed835e 274 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
c8b97818
CM		 275 return 0;
276}
277
771ed689
CM		 278struct async_extent {
279 u64 start;
280 u64 ram_size;
281 u64 compressed_size;
282 struct page **pages;
283 unsigned long nr_pages;
261507a0 284 int compress_type;
771ed689
CM		 285 struct list_head list;
286};
287
288struct async_cow {
289 struct inode *inode;
290 struct btrfs_root *root;
291 struct page *locked_page;
292 u64 start;
293 u64 end;
294 struct list_head extents;
295 struct btrfs_work work;
296};
297
298static noinline int add_async_extent(struct async_cow *cow,
299 u64 start, u64 ram_size,
300 u64 compressed_size,
301 struct page **pages,
261507a0
LZ		 302 unsigned long nr_pages,
303 int compress_type)
771ed689
CM		 304{
305 struct async_extent *async_extent;
306
307 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
79787eaa 308 BUG_ON(!async_extent); /* -ENOMEM */
771ed689
CM		 309 async_extent->start = start;
310 async_extent->ram_size = ram_size;
311 async_extent->compressed_size = compressed_size;
312 async_extent->pages = pages;
313 async_extent->nr_pages = nr_pages;
261507a0 314 async_extent->compress_type = compress_type;
771ed689
CM		 315 list_add_tail(&async_extent->list, &cow->extents);
316 return 0;
317}
318
d352ac68 319/*
771ed689
CM		 320 * we create compressed extents in two phases. The first
321 * phase compresses a range of pages that have already been
322 * locked (both pages and state bits are locked).
c8b97818 323 *
771ed689
CM		 324 * This is done inside an ordered work queue, and the compression
325 * is spread across many cpus. The actual IO submission is step
326 * two, and the ordered work queue takes care of making sure that
327 * happens in the same order things were put onto the queue by
328 * writepages and friends.
c8b97818 329 *
771ed689
CM		 330 * If this code finds it can't get good compression, it puts an
331 * entry onto the work queue to write the uncompressed bytes. This
332 * makes sure that both compressed inodes and uncompressed inodes
b2570314
AB		 333 * are written in the same order that the flusher thread sent them
334 * down.
d352ac68 335 */
771ed689
CM		 336static noinline int compress_file_range(struct inode *inode,
337 struct page *locked_page,
338 u64 start, u64 end,
339 struct async_cow *async_cow,
340 int *num_added)
b888db2b
CM		 341{
342 struct btrfs_root *root = BTRFS_I(inode)->root;
343 struct btrfs_trans_handle *trans;
db94535d 344 u64 num_bytes;
db94535d 345 u64 blocksize = root->sectorsize;
c8b97818 346 u64 actual_end;
42dc7bab 347 u64 isize = i_size_read(inode);
e6dcd2dc 348 int ret = 0;
c8b97818
CM		 349 struct page **pages = NULL;
350 unsigned long nr_pages;
351 unsigned long nr_pages_ret = 0;
352 unsigned long total_compressed = 0;
353 unsigned long total_in = 0;
354 unsigned long max_compressed = 128 * 1024;
771ed689 355 unsigned long max_uncompressed = 128 * 1024;
c8b97818
CM		 356 int i;
357 int will_compress;
261507a0 358 int compress_type = root->fs_info->compress_type;
4adaa611 359 int redirty = 0;
b888db2b 360
4cb13e5d
LB		 361 /* if this is a small write inside eof, kick off a defrag */
362 if ((end - start + 1) < 16 * 1024 &&
363 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
4cb5300b
CM		 364 btrfs_add_inode_defrag(NULL, inode);
365
42dc7bab 366 actual_end = min_t(u64, isize, end + 1);
c8b97818
CM		 367again:
368 will_compress = 0;
369 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
370 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
be20aa9d 371
f03d9301
CM		 372 /*
373 * we don't want to send crud past the end of i_size through
374 * compression, that's just a waste of CPU time. So, if the
375 * end of the file is before the start of our current
376 * requested range of bytes, we bail out to the uncompressed
377 * cleanup code that can deal with all of this.
378 *
379 * It isn't really the fastest way to fix things, but this is a
380 * very uncommon corner.
381 */
382 if (actual_end <= start)
383 goto cleanup_and_bail_uncompressed;
384
c8b97818
CM		 385 total_compressed = actual_end - start;
386
387 /* we want to make sure that amount of ram required to uncompress
388 * an extent is reasonable, so we limit the total size in ram
771ed689
CM		 389 * of a compressed extent to 128k. This is a crucial number
390 * because it also controls how easily we can spread reads across
391 * cpus for decompression.
392 *
393 * We also want to make sure the amount of IO required to do
394 * a random read is reasonably small, so we limit the size of
395 * a compressed extent to 128k.
c8b97818
CM		 396 */
397 total_compressed = min(total_compressed, max_uncompressed);
fda2832f 398 num_bytes = ALIGN(end - start + 1, blocksize);
be20aa9d 399 num_bytes = max(blocksize, num_bytes);
c8b97818
CM		 400 total_in = 0;
401 ret = 0;
db94535d 402
771ed689
CM		 403 /*
404 * we do compression for mount -o compress and when the
405 * inode has not been flagged as nocompress. This flag can
406 * change at any time if we discover bad compression ratios.
c8b97818 407 */
6cbff00f 408 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
1e701a32 409 (btrfs_test_opt(root, COMPRESS) ||
75e7cb7f
LB		 410 (BTRFS_I(inode)->force_compress) ||
411 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
c8b97818 412 WARN_ON(pages);
cfbc246e 413 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
560f7d75
LZ		 414 if (!pages) {
415 /* just bail out to the uncompressed code */
416 goto cont;
417 }
c8b97818 418
261507a0
LZ		 419 if (BTRFS_I(inode)->force_compress)
420 compress_type = BTRFS_I(inode)->force_compress;
421
4adaa611
CM		 422 /*
423 * we need to call clear_page_dirty_for_io on each
424 * page in the range. Otherwise applications with the file
425 * mmap'd can wander in and change the page contents while
426 * we are compressing them.
427 *
428 * If the compression fails for any reason, we set the pages
429 * dirty again later on.
430 */
431 extent_range_clear_dirty_for_io(inode, start, end);
432 redirty = 1;
261507a0
LZ		 433 ret = btrfs_compress_pages(compress_type,
434 inode->i_mapping, start,
435 total_compressed, pages,
436 nr_pages, &nr_pages_ret,
437 &total_in,
438 &total_compressed,
439 max_compressed);
c8b97818
CM		 440
441 if (!ret) {
442 unsigned long offset = total_compressed &
443 (PAGE_CACHE_SIZE - 1);
444 struct page *page = pages[nr_pages_ret - 1];
445 char *kaddr;
446
447 /* zero the tail end of the last page, we might be
448 * sending it down to disk
449 */
450 if (offset) {
7ac687d9 451 kaddr = kmap_atomic(page);
c8b97818
CM		 452 memset(kaddr + offset, 0,
453 PAGE_CACHE_SIZE - offset);
7ac687d9 454 kunmap_atomic(kaddr);
c8b97818
CM		 455 }
456 will_compress = 1;
457 }
458 }
560f7d75 459cont:
c8b97818 460 if (start == 0) {
7a7eaa40 461 trans = btrfs_join_transaction(root);
79787eaa
JM		 462 if (IS_ERR(trans)) {
463 ret = PTR_ERR(trans);
464 trans = NULL;
465 goto cleanup_and_out;
466 }
0ca1f7ce 467 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
771ed689 468
c8b97818 469 /* lets try to make an inline extent */
771ed689 470 if (ret || total_in < (actual_end - start)) {
c8b97818 471 /* we didn't compress the entire range, try
771ed689 472 * to make an uncompressed inline extent.
c8b97818
CM		 473 */
474 ret = cow_file_range_inline(trans, root, inode,
fe3f566c 475 start, end, 0, 0, NULL);
c8b97818 476 } else {
771ed689 477 /* try making a compressed inline extent */
c8b97818
CM		 478 ret = cow_file_range_inline(trans, root, inode,
479 start, end,
fe3f566c
LZ		 480 total_compressed,
481 compress_type, pages);
c8b97818 482 }
79787eaa 483 if (ret <= 0) {
771ed689 484 /*
79787eaa
JM		 485 * inline extent creation worked or returned error,
486 * we don't need to create any more async work items.
487 * Unlock and free up our temp pages.
771ed689 488 */
c8b97818 489 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 490 &BTRFS_I(inode)->io_tree,
491 start, end, NULL,
492 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
a3429ab7 493 EXTENT_CLEAR_DELALLOC |
a791e35e 494 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
c2167754
YZ		 495
496 btrfs_end_transaction(trans, root);
c8b97818
CM		 497 goto free_pages_out;
498 }
c2167754 499 btrfs_end_transaction(trans, root);
c8b97818
CM		 500 }
501
502 if (will_compress) {
503 /*
504 * we aren't doing an inline extent round the compressed size
505 * up to a block size boundary so the allocator does sane
506 * things
507 */
fda2832f 508 total_compressed = ALIGN(total_compressed, blocksize);
c8b97818
CM		 509
510 /*
511 * one last check to make sure the compression is really a
512 * win, compare the page count read with the blocks on disk
513 */
fda2832f 514 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
c8b97818
CM		 515 if (total_compressed >= total_in) {
516 will_compress = 0;
517 } else {
c8b97818
CM		 518 num_bytes = total_in;
519 }
520 }
521 if (!will_compress && pages) {
522 /*
523 * the compression code ran but failed to make things smaller,
524 * free any pages it allocated and our page pointer array
525 */
526 for (i = 0; i < nr_pages_ret; i++) {
70b99e69 527 WARN_ON(pages[i]->mapping);
c8b97818
CM		 528 page_cache_release(pages[i]);
529 }
530 kfree(pages);
531 pages = NULL;
532 total_compressed = 0;
533 nr_pages_ret = 0;
534
535 /* flag the file so we don't compress in the future */
1e701a32
CM		 536 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
537 !(BTRFS_I(inode)->force_compress)) {
a555f810 538 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
1e701a32 539 }
c8b97818 540 }
771ed689
CM		 541 if (will_compress) {
542 *num_added += 1;
c8b97818 543
771ed689
CM		 544 /* the async work queues will take care of doing actual
545 * allocation on disk for these compressed pages,
546 * and will submit them to the elevator.
547 */
548 add_async_extent(async_cow, start, num_bytes,
261507a0
LZ		 549 total_compressed, pages, nr_pages_ret,
550 compress_type);
179e29e4 551
24ae6365 552 if (start + num_bytes < end) {
771ed689
CM		 553 start += num_bytes;
554 pages = NULL;
555 cond_resched();
556 goto again;
557 }
558 } else {
f03d9301 559cleanup_and_bail_uncompressed:
771ed689
CM		 560 /*
561 * No compression, but we still need to write the pages in
562 * the file we've been given so far. redirty the locked
563 * page if it corresponds to our extent and set things up
564 * for the async work queue to run cow_file_range to do
565 * the normal delalloc dance
566 */
567 if (page_offset(locked_page) >= start &&
568 page_offset(locked_page) <= end) {
569 __set_page_dirty_nobuffers(locked_page);
570 /* unlocked later on in the async handlers */
571 }
4adaa611
CM		 572 if (redirty)
573 extent_range_redirty_for_io(inode, start, end);
261507a0
LZ		 574 add_async_extent(async_cow, start, end - start + 1,
575 0, NULL, 0, BTRFS_COMPRESS_NONE);
771ed689
CM		 576 *num_added += 1;
577 }
3b951516 578
771ed689 579out:
79787eaa 580 return ret;
771ed689
CM		 581
582free_pages_out:
583 for (i = 0; i < nr_pages_ret; i++) {
584 WARN_ON(pages[i]->mapping);
585 page_cache_release(pages[i]);
586 }
d397712b 587 kfree(pages);
771ed689
CM		 588
589 goto out;
79787eaa
JM		 590
591cleanup_and_out:
592 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
593 start, end, NULL,
594 EXTENT_CLEAR_UNLOCK_PAGE |
595 EXTENT_CLEAR_DIRTY |
596 EXTENT_CLEAR_DELALLOC |
597 EXTENT_SET_WRITEBACK |
598 EXTENT_END_WRITEBACK);
599 if (!trans || IS_ERR(trans))
600 btrfs_error(root->fs_info, ret, "Failed to join transaction");
601 else
602 btrfs_abort_transaction(trans, root, ret);
603 goto free_pages_out;
771ed689
CM		 604}
605
606/*
607 * phase two of compressed writeback. This is the ordered portion
608 * of the code, which only gets called in the order the work was
609 * queued. We walk all the async extents created by compress_file_range
610 * and send them down to the disk.
611 */
612static noinline int submit_compressed_extents(struct inode *inode,
613 struct async_cow *async_cow)
614{
615 struct async_extent *async_extent;
616 u64 alloc_hint = 0;
617 struct btrfs_trans_handle *trans;
618 struct btrfs_key ins;
619 struct extent_map *em;
620 struct btrfs_root *root = BTRFS_I(inode)->root;
621 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
622 struct extent_io_tree *io_tree;
f5a84ee3 623 int ret = 0;
771ed689
CM		 624
625 if (list_empty(&async_cow->extents))
626 return 0;
627
3e04e7f1 628again:
d397712b 629 while (!list_empty(&async_cow->extents)) {
771ed689
CM		 630 async_extent = list_entry(async_cow->extents.next,
631 struct async_extent, list);
632 list_del(&async_extent->list);
c8b97818 633
771ed689
CM		 634 io_tree = &BTRFS_I(inode)->io_tree;
635
f5a84ee3 636retry:
771ed689
CM		 637 /* did the compression code fall back to uncompressed IO? */
638 if (!async_extent->pages) {
639 int page_started = 0;
640 unsigned long nr_written = 0;
641
642 lock_extent(io_tree, async_extent->start,
2ac55d41 643 async_extent->start +
d0082371 644 async_extent->ram_size - 1);
771ed689
CM		 645
646 /* allocate blocks */
f5a84ee3
JB		 647 ret = cow_file_range(inode, async_cow->locked_page,
648 async_extent->start,
649 async_extent->start +
650 async_extent->ram_size - 1,
651 &page_started, &nr_written, 0);
771ed689 652
79787eaa
JM		 653 /* JDM XXX */
654
771ed689
CM		 655 /*
656 * if page_started, cow_file_range inserted an
657 * inline extent and took care of all the unlocking
658 * and IO for us. Otherwise, we need to submit
659 * all those pages down to the drive.
660 */
f5a84ee3 661 if (!page_started && !ret)
771ed689
CM		 662 extent_write_locked_range(io_tree,
663 inode, async_extent->start,
d397712b 664 async_extent->start +
771ed689
CM		 665 async_extent->ram_size - 1,
666 btrfs_get_extent,
667 WB_SYNC_ALL);
3e04e7f1
JB		 668 else if (ret)
669 unlock_page(async_cow->locked_page);
771ed689
CM		 670 kfree(async_extent);
671 cond_resched();
672 continue;
673 }
674
675 lock_extent(io_tree, async_extent->start,
d0082371 676 async_extent->start + async_extent->ram_size - 1);
771ed689 677
7a7eaa40 678 trans = btrfs_join_transaction(root);
79787eaa
JM		 679 if (IS_ERR(trans)) {
680 ret = PTR_ERR(trans);
681 } else {
682 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
683 ret = btrfs_reserve_extent(trans, root,
771ed689
CM		 684 async_extent->compressed_size,
685 async_extent->compressed_size,
81c9ad23 686 0, alloc_hint, &ins, 1);
962197ba 687 if (ret && ret != -ENOSPC)
79787eaa
JM		 688 btrfs_abort_transaction(trans, root, ret);
689 btrfs_end_transaction(trans, root);
690 }
c2167754 691
f5a84ee3
JB		 692 if (ret) {
693 int i;
3e04e7f1 694
f5a84ee3
JB		 695 for (i = 0; i < async_extent->nr_pages; i++) {
696 WARN_ON(async_extent->pages[i]->mapping);
697 page_cache_release(async_extent->pages[i]);
698 }
699 kfree(async_extent->pages);
700 async_extent->nr_pages = 0;
701 async_extent->pages = NULL;
3e04e7f1 702
79787eaa
JM		 703 if (ret == -ENOSPC)
704 goto retry;
3e04e7f1 705 goto out_free;
f5a84ee3
JB		 706 }
707
c2167754
YZ		 708 /*
709 * here we're doing allocation and writeback of the
710 * compressed pages
711 */
712 btrfs_drop_extent_cache(inode, async_extent->start,
713 async_extent->start +
714 async_extent->ram_size - 1, 0);
715
172ddd60 716 em = alloc_extent_map();
b9aa55be
LB		 717 if (!em) {
718 ret = -ENOMEM;
3e04e7f1 719 goto out_free_reserve;
b9aa55be 720 }
771ed689
CM		 721 em->start = async_extent->start;
722 em->len = async_extent->ram_size;
445a6944 723 em->orig_start = em->start;
2ab28f32
JB		 724 em->mod_start = em->start;
725 em->mod_len = em->len;
c8b97818 726
771ed689
CM		 727 em->block_start = ins.objectid;
728 em->block_len = ins.offset;
b4939680 729 em->orig_block_len = ins.offset;
cc95bef6 730 em->ram_bytes = async_extent->ram_size;
771ed689 731 em->bdev = root->fs_info->fs_devices->latest_bdev;
261507a0 732 em->compress_type = async_extent->compress_type;
771ed689
CM		 733 set_bit(EXTENT_FLAG_PINNED, &em->flags);
734 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
70c8a91c 735 em->generation = -1;
771ed689 736
d397712b 737 while (1) {
890871be 738 write_lock(&em_tree->lock);
09a2a8f9 739 ret = add_extent_mapping(em_tree, em, 1);
890871be 740 write_unlock(&em_tree->lock);
771ed689
CM		 741 if (ret != -EEXIST) {
742 free_extent_map(em);
743 break;
744 }
745 btrfs_drop_extent_cache(inode, async_extent->start,
746 async_extent->start +
747 async_extent->ram_size - 1, 0);
748 }
749
3e04e7f1
JB		 750 if (ret)
751 goto out_free_reserve;
752
261507a0
LZ		 753 ret = btrfs_add_ordered_extent_compress(inode,
754 async_extent->start,
755 ins.objectid,
756 async_extent->ram_size,
757 ins.offset,
758 BTRFS_ORDERED_COMPRESSED,
759 async_extent->compress_type);
3e04e7f1
JB		 760 if (ret)
761 goto out_free_reserve;
771ed689 762
771ed689
CM		 763 /*
764 * clear dirty, set writeback and unlock the pages.
765 */
766 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 767 &BTRFS_I(inode)->io_tree,
768 async_extent->start,
769 async_extent->start +
770 async_extent->ram_size - 1,
771 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
772 EXTENT_CLEAR_UNLOCK |
a3429ab7 773 EXTENT_CLEAR_DELALLOC |
a791e35e 774 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
771ed689
CM		 775
776 ret = btrfs_submit_compressed_write(inode,
d397712b
CM		 777 async_extent->start,
778 async_extent->ram_size,
779 ins.objectid,
780 ins.offset, async_extent->pages,
781 async_extent->nr_pages);
771ed689
CM		 782 alloc_hint = ins.objectid + ins.offset;
783 kfree(async_extent);
3e04e7f1
JB		 784 if (ret)
785 goto out;
771ed689
CM		 786 cond_resched();
787 }
79787eaa
JM		 788 ret = 0;
789out:
790 return ret;
3e04e7f1
JB		 791out_free_reserve:
792 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
79787eaa 793out_free:
3e04e7f1
JB		 794 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
795 async_extent->start,
796 async_extent->start +
797 async_extent->ram_size - 1,
798 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
799 EXTENT_CLEAR_UNLOCK |
800 EXTENT_CLEAR_DELALLOC |
801 EXTENT_CLEAR_DIRTY |
802 EXTENT_SET_WRITEBACK |
803 EXTENT_END_WRITEBACK);
79787eaa 804 kfree(async_extent);
3e04e7f1 805 goto again;
771ed689
CM		 806}
807
4b46fce2
JB		 808static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
809 u64 num_bytes)
810{
811 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
812 struct extent_map *em;
813 u64 alloc_hint = 0;
814
815 read_lock(&em_tree->lock);
816 em = search_extent_mapping(em_tree, start, num_bytes);
817 if (em) {
818 /*
819 * if block start isn't an actual block number then find the
820 * first block in this inode and use that as a hint. If that
821 * block is also bogus then just don't worry about it.
822 */
823 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
824 free_extent_map(em);
825 em = search_extent_mapping(em_tree, 0, 0);
826 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
827 alloc_hint = em->block_start;
828 if (em)
829 free_extent_map(em);
830 } else {
831 alloc_hint = em->block_start;
832 free_extent_map(em);
833 }
834 }
835 read_unlock(&em_tree->lock);
836
837 return alloc_hint;
838}
839
771ed689
CM		 840/*
841 * when extent_io.c finds a delayed allocation range in the file,
842 * the call backs end up in this code. The basic idea is to
843 * allocate extents on disk for the range, and create ordered data structs
844 * in ram to track those extents.
845 *
846 * locked_page is the page that writepage had locked already. We use
847 * it to make sure we don't do extra locks or unlocks.
848 *
849 * *page_started is set to one if we unlock locked_page and do everything
850 * required to start IO on it. It may be clean and already done with
851 * IO when we return.
852 */
b7d5b0a8
MX		 853static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
854 struct inode *inode,
855 struct btrfs_root *root,
856 struct page *locked_page,
857 u64 start, u64 end, int *page_started,
858 unsigned long *nr_written,
859 int unlock)
771ed689 860{
771ed689
CM		 861 u64 alloc_hint = 0;
862 u64 num_bytes;
863 unsigned long ram_size;
864 u64 disk_num_bytes;
865 u64 cur_alloc_size;
866 u64 blocksize = root->sectorsize;
771ed689
CM		 867 struct btrfs_key ins;
868 struct extent_map *em;
869 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
870 int ret = 0;
871
83eea1f1 872 BUG_ON(btrfs_is_free_space_inode(inode));
771ed689 873
fda2832f 874 num_bytes = ALIGN(end - start + 1, blocksize);
771ed689
CM		 875 num_bytes = max(blocksize, num_bytes);
876 disk_num_bytes = num_bytes;
771ed689 877
4cb5300b 878 /* if this is a small write inside eof, kick off defrag */
4cb13e5d
LB		 879 if (num_bytes < 64 * 1024 &&
880 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
4cb5300b
CM		 881 btrfs_add_inode_defrag(trans, inode);
882
771ed689
CM		 883 if (start == 0) {
884 /* lets try to make an inline extent */
885 ret = cow_file_range_inline(trans, root, inode,
fe3f566c 886 start, end, 0, 0, NULL);
771ed689
CM		 887 if (ret == 0) {
888 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 889 &BTRFS_I(inode)->io_tree,
890 start, end, NULL,
891 EXTENT_CLEAR_UNLOCK_PAGE |
892 EXTENT_CLEAR_UNLOCK |
893 EXTENT_CLEAR_DELALLOC |
894 EXTENT_CLEAR_DIRTY |
895 EXTENT_SET_WRITEBACK |
896 EXTENT_END_WRITEBACK);
c2167754 897
771ed689
CM		 898 *nr_written = *nr_written +
899 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
900 *page_started = 1;
771ed689 901 goto out;
79787eaa
JM		 902 } else if (ret < 0) {
903 btrfs_abort_transaction(trans, root, ret);
904 goto out_unlock;
771ed689
CM		 905 }
906 }
907
908 BUG_ON(disk_num_bytes >
6c41761f 909 btrfs_super_total_bytes(root->fs_info->super_copy));
771ed689 910
4b46fce2 911 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
771ed689
CM		 912 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
913
d397712b 914 while (disk_num_bytes > 0) {
a791e35e
CM		 915 unsigned long op;
916
287a0ab9 917 cur_alloc_size = disk_num_bytes;
e6dcd2dc 918 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
771ed689 919 root->sectorsize, 0, alloc_hint,
81c9ad23 920 &ins, 1);
79787eaa
JM		 921 if (ret < 0) {
922 btrfs_abort_transaction(trans, root, ret);
923 goto out_unlock;
924 }
d397712b 925
172ddd60 926 em = alloc_extent_map();
b9aa55be
LB		 927 if (!em) {
928 ret = -ENOMEM;
ace68bac 929 goto out_reserve;
b9aa55be 930 }
e6dcd2dc 931 em->start = start;
445a6944 932 em->orig_start = em->start;
771ed689
CM		 933 ram_size = ins.offset;
934 em->len = ins.offset;
2ab28f32
JB		 935 em->mod_start = em->start;
936 em->mod_len = em->len;
c8b97818 937
e6dcd2dc 938 em->block_start = ins.objectid;
c8b97818 939 em->block_len = ins.offset;
b4939680 940 em->orig_block_len = ins.offset;
cc95bef6 941 em->ram_bytes = ram_size;
e6dcd2dc 942 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 943 set_bit(EXTENT_FLAG_PINNED, &em->flags);
70c8a91c 944 em->generation = -1;
c8b97818 945
d397712b 946 while (1) {
890871be 947 write_lock(&em_tree->lock);
09a2a8f9 948 ret = add_extent_mapping(em_tree, em, 1);
890871be 949 write_unlock(&em_tree->lock);
e6dcd2dc
CM		 950 if (ret != -EEXIST) {
951 free_extent_map(em);
952 break;
953 }
954 btrfs_drop_extent_cache(inode, start,
c8b97818 955 start + ram_size - 1, 0);
e6dcd2dc 956 }
ace68bac
LB		 957 if (ret)
958 goto out_reserve;
e6dcd2dc 959
98d20f67 960 cur_alloc_size = ins.offset;
e6dcd2dc 961 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771ed689 962 ram_size, cur_alloc_size, 0);
ace68bac
LB		 963 if (ret)
964 goto out_reserve;
c8b97818 965
17d217fe
YZ		 966 if (root->root_key.objectid ==
967 BTRFS_DATA_RELOC_TREE_OBJECTID) {
968 ret = btrfs_reloc_clone_csums(inode, start,
969 cur_alloc_size);
79787eaa
JM		 970 if (ret) {
971 btrfs_abort_transaction(trans, root, ret);
ace68bac 972 goto out_reserve;
79787eaa 973 }
17d217fe
YZ		 974 }
975
d397712b 976 if (disk_num_bytes < cur_alloc_size)
3b951516 977 break;
d397712b 978
c8b97818
CM		 979 /* we're not doing compressed IO, don't unlock the first
980 * page (which the caller expects to stay locked), don't
981 * clear any dirty bits and don't set any writeback bits
8b62b72b
CM		 982 *
983 * Do set the Private2 bit so we know this page was properly
984 * setup for writepage
c8b97818 985 */
a791e35e
CM		 986 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
987 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
988 EXTENT_SET_PRIVATE2;
989
c8b97818
CM		 990 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
991 start, start + ram_size - 1,
a791e35e 992 locked_page, op);
c8b97818 993 disk_num_bytes -= cur_alloc_size;
c59f8951
CM		 994 num_bytes -= cur_alloc_size;
995 alloc_hint = ins.objectid + ins.offset;
996 start += cur_alloc_size;
b888db2b 997 }
79787eaa 998out:
be20aa9d 999 return ret;
b7d5b0a8 1000
ace68bac
LB		 1001out_reserve:
1002 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
79787eaa
JM		 1003out_unlock:
1004 extent_clear_unlock_delalloc(inode,
1005 &BTRFS_I(inode)->io_tree,
beb42dd7 1006 start, end, locked_page,
79787eaa
JM		 1007 EXTENT_CLEAR_UNLOCK_PAGE |
1008 EXTENT_CLEAR_UNLOCK |
1009 EXTENT_CLEAR_DELALLOC |
1010 EXTENT_CLEAR_DIRTY |
1011 EXTENT_SET_WRITEBACK |
1012 EXTENT_END_WRITEBACK);
1013
1014 goto out;
771ed689 1015}
c8b97818 1016
b7d5b0a8
MX		 1017static noinline int cow_file_range(struct inode *inode,
1018 struct page *locked_page,
1019 u64 start, u64 end, int *page_started,
1020 unsigned long *nr_written,
1021 int unlock)
1022{
1023 struct btrfs_trans_handle *trans;
1024 struct btrfs_root *root = BTRFS_I(inode)->root;
1025 int ret;
1026
1027 trans = btrfs_join_transaction(root);
1028 if (IS_ERR(trans)) {
1029 extent_clear_unlock_delalloc(inode,
1030 &BTRFS_I(inode)->io_tree,
1031 start, end, locked_page,
1032 EXTENT_CLEAR_UNLOCK_PAGE |
1033 EXTENT_CLEAR_UNLOCK |
1034 EXTENT_CLEAR_DELALLOC |
1035 EXTENT_CLEAR_DIRTY |
1036 EXTENT_SET_WRITEBACK |
1037 EXTENT_END_WRITEBACK);
1038 return PTR_ERR(trans);
1039 }
1040 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1041
1042 ret = __cow_file_range(trans, inode, root, locked_page, start, end,
1043 page_started, nr_written, unlock);
1044
1045 btrfs_end_transaction(trans, root);
1046
1047 return ret;
1048}
1049
771ed689
CM		 1050/*
1051 * work queue call back to started compression on a file and pages
1052 */
1053static noinline void async_cow_start(struct btrfs_work *work)
1054{
1055 struct async_cow *async_cow;
1056 int num_added = 0;
1057 async_cow = container_of(work, struct async_cow, work);
1058
1059 compress_file_range(async_cow->inode, async_cow->locked_page,
1060 async_cow->start, async_cow->end, async_cow,
1061 &num_added);
8180ef88 1062 if (num_added == 0) {
cb77fcd8 1063 btrfs_add_delayed_iput(async_cow->inode);
771ed689 1064 async_cow->inode = NULL;
8180ef88 1065 }
771ed689
CM		 1066}
1067
1068/*
1069 * work queue call back to submit previously compressed pages
1070 */
1071static noinline void async_cow_submit(struct btrfs_work *work)
1072{
1073 struct async_cow *async_cow;
1074 struct btrfs_root *root;
1075 unsigned long nr_pages;
1076
1077 async_cow = container_of(work, struct async_cow, work);
1078
1079 root = async_cow->root;
1080 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1081 PAGE_CACHE_SHIFT;
1082
66657b31 1083 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
287082b0 1084 5 * 1024 * 1024 &&
771ed689
CM		 1085 waitqueue_active(&root->fs_info->async_submit_wait))
1086 wake_up(&root->fs_info->async_submit_wait);
1087
d397712b 1088 if (async_cow->inode)
771ed689 1089 submit_compressed_extents(async_cow->inode, async_cow);
771ed689 1090}
c8b97818 1091
771ed689
CM		 1092static noinline void async_cow_free(struct btrfs_work *work)
1093{
1094 struct async_cow *async_cow;
1095 async_cow = container_of(work, struct async_cow, work);
8180ef88 1096 if (async_cow->inode)
cb77fcd8 1097 btrfs_add_delayed_iput(async_cow->inode);
771ed689
CM		 1098 kfree(async_cow);
1099}
1100
1101static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1102 u64 start, u64 end, int *page_started,
1103 unsigned long *nr_written)
1104{
1105 struct async_cow *async_cow;
1106 struct btrfs_root *root = BTRFS_I(inode)->root;
1107 unsigned long nr_pages;
1108 u64 cur_end;
287082b0 1109 int limit = 10 * 1024 * 1024;
771ed689 1110
a3429ab7
CM		 1111 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1112 1, 0, NULL, GFP_NOFS);
d397712b 1113 while (start < end) {
771ed689 1114 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
79787eaa 1115 BUG_ON(!async_cow); /* -ENOMEM */
8180ef88 1116 async_cow->inode = igrab(inode);
771ed689
CM		 1117 async_cow->root = root;
1118 async_cow->locked_page = locked_page;
1119 async_cow->start = start;
1120
6cbff00f 1121 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
771ed689
CM		 1122 cur_end = end;
1123 else
1124 cur_end = min(end, start + 512 * 1024 - 1);
1125
1126 async_cow->end = cur_end;
1127 INIT_LIST_HEAD(&async_cow->extents);
1128
1129 async_cow->work.func = async_cow_start;
1130 async_cow->work.ordered_func = async_cow_submit;
1131 async_cow->work.ordered_free = async_cow_free;
1132 async_cow->work.flags = 0;
1133
771ed689
CM		 1134 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1135 PAGE_CACHE_SHIFT;
1136 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1137
1138 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1139 &async_cow->work);
1140
1141 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1142 wait_event(root->fs_info->async_submit_wait,
1143 (atomic_read(&root->fs_info->async_delalloc_pages) <
1144 limit));
1145 }
1146
d397712b 1147 while (atomic_read(&root->fs_info->async_submit_draining) &&
771ed689
CM		 1148 atomic_read(&root->fs_info->async_delalloc_pages)) {
1149 wait_event(root->fs_info->async_submit_wait,
1150 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1151 0));
1152 }
1153
1154 *nr_written += nr_pages;
1155 start = cur_end + 1;
1156 }
1157 *page_started = 1;
1158 return 0;
be20aa9d
CM		 1159}
1160
d397712b 1161static noinline int csum_exist_in_range(struct btrfs_root *root,
17d217fe
YZ		 1162 u64 bytenr, u64 num_bytes)
1163{
1164 int ret;
1165 struct btrfs_ordered_sum *sums;
1166 LIST_HEAD(list);
1167
07d400a6 1168 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
a2de733c 1169 bytenr + num_bytes - 1, &list, 0);
17d217fe
YZ		 1170 if (ret == 0 && list_empty(&list))
1171 return 0;
1172
1173 while (!list_empty(&list)) {
1174 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1175 list_del(&sums->list);
1176 kfree(sums);
1177 }
1178 return 1;
1179}
1180
d352ac68
CM		 1181/*
1182 * when nowcow writeback call back. This checks for snapshots or COW copies
1183 * of the extents that exist in the file, and COWs the file as required.
1184 *
1185 * If no cow copies or snapshots exist, we write directly to the existing
1186 * blocks on disk
1187 */
7f366cfe
CM		 1188static noinline int run_delalloc_nocow(struct inode *inode,
1189 struct page *locked_page,
771ed689
CM		 1190 u64 start, u64 end, int *page_started, int force,
1191 unsigned long *nr_written)
be20aa9d 1192{
be20aa9d 1193 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 1194 struct btrfs_trans_handle *trans;
be20aa9d 1195 struct extent_buffer *leaf;
be20aa9d 1196 struct btrfs_path *path;
80ff3856 1197 struct btrfs_file_extent_item *fi;
be20aa9d 1198 struct btrfs_key found_key;
80ff3856
YZ		 1199 u64 cow_start;
1200 u64 cur_offset;
1201 u64 extent_end;
5d4f98a2 1202 u64 extent_offset;
80ff3856
YZ		 1203 u64 disk_bytenr;
1204 u64 num_bytes;
b4939680 1205 u64 disk_num_bytes;
cc95bef6 1206 u64 ram_bytes;
80ff3856 1207 int extent_type;
79787eaa 1208 int ret, err;
d899e052 1209 int type;
80ff3856
YZ		 1210 int nocow;
1211 int check_prev = 1;
82d5902d 1212 bool nolock;
33345d01 1213 u64 ino = btrfs_ino(inode);
be20aa9d
CM		 1214
1215 path = btrfs_alloc_path();
17ca04af
JB		 1216 if (!path) {
1217 extent_clear_unlock_delalloc(inode,
1218 &BTRFS_I(inode)->io_tree,
1219 start, end, locked_page,
1220 EXTENT_CLEAR_UNLOCK_PAGE |
1221 EXTENT_CLEAR_UNLOCK |
1222 EXTENT_CLEAR_DELALLOC |
1223 EXTENT_CLEAR_DIRTY |
1224 EXTENT_SET_WRITEBACK |
1225 EXTENT_END_WRITEBACK);
d8926bb3 1226 return -ENOMEM;
17ca04af 1227 }
82d5902d 1228
83eea1f1 1229 nolock = btrfs_is_free_space_inode(inode);
82d5902d
LZ		 1230
1231 if (nolock)
7a7eaa40 1232 trans = btrfs_join_transaction_nolock(root);
82d5902d 1233 else
7a7eaa40 1234 trans = btrfs_join_transaction(root);
ff5714cc 1235
79787eaa 1236 if (IS_ERR(trans)) {
17ca04af
JB		 1237 extent_clear_unlock_delalloc(inode,
1238 &BTRFS_I(inode)->io_tree,
1239 start, end, locked_page,
1240 EXTENT_CLEAR_UNLOCK_PAGE |
1241 EXTENT_CLEAR_UNLOCK |
1242 EXTENT_CLEAR_DELALLOC |
1243 EXTENT_CLEAR_DIRTY |
1244 EXTENT_SET_WRITEBACK |
1245 EXTENT_END_WRITEBACK);
79787eaa
JM		 1246 btrfs_free_path(path);
1247 return PTR_ERR(trans);
1248 }
1249
74b21075 1250 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
be20aa9d 1251
80ff3856
YZ		 1252 cow_start = (u64)-1;
1253 cur_offset = start;
1254 while (1) {
33345d01 1255 ret = btrfs_lookup_file_extent(trans, root, path, ino,
80ff3856 1256 cur_offset, 0);
79787eaa
JM		 1257 if (ret < 0) {
1258 btrfs_abort_transaction(trans, root, ret);
1259 goto error;
1260 }
80ff3856
YZ		 1261 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1262 leaf = path->nodes[0];
1263 btrfs_item_key_to_cpu(leaf, &found_key,
1264 path->slots[0] - 1);
33345d01 1265 if (found_key.objectid == ino &&
80ff3856
YZ		 1266 found_key.type == BTRFS_EXTENT_DATA_KEY)
1267 path->slots[0]--;
1268 }
1269 check_prev = 0;
1270next_slot:
1271 leaf = path->nodes[0];
1272 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1273 ret = btrfs_next_leaf(root, path);
79787eaa
JM		 1274 if (ret < 0) {
1275 btrfs_abort_transaction(trans, root, ret);
1276 goto error;
1277 }
80ff3856
YZ		 1278 if (ret > 0)
1279 break;
1280 leaf = path->nodes[0];
1281 }
be20aa9d 1282
80ff3856
YZ		 1283 nocow = 0;
1284 disk_bytenr = 0;
17d217fe 1285 num_bytes = 0;
80ff3856
YZ		 1286 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1287
33345d01 1288 if (found_key.objectid > ino ||
80ff3856
YZ		 1289 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1290 found_key.offset > end)
1291 break;
1292
1293 if (found_key.offset > cur_offset) {
1294 extent_end = found_key.offset;
e9061e21 1295 extent_type = 0;
80ff3856
YZ		 1296 goto out_check;
1297 }
1298
1299 fi = btrfs_item_ptr(leaf, path->slots[0],
1300 struct btrfs_file_extent_item);
1301 extent_type = btrfs_file_extent_type(leaf, fi);
1302
cc95bef6 1303 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
d899e052
YZ		 1304 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1305 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856 1306 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5d4f98a2 1307 extent_offset = btrfs_file_extent_offset(leaf, fi);
80ff3856
YZ		 1308 extent_end = found_key.offset +
1309 btrfs_file_extent_num_bytes(leaf, fi);
b4939680
JB		 1310 disk_num_bytes =
1311 btrfs_file_extent_disk_num_bytes(leaf, fi);
80ff3856
YZ		 1312 if (extent_end <= start) {
1313 path->slots[0]++;
1314 goto next_slot;
1315 }
17d217fe
YZ		 1316 if (disk_bytenr == 0)
1317 goto out_check;
80ff3856
YZ		 1318 if (btrfs_file_extent_compression(leaf, fi) ||
1319 btrfs_file_extent_encryption(leaf, fi) ||
1320 btrfs_file_extent_other_encoding(leaf, fi))
1321 goto out_check;
d899e052
YZ		 1322 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1323 goto out_check;
d2fb3437 1324 if (btrfs_extent_readonly(root, disk_bytenr))
80ff3856 1325 goto out_check;
33345d01 1326 if (btrfs_cross_ref_exist(trans, root, ino,
5d4f98a2
YZ		 1327 found_key.offset -
1328 extent_offset, disk_bytenr))
17d217fe 1329 goto out_check;
5d4f98a2 1330 disk_bytenr += extent_offset;
17d217fe
YZ		 1331 disk_bytenr += cur_offset - found_key.offset;
1332 num_bytes = min(end + 1, extent_end) - cur_offset;
1333 /*
1334 * force cow if csum exists in the range.
1335 * this ensure that csum for a given extent are
1336 * either valid or do not exist.
1337 */
1338 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1339 goto out_check;
80ff3856
YZ		 1340 nocow = 1;
1341 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1342 extent_end = found_key.offset +
1343 btrfs_file_extent_inline_len(leaf, fi);
1344 extent_end = ALIGN(extent_end, root->sectorsize);
1345 } else {
1346 BUG_ON(1);
1347 }
1348out_check:
1349 if (extent_end <= start) {
1350 path->slots[0]++;
1351 goto next_slot;
1352 }
1353 if (!nocow) {
1354 if (cow_start == (u64)-1)
1355 cow_start = cur_offset;
1356 cur_offset = extent_end;
1357 if (cur_offset > end)
1358 break;
1359 path->slots[0]++;
1360 goto next_slot;
7ea394f1
YZ		 1361 }
1362
b3b4aa74 1363 btrfs_release_path(path);
80ff3856 1364 if (cow_start != (u64)-1) {
b7d5b0a8
MX		 1365 ret = __cow_file_range(trans, inode, root, locked_page,
1366 cow_start, found_key.offset - 1,
1367 page_started, nr_written, 1);
79787eaa
JM		 1368 if (ret) {
1369 btrfs_abort_transaction(trans, root, ret);
1370 goto error;
1371 }
80ff3856 1372 cow_start = (u64)-1;
7ea394f1 1373 }
80ff3856 1374
d899e052
YZ		 1375 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1376 struct extent_map *em;
1377 struct extent_map_tree *em_tree;
1378 em_tree = &BTRFS_I(inode)->extent_tree;
172ddd60 1379 em = alloc_extent_map();
79787eaa 1380 BUG_ON(!em); /* -ENOMEM */
d899e052 1381 em->start = cur_offset;
70c8a91c 1382 em->orig_start = found_key.offset - extent_offset;
d899e052
YZ		 1383 em->len = num_bytes;
1384 em->block_len = num_bytes;
1385 em->block_start = disk_bytenr;
b4939680 1386 em->orig_block_len = disk_num_bytes;
cc95bef6 1387 em->ram_bytes = ram_bytes;
d899e052 1388 em->bdev = root->fs_info->fs_devices->latest_bdev;
2ab28f32
JB		 1389 em->mod_start = em->start;
1390 em->mod_len = em->len;
d899e052 1391 set_bit(EXTENT_FLAG_PINNED, &em->flags);
b11e234d 1392 set_bit(EXTENT_FLAG_FILLING, &em->flags);
70c8a91c 1393 em->generation = -1;
d899e052 1394 while (1) {
890871be 1395 write_lock(&em_tree->lock);
09a2a8f9 1396 ret = add_extent_mapping(em_tree, em, 1);
890871be 1397 write_unlock(&em_tree->lock);
d899e052
YZ		 1398 if (ret != -EEXIST) {
1399 free_extent_map(em);
1400 break;
1401 }
1402 btrfs_drop_extent_cache(inode, em->start,
1403 em->start + em->len - 1, 0);
1404 }
1405 type = BTRFS_ORDERED_PREALLOC;
1406 } else {
1407 type = BTRFS_ORDERED_NOCOW;
1408 }
80ff3856
YZ		 1409
1410 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052 1411 num_bytes, num_bytes, type);
79787eaa 1412 BUG_ON(ret); /* -ENOMEM */
771ed689 1413
efa56464
YZ		 1414 if (root->root_key.objectid ==
1415 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1416 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1417 num_bytes);
79787eaa
JM		 1418 if (ret) {
1419 btrfs_abort_transaction(trans, root, ret);
1420 goto error;
1421 }
efa56464
YZ		 1422 }
1423
d899e052 1424 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
a791e35e
CM		 1425 cur_offset, cur_offset + num_bytes - 1,
1426 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1427 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1428 EXTENT_SET_PRIVATE2);
80ff3856
YZ		 1429 cur_offset = extent_end;
1430 if (cur_offset > end)
1431 break;
be20aa9d 1432 }
b3b4aa74 1433 btrfs_release_path(path);
80ff3856 1434
17ca04af 1435 if (cur_offset <= end && cow_start == (u64)-1) {
80ff3856 1436 cow_start = cur_offset;
17ca04af
JB		 1437 cur_offset = end;
1438 }
1439
80ff3856 1440 if (cow_start != (u64)-1) {
b7d5b0a8
MX		 1441 ret = __cow_file_range(trans, inode, root, locked_page,
1442 cow_start, end,
1443 page_started, nr_written, 1);
79787eaa
JM		 1444 if (ret) {
1445 btrfs_abort_transaction(trans, root, ret);
1446 goto error;
1447 }
80ff3856
YZ		 1448 }
1449
79787eaa 1450error:
a698d075 1451 err = btrfs_end_transaction(trans, root);
79787eaa
JM		 1452 if (!ret)
1453 ret = err;
1454
17ca04af
JB		 1455 if (ret && cur_offset < end)
1456 extent_clear_unlock_delalloc(inode,
1457 &BTRFS_I(inode)->io_tree,
1458 cur_offset, end, locked_page,
1459 EXTENT_CLEAR_UNLOCK_PAGE |
1460 EXTENT_CLEAR_UNLOCK |
1461 EXTENT_CLEAR_DELALLOC |
1462 EXTENT_CLEAR_DIRTY |
1463 EXTENT_SET_WRITEBACK |
1464 EXTENT_END_WRITEBACK);
1465
7ea394f1 1466 btrfs_free_path(path);
79787eaa 1467 return ret;
be20aa9d
CM		 1468}
1469
d352ac68
CM		 1470/*
1471 * extent_io.c call back to do delayed allocation processing
1472 */
c8b97818 1473static int run_delalloc_range(struct inode *inode, struct page *locked_page,
771ed689
CM		 1474 u64 start, u64 end, int *page_started,
1475 unsigned long *nr_written)
be20aa9d 1476{
be20aa9d 1477 int ret;
7f366cfe 1478 struct btrfs_root *root = BTRFS_I(inode)->root;
a2135011 1479
7ddf5a42 1480 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
c8b97818 1481 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1482 page_started, 1, nr_written);
7ddf5a42 1483 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
d899e052 1484 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1485 page_started, 0, nr_written);
7ddf5a42
JB		 1486 } else if (!btrfs_test_opt(root, COMPRESS) &&
1487 !(BTRFS_I(inode)->force_compress) &&
1488 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
7f366cfe
CM		 1489 ret = cow_file_range(inode, locked_page, start, end,
1490 page_started, nr_written, 1);
7ddf5a42
JB		 1491 } else {
1492 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1493 &BTRFS_I(inode)->runtime_flags);
771ed689 1494 ret = cow_file_range_async(inode, locked_page, start, end,
d397712b 1495 page_started, nr_written);
7ddf5a42 1496 }
b888db2b
CM		 1497 return ret;
1498}
1499
1bf85046
JM		 1500static void btrfs_split_extent_hook(struct inode *inode,
1501 struct extent_state *orig, u64 split)
9ed74f2d 1502{
0ca1f7ce 1503 /* not delalloc, ignore it */
9ed74f2d 1504 if (!(orig->state & EXTENT_DELALLOC))
1bf85046 1505 return;
9ed74f2d 1506
9e0baf60
JB		 1507 spin_lock(&BTRFS_I(inode)->lock);
1508 BTRFS_I(inode)->outstanding_extents++;
1509 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1510}
1511
1512/*
1513 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1514 * extents so we can keep track of new extents that are just merged onto old
1515 * extents, such as when we are doing sequential writes, so we can properly
1516 * account for the metadata space we'll need.
1517 */
1bf85046
JM		 1518static void btrfs_merge_extent_hook(struct inode *inode,
1519 struct extent_state *new,
1520 struct extent_state *other)
9ed74f2d 1521{
9ed74f2d
JB		 1522 /* not delalloc, ignore it */
1523 if (!(other->state & EXTENT_DELALLOC))
1bf85046 1524 return;
9ed74f2d 1525
9e0baf60
JB		 1526 spin_lock(&BTRFS_I(inode)->lock);
1527 BTRFS_I(inode)->outstanding_extents--;
1528 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1529}
1530
eb73c1b7
MX		 1531static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1532 struct inode *inode)
1533{
1534 spin_lock(&root->delalloc_lock);
1535 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1536 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1537 &root->delalloc_inodes);
1538 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1539 &BTRFS_I(inode)->runtime_flags);
1540 root->nr_delalloc_inodes++;
1541 if (root->nr_delalloc_inodes == 1) {
1542 spin_lock(&root->fs_info->delalloc_root_lock);
1543 BUG_ON(!list_empty(&root->delalloc_root));
1544 list_add_tail(&root->delalloc_root,
1545 &root->fs_info->delalloc_roots);
1546 spin_unlock(&root->fs_info->delalloc_root_lock);
1547 }
1548 }
1549 spin_unlock(&root->delalloc_lock);
1550}
1551
1552static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1553 struct inode *inode)
1554{
1555 spin_lock(&root->delalloc_lock);
1556 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1557 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1558 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1559 &BTRFS_I(inode)->runtime_flags);
1560 root->nr_delalloc_inodes--;
1561 if (!root->nr_delalloc_inodes) {
1562 spin_lock(&root->fs_info->delalloc_root_lock);
1563 BUG_ON(list_empty(&root->delalloc_root));
1564 list_del_init(&root->delalloc_root);
1565 spin_unlock(&root->fs_info->delalloc_root_lock);
1566 }
1567 }
1568 spin_unlock(&root->delalloc_lock);
1569}
1570
d352ac68
CM		 1571/*
1572 * extent_io.c set_bit_hook, used to track delayed allocation
1573 * bytes in this file, and to maintain the list of inodes that
1574 * have pending delalloc work to be done.
1575 */
1bf85046 1576static void btrfs_set_bit_hook(struct inode *inode,
41074888 1577 struct extent_state *state, unsigned long *bits)
291d673e 1578{
9ed74f2d 1579
75eff68e
CM		 1580 /*
1581 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1582 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1583 * bit, which is only set or cleared with irqs on
1584 */
0ca1f7ce 1585 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1586 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1587 u64 len = state->end + 1 - state->start;
83eea1f1 1588 bool do_list = !btrfs_is_free_space_inode(inode);
9ed74f2d 1589
9e0baf60 1590 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1591 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1592 } else {
1593 spin_lock(&BTRFS_I(inode)->lock);
1594 BTRFS_I(inode)->outstanding_extents++;
1595 spin_unlock(&BTRFS_I(inode)->lock);
1596 }
287a0ab9 1597
963d678b
MX		 1598 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1599 root->fs_info->delalloc_batch);
df0af1a5 1600 spin_lock(&BTRFS_I(inode)->lock);
0ca1f7ce 1601 BTRFS_I(inode)->delalloc_bytes += len;
df0af1a5 1602 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
eb73c1b7
MX		 1603 &BTRFS_I(inode)->runtime_flags))
1604 btrfs_add_delalloc_inodes(root, inode);
df0af1a5 1605 spin_unlock(&BTRFS_I(inode)->lock);
291d673e 1606 }
291d673e
CM		 1607}
1608
d352ac68
CM		 1609/*
1610 * extent_io.c clear_bit_hook, see set_bit_hook for why
1611 */
1bf85046 1612static void btrfs_clear_bit_hook(struct inode *inode,
41074888
DS		 1613 struct extent_state *state,
1614 unsigned long *bits)
291d673e 1615{
75eff68e
CM		 1616 /*
1617 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1618 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1619 * bit, which is only set or cleared with irqs on
1620 */
0ca1f7ce 1621 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1622 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1623 u64 len = state->end + 1 - state->start;
83eea1f1 1624 bool do_list = !btrfs_is_free_space_inode(inode);
bcbfce8a 1625
9e0baf60 1626 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1627 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1628 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1629 spin_lock(&BTRFS_I(inode)->lock);
1630 BTRFS_I(inode)->outstanding_extents--;
1631 spin_unlock(&BTRFS_I(inode)->lock);
1632 }
0ca1f7ce
YZ		 1633
1634 if (*bits & EXTENT_DO_ACCOUNTING)
1635 btrfs_delalloc_release_metadata(inode, len);
1636
0cb59c99
JB		 1637 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1638 && do_list)
0ca1f7ce 1639 btrfs_free_reserved_data_space(inode, len);
9ed74f2d 1640
963d678b
MX		 1641 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1642 root->fs_info->delalloc_batch);
df0af1a5 1643 spin_lock(&BTRFS_I(inode)->lock);
0ca1f7ce 1644 BTRFS_I(inode)->delalloc_bytes -= len;
0cb59c99 1645 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
df0af1a5 1646 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
eb73c1b7
MX		 1647 &BTRFS_I(inode)->runtime_flags))
1648 btrfs_del_delalloc_inode(root, inode);
df0af1a5 1649 spin_unlock(&BTRFS_I(inode)->lock);
291d673e 1650 }
291d673e
CM		 1651}
1652
d352ac68
CM		 1653/*
1654 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1655 * we don't create bios that span stripes or chunks
1656 */
64a16701 1657int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
c8b97818
CM		 1658 size_t size, struct bio *bio,
1659 unsigned long bio_flags)
239b14b3
CM		 1660{
1661 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
a62b9401 1662 u64 logical = (u64)bio->bi_sector << 9;
239b14b3
CM		 1663 u64 length = 0;
1664 u64 map_length;
239b14b3
CM		 1665 int ret;
1666
771ed689
CM		 1667 if (bio_flags & EXTENT_BIO_COMPRESSED)
1668 return 0;
1669
f2d8d74d 1670 length = bio->bi_size;
239b14b3 1671 map_length = length;
64a16701 1672 ret = btrfs_map_block(root->fs_info, rw, logical,
f188591e 1673 &map_length, NULL, 0);
3ec706c8 1674 /* Will always return 0 with map_multi == NULL */
3444a972 1675 BUG_ON(ret < 0);
d397712b 1676 if (map_length < length + size)
239b14b3 1677 return 1;
3444a972 1678 return 0;
239b14b3
CM		 1679}
1680
d352ac68
CM		 1681/*
1682 * in order to insert checksums into the metadata in large chunks,
1683 * we wait until bio submission time. All the pages in the bio are
1684 * checksummed and sums are attached onto the ordered extent record.
1685 *
1686 * At IO completion time the cums attached on the ordered extent record
1687 * are inserted into the btree
1688 */
d397712b
CM		 1689static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1690 struct bio *bio, int mirror_num,
eaf25d93
CM		 1691 unsigned long bio_flags,
1692 u64 bio_offset)
065631f6 1693{
065631f6 1694 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 1695 int ret = 0;
e015640f 1696
d20f7043 1697 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
79787eaa 1698 BUG_ON(ret); /* -ENOMEM */
4a69a410
CM		 1699 return 0;
1700}
e015640f 1701
4a69a410
CM		 1702/*
1703 * in order to insert checksums into the metadata in large chunks,
1704 * we wait until bio submission time. All the pages in the bio are
1705 * checksummed and sums are attached onto the ordered extent record.
1706 *
1707 * At IO completion time the cums attached on the ordered extent record
1708 * are inserted into the btree
1709 */
b2950863 1710static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1711 int mirror_num, unsigned long bio_flags,
1712 u64 bio_offset)
4a69a410
CM		 1713{
1714 struct btrfs_root *root = BTRFS_I(inode)->root;
61891923
SB		 1715 int ret;
1716
1717 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1718 if (ret)
1719 bio_endio(bio, ret);
1720 return ret;
44b8bd7e
CM		 1721}
1722
d352ac68 1723/*
cad321ad
CM		 1724 * extent_io.c submission hook. This does the right thing for csum calculation
1725 * on write, or reading the csums from the tree before a read
d352ac68 1726 */
b2950863 1727static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1728 int mirror_num, unsigned long bio_flags,
1729 u64 bio_offset)
44b8bd7e
CM		 1730{
1731 struct btrfs_root *root = BTRFS_I(inode)->root;
1732 int ret = 0;
19b9bdb0 1733 int skip_sum;
0417341e 1734 int metadata = 0;
b812ce28 1735 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
44b8bd7e 1736
6cbff00f 1737 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
cad321ad 1738
83eea1f1 1739 if (btrfs_is_free_space_inode(inode))
0417341e
JM		 1740 metadata = 2;
1741
7b6d91da 1742 if (!(rw & REQ_WRITE)) {
5fd02043
JB		 1743 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1744 if (ret)
61891923 1745 goto out;
5fd02043 1746
d20f7043 1747 if (bio_flags & EXTENT_BIO_COMPRESSED) {
61891923
SB		 1748 ret = btrfs_submit_compressed_read(inode, bio,
1749 mirror_num,
1750 bio_flags);
1751 goto out;
c2db1073
TI		 1752 } else if (!skip_sum) {
1753 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1754 if (ret)
61891923 1755 goto out;
c2db1073 1756 }
4d1b5fb4 1757 goto mapit;
b812ce28 1758 } else if (async && !skip_sum) {
17d217fe
YZ		 1759 /* csum items have already been cloned */
1760 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1761 goto mapit;
19b9bdb0 1762 /* we're doing a write, do the async checksumming */
61891923 1763 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 1764 inode, rw, bio, mirror_num,
eaf25d93
CM		 1765 bio_flags, bio_offset,
1766 __btrfs_submit_bio_start,
4a69a410 1767 __btrfs_submit_bio_done);
61891923 1768 goto out;
b812ce28
JB		 1769 } else if (!skip_sum) {
1770 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1771 if (ret)
1772 goto out;
19b9bdb0
CM		 1773 }
1774
0b86a832 1775mapit:
61891923
SB		 1776 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1777
1778out:
1779 if (ret < 0)
1780 bio_endio(bio, ret);
1781 return ret;
065631f6 1782}
6885f308 1783
d352ac68
CM		 1784/*
1785 * given a list of ordered sums record them in the inode. This happens
1786 * at IO completion time based on sums calculated at bio submission time.
1787 */
ba1da2f4 1788static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM		 1789 struct inode *inode, u64 file_offset,
1790 struct list_head *list)
1791{
e6dcd2dc
CM		 1792 struct btrfs_ordered_sum *sum;
1793
c6e30871 1794 list_for_each_entry(sum, list, list) {
39847c4d 1795 trans->adding_csums = 1;
d20f7043
CM		 1796 btrfs_csum_file_blocks(trans,
1797 BTRFS_I(inode)->root->fs_info->csum_root, sum);
39847c4d 1798 trans->adding_csums = 0;
e6dcd2dc
CM		 1799 }
1800 return 0;
1801}
1802
2ac55d41
JB		 1803int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1804 struct extent_state **cached_state)
ea8c2819 1805{
6c1500f2 1806 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
ea8c2819 1807 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2ac55d41 1808 cached_state, GFP_NOFS);
ea8c2819
CM		 1809}
1810
d352ac68 1811/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM		 1812struct btrfs_writepage_fixup {
1813 struct page *page;
1814 struct btrfs_work work;
1815};
1816
b2950863 1817static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
247e743c
CM		 1818{
1819 struct btrfs_writepage_fixup *fixup;
1820 struct btrfs_ordered_extent *ordered;
2ac55d41 1821 struct extent_state *cached_state = NULL;
247e743c
CM		 1822 struct page *page;
1823 struct inode *inode;
1824 u64 page_start;
1825 u64 page_end;
87826df0 1826 int ret;
247e743c
CM		 1827
1828 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1829 page = fixup->page;
4a096752 1830again:
247e743c
CM		 1831 lock_page(page);
1832 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1833 ClearPageChecked(page);
1834 goto out_page;
1835 }
1836
1837 inode = page->mapping->host;
1838 page_start = page_offset(page);
1839 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1840
2ac55d41 1841 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
d0082371 1842 &cached_state);
4a096752
CM		 1843
1844 /* already ordered? We're done */
8b62b72b 1845 if (PagePrivate2(page))
247e743c 1846 goto out;
4a096752
CM		 1847
1848 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1849 if (ordered) {
2ac55d41
JB		 1850 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1851 page_end, &cached_state, GFP_NOFS);
4a096752
CM		 1852 unlock_page(page);
1853 btrfs_start_ordered_extent(inode, ordered, 1);
87826df0 1854 btrfs_put_ordered_extent(ordered);
4a096752
CM		 1855 goto again;
1856 }
247e743c 1857
87826df0
JM		 1858 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1859 if (ret) {
1860 mapping_set_error(page->mapping, ret);
1861 end_extent_writepage(page, ret, page_start, page_end);
1862 ClearPageChecked(page);
1863 goto out;
1864 }
1865
2ac55d41 1866 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
247e743c 1867 ClearPageChecked(page);
87826df0 1868 set_page_dirty(page);
247e743c 1869out:
2ac55d41
JB		 1870 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1871 &cached_state, GFP_NOFS);
247e743c
CM		 1872out_page:
1873 unlock_page(page);
1874 page_cache_release(page);
b897abec 1875 kfree(fixup);
247e743c
CM		 1876}
1877
1878/*
1879 * There are a few paths in the higher layers of the kernel that directly
1880 * set the page dirty bit without asking the filesystem if it is a
1881 * good idea. This causes problems because we want to make sure COW
1882 * properly happens and the data=ordered rules are followed.
1883 *
c8b97818 1884 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 1885 * hasn't been properly setup for IO. We kick off an async process
1886 * to fix it up. The async helper will wait for ordered extents, set
1887 * the delalloc bit and make it safe to write the page.
1888 */
b2950863 1889static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
247e743c
CM		 1890{
1891 struct inode *inode = page->mapping->host;
1892 struct btrfs_writepage_fixup *fixup;
1893 struct btrfs_root *root = BTRFS_I(inode)->root;
247e743c 1894
8b62b72b
CM		 1895 /* this page is properly in the ordered list */
1896 if (TestClearPagePrivate2(page))
247e743c
CM		 1897 return 0;
1898
1899 if (PageChecked(page))
1900 return -EAGAIN;
1901
1902 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1903 if (!fixup)
1904 return -EAGAIN;
f421950f 1905
247e743c
CM		 1906 SetPageChecked(page);
1907 page_cache_get(page);
1908 fixup->work.func = btrfs_writepage_fixup_worker;
1909 fixup->page = page;
1910 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
87826df0 1911 return -EBUSY;
247e743c
CM		 1912}
1913
d899e052
YZ		 1914static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1915 struct inode *inode, u64 file_pos,
1916 u64 disk_bytenr, u64 disk_num_bytes,
1917 u64 num_bytes, u64 ram_bytes,
1918 u8 compression, u8 encryption,
1919 u16 other_encoding, int extent_type)
1920{
1921 struct btrfs_root *root = BTRFS_I(inode)->root;
1922 struct btrfs_file_extent_item *fi;
1923 struct btrfs_path *path;
1924 struct extent_buffer *leaf;
1925 struct btrfs_key ins;
d899e052
YZ		 1926 int ret;
1927
1928 path = btrfs_alloc_path();
d8926bb3
MF		 1929 if (!path)
1930 return -ENOMEM;
d899e052 1931
b9473439 1932 path->leave_spinning = 1;
a1ed835e
CM		 1933
1934 /*
1935 * we may be replacing one extent in the tree with another.
1936 * The new extent is pinned in the extent map, and we don't want
1937 * to drop it from the cache until it is completely in the btree.
1938 *
1939 * So, tell btrfs_drop_extents to leave this extent in the cache.
1940 * the caller is expected to unpin it and allow it to be merged
1941 * with the others.
1942 */
5dc562c5 1943 ret = btrfs_drop_extents(trans, root, inode, file_pos,
2671485d 1944 file_pos + num_bytes, 0);
79787eaa
JM		 1945 if (ret)
1946 goto out;
d899e052 1947
33345d01 1948 ins.objectid = btrfs_ino(inode);
d899e052
YZ		 1949 ins.offset = file_pos;
1950 ins.type = BTRFS_EXTENT_DATA_KEY;
1951 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
79787eaa
JM		 1952 if (ret)
1953 goto out;
d899e052
YZ		 1954 leaf = path->nodes[0];
1955 fi = btrfs_item_ptr(leaf, path->slots[0],
1956 struct btrfs_file_extent_item);
1957 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1958 btrfs_set_file_extent_type(leaf, fi, extent_type);
1959 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1960 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1961 btrfs_set_file_extent_offset(leaf, fi, 0);
1962 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1963 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1964 btrfs_set_file_extent_compression(leaf, fi, compression);
1965 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1966 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
b9473439 1967
d899e052 1968 btrfs_mark_buffer_dirty(leaf);
ce195332 1969 btrfs_release_path(path);
d899e052
YZ		 1970
1971 inode_add_bytes(inode, num_bytes);
d899e052
YZ		 1972
1973 ins.objectid = disk_bytenr;
1974 ins.offset = disk_num_bytes;
1975 ins.type = BTRFS_EXTENT_ITEM_KEY;
5d4f98a2
YZ		 1976 ret = btrfs_alloc_reserved_file_extent(trans, root,
1977 root->root_key.objectid,
33345d01 1978 btrfs_ino(inode), file_pos, &ins);
79787eaa 1979out:
d899e052 1980 btrfs_free_path(path);
b9473439 1981
79787eaa 1982 return ret;
d899e052
YZ		 1983}
1984
38c227d8
LB		 1985/* snapshot-aware defrag */
1986struct sa_defrag_extent_backref {
1987 struct rb_node node;
1988 struct old_sa_defrag_extent *old;
1989 u64 root_id;
1990 u64 inum;
1991 u64 file_pos;
1992 u64 extent_offset;
1993 u64 num_bytes;
1994 u64 generation;
1995};
1996
1997struct old_sa_defrag_extent {
1998 struct list_head list;
1999 struct new_sa_defrag_extent *new;
2000
2001 u64 extent_offset;
2002 u64 bytenr;
2003 u64 offset;
2004 u64 len;
2005 int count;
2006};
2007
2008struct new_sa_defrag_extent {
2009 struct rb_root root;
2010 struct list_head head;
2011 struct btrfs_path *path;
2012 struct inode *inode;
2013 u64 file_pos;
2014 u64 len;
2015 u64 bytenr;
2016 u64 disk_len;
2017 u8 compress_type;
2018};
2019
2020static int backref_comp(struct sa_defrag_extent_backref *b1,
2021 struct sa_defrag_extent_backref *b2)
2022{
2023 if (b1->root_id < b2->root_id)
2024 return -1;
2025 else if (b1->root_id > b2->root_id)
2026 return 1;
2027
2028 if (b1->inum < b2->inum)
2029 return -1;
2030 else if (b1->inum > b2->inum)
2031 return 1;
2032
2033 if (b1->file_pos < b2->file_pos)
2034 return -1;
2035 else if (b1->file_pos > b2->file_pos)
2036 return 1;
2037
2038 /*
2039 * [------------------------------] ===> (a range of space)
2040 * |<--->| |<---->| =============> (fs/file tree A)
2041 * |<---------------------------->| ===> (fs/file tree B)
2042 *
2043 * A range of space can refer to two file extents in one tree while
2044 * refer to only one file extent in another tree.
2045 *
2046 * So we may process a disk offset more than one time(two extents in A)
2047 * and locate at the same extent(one extent in B), then insert two same
2048 * backrefs(both refer to the extent in B).
2049 */
2050 return 0;
2051}
2052
2053static void backref_insert(struct rb_root *root,
2054 struct sa_defrag_extent_backref *backref)
2055{
2056 struct rb_node **p = &root->rb_node;
2057 struct rb_node *parent = NULL;
2058 struct sa_defrag_extent_backref *entry;
2059 int ret;
2060
2061 while (*p) {
2062 parent = *p;
2063 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2064
2065 ret = backref_comp(backref, entry);
2066 if (ret < 0)
2067 p = &(*p)->rb_left;
2068 else
2069 p = &(*p)->rb_right;
2070 }
2071
2072 rb_link_node(&backref->node, parent, p);
2073 rb_insert_color(&backref->node, root);
2074}
2075
2076/*
2077 * Note the backref might has changed, and in this case we just return 0.
2078 */
2079static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2080 void *ctx)
2081{
2082 struct btrfs_file_extent_item *extent;
2083 struct btrfs_fs_info *fs_info;
2084 struct old_sa_defrag_extent *old = ctx;
2085 struct new_sa_defrag_extent *new = old->new;
2086 struct btrfs_path *path = new->path;
2087 struct btrfs_key key;
2088 struct btrfs_root *root;
2089 struct sa_defrag_extent_backref *backref;
2090 struct extent_buffer *leaf;
2091 struct inode *inode = new->inode;
2092 int slot;
2093 int ret;
2094 u64 extent_offset;
2095 u64 num_bytes;
2096
2097 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2098 inum == btrfs_ino(inode))
2099 return 0;
2100
2101 key.objectid = root_id;
2102 key.type = BTRFS_ROOT_ITEM_KEY;
2103 key.offset = (u64)-1;
2104
2105 fs_info = BTRFS_I(inode)->root->fs_info;
2106 root = btrfs_read_fs_root_no_name(fs_info, &key);
2107 if (IS_ERR(root)) {
2108 if (PTR_ERR(root) == -ENOENT)
2109 return 0;
2110 WARN_ON(1);
2111 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2112 inum, offset, root_id);
2113 return PTR_ERR(root);
2114 }
2115
2116 key.objectid = inum;
2117 key.type = BTRFS_EXTENT_DATA_KEY;
2118 if (offset > (u64)-1 << 32)
2119 key.offset = 0;
2120 else
2121 key.offset = offset;
2122
2123 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2124 if (ret < 0) {
2125 WARN_ON(1);
2126 return ret;
2127 }
2128
2129 while (1) {
2130 cond_resched();
2131
2132 leaf = path->nodes[0];
2133 slot = path->slots[0];
2134
2135 if (slot >= btrfs_header_nritems(leaf)) {
2136 ret = btrfs_next_leaf(root, path);
2137 if (ret < 0) {
2138 goto out;
2139 } else if (ret > 0) {
2140 ret = 0;
2141 goto out;
2142 }
2143 continue;
2144 }
2145
2146 path->slots[0]++;
2147
2148 btrfs_item_key_to_cpu(leaf, &key, slot);
2149
2150 if (key.objectid > inum)
2151 goto out;
2152
2153 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2154 continue;
2155
2156 extent = btrfs_item_ptr(leaf, slot,
2157 struct btrfs_file_extent_item);
2158
2159 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2160 continue;
2161
2162 extent_offset = btrfs_file_extent_offset(leaf, extent);
2163 if (key.offset - extent_offset != offset)
2164 continue;
2165
2166 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2167 if (extent_offset >= old->extent_offset + old->offset +
2168 old->len || extent_offset + num_bytes <=
2169 old->extent_offset + old->offset)
2170 continue;
2171
2172 break;
2173 }
2174
2175 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2176 if (!backref) {
2177 ret = -ENOENT;
2178 goto out;
2179 }
2180
2181 backref->root_id = root_id;
2182 backref->inum = inum;
2183 backref->file_pos = offset + extent_offset;
2184 backref->num_bytes = num_bytes;
2185 backref->extent_offset = extent_offset;
2186 backref->generation = btrfs_file_extent_generation(leaf, extent);
2187 backref->old = old;
2188 backref_insert(&new->root, backref);
2189 old->count++;
2190out:
2191 btrfs_release_path(path);
2192 WARN_ON(ret);
2193 return ret;
2194}
2195
2196static noinline bool record_extent_backrefs(struct btrfs_path *path,
2197 struct new_sa_defrag_extent *new)
2198{
2199 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2200 struct old_sa_defrag_extent *old, *tmp;
2201 int ret;
2202
2203 new->path = path;
2204
2205 list_for_each_entry_safe(old, tmp, &new->head, list) {
2206 ret = iterate_inodes_from_logical(old->bytenr, fs_info,
2207 path, record_one_backref,
2208 old);
2209 BUG_ON(ret < 0 && ret != -ENOENT);
2210
2211 /* no backref to be processed for this extent */
2212 if (!old->count) {
2213 list_del(&old->list);
2214 kfree(old);
2215 }
2216 }
2217
2218 if (list_empty(&new->head))
2219 return false;
2220
2221 return true;
2222}
2223
2224static int relink_is_mergable(struct extent_buffer *leaf,
2225 struct btrfs_file_extent_item *fi,
2226 u64 disk_bytenr)
2227{
2228 if (btrfs_file_extent_disk_bytenr(leaf, fi) != disk_bytenr)
2229 return 0;
2230
2231 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2232 return 0;
2233
2234 if (btrfs_file_extent_compression(leaf, fi) ||
2235 btrfs_file_extent_encryption(leaf, fi) ||
2236 btrfs_file_extent_other_encoding(leaf, fi))
2237 return 0;
2238
2239 return 1;
2240}
2241
2242/*
2243 * Note the backref might has changed, and in this case we just return 0.
2244 */
2245static noinline int relink_extent_backref(struct btrfs_path *path,
2246 struct sa_defrag_extent_backref *prev,
2247 struct sa_defrag_extent_backref *backref)
2248{
2249 struct btrfs_file_extent_item *extent;
2250 struct btrfs_file_extent_item *item;
2251 struct btrfs_ordered_extent *ordered;
2252 struct btrfs_trans_handle *trans;
2253 struct btrfs_fs_info *fs_info;
2254 struct btrfs_root *root;
2255 struct btrfs_key key;
2256 struct extent_buffer *leaf;
2257 struct old_sa_defrag_extent *old = backref->old;
2258 struct new_sa_defrag_extent *new = old->new;
2259 struct inode *src_inode = new->inode;
2260 struct inode *inode;
2261 struct extent_state *cached = NULL;
2262 int ret = 0;
2263 u64 start;
2264 u64 len;
2265 u64 lock_start;
2266 u64 lock_end;
2267 bool merge = false;
2268 int index;
2269
2270 if (prev && prev->root_id == backref->root_id &&
2271 prev->inum == backref->inum &&
2272 prev->file_pos + prev->num_bytes == backref->file_pos)
2273 merge = true;
2274
2275 /* step 1: get root */
2276 key.objectid = backref->root_id;
2277 key.type = BTRFS_ROOT_ITEM_KEY;
2278 key.offset = (u64)-1;
2279
2280 fs_info = BTRFS_I(src_inode)->root->fs_info;
2281 index = srcu_read_lock(&fs_info->subvol_srcu);
2282
2283 root = btrfs_read_fs_root_no_name(fs_info, &key);
2284 if (IS_ERR(root)) {
2285 srcu_read_unlock(&fs_info->subvol_srcu, index);
2286 if (PTR_ERR(root) == -ENOENT)
2287 return 0;
2288 return PTR_ERR(root);
2289 }
38c227d8
LB		 2290
2291 /* step 2: get inode */
2292 key.objectid = backref->inum;
2293 key.type = BTRFS_INODE_ITEM_KEY;
2294 key.offset = 0;
2295
2296 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2297 if (IS_ERR(inode)) {
2298 srcu_read_unlock(&fs_info->subvol_srcu, index);
2299 return 0;
2300 }
2301
2302 srcu_read_unlock(&fs_info->subvol_srcu, index);
2303
2304 /* step 3: relink backref */
2305 lock_start = backref->file_pos;
2306 lock_end = backref->file_pos + backref->num_bytes - 1;
2307 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2308 0, &cached);
2309
2310 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2311 if (ordered) {
2312 btrfs_put_ordered_extent(ordered);
2313 goto out_unlock;
2314 }
2315
2316 trans = btrfs_join_transaction(root);
2317 if (IS_ERR(trans)) {
2318 ret = PTR_ERR(trans);
2319 goto out_unlock;
2320 }
2321
2322 key.objectid = backref->inum;
2323 key.type = BTRFS_EXTENT_DATA_KEY;
2324 key.offset = backref->file_pos;
2325
2326 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2327 if (ret < 0) {
2328 goto out_free_path;
2329 } else if (ret > 0) {
2330 ret = 0;
2331 goto out_free_path;
2332 }
2333
2334 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2335 struct btrfs_file_extent_item);
2336
2337 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2338 backref->generation)
2339 goto out_free_path;
2340
2341 btrfs_release_path(path);
2342
2343 start = backref->file_pos;
2344 if (backref->extent_offset < old->extent_offset + old->offset)
2345 start += old->extent_offset + old->offset -
2346 backref->extent_offset;
2347
2348 len = min(backref->extent_offset + backref->num_bytes,
2349 old->extent_offset + old->offset + old->len);
2350 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2351
2352 ret = btrfs_drop_extents(trans, root, inode, start,
2353 start + len, 1);
2354 if (ret)
2355 goto out_free_path;
2356again:
2357 key.objectid = btrfs_ino(inode);
2358 key.type = BTRFS_EXTENT_DATA_KEY;
2359 key.offset = start;
2360
a09a0a70 2361 path->leave_spinning = 1;
38c227d8
LB		 2362 if (merge) {
2363 struct btrfs_file_extent_item *fi;
2364 u64 extent_len;
2365 struct btrfs_key found_key;
2366
2367 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2368 if (ret < 0)
2369 goto out_free_path;
2370
2371 path->slots[0]--;
2372 leaf = path->nodes[0];
2373 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2374
2375 fi = btrfs_item_ptr(leaf, path->slots[0],
2376 struct btrfs_file_extent_item);
2377 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2378
2379 if (relink_is_mergable(leaf, fi, new->bytenr) &&
2380 extent_len + found_key.offset == start) {
2381 btrfs_set_file_extent_num_bytes(leaf, fi,
2382 extent_len + len);
2383 btrfs_mark_buffer_dirty(leaf);
2384 inode_add_bytes(inode, len);
2385
2386 ret = 1;
2387 goto out_free_path;
2388 } else {
2389 merge = false;
2390 btrfs_release_path(path);
2391 goto again;
2392 }
2393 }
2394
2395 ret = btrfs_insert_empty_item(trans, root, path, &key,
2396 sizeof(*extent));
2397 if (ret) {
2398 btrfs_abort_transaction(trans, root, ret);
2399 goto out_free_path;
2400 }
2401
2402 leaf = path->nodes[0];
2403 item = btrfs_item_ptr(leaf, path->slots[0],
2404 struct btrfs_file_extent_item);
2405 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2406 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2407 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2408 btrfs_set_file_extent_num_bytes(leaf, item, len);
2409 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2410 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2411 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2412 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2413 btrfs_set_file_extent_encryption(leaf, item, 0);
2414 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2415
2416 btrfs_mark_buffer_dirty(leaf);
2417 inode_add_bytes(inode, len);
a09a0a70 2418 btrfs_release_path(path);
38c227d8
LB		 2419
2420 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2421 new->disk_len, 0,
2422 backref->root_id, backref->inum,
2423 new->file_pos, 0); /* start - extent_offset */
2424 if (ret) {
2425 btrfs_abort_transaction(trans, root, ret);
2426 goto out_free_path;
2427 }
2428
2429 ret = 1;
2430out_free_path:
2431 btrfs_release_path(path);
a09a0a70 2432 path->leave_spinning = 0;
38c227d8
LB		 2433 btrfs_end_transaction(trans, root);
2434out_unlock:
2435 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2436 &cached, GFP_NOFS);
2437 iput(inode);
2438 return ret;
2439}
2440
2441static void relink_file_extents(struct new_sa_defrag_extent *new)
2442{
2443 struct btrfs_path *path;
2444 struct old_sa_defrag_extent *old, *tmp;
2445 struct sa_defrag_extent_backref *backref;
2446 struct sa_defrag_extent_backref *prev = NULL;
2447 struct inode *inode;
2448 struct btrfs_root *root;
2449 struct rb_node *node;
2450 int ret;
2451
2452 inode = new->inode;
2453 root = BTRFS_I(inode)->root;
2454
2455 path = btrfs_alloc_path();
2456 if (!path)
2457 return;
2458
2459 if (!record_extent_backrefs(path, new)) {
2460 btrfs_free_path(path);
2461 goto out;
2462 }
2463 btrfs_release_path(path);
2464
2465 while (1) {
2466 node = rb_first(&new->root);
2467 if (!node)
2468 break;
2469 rb_erase(node, &new->root);
2470
2471 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2472
2473 ret = relink_extent_backref(path, prev, backref);
2474 WARN_ON(ret < 0);
2475
2476 kfree(prev);
2477
2478 if (ret == 1)
2479 prev = backref;
2480 else
2481 prev = NULL;
2482 cond_resched();
2483 }
2484 kfree(prev);
2485
2486 btrfs_free_path(path);
2487
2488 list_for_each_entry_safe(old, tmp, &new->head, list) {
2489 list_del(&old->list);
2490 kfree(old);
2491 }
2492out:
2493 atomic_dec(&root->fs_info->defrag_running);
2494 wake_up(&root->fs_info->transaction_wait);
2495
2496 kfree(new);
2497}
2498
2499static struct new_sa_defrag_extent *
2500record_old_file_extents(struct inode *inode,
2501 struct btrfs_ordered_extent *ordered)
2502{
2503 struct btrfs_root *root = BTRFS_I(inode)->root;
2504 struct btrfs_path *path;
2505 struct btrfs_key key;
2506 struct old_sa_defrag_extent *old, *tmp;
2507 struct new_sa_defrag_extent *new;
2508 int ret;
2509
2510 new = kmalloc(sizeof(*new), GFP_NOFS);
2511 if (!new)
2512 return NULL;
2513
2514 new->inode = inode;
2515 new->file_pos = ordered->file_offset;
2516 new->len = ordered->len;
2517 new->bytenr = ordered->start;
2518 new->disk_len = ordered->disk_len;
2519 new->compress_type = ordered->compress_type;
2520 new->root = RB_ROOT;
2521 INIT_LIST_HEAD(&new->head);
2522
2523 path = btrfs_alloc_path();
2524 if (!path)
2525 goto out_kfree;
2526
2527 key.objectid = btrfs_ino(inode);
2528 key.type = BTRFS_EXTENT_DATA_KEY;
2529 key.offset = new->file_pos;
2530
2531 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2532 if (ret < 0)
2533 goto out_free_path;
2534 if (ret > 0 && path->slots[0] > 0)
2535 path->slots[0]--;
2536
2537 /* find out all the old extents for the file range */
2538 while (1) {
2539 struct btrfs_file_extent_item *extent;
2540 struct extent_buffer *l;
2541 int slot;
2542 u64 num_bytes;
2543 u64 offset;
2544 u64 end;
2545 u64 disk_bytenr;
2546 u64 extent_offset;
2547
2548 l = path->nodes[0];
2549 slot = path->slots[0];
2550
2551 if (slot >= btrfs_header_nritems(l)) {
2552 ret = btrfs_next_leaf(root, path);
2553 if (ret < 0)
2554 goto out_free_list;
2555 else if (ret > 0)
2556 break;
2557 continue;
2558 }
2559
2560 btrfs_item_key_to_cpu(l, &key, slot);
2561
2562 if (key.objectid != btrfs_ino(inode))
2563 break;
2564 if (key.type != BTRFS_EXTENT_DATA_KEY)
2565 break;
2566 if (key.offset >= new->file_pos + new->len)
2567 break;
2568
2569 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2570
2571 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2572 if (key.offset + num_bytes < new->file_pos)
2573 goto next;
2574
2575 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2576 if (!disk_bytenr)
2577 goto next;
2578
2579 extent_offset = btrfs_file_extent_offset(l, extent);
2580
2581 old = kmalloc(sizeof(*old), GFP_NOFS);
2582 if (!old)
2583 goto out_free_list;
2584
2585 offset = max(new->file_pos, key.offset);
2586 end = min(new->file_pos + new->len, key.offset + num_bytes);
2587
2588 old->bytenr = disk_bytenr;
2589 old->extent_offset = extent_offset;
2590 old->offset = offset - key.offset;
2591 old->len = end - offset;
2592 old->new = new;
2593 old->count = 0;
2594 list_add_tail(&old->list, &new->head);
2595next:
2596 path->slots[0]++;
2597 cond_resched();
2598 }
2599
2600 btrfs_free_path(path);
2601 atomic_inc(&root->fs_info->defrag_running);
2602
2603 return new;
2604
2605out_free_list:
2606 list_for_each_entry_safe(old, tmp, &new->head, list) {
2607 list_del(&old->list);
2608 kfree(old);
2609 }
2610out_free_path:
2611 btrfs_free_path(path);
2612out_kfree:
2613 kfree(new);
2614 return NULL;
2615}
2616
5d13a98f
CM		 2617/*
2618 * helper function for btrfs_finish_ordered_io, this
2619 * just reads in some of the csum leaves to prime them into ram
2620 * before we start the transaction. It limits the amount of btree
2621 * reads required while inside the transaction.
2622 */
d352ac68
CM		 2623/* as ordered data IO finishes, this gets called so we can finish
2624 * an ordered extent if the range of bytes in the file it covers are
2625 * fully written.
2626 */
5fd02043 2627static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
e6dcd2dc 2628{
5fd02043 2629 struct inode *inode = ordered_extent->inode;
e6dcd2dc 2630 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 2631 struct btrfs_trans_handle *trans = NULL;
e6dcd2dc 2632 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2ac55d41 2633 struct extent_state *cached_state = NULL;
38c227d8 2634 struct new_sa_defrag_extent *new = NULL;
261507a0 2635 int compress_type = 0;
e6dcd2dc 2636 int ret;
82d5902d 2637 bool nolock;
e6dcd2dc 2638
83eea1f1 2639 nolock = btrfs_is_free_space_inode(inode);
0cb59c99 2640
5fd02043
JB		 2641 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2642 ret = -EIO;
2643 goto out;
2644 }
2645
c2167754 2646 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
79787eaa 2647 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
6c760c07
JB		 2648 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2649 if (nolock)
2650 trans = btrfs_join_transaction_nolock(root);
2651 else
2652 trans = btrfs_join_transaction(root);
2653 if (IS_ERR(trans)) {
2654 ret = PTR_ERR(trans);
2655 trans = NULL;
2656 goto out;
c2167754 2657 }
6c760c07
JB		 2658 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2659 ret = btrfs_update_inode_fallback(trans, root, inode);
2660 if (ret) /* -ENOMEM or corruption */
2661 btrfs_abort_transaction(trans, root, ret);
c2167754
YZ		 2662 goto out;
2663 }
e6dcd2dc 2664
2ac55d41
JB		 2665 lock_extent_bits(io_tree, ordered_extent->file_offset,
2666 ordered_extent->file_offset + ordered_extent->len - 1,
d0082371 2667 0, &cached_state);
e6dcd2dc 2668
38c227d8
LB		 2669 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2670 ordered_extent->file_offset + ordered_extent->len - 1,
2671 EXTENT_DEFRAG, 1, cached_state);
2672 if (ret) {
2673 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2674 if (last_snapshot >= BTRFS_I(inode)->generation)
2675 /* the inode is shared */
2676 new = record_old_file_extents(inode, ordered_extent);
2677
2678 clear_extent_bit(io_tree, ordered_extent->file_offset,
2679 ordered_extent->file_offset + ordered_extent->len - 1,
2680 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2681 }
2682
0cb59c99 2683 if (nolock)
7a7eaa40 2684 trans = btrfs_join_transaction_nolock(root);
0cb59c99 2685 else
7a7eaa40 2686 trans = btrfs_join_transaction(root);
79787eaa
JM		 2687 if (IS_ERR(trans)) {
2688 ret = PTR_ERR(trans);
2689 trans = NULL;
2690 goto out_unlock;
2691 }
0ca1f7ce 2692 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
c2167754 2693
c8b97818 2694 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
261507a0 2695 compress_type = ordered_extent->compress_type;
d899e052 2696 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
261507a0 2697 BUG_ON(compress_type);
920bbbfb 2698 ret = btrfs_mark_extent_written(trans, inode,
d899e052
YZ		 2699 ordered_extent->file_offset,
2700 ordered_extent->file_offset +
2701 ordered_extent->len);
d899e052 2702 } else {
0af3d00b 2703 BUG_ON(root == root->fs_info->tree_root);
d899e052
YZ		 2704 ret = insert_reserved_file_extent(trans, inode,
2705 ordered_extent->file_offset,
2706 ordered_extent->start,
2707 ordered_extent->disk_len,
2708 ordered_extent->len,
2709 ordered_extent->len,
261507a0 2710 compress_type, 0, 0,
d899e052 2711 BTRFS_FILE_EXTENT_REG);
d899e052 2712 }
5dc562c5
JB		 2713 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2714 ordered_extent->file_offset, ordered_extent->len,
2715 trans->transid);
79787eaa
JM		 2716 if (ret < 0) {
2717 btrfs_abort_transaction(trans, root, ret);
5fd02043 2718 goto out_unlock;
79787eaa 2719 }
2ac55d41 2720
e6dcd2dc
CM		 2721 add_pending_csums(trans, inode, ordered_extent->file_offset,
2722 &ordered_extent->list);
2723
6c760c07
JB		 2724 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2725 ret = btrfs_update_inode_fallback(trans, root, inode);
2726 if (ret) { /* -ENOMEM or corruption */
2727 btrfs_abort_transaction(trans, root, ret);
2728 goto out_unlock;
1ef30be1
JB		 2729 }
2730 ret = 0;
5fd02043
JB		 2731out_unlock:
2732 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2733 ordered_extent->file_offset +
2734 ordered_extent->len - 1, &cached_state, GFP_NOFS);
c2167754 2735out:
5b0e95bf 2736 if (root != root->fs_info->tree_root)
0cb59c99 2737 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
a698d075
MX		 2738 if (trans)
2739 btrfs_end_transaction(trans, root);
0cb59c99 2740
0bec9ef5 2741 if (ret) {
5fd02043
JB		 2742 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2743 ordered_extent->file_offset +
2744 ordered_extent->len - 1, NULL, GFP_NOFS);
2745
0bec9ef5
JB		 2746 /*
2747 * If the ordered extent had an IOERR or something else went
2748 * wrong we need to return the space for this ordered extent
2749 * back to the allocator.
2750 */
2751 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2752 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2753 btrfs_free_reserved_extent(root, ordered_extent->start,
2754 ordered_extent->disk_len);
2755 }
2756
2757
5fd02043 2758 /*
8bad3c02
LB		 2759 * This needs to be done to make sure anybody waiting knows we are done
2760 * updating everything for this ordered extent.
5fd02043
JB		 2761 */
2762 btrfs_remove_ordered_extent(inode, ordered_extent);
2763
38c227d8
LB		 2764 /* for snapshot-aware defrag */
2765 if (new)
2766 relink_file_extents(new);
2767
e6dcd2dc
CM		 2768 /* once for us */
2769 btrfs_put_ordered_extent(ordered_extent);
2770 /* once for the tree */
2771 btrfs_put_ordered_extent(ordered_extent);
2772
5fd02043
JB		 2773 return ret;
2774}
2775
2776static void finish_ordered_fn(struct btrfs_work *work)
2777{
2778 struct btrfs_ordered_extent *ordered_extent;
2779 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2780 btrfs_finish_ordered_io(ordered_extent);
e6dcd2dc
CM		 2781}
2782
b2950863 2783static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM		 2784 struct extent_state *state, int uptodate)
2785{
5fd02043
JB		 2786 struct inode *inode = page->mapping->host;
2787 struct btrfs_root *root = BTRFS_I(inode)->root;
2788 struct btrfs_ordered_extent *ordered_extent = NULL;
2789 struct btrfs_workers *workers;
2790
1abe9b8a 2791 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2792
8b62b72b 2793 ClearPagePrivate2(page);
5fd02043
JB		 2794 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2795 end - start + 1, uptodate))
2796 return 0;
2797
2798 ordered_extent->work.func = finish_ordered_fn;
2799 ordered_extent->work.flags = 0;
2800
83eea1f1 2801 if (btrfs_is_free_space_inode(inode))
5fd02043
JB		 2802 workers = &root->fs_info->endio_freespace_worker;
2803 else
2804 workers = &root->fs_info->endio_write_workers;
2805 btrfs_queue_worker(workers, &ordered_extent->work);
2806
2807 return 0;
211f90e6
CM		 2808}
2809
d352ac68
CM		 2810/*
2811 * when reads are done, we need to check csums to verify the data is correct
4a54c8c1
JS		 2812 * if there's a match, we allow the bio to finish. If not, the code in
2813 * extent_io.c will try to find good copies for us.
d352ac68 2814 */
b2950863 2815static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
5cf1ab56 2816 struct extent_state *state, int mirror)
07157aac 2817{
4eee4fa4 2818 size_t offset = start - page_offset(page);
07157aac 2819 struct inode *inode = page->mapping->host;
d1310b2e 2820 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
07157aac 2821 char *kaddr;
aadfeb6e 2822 u64 private = ~(u32)0;
07157aac 2823 int ret;
ff79f819
CM		 2824 struct btrfs_root *root = BTRFS_I(inode)->root;
2825 u32 csum = ~(u32)0;
c2cf52eb
SK		 2826 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2827 DEFAULT_RATELIMIT_BURST);
d1310b2e 2828
d20f7043
CM		 2829 if (PageChecked(page)) {
2830 ClearPageChecked(page);
2831 goto good;
2832 }
6cbff00f
CH		 2833
2834 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
08d2f347 2835 goto good;
17d217fe
YZ		 2836
2837 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
9655d298 2838 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
17d217fe
YZ		 2839 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2840 GFP_NOFS);
b6cda9bc 2841 return 0;
17d217fe 2842 }
d20f7043 2843
c2e639f0 2844 if (state && state->start == start) {
70dec807
CM		 2845 private = state->private;
2846 ret = 0;
2847 } else {
2848 ret = get_state_private(io_tree, start, &private);
2849 }
7ac687d9 2850 kaddr = kmap_atomic(page);
d397712b 2851 if (ret)
07157aac 2852 goto zeroit;
d397712b 2853
b0496686 2854 csum = btrfs_csum_data(kaddr + offset, csum, end - start + 1);
ff79f819 2855 btrfs_csum_final(csum, (char *)&csum);
d397712b 2856 if (csum != private)
07157aac 2857 goto zeroit;
d397712b 2858
7ac687d9 2859 kunmap_atomic(kaddr);
d20f7043 2860good:
07157aac
CM		 2861 return 0;
2862
2863zeroit:
c2cf52eb
SK		 2864 if (__ratelimit(&_rs))
2865 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u private %llu",
2866 (unsigned long long)btrfs_ino(page->mapping->host),
2867 (unsigned long long)start, csum,
2868 (unsigned long long)private);
db94535d
CM		 2869 memset(kaddr + offset, 1, end - start + 1);
2870 flush_dcache_page(page);
7ac687d9 2871 kunmap_atomic(kaddr);
3b951516
CM		 2872 if (private == 0)
2873 return 0;
7e38326f 2874 return -EIO;
07157aac 2875}
b888db2b 2876
24bbcf04
YZ		 2877struct delayed_iput {
2878 struct list_head list;
2879 struct inode *inode;
2880};
2881
79787eaa
JM		 2882/* JDM: If this is fs-wide, why can't we add a pointer to
2883 * btrfs_inode instead and avoid the allocation? */
24bbcf04
YZ		 2884void btrfs_add_delayed_iput(struct inode *inode)
2885{
2886 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2887 struct delayed_iput *delayed;
2888
2889 if (atomic_add_unless(&inode->i_count, -1, 1))
2890 return;
2891
2892 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2893 delayed->inode = inode;
2894
2895 spin_lock(&fs_info->delayed_iput_lock);
2896 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2897 spin_unlock(&fs_info->delayed_iput_lock);
2898}
2899
2900void btrfs_run_delayed_iputs(struct btrfs_root *root)
2901{
2902 LIST_HEAD(list);
2903 struct btrfs_fs_info *fs_info = root->fs_info;
2904 struct delayed_iput *delayed;
2905 int empty;
2906
2907 spin_lock(&fs_info->delayed_iput_lock);
2908 empty = list_empty(&fs_info->delayed_iputs);
2909 spin_unlock(&fs_info->delayed_iput_lock);
2910 if (empty)
2911 return;
2912
24bbcf04
YZ		 2913 spin_lock(&fs_info->delayed_iput_lock);
2914 list_splice_init(&fs_info->delayed_iputs, &list);
2915 spin_unlock(&fs_info->delayed_iput_lock);
2916
2917 while (!list_empty(&list)) {
2918 delayed = list_entry(list.next, struct delayed_iput, list);
2919 list_del(&delayed->list);
2920 iput(delayed->inode);
2921 kfree(delayed);
2922 }
24bbcf04
YZ		 2923}
2924
d68fc57b 2925/*
42b2aa86 2926 * This is called in transaction commit time. If there are no orphan
d68fc57b
YZ		 2927 * files in the subvolume, it removes orphan item and frees block_rsv
2928 * structure.
2929 */
2930void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2931 struct btrfs_root *root)
2932{
90290e19 2933 struct btrfs_block_rsv *block_rsv;
d68fc57b
YZ		 2934 int ret;
2935
8a35d95f 2936 if (atomic_read(&root->orphan_inodes) ||
d68fc57b
YZ		 2937 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2938 return;
2939
90290e19 2940 spin_lock(&root->orphan_lock);
8a35d95f 2941 if (atomic_read(&root->orphan_inodes)) {
90290e19
JB		 2942 spin_unlock(&root->orphan_lock);
2943 return;
2944 }
2945
2946 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2947 spin_unlock(&root->orphan_lock);
2948 return;
2949 }
2950
2951 block_rsv = root->orphan_block_rsv;
2952 root->orphan_block_rsv = NULL;
2953 spin_unlock(&root->orphan_lock);
2954
d68fc57b
YZ		 2955 if (root->orphan_item_inserted &&
2956 btrfs_root_refs(&root->root_item) > 0) {
2957 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2958 root->root_key.objectid);
2959 BUG_ON(ret);
2960 root->orphan_item_inserted = 0;
2961 }
2962
90290e19
JB		 2963 if (block_rsv) {
2964 WARN_ON(block_rsv->size > 0);
2965 btrfs_free_block_rsv(root, block_rsv);
d68fc57b
YZ		 2966 }
2967}
2968
7b128766
JB		 2969/*
2970 * This creates an orphan entry for the given inode in case something goes
2971 * wrong in the middle of an unlink/truncate.
d68fc57b
YZ		 2972 *
2973 * NOTE: caller of this function should reserve 5 units of metadata for
2974 * this function.
7b128766
JB		 2975 */
2976int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2977{
2978 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 2979 struct btrfs_block_rsv *block_rsv = NULL;
2980 int reserve = 0;
2981 int insert = 0;
2982 int ret;
7b128766 2983
d68fc57b 2984 if (!root->orphan_block_rsv) {
66d8f3dd 2985 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
b532402e
TI		 2986 if (!block_rsv)
2987 return -ENOMEM;
d68fc57b 2988 }
7b128766 2989
d68fc57b
YZ		 2990 spin_lock(&root->orphan_lock);
2991 if (!root->orphan_block_rsv) {
2992 root->orphan_block_rsv = block_rsv;
2993 } else if (block_rsv) {
2994 btrfs_free_block_rsv(root, block_rsv);
2995 block_rsv = NULL;
7b128766 2996 }
7b128766 2997
8a35d95f
JB		 2998 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2999 &BTRFS_I(inode)->runtime_flags)) {
d68fc57b
YZ		 3000#if 0
3001 /*
3002 * For proper ENOSPC handling, we should do orphan
3003 * cleanup when mounting. But this introduces backward
3004 * compatibility issue.
3005 */
3006 if (!xchg(&root->orphan_item_inserted, 1))
3007 insert = 2;
3008 else
3009 insert = 1;
3010#endif
3011 insert = 1;
321f0e70 3012 atomic_inc(&root->orphan_inodes);
7b128766
JB		 3013 }
3014
72ac3c0d
JB		 3015 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3016 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3017 reserve = 1;
d68fc57b 3018 spin_unlock(&root->orphan_lock);
7b128766 3019
d68fc57b
YZ		 3020 /* grab metadata reservation from transaction handle */
3021 if (reserve) {
3022 ret = btrfs_orphan_reserve_metadata(trans, inode);
79787eaa 3023 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
d68fc57b 3024 }
7b128766 3025
d68fc57b
YZ		 3026 /* insert an orphan item to track this unlinked/truncated file */
3027 if (insert >= 1) {
33345d01 3028 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
79787eaa 3029 if (ret && ret != -EEXIST) {
8a35d95f
JB		 3030 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3031 &BTRFS_I(inode)->runtime_flags);
79787eaa
JM		 3032 btrfs_abort_transaction(trans, root, ret);
3033 return ret;
3034 }
3035 ret = 0;
d68fc57b
YZ		 3036 }
3037
3038 /* insert an orphan item to track subvolume contains orphan files */
3039 if (insert >= 2) {
3040 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3041 root->root_key.objectid);
79787eaa
JM		 3042 if (ret && ret != -EEXIST) {
3043 btrfs_abort_transaction(trans, root, ret);
3044 return ret;
3045 }
d68fc57b
YZ		 3046 }
3047 return 0;
7b128766
JB		 3048}
3049
3050/*
3051 * We have done the truncate/delete so we can go ahead and remove the orphan
3052 * item for this particular inode.
3053 */
48a3b636
ES		 3054static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3055 struct inode *inode)
7b128766
JB		 3056{
3057 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 3058 int delete_item = 0;
3059 int release_rsv = 0;
7b128766
JB		 3060 int ret = 0;
3061
d68fc57b 3062 spin_lock(&root->orphan_lock);
8a35d95f
JB		 3063 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3064 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3065 delete_item = 1;
7b128766 3066
72ac3c0d
JB		 3067 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3068 &BTRFS_I(inode)->runtime_flags))
d68fc57b 3069 release_rsv = 1;
d68fc57b 3070 spin_unlock(&root->orphan_lock);
7b128766 3071
d68fc57b 3072 if (trans && delete_item) {
33345d01 3073 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
79787eaa 3074 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
d68fc57b 3075 }
7b128766 3076
8a35d95f 3077 if (release_rsv) {
d68fc57b 3078 btrfs_orphan_release_metadata(inode);
8a35d95f
JB		 3079 atomic_dec(&root->orphan_inodes);
3080 }
7b128766 3081
d68fc57b 3082 return 0;
7b128766
JB		 3083}
3084
3085/*
3086 * this cleans up any orphans that may be left on the list from the last use
3087 * of this root.
3088 */
66b4ffd1 3089int btrfs_orphan_cleanup(struct btrfs_root *root)
7b128766
JB		 3090{
3091 struct btrfs_path *path;
3092 struct extent_buffer *leaf;
7b128766
JB		 3093 struct btrfs_key key, found_key;
3094 struct btrfs_trans_handle *trans;
3095 struct inode *inode;
8f6d7f4f 3096 u64 last_objectid = 0;
7b128766
JB		 3097 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3098
d68fc57b 3099 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
66b4ffd1 3100 return 0;
c71bf099
YZ		 3101
3102 path = btrfs_alloc_path();
66b4ffd1
JB		 3103 if (!path) {
3104 ret = -ENOMEM;
3105 goto out;
3106 }
7b128766
JB		 3107 path->reada = -1;
3108
3109 key.objectid = BTRFS_ORPHAN_OBJECTID;
3110 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3111 key.offset = (u64)-1;
3112
7b128766
JB		 3113 while (1) {
3114 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
66b4ffd1
JB		 3115 if (ret < 0)
3116 goto out;
7b128766
JB		 3117
3118 /*
3119 * if ret == 0 means we found what we were searching for, which
25985edc 3120 * is weird, but possible, so only screw with path if we didn't
7b128766
JB		 3121 * find the key and see if we have stuff that matches
3122 */
3123 if (ret > 0) {
66b4ffd1 3124 ret = 0;
7b128766
JB		 3125 if (path->slots[0] == 0)
3126 break;
3127 path->slots[0]--;
3128 }
3129
3130 /* pull out the item */
3131 leaf = path->nodes[0];
7b128766
JB		 3132 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3133
3134 /* make sure the item matches what we want */
3135 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3136 break;
3137 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3138 break;
3139
3140 /* release the path since we're done with it */
b3b4aa74 3141 btrfs_release_path(path);
7b128766
JB		 3142
3143 /*
3144 * this is where we are basically btrfs_lookup, without the
3145 * crossing root thing. we store the inode number in the
3146 * offset of the orphan item.
3147 */
8f6d7f4f
JB		 3148
3149 if (found_key.offset == last_objectid) {
c2cf52eb
SK		 3150 btrfs_err(root->fs_info,
3151 "Error removing orphan entry, stopping orphan cleanup");
8f6d7f4f
JB		 3152 ret = -EINVAL;
3153 goto out;
3154 }
3155
3156 last_objectid = found_key.offset;
3157
5d4f98a2
YZ		 3158 found_key.objectid = found_key.offset;
3159 found_key.type = BTRFS_INODE_ITEM_KEY;
3160 found_key.offset = 0;
73f73415 3161 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
a8c9e576
JB		 3162 ret = PTR_RET(inode);
3163 if (ret && ret != -ESTALE)
66b4ffd1 3164 goto out;
7b128766 3165
f8e9e0b0
AJ		 3166 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3167 struct btrfs_root *dead_root;
3168 struct btrfs_fs_info *fs_info = root->fs_info;
3169 int is_dead_root = 0;
3170
3171 /*
3172 * this is an orphan in the tree root. Currently these
3173 * could come from 2 sources:
3174 * a) a snapshot deletion in progress
3175 * b) a free space cache inode
3176 * We need to distinguish those two, as the snapshot
3177 * orphan must not get deleted.
3178 * find_dead_roots already ran before us, so if this
3179 * is a snapshot deletion, we should find the root
3180 * in the dead_roots list
3181 */
3182 spin_lock(&fs_info->trans_lock);
3183 list_for_each_entry(dead_root, &fs_info->dead_roots,
3184 root_list) {
3185 if (dead_root->root_key.objectid ==
3186 found_key.objectid) {
3187 is_dead_root = 1;
3188 break;
3189 }
3190 }
3191 spin_unlock(&fs_info->trans_lock);
3192 if (is_dead_root) {
3193 /* prevent this orphan from being found again */
3194 key.offset = found_key.objectid - 1;
3195 continue;
3196 }
3197 }
7b128766 3198 /*
a8c9e576
JB		 3199 * Inode is already gone but the orphan item is still there,
3200 * kill the orphan item.
7b128766 3201 */
a8c9e576
JB		 3202 if (ret == -ESTALE) {
3203 trans = btrfs_start_transaction(root, 1);
66b4ffd1
JB		 3204 if (IS_ERR(trans)) {
3205 ret = PTR_ERR(trans);
3206 goto out;
3207 }
c2cf52eb
SK		 3208 btrfs_debug(root->fs_info, "auto deleting %Lu",
3209 found_key.objectid);
a8c9e576
JB		 3210 ret = btrfs_del_orphan_item(trans, root,
3211 found_key.objectid);
79787eaa 3212 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
5b21f2ed 3213 btrfs_end_transaction(trans, root);
7b128766
JB		 3214 continue;
3215 }
3216
a8c9e576
JB		 3217 /*
3218 * add this inode to the orphan list so btrfs_orphan_del does
3219 * the proper thing when we hit it
3220 */
8a35d95f
JB		 3221 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3222 &BTRFS_I(inode)->runtime_flags);
925396ec 3223 atomic_inc(&root->orphan_inodes);
a8c9e576 3224
7b128766
JB		 3225 /* if we have links, this was a truncate, lets do that */
3226 if (inode->i_nlink) {
a41ad394
JB		 3227 if (!S_ISREG(inode->i_mode)) {
3228 WARN_ON(1);
3229 iput(inode);
3230 continue;
3231 }
7b128766 3232 nr_truncate++;
f3fe820c
JB		 3233
3234 /* 1 for the orphan item deletion. */
3235 trans = btrfs_start_transaction(root, 1);
3236 if (IS_ERR(trans)) {
3237 ret = PTR_ERR(trans);
3238 goto out;
3239 }
3240 ret = btrfs_orphan_add(trans, inode);
3241 btrfs_end_transaction(trans, root);
3242 if (ret)
3243 goto out;
3244
66b4ffd1 3245 ret = btrfs_truncate(inode);
4a7d0f68
JB		 3246 if (ret)
3247 btrfs_orphan_del(NULL, inode);
7b128766
JB		 3248 } else {
3249 nr_unlink++;
3250 }
3251
3252 /* this will do delete_inode and everything for us */
3253 iput(inode);
66b4ffd1
JB		 3254 if (ret)
3255 goto out;
7b128766 3256 }
3254c876
MX		 3257 /* release the path since we're done with it */
3258 btrfs_release_path(path);
3259
d68fc57b
YZ		 3260 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3261
3262 if (root->orphan_block_rsv)
3263 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3264 (u64)-1);
3265
3266 if (root->orphan_block_rsv || root->orphan_item_inserted) {
7a7eaa40 3267 trans = btrfs_join_transaction(root);
66b4ffd1
JB		 3268 if (!IS_ERR(trans))
3269 btrfs_end_transaction(trans, root);
d68fc57b 3270 }
7b128766
JB		 3271
3272 if (nr_unlink)
4884b476 3273 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
7b128766 3274 if (nr_truncate)
4884b476 3275 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
66b4ffd1
JB		 3276
3277out:
3278 if (ret)
c2cf52eb
SK		 3279 btrfs_crit(root->fs_info,
3280 "could not do orphan cleanup %d", ret);
66b4ffd1
JB		 3281 btrfs_free_path(path);
3282 return ret;
7b128766
JB		 3283}
3284
46a53cca
CM		 3285/*
3286 * very simple check to peek ahead in the leaf looking for xattrs. If we
3287 * don't find any xattrs, we know there can't be any acls.
3288 *
3289 * slot is the slot the inode is in, objectid is the objectid of the inode
3290 */
3291static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3292 int slot, u64 objectid)
3293{
3294 u32 nritems = btrfs_header_nritems(leaf);
3295 struct btrfs_key found_key;
3296 int scanned = 0;
3297
3298 slot++;
3299 while (slot < nritems) {
3300 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3301
3302 /* we found a different objectid, there must not be acls */
3303 if (found_key.objectid != objectid)
3304 return 0;
3305
3306 /* we found an xattr, assume we've got an acl */
3307 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
3308 return 1;
3309
3310 /*
3311 * we found a key greater than an xattr key, there can't
3312 * be any acls later on
3313 */
3314 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3315 return 0;
3316
3317 slot++;
3318 scanned++;
3319
3320 /*
3321 * it goes inode, inode backrefs, xattrs, extents,
3322 * so if there are a ton of hard links to an inode there can
3323 * be a lot of backrefs. Don't waste time searching too hard,
3324 * this is just an optimization
3325 */
3326 if (scanned >= 8)
3327 break;
3328 }
3329 /* we hit the end of the leaf before we found an xattr or
3330 * something larger than an xattr. We have to assume the inode
3331 * has acls
3332 */
3333 return 1;
3334}
3335
d352ac68
CM		 3336/*
3337 * read an inode from the btree into the in-memory inode
3338 */
5d4f98a2 3339static void btrfs_read_locked_inode(struct inode *inode)
39279cc3
CM		 3340{
3341 struct btrfs_path *path;
5f39d397 3342 struct extent_buffer *leaf;
39279cc3 3343 struct btrfs_inode_item *inode_item;
0b86a832 3344 struct btrfs_timespec *tspec;
39279cc3
CM		 3345 struct btrfs_root *root = BTRFS_I(inode)->root;
3346 struct btrfs_key location;
46a53cca 3347 int maybe_acls;
618e21d5 3348 u32 rdev;
39279cc3 3349 int ret;
2f7e33d4
MX		 3350 bool filled = false;
3351
3352 ret = btrfs_fill_inode(inode, &rdev);
3353 if (!ret)
3354 filled = true;
39279cc3
CM		 3355
3356 path = btrfs_alloc_path();
1748f843
MF		 3357 if (!path)
3358 goto make_bad;
3359
d90c7321 3360 path->leave_spinning = 1;
39279cc3 3361 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 3362
39279cc3 3363 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 3364 if (ret)
39279cc3 3365 goto make_bad;
39279cc3 3366
5f39d397 3367 leaf = path->nodes[0];
2f7e33d4
MX		 3368
3369 if (filled)
3370 goto cache_acl;
3371
5f39d397
CM		 3372 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3373 struct btrfs_inode_item);
5f39d397 3374 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
bfe86848 3375 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2f2f43d3
EB		 3376 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3377 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
dbe674a9 3378 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397
CM		 3379
3380 tspec = btrfs_inode_atime(inode_item);
3381 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3382 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3383
3384 tspec = btrfs_inode_mtime(inode_item);
3385 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3386 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3387
3388 tspec = btrfs_inode_ctime(inode_item);
3389 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3390 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3391
a76a3cd4 3392 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 3393 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
5dc562c5
JB		 3394 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3395
3396 /*
3397 * If we were modified in the current generation and evicted from memory
3398 * and then re-read we need to do a full sync since we don't have any
3399 * idea about which extents were modified before we were evicted from
3400 * cache.
3401 */
3402 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3403 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3404 &BTRFS_I(inode)->runtime_flags);
3405
0c4d2d95 3406 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
e02119d5 3407 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 3408 inode->i_rdev = 0;
5f39d397
CM		 3409 rdev = btrfs_inode_rdev(leaf, inode_item);
3410
aec7477b 3411 BTRFS_I(inode)->index_cnt = (u64)-1;
d2fb3437 3412 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2f7e33d4 3413cache_acl:
46a53cca
CM		 3414 /*
3415 * try to precache a NULL acl entry for files that don't have
3416 * any xattrs or acls
3417 */
33345d01
LZ		 3418 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3419 btrfs_ino(inode));
72c04902
AV		 3420 if (!maybe_acls)
3421 cache_no_acl(inode);
46a53cca 3422
39279cc3 3423 btrfs_free_path(path);
39279cc3 3424
39279cc3 3425 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 3426 case S_IFREG:
3427 inode->i_mapping->a_ops = &btrfs_aops;
04160088 3428 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 3429 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 3430 inode->i_fop = &btrfs_file_operations;
3431 inode->i_op = &btrfs_file_inode_operations;
3432 break;
3433 case S_IFDIR:
3434 inode->i_fop = &btrfs_dir_file_operations;
3435 if (root == root->fs_info->tree_root)
3436 inode->i_op = &btrfs_dir_ro_inode_operations;
3437 else
3438 inode->i_op = &btrfs_dir_inode_operations;
3439 break;
3440 case S_IFLNK:
3441 inode->i_op = &btrfs_symlink_inode_operations;
3442 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 3443 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3 3444 break;
618e21d5 3445 default:
0279b4cd 3446 inode->i_op = &btrfs_special_inode_operations;
618e21d5
JB		 3447 init_special_inode(inode, inode->i_mode, rdev);
3448 break;
39279cc3 3449 }
6cbff00f
CH		 3450
3451 btrfs_update_iflags(inode);
39279cc3
CM		 3452 return;
3453
3454make_bad:
39279cc3 3455 btrfs_free_path(path);
39279cc3
CM		 3456 make_bad_inode(inode);
3457}
3458
d352ac68
CM		 3459/*
3460 * given a leaf and an inode, copy the inode fields into the leaf
3461 */
e02119d5
CM		 3462static void fill_inode_item(struct btrfs_trans_handle *trans,
3463 struct extent_buffer *leaf,
5f39d397 3464 struct btrfs_inode_item *item,
39279cc3
CM		 3465 struct inode *inode)
3466{
51fab693
LB		 3467 struct btrfs_map_token token;
3468
3469 btrfs_init_map_token(&token);
5f39d397 3470
51fab693
LB		 3471 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3472 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3473 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3474 &token);
3475 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3476 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
5f39d397 3477
51fab693
LB		 3478 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3479 inode->i_atime.tv_sec, &token);
3480 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3481 inode->i_atime.tv_nsec, &token);
5f39d397 3482
51fab693
LB		 3483 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3484 inode->i_mtime.tv_sec, &token);
3485 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3486 inode->i_mtime.tv_nsec, &token);
5f39d397 3487
51fab693
LB		 3488 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3489 inode->i_ctime.tv_sec, &token);
3490 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3491 inode->i_ctime.tv_nsec, &token);
5f39d397 3492
51fab693
LB		 3493 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3494 &token);
3495 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3496 &token);
3497 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3498 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3499 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3500 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3501 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
39279cc3
CM		 3502}
3503
d352ac68
CM		 3504/*
3505 * copy everything in the in-memory inode into the btree.
3506 */
2115133f 3507static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
d397712b 3508 struct btrfs_root *root, struct inode *inode)
39279cc3
CM		 3509{
3510 struct btrfs_inode_item *inode_item;
3511 struct btrfs_path *path;
5f39d397 3512 struct extent_buffer *leaf;
39279cc3
CM		 3513 int ret;
3514
3515 path = btrfs_alloc_path();
16cdcec7
MX		 3516 if (!path)
3517 return -ENOMEM;
3518
b9473439 3519 path->leave_spinning = 1;
16cdcec7
MX		 3520 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3521 1);
39279cc3
CM		 3522 if (ret) {
3523 if (ret > 0)
3524 ret = -ENOENT;
3525 goto failed;
3526 }
3527
b4ce94de 3528 btrfs_unlock_up_safe(path, 1);
5f39d397
CM		 3529 leaf = path->nodes[0];
3530 inode_item = btrfs_item_ptr(leaf, path->slots[0],
16cdcec7 3531 struct btrfs_inode_item);
39279cc3 3532
e02119d5 3533 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 3534 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 3535 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 3536 ret = 0;
3537failed:
39279cc3
CM		 3538 btrfs_free_path(path);
3539 return ret;
3540}
3541
2115133f
CM		 3542/*
3543 * copy everything in the in-memory inode into the btree.
3544 */
3545noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3546 struct btrfs_root *root, struct inode *inode)
3547{
3548 int ret;
3549
3550 /*
3551 * If the inode is a free space inode, we can deadlock during commit
3552 * if we put it into the delayed code.
3553 *
3554 * The data relocation inode should also be directly updated
3555 * without delay
3556 */
83eea1f1 3557 if (!btrfs_is_free_space_inode(inode)
2115133f 3558 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
8ea05e3a
AB		 3559 btrfs_update_root_times(trans, root);
3560
2115133f
CM		 3561 ret = btrfs_delayed_update_inode(trans, root, inode);
3562 if (!ret)
3563 btrfs_set_inode_last_trans(trans, inode);
3564 return ret;
3565 }
3566
3567 return btrfs_update_inode_item(trans, root, inode);
3568}
3569
be6aef60
JB		 3570noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3571 struct btrfs_root *root,
3572 struct inode *inode)
2115133f
CM		 3573{
3574 int ret;
3575
3576 ret = btrfs_update_inode(trans, root, inode);
3577 if (ret == -ENOSPC)
3578 return btrfs_update_inode_item(trans, root, inode);
3579 return ret;
3580}
3581
d352ac68
CM		 3582/*
3583 * unlink helper that gets used here in inode.c and in the tree logging
3584 * recovery code. It remove a link in a directory with a given name, and
3585 * also drops the back refs in the inode to the directory
3586 */
92986796
AV		 3587static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3588 struct btrfs_root *root,
3589 struct inode *dir, struct inode *inode,
3590 const char *name, int name_len)
39279cc3
CM		 3591{
3592 struct btrfs_path *path;
39279cc3 3593 int ret = 0;
5f39d397 3594 struct extent_buffer *leaf;
39279cc3 3595 struct btrfs_dir_item *di;
5f39d397 3596 struct btrfs_key key;
aec7477b 3597 u64 index;
33345d01
LZ		 3598 u64 ino = btrfs_ino(inode);
3599 u64 dir_ino = btrfs_ino(dir);
39279cc3
CM		 3600
3601 path = btrfs_alloc_path();
54aa1f4d
CM		 3602 if (!path) {
3603 ret = -ENOMEM;
554233a6 3604 goto out;
54aa1f4d
CM		 3605 }
3606
b9473439 3607 path->leave_spinning = 1;
33345d01 3608 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
39279cc3
CM		 3609 name, name_len, -1);
3610 if (IS_ERR(di)) {
3611 ret = PTR_ERR(di);
3612 goto err;
3613 }
3614 if (!di) {
3615 ret = -ENOENT;
3616 goto err;
3617 }
5f39d397
CM		 3618 leaf = path->nodes[0];
3619 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 3620 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 3621 if (ret)
3622 goto err;
b3b4aa74 3623 btrfs_release_path(path);
39279cc3 3624
33345d01
LZ		 3625 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3626 dir_ino, &index);
aec7477b 3627 if (ret) {
c2cf52eb
SK		 3628 btrfs_info(root->fs_info,
3629 "failed to delete reference to %.*s, inode %llu parent %llu",
3630 name_len, name,
3631 (unsigned long long)ino, (unsigned long long)dir_ino);
79787eaa 3632 btrfs_abort_transaction(trans, root, ret);
aec7477b
JB		 3633 goto err;
3634 }
3635
16cdcec7 3636 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 3637 if (ret) {
3638 btrfs_abort_transaction(trans, root, ret);
39279cc3 3639 goto err;
79787eaa 3640 }
39279cc3 3641
e02119d5 3642 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
33345d01 3643 inode, dir_ino);
79787eaa
JM		 3644 if (ret != 0 && ret != -ENOENT) {
3645 btrfs_abort_transaction(trans, root, ret);
3646 goto err;
3647 }
e02119d5
CM		 3648
3649 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3650 dir, index);
6418c961
CM		 3651 if (ret == -ENOENT)
3652 ret = 0;
d4e3991b
ZB		 3653 else if (ret)
3654 btrfs_abort_transaction(trans, root, ret);
39279cc3
CM		 3655err:
3656 btrfs_free_path(path);
e02119d5
CM		 3657 if (ret)
3658 goto out;
3659
3660 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95
JB		 3661 inode_inc_iversion(inode);
3662 inode_inc_iversion(dir);
e02119d5 3663 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
b9959295 3664 ret = btrfs_update_inode(trans, root, dir);
e02119d5 3665out:
39279cc3
CM		 3666 return ret;
3667}
3668
92986796
AV		 3669int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3670 struct btrfs_root *root,
3671 struct inode *dir, struct inode *inode,
3672 const char *name, int name_len)
3673{
3674 int ret;
3675 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3676 if (!ret) {
3677 btrfs_drop_nlink(inode);
3678 ret = btrfs_update_inode(trans, root, inode);
3679 }
3680 return ret;
3681}
3682
3683
a22285a6
YZ		 3684/* helper to check if there is any shared block in the path */
3685static int check_path_shared(struct btrfs_root *root,
3686 struct btrfs_path *path)
39279cc3 3687{
a22285a6
YZ		 3688 struct extent_buffer *eb;
3689 int level;
0e4dcbef 3690 u64 refs = 1;
5df6a9f6 3691
a22285a6 3692 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
dedefd72
JB		 3693 int ret;
3694
a22285a6
YZ		 3695 if (!path->nodes[level])
3696 break;
3697 eb = path->nodes[level];
3698 if (!btrfs_block_can_be_shared(root, eb))
3699 continue;
3173a18f 3700 ret = btrfs_lookup_extent_info(NULL, root, eb->start, level, 1,
a22285a6
YZ		 3701 &refs, NULL);
3702 if (refs > 1)
3703 return 1;
5df6a9f6 3704 }
dedefd72 3705 return 0;
39279cc3
CM		 3706}
3707
a22285a6
YZ		 3708/*
3709 * helper to start transaction for unlink and rmdir.
3710 *
3711 * unlink and rmdir are special in btrfs, they do not always free space.
3712 * so in enospc case, we should make sure they will free space before
3713 * allowing them to use the global metadata reservation.
3714 */
3715static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
3716 struct dentry *dentry)
4df27c4d 3717{
39279cc3 3718 struct btrfs_trans_handle *trans;
a22285a6 3719 struct btrfs_root *root = BTRFS_I(dir)->root;
4df27c4d 3720 struct btrfs_path *path;
4df27c4d 3721 struct btrfs_dir_item *di;
7b128766 3722 struct inode *inode = dentry->d_inode;
4df27c4d 3723 u64 index;
a22285a6
YZ		 3724 int check_link = 1;
3725 int err = -ENOSPC;
4df27c4d 3726 int ret;
33345d01
LZ		 3727 u64 ino = btrfs_ino(inode);
3728 u64 dir_ino = btrfs_ino(dir);
4df27c4d 3729
e70bea5f
JB		 3730 /*
3731 * 1 for the possible orphan item
3732 * 1 for the dir item
3733 * 1 for the dir index
3734 * 1 for the inode ref
e70bea5f
JB		 3735 * 1 for the inode
3736 */
6e137ed3 3737 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 3738 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3739 return trans;
4df27c4d 3740
33345d01 3741 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
a22285a6 3742 return ERR_PTR(-ENOSPC);
4df27c4d 3743
a22285a6
YZ		 3744 /* check if there is someone else holds reference */
3745 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
3746 return ERR_PTR(-ENOSPC);
4df27c4d 3747
a22285a6
YZ		 3748 if (atomic_read(&inode->i_count) > 2)
3749 return ERR_PTR(-ENOSPC);
4df27c4d 3750
a22285a6
YZ		 3751 if (xchg(&root->fs_info->enospc_unlink, 1))
3752 return ERR_PTR(-ENOSPC);
3753
3754 path = btrfs_alloc_path();
3755 if (!path) {
3756 root->fs_info->enospc_unlink = 0;
3757 return ERR_PTR(-ENOMEM);
4df27c4d
YZ		 3758 }
3759
3880a1b4
JB		 3760 /* 1 for the orphan item */
3761 trans = btrfs_start_transaction(root, 1);
5df6a9f6 3762 if (IS_ERR(trans)) {
a22285a6
YZ		 3763 btrfs_free_path(path);
3764 root->fs_info->enospc_unlink = 0;
3765 return trans;
3766 }
4df27c4d 3767
a22285a6
YZ		 3768 path->skip_locking = 1;
3769 path->search_commit_root = 1;
4df27c4d 3770
a22285a6
YZ		 3771 ret = btrfs_lookup_inode(trans, root, path,
3772 &BTRFS_I(dir)->location, 0);
3773 if (ret < 0) {
3774 err = ret;
3775 goto out;
3776 }
3777 if (ret == 0) {
3778 if (check_path_shared(root, path))
3779 goto out;
3780 } else {
3781 check_link = 0;
5df6a9f6 3782 }
b3b4aa74 3783 btrfs_release_path(path);
a22285a6
YZ		 3784
3785 ret = btrfs_lookup_inode(trans, root, path,
3786 &BTRFS_I(inode)->location, 0);
3787 if (ret < 0) {
3788 err = ret;
3789 goto out;
3790 }
3791 if (ret == 0) {
3792 if (check_path_shared(root, path))
3793 goto out;
3794 } else {
3795 check_link = 0;
3796 }
b3b4aa74 3797 btrfs_release_path(path);
a22285a6
YZ		 3798
3799 if (ret == 0 && S_ISREG(inode->i_mode)) {
3800 ret = btrfs_lookup_file_extent(trans, root, path,
33345d01 3801 ino, (u64)-1, 0);
a22285a6
YZ		 3802 if (ret < 0) {
3803 err = ret;
3804 goto out;
3805 }
79787eaa 3806 BUG_ON(ret == 0); /* Corruption */
a22285a6
YZ		 3807 if (check_path_shared(root, path))
3808 goto out;
b3b4aa74 3809 btrfs_release_path(path);
a22285a6
YZ		 3810 }
3811
3812 if (!check_link) {
3813 err = 0;
3814 goto out;
3815 }
3816
33345d01 3817 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
a22285a6
YZ		 3818 dentry->d_name.name, dentry->d_name.len, 0);
3819 if (IS_ERR(di)) {
3820 err = PTR_ERR(di);
3821 goto out;
3822 }
3823 if (di) {
3824 if (check_path_shared(root, path))
3825 goto out;
3826 } else {
3827 err = 0;
3828 goto out;
3829 }
b3b4aa74 3830 btrfs_release_path(path);
a22285a6 3831
f186373f
MF		 3832 ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3833 dentry->d_name.len, ino, dir_ino, 0,
3834 &index);
3835 if (ret) {
3836 err = ret;
a22285a6
YZ		 3837 goto out;
3838 }
f186373f 3839
a22285a6
YZ		 3840 if (check_path_shared(root, path))
3841 goto out;
f186373f 3842
b3b4aa74 3843 btrfs_release_path(path);
a22285a6 3844
16cdcec7
MX		 3845 /*
3846 * This is a commit root search, if we can lookup inode item and other
3847 * relative items in the commit root, it means the transaction of
3848 * dir/file creation has been committed, and the dir index item that we
3849 * delay to insert has also been inserted into the commit root. So
3850 * we needn't worry about the delayed insertion of the dir index item
3851 * here.
3852 */
33345d01 3853 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
a22285a6
YZ		 3854 dentry->d_name.name, dentry->d_name.len, 0);
3855 if (IS_ERR(di)) {
3856 err = PTR_ERR(di);
3857 goto out;
3858 }
3859 BUG_ON(ret == -ENOENT);
3860 if (check_path_shared(root, path))
3861 goto out;
3862
3863 err = 0;
3864out:
3865 btrfs_free_path(path);
3880a1b4
JB		 3866 /* Migrate the orphan reservation over */
3867 if (!err)
3868 err = btrfs_block_rsv_migrate(trans->block_rsv,
3869 &root->fs_info->global_block_rsv,
5a77d76c 3870 trans->bytes_reserved);
3880a1b4 3871
a22285a6
YZ		 3872 if (err) {
3873 btrfs_end_transaction(trans, root);
3874 root->fs_info->enospc_unlink = 0;
3875 return ERR_PTR(err);
3876 }
3877
3878 trans->block_rsv = &root->fs_info->global_block_rsv;
3879 return trans;
3880}
3881
3882static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3883 struct btrfs_root *root)
3884{
66d8f3dd 3885 if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
5a77d76c
JB		 3886 btrfs_block_rsv_release(root, trans->block_rsv,
3887 trans->bytes_reserved);
3888 trans->block_rsv = &root->fs_info->trans_block_rsv;
a22285a6
YZ		 3889 BUG_ON(!root->fs_info->enospc_unlink);
3890 root->fs_info->enospc_unlink = 0;
3891 }
7ad85bb7 3892 btrfs_end_transaction(trans, root);
a22285a6
YZ		 3893}
3894
3895static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3896{
3897 struct btrfs_root *root = BTRFS_I(dir)->root;
3898 struct btrfs_trans_handle *trans;
3899 struct inode *inode = dentry->d_inode;
3900 int ret;
a22285a6
YZ		 3901
3902 trans = __unlink_start_trans(dir, dentry);
3903 if (IS_ERR(trans))
3904 return PTR_ERR(trans);
5f39d397 3905
12fcfd22
CM		 3906 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3907
e02119d5
CM		 3908 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3909 dentry->d_name.name, dentry->d_name.len);
b532402e
TI		 3910 if (ret)
3911 goto out;
7b128766 3912
a22285a6 3913 if (inode->i_nlink == 0) {
7b128766 3914 ret = btrfs_orphan_add(trans, inode);
b532402e
TI		 3915 if (ret)
3916 goto out;
a22285a6 3917 }
7b128766 3918
b532402e 3919out:
a22285a6 3920 __unlink_end_trans(trans, root);
b53d3f5d 3921 btrfs_btree_balance_dirty(root);
39279cc3
CM		 3922 return ret;
3923}
3924
4df27c4d
YZ		 3925int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3926 struct btrfs_root *root,
3927 struct inode *dir, u64 objectid,
3928 const char *name, int name_len)
3929{
3930 struct btrfs_path *path;
3931 struct extent_buffer *leaf;
3932 struct btrfs_dir_item *di;
3933 struct btrfs_key key;
3934 u64 index;
3935 int ret;
33345d01 3936 u64 dir_ino = btrfs_ino(dir);
4df27c4d
YZ		 3937
3938 path = btrfs_alloc_path();
3939 if (!path)
3940 return -ENOMEM;
3941
33345d01 3942 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4df27c4d 3943 name, name_len, -1);
79787eaa
JM		 3944 if (IS_ERR_OR_NULL(di)) {
3945 if (!di)
3946 ret = -ENOENT;
3947 else
3948 ret = PTR_ERR(di);
3949 goto out;
3950 }
4df27c4d
YZ		 3951
3952 leaf = path->nodes[0];
3953 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3954 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3955 ret = btrfs_delete_one_dir_name(trans, root, path, di);
79787eaa
JM		 3956 if (ret) {
3957 btrfs_abort_transaction(trans, root, ret);
3958 goto out;
3959 }
b3b4aa74 3960 btrfs_release_path(path);
4df27c4d
YZ		 3961
3962 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3963 objectid, root->root_key.objectid,
33345d01 3964 dir_ino, &index, name, name_len);
4df27c4d 3965 if (ret < 0) {
79787eaa
JM		 3966 if (ret != -ENOENT) {
3967 btrfs_abort_transaction(trans, root, ret);
3968 goto out;
3969 }
33345d01 3970 di = btrfs_search_dir_index_item(root, path, dir_ino,
4df27c4d 3971 name, name_len);
79787eaa
JM		 3972 if (IS_ERR_OR_NULL(di)) {
3973 if (!di)
3974 ret = -ENOENT;
3975 else
3976 ret = PTR_ERR(di);
3977 btrfs_abort_transaction(trans, root, ret);
3978 goto out;
3979 }
4df27c4d
YZ		 3980
3981 leaf = path->nodes[0];
3982 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
b3b4aa74 3983 btrfs_release_path(path);
4df27c4d
YZ		 3984 index = key.offset;
3985 }
945d8962 3986 btrfs_release_path(path);
4df27c4d 3987
16cdcec7 3988 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 3989 if (ret) {
3990 btrfs_abort_transaction(trans, root, ret);
3991 goto out;
3992 }
4df27c4d
YZ		 3993
3994 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95 3995 inode_inc_iversion(dir);
4df27c4d 3996 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
5a24e84c 3997 ret = btrfs_update_inode_fallback(trans, root, dir);
79787eaa
JM		 3998 if (ret)
3999 btrfs_abort_transaction(trans, root, ret);
4000out:
71d7aed0 4001 btrfs_free_path(path);
79787eaa 4002 return ret;
4df27c4d
YZ		 4003}
4004
39279cc3
CM		 4005static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4006{
4007 struct inode *inode = dentry->d_inode;
1832a6d5 4008 int err = 0;
39279cc3 4009 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 4010 struct btrfs_trans_handle *trans;
39279cc3 4011
b3ae244e 4012 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
134d4512 4013 return -ENOTEMPTY;
b3ae244e
DS		 4014 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4015 return -EPERM;
134d4512 4016
a22285a6
YZ		 4017 trans = __unlink_start_trans(dir, dentry);
4018 if (IS_ERR(trans))
5df6a9f6 4019 return PTR_ERR(trans);
5df6a9f6 4020
33345d01 4021 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4df27c4d
YZ		 4022 err = btrfs_unlink_subvol(trans, root, dir,
4023 BTRFS_I(inode)->location.objectid,
4024 dentry->d_name.name,
4025 dentry->d_name.len);
4026 goto out;
4027 }
4028
7b128766
JB		 4029 err = btrfs_orphan_add(trans, inode);
4030 if (err)
4df27c4d 4031 goto out;
7b128766 4032
39279cc3 4033 /* now the directory is empty */
e02119d5
CM		 4034 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
4035 dentry->d_name.name, dentry->d_name.len);
d397712b 4036 if (!err)
dbe674a9 4037 btrfs_i_size_write(inode, 0);
4df27c4d 4038out:
a22285a6 4039 __unlink_end_trans(trans, root);
b53d3f5d 4040 btrfs_btree_balance_dirty(root);
3954401f 4041
39279cc3
CM		 4042 return err;
4043}
4044
39279cc3
CM		 4045/*
4046 * this can truncate away extent items, csum items and directory items.
4047 * It starts at a high offset and removes keys until it can't find
d352ac68 4048 * any higher than new_size
39279cc3
CM		 4049 *
4050 * csum items that cross the new i_size are truncated to the new size
4051 * as well.
7b128766
JB		 4052 *
4053 * min_type is the minimum key type to truncate down to. If set to 0, this
4054 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 4055 */
8082510e
YZ		 4056int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4057 struct btrfs_root *root,
4058 struct inode *inode,
4059 u64 new_size, u32 min_type)
39279cc3 4060{
39279cc3 4061 struct btrfs_path *path;
5f39d397 4062 struct extent_buffer *leaf;
39279cc3 4063 struct btrfs_file_extent_item *fi;
8082510e
YZ		 4064 struct btrfs_key key;
4065 struct btrfs_key found_key;
39279cc3 4066 u64 extent_start = 0;
db94535d 4067 u64 extent_num_bytes = 0;
5d4f98a2 4068 u64 extent_offset = 0;
39279cc3 4069 u64 item_end = 0;
8082510e 4070 u32 found_type = (u8)-1;
39279cc3
CM		 4071 int found_extent;
4072 int del_item;
85e21bac
CM		 4073 int pending_del_nr = 0;
4074 int pending_del_slot = 0;
179e29e4 4075 int extent_type = -1;
8082510e
YZ		 4076 int ret;
4077 int err = 0;
33345d01 4078 u64 ino = btrfs_ino(inode);
8082510e
YZ		 4079
4080 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
39279cc3 4081
0eb0e19c
MF		 4082 path = btrfs_alloc_path();
4083 if (!path)
4084 return -ENOMEM;
4085 path->reada = -1;
4086
5dc562c5
JB		 4087 /*
4088 * We want to drop from the next block forward in case this new size is
4089 * not block aligned since we will be keeping the last block of the
4090 * extent just the way it is.
4091 */
0af3d00b 4092 if (root->ref_cows || root == root->fs_info->tree_root)
fda2832f
QW		 4093 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4094 root->sectorsize), (u64)-1, 0);
8082510e 4095
16cdcec7
MX		 4096 /*
4097 * This function is also used to drop the items in the log tree before
4098 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4099 * it is used to drop the loged items. So we shouldn't kill the delayed
4100 * items.
4101 */
4102 if (min_type == 0 && root == BTRFS_I(inode)->root)
4103 btrfs_kill_delayed_inode_items(inode);
4104
33345d01 4105 key.objectid = ino;
39279cc3 4106 key.offset = (u64)-1;
5f39d397
CM		 4107 key.type = (u8)-1;
4108
85e21bac 4109search_again:
b9473439 4110 path->leave_spinning = 1;
85e21bac 4111 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8082510e
YZ		 4112 if (ret < 0) {
4113 err = ret;
4114 goto out;
4115 }
d397712b 4116
85e21bac 4117 if (ret > 0) {
e02119d5
CM		 4118 /* there are no items in the tree for us to truncate, we're
4119 * done
4120 */
8082510e
YZ		 4121 if (path->slots[0] == 0)
4122 goto out;
85e21bac
CM		 4123 path->slots[0]--;
4124 }
4125
d397712b 4126 while (1) {
39279cc3 4127 fi = NULL;
5f39d397
CM		 4128 leaf = path->nodes[0];
4129 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4130 found_type = btrfs_key_type(&found_key);
39279cc3 4131
33345d01 4132 if (found_key.objectid != ino)
39279cc3 4133 break;
5f39d397 4134
85e21bac 4135 if (found_type < min_type)
39279cc3
CM		 4136 break;
4137
5f39d397 4138 item_end = found_key.offset;
39279cc3 4139 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 4140 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 4141 struct btrfs_file_extent_item);
179e29e4
CM		 4142 extent_type = btrfs_file_extent_type(leaf, fi);
4143 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 4144 item_end +=
db94535d 4145 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 4146 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 4147 item_end += btrfs_file_extent_inline_len(leaf,
c8b97818 4148 fi);
39279cc3 4149 }
008630c1 4150 item_end--;
39279cc3 4151 }
8082510e
YZ		 4152 if (found_type > min_type) {
4153 del_item = 1;
4154 } else {
4155 if (item_end < new_size)
b888db2b 4156 break;
8082510e
YZ		 4157 if (found_key.offset >= new_size)
4158 del_item = 1;
4159 else
4160 del_item = 0;
39279cc3 4161 }
39279cc3 4162 found_extent = 0;
39279cc3 4163 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 4164 if (found_type != BTRFS_EXTENT_DATA_KEY)
4165 goto delete;
4166
4167 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 4168 u64 num_dec;
db94535d 4169 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
f70a9a6b 4170 if (!del_item) {
db94535d
CM		 4171 u64 orig_num_bytes =
4172 btrfs_file_extent_num_bytes(leaf, fi);
fda2832f
QW		 4173 extent_num_bytes = ALIGN(new_size -
4174 found_key.offset,
4175 root->sectorsize);
db94535d
CM		 4176 btrfs_set_file_extent_num_bytes(leaf, fi,
4177 extent_num_bytes);
4178 num_dec = (orig_num_bytes -
9069218d 4179 extent_num_bytes);
e02119d5 4180 if (root->ref_cows && extent_start != 0)
a76a3cd4 4181 inode_sub_bytes(inode, num_dec);
5f39d397 4182 btrfs_mark_buffer_dirty(leaf);
39279cc3 4183 } else {
db94535d
CM		 4184 extent_num_bytes =
4185 btrfs_file_extent_disk_num_bytes(leaf,
4186 fi);
5d4f98a2
YZ		 4187 extent_offset = found_key.offset -
4188 btrfs_file_extent_offset(leaf, fi);
4189
39279cc3 4190 /* FIXME blocksize != 4096 */
9069218d 4191 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 4192 if (extent_start != 0) {
4193 found_extent = 1;
e02119d5 4194 if (root->ref_cows)
a76a3cd4 4195 inode_sub_bytes(inode, num_dec);
e02119d5 4196 }
39279cc3 4197 }
9069218d 4198 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 4199 /*
4200 * we can't truncate inline items that have had
4201 * special encodings
4202 */
4203 if (!del_item &&
4204 btrfs_file_extent_compression(leaf, fi) == 0 &&
4205 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4206 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM		 4207 u32 size = new_size - found_key.offset;
4208
4209 if (root->ref_cows) {
a76a3cd4
YZ		 4210 inode_sub_bytes(inode, item_end + 1 -
4211 new_size);
e02119d5
CM		 4212 }
4213 size =
4214 btrfs_file_extent_calc_inline_size(size);
afe5fea7 4215 btrfs_truncate_item(root, path, size, 1);
e02119d5 4216 } else if (root->ref_cows) {
a76a3cd4
YZ		 4217 inode_sub_bytes(inode, item_end + 1 -
4218 found_key.offset);
9069218d 4219 }
39279cc3 4220 }
179e29e4 4221delete:
39279cc3 4222 if (del_item) {
85e21bac
CM		 4223 if (!pending_del_nr) {
4224 /* no pending yet, add ourselves */
4225 pending_del_slot = path->slots[0];
4226 pending_del_nr = 1;
4227 } else if (pending_del_nr &&
4228 path->slots[0] + 1 == pending_del_slot) {
4229 /* hop on the pending chunk */
4230 pending_del_nr++;
4231 pending_del_slot = path->slots[0];
4232 } else {
d397712b 4233 BUG();
85e21bac 4234 }
39279cc3
CM		 4235 } else {
4236 break;
4237 }
0af3d00b
JB		 4238 if (found_extent && (root->ref_cows ||
4239 root == root->fs_info->tree_root)) {
b9473439 4240 btrfs_set_path_blocking(path);
39279cc3 4241 ret = btrfs_free_extent(trans, root, extent_start,
5d4f98a2
YZ		 4242 extent_num_bytes, 0,
4243 btrfs_header_owner(leaf),
66d7e7f0 4244 ino, extent_offset, 0);
39279cc3
CM		 4245 BUG_ON(ret);
4246 }
85e21bac 4247
8082510e
YZ		 4248 if (found_type == BTRFS_INODE_ITEM_KEY)
4249 break;
4250
4251 if (path->slots[0] == 0 ||
4252 path->slots[0] != pending_del_slot) {
8082510e
YZ		 4253 if (pending_del_nr) {
4254 ret = btrfs_del_items(trans, root, path,
4255 pending_del_slot,
4256 pending_del_nr);
79787eaa
JM		 4257 if (ret) {
4258 btrfs_abort_transaction(trans,
4259 root, ret);
4260 goto error;
4261 }
8082510e
YZ		 4262 pending_del_nr = 0;
4263 }
b3b4aa74 4264 btrfs_release_path(path);
85e21bac 4265 goto search_again;
8082510e
YZ		 4266 } else {
4267 path->slots[0]--;
85e21bac 4268 }
39279cc3 4269 }
8082510e 4270out:
85e21bac
CM		 4271 if (pending_del_nr) {
4272 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4273 pending_del_nr);
79787eaa
JM		 4274 if (ret)
4275 btrfs_abort_transaction(trans, root, ret);
85e21bac 4276 }
79787eaa 4277error:
39279cc3 4278 btrfs_free_path(path);
8082510e 4279 return err;
39279cc3
CM		 4280}
4281
4282/*
2aaa6655
JB		 4283 * btrfs_truncate_page - read, zero a chunk and write a page
4284 * @inode - inode that we're zeroing
4285 * @from - the offset to start zeroing
4286 * @len - the length to zero, 0 to zero the entire range respective to the
4287 * offset
4288 * @front - zero up to the offset instead of from the offset on
4289 *
4290 * This will find the page for the "from" offset and cow the page and zero the
4291 * part we want to zero. This is used with truncate and hole punching.
39279cc3 4292 */
2aaa6655
JB		 4293int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4294 int front)
39279cc3 4295{
2aaa6655 4296 struct address_space *mapping = inode->i_mapping;
db94535d 4297 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 4298 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4299 struct btrfs_ordered_extent *ordered;
2ac55d41 4300 struct extent_state *cached_state = NULL;
e6dcd2dc 4301 char *kaddr;
db94535d 4302 u32 blocksize = root->sectorsize;
39279cc3
CM		 4303 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4304 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4305 struct page *page;
3b16a4e3 4306 gfp_t mask = btrfs_alloc_write_mask(mapping);
39279cc3 4307 int ret = 0;
a52d9a80 4308 u64 page_start;
e6dcd2dc 4309 u64 page_end;
39279cc3 4310
2aaa6655
JB		 4311 if ((offset & (blocksize - 1)) == 0 &&
4312 (!len || ((len & (blocksize - 1)) == 0)))
39279cc3 4313 goto out;
0ca1f7ce 4314 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
5d5e103a
JB		 4315 if (ret)
4316 goto out;
39279cc3 4317
211c17f5 4318again:
3b16a4e3 4319 page = find_or_create_page(mapping, index, mask);
5d5e103a 4320 if (!page) {
0ca1f7ce 4321 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
ac6a2b36 4322 ret = -ENOMEM;
39279cc3 4323 goto out;
5d5e103a 4324 }
e6dcd2dc
CM		 4325
4326 page_start = page_offset(page);
4327 page_end = page_start + PAGE_CACHE_SIZE - 1;
4328
39279cc3 4329 if (!PageUptodate(page)) {
9ebefb18 4330 ret = btrfs_readpage(NULL, page);
39279cc3 4331 lock_page(page);
211c17f5
CM		 4332 if (page->mapping != mapping) {
4333 unlock_page(page);
4334 page_cache_release(page);
4335 goto again;
4336 }
39279cc3
CM		 4337 if (!PageUptodate(page)) {
4338 ret = -EIO;
89642229 4339 goto out_unlock;
39279cc3
CM		 4340 }
4341 }
211c17f5 4342 wait_on_page_writeback(page);
e6dcd2dc 4343
d0082371 4344 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 4345 set_page_extent_mapped(page);
4346
4347 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4348 if (ordered) {
2ac55d41
JB		 4349 unlock_extent_cached(io_tree, page_start, page_end,
4350 &cached_state, GFP_NOFS);
e6dcd2dc
CM		 4351 unlock_page(page);
4352 page_cache_release(page);
eb84ae03 4353 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 4354 btrfs_put_ordered_extent(ordered);
4355 goto again;
4356 }
4357
2ac55d41 4358 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 4359 EXTENT_DIRTY | EXTENT_DELALLOC |
4360 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 4361 0, 0, &cached_state, GFP_NOFS);
5d5e103a 4362
2ac55d41
JB		 4363 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4364 &cached_state);
9ed74f2d 4365 if (ret) {
2ac55d41
JB		 4366 unlock_extent_cached(io_tree, page_start, page_end,
4367 &cached_state, GFP_NOFS);
9ed74f2d
JB		 4368 goto out_unlock;
4369 }
4370
e6dcd2dc 4371 if (offset != PAGE_CACHE_SIZE) {
2aaa6655
JB		 4372 if (!len)
4373 len = PAGE_CACHE_SIZE - offset;
e6dcd2dc 4374 kaddr = kmap(page);
2aaa6655
JB		 4375 if (front)
4376 memset(kaddr, 0, offset);
4377 else
4378 memset(kaddr + offset, 0, len);
e6dcd2dc
CM		 4379 flush_dcache_page(page);
4380 kunmap(page);
4381 }
247e743c 4382 ClearPageChecked(page);
e6dcd2dc 4383 set_page_dirty(page);
2ac55d41
JB		 4384 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4385 GFP_NOFS);
39279cc3 4386
89642229 4387out_unlock:
5d5e103a 4388 if (ret)
0ca1f7ce 4389 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
39279cc3
CM		 4390 unlock_page(page);
4391 page_cache_release(page);
4392out:
4393 return ret;
4394}
4395
695a0d0d
JB		 4396/*
4397 * This function puts in dummy file extents for the area we're creating a hole
4398 * for. So if we are truncating this file to a larger size we need to insert
4399 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4400 * the range between oldsize and size
4401 */
a41ad394 4402int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
39279cc3 4403{
9036c102
YZ		 4404 struct btrfs_trans_handle *trans;
4405 struct btrfs_root *root = BTRFS_I(inode)->root;
4406 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a22285a6 4407 struct extent_map *em = NULL;
2ac55d41 4408 struct extent_state *cached_state = NULL;
5dc562c5 4409 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
fda2832f
QW		 4410 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4411 u64 block_end = ALIGN(size, root->sectorsize);
9036c102
YZ		 4412 u64 last_byte;
4413 u64 cur_offset;
4414 u64 hole_size;
9ed74f2d 4415 int err = 0;
39279cc3 4416
9036c102
YZ		 4417 if (size <= hole_start)
4418 return 0;
4419
9036c102
YZ		 4420 while (1) {
4421 struct btrfs_ordered_extent *ordered;
4422 btrfs_wait_ordered_range(inode, hole_start,
4423 block_end - hole_start);
2ac55d41 4424 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
d0082371 4425 &cached_state);
9036c102
YZ		 4426 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
4427 if (!ordered)
4428 break;
2ac55d41
JB		 4429 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4430 &cached_state, GFP_NOFS);
9036c102
YZ		 4431 btrfs_put_ordered_extent(ordered);
4432 }
39279cc3 4433
9036c102
YZ		 4434 cur_offset = hole_start;
4435 while (1) {
4436 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4437 block_end - cur_offset, 0);
79787eaa
JM		 4438 if (IS_ERR(em)) {
4439 err = PTR_ERR(em);
f2767956 4440 em = NULL;
79787eaa
JM		 4441 break;
4442 }
9036c102 4443 last_byte = min(extent_map_end(em), block_end);
fda2832f 4444 last_byte = ALIGN(last_byte , root->sectorsize);
8082510e 4445 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5dc562c5 4446 struct extent_map *hole_em;
9036c102 4447 hole_size = last_byte - cur_offset;
9ed74f2d 4448
3642320e 4449 trans = btrfs_start_transaction(root, 3);
a22285a6
YZ		 4450 if (IS_ERR(trans)) {
4451 err = PTR_ERR(trans);
9ed74f2d 4452 break;
a22285a6 4453 }
8082510e 4454
5dc562c5
JB		 4455 err = btrfs_drop_extents(trans, root, inode,
4456 cur_offset,
2671485d 4457 cur_offset + hole_size, 1);
5b397377 4458 if (err) {
79787eaa 4459 btrfs_abort_transaction(trans, root, err);
5b397377 4460 btrfs_end_transaction(trans, root);
3893e33b 4461 break;
5b397377 4462 }
8082510e 4463
9036c102 4464 err = btrfs_insert_file_extent(trans, root,
33345d01 4465 btrfs_ino(inode), cur_offset, 0,
9036c102
YZ		 4466 0, hole_size, 0, hole_size,
4467 0, 0, 0);
5b397377 4468 if (err) {
79787eaa 4469 btrfs_abort_transaction(trans, root, err);
5b397377 4470 btrfs_end_transaction(trans, root);
3893e33b 4471 break;
5b397377 4472 }
8082510e 4473
5dc562c5
JB		 4474 btrfs_drop_extent_cache(inode, cur_offset,
4475 cur_offset + hole_size - 1, 0);
4476 hole_em = alloc_extent_map();
4477 if (!hole_em) {
4478 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4479 &BTRFS_I(inode)->runtime_flags);
4480 goto next;
4481 }
4482 hole_em->start = cur_offset;
4483 hole_em->len = hole_size;
4484 hole_em->orig_start = cur_offset;
8082510e 4485
5dc562c5
JB		 4486 hole_em->block_start = EXTENT_MAP_HOLE;
4487 hole_em->block_len = 0;
b4939680 4488 hole_em->orig_block_len = 0;
cc95bef6 4489 hole_em->ram_bytes = hole_size;
5dc562c5
JB		 4490 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4491 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4492 hole_em->generation = trans->transid;
8082510e 4493
5dc562c5
JB		 4494 while (1) {
4495 write_lock(&em_tree->lock);
09a2a8f9 4496 err = add_extent_mapping(em_tree, hole_em, 1);
5dc562c5
JB		 4497 write_unlock(&em_tree->lock);
4498 if (err != -EEXIST)
4499 break;
4500 btrfs_drop_extent_cache(inode, cur_offset,
4501 cur_offset +
4502 hole_size - 1, 0);
4503 }
4504 free_extent_map(hole_em);
4505next:
3642320e 4506 btrfs_update_inode(trans, root, inode);
8082510e 4507 btrfs_end_transaction(trans, root);
9036c102
YZ		 4508 }
4509 free_extent_map(em);
a22285a6 4510 em = NULL;
9036c102 4511 cur_offset = last_byte;
8082510e 4512 if (cur_offset >= block_end)
9036c102
YZ		 4513 break;
4514 }
1832a6d5 4515
a22285a6 4516 free_extent_map(em);
2ac55d41
JB		 4517 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4518 GFP_NOFS);
9036c102
YZ		 4519 return err;
4520}
39279cc3 4521
3972f260 4522static int btrfs_setsize(struct inode *inode, struct iattr *attr)
8082510e 4523{
f4a2f4c5
MX		 4524 struct btrfs_root *root = BTRFS_I(inode)->root;
4525 struct btrfs_trans_handle *trans;
a41ad394 4526 loff_t oldsize = i_size_read(inode);
3972f260
ES		 4527 loff_t newsize = attr->ia_size;
4528 int mask = attr->ia_valid;
8082510e
YZ		 4529 int ret;
4530
a41ad394 4531 if (newsize == oldsize)
8082510e
YZ		 4532 return 0;
4533
3972f260
ES		 4534 /*
4535 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4536 * special case where we need to update the times despite not having
4537 * these flags set. For all other operations the VFS set these flags
4538 * explicitly if it wants a timestamp update.
4539 */
4540 if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4541 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4542
a41ad394 4543 if (newsize > oldsize) {
a41ad394
JB		 4544 truncate_pagecache(inode, oldsize, newsize);
4545 ret = btrfs_cont_expand(inode, oldsize, newsize);
f4a2f4c5 4546 if (ret)
8082510e 4547 return ret;
8082510e 4548
f4a2f4c5
MX		 4549 trans = btrfs_start_transaction(root, 1);
4550 if (IS_ERR(trans))
4551 return PTR_ERR(trans);
4552
4553 i_size_write(inode, newsize);
4554 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4555 ret = btrfs_update_inode(trans, root, inode);
7ad85bb7 4556 btrfs_end_transaction(trans, root);
a41ad394 4557 } else {
8082510e 4558
a41ad394
JB		 4559 /*
4560 * We're truncating a file that used to have good data down to
4561 * zero. Make sure it gets into the ordered flush list so that
4562 * any new writes get down to disk quickly.
4563 */
4564 if (newsize == 0)
72ac3c0d
JB		 4565 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4566 &BTRFS_I(inode)->runtime_flags);
8082510e 4567
f3fe820c
JB		 4568 /*
4569 * 1 for the orphan item we're going to add
4570 * 1 for the orphan item deletion.
4571 */
4572 trans = btrfs_start_transaction(root, 2);
4573 if (IS_ERR(trans))
4574 return PTR_ERR(trans);
4575
4576 /*
4577 * We need to do this in case we fail at _any_ point during the
4578 * actual truncate. Once we do the truncate_setsize we could
4579 * invalidate pages which forces any outstanding ordered io to
4580 * be instantly completed which will give us extents that need
4581 * to be truncated. If we fail to get an orphan inode down we
4582 * could have left over extents that were never meant to live,
4583 * so we need to garuntee from this point on that everything
4584 * will be consistent.
4585 */
4586 ret = btrfs_orphan_add(trans, inode);
4587 btrfs_end_transaction(trans, root);
4588 if (ret)
4589 return ret;
4590
a41ad394
JB		 4591 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4592 truncate_setsize(inode, newsize);
2e60a51e
MX		 4593
4594 /* Disable nonlocked read DIO to avoid the end less truncate */
4595 btrfs_inode_block_unlocked_dio(inode);
4596 inode_dio_wait(inode);
4597 btrfs_inode_resume_unlocked_dio(inode);
4598
a41ad394 4599 ret = btrfs_truncate(inode);
f3fe820c
JB		 4600 if (ret && inode->i_nlink)
4601 btrfs_orphan_del(NULL, inode);
8082510e
YZ		 4602 }
4603
a41ad394 4604 return ret;
8082510e
YZ		 4605}
4606
9036c102
YZ		 4607static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4608{
4609 struct inode *inode = dentry->d_inode;
b83cc969 4610 struct btrfs_root *root = BTRFS_I(inode)->root;
9036c102 4611 int err;
39279cc3 4612
b83cc969
LZ		 4613 if (btrfs_root_readonly(root))
4614 return -EROFS;
4615
9036c102
YZ		 4616 err = inode_change_ok(inode, attr);
4617 if (err)
4618 return err;
2bf5a725 4619
5a3f23d5 4620 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3972f260 4621 err = btrfs_setsize(inode, attr);
8082510e
YZ		 4622 if (err)
4623 return err;
39279cc3 4624 }
9036c102 4625
1025774c
CH		 4626 if (attr->ia_valid) {
4627 setattr_copy(inode, attr);
0c4d2d95 4628 inode_inc_iversion(inode);
22c44fe6 4629 err = btrfs_dirty_inode(inode);
1025774c 4630
22c44fe6 4631 if (!err && attr->ia_valid & ATTR_MODE)
1025774c
CH		 4632 err = btrfs_acl_chmod(inode);
4633 }
33268eaf 4634
39279cc3
CM		 4635 return err;
4636}
61295eb8 4637
bd555975 4638void btrfs_evict_inode(struct inode *inode)
39279cc3
CM		 4639{
4640 struct btrfs_trans_handle *trans;
4641 struct btrfs_root *root = BTRFS_I(inode)->root;
726c35fa 4642 struct btrfs_block_rsv *rsv, *global_rsv;
07127184 4643 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3
CM		 4644 int ret;
4645
1abe9b8a 4646 trace_btrfs_inode_evict(inode);
4647
39279cc3 4648 truncate_inode_pages(&inode->i_data, 0);
0af3d00b 4649 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
83eea1f1 4650 btrfs_is_free_space_inode(inode)))
bd555975
AV		 4651 goto no_delete;
4652
39279cc3 4653 if (is_bad_inode(inode)) {
7b128766 4654 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 4655 goto no_delete;
4656 }
bd555975 4657 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4a096752 4658 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 4659
c71bf099 4660 if (root->fs_info->log_root_recovering) {
6bf02314 4661 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8a35d95f 4662 &BTRFS_I(inode)->runtime_flags));
c71bf099
YZ		 4663 goto no_delete;
4664 }
4665
76dda93c
YZ		 4666 if (inode->i_nlink > 0) {
4667 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
4668 goto no_delete;
4669 }
4670
0e8c36a9
MX		 4671 ret = btrfs_commit_inode_delayed_inode(inode);
4672 if (ret) {
4673 btrfs_orphan_del(NULL, inode);
4674 goto no_delete;
4675 }
4676
66d8f3dd 4677 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4289a667
JB		 4678 if (!rsv) {
4679 btrfs_orphan_del(NULL, inode);
4680 goto no_delete;
4681 }
4a338542 4682 rsv->size = min_size;
ca7e70f5 4683 rsv->failfast = 1;
726c35fa 4684 global_rsv = &root->fs_info->global_block_rsv;
4289a667 4685
dbe674a9 4686 btrfs_i_size_write(inode, 0);
5f39d397 4687
4289a667 4688 /*
8407aa46
MX		 4689 * This is a bit simpler than btrfs_truncate since we've already
4690 * reserved our space for our orphan item in the unlink, so we just
4691 * need to reserve some slack space in case we add bytes and update
4692 * inode item when doing the truncate.
4289a667 4693 */
8082510e 4694 while (1) {
08e007d2
MX		 4695 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4696 BTRFS_RESERVE_FLUSH_LIMIT);
726c35fa
JB		 4697
4698 /*
4699 * Try and steal from the global reserve since we will
4700 * likely not use this space anyway, we want to try as
4701 * hard as possible to get this to work.
4702 */
4703 if (ret)
4704 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
d68fc57b 4705
d68fc57b 4706 if (ret) {
c2cf52eb
SK		 4707 btrfs_warn(root->fs_info,
4708 "Could not get space for a delete, will truncate on mount %d",
4709 ret);
4289a667
JB		 4710 btrfs_orphan_del(NULL, inode);
4711 btrfs_free_block_rsv(root, rsv);
4712 goto no_delete;
d68fc57b 4713 }
7b128766 4714
0e8c36a9 4715 trans = btrfs_join_transaction(root);
4289a667
JB		 4716 if (IS_ERR(trans)) {
4717 btrfs_orphan_del(NULL, inode);
4718 btrfs_free_block_rsv(root, rsv);
4719 goto no_delete;
d68fc57b 4720 }
7b128766 4721
4289a667
JB		 4722 trans->block_rsv = rsv;
4723
d68fc57b 4724 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
ca7e70f5 4725 if (ret != -ENOSPC)
8082510e 4726 break;
85e21bac 4727
8407aa46 4728 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e
YZ		 4729 btrfs_end_transaction(trans, root);
4730 trans = NULL;
b53d3f5d 4731 btrfs_btree_balance_dirty(root);
8082510e 4732 }
5f39d397 4733
4289a667
JB		 4734 btrfs_free_block_rsv(root, rsv);
4735
8082510e 4736 if (ret == 0) {
4289a667 4737 trans->block_rsv = root->orphan_block_rsv;
8082510e
YZ		 4738 ret = btrfs_orphan_del(trans, inode);
4739 BUG_ON(ret);
4740 }
54aa1f4d 4741
4289a667 4742 trans->block_rsv = &root->fs_info->trans_block_rsv;
581bb050
LZ		 4743 if (!(root == root->fs_info->tree_root ||
4744 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
33345d01 4745 btrfs_return_ino(root, btrfs_ino(inode));
581bb050 4746
54aa1f4d 4747 btrfs_end_transaction(trans, root);
b53d3f5d 4748 btrfs_btree_balance_dirty(root);
39279cc3 4749no_delete:
89042e5a 4750 btrfs_remove_delayed_node(inode);
dbd5768f 4751 clear_inode(inode);
8082510e 4752 return;
39279cc3
CM		 4753}
4754
4755/*
4756 * this returns the key found in the dir entry in the location pointer.
4757 * If no dir entries were found, location->objectid is 0.
4758 */
4759static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4760 struct btrfs_key *location)
4761{
4762 const char *name = dentry->d_name.name;
4763 int namelen = dentry->d_name.len;
4764 struct btrfs_dir_item *di;
4765 struct btrfs_path *path;
4766 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 4767 int ret = 0;
39279cc3
CM		 4768
4769 path = btrfs_alloc_path();
d8926bb3
MF		 4770 if (!path)
4771 return -ENOMEM;
3954401f 4772
33345d01 4773 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
39279cc3 4774 namelen, 0);
0d9f7f3e
Y		 4775 if (IS_ERR(di))
4776 ret = PTR_ERR(di);
d397712b 4777
c704005d 4778 if (IS_ERR_OR_NULL(di))
3954401f 4779 goto out_err;
d397712b 4780
5f39d397 4781 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 4782out:
39279cc3
CM		 4783 btrfs_free_path(path);
4784 return ret;
3954401f
CM		 4785out_err:
4786 location->objectid = 0;
4787 goto out;
39279cc3
CM		 4788}
4789
4790/*
4791 * when we hit a tree root in a directory, the btrfs part of the inode
4792 * needs to be changed to reflect the root directory of the tree root. This
4793 * is kind of like crossing a mount point.
4794 */
4795static int fixup_tree_root_location(struct btrfs_root *root,
4df27c4d
YZ		 4796 struct inode *dir,
4797 struct dentry *dentry,
4798 struct btrfs_key *location,
4799 struct btrfs_root **sub_root)
39279cc3 4800{
4df27c4d
YZ		 4801 struct btrfs_path *path;
4802 struct btrfs_root *new_root;
4803 struct btrfs_root_ref *ref;
4804 struct extent_buffer *leaf;
4805 int ret;
4806 int err = 0;
39279cc3 4807
4df27c4d
YZ		 4808 path = btrfs_alloc_path();
4809 if (!path) {
4810 err = -ENOMEM;
4811 goto out;
4812 }
39279cc3 4813
4df27c4d
YZ		 4814 err = -ENOENT;
4815 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4816 BTRFS_I(dir)->root->root_key.objectid,
4817 location->objectid);
4818 if (ret) {
4819 if (ret < 0)
4820 err = ret;
4821 goto out;
4822 }
39279cc3 4823
4df27c4d
YZ		 4824 leaf = path->nodes[0];
4825 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
33345d01 4826 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4df27c4d
YZ		 4827 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4828 goto out;
39279cc3 4829
4df27c4d
YZ		 4830 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4831 (unsigned long)(ref + 1),
4832 dentry->d_name.len);
4833 if (ret)
4834 goto out;
4835
b3b4aa74 4836 btrfs_release_path(path);
4df27c4d
YZ		 4837
4838 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4839 if (IS_ERR(new_root)) {
4840 err = PTR_ERR(new_root);
4841 goto out;
4842 }
4843
4df27c4d
YZ		 4844 *sub_root = new_root;
4845 location->objectid = btrfs_root_dirid(&new_root->root_item);
4846 location->type = BTRFS_INODE_ITEM_KEY;
4847 location->offset = 0;
4848 err = 0;
4849out:
4850 btrfs_free_path(path);
4851 return err;
39279cc3
CM		 4852}
4853
5d4f98a2
YZ		 4854static void inode_tree_add(struct inode *inode)
4855{
4856 struct btrfs_root *root = BTRFS_I(inode)->root;
4857 struct btrfs_inode *entry;
03e860bd
FNP		 4858 struct rb_node **p;
4859 struct rb_node *parent;
33345d01 4860 u64 ino = btrfs_ino(inode);
5d4f98a2 4861
1d3382cb 4862 if (inode_unhashed(inode))
76dda93c 4863 return;
e1409cef
MX		 4864again:
4865 parent = NULL;
5d4f98a2 4866 spin_lock(&root->inode_lock);
e1409cef 4867 p = &root->inode_tree.rb_node;
5d4f98a2
YZ		 4868 while (*p) {
4869 parent = *p;
4870 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4871
33345d01 4872 if (ino < btrfs_ino(&entry->vfs_inode))
03e860bd 4873 p = &parent->rb_left;
33345d01 4874 else if (ino > btrfs_ino(&entry->vfs_inode))
03e860bd 4875 p = &parent->rb_right;
5d4f98a2
YZ		 4876 else {
4877 WARN_ON(!(entry->vfs_inode.i_state &
a4ffdde6 4878 (I_WILL_FREE | I_FREEING)));
03e860bd
FNP		 4879 rb_erase(parent, &root->inode_tree);
4880 RB_CLEAR_NODE(parent);
4881 spin_unlock(&root->inode_lock);
4882 goto again;
5d4f98a2
YZ		 4883 }
4884 }
4885 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4886 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4887 spin_unlock(&root->inode_lock);
4888}
4889
4890static void inode_tree_del(struct inode *inode)
4891{
4892 struct btrfs_root *root = BTRFS_I(inode)->root;
76dda93c 4893 int empty = 0;
5d4f98a2 4894
03e860bd 4895 spin_lock(&root->inode_lock);
5d4f98a2 4896 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5d4f98a2 4897 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5d4f98a2 4898 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
76dda93c 4899 empty = RB_EMPTY_ROOT(&root->inode_tree);
5d4f98a2 4900 }
03e860bd 4901 spin_unlock(&root->inode_lock);
76dda93c 4902
0af3d00b
JB		 4903 /*
4904 * Free space cache has inodes in the tree root, but the tree root has a
4905 * root_refs of 0, so this could end up dropping the tree root as a
4906 * snapshot, so we need the extra !root->fs_info->tree_root check to
4907 * make sure we don't drop it.
4908 */
4909 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4910 root != root->fs_info->tree_root) {
76dda93c
YZ		 4911 synchronize_srcu(&root->fs_info->subvol_srcu);
4912 spin_lock(&root->inode_lock);
4913 empty = RB_EMPTY_ROOT(&root->inode_tree);
4914 spin_unlock(&root->inode_lock);
4915 if (empty)
4916 btrfs_add_dead_root(root);
4917 }
4918}
4919
143bede5 4920void btrfs_invalidate_inodes(struct btrfs_root *root)
76dda93c
YZ		 4921{
4922 struct rb_node *node;
4923 struct rb_node *prev;
4924 struct btrfs_inode *entry;
4925 struct inode *inode;
4926 u64 objectid = 0;
4927
4928 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4929
4930 spin_lock(&root->inode_lock);
4931again:
4932 node = root->inode_tree.rb_node;
4933 prev = NULL;
4934 while (node) {
4935 prev = node;
4936 entry = rb_entry(node, struct btrfs_inode, rb_node);
4937
33345d01 4938 if (objectid < btrfs_ino(&entry->vfs_inode))
76dda93c 4939 node = node->rb_left;
33345d01 4940 else if (objectid > btrfs_ino(&entry->vfs_inode))
76dda93c
YZ		 4941 node = node->rb_right;
4942 else
4943 break;
4944 }
4945 if (!node) {
4946 while (prev) {
4947 entry = rb_entry(prev, struct btrfs_inode, rb_node);
33345d01 4948 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
76dda93c
YZ		 4949 node = prev;
4950 break;
4951 }
4952 prev = rb_next(prev);
4953 }
4954 }
4955 while (node) {
4956 entry = rb_entry(node, struct btrfs_inode, rb_node);
33345d01 4957 objectid = btrfs_ino(&entry->vfs_inode) + 1;
76dda93c
YZ		 4958 inode = igrab(&entry->vfs_inode);
4959 if (inode) {
4960 spin_unlock(&root->inode_lock);
4961 if (atomic_read(&inode->i_count) > 1)
4962 d_prune_aliases(inode);
4963 /*
45321ac5 4964 * btrfs_drop_inode will have it removed from
76dda93c
YZ		 4965 * the inode cache when its usage count
4966 * hits zero.
4967 */
4968 iput(inode);
4969 cond_resched();
4970 spin_lock(&root->inode_lock);
4971 goto again;
4972 }
4973
4974 if (cond_resched_lock(&root->inode_lock))
4975 goto again;
4976
4977 node = rb_next(node);
4978 }
4979 spin_unlock(&root->inode_lock);
5d4f98a2
YZ		 4980}
4981
e02119d5
CM		 4982static int btrfs_init_locked_inode(struct inode *inode, void *p)
4983{
4984 struct btrfs_iget_args *args = p;
4985 inode->i_ino = args->ino;
e02119d5 4986 BTRFS_I(inode)->root = args->root;
39279cc3
CM		 4987 return 0;
4988}
4989
4990static int btrfs_find_actor(struct inode *inode, void *opaque)
4991{
4992 struct btrfs_iget_args *args = opaque;
33345d01 4993 return args->ino == btrfs_ino(inode) &&
d397712b 4994 args->root == BTRFS_I(inode)->root;
39279cc3
CM		 4995}
4996
5d4f98a2
YZ		 4997static struct inode *btrfs_iget_locked(struct super_block *s,
4998 u64 objectid,
4999 struct btrfs_root *root)
39279cc3
CM		 5000{
5001 struct inode *inode;
5002 struct btrfs_iget_args args;
5003 args.ino = objectid;
5004 args.root = root;
5005
5006 inode = iget5_locked(s, objectid, btrfs_find_actor,
5007 btrfs_init_locked_inode,
5008 (void *)&args);
5009 return inode;
5010}
5011
1a54ef8c
BR		 5012/* Get an inode object given its location and corresponding root.
5013 * Returns in *is_new if the inode was read from disk
5014 */
5015struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
73f73415 5016 struct btrfs_root *root, int *new)
1a54ef8c
BR		 5017{
5018 struct inode *inode;
5019
5020 inode = btrfs_iget_locked(s, location->objectid, root);
5021 if (!inode)
5d4f98a2 5022 return ERR_PTR(-ENOMEM);
1a54ef8c
BR		 5023
5024 if (inode->i_state & I_NEW) {
5025 BTRFS_I(inode)->root = root;
5026 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
5027 btrfs_read_locked_inode(inode);
1748f843
MF		 5028 if (!is_bad_inode(inode)) {
5029 inode_tree_add(inode);
5030 unlock_new_inode(inode);
5031 if (new)
5032 *new = 1;
5033 } else {
e0b6d65b
ST		 5034 unlock_new_inode(inode);
5035 iput(inode);
5036 inode = ERR_PTR(-ESTALE);
1748f843
MF		 5037 }
5038 }
5039
1a54ef8c
BR		 5040 return inode;
5041}
5042
4df27c4d
YZ		 5043static struct inode *new_simple_dir(struct super_block *s,
5044 struct btrfs_key *key,
5045 struct btrfs_root *root)
5046{
5047 struct inode *inode = new_inode(s);
5048
5049 if (!inode)
5050 return ERR_PTR(-ENOMEM);
5051
4df27c4d
YZ		 5052 BTRFS_I(inode)->root = root;
5053 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
72ac3c0d 5054 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4df27c4d
YZ		 5055
5056 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
848cce0d 5057 inode->i_op = &btrfs_dir_ro_inode_operations;
4df27c4d
YZ		 5058 inode->i_fop = &simple_dir_operations;
5059 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5060 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5061
5062 return inode;
5063}
5064
3de4586c 5065struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3 5066{
d397712b 5067 struct inode *inode;
4df27c4d 5068 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3
CM		 5069 struct btrfs_root *sub_root = root;
5070 struct btrfs_key location;
76dda93c 5071 int index;
b4aff1f8 5072 int ret = 0;
39279cc3
CM		 5073
5074 if (dentry->d_name.len > BTRFS_NAME_LEN)
5075 return ERR_PTR(-ENAMETOOLONG);
5f39d397 5076
39e3c955 5077 ret = btrfs_inode_by_name(dir, dentry, &location);
39279cc3
CM		 5078 if (ret < 0)
5079 return ERR_PTR(ret);
5f39d397 5080
4df27c4d
YZ		 5081 if (location.objectid == 0)
5082 return NULL;
5083
5084 if (location.type == BTRFS_INODE_ITEM_KEY) {
73f73415 5085 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4df27c4d
YZ		 5086 return inode;
5087 }
5088
5089 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5090
76dda93c 5091 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4df27c4d
YZ		 5092 ret = fixup_tree_root_location(root, dir, dentry,
5093 &location, &sub_root);
5094 if (ret < 0) {
5095 if (ret != -ENOENT)
5096 inode = ERR_PTR(ret);
5097 else
5098 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5099 } else {
73f73415 5100 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
39279cc3 5101 }
76dda93c
YZ		 5102 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5103
34d19bad 5104 if (!IS_ERR(inode) && root != sub_root) {
c71bf099
YZ		 5105 down_read(&root->fs_info->cleanup_work_sem);
5106 if (!(inode->i_sb->s_flags & MS_RDONLY))
66b4ffd1 5107 ret = btrfs_orphan_cleanup(sub_root);
c71bf099 5108 up_read(&root->fs_info->cleanup_work_sem);
66b4ffd1
JB		 5109 if (ret)
5110 inode = ERR_PTR(ret);
c71bf099
YZ		 5111 }
5112
3de4586c
CM		 5113 return inode;
5114}
5115
fe15ce44 5116static int btrfs_dentry_delete(const struct dentry *dentry)
76dda93c
YZ		 5117{
5118 struct btrfs_root *root;
848cce0d 5119 struct inode *inode = dentry->d_inode;
76dda93c 5120
848cce0d
LZ		 5121 if (!inode && !IS_ROOT(dentry))
5122 inode = dentry->d_parent->d_inode;
76dda93c 5123
848cce0d
LZ		 5124 if (inode) {
5125 root = BTRFS_I(inode)->root;
efefb143
YZ		 5126 if (btrfs_root_refs(&root->root_item) == 0)
5127 return 1;
848cce0d
LZ		 5128
5129 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5130 return 1;
efefb143 5131 }
76dda93c
YZ		 5132 return 0;
5133}
5134
b4aff1f8
JB		 5135static void btrfs_dentry_release(struct dentry *dentry)
5136{
5137 if (dentry->d_fsdata)
5138 kfree(dentry->d_fsdata);
5139}
5140
3de4586c 5141static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
00cd8dd3 5142 unsigned int flags)
3de4586c 5143{
a66e7cc6
JB		 5144 struct dentry *ret;
5145
5146 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
a66e7cc6 5147 return ret;
39279cc3
CM		 5148}
5149
16cdcec7 5150unsigned char btrfs_filetype_table[] = {
39279cc3
CM		 5151 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5152};
5153
cbdf5a24
DW		 5154static int btrfs_real_readdir(struct file *filp, void *dirent,
5155 filldir_t filldir)
39279cc3 5156{
496ad9aa 5157 struct inode *inode = file_inode(filp);
39279cc3
CM		 5158 struct btrfs_root *root = BTRFS_I(inode)->root;
5159 struct btrfs_item *item;
5160 struct btrfs_dir_item *di;
5161 struct btrfs_key key;
5f39d397 5162 struct btrfs_key found_key;
39279cc3 5163 struct btrfs_path *path;
16cdcec7
MX		 5164 struct list_head ins_list;
5165 struct list_head del_list;
39279cc3 5166 int ret;
5f39d397 5167 struct extent_buffer *leaf;
39279cc3 5168 int slot;
39279cc3
CM		 5169 unsigned char d_type;
5170 int over = 0;
5171 u32 di_cur;
5172 u32 di_total;
5173 u32 di_len;
5174 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 5175 char tmp_name[32];
5176 char *name_ptr;
5177 int name_len;
16cdcec7 5178 int is_curr = 0; /* filp->f_pos points to the current index? */
39279cc3
CM		 5179
5180 /* FIXME, use a real flag for deciding about the key type */
5181 if (root->fs_info->tree_root == root)
5182 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 5183
3954401f
CM		 5184 /* special case for "." */
5185 if (filp->f_pos == 0) {
3765fefa
HS		 5186 over = filldir(dirent, ".", 1,
5187 filp->f_pos, btrfs_ino(inode), DT_DIR);
3954401f
CM		 5188 if (over)
5189 return 0;
5190 filp->f_pos = 1;
5191 }
3954401f
CM		 5192 /* special case for .., just use the back ref */
5193 if (filp->f_pos == 1) {
5ecc7e5d 5194 u64 pino = parent_ino(filp->f_path.dentry);
3954401f 5195 over = filldir(dirent, "..", 2,
3765fefa 5196 filp->f_pos, pino, DT_DIR);
3954401f 5197 if (over)
49593bfa 5198 return 0;
3954401f
CM		 5199 filp->f_pos = 2;
5200 }
49593bfa 5201 path = btrfs_alloc_path();
16cdcec7
MX		 5202 if (!path)
5203 return -ENOMEM;
ff5714cc 5204
026fd317 5205 path->reada = 1;
49593bfa 5206
16cdcec7
MX		 5207 if (key_type == BTRFS_DIR_INDEX_KEY) {
5208 INIT_LIST_HEAD(&ins_list);
5209 INIT_LIST_HEAD(&del_list);
5210 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5211 }
5212
39279cc3
CM		 5213 btrfs_set_key_type(&key, key_type);
5214 key.offset = filp->f_pos;
33345d01 5215 key.objectid = btrfs_ino(inode);
5f39d397 5216
39279cc3
CM		 5217 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5218 if (ret < 0)
5219 goto err;
49593bfa
DW		 5220
5221 while (1) {
5f39d397 5222 leaf = path->nodes[0];
39279cc3 5223 slot = path->slots[0];
b9e03af0
LZ		 5224 if (slot >= btrfs_header_nritems(leaf)) {
5225 ret = btrfs_next_leaf(root, path);
5226 if (ret < 0)
5227 goto err;
5228 else if (ret > 0)
5229 break;
5230 continue;
39279cc3 5231 }
3de4586c 5232
5f39d397
CM		 5233 item = btrfs_item_nr(leaf, slot);
5234 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5235
5236 if (found_key.objectid != key.objectid)
39279cc3 5237 break;
5f39d397 5238 if (btrfs_key_type(&found_key) != key_type)
39279cc3 5239 break;
5f39d397 5240 if (found_key.offset < filp->f_pos)
b9e03af0 5241 goto next;
16cdcec7
MX		 5242 if (key_type == BTRFS_DIR_INDEX_KEY &&
5243 btrfs_should_delete_dir_index(&del_list,
5244 found_key.offset))
5245 goto next;
5f39d397
CM		 5246
5247 filp->f_pos = found_key.offset;
16cdcec7 5248 is_curr = 1;
49593bfa 5249
39279cc3
CM		 5250 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5251 di_cur = 0;
5f39d397 5252 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 5253
5254 while (di_cur < di_total) {
5f39d397
CM		 5255 struct btrfs_key location;
5256
22a94d44
JB		 5257 if (verify_dir_item(root, leaf, di))
5258 break;
5259
5f39d397 5260 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 5261 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 5262 name_ptr = tmp_name;
5263 } else {
5264 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 5265 if (!name_ptr) {
5266 ret = -ENOMEM;
5267 goto err;
5268 }
5f39d397
CM		 5269 }
5270 read_extent_buffer(leaf, name_ptr,
5271 (unsigned long)(di + 1), name_len);
5272
5273 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5274 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c 5275
fede766f 5276
3de4586c 5277 /* is this a reference to our own snapshot? If so
8c9c2bf7
AJ		 5278 * skip it.
5279 *
5280 * In contrast to old kernels, we insert the snapshot's
5281 * dir item and dir index after it has been created, so
5282 * we won't find a reference to our own snapshot. We
5283 * still keep the following code for backward
5284 * compatibility.
3de4586c
CM		 5285 */
5286 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5287 location.objectid == root->root_key.objectid) {
5288 over = 0;
5289 goto skip;
5290 }
5f39d397 5291 over = filldir(dirent, name_ptr, name_len,
49593bfa 5292 found_key.offset, location.objectid,
39279cc3 5293 d_type);
5f39d397 5294
3de4586c 5295skip:
5f39d397
CM		 5296 if (name_ptr != tmp_name)
5297 kfree(name_ptr);
5298
39279cc3
CM		 5299 if (over)
5300 goto nopos;
5103e947 5301 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 5302 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 5303 di_cur += di_len;
5304 di = (struct btrfs_dir_item *)((char *)di + di_len);
5305 }
b9e03af0
LZ		 5306next:
5307 path->slots[0]++;
39279cc3 5308 }
49593bfa 5309
16cdcec7
MX		 5310 if (key_type == BTRFS_DIR_INDEX_KEY) {
5311 if (is_curr)
5312 filp->f_pos++;
5313 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
5314 &ins_list);
5315 if (ret)
5316 goto nopos;
5317 }
5318
49593bfa 5319 /* Reached end of directory/root. Bump pos past the last item. */
5e591a07 5320 if (key_type == BTRFS_DIR_INDEX_KEY)
406266ab
JE		 5321 /*
5322 * 32-bit glibc will use getdents64, but then strtol -
5323 * so the last number we can serve is this.
5324 */
5325 filp->f_pos = 0x7fffffff;
5e591a07
YZ		 5326 else
5327 filp->f_pos++;
39279cc3
CM		 5328nopos:
5329 ret = 0;
5330err:
16cdcec7
MX		 5331 if (key_type == BTRFS_DIR_INDEX_KEY)
5332 btrfs_put_delayed_items(&ins_list, &del_list);
39279cc3 5333 btrfs_free_path(path);
39279cc3
CM		 5334 return ret;
5335}
5336
a9185b41 5337int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
39279cc3
CM		 5338{
5339 struct btrfs_root *root = BTRFS_I(inode)->root;
5340 struct btrfs_trans_handle *trans;
5341 int ret = 0;
0af3d00b 5342 bool nolock = false;
39279cc3 5343
72ac3c0d 5344 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4ca8b41e
CM		 5345 return 0;
5346
83eea1f1 5347 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
82d5902d 5348 nolock = true;
0af3d00b 5349
a9185b41 5350 if (wbc->sync_mode == WB_SYNC_ALL) {
0af3d00b 5351 if (nolock)
7a7eaa40 5352 trans = btrfs_join_transaction_nolock(root);
0af3d00b 5353 else
7a7eaa40 5354 trans = btrfs_join_transaction(root);
3612b495
TI		 5355 if (IS_ERR(trans))
5356 return PTR_ERR(trans);
a698d075 5357 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 5358 }
5359 return ret;
5360}
5361
5362/*
54aa1f4d 5363 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 5364 * inode changes. But, it is most likely to find the inode in cache.
5365 * FIXME, needs more benchmarking...there are no reasons other than performance
5366 * to keep or drop this code.
5367 */
48a3b636 5368static int btrfs_dirty_inode(struct inode *inode)
39279cc3
CM		 5369{
5370 struct btrfs_root *root = BTRFS_I(inode)->root;
5371 struct btrfs_trans_handle *trans;
8929ecfa
YZ		 5372 int ret;
5373
72ac3c0d 5374 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
22c44fe6 5375 return 0;
39279cc3 5376
7a7eaa40 5377 trans = btrfs_join_transaction(root);
22c44fe6
JB		 5378 if (IS_ERR(trans))
5379 return PTR_ERR(trans);
8929ecfa
YZ		 5380
5381 ret = btrfs_update_inode(trans, root, inode);
94b60442
CM		 5382 if (ret && ret == -ENOSPC) {
5383 /* whoops, lets try again with the full transaction */
5384 btrfs_end_transaction(trans, root);
5385 trans = btrfs_start_transaction(root, 1);
22c44fe6
JB		 5386 if (IS_ERR(trans))
5387 return PTR_ERR(trans);
8929ecfa 5388
94b60442 5389 ret = btrfs_update_inode(trans, root, inode);
94b60442 5390 }
39279cc3 5391 btrfs_end_transaction(trans, root);
16cdcec7
MX		 5392 if (BTRFS_I(inode)->delayed_node)
5393 btrfs_balance_delayed_items(root);
22c44fe6
JB		 5394
5395 return ret;
5396}
5397
5398/*
5399 * This is a copy of file_update_time. We need this so we can return error on
5400 * ENOSPC for updating the inode in the case of file write and mmap writes.
5401 */
e41f941a
JB		 5402static int btrfs_update_time(struct inode *inode, struct timespec *now,
5403 int flags)
22c44fe6 5404{
2bc55652
AB		 5405 struct btrfs_root *root = BTRFS_I(inode)->root;
5406
5407 if (btrfs_root_readonly(root))
5408 return -EROFS;
5409
e41f941a 5410 if (flags & S_VERSION)
22c44fe6 5411 inode_inc_iversion(inode);
e41f941a
JB		 5412 if (flags & S_CTIME)
5413 inode->i_ctime = *now;
5414 if (flags & S_MTIME)
5415 inode->i_mtime = *now;
5416 if (flags & S_ATIME)
5417 inode->i_atime = *now;
5418 return btrfs_dirty_inode(inode);
39279cc3
CM		 5419}
5420
d352ac68
CM		 5421/*
5422 * find the highest existing sequence number in a directory
5423 * and then set the in-memory index_cnt variable to reflect
5424 * free sequence numbers
5425 */
aec7477b
JB		 5426static int btrfs_set_inode_index_count(struct inode *inode)
5427{
5428 struct btrfs_root *root = BTRFS_I(inode)->root;
5429 struct btrfs_key key, found_key;
5430 struct btrfs_path *path;
5431 struct extent_buffer *leaf;
5432 int ret;
5433
33345d01 5434 key.objectid = btrfs_ino(inode);
aec7477b
JB		 5435 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5436 key.offset = (u64)-1;
5437
5438 path = btrfs_alloc_path();
5439 if (!path)
5440 return -ENOMEM;
5441
5442 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5443 if (ret < 0)
5444 goto out;
5445 /* FIXME: we should be able to handle this */
5446 if (ret == 0)
5447 goto out;
5448 ret = 0;
5449
5450 /*
5451 * MAGIC NUMBER EXPLANATION:
5452 * since we search a directory based on f_pos we have to start at 2
5453 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5454 * else has to start at 2
5455 */
5456 if (path->slots[0] == 0) {
5457 BTRFS_I(inode)->index_cnt = 2;
5458 goto out;
5459 }
5460
5461 path->slots[0]--;
5462
5463 leaf = path->nodes[0];
5464 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5465
33345d01 5466 if (found_key.objectid != btrfs_ino(inode) ||
aec7477b
JB		 5467 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5468 BTRFS_I(inode)->index_cnt = 2;
5469 goto out;
5470 }
5471
5472 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5473out:
5474 btrfs_free_path(path);
5475 return ret;
5476}
5477
d352ac68
CM		 5478/*
5479 * helper to find a free sequence number in a given directory. This current
5480 * code is very simple, later versions will do smarter things in the btree
5481 */
3de4586c 5482int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 5483{
5484 int ret = 0;
5485
5486 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
16cdcec7
MX		 5487 ret = btrfs_inode_delayed_dir_index_count(dir);
5488 if (ret) {
5489 ret = btrfs_set_inode_index_count(dir);
5490 if (ret)
5491 return ret;
5492 }
aec7477b
JB		 5493 }
5494
00e4e6b3 5495 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 5496 BTRFS_I(dir)->index_cnt++;
5497
5498 return ret;
5499}
5500
39279cc3
CM		 5501static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5502 struct btrfs_root *root,
aec7477b 5503 struct inode *dir,
9c58309d 5504 const char *name, int name_len,
175a4eb7
AV		 5505 u64 ref_objectid, u64 objectid,
5506 umode_t mode, u64 *index)
39279cc3
CM		 5507{
5508 struct inode *inode;
5f39d397 5509 struct btrfs_inode_item *inode_item;
39279cc3 5510 struct btrfs_key *location;
5f39d397 5511 struct btrfs_path *path;
9c58309d
CM		 5512 struct btrfs_inode_ref *ref;
5513 struct btrfs_key key[2];
5514 u32 sizes[2];
5515 unsigned long ptr;
39279cc3
CM		 5516 int ret;
5517 int owner;
5518
5f39d397 5519 path = btrfs_alloc_path();
d8926bb3
MF		 5520 if (!path)
5521 return ERR_PTR(-ENOMEM);
5f39d397 5522
39279cc3 5523 inode = new_inode(root->fs_info->sb);
8fb27640
YS		 5524 if (!inode) {
5525 btrfs_free_path(path);
39279cc3 5526 return ERR_PTR(-ENOMEM);
8fb27640 5527 }
39279cc3 5528
581bb050
LZ		 5529 /*
5530 * we have to initialize this early, so we can reclaim the inode
5531 * number if we fail afterwards in this function.
5532 */
5533 inode->i_ino = objectid;
5534
aec7477b 5535 if (dir) {
1abe9b8a 5536 trace_btrfs_inode_request(dir);
5537
3de4586c 5538 ret = btrfs_set_inode_index(dir, index);
09771430 5539 if (ret) {
8fb27640 5540 btrfs_free_path(path);
09771430 5541 iput(inode);
aec7477b 5542 return ERR_PTR(ret);
09771430 5543 }
aec7477b
JB		 5544 }
5545 /*
5546 * index_cnt is ignored for everything but a dir,
5547 * btrfs_get_inode_index_count has an explanation for the magic
5548 * number
5549 */
5550 BTRFS_I(inode)->index_cnt = 2;
39279cc3 5551 BTRFS_I(inode)->root = root;
e02119d5 5552 BTRFS_I(inode)->generation = trans->transid;
76195853 5553 inode->i_generation = BTRFS_I(inode)->generation;
b888db2b 5554
5dc562c5
JB		 5555 /*
5556 * We could have gotten an inode number from somebody who was fsynced
5557 * and then removed in this same transaction, so let's just set full
5558 * sync since it will be a full sync anyway and this will blow away the
5559 * old info in the log.
5560 */
5561 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5562
569254b0 5563 if (S_ISDIR(mode))
39279cc3
CM		 5564 owner = 0;
5565 else
5566 owner = 1;
9c58309d
CM		 5567
5568 key[0].objectid = objectid;
5569 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5570 key[0].offset = 0;
5571
f186373f
MF		 5572 /*
5573 * Start new inodes with an inode_ref. This is slightly more
5574 * efficient for small numbers of hard links since they will
5575 * be packed into one item. Extended refs will kick in if we
5576 * add more hard links than can fit in the ref item.
5577 */
9c58309d
CM		 5578 key[1].objectid = objectid;
5579 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5580 key[1].offset = ref_objectid;
5581
5582 sizes[0] = sizeof(struct btrfs_inode_item);
5583 sizes[1] = name_len + sizeof(*ref);
5584
b9473439 5585 path->leave_spinning = 1;
9c58309d
CM		 5586 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5587 if (ret != 0)
5f39d397
CM		 5588 goto fail;
5589
ecc11fab 5590 inode_init_owner(inode, dir, mode);
a76a3cd4 5591 inode_set_bytes(inode, 0);
39279cc3 5592 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5f39d397
CM		 5593 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5594 struct btrfs_inode_item);
293f7e07
LZ		 5595 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5596 sizeof(*inode_item));
e02119d5 5597 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d
CM		 5598
5599 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5600 struct btrfs_inode_ref);
5601 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
00e4e6b3 5602 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
9c58309d
CM		 5603 ptr = (unsigned long)(ref + 1);
5604 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5605
5f39d397
CM		 5606 btrfs_mark_buffer_dirty(path->nodes[0]);
5607 btrfs_free_path(path);
5608
39279cc3
CM		 5609 location = &BTRFS_I(inode)->location;
5610 location->objectid = objectid;
39279cc3
CM		 5611 location->offset = 0;
5612 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5613
6cbff00f
CH		 5614 btrfs_inherit_iflags(inode, dir);
5615
569254b0 5616 if (S_ISREG(mode)) {
94272164
CM		 5617 if (btrfs_test_opt(root, NODATASUM))
5618 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
213490b3 5619 if (btrfs_test_opt(root, NODATACOW))
f2bdf9a8
JB		 5620 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5621 BTRFS_INODE_NODATASUM;
94272164
CM		 5622 }
5623
39279cc3 5624 insert_inode_hash(inode);
5d4f98a2 5625 inode_tree_add(inode);
1abe9b8a 5626
5627 trace_btrfs_inode_new(inode);
1973f0fa 5628 btrfs_set_inode_last_trans(trans, inode);
1abe9b8a 5629
8ea05e3a
AB		 5630 btrfs_update_root_times(trans, root);
5631
39279cc3 5632 return inode;
5f39d397 5633fail:
aec7477b
JB		 5634 if (dir)
5635 BTRFS_I(dir)->index_cnt--;
5f39d397 5636 btrfs_free_path(path);
09771430 5637 iput(inode);
5f39d397 5638 return ERR_PTR(ret);
39279cc3
CM		 5639}
5640
5641static inline u8 btrfs_inode_type(struct inode *inode)
5642{
5643 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5644}
5645
d352ac68
CM		 5646/*
5647 * utility function to add 'inode' into 'parent_inode' with
5648 * a give name and a given sequence number.
5649 * if 'add_backref' is true, also insert a backref from the
5650 * inode to the parent directory.
5651 */
e02119d5
CM		 5652int btrfs_add_link(struct btrfs_trans_handle *trans,
5653 struct inode *parent_inode, struct inode *inode,
5654 const char *name, int name_len, int add_backref, u64 index)
39279cc3 5655{
4df27c4d 5656 int ret = 0;
39279cc3 5657 struct btrfs_key key;
e02119d5 5658 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
33345d01
LZ		 5659 u64 ino = btrfs_ino(inode);
5660 u64 parent_ino = btrfs_ino(parent_inode);
5f39d397 5661
33345d01 5662 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 5663 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5664 } else {
33345d01 5665 key.objectid = ino;
4df27c4d
YZ		 5666 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5667 key.offset = 0;
5668 }
5669
33345d01 5670 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 5671 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5672 key.objectid, root->root_key.objectid,
33345d01 5673 parent_ino, index, name, name_len);
4df27c4d 5674 } else if (add_backref) {
33345d01
LZ		 5675 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5676 parent_ino, index);
4df27c4d 5677 }
39279cc3 5678
79787eaa
JM		 5679 /* Nothing to clean up yet */
5680 if (ret)
5681 return ret;
4df27c4d 5682
79787eaa
JM		 5683 ret = btrfs_insert_dir_item(trans, root, name, name_len,
5684 parent_inode, &key,
5685 btrfs_inode_type(inode), index);
9c52057c 5686 if (ret == -EEXIST || ret == -EOVERFLOW)
79787eaa
JM		 5687 goto fail_dir_item;
5688 else if (ret) {
5689 btrfs_abort_transaction(trans, root, ret);
5690 return ret;
39279cc3 5691 }
79787eaa
JM		 5692
5693 btrfs_i_size_write(parent_inode, parent_inode->i_size +
5694 name_len * 2);
0c4d2d95 5695 inode_inc_iversion(parent_inode);
79787eaa
JM		 5696 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5697 ret = btrfs_update_inode(trans, root, parent_inode);
5698 if (ret)
5699 btrfs_abort_transaction(trans, root, ret);
39279cc3 5700 return ret;
fe66a05a
CM		 5701
5702fail_dir_item:
5703 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5704 u64 local_index;
5705 int err;
5706 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5707 key.objectid, root->root_key.objectid,
5708 parent_ino, &local_index, name, name_len);
5709
5710 } else if (add_backref) {
5711 u64 local_index;
5712 int err;
5713
5714 err = btrfs_del_inode_ref(trans, root, name, name_len,
5715 ino, parent_ino, &local_index);
5716 }
5717 return ret;
39279cc3
CM		 5718}
5719
5720static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
a1b075d2
JB		 5721 struct inode *dir, struct dentry *dentry,
5722 struct inode *inode, int backref, u64 index)
39279cc3 5723{
a1b075d2
JB		 5724 int err = btrfs_add_link(trans, dir, inode,
5725 dentry->d_name.name, dentry->d_name.len,
5726 backref, index);
39279cc3
CM		 5727 if (err > 0)
5728 err = -EEXIST;
5729 return err;
5730}
5731
618e21d5 5732static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1a67aafb 5733 umode_t mode, dev_t rdev)
618e21d5
JB		 5734{
5735 struct btrfs_trans_handle *trans;
5736 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 5737 struct inode *inode = NULL;
618e21d5
JB		 5738 int err;
5739 int drop_inode = 0;
5740 u64 objectid;
00e4e6b3 5741 u64 index = 0;
618e21d5
JB		 5742
5743 if (!new_valid_dev(rdev))
5744 return -EINVAL;
5745
9ed74f2d
JB		 5746 /*
5747 * 2 for inode item and ref
5748 * 2 for dir items
5749 * 1 for xattr if selinux is on
5750 */
a22285a6
YZ		 5751 trans = btrfs_start_transaction(root, 5);
5752 if (IS_ERR(trans))
5753 return PTR_ERR(trans);
1832a6d5 5754
581bb050
LZ		 5755 err = btrfs_find_free_ino(root, &objectid);
5756 if (err)
5757 goto out_unlock;
5758
aec7477b 5759 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 5760 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 5761 mode, &index);
7cf96da3
TI		 5762 if (IS_ERR(inode)) {
5763 err = PTR_ERR(inode);
618e21d5 5764 goto out_unlock;
7cf96da3 5765 }
618e21d5 5766
2a7dba39 5767 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 5768 if (err) {
5769 drop_inode = 1;
5770 goto out_unlock;
5771 }
5772
ad19db71
CS		 5773 /*
5774 * If the active LSM wants to access the inode during
5775 * d_instantiate it needs these. Smack checks to see
5776 * if the filesystem supports xattrs by looking at the
5777 * ops vector.
5778 */
5779
5780 inode->i_op = &btrfs_special_inode_operations;
a1b075d2 5781 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
618e21d5
JB		 5782 if (err)
5783 drop_inode = 1;
5784 else {
618e21d5 5785 init_special_inode(inode, inode->i_mode, rdev);
1b4ab1bb 5786 btrfs_update_inode(trans, root, inode);
08c422c2 5787 d_instantiate(dentry, inode);
618e21d5 5788 }
618e21d5 5789out_unlock:
7ad85bb7 5790 btrfs_end_transaction(trans, root);
b53d3f5d 5791 btrfs_btree_balance_dirty(root);
618e21d5
JB		 5792 if (drop_inode) {
5793 inode_dec_link_count(inode);
5794 iput(inode);
5795 }
618e21d5
JB		 5796 return err;
5797}
5798
39279cc3 5799static int btrfs_create(struct inode *dir, struct dentry *dentry,
ebfc3b49 5800 umode_t mode, bool excl)
39279cc3
CM		 5801{
5802 struct btrfs_trans_handle *trans;
5803 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 5804 struct inode *inode = NULL;
43baa579 5805 int drop_inode_on_err = 0;
a22285a6 5806 int err;
39279cc3 5807 u64 objectid;
00e4e6b3 5808 u64 index = 0;
39279cc3 5809
9ed74f2d
JB		 5810 /*
5811 * 2 for inode item and ref
5812 * 2 for dir items
5813 * 1 for xattr if selinux is on
5814 */
a22285a6
YZ		 5815 trans = btrfs_start_transaction(root, 5);
5816 if (IS_ERR(trans))
5817 return PTR_ERR(trans);
9ed74f2d 5818
581bb050
LZ		 5819 err = btrfs_find_free_ino(root, &objectid);
5820 if (err)
5821 goto out_unlock;
5822
aec7477b 5823 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 5824 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 5825 mode, &index);
7cf96da3
TI		 5826 if (IS_ERR(inode)) {
5827 err = PTR_ERR(inode);
39279cc3 5828 goto out_unlock;
7cf96da3 5829 }
43baa579 5830 drop_inode_on_err = 1;
39279cc3 5831
2a7dba39 5832 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
43baa579 5833 if (err)
33268eaf 5834 goto out_unlock;
33268eaf 5835
9185aa58
FB		 5836 err = btrfs_update_inode(trans, root, inode);
5837 if (err)
5838 goto out_unlock;
5839
ad19db71
CS		 5840 /*
5841 * If the active LSM wants to access the inode during
5842 * d_instantiate it needs these. Smack checks to see
5843 * if the filesystem supports xattrs by looking at the
5844 * ops vector.
5845 */
5846 inode->i_fop = &btrfs_file_operations;
5847 inode->i_op = &btrfs_file_inode_operations;
5848
a1b075d2 5849 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3 5850 if (err)
43baa579
FB		 5851 goto out_unlock;
5852
5853 inode->i_mapping->a_ops = &btrfs_aops;
5854 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5855 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5856 d_instantiate(dentry, inode);
5857
39279cc3 5858out_unlock:
7ad85bb7 5859 btrfs_end_transaction(trans, root);
43baa579 5860 if (err && drop_inode_on_err) {
39279cc3
CM		 5861 inode_dec_link_count(inode);
5862 iput(inode);
5863 }
b53d3f5d 5864 btrfs_btree_balance_dirty(root);
39279cc3
CM		 5865 return err;
5866}
5867
5868static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5869 struct dentry *dentry)
5870{
5871 struct btrfs_trans_handle *trans;
5872 struct btrfs_root *root = BTRFS_I(dir)->root;
5873 struct inode *inode = old_dentry->d_inode;
00e4e6b3 5874 u64 index;
39279cc3
CM		 5875 int err;
5876 int drop_inode = 0;
5877
4a8be425
TH		 5878 /* do not allow sys_link's with other subvols of the same device */
5879 if (root->objectid != BTRFS_I(inode)->root->objectid)
3ab3564f 5880 return -EXDEV;
4a8be425 5881
f186373f 5882 if (inode->i_nlink >= BTRFS_LINK_MAX)
c055e99e 5883 return -EMLINK;
4a8be425 5884
3de4586c 5885 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 5886 if (err)
5887 goto fail;
5888
a22285a6 5889 /*
7e6b6465 5890 * 2 items for inode and inode ref
a22285a6 5891 * 2 items for dir items
7e6b6465 5892 * 1 item for parent inode
a22285a6 5893 */
7e6b6465 5894 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 5895 if (IS_ERR(trans)) {
5896 err = PTR_ERR(trans);
5897 goto fail;
5898 }
5f39d397 5899
3153495d 5900 btrfs_inc_nlink(inode);
0c4d2d95 5901 inode_inc_iversion(inode);
3153495d 5902 inode->i_ctime = CURRENT_TIME;
7de9c6ee 5903 ihold(inode);
e9976151 5904 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
aec7477b 5905
a1b075d2 5906 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5f39d397 5907
a5719521 5908 if (err) {
54aa1f4d 5909 drop_inode = 1;
a5719521 5910 } else {
10d9f309 5911 struct dentry *parent = dentry->d_parent;
a5719521 5912 err = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 5913 if (err)
5914 goto fail;
08c422c2 5915 d_instantiate(dentry, inode);
6a912213 5916 btrfs_log_new_name(trans, inode, NULL, parent);
a5719521 5917 }
39279cc3 5918
7ad85bb7 5919 btrfs_end_transaction(trans, root);
1832a6d5 5920fail:
39279cc3
CM		 5921 if (drop_inode) {
5922 inode_dec_link_count(inode);
5923 iput(inode);
5924 }
b53d3f5d 5925 btrfs_btree_balance_dirty(root);
39279cc3
CM		 5926 return err;
5927}
5928
18bb1db3 5929static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
39279cc3 5930{
b9d86667 5931 struct inode *inode = NULL;
39279cc3
CM		 5932 struct btrfs_trans_handle *trans;
5933 struct btrfs_root *root = BTRFS_I(dir)->root;
5934 int err = 0;
5935 int drop_on_err = 0;
b9d86667 5936 u64 objectid = 0;
00e4e6b3 5937 u64 index = 0;
39279cc3 5938
9ed74f2d
JB		 5939 /*
5940 * 2 items for inode and ref
5941 * 2 items for dir items
5942 * 1 for xattr if selinux is on
5943 */
a22285a6
YZ		 5944 trans = btrfs_start_transaction(root, 5);
5945 if (IS_ERR(trans))
5946 return PTR_ERR(trans);
39279cc3 5947
581bb050
LZ		 5948 err = btrfs_find_free_ino(root, &objectid);
5949 if (err)
5950 goto out_fail;
5951
aec7477b 5952 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 5953 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 5954 S_IFDIR | mode, &index);
39279cc3
CM		 5955 if (IS_ERR(inode)) {
5956 err = PTR_ERR(inode);
5957 goto out_fail;
5958 }
5f39d397 5959
39279cc3 5960 drop_on_err = 1;
33268eaf 5961
2a7dba39 5962 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 5963 if (err)
5964 goto out_fail;
5965
39279cc3
CM		 5966 inode->i_op = &btrfs_dir_inode_operations;
5967 inode->i_fop = &btrfs_dir_file_operations;
39279cc3 5968
dbe674a9 5969 btrfs_i_size_write(inode, 0);
39279cc3
CM		 5970 err = btrfs_update_inode(trans, root, inode);
5971 if (err)
5972 goto out_fail;
5f39d397 5973
a1b075d2
JB		 5974 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5975 dentry->d_name.len, 0, index);
39279cc3
CM		 5976 if (err)
5977 goto out_fail;
5f39d397 5978
39279cc3
CM		 5979 d_instantiate(dentry, inode);
5980 drop_on_err = 0;
39279cc3
CM		 5981
5982out_fail:
7ad85bb7 5983 btrfs_end_transaction(trans, root);
39279cc3
CM		 5984 if (drop_on_err)
5985 iput(inode);
b53d3f5d 5986 btrfs_btree_balance_dirty(root);
39279cc3
CM		 5987 return err;
5988}
5989
d352ac68
CM		 5990/* helper for btfs_get_extent. Given an existing extent in the tree,
5991 * and an extent that you want to insert, deal with overlap and insert
5992 * the new extent into the tree.
5993 */
3b951516
CM		 5994static int merge_extent_mapping(struct extent_map_tree *em_tree,
5995 struct extent_map *existing,
e6dcd2dc
CM		 5996 struct extent_map *em,
5997 u64 map_start, u64 map_len)
3b951516
CM		 5998{
5999 u64 start_diff;
3b951516 6000
e6dcd2dc
CM		 6001 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6002 start_diff = map_start - em->start;
6003 em->start = map_start;
6004 em->len = map_len;
c8b97818
CM		 6005 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6006 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 6007 em->block_start += start_diff;
c8b97818
CM		 6008 em->block_len -= start_diff;
6009 }
09a2a8f9 6010 return add_extent_mapping(em_tree, em, 0);
3b951516
CM		 6011}
6012
c8b97818
CM		 6013static noinline int uncompress_inline(struct btrfs_path *path,
6014 struct inode *inode, struct page *page,
6015 size_t pg_offset, u64 extent_offset,
6016 struct btrfs_file_extent_item *item)
6017{
6018 int ret;
6019 struct extent_buffer *leaf = path->nodes[0];
6020 char *tmp;
6021 size_t max_size;
6022 unsigned long inline_size;
6023 unsigned long ptr;
261507a0 6024 int compress_type;
c8b97818
CM		 6025
6026 WARN_ON(pg_offset != 0);
261507a0 6027 compress_type = btrfs_file_extent_compression(leaf, item);
c8b97818
CM		 6028 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6029 inline_size = btrfs_file_extent_inline_item_len(leaf,
6030 btrfs_item_nr(leaf, path->slots[0]));
6031 tmp = kmalloc(inline_size, GFP_NOFS);
8d413713
TI		 6032 if (!tmp)
6033 return -ENOMEM;
c8b97818
CM		 6034 ptr = btrfs_file_extent_inline_start(item);
6035
6036 read_extent_buffer(leaf, tmp, ptr, inline_size);
6037
5b050f04 6038 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
261507a0
LZ		 6039 ret = btrfs_decompress(compress_type, tmp, page,
6040 extent_offset, inline_size, max_size);
c8b97818 6041 if (ret) {
7ac687d9 6042 char *kaddr = kmap_atomic(page);
c8b97818
CM		 6043 unsigned long copy_size = min_t(u64,
6044 PAGE_CACHE_SIZE - pg_offset,
6045 max_size - extent_offset);
6046 memset(kaddr + pg_offset, 0, copy_size);
7ac687d9 6047 kunmap_atomic(kaddr);
c8b97818
CM		 6048 }
6049 kfree(tmp);
6050 return 0;
6051}
6052
d352ac68
CM		 6053/*
6054 * a bit scary, this does extent mapping from logical file offset to the disk.
d397712b
CM		 6055 * the ugly parts come from merging extents from the disk with the in-ram
6056 * representation. This gets more complex because of the data=ordered code,
d352ac68
CM		 6057 * where the in-ram extents might be locked pending data=ordered completion.
6058 *
6059 * This also copies inline extents directly into the page.
6060 */
d397712b 6061
a52d9a80 6062struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 6063 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 6064 int create)
6065{
6066 int ret;
6067 int err = 0;
db94535d 6068 u64 bytenr;
a52d9a80
CM		 6069 u64 extent_start = 0;
6070 u64 extent_end = 0;
33345d01 6071 u64 objectid = btrfs_ino(inode);
a52d9a80 6072 u32 found_type;
f421950f 6073 struct btrfs_path *path = NULL;
a52d9a80
CM		 6074 struct btrfs_root *root = BTRFS_I(inode)->root;
6075 struct btrfs_file_extent_item *item;
5f39d397
CM		 6076 struct extent_buffer *leaf;
6077 struct btrfs_key found_key;
a52d9a80
CM		 6078 struct extent_map *em = NULL;
6079 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 6080 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 6081 struct btrfs_trans_handle *trans = NULL;
261507a0 6082 int compress_type;
a52d9a80 6083
a52d9a80 6084again:
890871be 6085 read_lock(&em_tree->lock);
d1310b2e 6086 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 6087 if (em)
6088 em->bdev = root->fs_info->fs_devices->latest_bdev;
890871be 6089 read_unlock(&em_tree->lock);
d1310b2e 6090
a52d9a80 6091 if (em) {
e1c4b745
CM		 6092 if (em->start > start || em->start + em->len <= start)
6093 free_extent_map(em);
6094 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 6095 free_extent_map(em);
6096 else
6097 goto out;
a52d9a80 6098 }
172ddd60 6099 em = alloc_extent_map();
a52d9a80 6100 if (!em) {
d1310b2e
CM		 6101 err = -ENOMEM;
6102 goto out;
a52d9a80 6103 }
e6dcd2dc 6104 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 6105 em->start = EXTENT_MAP_HOLE;
445a6944 6106 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 6107 em->len = (u64)-1;
c8b97818 6108 em->block_len = (u64)-1;
f421950f
CM		 6109
6110 if (!path) {
6111 path = btrfs_alloc_path();
026fd317
JB		 6112 if (!path) {
6113 err = -ENOMEM;
6114 goto out;
6115 }
6116 /*
6117 * Chances are we'll be called again, so go ahead and do
6118 * readahead
6119 */
6120 path->reada = 1;
f421950f
CM		 6121 }
6122
179e29e4
CM		 6123 ret = btrfs_lookup_file_extent(trans, root, path,
6124 objectid, start, trans != NULL);
a52d9a80
CM		 6125 if (ret < 0) {
6126 err = ret;
6127 goto out;
6128 }
6129
6130 if (ret != 0) {
6131 if (path->slots[0] == 0)
6132 goto not_found;
6133 path->slots[0]--;
6134 }
6135
5f39d397
CM		 6136 leaf = path->nodes[0];
6137 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 6138 struct btrfs_file_extent_item);
a52d9a80 6139 /* are we inside the extent that was found? */
5f39d397
CM		 6140 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6141 found_type = btrfs_key_type(&found_key);
6142 if (found_key.objectid != objectid ||
a52d9a80
CM		 6143 found_type != BTRFS_EXTENT_DATA_KEY) {
6144 goto not_found;
6145 }
6146
5f39d397
CM		 6147 found_type = btrfs_file_extent_type(leaf, item);
6148 extent_start = found_key.offset;
261507a0 6149 compress_type = btrfs_file_extent_compression(leaf, item);
d899e052
YZ		 6150 if (found_type == BTRFS_FILE_EXTENT_REG ||
6151 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 6152 extent_end = extent_start +
db94535d 6153 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 6154 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6155 size_t size;
6156 size = btrfs_file_extent_inline_len(leaf, item);
fda2832f 6157 extent_end = ALIGN(extent_start + size, root->sectorsize);
9036c102
YZ		 6158 }
6159
6160 if (start >= extent_end) {
6161 path->slots[0]++;
6162 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6163 ret = btrfs_next_leaf(root, path);
6164 if (ret < 0) {
6165 err = ret;
6166 goto out;
a52d9a80 6167 }
9036c102
YZ		 6168 if (ret > 0)
6169 goto not_found;
6170 leaf = path->nodes[0];
a52d9a80 6171 }
9036c102
YZ		 6172 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6173 if (found_key.objectid != objectid ||
6174 found_key.type != BTRFS_EXTENT_DATA_KEY)
6175 goto not_found;
6176 if (start + len <= found_key.offset)
6177 goto not_found;
6178 em->start = start;
70c8a91c 6179 em->orig_start = start;
9036c102
YZ		 6180 em->len = found_key.offset - start;
6181 goto not_found_em;
6182 }
6183
cc95bef6 6184 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
d899e052
YZ		 6185 if (found_type == BTRFS_FILE_EXTENT_REG ||
6186 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
9036c102
YZ		 6187 em->start = extent_start;
6188 em->len = extent_end - extent_start;
ff5b7ee3
YZ		 6189 em->orig_start = extent_start -
6190 btrfs_file_extent_offset(leaf, item);
b4939680
JB		 6191 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6192 item);
db94535d
CM		 6193 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6194 if (bytenr == 0) {
5f39d397 6195 em->block_start = EXTENT_MAP_HOLE;
a52d9a80
CM		 6196 goto insert;
6197 }
261507a0 6198 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818 6199 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0 6200 em->compress_type = compress_type;
c8b97818 6201 em->block_start = bytenr;
b4939680 6202 em->block_len = em->orig_block_len;
c8b97818
CM		 6203 } else {
6204 bytenr += btrfs_file_extent_offset(leaf, item);
6205 em->block_start = bytenr;
6206 em->block_len = em->len;
d899e052
YZ		 6207 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6208 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
c8b97818 6209 }
a52d9a80
CM		 6210 goto insert;
6211 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 6212 unsigned long ptr;
a52d9a80 6213 char *map;
3326d1b0
CM		 6214 size_t size;
6215 size_t extent_offset;
6216 size_t copy_size;
a52d9a80 6217
689f9346 6218 em->block_start = EXTENT_MAP_INLINE;
c8b97818 6219 if (!page || create) {
689f9346 6220 em->start = extent_start;
9036c102 6221 em->len = extent_end - extent_start;
689f9346
Y		 6222 goto out;
6223 }
5f39d397 6224
9036c102
YZ		 6225 size = btrfs_file_extent_inline_len(leaf, item);
6226 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 6227 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 6228 size - extent_offset);
3326d1b0 6229 em->start = extent_start + extent_offset;
fda2832f 6230 em->len = ALIGN(copy_size, root->sectorsize);
b4939680 6231 em->orig_block_len = em->len;
70c8a91c 6232 em->orig_start = em->start;
261507a0 6233 if (compress_type) {
c8b97818 6234 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0
LZ		 6235 em->compress_type = compress_type;
6236 }
689f9346 6237 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 6238 if (create == 0 && !PageUptodate(page)) {
261507a0
LZ		 6239 if (btrfs_file_extent_compression(leaf, item) !=
6240 BTRFS_COMPRESS_NONE) {
c8b97818
CM		 6241 ret = uncompress_inline(path, inode, page,
6242 pg_offset,
6243 extent_offset, item);
79787eaa 6244 BUG_ON(ret); /* -ENOMEM */
c8b97818
CM		 6245 } else {
6246 map = kmap(page);
6247 read_extent_buffer(leaf, map + pg_offset, ptr,
6248 copy_size);
93c82d57
CM		 6249 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6250 memset(map + pg_offset + copy_size, 0,
6251 PAGE_CACHE_SIZE - pg_offset -
6252 copy_size);
6253 }
c8b97818
CM		 6254 kunmap(page);
6255 }
179e29e4
CM		 6256 flush_dcache_page(page);
6257 } else if (create && PageUptodate(page)) {
6bf7e080 6258 BUG();
179e29e4
CM		 6259 if (!trans) {
6260 kunmap(page);
6261 free_extent_map(em);
6262 em = NULL;
ff5714cc 6263
b3b4aa74 6264 btrfs_release_path(path);
7a7eaa40 6265 trans = btrfs_join_transaction(root);
ff5714cc 6266
3612b495
TI		 6267 if (IS_ERR(trans))
6268 return ERR_CAST(trans);
179e29e4
CM		 6269 goto again;
6270 }
c8b97818 6271 map = kmap(page);
70dec807 6272 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 6273 copy_size);
c8b97818 6274 kunmap(page);
179e29e4 6275 btrfs_mark_buffer_dirty(leaf);
a52d9a80 6276 }
d1310b2e 6277 set_extent_uptodate(io_tree, em->start,
507903b8 6278 extent_map_end(em) - 1, NULL, GFP_NOFS);
a52d9a80
CM		 6279 goto insert;
6280 } else {
31b1a2bd 6281 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
a52d9a80
CM		 6282 }
6283not_found:
6284 em->start = start;
70c8a91c 6285 em->orig_start = start;
d1310b2e 6286 em->len = len;
a52d9a80 6287not_found_em:
5f39d397 6288 em->block_start = EXTENT_MAP_HOLE;
9036c102 6289 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80 6290insert:
b3b4aa74 6291 btrfs_release_path(path);
d1310b2e 6292 if (em->start > start || extent_map_end(em) <= start) {
c2cf52eb
SK		 6293 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6294 (unsigned long long)em->start,
6295 (unsigned long long)em->len,
6296 (unsigned long long)start,
6297 (unsigned long long)len);
a52d9a80
CM		 6298 err = -EIO;
6299 goto out;
6300 }
d1310b2e
CM		 6301
6302 err = 0;
890871be 6303 write_lock(&em_tree->lock);
09a2a8f9 6304 ret = add_extent_mapping(em_tree, em, 0);
3b951516
CM		 6305 /* it is possible that someone inserted the extent into the tree
6306 * while we had the lock dropped. It is also possible that
6307 * an overlapping map exists in the tree
6308 */
a52d9a80 6309 if (ret == -EEXIST) {
3b951516 6310 struct extent_map *existing;
e6dcd2dc
CM		 6311
6312 ret = 0;
6313
3b951516 6314 existing = lookup_extent_mapping(em_tree, start, len);
e1c4b745
CM		 6315 if (existing && (existing->start > start ||
6316 existing->start + existing->len <= start)) {
6317 free_extent_map(existing);
6318 existing = NULL;
6319 }
3b951516
CM		 6320 if (!existing) {
6321 existing = lookup_extent_mapping(em_tree, em->start,
6322 em->len);
6323 if (existing) {
6324 err = merge_extent_mapping(em_tree, existing,
e6dcd2dc
CM		 6325 em, start,
6326 root->sectorsize);
3b951516
CM		 6327 free_extent_map(existing);
6328 if (err) {
6329 free_extent_map(em);
6330 em = NULL;
6331 }
6332 } else {
6333 err = -EIO;
3b951516
CM		 6334 free_extent_map(em);
6335 em = NULL;
6336 }
6337 } else {
6338 free_extent_map(em);
6339 em = existing;
e6dcd2dc 6340 err = 0;
a52d9a80 6341 }
a52d9a80 6342 }
890871be 6343 write_unlock(&em_tree->lock);
a52d9a80 6344out:
1abe9b8a 6345
f0bd95ea
TI		 6346 if (em)
6347 trace_btrfs_get_extent(root, em);
1abe9b8a 6348
f421950f
CM		 6349 if (path)
6350 btrfs_free_path(path);
a52d9a80
CM		 6351 if (trans) {
6352 ret = btrfs_end_transaction(trans, root);
d397712b 6353 if (!err)
a52d9a80
CM		 6354 err = ret;
6355 }
a52d9a80
CM		 6356 if (err) {
6357 free_extent_map(em);
a52d9a80
CM		 6358 return ERR_PTR(err);
6359 }
79787eaa 6360 BUG_ON(!em); /* Error is always set */
a52d9a80
CM		 6361 return em;
6362}
6363
ec29ed5b
CM		 6364struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6365 size_t pg_offset, u64 start, u64 len,
6366 int create)
6367{
6368 struct extent_map *em;
6369 struct extent_map *hole_em = NULL;
6370 u64 range_start = start;
6371 u64 end;
6372 u64 found;
6373 u64 found_end;
6374 int err = 0;
6375
6376 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6377 if (IS_ERR(em))
6378 return em;
6379 if (em) {
6380 /*
f9e4fb53
LB		 6381 * if our em maps to
6382 * - a hole or
6383 * - a pre-alloc extent,
6384 * there might actually be delalloc bytes behind it.
ec29ed5b 6385 */
f9e4fb53
LB		 6386 if (em->block_start != EXTENT_MAP_HOLE &&
6387 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
ec29ed5b
CM		 6388 return em;
6389 else
6390 hole_em = em;
6391 }
6392
6393 /* check to see if we've wrapped (len == -1 or similar) */
6394 end = start + len;
6395 if (end < start)
6396 end = (u64)-1;
6397 else
6398 end -= 1;
6399
6400 em = NULL;
6401
6402 /* ok, we didn't find anything, lets look for delalloc */
6403 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6404 end, len, EXTENT_DELALLOC, 1);
6405 found_end = range_start + found;
6406 if (found_end < range_start)
6407 found_end = (u64)-1;
6408
6409 /*
6410 * we didn't find anything useful, return
6411 * the original results from get_extent()
6412 */
6413 if (range_start > end || found_end <= start) {
6414 em = hole_em;
6415 hole_em = NULL;
6416 goto out;
6417 }
6418
6419 /* adjust the range_start to make sure it doesn't
6420 * go backwards from the start they passed in
6421 */
6422 range_start = max(start,range_start);
6423 found = found_end - range_start;
6424
6425 if (found > 0) {
6426 u64 hole_start = start;
6427 u64 hole_len = len;
6428
172ddd60 6429 em = alloc_extent_map();
ec29ed5b
CM		 6430 if (!em) {
6431 err = -ENOMEM;
6432 goto out;
6433 }
6434 /*
6435 * when btrfs_get_extent can't find anything it
6436 * returns one huge hole
6437 *
6438 * make sure what it found really fits our range, and
6439 * adjust to make sure it is based on the start from
6440 * the caller
6441 */
6442 if (hole_em) {
6443 u64 calc_end = extent_map_end(hole_em);
6444
6445 if (calc_end <= start || (hole_em->start > end)) {
6446 free_extent_map(hole_em);
6447 hole_em = NULL;
6448 } else {
6449 hole_start = max(hole_em->start, start);
6450 hole_len = calc_end - hole_start;
6451 }
6452 }
6453 em->bdev = NULL;
6454 if (hole_em && range_start > hole_start) {
6455 /* our hole starts before our delalloc, so we
6456 * have to return just the parts of the hole
6457 * that go until the delalloc starts
6458 */
6459 em->len = min(hole_len,
6460 range_start - hole_start);
6461 em->start = hole_start;
6462 em->orig_start = hole_start;
6463 /*
6464 * don't adjust block start at all,
6465 * it is fixed at EXTENT_MAP_HOLE
6466 */
6467 em->block_start = hole_em->block_start;
6468 em->block_len = hole_len;
f9e4fb53
LB		 6469 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6470 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
ec29ed5b
CM		 6471 } else {
6472 em->start = range_start;
6473 em->len = found;
6474 em->orig_start = range_start;
6475 em->block_start = EXTENT_MAP_DELALLOC;
6476 em->block_len = found;
6477 }
6478 } else if (hole_em) {
6479 return hole_em;
6480 }
6481out:
6482
6483 free_extent_map(hole_em);
6484 if (err) {
6485 free_extent_map(em);
6486 return ERR_PTR(err);
6487 }
6488 return em;
6489}
6490
4b46fce2
JB		 6491static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6492 u64 start, u64 len)
6493{
6494 struct btrfs_root *root = BTRFS_I(inode)->root;
6495 struct btrfs_trans_handle *trans;
70c8a91c 6496 struct extent_map *em;
4b46fce2
JB		 6497 struct btrfs_key ins;
6498 u64 alloc_hint;
6499 int ret;
4b46fce2 6500
7a7eaa40 6501 trans = btrfs_join_transaction(root);
3612b495
TI		 6502 if (IS_ERR(trans))
6503 return ERR_CAST(trans);
4b46fce2
JB		 6504
6505 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
6506
6507 alloc_hint = get_extent_allocation_hint(inode, start, len);
6508 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
81c9ad23 6509 alloc_hint, &ins, 1);
4b46fce2
JB		 6510 if (ret) {
6511 em = ERR_PTR(ret);
6512 goto out;
6513 }
6514
70c8a91c 6515 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
cc95bef6 6516 ins.offset, ins.offset, ins.offset, 0);
70c8a91c
JB		 6517 if (IS_ERR(em))
6518 goto out;
4b46fce2
JB		 6519
6520 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6521 ins.offset, ins.offset, 0);
6522 if (ret) {
6523 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6524 em = ERR_PTR(ret);
6525 }
6526out:
6527 btrfs_end_transaction(trans, root);
6528 return em;
6529}
6530
46bfbb5c
CM		 6531/*
6532 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6533 * block must be cow'd
6534 */
6535static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
eb384b55
JB		 6536 struct inode *inode, u64 offset, u64 *len,
6537 u64 *orig_start, u64 *orig_block_len,
6538 u64 *ram_bytes)
46bfbb5c
CM		 6539{
6540 struct btrfs_path *path;
6541 int ret;
6542 struct extent_buffer *leaf;
6543 struct btrfs_root *root = BTRFS_I(inode)->root;
6544 struct btrfs_file_extent_item *fi;
6545 struct btrfs_key key;
6546 u64 disk_bytenr;
6547 u64 backref_offset;
6548 u64 extent_end;
6549 u64 num_bytes;
6550 int slot;
6551 int found_type;
6552
6553 path = btrfs_alloc_path();
6554 if (!path)
6555 return -ENOMEM;
6556
33345d01 6557 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
46bfbb5c
CM		 6558 offset, 0);
6559 if (ret < 0)
6560 goto out;
6561
6562 slot = path->slots[0];
6563 if (ret == 1) {
6564 if (slot == 0) {
6565 /* can't find the item, must cow */
6566 ret = 0;
6567 goto out;
6568 }
6569 slot--;
6570 }
6571 ret = 0;
6572 leaf = path->nodes[0];
6573 btrfs_item_key_to_cpu(leaf, &key, slot);
33345d01 6574 if (key.objectid != btrfs_ino(inode) ||
46bfbb5c
CM		 6575 key.type != BTRFS_EXTENT_DATA_KEY) {
6576 /* not our file or wrong item type, must cow */
6577 goto out;
6578 }
6579
6580 if (key.offset > offset) {
6581 /* Wrong offset, must cow */
6582 goto out;
6583 }
6584
6585 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6586 found_type = btrfs_file_extent_type(leaf, fi);
6587 if (found_type != BTRFS_FILE_EXTENT_REG &&
6588 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6589 /* not a regular extent, must cow */
6590 goto out;
6591 }
6592 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6593 backref_offset = btrfs_file_extent_offset(leaf, fi);
6594
eb384b55
JB		 6595 *orig_start = key.offset - backref_offset;
6596 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6597 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6598
46bfbb5c 6599 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
eb384b55 6600 if (extent_end < offset + *len) {
46bfbb5c
CM		 6601 /* extent doesn't include our full range, must cow */
6602 goto out;
6603 }
6604
6605 if (btrfs_extent_readonly(root, disk_bytenr))
6606 goto out;
6607
6608 /*
6609 * look for other files referencing this extent, if we
6610 * find any we must cow
6611 */
33345d01 6612 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
46bfbb5c
CM		 6613 key.offset - backref_offset, disk_bytenr))
6614 goto out;
6615
6616 /*
6617 * adjust disk_bytenr and num_bytes to cover just the bytes
6618 * in this extent we are about to write. If there
6619 * are any csums in that range we have to cow in order
6620 * to keep the csums correct
6621 */
6622 disk_bytenr += backref_offset;
6623 disk_bytenr += offset - key.offset;
eb384b55 6624 num_bytes = min(offset + *len, extent_end) - offset;
46bfbb5c
CM		 6625 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6626 goto out;
6627 /*
6628 * all of the above have passed, it is safe to overwrite this extent
6629 * without cow
6630 */
eb384b55 6631 *len = num_bytes;
46bfbb5c
CM		 6632 ret = 1;
6633out:
6634 btrfs_free_path(path);
6635 return ret;
6636}
6637
eb838e73
JB		 6638static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6639 struct extent_state **cached_state, int writing)
6640{
6641 struct btrfs_ordered_extent *ordered;
6642 int ret = 0;
6643
6644 while (1) {
6645 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6646 0, cached_state);
6647 /*
6648 * We're concerned with the entire range that we're going to be
6649 * doing DIO to, so we need to make sure theres no ordered
6650 * extents in this range.
6651 */
6652 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6653 lockend - lockstart + 1);
6654
6655 /*
6656 * We need to make sure there are no buffered pages in this
6657 * range either, we could have raced between the invalidate in
6658 * generic_file_direct_write and locking the extent. The
6659 * invalidate needs to happen so that reads after a write do not
6660 * get stale data.
6661 */
6662 if (!ordered && (!writing ||
6663 !test_range_bit(&BTRFS_I(inode)->io_tree,
6664 lockstart, lockend, EXTENT_UPTODATE, 0,
6665 *cached_state)))
6666 break;
6667
6668 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6669 cached_state, GFP_NOFS);
6670
6671 if (ordered) {
6672 btrfs_start_ordered_extent(inode, ordered, 1);
6673 btrfs_put_ordered_extent(ordered);
6674 } else {
6675 /* Screw you mmap */
6676 ret = filemap_write_and_wait_range(inode->i_mapping,
6677 lockstart,
6678 lockend);
6679 if (ret)
6680 break;
6681
6682 /*
6683 * If we found a page that couldn't be invalidated just
6684 * fall back to buffered.
6685 */
6686 ret = invalidate_inode_pages2_range(inode->i_mapping,
6687 lockstart >> PAGE_CACHE_SHIFT,
6688 lockend >> PAGE_CACHE_SHIFT);
6689 if (ret)
6690 break;
6691 }
6692
6693 cond_resched();
6694 }
6695
6696 return ret;
6697}
6698
69ffb543
JB		 6699static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6700 u64 len, u64 orig_start,
6701 u64 block_start, u64 block_len,
cc95bef6
JB		 6702 u64 orig_block_len, u64 ram_bytes,
6703 int type)
69ffb543
JB		 6704{
6705 struct extent_map_tree *em_tree;
6706 struct extent_map *em;
6707 struct btrfs_root *root = BTRFS_I(inode)->root;
6708 int ret;
6709
6710 em_tree = &BTRFS_I(inode)->extent_tree;
6711 em = alloc_extent_map();
6712 if (!em)
6713 return ERR_PTR(-ENOMEM);
6714
6715 em->start = start;
6716 em->orig_start = orig_start;
2ab28f32
JB		 6717 em->mod_start = start;
6718 em->mod_len = len;
69ffb543
JB		 6719 em->len = len;
6720 em->block_len = block_len;
6721 em->block_start = block_start;
6722 em->bdev = root->fs_info->fs_devices->latest_bdev;
b4939680 6723 em->orig_block_len = orig_block_len;
cc95bef6 6724 em->ram_bytes = ram_bytes;
70c8a91c 6725 em->generation = -1;
69ffb543
JB		 6726 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6727 if (type == BTRFS_ORDERED_PREALLOC)
b11e234d 6728 set_bit(EXTENT_FLAG_FILLING, &em->flags);
69ffb543
JB		 6729
6730 do {
6731 btrfs_drop_extent_cache(inode, em->start,
6732 em->start + em->len - 1, 0);
6733 write_lock(&em_tree->lock);
09a2a8f9 6734 ret = add_extent_mapping(em_tree, em, 1);
69ffb543
JB		 6735 write_unlock(&em_tree->lock);
6736 } while (ret == -EEXIST);
6737
6738 if (ret) {
6739 free_extent_map(em);
6740 return ERR_PTR(ret);
6741 }
6742
6743 return em;
6744}
6745
6746
4b46fce2
JB		 6747static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6748 struct buffer_head *bh_result, int create)
6749{
6750 struct extent_map *em;
6751 struct btrfs_root *root = BTRFS_I(inode)->root;
eb838e73 6752 struct extent_state *cached_state = NULL;
4b46fce2 6753 u64 start = iblock << inode->i_blkbits;
eb838e73 6754 u64 lockstart, lockend;
4b46fce2 6755 u64 len = bh_result->b_size;
46bfbb5c 6756 struct btrfs_trans_handle *trans;
eb838e73 6757 int unlock_bits = EXTENT_LOCKED;
0934856d 6758 int ret = 0;
eb838e73 6759
172a5049 6760 if (create)
eb838e73 6761 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
172a5049 6762 else
c329861d 6763 len = min_t(u64, len, root->sectorsize);
eb838e73 6764
c329861d
JB		 6765 lockstart = start;
6766 lockend = start + len - 1;
6767
eb838e73
JB		 6768 /*
6769 * If this errors out it's because we couldn't invalidate pagecache for
6770 * this range and we need to fallback to buffered.
6771 */
6772 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6773 return -ENOTBLK;
6774
4b46fce2 6775 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
eb838e73
JB		 6776 if (IS_ERR(em)) {
6777 ret = PTR_ERR(em);
6778 goto unlock_err;
6779 }
4b46fce2
JB		 6780
6781 /*
6782 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6783 * io. INLINE is special, and we could probably kludge it in here, but
6784 * it's still buffered so for safety lets just fall back to the generic
6785 * buffered path.
6786 *
6787 * For COMPRESSED we _have_ to read the entire extent in so we can
6788 * decompress it, so there will be buffering required no matter what we
6789 * do, so go ahead and fallback to buffered.
6790 *
6791 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6792 * to buffered IO. Don't blame me, this is the price we pay for using
6793 * the generic code.
6794 */
6795 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6796 em->block_start == EXTENT_MAP_INLINE) {
6797 free_extent_map(em);
eb838e73
JB		 6798 ret = -ENOTBLK;
6799 goto unlock_err;
4b46fce2
JB		 6800 }
6801
6802 /* Just a good old fashioned hole, return */
6803 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6804 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6805 free_extent_map(em);
eb838e73 6806 goto unlock_err;
4b46fce2
JB		 6807 }
6808
6809 /*
6810 * We don't allocate a new extent in the following cases
6811 *
6812 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6813 * existing extent.
6814 * 2) The extent is marked as PREALLOC. We're good to go here and can
6815 * just use the extent.
6816 *
6817 */
46bfbb5c 6818 if (!create) {
eb838e73
JB		 6819 len = min(len, em->len - (start - em->start));
6820 lockstart = start + len;
6821 goto unlock;
46bfbb5c 6822 }
4b46fce2
JB		 6823
6824 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6825 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6826 em->block_start != EXTENT_MAP_HOLE)) {
4b46fce2
JB		 6827 int type;
6828 int ret;
eb384b55 6829 u64 block_start, orig_start, orig_block_len, ram_bytes;
4b46fce2
JB		 6830
6831 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6832 type = BTRFS_ORDERED_PREALLOC;
6833 else
6834 type = BTRFS_ORDERED_NOCOW;
46bfbb5c 6835 len = min(len, em->len - (start - em->start));
4b46fce2 6836 block_start = em->block_start + (start - em->start);
46bfbb5c
CM		 6837
6838 /*
6839 * we're not going to log anything, but we do need
6840 * to make sure the current transaction stays open
6841 * while we look for nocow cross refs
6842 */
7a7eaa40 6843 trans = btrfs_join_transaction(root);
3612b495 6844 if (IS_ERR(trans))
46bfbb5c
CM		 6845 goto must_cow;
6846
eb384b55
JB		 6847 if (can_nocow_odirect(trans, inode, start, &len, &orig_start,
6848 &orig_block_len, &ram_bytes) == 1) {
69ffb543
JB		 6849 if (type == BTRFS_ORDERED_PREALLOC) {
6850 free_extent_map(em);
6851 em = create_pinned_em(inode, start, len,
6852 orig_start,
b4939680 6853 block_start, len,
cc95bef6
JB		 6854 orig_block_len,
6855 ram_bytes, type);
69ffb543
JB		 6856 if (IS_ERR(em)) {
6857 btrfs_end_transaction(trans, root);
6858 goto unlock_err;
6859 }
6860 }
6861
46bfbb5c
CM		 6862 ret = btrfs_add_ordered_extent_dio(inode, start,
6863 block_start, len, len, type);
6864 btrfs_end_transaction(trans, root);
6865 if (ret) {
6866 free_extent_map(em);
eb838e73 6867 goto unlock_err;
46bfbb5c
CM		 6868 }
6869 goto unlock;
4b46fce2 6870 }
46bfbb5c 6871 btrfs_end_transaction(trans, root);
4b46fce2 6872 }
46bfbb5c
CM		 6873must_cow:
6874 /*
6875 * this will cow the extent, reset the len in case we changed
6876 * it above
6877 */
6878 len = bh_result->b_size;
70c8a91c
JB		 6879 free_extent_map(em);
6880 em = btrfs_new_extent_direct(inode, start, len);
eb838e73
JB		 6881 if (IS_ERR(em)) {
6882 ret = PTR_ERR(em);
6883 goto unlock_err;
6884 }
46bfbb5c
CM		 6885 len = min(len, em->len - (start - em->start));
6886unlock:
4b46fce2
JB		 6887 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6888 inode->i_blkbits;
46bfbb5c 6889 bh_result->b_size = len;
4b46fce2
JB		 6890 bh_result->b_bdev = em->bdev;
6891 set_buffer_mapped(bh_result);
c3473e83
JB		 6892 if (create) {
6893 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6894 set_buffer_new(bh_result);
6895
6896 /*
6897 * Need to update the i_size under the extent lock so buffered
6898 * readers will get the updated i_size when we unlock.
6899 */
6900 if (start + len > i_size_read(inode))
6901 i_size_write(inode, start + len);
0934856d 6902
172a5049
MX		 6903 spin_lock(&BTRFS_I(inode)->lock);
6904 BTRFS_I(inode)->outstanding_extents++;
6905 spin_unlock(&BTRFS_I(inode)->lock);
6906
0934856d
MX		 6907 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6908 lockstart + len - 1, EXTENT_DELALLOC, NULL,
6909 &cached_state, GFP_NOFS);
6910 BUG_ON(ret);
c3473e83 6911 }
4b46fce2 6912
eb838e73
JB		 6913 /*
6914 * In the case of write we need to clear and unlock the entire range,
6915 * in the case of read we need to unlock only the end area that we
6916 * aren't using if there is any left over space.
6917 */
24c03fa5 6918 if (lockstart < lockend) {
0934856d
MX		 6919 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6920 lockend, unlock_bits, 1, 0,
6921 &cached_state, GFP_NOFS);
24c03fa5 6922 } else {
eb838e73 6923 free_extent_state(cached_state);
24c03fa5 6924 }
eb838e73 6925
4b46fce2
JB		 6926 free_extent_map(em);
6927
6928 return 0;
eb838e73
JB		 6929
6930unlock_err:
eb838e73
JB		 6931 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6932 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6933 return ret;
4b46fce2
JB		 6934}
6935
6936struct btrfs_dio_private {
6937 struct inode *inode;
6938 u64 logical_offset;
6939 u64 disk_bytenr;
6940 u64 bytes;
4b46fce2 6941 void *private;
e65e1535
MX		 6942
6943 /* number of bios pending for this dio */
6944 atomic_t pending_bios;
6945
6946 /* IO errors */
6947 int errors;
6948
9be3395b 6949 /* orig_bio is our btrfs_io_bio */
e65e1535 6950 struct bio *orig_bio;
9be3395b
CM		 6951
6952 /* dio_bio came from fs/direct-io.c */
6953 struct bio *dio_bio;
4b46fce2
JB		 6954};
6955
6956static void btrfs_endio_direct_read(struct bio *bio, int err)
6957{
e65e1535 6958 struct btrfs_dio_private *dip = bio->bi_private;
4b46fce2
JB		 6959 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6960 struct bio_vec *bvec = bio->bi_io_vec;
4b46fce2 6961 struct inode *inode = dip->inode;
c2cf52eb 6962 struct btrfs_root *root = BTRFS_I(inode)->root;
9be3395b 6963 struct bio *dio_bio;
4b46fce2 6964 u64 start;
4b46fce2
JB		 6965
6966 start = dip->logical_offset;
6967 do {
6968 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6969 struct page *page = bvec->bv_page;
6970 char *kaddr;
6971 u32 csum = ~(u32)0;
c329861d 6972 u64 private = ~(u32)0;
4b46fce2
JB		 6973 unsigned long flags;
6974
c329861d
JB		 6975 if (get_state_private(&BTRFS_I(inode)->io_tree,
6976 start, &private))
6977 goto failed;
4b46fce2 6978 local_irq_save(flags);
7ac687d9 6979 kaddr = kmap_atomic(page);
b0496686 6980 csum = btrfs_csum_data(kaddr + bvec->bv_offset,
4b46fce2
JB		 6981 csum, bvec->bv_len);
6982 btrfs_csum_final(csum, (char *)&csum);
7ac687d9 6983 kunmap_atomic(kaddr);
4b46fce2
JB		 6984 local_irq_restore(flags);
6985
6986 flush_dcache_page(bvec->bv_page);
c329861d
JB		 6987 if (csum != private) {
6988failed:
c2cf52eb
SK		 6989 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u private %u",
6990 (unsigned long long)btrfs_ino(inode),
6991 (unsigned long long)start,
6992 csum, (unsigned)private);
4b46fce2
JB		 6993 err = -EIO;
6994 }
6995 }
6996
6997 start += bvec->bv_len;
4b46fce2
JB		 6998 bvec++;
6999 } while (bvec <= bvec_end);
7000
7001 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
d0082371 7002 dip->logical_offset + dip->bytes - 1);
9be3395b 7003 dio_bio = dip->dio_bio;
4b46fce2 7004
4b46fce2 7005 kfree(dip);
c0da7aa1
JB		 7006
7007 /* If we had a csum failure make sure to clear the uptodate flag */
7008 if (err)
9be3395b
CM		 7009 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7010 dio_end_io(dio_bio, err);
7011 bio_put(bio);
4b46fce2
JB		 7012}
7013
7014static void btrfs_endio_direct_write(struct bio *bio, int err)
7015{
7016 struct btrfs_dio_private *dip = bio->bi_private;
7017 struct inode *inode = dip->inode;
7018 struct btrfs_root *root = BTRFS_I(inode)->root;
4b46fce2 7019 struct btrfs_ordered_extent *ordered = NULL;
163cf09c
CM		 7020 u64 ordered_offset = dip->logical_offset;
7021 u64 ordered_bytes = dip->bytes;
9be3395b 7022 struct bio *dio_bio;
4b46fce2
JB		 7023 int ret;
7024
7025 if (err)
7026 goto out_done;
163cf09c
CM		 7027again:
7028 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7029 &ordered_offset,
5fd02043 7030 ordered_bytes, !err);
4b46fce2 7031 if (!ret)
163cf09c 7032 goto out_test;
4b46fce2 7033
5fd02043
JB		 7034 ordered->work.func = finish_ordered_fn;
7035 ordered->work.flags = 0;
7036 btrfs_queue_worker(&root->fs_info->endio_write_workers,
7037 &ordered->work);
163cf09c
CM		 7038out_test:
7039 /*
7040 * our bio might span multiple ordered extents. If we haven't
7041 * completed the accounting for the whole dio, go back and try again
7042 */
7043 if (ordered_offset < dip->logical_offset + dip->bytes) {
7044 ordered_bytes = dip->logical_offset + dip->bytes -
7045 ordered_offset;
5fd02043 7046 ordered = NULL;
163cf09c
CM		 7047 goto again;
7048 }
4b46fce2 7049out_done:
9be3395b 7050 dio_bio = dip->dio_bio;
4b46fce2 7051
4b46fce2 7052 kfree(dip);
c0da7aa1
JB		 7053
7054 /* If we had an error make sure to clear the uptodate flag */
7055 if (err)
9be3395b
CM		 7056 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7057 dio_end_io(dio_bio, err);
7058 bio_put(bio);
4b46fce2
JB		 7059}
7060
eaf25d93
CM		 7061static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7062 struct bio *bio, int mirror_num,
7063 unsigned long bio_flags, u64 offset)
7064{
7065 int ret;
7066 struct btrfs_root *root = BTRFS_I(inode)->root;
7067 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
79787eaa 7068 BUG_ON(ret); /* -ENOMEM */
eaf25d93
CM		 7069 return 0;
7070}
7071
e65e1535
MX		 7072static void btrfs_end_dio_bio(struct bio *bio, int err)
7073{
7074 struct btrfs_dio_private *dip = bio->bi_private;
7075
7076 if (err) {
33345d01 7077 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
3dd1462e 7078 "sector %#Lx len %u err no %d\n",
33345d01 7079 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
3dd1462e 7080 (unsigned long long)bio->bi_sector, bio->bi_size, err);
e65e1535
MX		 7081 dip->errors = 1;
7082
7083 /*
7084 * before atomic variable goto zero, we must make sure
7085 * dip->errors is perceived to be set.
7086 */
7087 smp_mb__before_atomic_dec();
7088 }
7089
7090 /* if there are more bios still pending for this dio, just exit */
7091 if (!atomic_dec_and_test(&dip->pending_bios))
7092 goto out;
7093
9be3395b 7094 if (dip->errors) {
e65e1535 7095 bio_io_error(dip->orig_bio);
9be3395b
CM		 7096 } else {
7097 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
e65e1535
MX		 7098 bio_endio(dip->orig_bio, 0);
7099 }
7100out:
7101 bio_put(bio);
7102}
7103
7104static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7105 u64 first_sector, gfp_t gfp_flags)
7106{
7107 int nr_vecs = bio_get_nr_vecs(bdev);
7108 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7109}
7110
7111static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7112 int rw, u64 file_offset, int skip_sum,
c329861d 7113 int async_submit)
e65e1535
MX		 7114{
7115 int write = rw & REQ_WRITE;
7116 struct btrfs_root *root = BTRFS_I(inode)->root;
7117 int ret;
7118
b812ce28
JB		 7119 if (async_submit)
7120 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7121
e65e1535 7122 bio_get(bio);
5fd02043
JB		 7123
7124 if (!write) {
7125 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7126 if (ret)
7127 goto err;
7128 }
e65e1535 7129
1ae39938
JB		 7130 if (skip_sum)
7131 goto map;
7132
7133 if (write && async_submit) {
e65e1535
MX		 7134 ret = btrfs_wq_submit_bio(root->fs_info,
7135 inode, rw, bio, 0, 0,
7136 file_offset,
7137 __btrfs_submit_bio_start_direct_io,
7138 __btrfs_submit_bio_done);
7139 goto err;
1ae39938
JB		 7140 } else if (write) {
7141 /*
7142 * If we aren't doing async submit, calculate the csum of the
7143 * bio now.
7144 */
7145 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7146 if (ret)
7147 goto err;
c2db1073 7148 } else if (!skip_sum) {
c329861d 7149 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
c2db1073
TI		 7150 if (ret)
7151 goto err;
7152 }
e65e1535 7153
1ae39938
JB		 7154map:
7155 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
e65e1535
MX		 7156err:
7157 bio_put(bio);
7158 return ret;
7159}
7160
7161static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7162 int skip_sum)
7163{
7164 struct inode *inode = dip->inode;
7165 struct btrfs_root *root = BTRFS_I(inode)->root;
e65e1535
MX		 7166 struct bio *bio;
7167 struct bio *orig_bio = dip->orig_bio;
7168 struct bio_vec *bvec = orig_bio->bi_io_vec;
7169 u64 start_sector = orig_bio->bi_sector;
7170 u64 file_offset = dip->logical_offset;
7171 u64 submit_len = 0;
7172 u64 map_length;
7173 int nr_pages = 0;
e65e1535 7174 int ret = 0;
1ae39938 7175 int async_submit = 0;
e65e1535 7176
e65e1535 7177 map_length = orig_bio->bi_size;
53b381b3 7178 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
e65e1535
MX		 7179 &map_length, NULL, 0);
7180 if (ret) {
64728bbb 7181 bio_put(orig_bio);
e65e1535
MX		 7182 return -EIO;
7183 }
02f57c7a
JB		 7184 if (map_length >= orig_bio->bi_size) {
7185 bio = orig_bio;
7186 goto submit;
7187 }
7188
53b381b3
DW		 7189 /* async crcs make it difficult to collect full stripe writes. */
7190 if (btrfs_get_alloc_profile(root, 1) &
7191 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7192 async_submit = 0;
7193 else
7194 async_submit = 1;
7195
02f57c7a
JB		 7196 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7197 if (!bio)
7198 return -ENOMEM;
7199 bio->bi_private = dip;
7200 bio->bi_end_io = btrfs_end_dio_bio;
7201 atomic_inc(&dip->pending_bios);
7202
e65e1535
MX		 7203 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7204 if (unlikely(map_length < submit_len + bvec->bv_len ||
7205 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7206 bvec->bv_offset) < bvec->bv_len)) {
7207 /*
7208 * inc the count before we submit the bio so
7209 * we know the end IO handler won't happen before
7210 * we inc the count. Otherwise, the dip might get freed
7211 * before we're done setting it up
7212 */
7213 atomic_inc(&dip->pending_bios);
7214 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7215 file_offset, skip_sum,
c329861d 7216 async_submit);
e65e1535
MX		 7217 if (ret) {
7218 bio_put(bio);
7219 atomic_dec(&dip->pending_bios);
7220 goto out_err;
7221 }
7222
e65e1535
MX		 7223 start_sector += submit_len >> 9;
7224 file_offset += submit_len;
7225
7226 submit_len = 0;
7227 nr_pages = 0;
7228
7229 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7230 start_sector, GFP_NOFS);
7231 if (!bio)
7232 goto out_err;
7233 bio->bi_private = dip;
7234 bio->bi_end_io = btrfs_end_dio_bio;
7235
7236 map_length = orig_bio->bi_size;
53b381b3 7237 ret = btrfs_map_block(root->fs_info, rw,
3ec706c8 7238 start_sector << 9,
e65e1535
MX		 7239 &map_length, NULL, 0);
7240 if (ret) {
7241 bio_put(bio);
7242 goto out_err;
7243 }
7244 } else {
7245 submit_len += bvec->bv_len;
7246 nr_pages ++;
7247 bvec++;
7248 }
7249 }
7250
02f57c7a 7251submit:
e65e1535 7252 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
c329861d 7253 async_submit);
e65e1535
MX		 7254 if (!ret)
7255 return 0;
7256
7257 bio_put(bio);
7258out_err:
7259 dip->errors = 1;
7260 /*
7261 * before atomic variable goto zero, we must
7262 * make sure dip->errors is perceived to be set.
7263 */
7264 smp_mb__before_atomic_dec();
7265 if (atomic_dec_and_test(&dip->pending_bios))
7266 bio_io_error(dip->orig_bio);
7267
7268 /* bio_end_io() will handle error, so we needn't return it */
7269 return 0;
7270}
7271
9be3395b
CM		 7272static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7273 struct inode *inode, loff_t file_offset)
4b46fce2
JB		 7274{
7275 struct btrfs_root *root = BTRFS_I(inode)->root;
7276 struct btrfs_dio_private *dip;
9be3395b
CM		 7277 struct bio_vec *bvec = dio_bio->bi_io_vec;
7278 struct bio *io_bio;
4b46fce2 7279 int skip_sum;
7b6d91da 7280 int write = rw & REQ_WRITE;
4b46fce2
JB		 7281 int ret = 0;
7282
7283 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7284
9be3395b
CM		 7285 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7286
7287 if (!io_bio) {
7288 ret = -ENOMEM;
7289 goto free_ordered;
7290 }
7291
4b46fce2
JB		 7292 dip = kmalloc(sizeof(*dip), GFP_NOFS);
7293 if (!dip) {
7294 ret = -ENOMEM;
9be3395b 7295 goto free_io_bio;
4b46fce2 7296 }
4b46fce2 7297
9be3395b
CM		 7298 dip->private = dio_bio->bi_private;
7299 io_bio->bi_private = dio_bio->bi_private;
4b46fce2
JB		 7300 dip->inode = inode;
7301 dip->logical_offset = file_offset;
7302
4b46fce2
JB		 7303 dip->bytes = 0;
7304 do {
7305 dip->bytes += bvec->bv_len;
7306 bvec++;
9be3395b 7307 } while (bvec <= (dio_bio->bi_io_vec + dio_bio->bi_vcnt - 1));
4b46fce2 7308
9be3395b
CM		 7309 dip->disk_bytenr = (u64)dio_bio->bi_sector << 9;
7310 io_bio->bi_private = dip;
e65e1535 7311 dip->errors = 0;
9be3395b
CM		 7312 dip->orig_bio = io_bio;
7313 dip->dio_bio = dio_bio;
e65e1535 7314 atomic_set(&dip->pending_bios, 0);
4b46fce2
JB		 7315
7316 if (write)
9be3395b 7317 io_bio->bi_end_io = btrfs_endio_direct_write;
4b46fce2 7318 else
9be3395b 7319 io_bio->bi_end_io = btrfs_endio_direct_read;
4b46fce2 7320
e65e1535
MX		 7321 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7322 if (!ret)
eaf25d93 7323 return;
9be3395b
CM		 7324
7325free_io_bio:
7326 bio_put(io_bio);
7327
4b46fce2
JB		 7328free_ordered:
7329 /*
7330 * If this is a write, we need to clean up the reserved space and kill
7331 * the ordered extent.
7332 */
7333 if (write) {
7334 struct btrfs_ordered_extent *ordered;
955256f2 7335 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
4b46fce2
JB		 7336 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7337 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7338 btrfs_free_reserved_extent(root, ordered->start,
7339 ordered->disk_len);
7340 btrfs_put_ordered_extent(ordered);
7341 btrfs_put_ordered_extent(ordered);
7342 }
9be3395b 7343 bio_endio(dio_bio, ret);
4b46fce2
JB		 7344}
7345
5a5f79b5
CM		 7346static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7347 const struct iovec *iov, loff_t offset,
7348 unsigned long nr_segs)
7349{
7350 int seg;
a1b75f7d 7351 int i;
5a5f79b5
CM		 7352 size_t size;
7353 unsigned long addr;
7354 unsigned blocksize_mask = root->sectorsize - 1;
7355 ssize_t retval = -EINVAL;
7356 loff_t end = offset;
7357
7358 if (offset & blocksize_mask)
7359 goto out;
7360
7361 /* Check the memory alignment. Blocks cannot straddle pages */
7362 for (seg = 0; seg < nr_segs; seg++) {
7363 addr = (unsigned long)iov[seg].iov_base;
7364 size = iov[seg].iov_len;
7365 end += size;
a1b75f7d 7366 if ((addr & blocksize_mask) || (size & blocksize_mask))
5a5f79b5 7367 goto out;
a1b75f7d
JB		 7368
7369 /* If this is a write we don't need to check anymore */
7370 if (rw & WRITE)
7371 continue;
7372
7373 /*
7374 * Check to make sure we don't have duplicate iov_base's in this
7375 * iovec, if so return EINVAL, otherwise we'll get csum errors
7376 * when reading back.
7377 */
7378 for (i = seg + 1; i < nr_segs; i++) {
7379 if (iov[seg].iov_base == iov[i].iov_base)
7380 goto out;
7381 }
5a5f79b5
CM		 7382 }
7383 retval = 0;
7384out:
7385 return retval;
7386}
eb838e73 7387
16432985
CM		 7388static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7389 const struct iovec *iov, loff_t offset,
7390 unsigned long nr_segs)
7391{
4b46fce2
JB		 7392 struct file *file = iocb->ki_filp;
7393 struct inode *inode = file->f_mapping->host;
0934856d 7394 size_t count = 0;
2e60a51e 7395 int flags = 0;
38851cc1
MX		 7396 bool wakeup = true;
7397 bool relock = false;
0934856d 7398 ssize_t ret;
4b46fce2 7399
5a5f79b5 7400 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
eb838e73 7401 offset, nr_segs))
5a5f79b5 7402 return 0;
3f7c579c 7403
38851cc1
MX		 7404 atomic_inc(&inode->i_dio_count);
7405 smp_mb__after_atomic_inc();
7406
0934856d
MX		 7407 if (rw & WRITE) {
7408 count = iov_length(iov, nr_segs);
38851cc1
MX		 7409 /*
7410 * If the write DIO is beyond the EOF, we need update
7411 * the isize, but it is protected by i_mutex. So we can
7412 * not unlock the i_mutex at this case.
7413 */
7414 if (offset + count <= inode->i_size) {
7415 mutex_unlock(&inode->i_mutex);
7416 relock = true;
7417 }
0934856d
MX		 7418 ret = btrfs_delalloc_reserve_space(inode, count);
7419 if (ret)
38851cc1
MX		 7420 goto out;
7421 } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7422 &BTRFS_I(inode)->runtime_flags))) {
7423 inode_dio_done(inode);
7424 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7425 wakeup = false;
0934856d
MX		 7426 }
7427
7428 ret = __blockdev_direct_IO(rw, iocb, inode,
7429 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7430 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
2e60a51e 7431 btrfs_submit_direct, flags);
0934856d
MX		 7432 if (rw & WRITE) {
7433 if (ret < 0 && ret != -EIOCBQUEUED)
7434 btrfs_delalloc_release_space(inode, count);
172a5049 7435 else if (ret >= 0 && (size_t)ret < count)
0934856d
MX		 7436 btrfs_delalloc_release_space(inode,
7437 count - (size_t)ret);
172a5049
MX		 7438 else
7439 btrfs_delalloc_release_metadata(inode, 0);
0934856d 7440 }
38851cc1 7441out:
2e60a51e
MX		 7442 if (wakeup)
7443 inode_dio_done(inode);
38851cc1
MX		 7444 if (relock)
7445 mutex_lock(&inode->i_mutex);
0934856d
MX		 7446
7447 return ret;
16432985
CM		 7448}
7449
05dadc09
TI		 7450#define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
7451
1506fcc8
YS		 7452static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7453 __u64 start, __u64 len)
7454{
05dadc09
TI		 7455 int ret;
7456
7457 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7458 if (ret)
7459 return ret;
7460
ec29ed5b 7461 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
1506fcc8
YS		 7462}
7463
a52d9a80 7464int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 7465{
d1310b2e
CM		 7466 struct extent_io_tree *tree;
7467 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 7468 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
9ebefb18 7469}
1832a6d5 7470
a52d9a80 7471static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 7472{
d1310b2e 7473 struct extent_io_tree *tree;
b888db2b
CM		 7474
7475
7476 if (current->flags & PF_MEMALLOC) {
7477 redirty_page_for_writepage(wbc, page);
7478 unlock_page(page);
7479 return 0;
7480 }
d1310b2e 7481 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 7482 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM		 7483}
7484
48a3b636
ES		 7485static int btrfs_writepages(struct address_space *mapping,
7486 struct writeback_control *wbc)
b293f02e 7487{
d1310b2e 7488 struct extent_io_tree *tree;
771ed689 7489
d1310b2e 7490 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 7491 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7492}
7493
3ab2fb5a
CM		 7494static int
7495btrfs_readpages(struct file *file, struct address_space *mapping,
7496 struct list_head *pages, unsigned nr_pages)
7497{
d1310b2e
CM		 7498 struct extent_io_tree *tree;
7499 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 7500 return extent_readpages(tree, mapping, pages, nr_pages,
7501 btrfs_get_extent);
7502}
e6dcd2dc 7503static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 7504{
d1310b2e
CM		 7505 struct extent_io_tree *tree;
7506 struct extent_map_tree *map;
a52d9a80 7507 int ret;
8c2383c3 7508
d1310b2e
CM		 7509 tree = &BTRFS_I(page->mapping->host)->io_tree;
7510 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 7511 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 7512 if (ret == 1) {
7513 ClearPagePrivate(page);
7514 set_page_private(page, 0);
7515 page_cache_release(page);
39279cc3 7516 }
a52d9a80 7517 return ret;
39279cc3
CM		 7518}
7519
e6dcd2dc
CM		 7520static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7521{
98509cfc
CM		 7522 if (PageWriteback(page) || PageDirty(page))
7523 return 0;
b335b003 7524 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
e6dcd2dc
CM		 7525}
7526
a52d9a80 7527static void btrfs_invalidatepage(struct page *page, unsigned long offset)
39279cc3 7528{
5fd02043 7529 struct inode *inode = page->mapping->host;
d1310b2e 7530 struct extent_io_tree *tree;
e6dcd2dc 7531 struct btrfs_ordered_extent *ordered;
2ac55d41 7532 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 7533 u64 page_start = page_offset(page);
7534 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
39279cc3 7535
8b62b72b
CM		 7536 /*
7537 * we have the page locked, so new writeback can't start,
7538 * and the dirty bit won't be cleared while we are here.
7539 *
7540 * Wait for IO on this page so that we can safely clear
7541 * the PagePrivate2 bit and do ordered accounting
7542 */
e6dcd2dc 7543 wait_on_page_writeback(page);
8b62b72b 7544
5fd02043 7545 tree = &BTRFS_I(inode)->io_tree;
e6dcd2dc
CM		 7546 if (offset) {
7547 btrfs_releasepage(page, GFP_NOFS);
7548 return;
7549 }
d0082371 7550 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
4eee4fa4 7551 ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
e6dcd2dc 7552 if (ordered) {
eb84ae03
CM		 7553 /*
7554 * IO on this page will never be started, so we need
7555 * to account for any ordered extents now
7556 */
e6dcd2dc
CM		 7557 clear_extent_bit(tree, page_start, page_end,
7558 EXTENT_DIRTY | EXTENT_DELALLOC |
9e8a4a8b
LB		 7559 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7560 EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
8b62b72b
CM		 7561 /*
7562 * whoever cleared the private bit is responsible
7563 * for the finish_ordered_io
7564 */
5fd02043
JB		 7565 if (TestClearPagePrivate2(page) &&
7566 btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
7567 PAGE_CACHE_SIZE, 1)) {
7568 btrfs_finish_ordered_io(ordered);
8b62b72b 7569 }
e6dcd2dc 7570 btrfs_put_ordered_extent(ordered);
2ac55d41 7571 cached_state = NULL;
d0082371 7572 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 7573 }
7574 clear_extent_bit(tree, page_start, page_end,
32c00aff 7575 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
9e8a4a8b
LB		 7576 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
7577 &cached_state, GFP_NOFS);
e6dcd2dc
CM		 7578 __btrfs_releasepage(page, GFP_NOFS);
7579
4a096752 7580 ClearPageChecked(page);
9ad6b7bc 7581 if (PagePrivate(page)) {
9ad6b7bc
CM		 7582 ClearPagePrivate(page);
7583 set_page_private(page, 0);
7584 page_cache_release(page);
7585 }
39279cc3
CM		 7586}
7587
9ebefb18
CM		 7588/*
7589 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7590 * called from a page fault handler when a page is first dirtied. Hence we must
7591 * be careful to check for EOF conditions here. We set the page up correctly
7592 * for a written page which means we get ENOSPC checking when writing into
7593 * holes and correct delalloc and unwritten extent mapping on filesystems that
7594 * support these features.
7595 *
7596 * We are not allowed to take the i_mutex here so we have to play games to
7597 * protect against truncate races as the page could now be beyond EOF. Because
7598 * vmtruncate() writes the inode size before removing pages, once we have the
7599 * page lock we can determine safely if the page is beyond EOF. If it is not
7600 * beyond EOF, then the page is guaranteed safe against truncation until we
7601 * unlock the page.
7602 */
c2ec175c 7603int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
9ebefb18 7604{
c2ec175c 7605 struct page *page = vmf->page;
496ad9aa 7606 struct inode *inode = file_inode(vma->vm_file);
1832a6d5 7607 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 7608 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7609 struct btrfs_ordered_extent *ordered;
2ac55d41 7610 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 7611 char *kaddr;
7612 unsigned long zero_start;
9ebefb18 7613 loff_t size;
1832a6d5 7614 int ret;
9998eb70 7615 int reserved = 0;
a52d9a80 7616 u64 page_start;
e6dcd2dc 7617 u64 page_end;
9ebefb18 7618
b2b5ef5c 7619 sb_start_pagefault(inode->i_sb);
0ca1f7ce 7620 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
9998eb70 7621 if (!ret) {
e41f941a 7622 ret = file_update_time(vma->vm_file);
9998eb70
CM		 7623 reserved = 1;
7624 }
56a76f82
NP		 7625 if (ret) {
7626 if (ret == -ENOMEM)
7627 ret = VM_FAULT_OOM;
7628 else /* -ENOSPC, -EIO, etc */
7629 ret = VM_FAULT_SIGBUS;
9998eb70
CM		 7630 if (reserved)
7631 goto out;
7632 goto out_noreserve;
56a76f82 7633 }
1832a6d5 7634
56a76f82 7635 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
e6dcd2dc 7636again:
9ebefb18 7637 lock_page(page);
9ebefb18 7638 size = i_size_read(inode);
e6dcd2dc
CM		 7639 page_start = page_offset(page);
7640 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 7641
9ebefb18 7642 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 7643 (page_start >= size)) {
9ebefb18
CM		 7644 /* page got truncated out from underneath us */
7645 goto out_unlock;
7646 }
e6dcd2dc
CM		 7647 wait_on_page_writeback(page);
7648
d0082371 7649 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 7650 set_page_extent_mapped(page);
7651
eb84ae03
CM		 7652 /*
7653 * we can't set the delalloc bits if there are pending ordered
7654 * extents. Drop our locks and wait for them to finish
7655 */
e6dcd2dc
CM		 7656 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7657 if (ordered) {
2ac55d41
JB		 7658 unlock_extent_cached(io_tree, page_start, page_end,
7659 &cached_state, GFP_NOFS);
e6dcd2dc 7660 unlock_page(page);
eb84ae03 7661 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 7662 btrfs_put_ordered_extent(ordered);
7663 goto again;
7664 }
7665
fbf19087
JB		 7666 /*
7667 * XXX - page_mkwrite gets called every time the page is dirtied, even
7668 * if it was already dirty, so for space accounting reasons we need to
7669 * clear any delalloc bits for the range we are fixing to save. There
7670 * is probably a better way to do this, but for now keep consistent with
7671 * prepare_pages in the normal write path.
7672 */
2ac55d41 7673 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 7674 EXTENT_DIRTY | EXTENT_DELALLOC |
7675 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 7676 0, 0, &cached_state, GFP_NOFS);
fbf19087 7677
2ac55d41
JB		 7678 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7679 &cached_state);
9ed74f2d 7680 if (ret) {
2ac55d41
JB		 7681 unlock_extent_cached(io_tree, page_start, page_end,
7682 &cached_state, GFP_NOFS);
9ed74f2d
JB		 7683 ret = VM_FAULT_SIGBUS;
7684 goto out_unlock;
7685 }
e6dcd2dc 7686 ret = 0;
9ebefb18
CM		 7687
7688 /* page is wholly or partially inside EOF */
a52d9a80 7689 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 7690 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 7691 else
e6dcd2dc 7692 zero_start = PAGE_CACHE_SIZE;
9ebefb18 7693
e6dcd2dc
CM		 7694 if (zero_start != PAGE_CACHE_SIZE) {
7695 kaddr = kmap(page);
7696 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7697 flush_dcache_page(page);
7698 kunmap(page);
7699 }
247e743c 7700 ClearPageChecked(page);
e6dcd2dc 7701 set_page_dirty(page);
50a9b214 7702 SetPageUptodate(page);
5a3f23d5 7703
257c62e1
CM		 7704 BTRFS_I(inode)->last_trans = root->fs_info->generation;
7705 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
46d8bc34 7706 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
257c62e1 7707
2ac55d41 7708 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9ebefb18
CM		 7709
7710out_unlock:
b2b5ef5c
JK		 7711 if (!ret) {
7712 sb_end_pagefault(inode->i_sb);
50a9b214 7713 return VM_FAULT_LOCKED;
b2b5ef5c 7714 }
9ebefb18 7715 unlock_page(page);
1832a6d5 7716out:
ec39e180 7717 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
9998eb70 7718out_noreserve:
b2b5ef5c 7719 sb_end_pagefault(inode->i_sb);
9ebefb18
CM		 7720 return ret;
7721}
7722
a41ad394 7723static int btrfs_truncate(struct inode *inode)
39279cc3
CM		 7724{
7725 struct btrfs_root *root = BTRFS_I(inode)->root;
fcb80c2a 7726 struct btrfs_block_rsv *rsv;
39279cc3 7727 int ret;
3893e33b 7728 int err = 0;
39279cc3 7729 struct btrfs_trans_handle *trans;
dbe674a9 7730 u64 mask = root->sectorsize - 1;
07127184 7731 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3 7732
2aaa6655 7733 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
5d5e103a 7734 if (ret)
a41ad394 7735 return ret;
8082510e 7736
4a096752 7737 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
8082510e 7738 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
39279cc3 7739
fcb80c2a
JB		 7740 /*
7741 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
7742 * 3 things going on here
7743 *
7744 * 1) We need to reserve space for our orphan item and the space to
7745 * delete our orphan item. Lord knows we don't want to have a dangling
7746 * orphan item because we didn't reserve space to remove it.
7747 *
7748 * 2) We need to reserve space to update our inode.
7749 *
7750 * 3) We need to have something to cache all the space that is going to
7751 * be free'd up by the truncate operation, but also have some slack
7752 * space reserved in case it uses space during the truncate (thank you
7753 * very much snapshotting).
7754 *
7755 * And we need these to all be seperate. The fact is we can use alot of
7756 * space doing the truncate, and we have no earthly idea how much space
7757 * we will use, so we need the truncate reservation to be seperate so it
7758 * doesn't end up using space reserved for updating the inode or
7759 * removing the orphan item. We also need to be able to stop the
7760 * transaction and start a new one, which means we need to be able to
7761 * update the inode several times, and we have no idea of knowing how
7762 * many times that will be, so we can't just reserve 1 item for the
7763 * entirety of the opration, so that has to be done seperately as well.
7764 * Then there is the orphan item, which does indeed need to be held on
7765 * to for the whole operation, and we need nobody to touch this reserved
7766 * space except the orphan code.
7767 *
7768 * So that leaves us with
7769 *
7770 * 1) root->orphan_block_rsv - for the orphan deletion.
7771 * 2) rsv - for the truncate reservation, which we will steal from the
7772 * transaction reservation.
7773 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7774 * updating the inode.
7775 */
66d8f3dd 7776 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
fcb80c2a
JB		 7777 if (!rsv)
7778 return -ENOMEM;
4a338542 7779 rsv->size = min_size;
ca7e70f5 7780 rsv->failfast = 1;
f0cd846e 7781
907cbceb 7782 /*
07127184 7783 * 1 for the truncate slack space
907cbceb
JB		 7784 * 1 for updating the inode.
7785 */
f3fe820c 7786 trans = btrfs_start_transaction(root, 2);
fcb80c2a
JB		 7787 if (IS_ERR(trans)) {
7788 err = PTR_ERR(trans);
7789 goto out;
7790 }
f0cd846e 7791
907cbceb
JB		 7792 /* Migrate the slack space for the truncate to our reserve */
7793 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7794 min_size);
fcb80c2a 7795 BUG_ON(ret);
f0cd846e 7796
5a3f23d5
CM		 7797 /*
7798 * setattr is responsible for setting the ordered_data_close flag,
7799 * but that is only tested during the last file release. That
7800 * could happen well after the next commit, leaving a great big
7801 * window where new writes may get lost if someone chooses to write
7802 * to this file after truncating to zero
7803 *
7804 * The inode doesn't have any dirty data here, and so if we commit
7805 * this is a noop. If someone immediately starts writing to the inode
7806 * it is very likely we'll catch some of their writes in this
7807 * transaction, and the commit will find this file on the ordered
7808 * data list with good things to send down.
7809 *
7810 * This is a best effort solution, there is still a window where
7811 * using truncate to replace the contents of the file will
7812 * end up with a zero length file after a crash.
7813 */
72ac3c0d
JB		 7814 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7815 &BTRFS_I(inode)->runtime_flags))
5a3f23d5
CM		 7816 btrfs_add_ordered_operation(trans, root, inode);
7817
5dc562c5
JB		 7818 /*
7819 * So if we truncate and then write and fsync we normally would just
7820 * write the extents that changed, which is a problem if we need to
7821 * first truncate that entire inode. So set this flag so we write out
7822 * all of the extents in the inode to the sync log so we're completely
7823 * safe.
7824 */
7825 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
ca7e70f5 7826 trans->block_rsv = rsv;
907cbceb 7827
8082510e
YZ		 7828 while (1) {
7829 ret = btrfs_truncate_inode_items(trans, root, inode,
7830 inode->i_size,
7831 BTRFS_EXTENT_DATA_KEY);
ca7e70f5 7832 if (ret != -ENOSPC) {
3893e33b 7833 err = ret;
8082510e 7834 break;
3893e33b 7835 }
39279cc3 7836
fcb80c2a 7837 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e 7838 ret = btrfs_update_inode(trans, root, inode);
3893e33b
JB		 7839 if (ret) {
7840 err = ret;
7841 break;
7842 }
ca7e70f5 7843
8082510e 7844 btrfs_end_transaction(trans, root);
b53d3f5d 7845 btrfs_btree_balance_dirty(root);
ca7e70f5
JB		 7846
7847 trans = btrfs_start_transaction(root, 2);
7848 if (IS_ERR(trans)) {
7849 ret = err = PTR_ERR(trans);
7850 trans = NULL;
7851 break;
7852 }
7853
7854 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7855 rsv, min_size);
7856 BUG_ON(ret); /* shouldn't happen */
7857 trans->block_rsv = rsv;
8082510e
YZ		 7858 }
7859
7860 if (ret == 0 && inode->i_nlink > 0) {
fcb80c2a 7861 trans->block_rsv = root->orphan_block_rsv;
8082510e 7862 ret = btrfs_orphan_del(trans, inode);
3893e33b
JB		 7863 if (ret)
7864 err = ret;
8082510e
YZ		 7865 }
7866
917c16b2
CM		 7867 if (trans) {
7868 trans->block_rsv = &root->fs_info->trans_block_rsv;
7869 ret = btrfs_update_inode(trans, root, inode);
7870 if (ret && !err)
7871 err = ret;
7b128766 7872
7ad85bb7 7873 ret = btrfs_end_transaction(trans, root);
b53d3f5d 7874 btrfs_btree_balance_dirty(root);
917c16b2 7875 }
fcb80c2a
JB		 7876
7877out:
7878 btrfs_free_block_rsv(root, rsv);
7879
3893e33b
JB		 7880 if (ret && !err)
7881 err = ret;
a41ad394 7882
3893e33b 7883 return err;
39279cc3
CM		 7884}
7885
d352ac68
CM		 7886/*
7887 * create a new subvolume directory/inode (helper for the ioctl).
7888 */
d2fb3437 7889int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
d82a6f1d 7890 struct btrfs_root *new_root, u64 new_dirid)
39279cc3 7891{
39279cc3 7892 struct inode *inode;
76dda93c 7893 int err;
00e4e6b3 7894 u64 index = 0;
39279cc3 7895
12fc9d09
FA		 7896 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7897 new_dirid, new_dirid,
7898 S_IFDIR | (~current_umask() & S_IRWXUGO),
7899 &index);
54aa1f4d 7900 if (IS_ERR(inode))
f46b5a66 7901 return PTR_ERR(inode);
39279cc3
CM		 7902 inode->i_op = &btrfs_dir_inode_operations;
7903 inode->i_fop = &btrfs_dir_file_operations;
7904
bfe86848 7905 set_nlink(inode, 1);
dbe674a9 7906 btrfs_i_size_write(inode, 0);
3b96362c 7907
76dda93c 7908 err = btrfs_update_inode(trans, new_root, inode);
cb8e7090 7909
76dda93c 7910 iput(inode);
ce598979 7911 return err;
39279cc3
CM		 7912}
7913
39279cc3
CM		 7914struct inode *btrfs_alloc_inode(struct super_block *sb)
7915{
7916 struct btrfs_inode *ei;
2ead6ae7 7917 struct inode *inode;
39279cc3
CM		 7918
7919 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7920 if (!ei)
7921 return NULL;
2ead6ae7
YZ		 7922
7923 ei->root = NULL;
2ead6ae7 7924 ei->generation = 0;
15ee9bc7 7925 ei->last_trans = 0;
257c62e1 7926 ei->last_sub_trans = 0;
e02119d5 7927 ei->logged_trans = 0;
2ead6ae7 7928 ei->delalloc_bytes = 0;
2ead6ae7
YZ		 7929 ei->disk_i_size = 0;
7930 ei->flags = 0;
7709cde3 7931 ei->csum_bytes = 0;
2ead6ae7
YZ		 7932 ei->index_cnt = (u64)-1;
7933 ei->last_unlink_trans = 0;
46d8bc34 7934 ei->last_log_commit = 0;
2ead6ae7 7935
9e0baf60
JB		 7936 spin_lock_init(&ei->lock);
7937 ei->outstanding_extents = 0;
7938 ei->reserved_extents = 0;
2ead6ae7 7939
72ac3c0d 7940 ei->runtime_flags = 0;
261507a0 7941 ei->force_compress = BTRFS_COMPRESS_NONE;
2ead6ae7 7942
16cdcec7
MX		 7943 ei->delayed_node = NULL;
7944
2ead6ae7 7945 inode = &ei->vfs_inode;
a8067e02 7946 extent_map_tree_init(&ei->extent_tree);
f993c883
DS		 7947 extent_io_tree_init(&ei->io_tree, &inode->i_data);
7948 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
0b32f4bb
JB		 7949 ei->io_tree.track_uptodate = 1;
7950 ei->io_failure_tree.track_uptodate = 1;
b812ce28 7951 atomic_set(&ei->sync_writers, 0);
2ead6ae7 7952 mutex_init(&ei->log_mutex);
f248679e 7953 mutex_init(&ei->delalloc_mutex);
e6dcd2dc 7954 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
2ead6ae7 7955 INIT_LIST_HEAD(&ei->delalloc_inodes);
5a3f23d5 7956 INIT_LIST_HEAD(&ei->ordered_operations);
2ead6ae7
YZ		 7957 RB_CLEAR_NODE(&ei->rb_node);
7958
7959 return inode;
39279cc3
CM		 7960}
7961
fa0d7e3d
NP		 7962static void btrfs_i_callback(struct rcu_head *head)
7963{
7964 struct inode *inode = container_of(head, struct inode, i_rcu);
fa0d7e3d
NP		 7965 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7966}
7967
39279cc3
CM		 7968void btrfs_destroy_inode(struct inode *inode)
7969{
e6dcd2dc 7970 struct btrfs_ordered_extent *ordered;
5a3f23d5
CM		 7971 struct btrfs_root *root = BTRFS_I(inode)->root;
7972
b3d9b7a3 7973 WARN_ON(!hlist_empty(&inode->i_dentry));
39279cc3 7974 WARN_ON(inode->i_data.nrpages);
9e0baf60
JB		 7975 WARN_ON(BTRFS_I(inode)->outstanding_extents);
7976 WARN_ON(BTRFS_I(inode)->reserved_extents);
7709cde3
JB		 7977 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7978 WARN_ON(BTRFS_I(inode)->csum_bytes);
39279cc3 7979
a6dbd429
JB		 7980 /*
7981 * This can happen where we create an inode, but somebody else also
7982 * created the same inode and we need to destroy the one we already
7983 * created.
7984 */
7985 if (!root)
7986 goto free;
7987
5a3f23d5
CM		 7988 /*
7989 * Make sure we're properly removed from the ordered operation
7990 * lists.
7991 */
7992 smp_mb();
7993 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
199c2a9c 7994 spin_lock(&root->fs_info->ordered_root_lock);
5a3f23d5 7995 list_del_init(&BTRFS_I(inode)->ordered_operations);
199c2a9c 7996 spin_unlock(&root->fs_info->ordered_root_lock);
5a3f23d5
CM		 7997 }
7998
8a35d95f
JB		 7999 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8000 &BTRFS_I(inode)->runtime_flags)) {
c2cf52eb
SK		 8001 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
8002 (unsigned long long)btrfs_ino(inode));
8a35d95f 8003 atomic_dec(&root->orphan_inodes);
7b128766 8004 }
7b128766 8005
d397712b 8006 while (1) {
e6dcd2dc
CM		 8007 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8008 if (!ordered)
8009 break;
8010 else {
c2cf52eb
SK		 8011 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
8012 (unsigned long long)ordered->file_offset,
8013 (unsigned long long)ordered->len);
e6dcd2dc
CM		 8014 btrfs_remove_ordered_extent(inode, ordered);
8015 btrfs_put_ordered_extent(ordered);
8016 btrfs_put_ordered_extent(ordered);
8017 }
8018 }
5d4f98a2 8019 inode_tree_del(inode);
5b21f2ed 8020 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
a6dbd429 8021free:
fa0d7e3d 8022 call_rcu(&inode->i_rcu, btrfs_i_callback);
39279cc3
CM		 8023}
8024
45321ac5 8025int btrfs_drop_inode(struct inode *inode)
76dda93c
YZ		 8026{
8027 struct btrfs_root *root = BTRFS_I(inode)->root;
45321ac5 8028
6379ef9f
NA		 8029 if (root == NULL)
8030 return 1;
8031
fa6ac876 8032 /* the snap/subvol tree is on deleting */
0af3d00b 8033 if (btrfs_root_refs(&root->root_item) == 0 &&
fa6ac876 8034 root != root->fs_info->tree_root)
45321ac5 8035 return 1;
76dda93c 8036 else
45321ac5 8037 return generic_drop_inode(inode);
76dda93c
YZ		 8038}
8039
0ee0fda0 8040static void init_once(void *foo)
39279cc3
CM		 8041{
8042 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8043
8044 inode_init_once(&ei->vfs_inode);
8045}
8046
8047void btrfs_destroy_cachep(void)
8048{
8c0a8537
KS		 8049 /*
8050 * Make sure all delayed rcu free inodes are flushed before we
8051 * destroy cache.
8052 */
8053 rcu_barrier();
39279cc3
CM		 8054 if (btrfs_inode_cachep)
8055 kmem_cache_destroy(btrfs_inode_cachep);
8056 if (btrfs_trans_handle_cachep)
8057 kmem_cache_destroy(btrfs_trans_handle_cachep);
8058 if (btrfs_transaction_cachep)
8059 kmem_cache_destroy(btrfs_transaction_cachep);
39279cc3
CM		 8060 if (btrfs_path_cachep)
8061 kmem_cache_destroy(btrfs_path_cachep);
dc89e982
JB		 8062 if (btrfs_free_space_cachep)
8063 kmem_cache_destroy(btrfs_free_space_cachep);
8ccf6f19
MX		 8064 if (btrfs_delalloc_work_cachep)
8065 kmem_cache_destroy(btrfs_delalloc_work_cachep);
39279cc3
CM		 8066}
8067
8068int btrfs_init_cachep(void)
8069{
837e1972 8070 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9601e3f6
CH		 8071 sizeof(struct btrfs_inode), 0,
8072 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
39279cc3
CM		 8073 if (!btrfs_inode_cachep)
8074 goto fail;
9601e3f6 8075
837e1972 8076 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9601e3f6
CH		 8077 sizeof(struct btrfs_trans_handle), 0,
8078 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 8079 if (!btrfs_trans_handle_cachep)
8080 goto fail;
9601e3f6 8081
837e1972 8082 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9601e3f6
CH		 8083 sizeof(struct btrfs_transaction), 0,
8084 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 8085 if (!btrfs_transaction_cachep)
8086 goto fail;
9601e3f6 8087
837e1972 8088 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9601e3f6
CH		 8089 sizeof(struct btrfs_path), 0,
8090 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 8091 if (!btrfs_path_cachep)
8092 goto fail;
9601e3f6 8093
837e1972 8094 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
dc89e982
JB		 8095 sizeof(struct btrfs_free_space), 0,
8096 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8097 if (!btrfs_free_space_cachep)
8098 goto fail;
8099
8ccf6f19
MX		 8100 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8101 sizeof(struct btrfs_delalloc_work), 0,
8102 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8103 NULL);
8104 if (!btrfs_delalloc_work_cachep)
8105 goto fail;
8106
39279cc3
CM		 8107 return 0;
8108fail:
8109 btrfs_destroy_cachep();
8110 return -ENOMEM;
8111}
8112
8113static int btrfs_getattr(struct vfsmount *mnt,
8114 struct dentry *dentry, struct kstat *stat)
8115{
df0af1a5 8116 u64 delalloc_bytes;
39279cc3 8117 struct inode *inode = dentry->d_inode;
fadc0d8b
DS		 8118 u32 blocksize = inode->i_sb->s_blocksize;
8119
39279cc3 8120 generic_fillattr(inode, stat);
0ee5dc67 8121 stat->dev = BTRFS_I(inode)->root->anon_dev;
d6667462 8122 stat->blksize = PAGE_CACHE_SIZE;
df0af1a5
MX		 8123
8124 spin_lock(&BTRFS_I(inode)->lock);
8125 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8126 spin_unlock(&BTRFS_I(inode)->lock);
fadc0d8b 8127 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
df0af1a5 8128 ALIGN(delalloc_bytes, blocksize)) >> 9;
39279cc3
CM		 8129 return 0;
8130}
8131
d397712b
CM		 8132static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8133 struct inode *new_dir, struct dentry *new_dentry)
39279cc3
CM		 8134{
8135 struct btrfs_trans_handle *trans;
8136 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4df27c4d 8137 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
39279cc3
CM		 8138 struct inode *new_inode = new_dentry->d_inode;
8139 struct inode *old_inode = old_dentry->d_inode;
8140 struct timespec ctime = CURRENT_TIME;
00e4e6b3 8141 u64 index = 0;
4df27c4d 8142 u64 root_objectid;
39279cc3 8143 int ret;
33345d01 8144 u64 old_ino = btrfs_ino(old_inode);
39279cc3 8145
33345d01 8146 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
f679a840
YZ		 8147 return -EPERM;
8148
4df27c4d 8149 /* we only allow rename subvolume link between subvolumes */
33345d01 8150 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
3394e160
CM		 8151 return -EXDEV;
8152
33345d01
LZ		 8153 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8154 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
39279cc3 8155 return -ENOTEMPTY;
5f39d397 8156
4df27c4d
YZ		 8157 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8158 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8159 return -ENOTEMPTY;
9c52057c
CM		 8160
8161
8162 /* check for collisions, even if the name isn't there */
8163 ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
8164 new_dentry->d_name.name,
8165 new_dentry->d_name.len);
8166
8167 if (ret) {
8168 if (ret == -EEXIST) {
8169 /* we shouldn't get
8170 * eexist without a new_inode */
8171 if (!new_inode) {
8172 WARN_ON(1);
8173 return ret;
8174 }
8175 } else {
8176 /* maybe -EOVERFLOW */
8177 return ret;
8178 }
8179 }
8180 ret = 0;
8181
5a3f23d5
CM		 8182 /*
8183 * we're using rename to replace one file with another.
8184 * and the replacement file is large. Start IO on it now so
8185 * we don't add too much work to the end of the transaction
8186 */
4baf8c92 8187 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
5a3f23d5
CM		 8188 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8189 filemap_flush(old_inode->i_mapping);
8190
76dda93c 8191 /* close the racy window with snapshot create/destroy ioctl */
33345d01 8192 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 8193 down_read(&root->fs_info->subvol_sem);
a22285a6
YZ		 8194 /*
8195 * We want to reserve the absolute worst case amount of items. So if
8196 * both inodes are subvols and we need to unlink them then that would
8197 * require 4 item modifications, but if they are both normal inodes it
8198 * would require 5 item modifications, so we'll assume their normal
8199 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8200 * should cover the worst case number of items we'll modify.
8201 */
6e137ed3 8202 trans = btrfs_start_transaction(root, 11);
b44c59a8
JL		 8203 if (IS_ERR(trans)) {
8204 ret = PTR_ERR(trans);
8205 goto out_notrans;
8206 }
76dda93c 8207
4df27c4d
YZ		 8208 if (dest != root)
8209 btrfs_record_root_in_trans(trans, dest);
5f39d397 8210
a5719521
YZ		 8211 ret = btrfs_set_inode_index(new_dir, &index);
8212 if (ret)
8213 goto out_fail;
5a3f23d5 8214
33345d01 8215 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 8216 /* force full log commit if subvolume involved. */
8217 root->fs_info->last_trans_log_full_commit = trans->transid;
8218 } else {
a5719521
YZ		 8219 ret = btrfs_insert_inode_ref(trans, dest,
8220 new_dentry->d_name.name,
8221 new_dentry->d_name.len,
33345d01
LZ		 8222 old_ino,
8223 btrfs_ino(new_dir), index);
a5719521
YZ		 8224 if (ret)
8225 goto out_fail;
4df27c4d
YZ		 8226 /*
8227 * this is an ugly little race, but the rename is required
8228 * to make sure that if we crash, the inode is either at the
8229 * old name or the new one. pinning the log transaction lets
8230 * us make sure we don't allow a log commit to come in after
8231 * we unlink the name but before we add the new name back in.
8232 */
8233 btrfs_pin_log_trans(root);
8234 }
5a3f23d5
CM		 8235 /*
8236 * make sure the inode gets flushed if it is replacing
8237 * something.
8238 */
33345d01 8239 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
5a3f23d5 8240 btrfs_add_ordered_operation(trans, root, old_inode);
5a3f23d5 8241
0c4d2d95
JB		 8242 inode_inc_iversion(old_dir);
8243 inode_inc_iversion(new_dir);
8244 inode_inc_iversion(old_inode);
39279cc3
CM		 8245 old_dir->i_ctime = old_dir->i_mtime = ctime;
8246 new_dir->i_ctime = new_dir->i_mtime = ctime;
8247 old_inode->i_ctime = ctime;
5f39d397 8248
12fcfd22
CM		 8249 if (old_dentry->d_parent != new_dentry->d_parent)
8250 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8251
33345d01 8252 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 8253 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8254 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8255 old_dentry->d_name.name,
8256 old_dentry->d_name.len);
8257 } else {
92986796
AV		 8258 ret = __btrfs_unlink_inode(trans, root, old_dir,
8259 old_dentry->d_inode,
8260 old_dentry->d_name.name,
8261 old_dentry->d_name.len);
8262 if (!ret)
8263 ret = btrfs_update_inode(trans, root, old_inode);
4df27c4d 8264 }
79787eaa
JM		 8265 if (ret) {
8266 btrfs_abort_transaction(trans, root, ret);
8267 goto out_fail;
8268 }
39279cc3
CM		 8269
8270 if (new_inode) {
0c4d2d95 8271 inode_inc_iversion(new_inode);
39279cc3 8272 new_inode->i_ctime = CURRENT_TIME;
33345d01 8273 if (unlikely(btrfs_ino(new_inode) ==
4df27c4d
YZ		 8274 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8275 root_objectid = BTRFS_I(new_inode)->location.objectid;
8276 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8277 root_objectid,
8278 new_dentry->d_name.name,
8279 new_dentry->d_name.len);
8280 BUG_ON(new_inode->i_nlink == 0);
8281 } else {
8282 ret = btrfs_unlink_inode(trans, dest, new_dir,
8283 new_dentry->d_inode,
8284 new_dentry->d_name.name,
8285 new_dentry->d_name.len);
8286 }
79787eaa 8287 if (!ret && new_inode->i_nlink == 0) {
e02119d5 8288 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4df27c4d 8289 BUG_ON(ret);
7b128766 8290 }
79787eaa
JM		 8291 if (ret) {
8292 btrfs_abort_transaction(trans, root, ret);
8293 goto out_fail;
8294 }
39279cc3 8295 }
aec7477b 8296
4df27c4d
YZ		 8297 ret = btrfs_add_link(trans, new_dir, old_inode,
8298 new_dentry->d_name.name,
a5719521 8299 new_dentry->d_name.len, 0, index);
79787eaa
JM		 8300 if (ret) {
8301 btrfs_abort_transaction(trans, root, ret);
8302 goto out_fail;
8303 }
39279cc3 8304
33345d01 8305 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
10d9f309 8306 struct dentry *parent = new_dentry->d_parent;
6a912213 8307 btrfs_log_new_name(trans, old_inode, old_dir, parent);
4df27c4d
YZ		 8308 btrfs_end_log_trans(root);
8309 }
39279cc3 8310out_fail:
7ad85bb7 8311 btrfs_end_transaction(trans, root);
b44c59a8 8312out_notrans:
33345d01 8313 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 8314 up_read(&root->fs_info->subvol_sem);
9ed74f2d 8315
39279cc3
CM		 8316 return ret;
8317}
8318
8ccf6f19
MX		 8319static void btrfs_run_delalloc_work(struct btrfs_work *work)
8320{
8321 struct btrfs_delalloc_work *delalloc_work;
8322
8323 delalloc_work = container_of(work, struct btrfs_delalloc_work,
8324 work);
8325 if (delalloc_work->wait)
8326 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
8327 else
8328 filemap_flush(delalloc_work->inode->i_mapping);
8329
8330 if (delalloc_work->delay_iput)
8331 btrfs_add_delayed_iput(delalloc_work->inode);
8332 else
8333 iput(delalloc_work->inode);
8334 complete(&delalloc_work->completion);
8335}
8336
8337struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8338 int wait, int delay_iput)
8339{
8340 struct btrfs_delalloc_work *work;
8341
8342 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8343 if (!work)
8344 return NULL;
8345
8346 init_completion(&work->completion);
8347 INIT_LIST_HEAD(&work->list);
8348 work->inode = inode;
8349 work->wait = wait;
8350 work->delay_iput = delay_iput;
8351 work->work.func = btrfs_run_delalloc_work;
8352
8353 return work;
8354}
8355
8356void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8357{
8358 wait_for_completion(&work->completion);
8359 kmem_cache_free(btrfs_delalloc_work_cachep, work);
8360}
8361
d352ac68
CM		 8362/*
8363 * some fairly slow code that needs optimization. This walks the list
8364 * of all the inodes with pending delalloc and forces them to disk.
8365 */
eb73c1b7 8366static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
ea8c2819 8367{
ea8c2819 8368 struct btrfs_inode *binode;
5b21f2ed 8369 struct inode *inode;
8ccf6f19
MX		 8370 struct btrfs_delalloc_work *work, *next;
8371 struct list_head works;
1eafa6c7 8372 struct list_head splice;
8ccf6f19 8373 int ret = 0;
ea8c2819 8374
8ccf6f19 8375 INIT_LIST_HEAD(&works);
1eafa6c7 8376 INIT_LIST_HEAD(&splice);
63607cc8 8377
eb73c1b7
MX		 8378 spin_lock(&root->delalloc_lock);
8379 list_splice_init(&root->delalloc_inodes, &splice);
1eafa6c7
MX		 8380 while (!list_empty(&splice)) {
8381 binode = list_entry(splice.next, struct btrfs_inode,
ea8c2819 8382 delalloc_inodes);
1eafa6c7 8383
eb73c1b7
MX		 8384 list_move_tail(&binode->delalloc_inodes,
8385 &root->delalloc_inodes);
5b21f2ed 8386 inode = igrab(&binode->vfs_inode);
df0af1a5 8387 if (!inode) {
eb73c1b7 8388 cond_resched_lock(&root->delalloc_lock);
1eafa6c7 8389 continue;
df0af1a5 8390 }
eb73c1b7 8391 spin_unlock(&root->delalloc_lock);
1eafa6c7
MX		 8392
8393 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8394 if (unlikely(!work)) {
8395 ret = -ENOMEM;
8396 goto out;
5b21f2ed 8397 }
1eafa6c7
MX		 8398 list_add_tail(&work->list, &works);
8399 btrfs_queue_worker(&root->fs_info->flush_workers,
8400 &work->work);
8401
5b21f2ed 8402 cond_resched();
eb73c1b7 8403 spin_lock(&root->delalloc_lock);
ea8c2819 8404 }
eb73c1b7 8405 spin_unlock(&root->delalloc_lock);
8c8bee1d 8406
1eafa6c7
MX		 8407 list_for_each_entry_safe(work, next, &works, list) {
8408 list_del_init(&work->list);
8409 btrfs_wait_and_free_delalloc_work(work);
8410 }
eb73c1b7
MX		 8411 return 0;
8412out:
8413 list_for_each_entry_safe(work, next, &works, list) {
8414 list_del_init(&work->list);
8415 btrfs_wait_and_free_delalloc_work(work);
8416 }
8417
8418 if (!list_empty_careful(&splice)) {
8419 spin_lock(&root->delalloc_lock);
8420 list_splice_tail(&splice, &root->delalloc_inodes);
8421 spin_unlock(&root->delalloc_lock);
8422 }
8423 return ret;
8424}
1eafa6c7 8425
eb73c1b7
MX		 8426int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8427{
8428 int ret;
8429
8430 if (root->fs_info->sb->s_flags & MS_RDONLY)
8431 return -EROFS;
8432
8433 ret = __start_delalloc_inodes(root, delay_iput);
8434 /*
8435 * the filemap_flush will queue IO into the worker threads, but
8c8bee1d
CM		 8436 * we have to make sure the IO is actually started and that
8437 * ordered extents get created before we return
8438 */
8439 atomic_inc(&root->fs_info->async_submit_draining);
d397712b 8440 while (atomic_read(&root->fs_info->nr_async_submits) ||
771ed689 8441 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 8442 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 8443 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8444 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 8445 }
8446 atomic_dec(&root->fs_info->async_submit_draining);
eb73c1b7
MX		 8447 return ret;
8448}
8449
8450int btrfs_start_all_delalloc_inodes(struct btrfs_fs_info *fs_info,
8451 int delay_iput)
8452{
8453 struct btrfs_root *root;
8454 struct list_head splice;
8455 int ret;
8456
8457 if (fs_info->sb->s_flags & MS_RDONLY)
8458 return -EROFS;
8459
8460 INIT_LIST_HEAD(&splice);
8461
8462 spin_lock(&fs_info->delalloc_root_lock);
8463 list_splice_init(&fs_info->delalloc_roots, &splice);
8464 while (!list_empty(&splice)) {
8465 root = list_first_entry(&splice, struct btrfs_root,
8466 delalloc_root);
8467 root = btrfs_grab_fs_root(root);
8468 BUG_ON(!root);
8469 list_move_tail(&root->delalloc_root,
8470 &fs_info->delalloc_roots);
8471 spin_unlock(&fs_info->delalloc_root_lock);
8472
8473 ret = __start_delalloc_inodes(root, delay_iput);
8474 btrfs_put_fs_root(root);
8475 if (ret)
8476 goto out;
8477
8478 spin_lock(&fs_info->delalloc_root_lock);
8ccf6f19 8479 }
eb73c1b7 8480 spin_unlock(&fs_info->delalloc_root_lock);
1eafa6c7 8481
eb73c1b7
MX		 8482 atomic_inc(&fs_info->async_submit_draining);
8483 while (atomic_read(&fs_info->nr_async_submits) ||
8484 atomic_read(&fs_info->async_delalloc_pages)) {
8485 wait_event(fs_info->async_submit_wait,
8486 (atomic_read(&fs_info->nr_async_submits) == 0 &&
8487 atomic_read(&fs_info->async_delalloc_pages) == 0));
8488 }
8489 atomic_dec(&fs_info->async_submit_draining);
8490 return 0;
8491out:
1eafa6c7 8492 if (!list_empty_careful(&splice)) {
eb73c1b7
MX		 8493 spin_lock(&fs_info->delalloc_root_lock);
8494 list_splice_tail(&splice, &fs_info->delalloc_roots);
8495 spin_unlock(&fs_info->delalloc_root_lock);
1eafa6c7 8496 }
8ccf6f19 8497 return ret;
ea8c2819
CM		 8498}
8499
39279cc3
CM		 8500static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8501 const char *symname)
8502{
8503 struct btrfs_trans_handle *trans;
8504 struct btrfs_root *root = BTRFS_I(dir)->root;
8505 struct btrfs_path *path;
8506 struct btrfs_key key;
1832a6d5 8507 struct inode *inode = NULL;
39279cc3
CM		 8508 int err;
8509 int drop_inode = 0;
8510 u64 objectid;
00e4e6b3 8511 u64 index = 0 ;
39279cc3
CM		 8512 int name_len;
8513 int datasize;
5f39d397 8514 unsigned long ptr;
39279cc3 8515 struct btrfs_file_extent_item *ei;
5f39d397 8516 struct extent_buffer *leaf;
39279cc3
CM		 8517
8518 name_len = strlen(symname) + 1;
8519 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8520 return -ENAMETOOLONG;
1832a6d5 8521
9ed74f2d
JB		 8522 /*
8523 * 2 items for inode item and ref
8524 * 2 items for dir items
8525 * 1 item for xattr if selinux is on
8526 */
a22285a6
YZ		 8527 trans = btrfs_start_transaction(root, 5);
8528 if (IS_ERR(trans))
8529 return PTR_ERR(trans);
1832a6d5 8530
581bb050
LZ		 8531 err = btrfs_find_free_ino(root, &objectid);
8532 if (err)
8533 goto out_unlock;
8534
aec7477b 8535 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 8536 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 8537 S_IFLNK|S_IRWXUGO, &index);
7cf96da3
TI		 8538 if (IS_ERR(inode)) {
8539 err = PTR_ERR(inode);
39279cc3 8540 goto out_unlock;
7cf96da3 8541 }
39279cc3 8542
2a7dba39 8543 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 8544 if (err) {
8545 drop_inode = 1;
8546 goto out_unlock;
8547 }
8548
ad19db71
CS		 8549 /*
8550 * If the active LSM wants to access the inode during
8551 * d_instantiate it needs these. Smack checks to see
8552 * if the filesystem supports xattrs by looking at the
8553 * ops vector.
8554 */
8555 inode->i_fop = &btrfs_file_operations;
8556 inode->i_op = &btrfs_file_inode_operations;
8557
a1b075d2 8558 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3
CM		 8559 if (err)
8560 drop_inode = 1;
8561 else {
8562 inode->i_mapping->a_ops = &btrfs_aops;
04160088 8563 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 8564 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3 8565 }
39279cc3
CM		 8566 if (drop_inode)
8567 goto out_unlock;
8568
8569 path = btrfs_alloc_path();
d8926bb3
MF		 8570 if (!path) {
8571 err = -ENOMEM;
8572 drop_inode = 1;
8573 goto out_unlock;
8574 }
33345d01 8575 key.objectid = btrfs_ino(inode);
39279cc3 8576 key.offset = 0;
39279cc3
CM		 8577 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8578 datasize = btrfs_file_extent_calc_inline_size(name_len);
8579 err = btrfs_insert_empty_item(trans, root, path, &key,
8580 datasize);
54aa1f4d
CM		 8581 if (err) {
8582 drop_inode = 1;
b0839166 8583 btrfs_free_path(path);
54aa1f4d
CM		 8584 goto out_unlock;
8585 }
5f39d397
CM		 8586 leaf = path->nodes[0];
8587 ei = btrfs_item_ptr(leaf, path->slots[0],
8588 struct btrfs_file_extent_item);
8589 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8590 btrfs_set_file_extent_type(leaf, ei,
39279cc3 8591 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 8592 btrfs_set_file_extent_encryption(leaf, ei, 0);
8593 btrfs_set_file_extent_compression(leaf, ei, 0);
8594 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8595 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8596
39279cc3 8597 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 8598 write_extent_buffer(leaf, symname, ptr, name_len);
8599 btrfs_mark_buffer_dirty(leaf);
39279cc3 8600 btrfs_free_path(path);
5f39d397 8601
39279cc3
CM		 8602 inode->i_op = &btrfs_symlink_inode_operations;
8603 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 8604 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d899e052 8605 inode_set_bytes(inode, name_len);
dbe674a9 8606 btrfs_i_size_write(inode, name_len - 1);
54aa1f4d
CM		 8607 err = btrfs_update_inode(trans, root, inode);
8608 if (err)
8609 drop_inode = 1;
39279cc3
CM		 8610
8611out_unlock:
08c422c2
AV		 8612 if (!err)
8613 d_instantiate(dentry, inode);
7ad85bb7 8614 btrfs_end_transaction(trans, root);
39279cc3
CM		 8615 if (drop_inode) {
8616 inode_dec_link_count(inode);
8617 iput(inode);
8618 }
b53d3f5d 8619 btrfs_btree_balance_dirty(root);
39279cc3
CM		 8620 return err;
8621}
16432985 8622
0af3d00b
JB		 8623static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8624 u64 start, u64 num_bytes, u64 min_size,
8625 loff_t actual_len, u64 *alloc_hint,
8626 struct btrfs_trans_handle *trans)
d899e052 8627{
5dc562c5
JB		 8628 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8629 struct extent_map *em;
d899e052
YZ		 8630 struct btrfs_root *root = BTRFS_I(inode)->root;
8631 struct btrfs_key ins;
d899e052 8632 u64 cur_offset = start;
55a61d1d 8633 u64 i_size;
154ea289 8634 u64 cur_bytes;
d899e052 8635 int ret = 0;
0af3d00b 8636 bool own_trans = true;
d899e052 8637
0af3d00b
JB		 8638 if (trans)
8639 own_trans = false;
d899e052 8640 while (num_bytes > 0) {
0af3d00b
JB		 8641 if (own_trans) {
8642 trans = btrfs_start_transaction(root, 3);
8643 if (IS_ERR(trans)) {
8644 ret = PTR_ERR(trans);
8645 break;
8646 }
5a303d5d
YZ		 8647 }
8648
154ea289
CM		 8649 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8650 cur_bytes = max(cur_bytes, min_size);
8651 ret = btrfs_reserve_extent(trans, root, cur_bytes,
24542bf7 8652 min_size, 0, *alloc_hint, &ins, 1);
5a303d5d 8653 if (ret) {
0af3d00b
JB		 8654 if (own_trans)
8655 btrfs_end_transaction(trans, root);
a22285a6 8656 break;
d899e052 8657 }
5a303d5d 8658
d899e052
YZ		 8659 ret = insert_reserved_file_extent(trans, inode,
8660 cur_offset, ins.objectid,
8661 ins.offset, ins.offset,
920bbbfb 8662 ins.offset, 0, 0, 0,
d899e052 8663 BTRFS_FILE_EXTENT_PREALLOC);
79787eaa
JM		 8664 if (ret) {
8665 btrfs_abort_transaction(trans, root, ret);
8666 if (own_trans)
8667 btrfs_end_transaction(trans, root);
8668 break;
8669 }
a1ed835e
CM		 8670 btrfs_drop_extent_cache(inode, cur_offset,
8671 cur_offset + ins.offset -1, 0);
5a303d5d 8672
5dc562c5
JB		 8673 em = alloc_extent_map();
8674 if (!em) {
8675 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8676 &BTRFS_I(inode)->runtime_flags);
8677 goto next;
8678 }
8679
8680 em->start = cur_offset;
8681 em->orig_start = cur_offset;
8682 em->len = ins.offset;
8683 em->block_start = ins.objectid;
8684 em->block_len = ins.offset;
b4939680 8685 em->orig_block_len = ins.offset;
cc95bef6 8686 em->ram_bytes = ins.offset;
5dc562c5
JB		 8687 em->bdev = root->fs_info->fs_devices->latest_bdev;
8688 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8689 em->generation = trans->transid;
8690
8691 while (1) {
8692 write_lock(&em_tree->lock);
09a2a8f9 8693 ret = add_extent_mapping(em_tree, em, 1);
5dc562c5
JB		 8694 write_unlock(&em_tree->lock);
8695 if (ret != -EEXIST)
8696 break;
8697 btrfs_drop_extent_cache(inode, cur_offset,
8698 cur_offset + ins.offset - 1,
8699 0);
8700 }
8701 free_extent_map(em);
8702next:
d899e052
YZ		 8703 num_bytes -= ins.offset;
8704 cur_offset += ins.offset;
efa56464 8705 *alloc_hint = ins.objectid + ins.offset;
5a303d5d 8706
0c4d2d95 8707 inode_inc_iversion(inode);
d899e052 8708 inode->i_ctime = CURRENT_TIME;
6cbff00f 8709 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
d899e052 8710 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
efa56464
YZ		 8711 (actual_len > inode->i_size) &&
8712 (cur_offset > inode->i_size)) {
d1ea6a61 8713 if (cur_offset > actual_len)
55a61d1d 8714 i_size = actual_len;
d1ea6a61 8715 else
55a61d1d
JB		 8716 i_size = cur_offset;
8717 i_size_write(inode, i_size);
8718 btrfs_ordered_update_i_size(inode, i_size, NULL);
5a303d5d
YZ		 8719 }
8720
d899e052 8721 ret = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 8722
8723 if (ret) {
8724 btrfs_abort_transaction(trans, root, ret);
8725 if (own_trans)
8726 btrfs_end_transaction(trans, root);
8727 break;
8728 }
d899e052 8729
0af3d00b
JB		 8730 if (own_trans)
8731 btrfs_end_transaction(trans, root);
5a303d5d 8732 }
d899e052
YZ		 8733 return ret;
8734}
8735
0af3d00b
JB		 8736int btrfs_prealloc_file_range(struct inode *inode, int mode,
8737 u64 start, u64 num_bytes, u64 min_size,
8738 loff_t actual_len, u64 *alloc_hint)
8739{
8740 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8741 min_size, actual_len, alloc_hint,
8742 NULL);
8743}
8744
8745int btrfs_prealloc_file_range_trans(struct inode *inode,
8746 struct btrfs_trans_handle *trans, int mode,
8747 u64 start, u64 num_bytes, u64 min_size,
8748 loff_t actual_len, u64 *alloc_hint)
8749{
8750 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8751 min_size, actual_len, alloc_hint, trans);
8752}
8753
e6dcd2dc
CM		 8754static int btrfs_set_page_dirty(struct page *page)
8755{
e6dcd2dc
CM		 8756 return __set_page_dirty_nobuffers(page);
8757}
8758
10556cb2 8759static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd 8760{
b83cc969 8761 struct btrfs_root *root = BTRFS_I(inode)->root;
cb6db4e5 8762 umode_t mode = inode->i_mode;
b83cc969 8763
cb6db4e5
JM		 8764 if (mask & MAY_WRITE &&
8765 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8766 if (btrfs_root_readonly(root))
8767 return -EROFS;
8768 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8769 return -EACCES;
8770 }
2830ba7f 8771 return generic_permission(inode, mask);
fdebe2bd 8772}
39279cc3 8773
6e1d5dcc 8774static const struct inode_operations btrfs_dir_inode_operations = {
3394e160 8775 .getattr = btrfs_getattr,
39279cc3
CM		 8776 .lookup = btrfs_lookup,
8777 .create = btrfs_create,
8778 .unlink = btrfs_unlink,
8779 .link = btrfs_link,
8780 .mkdir = btrfs_mkdir,
8781 .rmdir = btrfs_rmdir,
8782 .rename = btrfs_rename,
8783 .symlink = btrfs_symlink,
8784 .setattr = btrfs_setattr,
618e21d5 8785 .mknod = btrfs_mknod,
95819c05
CH		 8786 .setxattr = btrfs_setxattr,
8787 .getxattr = btrfs_getxattr,
5103e947 8788 .listxattr = btrfs_listxattr,
95819c05 8789 .removexattr = btrfs_removexattr,
fdebe2bd 8790 .permission = btrfs_permission,
4e34e719 8791 .get_acl = btrfs_get_acl,
39279cc3 8792};
6e1d5dcc 8793static const struct inode_operations btrfs_dir_ro_inode_operations = {
39279cc3 8794 .lookup = btrfs_lookup,
fdebe2bd 8795 .permission = btrfs_permission,
4e34e719 8796 .get_acl = btrfs_get_acl,
39279cc3 8797};
76dda93c 8798
828c0950 8799static const struct file_operations btrfs_dir_file_operations = {
39279cc3
CM		 8800 .llseek = generic_file_llseek,
8801 .read = generic_read_dir,
cbdf5a24 8802 .readdir = btrfs_real_readdir,
34287aa3 8803 .unlocked_ioctl = btrfs_ioctl,
39279cc3 8804#ifdef CONFIG_COMPAT
34287aa3 8805 .compat_ioctl = btrfs_ioctl,
39279cc3 8806#endif
6bf13c0c 8807 .release = btrfs_release_file,
e02119d5 8808 .fsync = btrfs_sync_file,
39279cc3
CM		 8809};
8810
d1310b2e 8811static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 8812 .fill_delalloc = run_delalloc_range,
065631f6 8813 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 8814 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 8815 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 8816 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 8817 .writepage_start_hook = btrfs_writepage_start_hook,
b0c68f8b
CM		 8818 .set_bit_hook = btrfs_set_bit_hook,
8819 .clear_bit_hook = btrfs_clear_bit_hook,
9ed74f2d
JB		 8820 .merge_extent_hook = btrfs_merge_extent_hook,
8821 .split_extent_hook = btrfs_split_extent_hook,
07157aac
CM		 8822};
8823
35054394
CM		 8824/*
8825 * btrfs doesn't support the bmap operation because swapfiles
8826 * use bmap to make a mapping of extents in the file. They assume
8827 * these extents won't change over the life of the file and they
8828 * use the bmap result to do IO directly to the drive.
8829 *
8830 * the btrfs bmap call would return logical addresses that aren't
8831 * suitable for IO and they also will change frequently as COW
8832 * operations happen. So, swapfile + btrfs == corruption.
8833 *
8834 * For now we're avoiding this by dropping bmap.
8835 */
7f09410b 8836static const struct address_space_operations btrfs_aops = {
39279cc3
CM		 8837 .readpage = btrfs_readpage,
8838 .writepage = btrfs_writepage,
b293f02e 8839 .writepages = btrfs_writepages,
3ab2fb5a 8840 .readpages = btrfs_readpages,
16432985 8841 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 8842 .invalidatepage = btrfs_invalidatepage,
8843 .releasepage = btrfs_releasepage,
e6dcd2dc 8844 .set_page_dirty = btrfs_set_page_dirty,
465fdd97 8845 .error_remove_page = generic_error_remove_page,
39279cc3
CM		 8846};
8847
7f09410b 8848static const struct address_space_operations btrfs_symlink_aops = {
39279cc3
CM		 8849 .readpage = btrfs_readpage,
8850 .writepage = btrfs_writepage,
2bf5a725
CM		 8851 .invalidatepage = btrfs_invalidatepage,
8852 .releasepage = btrfs_releasepage,
39279cc3
CM		 8853};
8854
6e1d5dcc 8855static const struct inode_operations btrfs_file_inode_operations = {
39279cc3
CM		 8856 .getattr = btrfs_getattr,
8857 .setattr = btrfs_setattr,
95819c05
CH		 8858 .setxattr = btrfs_setxattr,
8859 .getxattr = btrfs_getxattr,
5103e947 8860 .listxattr = btrfs_listxattr,
95819c05 8861 .removexattr = btrfs_removexattr,
fdebe2bd 8862 .permission = btrfs_permission,
1506fcc8 8863 .fiemap = btrfs_fiemap,
4e34e719 8864 .get_acl = btrfs_get_acl,
e41f941a 8865 .update_time = btrfs_update_time,
39279cc3 8866};
6e1d5dcc 8867static const struct inode_operations btrfs_special_inode_operations = {
618e21d5
JB		 8868 .getattr = btrfs_getattr,
8869 .setattr = btrfs_setattr,
fdebe2bd 8870 .permission = btrfs_permission,
95819c05
CH		 8871 .setxattr = btrfs_setxattr,
8872 .getxattr = btrfs_getxattr,
33268eaf 8873 .listxattr = btrfs_listxattr,
95819c05 8874 .removexattr = btrfs_removexattr,
4e34e719 8875 .get_acl = btrfs_get_acl,
e41f941a 8876 .update_time = btrfs_update_time,
618e21d5 8877};
6e1d5dcc 8878static const struct inode_operations btrfs_symlink_inode_operations = {
39279cc3
CM		 8879 .readlink = generic_readlink,
8880 .follow_link = page_follow_link_light,
8881 .put_link = page_put_link,
f209561a 8882 .getattr = btrfs_getattr,
22c44fe6 8883 .setattr = btrfs_setattr,
fdebe2bd 8884 .permission = btrfs_permission,
0279b4cd
JO		 8885 .setxattr = btrfs_setxattr,
8886 .getxattr = btrfs_getxattr,
8887 .listxattr = btrfs_listxattr,
8888 .removexattr = btrfs_removexattr,
4e34e719 8889 .get_acl = btrfs_get_acl,
e41f941a 8890 .update_time = btrfs_update_time,
39279cc3 8891};
76dda93c 8892
82d339d9 8893const struct dentry_operations btrfs_dentry_operations = {
76dda93c 8894 .d_delete = btrfs_dentry_delete,
b4aff1f8 8895 .d_release = btrfs_dentry_release,
76dda93c 8896};
Linux 6.x block layer and io_uring tree(s)