]>
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); |
4cb5300b | 153 | spin_unlock(&root->fs_info->defrag_inodes_lock); |
a0f98dde | 154 | return 0; |
4cb5300b CM |
155 | } |
156 | ||
157 | /* | |
158 | * must be called with the defrag_inodes lock held | |
159 | */ | |
160 | struct inode_defrag *btrfs_find_defrag_inode(struct btrfs_fs_info *info, u64 ino, | |
161 | struct rb_node **next) | |
162 | { | |
163 | struct inode_defrag *entry = NULL; | |
164 | struct rb_node *p; | |
165 | struct rb_node *parent = NULL; | |
166 | ||
167 | p = info->defrag_inodes.rb_node; | |
168 | while (p) { | |
169 | parent = p; | |
170 | entry = rb_entry(parent, struct inode_defrag, rb_node); | |
171 | ||
172 | if (ino < entry->ino) | |
173 | p = parent->rb_left; | |
174 | else if (ino > entry->ino) | |
175 | p = parent->rb_right; | |
176 | else | |
177 | return entry; | |
178 | } | |
179 | ||
180 | if (next) { | |
181 | while (parent && ino > entry->ino) { | |
182 | parent = rb_next(parent); | |
183 | entry = rb_entry(parent, struct inode_defrag, rb_node); | |
184 | } | |
185 | *next = parent; | |
186 | } | |
187 | return NULL; | |
188 | } | |
189 | ||
190 | /* | |
191 | * run through the list of inodes in the FS that need | |
192 | * defragging | |
193 | */ | |
194 | int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info) | |
195 | { | |
196 | struct inode_defrag *defrag; | |
197 | struct btrfs_root *inode_root; | |
198 | struct inode *inode; | |
199 | struct rb_node *n; | |
200 | struct btrfs_key key; | |
201 | struct btrfs_ioctl_defrag_range_args range; | |
202 | u64 first_ino = 0; | |
203 | int num_defrag; | |
204 | int defrag_batch = 1024; | |
205 | ||
206 | memset(&range, 0, sizeof(range)); | |
207 | range.len = (u64)-1; | |
208 | ||
209 | atomic_inc(&fs_info->defrag_running); | |
210 | spin_lock(&fs_info->defrag_inodes_lock); | |
211 | while(1) { | |
212 | n = NULL; | |
213 | ||
214 | /* find an inode to defrag */ | |
215 | defrag = btrfs_find_defrag_inode(fs_info, first_ino, &n); | |
216 | if (!defrag) { | |
217 | if (n) | |
218 | defrag = rb_entry(n, struct inode_defrag, rb_node); | |
219 | else if (first_ino) { | |
220 | first_ino = 0; | |
221 | continue; | |
222 | } else { | |
223 | break; | |
224 | } | |
225 | } | |
226 | ||
227 | /* remove it from the rbtree */ | |
228 | first_ino = defrag->ino + 1; | |
229 | rb_erase(&defrag->rb_node, &fs_info->defrag_inodes); | |
230 | ||
7841cb28 | 231 | if (btrfs_fs_closing(fs_info)) |
4cb5300b CM |
232 | goto next_free; |
233 | ||
234 | spin_unlock(&fs_info->defrag_inodes_lock); | |
235 | ||
236 | /* get the inode */ | |
237 | key.objectid = defrag->root; | |
238 | btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY); | |
239 | key.offset = (u64)-1; | |
240 | inode_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
241 | if (IS_ERR(inode_root)) | |
242 | goto next; | |
243 | ||
244 | key.objectid = defrag->ino; | |
245 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); | |
246 | key.offset = 0; | |
247 | ||
248 | inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL); | |
249 | if (IS_ERR(inode)) | |
250 | goto next; | |
251 | ||
252 | /* do a chunk of defrag */ | |
253 | BTRFS_I(inode)->in_defrag = 0; | |
254 | range.start = defrag->last_offset; | |
255 | num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid, | |
256 | defrag_batch); | |
257 | /* | |
258 | * if we filled the whole defrag batch, there | |
259 | * must be more work to do. Queue this defrag | |
260 | * again | |
261 | */ | |
262 | if (num_defrag == defrag_batch) { | |
263 | defrag->last_offset = range.start; | |
264 | __btrfs_add_inode_defrag(inode, defrag); | |
265 | /* | |
266 | * we don't want to kfree defrag, we added it back to | |
267 | * the rbtree | |
268 | */ | |
269 | defrag = NULL; | |
270 | } else if (defrag->last_offset && !defrag->cycled) { | |
271 | /* | |
272 | * we didn't fill our defrag batch, but | |
273 | * we didn't start at zero. Make sure we loop | |
274 | * around to the start of the file. | |
275 | */ | |
276 | defrag->last_offset = 0; | |
277 | defrag->cycled = 1; | |
278 | __btrfs_add_inode_defrag(inode, defrag); | |
279 | defrag = NULL; | |
280 | } | |
281 | ||
282 | iput(inode); | |
283 | next: | |
284 | spin_lock(&fs_info->defrag_inodes_lock); | |
285 | next_free: | |
286 | kfree(defrag); | |
287 | } | |
288 | spin_unlock(&fs_info->defrag_inodes_lock); | |
289 | ||
290 | atomic_dec(&fs_info->defrag_running); | |
291 | ||
292 | /* | |
293 | * during unmount, we use the transaction_wait queue to | |
294 | * wait for the defragger to stop | |
295 | */ | |
296 | wake_up(&fs_info->transaction_wait); | |
297 | return 0; | |
298 | } | |
39279cc3 | 299 | |
d352ac68 CM |
300 | /* simple helper to fault in pages and copy. This should go away |
301 | * and be replaced with calls into generic code. | |
302 | */ | |
d397712b | 303 | static noinline int btrfs_copy_from_user(loff_t pos, int num_pages, |
d0215f3e | 304 | size_t write_bytes, |
a1b32a59 | 305 | struct page **prepared_pages, |
11c65dcc | 306 | struct iov_iter *i) |
39279cc3 | 307 | { |
914ee295 | 308 | size_t copied = 0; |
d0215f3e | 309 | size_t total_copied = 0; |
11c65dcc | 310 | int pg = 0; |
39279cc3 CM |
311 | int offset = pos & (PAGE_CACHE_SIZE - 1); |
312 | ||
11c65dcc | 313 | while (write_bytes > 0) { |
39279cc3 CM |
314 | size_t count = min_t(size_t, |
315 | PAGE_CACHE_SIZE - offset, write_bytes); | |
11c65dcc | 316 | struct page *page = prepared_pages[pg]; |
914ee295 XZ |
317 | /* |
318 | * Copy data from userspace to the current page | |
319 | * | |
320 | * Disable pagefault to avoid recursive lock since | |
321 | * the pages are already locked | |
322 | */ | |
323 | pagefault_disable(); | |
324 | copied = iov_iter_copy_from_user_atomic(page, i, offset, count); | |
325 | pagefault_enable(); | |
11c65dcc | 326 | |
39279cc3 CM |
327 | /* Flush processor's dcache for this page */ |
328 | flush_dcache_page(page); | |
31339acd CM |
329 | |
330 | /* | |
331 | * if we get a partial write, we can end up with | |
332 | * partially up to date pages. These add | |
333 | * a lot of complexity, so make sure they don't | |
334 | * happen by forcing this copy to be retried. | |
335 | * | |
336 | * The rest of the btrfs_file_write code will fall | |
337 | * back to page at a time copies after we return 0. | |
338 | */ | |
339 | if (!PageUptodate(page) && copied < count) | |
340 | copied = 0; | |
341 | ||
11c65dcc JB |
342 | iov_iter_advance(i, copied); |
343 | write_bytes -= copied; | |
914ee295 | 344 | total_copied += copied; |
39279cc3 | 345 | |
914ee295 | 346 | /* Return to btrfs_file_aio_write to fault page */ |
9f570b8d | 347 | if (unlikely(copied == 0)) |
914ee295 | 348 | break; |
11c65dcc JB |
349 | |
350 | if (unlikely(copied < PAGE_CACHE_SIZE - offset)) { | |
351 | offset += copied; | |
352 | } else { | |
353 | pg++; | |
354 | offset = 0; | |
355 | } | |
39279cc3 | 356 | } |
914ee295 | 357 | return total_copied; |
39279cc3 CM |
358 | } |
359 | ||
d352ac68 CM |
360 | /* |
361 | * unlocks pages after btrfs_file_write is done with them | |
362 | */ | |
be1a12a0 | 363 | void btrfs_drop_pages(struct page **pages, size_t num_pages) |
39279cc3 CM |
364 | { |
365 | size_t i; | |
366 | for (i = 0; i < num_pages; i++) { | |
d352ac68 CM |
367 | /* page checked is some magic around finding pages that |
368 | * have been modified without going through btrfs_set_page_dirty | |
369 | * clear it here | |
370 | */ | |
4a096752 | 371 | ClearPageChecked(pages[i]); |
39279cc3 CM |
372 | unlock_page(pages[i]); |
373 | mark_page_accessed(pages[i]); | |
374 | page_cache_release(pages[i]); | |
375 | } | |
376 | } | |
377 | ||
d352ac68 CM |
378 | /* |
379 | * after copy_from_user, pages need to be dirtied and we need to make | |
380 | * sure holes are created between the current EOF and the start of | |
381 | * any next extents (if required). | |
382 | * | |
383 | * this also makes the decision about creating an inline extent vs | |
384 | * doing real data extents, marking pages dirty and delalloc as required. | |
385 | */ | |
be1a12a0 JB |
386 | int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode, |
387 | struct page **pages, size_t num_pages, | |
388 | loff_t pos, size_t write_bytes, | |
389 | struct extent_state **cached) | |
39279cc3 | 390 | { |
39279cc3 | 391 | int err = 0; |
a52d9a80 | 392 | int i; |
db94535d | 393 | u64 num_bytes; |
a52d9a80 CM |
394 | u64 start_pos; |
395 | u64 end_of_last_block; | |
396 | u64 end_pos = pos + write_bytes; | |
397 | loff_t isize = i_size_read(inode); | |
39279cc3 | 398 | |
5f39d397 | 399 | start_pos = pos & ~((u64)root->sectorsize - 1); |
db94535d CM |
400 | num_bytes = (write_bytes + pos - start_pos + |
401 | root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
39279cc3 | 402 | |
db94535d | 403 | end_of_last_block = start_pos + num_bytes - 1; |
2ac55d41 | 404 | err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block, |
be1a12a0 | 405 | cached); |
d0215f3e JB |
406 | if (err) |
407 | return err; | |
9ed74f2d | 408 | |
c8b97818 CM |
409 | for (i = 0; i < num_pages; i++) { |
410 | struct page *p = pages[i]; | |
411 | SetPageUptodate(p); | |
412 | ClearPageChecked(p); | |
413 | set_page_dirty(p); | |
a52d9a80 | 414 | } |
9f570b8d JB |
415 | |
416 | /* | |
417 | * we've only changed i_size in ram, and we haven't updated | |
418 | * the disk i_size. There is no need to log the inode | |
419 | * at this time. | |
420 | */ | |
421 | if (end_pos > isize) | |
a52d9a80 | 422 | i_size_write(inode, end_pos); |
a22285a6 | 423 | return 0; |
39279cc3 CM |
424 | } |
425 | ||
d352ac68 CM |
426 | /* |
427 | * this drops all the extents in the cache that intersect the range | |
428 | * [start, end]. Existing extents are split as required. | |
429 | */ | |
5b21f2ed ZY |
430 | int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end, |
431 | int skip_pinned) | |
a52d9a80 CM |
432 | { |
433 | struct extent_map *em; | |
3b951516 CM |
434 | struct extent_map *split = NULL; |
435 | struct extent_map *split2 = NULL; | |
a52d9a80 | 436 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
39b5637f | 437 | u64 len = end - start + 1; |
3b951516 CM |
438 | int ret; |
439 | int testend = 1; | |
5b21f2ed | 440 | unsigned long flags; |
c8b97818 | 441 | int compressed = 0; |
a52d9a80 | 442 | |
e6dcd2dc | 443 | WARN_ON(end < start); |
3b951516 | 444 | if (end == (u64)-1) { |
39b5637f | 445 | len = (u64)-1; |
3b951516 CM |
446 | testend = 0; |
447 | } | |
d397712b | 448 | while (1) { |
3b951516 | 449 | if (!split) |
172ddd60 | 450 | split = alloc_extent_map(); |
3b951516 | 451 | if (!split2) |
172ddd60 | 452 | split2 = alloc_extent_map(); |
c26a9203 | 453 | BUG_ON(!split || !split2); |
3b951516 | 454 | |
890871be | 455 | write_lock(&em_tree->lock); |
39b5637f | 456 | em = lookup_extent_mapping(em_tree, start, len); |
d1310b2e | 457 | if (!em) { |
890871be | 458 | write_unlock(&em_tree->lock); |
a52d9a80 | 459 | break; |
d1310b2e | 460 | } |
5b21f2ed ZY |
461 | flags = em->flags; |
462 | if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) { | |
55ef6899 | 463 | if (testend && em->start + em->len >= start + len) { |
5b21f2ed | 464 | free_extent_map(em); |
a1ed835e | 465 | write_unlock(&em_tree->lock); |
5b21f2ed ZY |
466 | break; |
467 | } | |
55ef6899 YZ |
468 | start = em->start + em->len; |
469 | if (testend) | |
5b21f2ed | 470 | len = start + len - (em->start + em->len); |
5b21f2ed | 471 | free_extent_map(em); |
a1ed835e | 472 | write_unlock(&em_tree->lock); |
5b21f2ed ZY |
473 | continue; |
474 | } | |
c8b97818 | 475 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
3ce7e67a | 476 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
a52d9a80 | 477 | remove_extent_mapping(em_tree, em); |
3b951516 CM |
478 | |
479 | if (em->block_start < EXTENT_MAP_LAST_BYTE && | |
480 | em->start < start) { | |
481 | split->start = em->start; | |
482 | split->len = start - em->start; | |
ff5b7ee3 | 483 | split->orig_start = em->orig_start; |
3b951516 | 484 | split->block_start = em->block_start; |
c8b97818 CM |
485 | |
486 | if (compressed) | |
487 | split->block_len = em->block_len; | |
488 | else | |
489 | split->block_len = split->len; | |
490 | ||
3b951516 | 491 | split->bdev = em->bdev; |
5b21f2ed | 492 | split->flags = flags; |
261507a0 | 493 | split->compress_type = em->compress_type; |
3b951516 CM |
494 | ret = add_extent_mapping(em_tree, split); |
495 | BUG_ON(ret); | |
496 | free_extent_map(split); | |
497 | split = split2; | |
498 | split2 = NULL; | |
499 | } | |
500 | if (em->block_start < EXTENT_MAP_LAST_BYTE && | |
501 | testend && em->start + em->len > start + len) { | |
502 | u64 diff = start + len - em->start; | |
503 | ||
504 | split->start = start + len; | |
505 | split->len = em->start + em->len - (start + len); | |
506 | split->bdev = em->bdev; | |
5b21f2ed | 507 | split->flags = flags; |
261507a0 | 508 | split->compress_type = em->compress_type; |
3b951516 | 509 | |
c8b97818 CM |
510 | if (compressed) { |
511 | split->block_len = em->block_len; | |
512 | split->block_start = em->block_start; | |
445a6944 | 513 | split->orig_start = em->orig_start; |
c8b97818 CM |
514 | } else { |
515 | split->block_len = split->len; | |
516 | split->block_start = em->block_start + diff; | |
445a6944 | 517 | split->orig_start = split->start; |
c8b97818 | 518 | } |
3b951516 CM |
519 | |
520 | ret = add_extent_mapping(em_tree, split); | |
521 | BUG_ON(ret); | |
522 | free_extent_map(split); | |
523 | split = NULL; | |
524 | } | |
890871be | 525 | write_unlock(&em_tree->lock); |
d1310b2e | 526 | |
a52d9a80 CM |
527 | /* once for us */ |
528 | free_extent_map(em); | |
529 | /* once for the tree*/ | |
530 | free_extent_map(em); | |
531 | } | |
3b951516 CM |
532 | if (split) |
533 | free_extent_map(split); | |
534 | if (split2) | |
535 | free_extent_map(split2); | |
a52d9a80 CM |
536 | return 0; |
537 | } | |
538 | ||
39279cc3 CM |
539 | /* |
540 | * this is very complex, but the basic idea is to drop all extents | |
541 | * in the range start - end. hint_block is filled in with a block number | |
542 | * that would be a good hint to the block allocator for this file. | |
543 | * | |
544 | * If an extent intersects the range but is not entirely inside the range | |
545 | * it is either truncated or split. Anything entirely inside the range | |
546 | * is deleted from the tree. | |
547 | */ | |
920bbbfb YZ |
548 | int btrfs_drop_extents(struct btrfs_trans_handle *trans, struct inode *inode, |
549 | u64 start, u64 end, u64 *hint_byte, int drop_cache) | |
39279cc3 | 550 | { |
920bbbfb | 551 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5f39d397 | 552 | struct extent_buffer *leaf; |
920bbbfb | 553 | struct btrfs_file_extent_item *fi; |
39279cc3 | 554 | struct btrfs_path *path; |
00f5c795 | 555 | struct btrfs_key key; |
920bbbfb | 556 | struct btrfs_key new_key; |
33345d01 | 557 | u64 ino = btrfs_ino(inode); |
920bbbfb YZ |
558 | u64 search_start = start; |
559 | u64 disk_bytenr = 0; | |
560 | u64 num_bytes = 0; | |
561 | u64 extent_offset = 0; | |
562 | u64 extent_end = 0; | |
563 | int del_nr = 0; | |
564 | int del_slot = 0; | |
565 | int extent_type; | |
ccd467d6 | 566 | int recow; |
00f5c795 | 567 | int ret; |
39279cc3 | 568 | |
a1ed835e CM |
569 | if (drop_cache) |
570 | btrfs_drop_extent_cache(inode, start, end - 1, 0); | |
a52d9a80 | 571 | |
39279cc3 CM |
572 | path = btrfs_alloc_path(); |
573 | if (!path) | |
574 | return -ENOMEM; | |
920bbbfb | 575 | |
d397712b | 576 | while (1) { |
ccd467d6 | 577 | recow = 0; |
33345d01 | 578 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
39279cc3 CM |
579 | search_start, -1); |
580 | if (ret < 0) | |
920bbbfb YZ |
581 | break; |
582 | if (ret > 0 && path->slots[0] > 0 && search_start == start) { | |
583 | leaf = path->nodes[0]; | |
584 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1); | |
33345d01 | 585 | if (key.objectid == ino && |
920bbbfb YZ |
586 | key.type == BTRFS_EXTENT_DATA_KEY) |
587 | path->slots[0]--; | |
39279cc3 | 588 | } |
920bbbfb | 589 | ret = 0; |
8c2383c3 | 590 | next_slot: |
5f39d397 | 591 | leaf = path->nodes[0]; |
920bbbfb YZ |
592 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { |
593 | BUG_ON(del_nr > 0); | |
594 | ret = btrfs_next_leaf(root, path); | |
595 | if (ret < 0) | |
596 | break; | |
597 | if (ret > 0) { | |
598 | ret = 0; | |
599 | break; | |
8c2383c3 | 600 | } |
920bbbfb YZ |
601 | leaf = path->nodes[0]; |
602 | recow = 1; | |
603 | } | |
604 | ||
605 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
33345d01 | 606 | if (key.objectid > ino || |
920bbbfb YZ |
607 | key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end) |
608 | break; | |
609 | ||
610 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
611 | struct btrfs_file_extent_item); | |
612 | extent_type = btrfs_file_extent_type(leaf, fi); | |
613 | ||
614 | if (extent_type == BTRFS_FILE_EXTENT_REG || | |
615 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
616 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
617 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
618 | extent_offset = btrfs_file_extent_offset(leaf, fi); | |
619 | extent_end = key.offset + | |
620 | btrfs_file_extent_num_bytes(leaf, fi); | |
621 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
622 | extent_end = key.offset + | |
623 | btrfs_file_extent_inline_len(leaf, fi); | |
8c2383c3 | 624 | } else { |
920bbbfb | 625 | WARN_ON(1); |
8c2383c3 | 626 | extent_end = search_start; |
39279cc3 CM |
627 | } |
628 | ||
920bbbfb YZ |
629 | if (extent_end <= search_start) { |
630 | path->slots[0]++; | |
8c2383c3 | 631 | goto next_slot; |
39279cc3 CM |
632 | } |
633 | ||
920bbbfb YZ |
634 | search_start = max(key.offset, start); |
635 | if (recow) { | |
b3b4aa74 | 636 | btrfs_release_path(path); |
920bbbfb | 637 | continue; |
39279cc3 | 638 | } |
6643558d | 639 | |
920bbbfb YZ |
640 | /* |
641 | * | - range to drop - | | |
642 | * | -------- extent -------- | | |
643 | */ | |
644 | if (start > key.offset && end < extent_end) { | |
645 | BUG_ON(del_nr > 0); | |
646 | BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE); | |
647 | ||
648 | memcpy(&new_key, &key, sizeof(new_key)); | |
649 | new_key.offset = start; | |
650 | ret = btrfs_duplicate_item(trans, root, path, | |
651 | &new_key); | |
652 | if (ret == -EAGAIN) { | |
b3b4aa74 | 653 | btrfs_release_path(path); |
920bbbfb | 654 | continue; |
6643558d | 655 | } |
920bbbfb YZ |
656 | if (ret < 0) |
657 | break; | |
658 | ||
659 | leaf = path->nodes[0]; | |
660 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
661 | struct btrfs_file_extent_item); | |
662 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
663 | start - key.offset); | |
664 | ||
665 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
666 | struct btrfs_file_extent_item); | |
667 | ||
668 | extent_offset += start - key.offset; | |
669 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); | |
670 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
671 | extent_end - start); | |
672 | btrfs_mark_buffer_dirty(leaf); | |
673 | ||
674 | if (disk_bytenr > 0) { | |
771ed689 | 675 | ret = btrfs_inc_extent_ref(trans, root, |
920bbbfb YZ |
676 | disk_bytenr, num_bytes, 0, |
677 | root->root_key.objectid, | |
678 | new_key.objectid, | |
679 | start - extent_offset); | |
771ed689 | 680 | BUG_ON(ret); |
920bbbfb | 681 | *hint_byte = disk_bytenr; |
771ed689 | 682 | } |
920bbbfb | 683 | key.offset = start; |
6643558d | 684 | } |
920bbbfb YZ |
685 | /* |
686 | * | ---- range to drop ----- | | |
687 | * | -------- extent -------- | | |
688 | */ | |
689 | if (start <= key.offset && end < extent_end) { | |
690 | BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE); | |
6643558d | 691 | |
920bbbfb YZ |
692 | memcpy(&new_key, &key, sizeof(new_key)); |
693 | new_key.offset = end; | |
694 | btrfs_set_item_key_safe(trans, root, path, &new_key); | |
6643558d | 695 | |
920bbbfb YZ |
696 | extent_offset += end - key.offset; |
697 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); | |
698 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
699 | extent_end - end); | |
700 | btrfs_mark_buffer_dirty(leaf); | |
701 | if (disk_bytenr > 0) { | |
702 | inode_sub_bytes(inode, end - key.offset); | |
703 | *hint_byte = disk_bytenr; | |
39279cc3 | 704 | } |
920bbbfb | 705 | break; |
39279cc3 | 706 | } |
771ed689 | 707 | |
920bbbfb YZ |
708 | search_start = extent_end; |
709 | /* | |
710 | * | ---- range to drop ----- | | |
711 | * | -------- extent -------- | | |
712 | */ | |
713 | if (start > key.offset && end >= extent_end) { | |
714 | BUG_ON(del_nr > 0); | |
715 | BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE); | |
8c2383c3 | 716 | |
920bbbfb YZ |
717 | btrfs_set_file_extent_num_bytes(leaf, fi, |
718 | start - key.offset); | |
719 | btrfs_mark_buffer_dirty(leaf); | |
720 | if (disk_bytenr > 0) { | |
721 | inode_sub_bytes(inode, extent_end - start); | |
722 | *hint_byte = disk_bytenr; | |
723 | } | |
724 | if (end == extent_end) | |
725 | break; | |
c8b97818 | 726 | |
920bbbfb YZ |
727 | path->slots[0]++; |
728 | goto next_slot; | |
31840ae1 ZY |
729 | } |
730 | ||
920bbbfb YZ |
731 | /* |
732 | * | ---- range to drop ----- | | |
733 | * | ------ extent ------ | | |
734 | */ | |
735 | if (start <= key.offset && end >= extent_end) { | |
736 | if (del_nr == 0) { | |
737 | del_slot = path->slots[0]; | |
738 | del_nr = 1; | |
739 | } else { | |
740 | BUG_ON(del_slot + del_nr != path->slots[0]); | |
741 | del_nr++; | |
742 | } | |
31840ae1 | 743 | |
920bbbfb | 744 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
a76a3cd4 | 745 | inode_sub_bytes(inode, |
920bbbfb YZ |
746 | extent_end - key.offset); |
747 | extent_end = ALIGN(extent_end, | |
748 | root->sectorsize); | |
749 | } else if (disk_bytenr > 0) { | |
31840ae1 | 750 | ret = btrfs_free_extent(trans, root, |
920bbbfb YZ |
751 | disk_bytenr, num_bytes, 0, |
752 | root->root_key.objectid, | |
5d4f98a2 | 753 | key.objectid, key.offset - |
920bbbfb | 754 | extent_offset); |
31840ae1 | 755 | BUG_ON(ret); |
920bbbfb YZ |
756 | inode_sub_bytes(inode, |
757 | extent_end - key.offset); | |
758 | *hint_byte = disk_bytenr; | |
31840ae1 | 759 | } |
31840ae1 | 760 | |
920bbbfb YZ |
761 | if (end == extent_end) |
762 | break; | |
763 | ||
764 | if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) { | |
765 | path->slots[0]++; | |
766 | goto next_slot; | |
767 | } | |
768 | ||
769 | ret = btrfs_del_items(trans, root, path, del_slot, | |
770 | del_nr); | |
771 | BUG_ON(ret); | |
772 | ||
773 | del_nr = 0; | |
774 | del_slot = 0; | |
775 | ||
b3b4aa74 | 776 | btrfs_release_path(path); |
920bbbfb | 777 | continue; |
39279cc3 | 778 | } |
920bbbfb YZ |
779 | |
780 | BUG_ON(1); | |
39279cc3 | 781 | } |
920bbbfb YZ |
782 | |
783 | if (del_nr > 0) { | |
784 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); | |
785 | BUG_ON(ret); | |
6643558d | 786 | } |
920bbbfb YZ |
787 | |
788 | btrfs_free_path(path); | |
39279cc3 CM |
789 | return ret; |
790 | } | |
791 | ||
d899e052 | 792 | static int extent_mergeable(struct extent_buffer *leaf, int slot, |
6c7d54ac YZ |
793 | u64 objectid, u64 bytenr, u64 orig_offset, |
794 | u64 *start, u64 *end) | |
d899e052 YZ |
795 | { |
796 | struct btrfs_file_extent_item *fi; | |
797 | struct btrfs_key key; | |
798 | u64 extent_end; | |
799 | ||
800 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) | |
801 | return 0; | |
802 | ||
803 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
804 | if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY) | |
805 | return 0; | |
806 | ||
807 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
808 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG || | |
809 | btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr || | |
6c7d54ac | 810 | btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset || |
d899e052 YZ |
811 | btrfs_file_extent_compression(leaf, fi) || |
812 | btrfs_file_extent_encryption(leaf, fi) || | |
813 | btrfs_file_extent_other_encoding(leaf, fi)) | |
814 | return 0; | |
815 | ||
816 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
817 | if ((*start && *start != key.offset) || (*end && *end != extent_end)) | |
818 | return 0; | |
819 | ||
820 | *start = key.offset; | |
821 | *end = extent_end; | |
822 | return 1; | |
823 | } | |
824 | ||
825 | /* | |
826 | * Mark extent in the range start - end as written. | |
827 | * | |
828 | * This changes extent type from 'pre-allocated' to 'regular'. If only | |
829 | * part of extent is marked as written, the extent will be split into | |
830 | * two or three. | |
831 | */ | |
832 | int btrfs_mark_extent_written(struct btrfs_trans_handle *trans, | |
d899e052 YZ |
833 | struct inode *inode, u64 start, u64 end) |
834 | { | |
920bbbfb | 835 | struct btrfs_root *root = BTRFS_I(inode)->root; |
d899e052 YZ |
836 | struct extent_buffer *leaf; |
837 | struct btrfs_path *path; | |
838 | struct btrfs_file_extent_item *fi; | |
839 | struct btrfs_key key; | |
920bbbfb | 840 | struct btrfs_key new_key; |
d899e052 YZ |
841 | u64 bytenr; |
842 | u64 num_bytes; | |
843 | u64 extent_end; | |
5d4f98a2 | 844 | u64 orig_offset; |
d899e052 YZ |
845 | u64 other_start; |
846 | u64 other_end; | |
920bbbfb YZ |
847 | u64 split; |
848 | int del_nr = 0; | |
849 | int del_slot = 0; | |
6c7d54ac | 850 | int recow; |
d899e052 | 851 | int ret; |
33345d01 | 852 | u64 ino = btrfs_ino(inode); |
d899e052 YZ |
853 | |
854 | btrfs_drop_extent_cache(inode, start, end - 1, 0); | |
855 | ||
856 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
857 | if (!path) |
858 | return -ENOMEM; | |
d899e052 | 859 | again: |
6c7d54ac | 860 | recow = 0; |
920bbbfb | 861 | split = start; |
33345d01 | 862 | key.objectid = ino; |
d899e052 | 863 | key.type = BTRFS_EXTENT_DATA_KEY; |
920bbbfb | 864 | key.offset = split; |
d899e052 YZ |
865 | |
866 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
41415730 JB |
867 | if (ret < 0) |
868 | goto out; | |
d899e052 YZ |
869 | if (ret > 0 && path->slots[0] > 0) |
870 | path->slots[0]--; | |
871 | ||
872 | leaf = path->nodes[0]; | |
873 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
33345d01 | 874 | BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY); |
d899e052 YZ |
875 | fi = btrfs_item_ptr(leaf, path->slots[0], |
876 | struct btrfs_file_extent_item); | |
920bbbfb YZ |
877 | BUG_ON(btrfs_file_extent_type(leaf, fi) != |
878 | BTRFS_FILE_EXTENT_PREALLOC); | |
d899e052 YZ |
879 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
880 | BUG_ON(key.offset > start || extent_end < end); | |
881 | ||
882 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
883 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
5d4f98a2 | 884 | orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi); |
6c7d54ac YZ |
885 | memcpy(&new_key, &key, sizeof(new_key)); |
886 | ||
887 | if (start == key.offset && end < extent_end) { | |
888 | other_start = 0; | |
889 | other_end = start; | |
890 | if (extent_mergeable(leaf, path->slots[0] - 1, | |
33345d01 | 891 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
892 | &other_start, &other_end)) { |
893 | new_key.offset = end; | |
894 | btrfs_set_item_key_safe(trans, root, path, &new_key); | |
895 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
896 | struct btrfs_file_extent_item); | |
897 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
898 | extent_end - end); | |
899 | btrfs_set_file_extent_offset(leaf, fi, | |
900 | end - orig_offset); | |
901 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
902 | struct btrfs_file_extent_item); | |
903 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
904 | end - other_start); | |
905 | btrfs_mark_buffer_dirty(leaf); | |
906 | goto out; | |
907 | } | |
908 | } | |
909 | ||
910 | if (start > key.offset && end == extent_end) { | |
911 | other_start = end; | |
912 | other_end = 0; | |
913 | if (extent_mergeable(leaf, path->slots[0] + 1, | |
33345d01 | 914 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
915 | &other_start, &other_end)) { |
916 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
917 | struct btrfs_file_extent_item); | |
918 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
919 | start - key.offset); | |
920 | path->slots[0]++; | |
921 | new_key.offset = start; | |
922 | btrfs_set_item_key_safe(trans, root, path, &new_key); | |
923 | ||
924 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
925 | struct btrfs_file_extent_item); | |
926 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
927 | other_end - start); | |
928 | btrfs_set_file_extent_offset(leaf, fi, | |
929 | start - orig_offset); | |
930 | btrfs_mark_buffer_dirty(leaf); | |
931 | goto out; | |
932 | } | |
933 | } | |
d899e052 | 934 | |
920bbbfb YZ |
935 | while (start > key.offset || end < extent_end) { |
936 | if (key.offset == start) | |
937 | split = end; | |
938 | ||
920bbbfb YZ |
939 | new_key.offset = split; |
940 | ret = btrfs_duplicate_item(trans, root, path, &new_key); | |
941 | if (ret == -EAGAIN) { | |
b3b4aa74 | 942 | btrfs_release_path(path); |
920bbbfb | 943 | goto again; |
d899e052 | 944 | } |
920bbbfb | 945 | BUG_ON(ret < 0); |
d899e052 | 946 | |
920bbbfb YZ |
947 | leaf = path->nodes[0]; |
948 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
d899e052 | 949 | struct btrfs_file_extent_item); |
d899e052 | 950 | btrfs_set_file_extent_num_bytes(leaf, fi, |
920bbbfb YZ |
951 | split - key.offset); |
952 | ||
953 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
954 | struct btrfs_file_extent_item); | |
955 | ||
956 | btrfs_set_file_extent_offset(leaf, fi, split - orig_offset); | |
957 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
958 | extent_end - split); | |
d899e052 YZ |
959 | btrfs_mark_buffer_dirty(leaf); |
960 | ||
920bbbfb YZ |
961 | ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0, |
962 | root->root_key.objectid, | |
33345d01 | 963 | ino, orig_offset); |
d899e052 | 964 | BUG_ON(ret); |
d899e052 | 965 | |
920bbbfb YZ |
966 | if (split == start) { |
967 | key.offset = start; | |
968 | } else { | |
969 | BUG_ON(start != key.offset); | |
d899e052 | 970 | path->slots[0]--; |
920bbbfb | 971 | extent_end = end; |
d899e052 | 972 | } |
6c7d54ac | 973 | recow = 1; |
d899e052 YZ |
974 | } |
975 | ||
920bbbfb YZ |
976 | other_start = end; |
977 | other_end = 0; | |
6c7d54ac | 978 | if (extent_mergeable(leaf, path->slots[0] + 1, |
33345d01 | 979 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
980 | &other_start, &other_end)) { |
981 | if (recow) { | |
b3b4aa74 | 982 | btrfs_release_path(path); |
6c7d54ac YZ |
983 | goto again; |
984 | } | |
920bbbfb YZ |
985 | extent_end = other_end; |
986 | del_slot = path->slots[0] + 1; | |
987 | del_nr++; | |
988 | ret = btrfs_free_extent(trans, root, bytenr, num_bytes, | |
989 | 0, root->root_key.objectid, | |
33345d01 | 990 | ino, orig_offset); |
920bbbfb | 991 | BUG_ON(ret); |
d899e052 | 992 | } |
920bbbfb YZ |
993 | other_start = 0; |
994 | other_end = start; | |
6c7d54ac | 995 | if (extent_mergeable(leaf, path->slots[0] - 1, |
33345d01 | 996 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
997 | &other_start, &other_end)) { |
998 | if (recow) { | |
b3b4aa74 | 999 | btrfs_release_path(path); |
6c7d54ac YZ |
1000 | goto again; |
1001 | } | |
920bbbfb YZ |
1002 | key.offset = other_start; |
1003 | del_slot = path->slots[0]; | |
1004 | del_nr++; | |
1005 | ret = btrfs_free_extent(trans, root, bytenr, num_bytes, | |
1006 | 0, root->root_key.objectid, | |
33345d01 | 1007 | ino, orig_offset); |
920bbbfb YZ |
1008 | BUG_ON(ret); |
1009 | } | |
1010 | if (del_nr == 0) { | |
3f6fae95 SL |
1011 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1012 | struct btrfs_file_extent_item); | |
920bbbfb YZ |
1013 | btrfs_set_file_extent_type(leaf, fi, |
1014 | BTRFS_FILE_EXTENT_REG); | |
1015 | btrfs_mark_buffer_dirty(leaf); | |
6c7d54ac | 1016 | } else { |
3f6fae95 SL |
1017 | fi = btrfs_item_ptr(leaf, del_slot - 1, |
1018 | struct btrfs_file_extent_item); | |
6c7d54ac YZ |
1019 | btrfs_set_file_extent_type(leaf, fi, |
1020 | BTRFS_FILE_EXTENT_REG); | |
1021 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
1022 | extent_end - key.offset); | |
1023 | btrfs_mark_buffer_dirty(leaf); | |
920bbbfb | 1024 | |
6c7d54ac YZ |
1025 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
1026 | BUG_ON(ret); | |
1027 | } | |
920bbbfb | 1028 | out: |
d899e052 YZ |
1029 | btrfs_free_path(path); |
1030 | return 0; | |
1031 | } | |
1032 | ||
b1bf862e CM |
1033 | /* |
1034 | * on error we return an unlocked page and the error value | |
1035 | * on success we return a locked page and 0 | |
1036 | */ | |
1037 | static int prepare_uptodate_page(struct page *page, u64 pos) | |
1038 | { | |
1039 | int ret = 0; | |
1040 | ||
1041 | if ((pos & (PAGE_CACHE_SIZE - 1)) && !PageUptodate(page)) { | |
1042 | ret = btrfs_readpage(NULL, page); | |
1043 | if (ret) | |
1044 | return ret; | |
1045 | lock_page(page); | |
1046 | if (!PageUptodate(page)) { | |
1047 | unlock_page(page); | |
1048 | return -EIO; | |
1049 | } | |
1050 | } | |
1051 | return 0; | |
1052 | } | |
1053 | ||
39279cc3 | 1054 | /* |
d352ac68 CM |
1055 | * this gets pages into the page cache and locks them down, it also properly |
1056 | * waits for data=ordered extents to finish before allowing the pages to be | |
1057 | * modified. | |
39279cc3 | 1058 | */ |
d397712b | 1059 | static noinline int prepare_pages(struct btrfs_root *root, struct file *file, |
98ed5174 CM |
1060 | struct page **pages, size_t num_pages, |
1061 | loff_t pos, unsigned long first_index, | |
1062 | unsigned long last_index, size_t write_bytes) | |
39279cc3 | 1063 | { |
2ac55d41 | 1064 | struct extent_state *cached_state = NULL; |
39279cc3 CM |
1065 | int i; |
1066 | unsigned long index = pos >> PAGE_CACHE_SHIFT; | |
6da6abae | 1067 | struct inode *inode = fdentry(file)->d_inode; |
39279cc3 | 1068 | int err = 0; |
b1bf862e | 1069 | int faili = 0; |
8c2383c3 | 1070 | u64 start_pos; |
e6dcd2dc | 1071 | u64 last_pos; |
8c2383c3 | 1072 | |
5f39d397 | 1073 | start_pos = pos & ~((u64)root->sectorsize - 1); |
e6dcd2dc | 1074 | last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT; |
39279cc3 | 1075 | |
9036c102 | 1076 | if (start_pos > inode->i_size) { |
a41ad394 | 1077 | err = btrfs_cont_expand(inode, i_size_read(inode), start_pos); |
9036c102 YZ |
1078 | if (err) |
1079 | return err; | |
1080 | } | |
1081 | ||
e6dcd2dc | 1082 | again: |
39279cc3 | 1083 | for (i = 0; i < num_pages; i++) { |
a94733d0 JB |
1084 | pages[i] = find_or_create_page(inode->i_mapping, index + i, |
1085 | GFP_NOFS); | |
39279cc3 | 1086 | if (!pages[i]) { |
b1bf862e CM |
1087 | faili = i - 1; |
1088 | err = -ENOMEM; | |
1089 | goto fail; | |
1090 | } | |
1091 | ||
1092 | if (i == 0) | |
1093 | err = prepare_uptodate_page(pages[i], pos); | |
1094 | if (i == num_pages - 1) | |
1095 | err = prepare_uptodate_page(pages[i], | |
1096 | pos + write_bytes); | |
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 CM |
1161 | unsigned long first_index; |
1162 | unsigned long last_index; | |
d0215f3e JB |
1163 | size_t num_written = 0; |
1164 | int nrptrs; | |
c9149235 | 1165 | int ret = 0; |
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; |
d0215f3e | 1175 | last_index = (pos + iov_iter_count(i)) >> PAGE_CACHE_SHIFT; |
39279cc3 | 1176 | |
d0215f3e | 1177 | while (iov_iter_count(i) > 0) { |
39279cc3 | 1178 | size_t offset = pos & (PAGE_CACHE_SIZE - 1); |
d0215f3e | 1179 | size_t write_bytes = min(iov_iter_count(i), |
11c65dcc | 1180 | nrptrs * (size_t)PAGE_CACHE_SIZE - |
8c2383c3 | 1181 | offset); |
3a90983d YZ |
1182 | size_t num_pages = (write_bytes + offset + |
1183 | PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
d0215f3e JB |
1184 | size_t dirty_pages; |
1185 | size_t copied; | |
39279cc3 | 1186 | |
8c2383c3 | 1187 | WARN_ON(num_pages > nrptrs); |
1832a6d5 | 1188 | |
914ee295 XZ |
1189 | /* |
1190 | * Fault pages before locking them in prepare_pages | |
1191 | * to avoid recursive lock | |
1192 | */ | |
d0215f3e | 1193 | if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) { |
914ee295 | 1194 | ret = -EFAULT; |
d0215f3e | 1195 | break; |
914ee295 XZ |
1196 | } |
1197 | ||
1198 | ret = btrfs_delalloc_reserve_space(inode, | |
1199 | num_pages << PAGE_CACHE_SHIFT); | |
1832a6d5 | 1200 | if (ret) |
d0215f3e | 1201 | break; |
1832a6d5 | 1202 | |
4a64001f JB |
1203 | /* |
1204 | * This is going to setup the pages array with the number of | |
1205 | * pages we want, so we don't really need to worry about the | |
1206 | * contents of pages from loop to loop | |
1207 | */ | |
39279cc3 CM |
1208 | ret = prepare_pages(root, file, pages, num_pages, |
1209 | pos, first_index, last_index, | |
8c2383c3 | 1210 | write_bytes); |
6a63209f | 1211 | if (ret) { |
914ee295 XZ |
1212 | btrfs_delalloc_release_space(inode, |
1213 | num_pages << PAGE_CACHE_SHIFT); | |
d0215f3e | 1214 | break; |
6a63209f | 1215 | } |
39279cc3 | 1216 | |
914ee295 | 1217 | copied = btrfs_copy_from_user(pos, num_pages, |
d0215f3e | 1218 | write_bytes, pages, i); |
b1bf862e CM |
1219 | |
1220 | /* | |
1221 | * if we have trouble faulting in the pages, fall | |
1222 | * back to one page at a time | |
1223 | */ | |
1224 | if (copied < write_bytes) | |
1225 | nrptrs = 1; | |
1226 | ||
1227 | if (copied == 0) | |
1228 | dirty_pages = 0; | |
1229 | else | |
1230 | dirty_pages = (copied + offset + | |
1231 | PAGE_CACHE_SIZE - 1) >> | |
1232 | PAGE_CACHE_SHIFT; | |
914ee295 | 1233 | |
d0215f3e JB |
1234 | /* |
1235 | * If we had a short copy we need to release the excess delaloc | |
1236 | * bytes we reserved. We need to increment outstanding_extents | |
1237 | * because btrfs_delalloc_release_space will decrement it, but | |
1238 | * we still have an outstanding extent for the chunk we actually | |
1239 | * managed to copy. | |
1240 | */ | |
914ee295 | 1241 | if (num_pages > dirty_pages) { |
9e0baf60 JB |
1242 | if (copied > 0) { |
1243 | spin_lock(&BTRFS_I(inode)->lock); | |
1244 | BTRFS_I(inode)->outstanding_extents++; | |
1245 | spin_unlock(&BTRFS_I(inode)->lock); | |
1246 | } | |
914ee295 XZ |
1247 | btrfs_delalloc_release_space(inode, |
1248 | (num_pages - dirty_pages) << | |
1249 | PAGE_CACHE_SHIFT); | |
1250 | } | |
1251 | ||
1252 | if (copied > 0) { | |
be1a12a0 JB |
1253 | ret = btrfs_dirty_pages(root, inode, pages, |
1254 | dirty_pages, pos, copied, | |
1255 | NULL); | |
d0215f3e JB |
1256 | if (ret) { |
1257 | btrfs_delalloc_release_space(inode, | |
1258 | dirty_pages << PAGE_CACHE_SHIFT); | |
1259 | btrfs_drop_pages(pages, num_pages); | |
1260 | break; | |
1261 | } | |
54aa1f4d | 1262 | } |
39279cc3 | 1263 | |
39279cc3 CM |
1264 | btrfs_drop_pages(pages, num_pages); |
1265 | ||
d0215f3e JB |
1266 | cond_resched(); |
1267 | ||
1268 | balance_dirty_pages_ratelimited_nr(inode->i_mapping, | |
1269 | dirty_pages); | |
1270 | if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1) | |
1271 | btrfs_btree_balance_dirty(root, 1); | |
1272 | btrfs_throttle(root); | |
cb843a6f | 1273 | |
914ee295 XZ |
1274 | pos += copied; |
1275 | num_written += copied; | |
d0215f3e | 1276 | } |
39279cc3 | 1277 | |
d0215f3e JB |
1278 | kfree(pages); |
1279 | ||
1280 | return num_written ? num_written : ret; | |
1281 | } | |
1282 | ||
1283 | static ssize_t __btrfs_direct_write(struct kiocb *iocb, | |
1284 | const struct iovec *iov, | |
1285 | unsigned long nr_segs, loff_t pos, | |
1286 | loff_t *ppos, size_t count, size_t ocount) | |
1287 | { | |
1288 | struct file *file = iocb->ki_filp; | |
1289 | struct inode *inode = fdentry(file)->d_inode; | |
1290 | struct iov_iter i; | |
1291 | ssize_t written; | |
1292 | ssize_t written_buffered; | |
1293 | loff_t endbyte; | |
1294 | int err; | |
1295 | ||
1296 | written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos, | |
1297 | count, ocount); | |
1298 | ||
1299 | /* | |
1300 | * the generic O_DIRECT will update in-memory i_size after the | |
1301 | * DIOs are done. But our endio handlers that update the on | |
1302 | * disk i_size never update past the in memory i_size. So we | |
1303 | * need one more update here to catch any additions to the | |
1304 | * file | |
1305 | */ | |
1306 | if (inode->i_size != BTRFS_I(inode)->disk_i_size) { | |
1307 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); | |
1308 | mark_inode_dirty(inode); | |
1309 | } | |
1310 | ||
1311 | if (written < 0 || written == count) | |
1312 | return written; | |
1313 | ||
1314 | pos += written; | |
1315 | count -= written; | |
1316 | iov_iter_init(&i, iov, nr_segs, count, written); | |
1317 | written_buffered = __btrfs_buffered_write(file, &i, pos); | |
1318 | if (written_buffered < 0) { | |
1319 | err = written_buffered; | |
1320 | goto out; | |
39279cc3 | 1321 | } |
d0215f3e JB |
1322 | endbyte = pos + written_buffered - 1; |
1323 | err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte); | |
1324 | if (err) | |
1325 | goto out; | |
1326 | written += written_buffered; | |
1327 | *ppos = pos + written_buffered; | |
1328 | invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT, | |
1329 | endbyte >> PAGE_CACHE_SHIFT); | |
39279cc3 | 1330 | out: |
d0215f3e JB |
1331 | return written ? written : err; |
1332 | } | |
5b92ee72 | 1333 | |
d0215f3e JB |
1334 | static ssize_t btrfs_file_aio_write(struct kiocb *iocb, |
1335 | const struct iovec *iov, | |
1336 | unsigned long nr_segs, loff_t pos) | |
1337 | { | |
1338 | struct file *file = iocb->ki_filp; | |
1339 | struct inode *inode = fdentry(file)->d_inode; | |
1340 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1341 | loff_t *ppos = &iocb->ki_pos; | |
1342 | ssize_t num_written = 0; | |
1343 | ssize_t err = 0; | |
1344 | size_t count, ocount; | |
1345 | ||
1346 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); | |
1347 | ||
1348 | mutex_lock(&inode->i_mutex); | |
1349 | ||
1350 | err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ); | |
1351 | if (err) { | |
1352 | mutex_unlock(&inode->i_mutex); | |
1353 | goto out; | |
1354 | } | |
1355 | count = ocount; | |
1356 | ||
1357 | current->backing_dev_info = inode->i_mapping->backing_dev_info; | |
1358 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); | |
1359 | if (err) { | |
1360 | mutex_unlock(&inode->i_mutex); | |
1361 | goto out; | |
1362 | } | |
1363 | ||
1364 | if (count == 0) { | |
1365 | mutex_unlock(&inode->i_mutex); | |
1366 | goto out; | |
1367 | } | |
1368 | ||
1369 | err = file_remove_suid(file); | |
1370 | if (err) { | |
1371 | mutex_unlock(&inode->i_mutex); | |
1372 | goto out; | |
1373 | } | |
1374 | ||
1375 | /* | |
1376 | * If BTRFS flips readonly due to some impossible error | |
1377 | * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR), | |
1378 | * although we have opened a file as writable, we have | |
1379 | * to stop this write operation to ensure FS consistency. | |
1380 | */ | |
1381 | if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) { | |
1382 | mutex_unlock(&inode->i_mutex); | |
1383 | err = -EROFS; | |
1384 | goto out; | |
1385 | } | |
1386 | ||
1387 | file_update_time(file); | |
1388 | BTRFS_I(inode)->sequence++; | |
1389 | ||
1390 | if (unlikely(file->f_flags & O_DIRECT)) { | |
1391 | num_written = __btrfs_direct_write(iocb, iov, nr_segs, | |
1392 | pos, ppos, count, ocount); | |
1393 | } else { | |
1394 | struct iov_iter i; | |
1395 | ||
1396 | iov_iter_init(&i, iov, nr_segs, count, num_written); | |
1397 | ||
1398 | num_written = __btrfs_buffered_write(file, &i, pos); | |
1399 | if (num_written > 0) | |
1400 | *ppos = pos + num_written; | |
1401 | } | |
1402 | ||
1403 | mutex_unlock(&inode->i_mutex); | |
2ff3e9b6 | 1404 | |
5a3f23d5 CM |
1405 | /* |
1406 | * we want to make sure fsync finds this change | |
1407 | * but we haven't joined a transaction running right now. | |
1408 | * | |
1409 | * Later on, someone is sure to update the inode and get the | |
1410 | * real transid recorded. | |
1411 | * | |
1412 | * We set last_trans now to the fs_info generation + 1, | |
1413 | * this will either be one more than the running transaction | |
1414 | * or the generation used for the next transaction if there isn't | |
1415 | * one running right now. | |
1416 | */ | |
1417 | BTRFS_I(inode)->last_trans = root->fs_info->generation + 1; | |
d0215f3e JB |
1418 | if (num_written > 0 || num_written == -EIOCBQUEUED) { |
1419 | err = generic_write_sync(file, pos, num_written); | |
1420 | if (err < 0 && num_written > 0) | |
2ff3e9b6 CM |
1421 | num_written = err; |
1422 | } | |
d0215f3e | 1423 | out: |
39279cc3 | 1424 | current->backing_dev_info = NULL; |
39279cc3 CM |
1425 | return num_written ? num_written : err; |
1426 | } | |
1427 | ||
d397712b | 1428 | int btrfs_release_file(struct inode *inode, struct file *filp) |
e1b81e67 | 1429 | { |
5a3f23d5 CM |
1430 | /* |
1431 | * ordered_data_close is set by settattr when we are about to truncate | |
1432 | * a file from a non-zero size to a zero size. This tries to | |
1433 | * flush down new bytes that may have been written if the | |
1434 | * application were using truncate to replace a file in place. | |
1435 | */ | |
1436 | if (BTRFS_I(inode)->ordered_data_close) { | |
1437 | BTRFS_I(inode)->ordered_data_close = 0; | |
1438 | btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode); | |
1439 | if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) | |
1440 | filemap_flush(inode->i_mapping); | |
1441 | } | |
6bf13c0c SW |
1442 | if (filp->private_data) |
1443 | btrfs_ioctl_trans_end(filp); | |
e1b81e67 M |
1444 | return 0; |
1445 | } | |
1446 | ||
d352ac68 CM |
1447 | /* |
1448 | * fsync call for both files and directories. This logs the inode into | |
1449 | * the tree log instead of forcing full commits whenever possible. | |
1450 | * | |
1451 | * It needs to call filemap_fdatawait so that all ordered extent updates are | |
1452 | * in the metadata btree are up to date for copying to the log. | |
1453 | * | |
1454 | * It drops the inode mutex before doing the tree log commit. This is an | |
1455 | * important optimization for directories because holding the mutex prevents | |
1456 | * new operations on the dir while we write to disk. | |
1457 | */ | |
7ea80859 | 1458 | int btrfs_sync_file(struct file *file, int datasync) |
39279cc3 | 1459 | { |
7ea80859 | 1460 | struct dentry *dentry = file->f_path.dentry; |
39279cc3 CM |
1461 | struct inode *inode = dentry->d_inode; |
1462 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
15ee9bc7 | 1463 | int ret = 0; |
39279cc3 CM |
1464 | struct btrfs_trans_handle *trans; |
1465 | ||
1abe9b8a | 1466 | trace_btrfs_sync_file(file, datasync); |
257c62e1 CM |
1467 | |
1468 | /* we wait first, since the writeback may change the inode */ | |
1469 | root->log_batch++; | |
1470 | /* the VFS called filemap_fdatawrite for us */ | |
1471 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
1472 | root->log_batch++; | |
1473 | ||
39279cc3 | 1474 | /* |
15ee9bc7 JB |
1475 | * check the transaction that last modified this inode |
1476 | * and see if its already been committed | |
39279cc3 | 1477 | */ |
15ee9bc7 JB |
1478 | if (!BTRFS_I(inode)->last_trans) |
1479 | goto out; | |
a2135011 | 1480 | |
257c62e1 CM |
1481 | /* |
1482 | * if the last transaction that changed this file was before | |
1483 | * the current transaction, we can bail out now without any | |
1484 | * syncing | |
1485 | */ | |
a4abeea4 | 1486 | smp_mb(); |
15ee9bc7 JB |
1487 | if (BTRFS_I(inode)->last_trans <= |
1488 | root->fs_info->last_trans_committed) { | |
1489 | BTRFS_I(inode)->last_trans = 0; | |
15ee9bc7 JB |
1490 | goto out; |
1491 | } | |
15ee9bc7 JB |
1492 | |
1493 | /* | |
a52d9a80 CM |
1494 | * ok we haven't committed the transaction yet, lets do a commit |
1495 | */ | |
6f902af4 | 1496 | if (file->private_data) |
6bf13c0c SW |
1497 | btrfs_ioctl_trans_end(file); |
1498 | ||
a22285a6 YZ |
1499 | trans = btrfs_start_transaction(root, 0); |
1500 | if (IS_ERR(trans)) { | |
1501 | ret = PTR_ERR(trans); | |
39279cc3 CM |
1502 | goto out; |
1503 | } | |
e02119d5 | 1504 | |
2cfbd50b | 1505 | ret = btrfs_log_dentry_safe(trans, root, dentry); |
d397712b | 1506 | if (ret < 0) |
e02119d5 | 1507 | goto out; |
49eb7e46 CM |
1508 | |
1509 | /* we've logged all the items and now have a consistent | |
1510 | * version of the file in the log. It is possible that | |
1511 | * someone will come in and modify the file, but that's | |
1512 | * fine because the log is consistent on disk, and we | |
1513 | * have references to all of the file's extents | |
1514 | * | |
1515 | * It is possible that someone will come in and log the | |
1516 | * file again, but that will end up using the synchronization | |
1517 | * inside btrfs_sync_log to keep things safe. | |
1518 | */ | |
2cfbd50b | 1519 | mutex_unlock(&dentry->d_inode->i_mutex); |
49eb7e46 | 1520 | |
257c62e1 CM |
1521 | if (ret != BTRFS_NO_LOG_SYNC) { |
1522 | if (ret > 0) { | |
12fcfd22 | 1523 | ret = btrfs_commit_transaction(trans, root); |
257c62e1 CM |
1524 | } else { |
1525 | ret = btrfs_sync_log(trans, root); | |
1526 | if (ret == 0) | |
1527 | ret = btrfs_end_transaction(trans, root); | |
1528 | else | |
1529 | ret = btrfs_commit_transaction(trans, root); | |
1530 | } | |
1531 | } else { | |
1532 | ret = btrfs_end_transaction(trans, root); | |
e02119d5 | 1533 | } |
2cfbd50b | 1534 | mutex_lock(&dentry->d_inode->i_mutex); |
39279cc3 | 1535 | out: |
014e4ac4 | 1536 | return ret > 0 ? -EIO : ret; |
39279cc3 CM |
1537 | } |
1538 | ||
f0f37e2f | 1539 | static const struct vm_operations_struct btrfs_file_vm_ops = { |
92fee66d | 1540 | .fault = filemap_fault, |
9ebefb18 CM |
1541 | .page_mkwrite = btrfs_page_mkwrite, |
1542 | }; | |
1543 | ||
1544 | static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma) | |
1545 | { | |
058a457e MX |
1546 | struct address_space *mapping = filp->f_mapping; |
1547 | ||
1548 | if (!mapping->a_ops->readpage) | |
1549 | return -ENOEXEC; | |
1550 | ||
9ebefb18 | 1551 | file_accessed(filp); |
058a457e MX |
1552 | vma->vm_ops = &btrfs_file_vm_ops; |
1553 | vma->vm_flags |= VM_CAN_NONLINEAR; | |
1554 | ||
9ebefb18 CM |
1555 | return 0; |
1556 | } | |
1557 | ||
2fe17c10 CH |
1558 | static long btrfs_fallocate(struct file *file, int mode, |
1559 | loff_t offset, loff_t len) | |
1560 | { | |
1561 | struct inode *inode = file->f_path.dentry->d_inode; | |
1562 | struct extent_state *cached_state = NULL; | |
1563 | u64 cur_offset; | |
1564 | u64 last_byte; | |
1565 | u64 alloc_start; | |
1566 | u64 alloc_end; | |
1567 | u64 alloc_hint = 0; | |
1568 | u64 locked_end; | |
1569 | u64 mask = BTRFS_I(inode)->root->sectorsize - 1; | |
1570 | struct extent_map *em; | |
1571 | int ret; | |
1572 | ||
1573 | alloc_start = offset & ~mask; | |
1574 | alloc_end = (offset + len + mask) & ~mask; | |
1575 | ||
1576 | /* We only support the FALLOC_FL_KEEP_SIZE mode */ | |
1577 | if (mode & ~FALLOC_FL_KEEP_SIZE) | |
1578 | return -EOPNOTSUPP; | |
1579 | ||
1580 | /* | |
1581 | * wait for ordered IO before we have any locks. We'll loop again | |
1582 | * below with the locks held. | |
1583 | */ | |
1584 | btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start); | |
1585 | ||
1586 | mutex_lock(&inode->i_mutex); | |
1587 | ret = inode_newsize_ok(inode, alloc_end); | |
1588 | if (ret) | |
1589 | goto out; | |
1590 | ||
1591 | if (alloc_start > inode->i_size) { | |
a41ad394 JB |
1592 | ret = btrfs_cont_expand(inode, i_size_read(inode), |
1593 | alloc_start); | |
2fe17c10 CH |
1594 | if (ret) |
1595 | goto out; | |
1596 | } | |
1597 | ||
1598 | ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start); | |
1599 | if (ret) | |
1600 | goto out; | |
1601 | ||
1602 | locked_end = alloc_end - 1; | |
1603 | while (1) { | |
1604 | struct btrfs_ordered_extent *ordered; | |
1605 | ||
1606 | /* the extent lock is ordered inside the running | |
1607 | * transaction | |
1608 | */ | |
1609 | lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start, | |
1610 | locked_end, 0, &cached_state, GFP_NOFS); | |
1611 | ordered = btrfs_lookup_first_ordered_extent(inode, | |
1612 | alloc_end - 1); | |
1613 | if (ordered && | |
1614 | ordered->file_offset + ordered->len > alloc_start && | |
1615 | ordered->file_offset < alloc_end) { | |
1616 | btrfs_put_ordered_extent(ordered); | |
1617 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, | |
1618 | alloc_start, locked_end, | |
1619 | &cached_state, GFP_NOFS); | |
1620 | /* | |
1621 | * we can't wait on the range with the transaction | |
1622 | * running or with the extent lock held | |
1623 | */ | |
1624 | btrfs_wait_ordered_range(inode, alloc_start, | |
1625 | alloc_end - alloc_start); | |
1626 | } else { | |
1627 | if (ordered) | |
1628 | btrfs_put_ordered_extent(ordered); | |
1629 | break; | |
1630 | } | |
1631 | } | |
1632 | ||
1633 | cur_offset = alloc_start; | |
1634 | while (1) { | |
1635 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
1636 | alloc_end - cur_offset, 0); | |
c704005d | 1637 | BUG_ON(IS_ERR_OR_NULL(em)); |
2fe17c10 CH |
1638 | last_byte = min(extent_map_end(em), alloc_end); |
1639 | last_byte = (last_byte + mask) & ~mask; | |
1640 | if (em->block_start == EXTENT_MAP_HOLE || | |
1641 | (cur_offset >= inode->i_size && | |
1642 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
1643 | ret = btrfs_prealloc_file_range(inode, mode, cur_offset, | |
1644 | last_byte - cur_offset, | |
1645 | 1 << inode->i_blkbits, | |
1646 | offset + len, | |
1647 | &alloc_hint); | |
1648 | if (ret < 0) { | |
1649 | free_extent_map(em); | |
1650 | break; | |
1651 | } | |
1652 | } | |
1653 | free_extent_map(em); | |
1654 | ||
1655 | cur_offset = last_byte; | |
1656 | if (cur_offset >= alloc_end) { | |
1657 | ret = 0; | |
1658 | break; | |
1659 | } | |
1660 | } | |
1661 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, | |
1662 | &cached_state, GFP_NOFS); | |
1663 | ||
1664 | btrfs_free_reserved_data_space(inode, alloc_end - alloc_start); | |
1665 | out: | |
1666 | mutex_unlock(&inode->i_mutex); | |
1667 | return ret; | |
1668 | } | |
1669 | ||
828c0950 | 1670 | const struct file_operations btrfs_file_operations = { |
39279cc3 CM |
1671 | .llseek = generic_file_llseek, |
1672 | .read = do_sync_read, | |
4a001071 | 1673 | .write = do_sync_write, |
9ebefb18 | 1674 | .aio_read = generic_file_aio_read, |
e9906a98 | 1675 | .splice_read = generic_file_splice_read, |
11c65dcc | 1676 | .aio_write = btrfs_file_aio_write, |
9ebefb18 | 1677 | .mmap = btrfs_file_mmap, |
39279cc3 | 1678 | .open = generic_file_open, |
e1b81e67 | 1679 | .release = btrfs_release_file, |
39279cc3 | 1680 | .fsync = btrfs_sync_file, |
2fe17c10 | 1681 | .fallocate = btrfs_fallocate, |
34287aa3 | 1682 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 1683 | #ifdef CONFIG_COMPAT |
34287aa3 | 1684 | .compat_ioctl = btrfs_ioctl, |
39279cc3 CM |
1685 | #endif |
1686 | }; |