Commit | Line | Data |
---|---|---|
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 | ||
39279cc3 CM |
19 | #include <linux/fs.h> |
20 | #include <linux/pagemap.h> | |
21 | #include <linux/highmem.h> | |
22 | #include <linux/time.h> | |
23 | #include <linux/init.h> | |
24 | #include <linux/string.h> | |
39279cc3 CM |
25 | #include <linux/backing-dev.h> |
26 | #include <linux/mpage.h> | |
2fe17c10 | 27 | #include <linux/falloc.h> |
39279cc3 CM |
28 | #include <linux/swap.h> |
29 | #include <linux/writeback.h> | |
30 | #include <linux/statfs.h> | |
31 | #include <linux/compat.h> | |
5a0e3ad6 | 32 | #include <linux/slab.h> |
39279cc3 CM |
33 | #include "ctree.h" |
34 | #include "disk-io.h" | |
35 | #include "transaction.h" | |
36 | #include "btrfs_inode.h" | |
37 | #include "ioctl.h" | |
38 | #include "print-tree.h" | |
e02119d5 CM |
39 | #include "tree-log.h" |
40 | #include "locking.h" | |
12fa8ec6 | 41 | #include "compat.h" |
39279cc3 | 42 | |
4cb5300b CM |
43 | /* |
44 | * when auto defrag is enabled we | |
45 | * queue up these defrag structs to remember which | |
46 | * inodes need defragging passes | |
47 | */ | |
48 | struct inode_defrag { | |
49 | struct rb_node rb_node; | |
50 | /* objectid */ | |
51 | u64 ino; | |
52 | /* | |
53 | * transid where the defrag was added, we search for | |
54 | * extents newer than this | |
55 | */ | |
56 | u64 transid; | |
57 | ||
58 | /* root objectid */ | |
59 | u64 root; | |
60 | ||
61 | /* last offset we were able to defrag */ | |
62 | u64 last_offset; | |
63 | ||
64 | /* if we've wrapped around back to zero once already */ | |
65 | int cycled; | |
66 | }; | |
67 | ||
68 | /* pop a record for an inode into the defrag tree. The lock | |
69 | * must be held already | |
70 | * | |
71 | * If you're inserting a record for an older transid than an | |
72 | * existing record, the transid already in the tree is lowered | |
73 | * | |
74 | * If an existing record is found the defrag item you | |
75 | * pass in is freed | |
76 | */ | |
a0f98dde | 77 | static void __btrfs_add_inode_defrag(struct inode *inode, |
4cb5300b CM |
78 | struct inode_defrag *defrag) |
79 | { | |
80 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
81 | struct inode_defrag *entry; | |
82 | struct rb_node **p; | |
83 | struct rb_node *parent = NULL; | |
84 | ||
85 | p = &root->fs_info->defrag_inodes.rb_node; | |
86 | while (*p) { | |
87 | parent = *p; | |
88 | entry = rb_entry(parent, struct inode_defrag, rb_node); | |
89 | ||
90 | if (defrag->ino < entry->ino) | |
91 | p = &parent->rb_left; | |
92 | else if (defrag->ino > entry->ino) | |
93 | p = &parent->rb_right; | |
94 | else { | |
95 | /* if we're reinserting an entry for | |
96 | * an old defrag run, make sure to | |
97 | * lower the transid of our existing record | |
98 | */ | |
99 | if (defrag->transid < entry->transid) | |
100 | entry->transid = defrag->transid; | |
101 | if (defrag->last_offset > entry->last_offset) | |
102 | entry->last_offset = defrag->last_offset; | |
103 | goto exists; | |
104 | } | |
105 | } | |
106 | BTRFS_I(inode)->in_defrag = 1; | |
107 | rb_link_node(&defrag->rb_node, parent, p); | |
108 | rb_insert_color(&defrag->rb_node, &root->fs_info->defrag_inodes); | |
a0f98dde | 109 | return; |
4cb5300b CM |
110 | |
111 | exists: | |
112 | kfree(defrag); | |
a0f98dde | 113 | return; |
4cb5300b CM |
114 | |
115 | } | |
116 | ||
117 | /* | |
118 | * insert a defrag record for this inode if auto defrag is | |
119 | * enabled | |
120 | */ | |
121 | int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans, | |
122 | struct inode *inode) | |
123 | { | |
124 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
125 | struct inode_defrag *defrag; | |
4cb5300b CM |
126 | u64 transid; |
127 | ||
128 | if (!btrfs_test_opt(root, AUTO_DEFRAG)) | |
129 | return 0; | |
130 | ||
7841cb28 | 131 | if (btrfs_fs_closing(root->fs_info)) |
4cb5300b CM |
132 | return 0; |
133 | ||
134 | if (BTRFS_I(inode)->in_defrag) | |
135 | return 0; | |
136 | ||
137 | if (trans) | |
138 | transid = trans->transid; | |
139 | else | |
140 | transid = BTRFS_I(inode)->root->last_trans; | |
141 | ||
142 | defrag = kzalloc(sizeof(*defrag), GFP_NOFS); | |
143 | if (!defrag) | |
144 | return -ENOMEM; | |
145 | ||
a4689d2b | 146 | defrag->ino = btrfs_ino(inode); |
4cb5300b CM |
147 | defrag->transid = transid; |
148 | defrag->root = root->root_key.objectid; | |
149 | ||
150 | spin_lock(&root->fs_info->defrag_inodes_lock); | |
151 | if (!BTRFS_I(inode)->in_defrag) | |
a0f98dde | 152 | __btrfs_add_inode_defrag(inode, defrag); |
f4ac904c DC |
153 | else |
154 | kfree(defrag); | |
4cb5300b | 155 | spin_unlock(&root->fs_info->defrag_inodes_lock); |
a0f98dde | 156 | return 0; |
4cb5300b CM |
157 | } |
158 | ||
159 | /* | |
160 | * must be called with the defrag_inodes lock held | |
161 | */ | |
162 | struct inode_defrag *btrfs_find_defrag_inode(struct btrfs_fs_info *info, u64 ino, | |
163 | struct rb_node **next) | |
164 | { | |
165 | struct inode_defrag *entry = NULL; | |
166 | struct rb_node *p; | |
167 | struct rb_node *parent = NULL; | |
168 | ||
169 | p = info->defrag_inodes.rb_node; | |
170 | while (p) { | |
171 | parent = p; | |
172 | entry = rb_entry(parent, struct inode_defrag, rb_node); | |
173 | ||
174 | if (ino < entry->ino) | |
175 | p = parent->rb_left; | |
176 | else if (ino > entry->ino) | |
177 | p = parent->rb_right; | |
178 | else | |
179 | return entry; | |
180 | } | |
181 | ||
182 | if (next) { | |
183 | while (parent && ino > entry->ino) { | |
184 | parent = rb_next(parent); | |
185 | entry = rb_entry(parent, struct inode_defrag, rb_node); | |
186 | } | |
187 | *next = parent; | |
188 | } | |
189 | return NULL; | |
190 | } | |
191 | ||
192 | /* | |
193 | * run through the list of inodes in the FS that need | |
194 | * defragging | |
195 | */ | |
196 | int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info) | |
197 | { | |
198 | struct inode_defrag *defrag; | |
199 | struct btrfs_root *inode_root; | |
200 | struct inode *inode; | |
201 | struct rb_node *n; | |
202 | struct btrfs_key key; | |
203 | struct btrfs_ioctl_defrag_range_args range; | |
204 | u64 first_ino = 0; | |
205 | int num_defrag; | |
206 | int defrag_batch = 1024; | |
207 | ||
208 | memset(&range, 0, sizeof(range)); | |
209 | range.len = (u64)-1; | |
210 | ||
211 | atomic_inc(&fs_info->defrag_running); | |
212 | spin_lock(&fs_info->defrag_inodes_lock); | |
213 | while(1) { | |
214 | n = NULL; | |
215 | ||
216 | /* find an inode to defrag */ | |
217 | defrag = btrfs_find_defrag_inode(fs_info, first_ino, &n); | |
218 | if (!defrag) { | |
219 | if (n) | |
220 | defrag = rb_entry(n, struct inode_defrag, rb_node); | |
221 | else if (first_ino) { | |
222 | first_ino = 0; | |
223 | continue; | |
224 | } else { | |
225 | break; | |
226 | } | |
227 | } | |
228 | ||
229 | /* remove it from the rbtree */ | |
230 | first_ino = defrag->ino + 1; | |
231 | rb_erase(&defrag->rb_node, &fs_info->defrag_inodes); | |
232 | ||
7841cb28 | 233 | if (btrfs_fs_closing(fs_info)) |
4cb5300b CM |
234 | goto next_free; |
235 | ||
236 | spin_unlock(&fs_info->defrag_inodes_lock); | |
237 | ||
238 | /* get the inode */ | |
239 | key.objectid = defrag->root; | |
240 | btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY); | |
241 | key.offset = (u64)-1; | |
242 | inode_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
243 | if (IS_ERR(inode_root)) | |
244 | goto next; | |
245 | ||
246 | key.objectid = defrag->ino; | |
247 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); | |
248 | key.offset = 0; | |
249 | ||
250 | inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL); | |
251 | if (IS_ERR(inode)) | |
252 | goto next; | |
253 | ||
254 | /* do a chunk of defrag */ | |
255 | BTRFS_I(inode)->in_defrag = 0; | |
256 | range.start = defrag->last_offset; | |
257 | num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid, | |
258 | defrag_batch); | |
259 | /* | |
260 | * if we filled the whole defrag batch, there | |
261 | * must be more work to do. Queue this defrag | |
262 | * again | |
263 | */ | |
264 | if (num_defrag == defrag_batch) { | |
265 | defrag->last_offset = range.start; | |
266 | __btrfs_add_inode_defrag(inode, defrag); | |
267 | /* | |
268 | * we don't want to kfree defrag, we added it back to | |
269 | * the rbtree | |
270 | */ | |
271 | defrag = NULL; | |
272 | } else if (defrag->last_offset && !defrag->cycled) { | |
273 | /* | |
274 | * we didn't fill our defrag batch, but | |
275 | * we didn't start at zero. Make sure we loop | |
276 | * around to the start of the file. | |
277 | */ | |
278 | defrag->last_offset = 0; | |
279 | defrag->cycled = 1; | |
280 | __btrfs_add_inode_defrag(inode, defrag); | |
281 | defrag = NULL; | |
282 | } | |
283 | ||
284 | iput(inode); | |
285 | next: | |
286 | spin_lock(&fs_info->defrag_inodes_lock); | |
287 | next_free: | |
288 | kfree(defrag); | |
289 | } | |
290 | spin_unlock(&fs_info->defrag_inodes_lock); | |
291 | ||
292 | atomic_dec(&fs_info->defrag_running); | |
293 | ||
294 | /* | |
295 | * during unmount, we use the transaction_wait queue to | |
296 | * wait for the defragger to stop | |
297 | */ | |
298 | wake_up(&fs_info->transaction_wait); | |
299 | return 0; | |
300 | } | |
39279cc3 | 301 | |
d352ac68 CM |
302 | /* simple helper to fault in pages and copy. This should go away |
303 | * and be replaced with calls into generic code. | |
304 | */ | |
d397712b | 305 | static noinline int btrfs_copy_from_user(loff_t pos, int num_pages, |
d0215f3e | 306 | size_t write_bytes, |
a1b32a59 | 307 | struct page **prepared_pages, |
11c65dcc | 308 | struct iov_iter *i) |
39279cc3 | 309 | { |
914ee295 | 310 | size_t copied = 0; |
d0215f3e | 311 | size_t total_copied = 0; |
11c65dcc | 312 | int pg = 0; |
39279cc3 CM |
313 | int offset = pos & (PAGE_CACHE_SIZE - 1); |
314 | ||
11c65dcc | 315 | while (write_bytes > 0) { |
39279cc3 CM |
316 | size_t count = min_t(size_t, |
317 | PAGE_CACHE_SIZE - offset, write_bytes); | |
11c65dcc | 318 | struct page *page = prepared_pages[pg]; |
914ee295 XZ |
319 | /* |
320 | * Copy data from userspace to the current page | |
321 | * | |
322 | * Disable pagefault to avoid recursive lock since | |
323 | * the pages are already locked | |
324 | */ | |
325 | pagefault_disable(); | |
326 | copied = iov_iter_copy_from_user_atomic(page, i, offset, count); | |
327 | pagefault_enable(); | |
11c65dcc | 328 | |
39279cc3 CM |
329 | /* Flush processor's dcache for this page */ |
330 | flush_dcache_page(page); | |
31339acd CM |
331 | |
332 | /* | |
333 | * if we get a partial write, we can end up with | |
334 | * partially up to date pages. These add | |
335 | * a lot of complexity, so make sure they don't | |
336 | * happen by forcing this copy to be retried. | |
337 | * | |
338 | * The rest of the btrfs_file_write code will fall | |
339 | * back to page at a time copies after we return 0. | |
340 | */ | |
341 | if (!PageUptodate(page) && copied < count) | |
342 | copied = 0; | |
343 | ||
11c65dcc JB |
344 | iov_iter_advance(i, copied); |
345 | write_bytes -= copied; | |
914ee295 | 346 | total_copied += copied; |
39279cc3 | 347 | |
914ee295 | 348 | /* Return to btrfs_file_aio_write to fault page */ |
9f570b8d | 349 | if (unlikely(copied == 0)) |
914ee295 | 350 | break; |
11c65dcc JB |
351 | |
352 | if (unlikely(copied < PAGE_CACHE_SIZE - offset)) { | |
353 | offset += copied; | |
354 | } else { | |
355 | pg++; | |
356 | offset = 0; | |
357 | } | |
39279cc3 | 358 | } |
914ee295 | 359 | return total_copied; |
39279cc3 CM |
360 | } |
361 | ||
d352ac68 CM |
362 | /* |
363 | * unlocks pages after btrfs_file_write is done with them | |
364 | */ | |
be1a12a0 | 365 | void btrfs_drop_pages(struct page **pages, size_t num_pages) |
39279cc3 CM |
366 | { |
367 | size_t i; | |
368 | for (i = 0; i < num_pages; i++) { | |
d352ac68 CM |
369 | /* page checked is some magic around finding pages that |
370 | * have been modified without going through btrfs_set_page_dirty | |
371 | * clear it here | |
372 | */ | |
4a096752 | 373 | ClearPageChecked(pages[i]); |
39279cc3 CM |
374 | unlock_page(pages[i]); |
375 | mark_page_accessed(pages[i]); | |
376 | page_cache_release(pages[i]); | |
377 | } | |
378 | } | |
379 | ||
d352ac68 CM |
380 | /* |
381 | * after copy_from_user, pages need to be dirtied and we need to make | |
382 | * sure holes are created between the current EOF and the start of | |
383 | * any next extents (if required). | |
384 | * | |
385 | * this also makes the decision about creating an inline extent vs | |
386 | * doing real data extents, marking pages dirty and delalloc as required. | |
387 | */ | |
be1a12a0 JB |
388 | int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode, |
389 | struct page **pages, size_t num_pages, | |
390 | loff_t pos, size_t write_bytes, | |
391 | struct extent_state **cached) | |
39279cc3 | 392 | { |
39279cc3 | 393 | int err = 0; |
a52d9a80 | 394 | int i; |
db94535d | 395 | u64 num_bytes; |
a52d9a80 CM |
396 | u64 start_pos; |
397 | u64 end_of_last_block; | |
398 | u64 end_pos = pos + write_bytes; | |
399 | loff_t isize = i_size_read(inode); | |
39279cc3 | 400 | |
5f39d397 | 401 | start_pos = pos & ~((u64)root->sectorsize - 1); |
db94535d CM |
402 | num_bytes = (write_bytes + pos - start_pos + |
403 | root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
39279cc3 | 404 | |
db94535d | 405 | end_of_last_block = start_pos + num_bytes - 1; |
2ac55d41 | 406 | err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block, |
be1a12a0 | 407 | cached); |
d0215f3e JB |
408 | if (err) |
409 | return err; | |
9ed74f2d | 410 | |
c8b97818 CM |
411 | for (i = 0; i < num_pages; i++) { |
412 | struct page *p = pages[i]; | |
413 | SetPageUptodate(p); | |
414 | ClearPageChecked(p); | |
415 | set_page_dirty(p); | |
a52d9a80 | 416 | } |
9f570b8d JB |
417 | |
418 | /* | |
419 | * we've only changed i_size in ram, and we haven't updated | |
420 | * the disk i_size. There is no need to log the inode | |
421 | * at this time. | |
422 | */ | |
423 | if (end_pos > isize) | |
a52d9a80 | 424 | i_size_write(inode, end_pos); |
a22285a6 | 425 | return 0; |
39279cc3 CM |
426 | } |
427 | ||
d352ac68 CM |
428 | /* |
429 | * this drops all the extents in the cache that intersect the range | |
430 | * [start, end]. Existing extents are split as required. | |
431 | */ | |
5b21f2ed ZY |
432 | int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end, |
433 | int skip_pinned) | |
a52d9a80 CM |
434 | { |
435 | struct extent_map *em; | |
3b951516 CM |
436 | struct extent_map *split = NULL; |
437 | struct extent_map *split2 = NULL; | |
a52d9a80 | 438 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
39b5637f | 439 | u64 len = end - start + 1; |
3b951516 CM |
440 | int ret; |
441 | int testend = 1; | |
5b21f2ed | 442 | unsigned long flags; |
c8b97818 | 443 | int compressed = 0; |
a52d9a80 | 444 | |
e6dcd2dc | 445 | WARN_ON(end < start); |
3b951516 | 446 | if (end == (u64)-1) { |
39b5637f | 447 | len = (u64)-1; |
3b951516 CM |
448 | testend = 0; |
449 | } | |
d397712b | 450 | while (1) { |
3b951516 | 451 | if (!split) |
172ddd60 | 452 | split = alloc_extent_map(); |
3b951516 | 453 | if (!split2) |
172ddd60 | 454 | split2 = alloc_extent_map(); |
c26a9203 | 455 | BUG_ON(!split || !split2); |
3b951516 | 456 | |
890871be | 457 | write_lock(&em_tree->lock); |
39b5637f | 458 | em = lookup_extent_mapping(em_tree, start, len); |
d1310b2e | 459 | if (!em) { |
890871be | 460 | write_unlock(&em_tree->lock); |
a52d9a80 | 461 | break; |
d1310b2e | 462 | } |
5b21f2ed ZY |
463 | flags = em->flags; |
464 | if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) { | |
55ef6899 | 465 | if (testend && em->start + em->len >= start + len) { |
5b21f2ed | 466 | free_extent_map(em); |
a1ed835e | 467 | write_unlock(&em_tree->lock); |
5b21f2ed ZY |
468 | break; |
469 | } | |
55ef6899 YZ |
470 | start = em->start + em->len; |
471 | if (testend) | |
5b21f2ed | 472 | len = start + len - (em->start + em->len); |
5b21f2ed | 473 | free_extent_map(em); |
a1ed835e | 474 | write_unlock(&em_tree->lock); |
5b21f2ed ZY |
475 | continue; |
476 | } | |
c8b97818 | 477 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
3ce7e67a | 478 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
a52d9a80 | 479 | remove_extent_mapping(em_tree, em); |
3b951516 CM |
480 | |
481 | if (em->block_start < EXTENT_MAP_LAST_BYTE && | |
482 | em->start < start) { | |
483 | split->start = em->start; | |
484 | split->len = start - em->start; | |
ff5b7ee3 | 485 | split->orig_start = em->orig_start; |
3b951516 | 486 | split->block_start = em->block_start; |
c8b97818 CM |
487 | |
488 | if (compressed) | |
489 | split->block_len = em->block_len; | |
490 | else | |
491 | split->block_len = split->len; | |
492 | ||
3b951516 | 493 | split->bdev = em->bdev; |
5b21f2ed | 494 | split->flags = flags; |
261507a0 | 495 | split->compress_type = em->compress_type; |
3b951516 CM |
496 | ret = add_extent_mapping(em_tree, split); |
497 | BUG_ON(ret); | |
498 | free_extent_map(split); | |
499 | split = split2; | |
500 | split2 = NULL; | |
501 | } | |
502 | if (em->block_start < EXTENT_MAP_LAST_BYTE && | |
503 | testend && em->start + em->len > start + len) { | |
504 | u64 diff = start + len - em->start; | |
505 | ||
506 | split->start = start + len; | |
507 | split->len = em->start + em->len - (start + len); | |
508 | split->bdev = em->bdev; | |
5b21f2ed | 509 | split->flags = flags; |
261507a0 | 510 | split->compress_type = em->compress_type; |
3b951516 | 511 | |
c8b97818 CM |
512 | if (compressed) { |
513 | split->block_len = em->block_len; | |
514 | split->block_start = em->block_start; | |
445a6944 | 515 | split->orig_start = em->orig_start; |
c8b97818 CM |
516 | } else { |
517 | split->block_len = split->len; | |
518 | split->block_start = em->block_start + diff; | |
445a6944 | 519 | split->orig_start = split->start; |
c8b97818 | 520 | } |
3b951516 CM |
521 | |
522 | ret = add_extent_mapping(em_tree, split); | |
523 | BUG_ON(ret); | |
524 | free_extent_map(split); | |
525 | split = NULL; | |
526 | } | |
890871be | 527 | write_unlock(&em_tree->lock); |
d1310b2e | 528 | |
a52d9a80 CM |
529 | /* once for us */ |
530 | free_extent_map(em); | |
531 | /* once for the tree*/ | |
532 | free_extent_map(em); | |
533 | } | |
3b951516 CM |
534 | if (split) |
535 | free_extent_map(split); | |
536 | if (split2) | |
537 | free_extent_map(split2); | |
a52d9a80 CM |
538 | return 0; |
539 | } | |
540 | ||
39279cc3 CM |
541 | /* |
542 | * this is very complex, but the basic idea is to drop all extents | |
543 | * in the range start - end. hint_block is filled in with a block number | |
544 | * that would be a good hint to the block allocator for this file. | |
545 | * | |
546 | * If an extent intersects the range but is not entirely inside the range | |
547 | * it is either truncated or split. Anything entirely inside the range | |
548 | * is deleted from the tree. | |
549 | */ | |
920bbbfb YZ |
550 | int btrfs_drop_extents(struct btrfs_trans_handle *trans, struct inode *inode, |
551 | u64 start, u64 end, u64 *hint_byte, int drop_cache) | |
39279cc3 | 552 | { |
920bbbfb | 553 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5f39d397 | 554 | struct extent_buffer *leaf; |
920bbbfb | 555 | struct btrfs_file_extent_item *fi; |
39279cc3 | 556 | struct btrfs_path *path; |
00f5c795 | 557 | struct btrfs_key key; |
920bbbfb | 558 | struct btrfs_key new_key; |
33345d01 | 559 | u64 ino = btrfs_ino(inode); |
920bbbfb YZ |
560 | u64 search_start = start; |
561 | u64 disk_bytenr = 0; | |
562 | u64 num_bytes = 0; | |
563 | u64 extent_offset = 0; | |
564 | u64 extent_end = 0; | |
565 | int del_nr = 0; | |
566 | int del_slot = 0; | |
567 | int extent_type; | |
ccd467d6 | 568 | int recow; |
00f5c795 | 569 | int ret; |
39279cc3 | 570 | |
a1ed835e CM |
571 | if (drop_cache) |
572 | btrfs_drop_extent_cache(inode, start, end - 1, 0); | |
a52d9a80 | 573 | |
39279cc3 CM |
574 | path = btrfs_alloc_path(); |
575 | if (!path) | |
576 | return -ENOMEM; | |
920bbbfb | 577 | |
d397712b | 578 | while (1) { |
ccd467d6 | 579 | recow = 0; |
33345d01 | 580 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
39279cc3 CM |
581 | search_start, -1); |
582 | if (ret < 0) | |
920bbbfb YZ |
583 | break; |
584 | if (ret > 0 && path->slots[0] > 0 && search_start == start) { | |
585 | leaf = path->nodes[0]; | |
586 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1); | |
33345d01 | 587 | if (key.objectid == ino && |
920bbbfb YZ |
588 | key.type == BTRFS_EXTENT_DATA_KEY) |
589 | path->slots[0]--; | |
39279cc3 | 590 | } |
920bbbfb | 591 | ret = 0; |
8c2383c3 | 592 | next_slot: |
5f39d397 | 593 | leaf = path->nodes[0]; |
920bbbfb YZ |
594 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { |
595 | BUG_ON(del_nr > 0); | |
596 | ret = btrfs_next_leaf(root, path); | |
597 | if (ret < 0) | |
598 | break; | |
599 | if (ret > 0) { | |
600 | ret = 0; | |
601 | break; | |
8c2383c3 | 602 | } |
920bbbfb YZ |
603 | leaf = path->nodes[0]; |
604 | recow = 1; | |
605 | } | |
606 | ||
607 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
33345d01 | 608 | if (key.objectid > ino || |
920bbbfb YZ |
609 | key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end) |
610 | break; | |
611 | ||
612 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
613 | struct btrfs_file_extent_item); | |
614 | extent_type = btrfs_file_extent_type(leaf, fi); | |
615 | ||
616 | if (extent_type == BTRFS_FILE_EXTENT_REG || | |
617 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
618 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
619 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
620 | extent_offset = btrfs_file_extent_offset(leaf, fi); | |
621 | extent_end = key.offset + | |
622 | btrfs_file_extent_num_bytes(leaf, fi); | |
623 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
624 | extent_end = key.offset + | |
625 | btrfs_file_extent_inline_len(leaf, fi); | |
8c2383c3 | 626 | } else { |
920bbbfb | 627 | WARN_ON(1); |
8c2383c3 | 628 | extent_end = search_start; |
39279cc3 CM |
629 | } |
630 | ||
920bbbfb YZ |
631 | if (extent_end <= search_start) { |
632 | path->slots[0]++; | |
8c2383c3 | 633 | goto next_slot; |
39279cc3 CM |
634 | } |
635 | ||
920bbbfb YZ |
636 | search_start = max(key.offset, start); |
637 | if (recow) { | |
b3b4aa74 | 638 | btrfs_release_path(path); |
920bbbfb | 639 | continue; |
39279cc3 | 640 | } |
6643558d | 641 | |
920bbbfb YZ |
642 | /* |
643 | * | - range to drop - | | |
644 | * | -------- extent -------- | | |
645 | */ | |
646 | if (start > key.offset && end < extent_end) { | |
647 | BUG_ON(del_nr > 0); | |
648 | BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE); | |
649 | ||
650 | memcpy(&new_key, &key, sizeof(new_key)); | |
651 | new_key.offset = start; | |
652 | ret = btrfs_duplicate_item(trans, root, path, | |
653 | &new_key); | |
654 | if (ret == -EAGAIN) { | |
b3b4aa74 | 655 | btrfs_release_path(path); |
920bbbfb | 656 | continue; |
6643558d | 657 | } |
920bbbfb YZ |
658 | if (ret < 0) |
659 | break; | |
660 | ||
661 | leaf = path->nodes[0]; | |
662 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
663 | struct btrfs_file_extent_item); | |
664 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
665 | start - key.offset); | |
666 | ||
667 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
668 | struct btrfs_file_extent_item); | |
669 | ||
670 | extent_offset += start - key.offset; | |
671 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); | |
672 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
673 | extent_end - start); | |
674 | btrfs_mark_buffer_dirty(leaf); | |
675 | ||
676 | if (disk_bytenr > 0) { | |
771ed689 | 677 | ret = btrfs_inc_extent_ref(trans, root, |
920bbbfb YZ |
678 | disk_bytenr, num_bytes, 0, |
679 | root->root_key.objectid, | |
680 | new_key.objectid, | |
681 | start - extent_offset); | |
771ed689 | 682 | BUG_ON(ret); |
920bbbfb | 683 | *hint_byte = disk_bytenr; |
771ed689 | 684 | } |
920bbbfb | 685 | key.offset = start; |
6643558d | 686 | } |
920bbbfb YZ |
687 | /* |
688 | * | ---- range to drop ----- | | |
689 | * | -------- extent -------- | | |
690 | */ | |
691 | if (start <= key.offset && end < extent_end) { | |
692 | BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE); | |
6643558d | 693 | |
920bbbfb YZ |
694 | memcpy(&new_key, &key, sizeof(new_key)); |
695 | new_key.offset = end; | |
696 | btrfs_set_item_key_safe(trans, root, path, &new_key); | |
6643558d | 697 | |
920bbbfb YZ |
698 | extent_offset += end - key.offset; |
699 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); | |
700 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
701 | extent_end - end); | |
702 | btrfs_mark_buffer_dirty(leaf); | |
703 | if (disk_bytenr > 0) { | |
704 | inode_sub_bytes(inode, end - key.offset); | |
705 | *hint_byte = disk_bytenr; | |
39279cc3 | 706 | } |
920bbbfb | 707 | break; |
39279cc3 | 708 | } |
771ed689 | 709 | |
920bbbfb YZ |
710 | search_start = extent_end; |
711 | /* | |
712 | * | ---- range to drop ----- | | |
713 | * | -------- extent -------- | | |
714 | */ | |
715 | if (start > key.offset && end >= extent_end) { | |
716 | BUG_ON(del_nr > 0); | |
717 | BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE); | |
8c2383c3 | 718 | |
920bbbfb YZ |
719 | btrfs_set_file_extent_num_bytes(leaf, fi, |
720 | start - key.offset); | |
721 | btrfs_mark_buffer_dirty(leaf); | |
722 | if (disk_bytenr > 0) { | |
723 | inode_sub_bytes(inode, extent_end - start); | |
724 | *hint_byte = disk_bytenr; | |
725 | } | |
726 | if (end == extent_end) | |
727 | break; | |
c8b97818 | 728 | |
920bbbfb YZ |
729 | path->slots[0]++; |
730 | goto next_slot; | |
31840ae1 ZY |
731 | } |
732 | ||
920bbbfb YZ |
733 | /* |
734 | * | ---- range to drop ----- | | |
735 | * | ------ extent ------ | | |
736 | */ | |
737 | if (start <= key.offset && end >= extent_end) { | |
738 | if (del_nr == 0) { | |
739 | del_slot = path->slots[0]; | |
740 | del_nr = 1; | |
741 | } else { | |
742 | BUG_ON(del_slot + del_nr != path->slots[0]); | |
743 | del_nr++; | |
744 | } | |
31840ae1 | 745 | |
920bbbfb | 746 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
a76a3cd4 | 747 | inode_sub_bytes(inode, |
920bbbfb YZ |
748 | extent_end - key.offset); |
749 | extent_end = ALIGN(extent_end, | |
750 | root->sectorsize); | |
751 | } else if (disk_bytenr > 0) { | |
31840ae1 | 752 | ret = btrfs_free_extent(trans, root, |
920bbbfb YZ |
753 | disk_bytenr, num_bytes, 0, |
754 | root->root_key.objectid, | |
5d4f98a2 | 755 | key.objectid, key.offset - |
920bbbfb | 756 | extent_offset); |
31840ae1 | 757 | BUG_ON(ret); |
920bbbfb YZ |
758 | inode_sub_bytes(inode, |
759 | extent_end - key.offset); | |
760 | *hint_byte = disk_bytenr; | |
31840ae1 | 761 | } |
31840ae1 | 762 | |
920bbbfb YZ |
763 | if (end == extent_end) |
764 | break; | |
765 | ||
766 | if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) { | |
767 | path->slots[0]++; | |
768 | goto next_slot; | |
769 | } | |
770 | ||
771 | ret = btrfs_del_items(trans, root, path, del_slot, | |
772 | del_nr); | |
773 | BUG_ON(ret); | |
774 | ||
775 | del_nr = 0; | |
776 | del_slot = 0; | |
777 | ||
b3b4aa74 | 778 | btrfs_release_path(path); |
920bbbfb | 779 | continue; |
39279cc3 | 780 | } |
920bbbfb YZ |
781 | |
782 | BUG_ON(1); | |
39279cc3 | 783 | } |
920bbbfb YZ |
784 | |
785 | if (del_nr > 0) { | |
786 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); | |
787 | BUG_ON(ret); | |
6643558d | 788 | } |
920bbbfb YZ |
789 | |
790 | btrfs_free_path(path); | |
39279cc3 CM |
791 | return ret; |
792 | } | |
793 | ||
d899e052 | 794 | static int extent_mergeable(struct extent_buffer *leaf, int slot, |
6c7d54ac YZ |
795 | u64 objectid, u64 bytenr, u64 orig_offset, |
796 | u64 *start, u64 *end) | |
d899e052 YZ |
797 | { |
798 | struct btrfs_file_extent_item *fi; | |
799 | struct btrfs_key key; | |
800 | u64 extent_end; | |
801 | ||
802 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) | |
803 | return 0; | |
804 | ||
805 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
806 | if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY) | |
807 | return 0; | |
808 | ||
809 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
810 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG || | |
811 | btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr || | |
6c7d54ac | 812 | btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset || |
d899e052 YZ |
813 | btrfs_file_extent_compression(leaf, fi) || |
814 | btrfs_file_extent_encryption(leaf, fi) || | |
815 | btrfs_file_extent_other_encoding(leaf, fi)) | |
816 | return 0; | |
817 | ||
818 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
819 | if ((*start && *start != key.offset) || (*end && *end != extent_end)) | |
820 | return 0; | |
821 | ||
822 | *start = key.offset; | |
823 | *end = extent_end; | |
824 | return 1; | |
825 | } | |
826 | ||
827 | /* | |
828 | * Mark extent in the range start - end as written. | |
829 | * | |
830 | * This changes extent type from 'pre-allocated' to 'regular'. If only | |
831 | * part of extent is marked as written, the extent will be split into | |
832 | * two or three. | |
833 | */ | |
834 | int btrfs_mark_extent_written(struct btrfs_trans_handle *trans, | |
d899e052 YZ |
835 | struct inode *inode, u64 start, u64 end) |
836 | { | |
920bbbfb | 837 | struct btrfs_root *root = BTRFS_I(inode)->root; |
d899e052 YZ |
838 | struct extent_buffer *leaf; |
839 | struct btrfs_path *path; | |
840 | struct btrfs_file_extent_item *fi; | |
841 | struct btrfs_key key; | |
920bbbfb | 842 | struct btrfs_key new_key; |
d899e052 YZ |
843 | u64 bytenr; |
844 | u64 num_bytes; | |
845 | u64 extent_end; | |
5d4f98a2 | 846 | u64 orig_offset; |
d899e052 YZ |
847 | u64 other_start; |
848 | u64 other_end; | |
920bbbfb YZ |
849 | u64 split; |
850 | int del_nr = 0; | |
851 | int del_slot = 0; | |
6c7d54ac | 852 | int recow; |
d899e052 | 853 | int ret; |
33345d01 | 854 | u64 ino = btrfs_ino(inode); |
d899e052 YZ |
855 | |
856 | btrfs_drop_extent_cache(inode, start, end - 1, 0); | |
857 | ||
858 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
859 | if (!path) |
860 | return -ENOMEM; | |
d899e052 | 861 | again: |
6c7d54ac | 862 | recow = 0; |
920bbbfb | 863 | split = start; |
33345d01 | 864 | key.objectid = ino; |
d899e052 | 865 | key.type = BTRFS_EXTENT_DATA_KEY; |
920bbbfb | 866 | key.offset = split; |
d899e052 YZ |
867 | |
868 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
41415730 JB |
869 | if (ret < 0) |
870 | goto out; | |
d899e052 YZ |
871 | if (ret > 0 && path->slots[0] > 0) |
872 | path->slots[0]--; | |
873 | ||
874 | leaf = path->nodes[0]; | |
875 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
33345d01 | 876 | BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY); |
d899e052 YZ |
877 | fi = btrfs_item_ptr(leaf, path->slots[0], |
878 | struct btrfs_file_extent_item); | |
920bbbfb YZ |
879 | BUG_ON(btrfs_file_extent_type(leaf, fi) != |
880 | BTRFS_FILE_EXTENT_PREALLOC); | |
d899e052 YZ |
881 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
882 | BUG_ON(key.offset > start || extent_end < end); | |
883 | ||
884 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
885 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
5d4f98a2 | 886 | orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi); |
6c7d54ac YZ |
887 | memcpy(&new_key, &key, sizeof(new_key)); |
888 | ||
889 | if (start == key.offset && end < extent_end) { | |
890 | other_start = 0; | |
891 | other_end = start; | |
892 | if (extent_mergeable(leaf, path->slots[0] - 1, | |
33345d01 | 893 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
894 | &other_start, &other_end)) { |
895 | new_key.offset = end; | |
896 | btrfs_set_item_key_safe(trans, root, path, &new_key); | |
897 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
898 | struct btrfs_file_extent_item); | |
899 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
900 | extent_end - end); | |
901 | btrfs_set_file_extent_offset(leaf, fi, | |
902 | end - orig_offset); | |
903 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
904 | struct btrfs_file_extent_item); | |
905 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
906 | end - other_start); | |
907 | btrfs_mark_buffer_dirty(leaf); | |
908 | goto out; | |
909 | } | |
910 | } | |
911 | ||
912 | if (start > key.offset && end == extent_end) { | |
913 | other_start = end; | |
914 | other_end = 0; | |
915 | if (extent_mergeable(leaf, path->slots[0] + 1, | |
33345d01 | 916 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
917 | &other_start, &other_end)) { |
918 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
919 | struct btrfs_file_extent_item); | |
920 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
921 | start - key.offset); | |
922 | path->slots[0]++; | |
923 | new_key.offset = start; | |
924 | btrfs_set_item_key_safe(trans, root, path, &new_key); | |
925 | ||
926 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
927 | struct btrfs_file_extent_item); | |
928 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
929 | other_end - start); | |
930 | btrfs_set_file_extent_offset(leaf, fi, | |
931 | start - orig_offset); | |
932 | btrfs_mark_buffer_dirty(leaf); | |
933 | goto out; | |
934 | } | |
935 | } | |
d899e052 | 936 | |
920bbbfb YZ |
937 | while (start > key.offset || end < extent_end) { |
938 | if (key.offset == start) | |
939 | split = end; | |
940 | ||
920bbbfb YZ |
941 | new_key.offset = split; |
942 | ret = btrfs_duplicate_item(trans, root, path, &new_key); | |
943 | if (ret == -EAGAIN) { | |
b3b4aa74 | 944 | btrfs_release_path(path); |
920bbbfb | 945 | goto again; |
d899e052 | 946 | } |
920bbbfb | 947 | BUG_ON(ret < 0); |
d899e052 | 948 | |
920bbbfb YZ |
949 | leaf = path->nodes[0]; |
950 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
d899e052 | 951 | struct btrfs_file_extent_item); |
d899e052 | 952 | btrfs_set_file_extent_num_bytes(leaf, fi, |
920bbbfb YZ |
953 | split - key.offset); |
954 | ||
955 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
956 | struct btrfs_file_extent_item); | |
957 | ||
958 | btrfs_set_file_extent_offset(leaf, fi, split - orig_offset); | |
959 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
960 | extent_end - split); | |
d899e052 YZ |
961 | btrfs_mark_buffer_dirty(leaf); |
962 | ||
920bbbfb YZ |
963 | ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0, |
964 | root->root_key.objectid, | |
33345d01 | 965 | ino, orig_offset); |
d899e052 | 966 | BUG_ON(ret); |
d899e052 | 967 | |
920bbbfb YZ |
968 | if (split == start) { |
969 | key.offset = start; | |
970 | } else { | |
971 | BUG_ON(start != key.offset); | |
d899e052 | 972 | path->slots[0]--; |
920bbbfb | 973 | extent_end = end; |
d899e052 | 974 | } |
6c7d54ac | 975 | recow = 1; |
d899e052 YZ |
976 | } |
977 | ||
920bbbfb YZ |
978 | other_start = end; |
979 | other_end = 0; | |
6c7d54ac | 980 | if (extent_mergeable(leaf, path->slots[0] + 1, |
33345d01 | 981 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
982 | &other_start, &other_end)) { |
983 | if (recow) { | |
b3b4aa74 | 984 | btrfs_release_path(path); |
6c7d54ac YZ |
985 | goto again; |
986 | } | |
920bbbfb YZ |
987 | extent_end = other_end; |
988 | del_slot = path->slots[0] + 1; | |
989 | del_nr++; | |
990 | ret = btrfs_free_extent(trans, root, bytenr, num_bytes, | |
991 | 0, root->root_key.objectid, | |
33345d01 | 992 | ino, orig_offset); |
920bbbfb | 993 | BUG_ON(ret); |
d899e052 | 994 | } |
920bbbfb YZ |
995 | other_start = 0; |
996 | other_end = start; | |
6c7d54ac | 997 | if (extent_mergeable(leaf, path->slots[0] - 1, |
33345d01 | 998 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
999 | &other_start, &other_end)) { |
1000 | if (recow) { | |
b3b4aa74 | 1001 | btrfs_release_path(path); |
6c7d54ac YZ |
1002 | goto again; |
1003 | } | |
920bbbfb YZ |
1004 | key.offset = other_start; |
1005 | del_slot = path->slots[0]; | |
1006 | del_nr++; | |
1007 | ret = btrfs_free_extent(trans, root, bytenr, num_bytes, | |
1008 | 0, root->root_key.objectid, | |
33345d01 | 1009 | ino, orig_offset); |
920bbbfb YZ |
1010 | BUG_ON(ret); |
1011 | } | |
1012 | if (del_nr == 0) { | |
3f6fae95 SL |
1013 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1014 | struct btrfs_file_extent_item); | |
920bbbfb YZ |
1015 | btrfs_set_file_extent_type(leaf, fi, |
1016 | BTRFS_FILE_EXTENT_REG); | |
1017 | btrfs_mark_buffer_dirty(leaf); | |
6c7d54ac | 1018 | } else { |
3f6fae95 SL |
1019 | fi = btrfs_item_ptr(leaf, del_slot - 1, |
1020 | struct btrfs_file_extent_item); | |
6c7d54ac YZ |
1021 | btrfs_set_file_extent_type(leaf, fi, |
1022 | BTRFS_FILE_EXTENT_REG); | |
1023 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
1024 | extent_end - key.offset); | |
1025 | btrfs_mark_buffer_dirty(leaf); | |
920bbbfb | 1026 | |
6c7d54ac YZ |
1027 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
1028 | BUG_ON(ret); | |
1029 | } | |
920bbbfb | 1030 | out: |
d899e052 YZ |
1031 | btrfs_free_path(path); |
1032 | return 0; | |
1033 | } | |
1034 | ||
b1bf862e CM |
1035 | /* |
1036 | * on error we return an unlocked page and the error value | |
1037 | * on success we return a locked page and 0 | |
1038 | */ | |
b6316429 JB |
1039 | static int prepare_uptodate_page(struct page *page, u64 pos, |
1040 | bool force_uptodate) | |
b1bf862e CM |
1041 | { |
1042 | int ret = 0; | |
1043 | ||
b6316429 JB |
1044 | if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) && |
1045 | !PageUptodate(page)) { | |
b1bf862e CM |
1046 | ret = btrfs_readpage(NULL, page); |
1047 | if (ret) | |
1048 | return ret; | |
1049 | lock_page(page); | |
1050 | if (!PageUptodate(page)) { | |
1051 | unlock_page(page); | |
1052 | return -EIO; | |
1053 | } | |
1054 | } | |
1055 | return 0; | |
1056 | } | |
1057 | ||
39279cc3 | 1058 | /* |
d352ac68 CM |
1059 | * this gets pages into the page cache and locks them down, it also properly |
1060 | * waits for data=ordered extents to finish before allowing the pages to be | |
1061 | * modified. | |
39279cc3 | 1062 | */ |
d397712b | 1063 | static noinline int prepare_pages(struct btrfs_root *root, struct file *file, |
98ed5174 CM |
1064 | struct page **pages, size_t num_pages, |
1065 | loff_t pos, unsigned long first_index, | |
b6316429 | 1066 | size_t write_bytes, bool force_uptodate) |
39279cc3 | 1067 | { |
2ac55d41 | 1068 | struct extent_state *cached_state = NULL; |
39279cc3 CM |
1069 | int i; |
1070 | unsigned long index = pos >> PAGE_CACHE_SHIFT; | |
6da6abae | 1071 | struct inode *inode = fdentry(file)->d_inode; |
3b16a4e3 | 1072 | gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); |
39279cc3 | 1073 | int err = 0; |
b1bf862e | 1074 | int faili = 0; |
8c2383c3 | 1075 | u64 start_pos; |
e6dcd2dc | 1076 | u64 last_pos; |
8c2383c3 | 1077 | |
5f39d397 | 1078 | start_pos = pos & ~((u64)root->sectorsize - 1); |
e6dcd2dc | 1079 | last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT; |
39279cc3 | 1080 | |
e6dcd2dc | 1081 | again: |
39279cc3 | 1082 | for (i = 0; i < num_pages; i++) { |
a94733d0 | 1083 | pages[i] = find_or_create_page(inode->i_mapping, index + i, |
3b16a4e3 | 1084 | mask); |
39279cc3 | 1085 | if (!pages[i]) { |
b1bf862e CM |
1086 | faili = i - 1; |
1087 | err = -ENOMEM; | |
1088 | goto fail; | |
1089 | } | |
1090 | ||
1091 | if (i == 0) | |
b6316429 JB |
1092 | err = prepare_uptodate_page(pages[i], pos, |
1093 | force_uptodate); | |
b1bf862e CM |
1094 | if (i == num_pages - 1) |
1095 | err = prepare_uptodate_page(pages[i], | |
b6316429 | 1096 | pos + write_bytes, false); |
b1bf862e CM |
1097 | if (err) { |
1098 | page_cache_release(pages[i]); | |
1099 | faili = i - 1; | |
1100 | goto fail; | |
39279cc3 | 1101 | } |
ccd467d6 | 1102 | wait_on_page_writeback(pages[i]); |
39279cc3 | 1103 | } |
b1bf862e | 1104 | err = 0; |
0762704b | 1105 | if (start_pos < inode->i_size) { |
e6dcd2dc | 1106 | struct btrfs_ordered_extent *ordered; |
2ac55d41 JB |
1107 | lock_extent_bits(&BTRFS_I(inode)->io_tree, |
1108 | start_pos, last_pos - 1, 0, &cached_state, | |
1109 | GFP_NOFS); | |
d397712b CM |
1110 | ordered = btrfs_lookup_first_ordered_extent(inode, |
1111 | last_pos - 1); | |
e6dcd2dc CM |
1112 | if (ordered && |
1113 | ordered->file_offset + ordered->len > start_pos && | |
1114 | ordered->file_offset < last_pos) { | |
1115 | btrfs_put_ordered_extent(ordered); | |
2ac55d41 JB |
1116 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, |
1117 | start_pos, last_pos - 1, | |
1118 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
1119 | for (i = 0; i < num_pages; i++) { |
1120 | unlock_page(pages[i]); | |
1121 | page_cache_release(pages[i]); | |
1122 | } | |
1123 | btrfs_wait_ordered_range(inode, start_pos, | |
1124 | last_pos - start_pos); | |
1125 | goto again; | |
1126 | } | |
1127 | if (ordered) | |
1128 | btrfs_put_ordered_extent(ordered); | |
1129 | ||
2ac55d41 | 1130 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos, |
32c00aff | 1131 | last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC | |
2ac55d41 | 1132 | EXTENT_DO_ACCOUNTING, 0, 0, &cached_state, |
0762704b | 1133 | GFP_NOFS); |
2ac55d41 JB |
1134 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, |
1135 | start_pos, last_pos - 1, &cached_state, | |
1136 | GFP_NOFS); | |
0762704b | 1137 | } |
e6dcd2dc | 1138 | for (i = 0; i < num_pages; i++) { |
f87f057b | 1139 | clear_page_dirty_for_io(pages[i]); |
e6dcd2dc CM |
1140 | set_page_extent_mapped(pages[i]); |
1141 | WARN_ON(!PageLocked(pages[i])); | |
1142 | } | |
39279cc3 | 1143 | return 0; |
b1bf862e CM |
1144 | fail: |
1145 | while (faili >= 0) { | |
1146 | unlock_page(pages[faili]); | |
1147 | page_cache_release(pages[faili]); | |
1148 | faili--; | |
1149 | } | |
1150 | return err; | |
1151 | ||
39279cc3 CM |
1152 | } |
1153 | ||
d0215f3e JB |
1154 | static noinline ssize_t __btrfs_buffered_write(struct file *file, |
1155 | struct iov_iter *i, | |
1156 | loff_t pos) | |
4b46fce2 | 1157 | { |
11c65dcc JB |
1158 | struct inode *inode = fdentry(file)->d_inode; |
1159 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
11c65dcc | 1160 | struct page **pages = NULL; |
39279cc3 | 1161 | unsigned long first_index; |
d0215f3e JB |
1162 | size_t num_written = 0; |
1163 | int nrptrs; | |
c9149235 | 1164 | int ret = 0; |
b6316429 | 1165 | bool force_page_uptodate = false; |
4b46fce2 | 1166 | |
d0215f3e | 1167 | nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) / |
11c65dcc JB |
1168 | PAGE_CACHE_SIZE, PAGE_CACHE_SIZE / |
1169 | (sizeof(struct page *))); | |
8c2383c3 | 1170 | pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL); |
d0215f3e JB |
1171 | if (!pages) |
1172 | return -ENOMEM; | |
ab93dbec | 1173 | |
39279cc3 | 1174 | first_index = pos >> PAGE_CACHE_SHIFT; |
39279cc3 | 1175 | |
d0215f3e | 1176 | while (iov_iter_count(i) > 0) { |
39279cc3 | 1177 | size_t offset = pos & (PAGE_CACHE_SIZE - 1); |
d0215f3e | 1178 | size_t write_bytes = min(iov_iter_count(i), |
11c65dcc | 1179 | nrptrs * (size_t)PAGE_CACHE_SIZE - |
8c2383c3 | 1180 | offset); |
3a90983d YZ |
1181 | size_t num_pages = (write_bytes + offset + |
1182 | PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
d0215f3e JB |
1183 | size_t dirty_pages; |
1184 | size_t copied; | |
39279cc3 | 1185 | |
8c2383c3 | 1186 | WARN_ON(num_pages > nrptrs); |
1832a6d5 | 1187 | |
914ee295 XZ |
1188 | /* |
1189 | * Fault pages before locking them in prepare_pages | |
1190 | * to avoid recursive lock | |
1191 | */ | |
d0215f3e | 1192 | if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) { |
914ee295 | 1193 | ret = -EFAULT; |
d0215f3e | 1194 | break; |
914ee295 XZ |
1195 | } |
1196 | ||
1197 | ret = btrfs_delalloc_reserve_space(inode, | |
1198 | num_pages << PAGE_CACHE_SHIFT); | |
1832a6d5 | 1199 | if (ret) |
d0215f3e | 1200 | break; |
1832a6d5 | 1201 | |
4a64001f JB |
1202 | /* |
1203 | * This is going to setup the pages array with the number of | |
1204 | * pages we want, so we don't really need to worry about the | |
1205 | * contents of pages from loop to loop | |
1206 | */ | |
39279cc3 | 1207 | ret = prepare_pages(root, file, pages, num_pages, |
b6316429 JB |
1208 | pos, first_index, write_bytes, |
1209 | force_page_uptodate); | |
6a63209f | 1210 | if (ret) { |
914ee295 XZ |
1211 | btrfs_delalloc_release_space(inode, |
1212 | num_pages << PAGE_CACHE_SHIFT); | |
d0215f3e | 1213 | break; |
6a63209f | 1214 | } |
39279cc3 | 1215 | |
914ee295 | 1216 | copied = btrfs_copy_from_user(pos, num_pages, |
d0215f3e | 1217 | write_bytes, pages, i); |
b1bf862e CM |
1218 | |
1219 | /* | |
1220 | * if we have trouble faulting in the pages, fall | |
1221 | * back to one page at a time | |
1222 | */ | |
1223 | if (copied < write_bytes) | |
1224 | nrptrs = 1; | |
1225 | ||
b6316429 JB |
1226 | if (copied == 0) { |
1227 | force_page_uptodate = true; | |
b1bf862e | 1228 | dirty_pages = 0; |
b6316429 JB |
1229 | } else { |
1230 | force_page_uptodate = false; | |
b1bf862e CM |
1231 | dirty_pages = (copied + offset + |
1232 | PAGE_CACHE_SIZE - 1) >> | |
1233 | PAGE_CACHE_SHIFT; | |
b6316429 | 1234 | } |
914ee295 | 1235 | |
d0215f3e JB |
1236 | /* |
1237 | * If we had a short copy we need to release the excess delaloc | |
1238 | * bytes we reserved. We need to increment outstanding_extents | |
1239 | * because btrfs_delalloc_release_space will decrement it, but | |
1240 | * we still have an outstanding extent for the chunk we actually | |
1241 | * managed to copy. | |
1242 | */ | |
914ee295 | 1243 | if (num_pages > dirty_pages) { |
9e0baf60 JB |
1244 | if (copied > 0) { |
1245 | spin_lock(&BTRFS_I(inode)->lock); | |
1246 | BTRFS_I(inode)->outstanding_extents++; | |
1247 | spin_unlock(&BTRFS_I(inode)->lock); | |
1248 | } | |
914ee295 XZ |
1249 | btrfs_delalloc_release_space(inode, |
1250 | (num_pages - dirty_pages) << | |
1251 | PAGE_CACHE_SHIFT); | |
1252 | } | |
1253 | ||
1254 | if (copied > 0) { | |
be1a12a0 JB |
1255 | ret = btrfs_dirty_pages(root, inode, pages, |
1256 | dirty_pages, pos, copied, | |
1257 | NULL); | |
d0215f3e JB |
1258 | if (ret) { |
1259 | btrfs_delalloc_release_space(inode, | |
1260 | dirty_pages << PAGE_CACHE_SHIFT); | |
1261 | btrfs_drop_pages(pages, num_pages); | |
1262 | break; | |
1263 | } | |
54aa1f4d | 1264 | } |
39279cc3 | 1265 | |
39279cc3 CM |
1266 | btrfs_drop_pages(pages, num_pages); |
1267 | ||
d0215f3e JB |
1268 | cond_resched(); |
1269 | ||
1270 | balance_dirty_pages_ratelimited_nr(inode->i_mapping, | |
1271 | dirty_pages); | |
1272 | if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1) | |
1273 | btrfs_btree_balance_dirty(root, 1); | |
1274 | btrfs_throttle(root); | |
cb843a6f | 1275 | |
914ee295 XZ |
1276 | pos += copied; |
1277 | num_written += copied; | |
d0215f3e | 1278 | } |
39279cc3 | 1279 | |
d0215f3e JB |
1280 | kfree(pages); |
1281 | ||
1282 | return num_written ? num_written : ret; | |
1283 | } | |
1284 | ||
1285 | static ssize_t __btrfs_direct_write(struct kiocb *iocb, | |
1286 | const struct iovec *iov, | |
1287 | unsigned long nr_segs, loff_t pos, | |
1288 | loff_t *ppos, size_t count, size_t ocount) | |
1289 | { | |
1290 | struct file *file = iocb->ki_filp; | |
1291 | struct inode *inode = fdentry(file)->d_inode; | |
1292 | struct iov_iter i; | |
1293 | ssize_t written; | |
1294 | ssize_t written_buffered; | |
1295 | loff_t endbyte; | |
1296 | int err; | |
1297 | ||
1298 | written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos, | |
1299 | count, ocount); | |
1300 | ||
1301 | /* | |
1302 | * the generic O_DIRECT will update in-memory i_size after the | |
1303 | * DIOs are done. But our endio handlers that update the on | |
1304 | * disk i_size never update past the in memory i_size. So we | |
1305 | * need one more update here to catch any additions to the | |
1306 | * file | |
1307 | */ | |
1308 | if (inode->i_size != BTRFS_I(inode)->disk_i_size) { | |
1309 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); | |
1310 | mark_inode_dirty(inode); | |
1311 | } | |
1312 | ||
1313 | if (written < 0 || written == count) | |
1314 | return written; | |
1315 | ||
1316 | pos += written; | |
1317 | count -= written; | |
1318 | iov_iter_init(&i, iov, nr_segs, count, written); | |
1319 | written_buffered = __btrfs_buffered_write(file, &i, pos); | |
1320 | if (written_buffered < 0) { | |
1321 | err = written_buffered; | |
1322 | goto out; | |
39279cc3 | 1323 | } |
d0215f3e JB |
1324 | endbyte = pos + written_buffered - 1; |
1325 | err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte); | |
1326 | if (err) | |
1327 | goto out; | |
1328 | written += written_buffered; | |
1329 | *ppos = pos + written_buffered; | |
1330 | invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT, | |
1331 | endbyte >> PAGE_CACHE_SHIFT); | |
39279cc3 | 1332 | out: |
d0215f3e JB |
1333 | return written ? written : err; |
1334 | } | |
5b92ee72 | 1335 | |
d0215f3e JB |
1336 | static ssize_t btrfs_file_aio_write(struct kiocb *iocb, |
1337 | const struct iovec *iov, | |
1338 | unsigned long nr_segs, loff_t pos) | |
1339 | { | |
1340 | struct file *file = iocb->ki_filp; | |
1341 | struct inode *inode = fdentry(file)->d_inode; | |
1342 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1343 | loff_t *ppos = &iocb->ki_pos; | |
0c1a98c8 | 1344 | u64 start_pos; |
d0215f3e JB |
1345 | ssize_t num_written = 0; |
1346 | ssize_t err = 0; | |
1347 | size_t count, ocount; | |
1348 | ||
1349 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); | |
1350 | ||
1351 | mutex_lock(&inode->i_mutex); | |
1352 | ||
1353 | err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ); | |
1354 | if (err) { | |
1355 | mutex_unlock(&inode->i_mutex); | |
1356 | goto out; | |
1357 | } | |
1358 | count = ocount; | |
1359 | ||
1360 | current->backing_dev_info = inode->i_mapping->backing_dev_info; | |
1361 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); | |
1362 | if (err) { | |
1363 | mutex_unlock(&inode->i_mutex); | |
1364 | goto out; | |
1365 | } | |
1366 | ||
1367 | if (count == 0) { | |
1368 | mutex_unlock(&inode->i_mutex); | |
1369 | goto out; | |
1370 | } | |
1371 | ||
1372 | err = file_remove_suid(file); | |
1373 | if (err) { | |
1374 | mutex_unlock(&inode->i_mutex); | |
1375 | goto out; | |
1376 | } | |
1377 | ||
1378 | /* | |
1379 | * If BTRFS flips readonly due to some impossible error | |
1380 | * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR), | |
1381 | * although we have opened a file as writable, we have | |
1382 | * to stop this write operation to ensure FS consistency. | |
1383 | */ | |
1384 | if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) { | |
1385 | mutex_unlock(&inode->i_mutex); | |
1386 | err = -EROFS; | |
1387 | goto out; | |
1388 | } | |
1389 | ||
1390 | file_update_time(file); | |
1391 | BTRFS_I(inode)->sequence++; | |
1392 | ||
0c1a98c8 MX |
1393 | start_pos = round_down(pos, root->sectorsize); |
1394 | if (start_pos > i_size_read(inode)) { | |
1395 | err = btrfs_cont_expand(inode, i_size_read(inode), start_pos); | |
1396 | if (err) { | |
1397 | mutex_unlock(&inode->i_mutex); | |
1398 | goto out; | |
1399 | } | |
1400 | } | |
1401 | ||
d0215f3e JB |
1402 | if (unlikely(file->f_flags & O_DIRECT)) { |
1403 | num_written = __btrfs_direct_write(iocb, iov, nr_segs, | |
1404 | pos, ppos, count, ocount); | |
1405 | } else { | |
1406 | struct iov_iter i; | |
1407 | ||
1408 | iov_iter_init(&i, iov, nr_segs, count, num_written); | |
1409 | ||
1410 | num_written = __btrfs_buffered_write(file, &i, pos); | |
1411 | if (num_written > 0) | |
1412 | *ppos = pos + num_written; | |
1413 | } | |
1414 | ||
1415 | mutex_unlock(&inode->i_mutex); | |
2ff3e9b6 | 1416 | |
5a3f23d5 CM |
1417 | /* |
1418 | * we want to make sure fsync finds this change | |
1419 | * but we haven't joined a transaction running right now. | |
1420 | * | |
1421 | * Later on, someone is sure to update the inode and get the | |
1422 | * real transid recorded. | |
1423 | * | |
1424 | * We set last_trans now to the fs_info generation + 1, | |
1425 | * this will either be one more than the running transaction | |
1426 | * or the generation used for the next transaction if there isn't | |
1427 | * one running right now. | |
1428 | */ | |
1429 | BTRFS_I(inode)->last_trans = root->fs_info->generation + 1; | |
d0215f3e JB |
1430 | if (num_written > 0 || num_written == -EIOCBQUEUED) { |
1431 | err = generic_write_sync(file, pos, num_written); | |
1432 | if (err < 0 && num_written > 0) | |
2ff3e9b6 CM |
1433 | num_written = err; |
1434 | } | |
d0215f3e | 1435 | out: |
39279cc3 | 1436 | current->backing_dev_info = NULL; |
39279cc3 CM |
1437 | return num_written ? num_written : err; |
1438 | } | |
1439 | ||
d397712b | 1440 | int btrfs_release_file(struct inode *inode, struct file *filp) |
e1b81e67 | 1441 | { |
5a3f23d5 CM |
1442 | /* |
1443 | * ordered_data_close is set by settattr when we are about to truncate | |
1444 | * a file from a non-zero size to a zero size. This tries to | |
1445 | * flush down new bytes that may have been written if the | |
1446 | * application were using truncate to replace a file in place. | |
1447 | */ | |
1448 | if (BTRFS_I(inode)->ordered_data_close) { | |
1449 | BTRFS_I(inode)->ordered_data_close = 0; | |
1450 | btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode); | |
1451 | if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) | |
1452 | filemap_flush(inode->i_mapping); | |
1453 | } | |
6bf13c0c SW |
1454 | if (filp->private_data) |
1455 | btrfs_ioctl_trans_end(filp); | |
e1b81e67 M |
1456 | return 0; |
1457 | } | |
1458 | ||
d352ac68 CM |
1459 | /* |
1460 | * fsync call for both files and directories. This logs the inode into | |
1461 | * the tree log instead of forcing full commits whenever possible. | |
1462 | * | |
1463 | * It needs to call filemap_fdatawait so that all ordered extent updates are | |
1464 | * in the metadata btree are up to date for copying to the log. | |
1465 | * | |
1466 | * It drops the inode mutex before doing the tree log commit. This is an | |
1467 | * important optimization for directories because holding the mutex prevents | |
1468 | * new operations on the dir while we write to disk. | |
1469 | */ | |
02c24a82 | 1470 | int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync) |
39279cc3 | 1471 | { |
7ea80859 | 1472 | struct dentry *dentry = file->f_path.dentry; |
39279cc3 CM |
1473 | struct inode *inode = dentry->d_inode; |
1474 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
15ee9bc7 | 1475 | int ret = 0; |
39279cc3 CM |
1476 | struct btrfs_trans_handle *trans; |
1477 | ||
1abe9b8a | 1478 | trace_btrfs_sync_file(file, datasync); |
257c62e1 | 1479 | |
02c24a82 JB |
1480 | ret = filemap_write_and_wait_range(inode->i_mapping, start, end); |
1481 | if (ret) | |
1482 | return ret; | |
1483 | mutex_lock(&inode->i_mutex); | |
1484 | ||
257c62e1 CM |
1485 | /* we wait first, since the writeback may change the inode */ |
1486 | root->log_batch++; | |
257c62e1 CM |
1487 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
1488 | root->log_batch++; | |
1489 | ||
39279cc3 | 1490 | /* |
15ee9bc7 JB |
1491 | * check the transaction that last modified this inode |
1492 | * and see if its already been committed | |
39279cc3 | 1493 | */ |
02c24a82 JB |
1494 | if (!BTRFS_I(inode)->last_trans) { |
1495 | mutex_unlock(&inode->i_mutex); | |
15ee9bc7 | 1496 | goto out; |
02c24a82 | 1497 | } |
a2135011 | 1498 | |
257c62e1 CM |
1499 | /* |
1500 | * if the last transaction that changed this file was before | |
1501 | * the current transaction, we can bail out now without any | |
1502 | * syncing | |
1503 | */ | |
a4abeea4 | 1504 | smp_mb(); |
15ee9bc7 JB |
1505 | if (BTRFS_I(inode)->last_trans <= |
1506 | root->fs_info->last_trans_committed) { | |
1507 | BTRFS_I(inode)->last_trans = 0; | |
02c24a82 | 1508 | mutex_unlock(&inode->i_mutex); |
15ee9bc7 JB |
1509 | goto out; |
1510 | } | |
15ee9bc7 JB |
1511 | |
1512 | /* | |
a52d9a80 CM |
1513 | * ok we haven't committed the transaction yet, lets do a commit |
1514 | */ | |
6f902af4 | 1515 | if (file->private_data) |
6bf13c0c SW |
1516 | btrfs_ioctl_trans_end(file); |
1517 | ||
a22285a6 YZ |
1518 | trans = btrfs_start_transaction(root, 0); |
1519 | if (IS_ERR(trans)) { | |
1520 | ret = PTR_ERR(trans); | |
02c24a82 | 1521 | mutex_unlock(&inode->i_mutex); |
39279cc3 CM |
1522 | goto out; |
1523 | } | |
e02119d5 | 1524 | |
2cfbd50b | 1525 | ret = btrfs_log_dentry_safe(trans, root, dentry); |
02c24a82 JB |
1526 | if (ret < 0) { |
1527 | mutex_unlock(&inode->i_mutex); | |
e02119d5 | 1528 | goto out; |
02c24a82 | 1529 | } |
49eb7e46 CM |
1530 | |
1531 | /* we've logged all the items and now have a consistent | |
1532 | * version of the file in the log. It is possible that | |
1533 | * someone will come in and modify the file, but that's | |
1534 | * fine because the log is consistent on disk, and we | |
1535 | * have references to all of the file's extents | |
1536 | * | |
1537 | * It is possible that someone will come in and log the | |
1538 | * file again, but that will end up using the synchronization | |
1539 | * inside btrfs_sync_log to keep things safe. | |
1540 | */ | |
02c24a82 | 1541 | mutex_unlock(&inode->i_mutex); |
49eb7e46 | 1542 | |
257c62e1 CM |
1543 | if (ret != BTRFS_NO_LOG_SYNC) { |
1544 | if (ret > 0) { | |
12fcfd22 | 1545 | ret = btrfs_commit_transaction(trans, root); |
257c62e1 CM |
1546 | } else { |
1547 | ret = btrfs_sync_log(trans, root); | |
1548 | if (ret == 0) | |
1549 | ret = btrfs_end_transaction(trans, root); | |
1550 | else | |
1551 | ret = btrfs_commit_transaction(trans, root); | |
1552 | } | |
1553 | } else { | |
1554 | ret = btrfs_end_transaction(trans, root); | |
e02119d5 | 1555 | } |
39279cc3 | 1556 | out: |
014e4ac4 | 1557 | return ret > 0 ? -EIO : ret; |
39279cc3 CM |
1558 | } |
1559 | ||
f0f37e2f | 1560 | static const struct vm_operations_struct btrfs_file_vm_ops = { |
92fee66d | 1561 | .fault = filemap_fault, |
9ebefb18 CM |
1562 | .page_mkwrite = btrfs_page_mkwrite, |
1563 | }; | |
1564 | ||
1565 | static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma) | |
1566 | { | |
058a457e MX |
1567 | struct address_space *mapping = filp->f_mapping; |
1568 | ||
1569 | if (!mapping->a_ops->readpage) | |
1570 | return -ENOEXEC; | |
1571 | ||
9ebefb18 | 1572 | file_accessed(filp); |
058a457e MX |
1573 | vma->vm_ops = &btrfs_file_vm_ops; |
1574 | vma->vm_flags |= VM_CAN_NONLINEAR; | |
1575 | ||
9ebefb18 CM |
1576 | return 0; |
1577 | } | |
1578 | ||
2fe17c10 CH |
1579 | static long btrfs_fallocate(struct file *file, int mode, |
1580 | loff_t offset, loff_t len) | |
1581 | { | |
1582 | struct inode *inode = file->f_path.dentry->d_inode; | |
1583 | struct extent_state *cached_state = NULL; | |
1584 | u64 cur_offset; | |
1585 | u64 last_byte; | |
1586 | u64 alloc_start; | |
1587 | u64 alloc_end; | |
1588 | u64 alloc_hint = 0; | |
1589 | u64 locked_end; | |
1590 | u64 mask = BTRFS_I(inode)->root->sectorsize - 1; | |
1591 | struct extent_map *em; | |
1592 | int ret; | |
1593 | ||
1594 | alloc_start = offset & ~mask; | |
1595 | alloc_end = (offset + len + mask) & ~mask; | |
1596 | ||
1597 | /* We only support the FALLOC_FL_KEEP_SIZE mode */ | |
1598 | if (mode & ~FALLOC_FL_KEEP_SIZE) | |
1599 | return -EOPNOTSUPP; | |
1600 | ||
1601 | /* | |
1602 | * wait for ordered IO before we have any locks. We'll loop again | |
1603 | * below with the locks held. | |
1604 | */ | |
1605 | btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start); | |
1606 | ||
1607 | mutex_lock(&inode->i_mutex); | |
1608 | ret = inode_newsize_ok(inode, alloc_end); | |
1609 | if (ret) | |
1610 | goto out; | |
1611 | ||
1612 | if (alloc_start > inode->i_size) { | |
a41ad394 JB |
1613 | ret = btrfs_cont_expand(inode, i_size_read(inode), |
1614 | alloc_start); | |
2fe17c10 CH |
1615 | if (ret) |
1616 | goto out; | |
1617 | } | |
1618 | ||
2fe17c10 CH |
1619 | locked_end = alloc_end - 1; |
1620 | while (1) { | |
1621 | struct btrfs_ordered_extent *ordered; | |
1622 | ||
1623 | /* the extent lock is ordered inside the running | |
1624 | * transaction | |
1625 | */ | |
1626 | lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start, | |
1627 | locked_end, 0, &cached_state, GFP_NOFS); | |
1628 | ordered = btrfs_lookup_first_ordered_extent(inode, | |
1629 | alloc_end - 1); | |
1630 | if (ordered && | |
1631 | ordered->file_offset + ordered->len > alloc_start && | |
1632 | ordered->file_offset < alloc_end) { | |
1633 | btrfs_put_ordered_extent(ordered); | |
1634 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, | |
1635 | alloc_start, locked_end, | |
1636 | &cached_state, GFP_NOFS); | |
1637 | /* | |
1638 | * we can't wait on the range with the transaction | |
1639 | * running or with the extent lock held | |
1640 | */ | |
1641 | btrfs_wait_ordered_range(inode, alloc_start, | |
1642 | alloc_end - alloc_start); | |
1643 | } else { | |
1644 | if (ordered) | |
1645 | btrfs_put_ordered_extent(ordered); | |
1646 | break; | |
1647 | } | |
1648 | } | |
1649 | ||
1650 | cur_offset = alloc_start; | |
1651 | while (1) { | |
f1e490a7 JB |
1652 | u64 actual_end; |
1653 | ||
2fe17c10 CH |
1654 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, |
1655 | alloc_end - cur_offset, 0); | |
c704005d | 1656 | BUG_ON(IS_ERR_OR_NULL(em)); |
2fe17c10 | 1657 | last_byte = min(extent_map_end(em), alloc_end); |
f1e490a7 | 1658 | actual_end = min_t(u64, extent_map_end(em), offset + len); |
2fe17c10 | 1659 | last_byte = (last_byte + mask) & ~mask; |
f1e490a7 | 1660 | |
2fe17c10 CH |
1661 | if (em->block_start == EXTENT_MAP_HOLE || |
1662 | (cur_offset >= inode->i_size && | |
1663 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
1b9c332b JB |
1664 | |
1665 | /* | |
1666 | * Make sure we have enough space before we do the | |
1667 | * allocation. | |
1668 | */ | |
1669 | ret = btrfs_check_data_free_space(inode, last_byte - | |
1670 | cur_offset); | |
1671 | if (ret) { | |
1672 | free_extent_map(em); | |
1673 | break; | |
1674 | } | |
1675 | ||
2fe17c10 CH |
1676 | ret = btrfs_prealloc_file_range(inode, mode, cur_offset, |
1677 | last_byte - cur_offset, | |
1678 | 1 << inode->i_blkbits, | |
1679 | offset + len, | |
1680 | &alloc_hint); | |
1b9c332b JB |
1681 | |
1682 | /* Let go of our reservation. */ | |
1683 | btrfs_free_reserved_data_space(inode, last_byte - | |
1684 | cur_offset); | |
2fe17c10 CH |
1685 | if (ret < 0) { |
1686 | free_extent_map(em); | |
1687 | break; | |
1688 | } | |
f1e490a7 JB |
1689 | } else if (actual_end > inode->i_size && |
1690 | !(mode & FALLOC_FL_KEEP_SIZE)) { | |
1691 | /* | |
1692 | * We didn't need to allocate any more space, but we | |
1693 | * still extended the size of the file so we need to | |
1694 | * update i_size. | |
1695 | */ | |
1696 | inode->i_ctime = CURRENT_TIME; | |
1697 | i_size_write(inode, actual_end); | |
1698 | btrfs_ordered_update_i_size(inode, actual_end, NULL); | |
2fe17c10 CH |
1699 | } |
1700 | free_extent_map(em); | |
1701 | ||
1702 | cur_offset = last_byte; | |
1703 | if (cur_offset >= alloc_end) { | |
1704 | ret = 0; | |
1705 | break; | |
1706 | } | |
1707 | } | |
1708 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, | |
1709 | &cached_state, GFP_NOFS); | |
2fe17c10 CH |
1710 | out: |
1711 | mutex_unlock(&inode->i_mutex); | |
1712 | return ret; | |
1713 | } | |
1714 | ||
b2675157 JB |
1715 | static int find_desired_extent(struct inode *inode, loff_t *offset, int origin) |
1716 | { | |
1717 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1718 | struct extent_map *em; | |
1719 | struct extent_state *cached_state = NULL; | |
1720 | u64 lockstart = *offset; | |
1721 | u64 lockend = i_size_read(inode); | |
1722 | u64 start = *offset; | |
1723 | u64 orig_start = *offset; | |
1724 | u64 len = i_size_read(inode); | |
1725 | u64 last_end = 0; | |
1726 | int ret = 0; | |
1727 | ||
1728 | lockend = max_t(u64, root->sectorsize, lockend); | |
1729 | if (lockend <= lockstart) | |
1730 | lockend = lockstart + root->sectorsize; | |
1731 | ||
1732 | len = lockend - lockstart + 1; | |
1733 | ||
1734 | len = max_t(u64, len, root->sectorsize); | |
1735 | if (inode->i_size == 0) | |
1736 | return -ENXIO; | |
1737 | ||
1738 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0, | |
1739 | &cached_state, GFP_NOFS); | |
1740 | ||
1741 | /* | |
1742 | * Delalloc is such a pain. If we have a hole and we have pending | |
1743 | * delalloc for a portion of the hole we will get back a hole that | |
1744 | * exists for the entire range since it hasn't been actually written | |
1745 | * yet. So to take care of this case we need to look for an extent just | |
1746 | * before the position we want in case there is outstanding delalloc | |
1747 | * going on here. | |
1748 | */ | |
1749 | if (origin == SEEK_HOLE && start != 0) { | |
1750 | if (start <= root->sectorsize) | |
1751 | em = btrfs_get_extent_fiemap(inode, NULL, 0, 0, | |
1752 | root->sectorsize, 0); | |
1753 | else | |
1754 | em = btrfs_get_extent_fiemap(inode, NULL, 0, | |
1755 | start - root->sectorsize, | |
1756 | root->sectorsize, 0); | |
1757 | if (IS_ERR(em)) { | |
1758 | ret = -ENXIO; | |
1759 | goto out; | |
1760 | } | |
1761 | last_end = em->start + em->len; | |
1762 | if (em->block_start == EXTENT_MAP_DELALLOC) | |
1763 | last_end = min_t(u64, last_end, inode->i_size); | |
1764 | free_extent_map(em); | |
1765 | } | |
1766 | ||
1767 | while (1) { | |
1768 | em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0); | |
1769 | if (IS_ERR(em)) { | |
1770 | ret = -ENXIO; | |
1771 | break; | |
1772 | } | |
1773 | ||
1774 | if (em->block_start == EXTENT_MAP_HOLE) { | |
1775 | if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) { | |
1776 | if (last_end <= orig_start) { | |
1777 | free_extent_map(em); | |
1778 | ret = -ENXIO; | |
1779 | break; | |
1780 | } | |
1781 | } | |
1782 | ||
1783 | if (origin == SEEK_HOLE) { | |
1784 | *offset = start; | |
1785 | free_extent_map(em); | |
1786 | break; | |
1787 | } | |
1788 | } else { | |
1789 | if (origin == SEEK_DATA) { | |
1790 | if (em->block_start == EXTENT_MAP_DELALLOC) { | |
1791 | if (start >= inode->i_size) { | |
1792 | free_extent_map(em); | |
1793 | ret = -ENXIO; | |
1794 | break; | |
1795 | } | |
1796 | } | |
1797 | ||
1798 | *offset = start; | |
1799 | free_extent_map(em); | |
1800 | break; | |
1801 | } | |
1802 | } | |
1803 | ||
1804 | start = em->start + em->len; | |
1805 | last_end = em->start + em->len; | |
1806 | ||
1807 | if (em->block_start == EXTENT_MAP_DELALLOC) | |
1808 | last_end = min_t(u64, last_end, inode->i_size); | |
1809 | ||
1810 | if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) { | |
1811 | free_extent_map(em); | |
1812 | ret = -ENXIO; | |
1813 | break; | |
1814 | } | |
1815 | free_extent_map(em); | |
1816 | cond_resched(); | |
1817 | } | |
1818 | if (!ret) | |
1819 | *offset = min(*offset, inode->i_size); | |
1820 | out: | |
1821 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
1822 | &cached_state, GFP_NOFS); | |
1823 | return ret; | |
1824 | } | |
1825 | ||
1826 | static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int origin) | |
1827 | { | |
1828 | struct inode *inode = file->f_mapping->host; | |
1829 | int ret; | |
1830 | ||
1831 | mutex_lock(&inode->i_mutex); | |
1832 | switch (origin) { | |
1833 | case SEEK_END: | |
1834 | case SEEK_CUR: | |
1835 | offset = generic_file_llseek_unlocked(file, offset, origin); | |
1836 | goto out; | |
1837 | case SEEK_DATA: | |
1838 | case SEEK_HOLE: | |
48802c8a JL |
1839 | if (offset >= i_size_read(inode)) { |
1840 | mutex_unlock(&inode->i_mutex); | |
1841 | return -ENXIO; | |
1842 | } | |
1843 | ||
b2675157 JB |
1844 | ret = find_desired_extent(inode, &offset, origin); |
1845 | if (ret) { | |
1846 | mutex_unlock(&inode->i_mutex); | |
1847 | return ret; | |
1848 | } | |
1849 | } | |
1850 | ||
9a4327ca | 1851 | if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET)) { |
48802c8a | 1852 | offset = -EINVAL; |
9a4327ca DC |
1853 | goto out; |
1854 | } | |
1855 | if (offset > inode->i_sb->s_maxbytes) { | |
48802c8a | 1856 | offset = -EINVAL; |
9a4327ca DC |
1857 | goto out; |
1858 | } | |
b2675157 JB |
1859 | |
1860 | /* Special lock needed here? */ | |
1861 | if (offset != file->f_pos) { | |
1862 | file->f_pos = offset; | |
1863 | file->f_version = 0; | |
1864 | } | |
1865 | out: | |
1866 | mutex_unlock(&inode->i_mutex); | |
1867 | return offset; | |
1868 | } | |
1869 | ||
828c0950 | 1870 | const struct file_operations btrfs_file_operations = { |
b2675157 | 1871 | .llseek = btrfs_file_llseek, |
39279cc3 | 1872 | .read = do_sync_read, |
4a001071 | 1873 | .write = do_sync_write, |
9ebefb18 | 1874 | .aio_read = generic_file_aio_read, |
e9906a98 | 1875 | .splice_read = generic_file_splice_read, |
11c65dcc | 1876 | .aio_write = btrfs_file_aio_write, |
9ebefb18 | 1877 | .mmap = btrfs_file_mmap, |
39279cc3 | 1878 | .open = generic_file_open, |
e1b81e67 | 1879 | .release = btrfs_release_file, |
39279cc3 | 1880 | .fsync = btrfs_sync_file, |
2fe17c10 | 1881 | .fallocate = btrfs_fallocate, |
34287aa3 | 1882 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 1883 | #ifdef CONFIG_COMPAT |
34287aa3 | 1884 | .compat_ioctl = btrfs_ioctl, |
39279cc3 CM |
1885 | #endif |
1886 | }; |