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" |
2aaa6655 | 42 | #include "volumes.h" |
39279cc3 | 43 | |
9247f317 | 44 | static struct kmem_cache *btrfs_inode_defrag_cachep; |
4cb5300b CM |
45 | /* |
46 | * when auto defrag is enabled we | |
47 | * queue up these defrag structs to remember which | |
48 | * inodes need defragging passes | |
49 | */ | |
50 | struct inode_defrag { | |
51 | struct rb_node rb_node; | |
52 | /* objectid */ | |
53 | u64 ino; | |
54 | /* | |
55 | * transid where the defrag was added, we search for | |
56 | * extents newer than this | |
57 | */ | |
58 | u64 transid; | |
59 | ||
60 | /* root objectid */ | |
61 | u64 root; | |
62 | ||
63 | /* last offset we were able to defrag */ | |
64 | u64 last_offset; | |
65 | ||
66 | /* if we've wrapped around back to zero once already */ | |
67 | int cycled; | |
68 | }; | |
69 | ||
762f2263 MX |
70 | static int __compare_inode_defrag(struct inode_defrag *defrag1, |
71 | struct inode_defrag *defrag2) | |
72 | { | |
73 | if (defrag1->root > defrag2->root) | |
74 | return 1; | |
75 | else if (defrag1->root < defrag2->root) | |
76 | return -1; | |
77 | else if (defrag1->ino > defrag2->ino) | |
78 | return 1; | |
79 | else if (defrag1->ino < defrag2->ino) | |
80 | return -1; | |
81 | else | |
82 | return 0; | |
83 | } | |
84 | ||
4cb5300b CM |
85 | /* pop a record for an inode into the defrag tree. The lock |
86 | * must be held already | |
87 | * | |
88 | * If you're inserting a record for an older transid than an | |
89 | * existing record, the transid already in the tree is lowered | |
90 | * | |
91 | * If an existing record is found the defrag item you | |
92 | * pass in is freed | |
93 | */ | |
8ddc4734 | 94 | static int __btrfs_add_inode_defrag(struct inode *inode, |
4cb5300b CM |
95 | struct inode_defrag *defrag) |
96 | { | |
97 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
98 | struct inode_defrag *entry; | |
99 | struct rb_node **p; | |
100 | struct rb_node *parent = NULL; | |
762f2263 | 101 | int ret; |
4cb5300b CM |
102 | |
103 | p = &root->fs_info->defrag_inodes.rb_node; | |
104 | while (*p) { | |
105 | parent = *p; | |
106 | entry = rb_entry(parent, struct inode_defrag, rb_node); | |
107 | ||
762f2263 MX |
108 | ret = __compare_inode_defrag(defrag, entry); |
109 | if (ret < 0) | |
4cb5300b | 110 | p = &parent->rb_left; |
762f2263 | 111 | else if (ret > 0) |
4cb5300b CM |
112 | p = &parent->rb_right; |
113 | else { | |
114 | /* if we're reinserting an entry for | |
115 | * an old defrag run, make sure to | |
116 | * lower the transid of our existing record | |
117 | */ | |
118 | if (defrag->transid < entry->transid) | |
119 | entry->transid = defrag->transid; | |
120 | if (defrag->last_offset > entry->last_offset) | |
121 | entry->last_offset = defrag->last_offset; | |
8ddc4734 | 122 | return -EEXIST; |
4cb5300b CM |
123 | } |
124 | } | |
72ac3c0d | 125 | set_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags); |
4cb5300b CM |
126 | rb_link_node(&defrag->rb_node, parent, p); |
127 | rb_insert_color(&defrag->rb_node, &root->fs_info->defrag_inodes); | |
8ddc4734 MX |
128 | return 0; |
129 | } | |
4cb5300b | 130 | |
8ddc4734 MX |
131 | static inline int __need_auto_defrag(struct btrfs_root *root) |
132 | { | |
133 | if (!btrfs_test_opt(root, AUTO_DEFRAG)) | |
134 | return 0; | |
135 | ||
136 | if (btrfs_fs_closing(root->fs_info)) | |
137 | return 0; | |
4cb5300b | 138 | |
8ddc4734 | 139 | return 1; |
4cb5300b CM |
140 | } |
141 | ||
142 | /* | |
143 | * insert a defrag record for this inode if auto defrag is | |
144 | * enabled | |
145 | */ | |
146 | int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans, | |
147 | struct inode *inode) | |
148 | { | |
149 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
150 | struct inode_defrag *defrag; | |
4cb5300b | 151 | u64 transid; |
8ddc4734 | 152 | int ret; |
4cb5300b | 153 | |
8ddc4734 | 154 | if (!__need_auto_defrag(root)) |
4cb5300b CM |
155 | return 0; |
156 | ||
72ac3c0d | 157 | if (test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags)) |
4cb5300b CM |
158 | return 0; |
159 | ||
160 | if (trans) | |
161 | transid = trans->transid; | |
162 | else | |
163 | transid = BTRFS_I(inode)->root->last_trans; | |
164 | ||
9247f317 | 165 | defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS); |
4cb5300b CM |
166 | if (!defrag) |
167 | return -ENOMEM; | |
168 | ||
a4689d2b | 169 | defrag->ino = btrfs_ino(inode); |
4cb5300b CM |
170 | defrag->transid = transid; |
171 | defrag->root = root->root_key.objectid; | |
172 | ||
173 | spin_lock(&root->fs_info->defrag_inodes_lock); | |
8ddc4734 MX |
174 | if (!test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags)) { |
175 | /* | |
176 | * If we set IN_DEFRAG flag and evict the inode from memory, | |
177 | * and then re-read this inode, this new inode doesn't have | |
178 | * IN_DEFRAG flag. At the case, we may find the existed defrag. | |
179 | */ | |
180 | ret = __btrfs_add_inode_defrag(inode, defrag); | |
181 | if (ret) | |
182 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
183 | } else { | |
9247f317 | 184 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); |
8ddc4734 | 185 | } |
4cb5300b | 186 | spin_unlock(&root->fs_info->defrag_inodes_lock); |
a0f98dde | 187 | return 0; |
4cb5300b CM |
188 | } |
189 | ||
8ddc4734 MX |
190 | /* |
191 | * Requeue the defrag object. If there is a defrag object that points to | |
192 | * the same inode in the tree, we will merge them together (by | |
193 | * __btrfs_add_inode_defrag()) and free the one that we want to requeue. | |
194 | */ | |
195 | void btrfs_requeue_inode_defrag(struct inode *inode, | |
196 | struct inode_defrag *defrag) | |
197 | { | |
198 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
199 | int ret; | |
200 | ||
201 | if (!__need_auto_defrag(root)) | |
202 | goto out; | |
203 | ||
204 | /* | |
205 | * Here we don't check the IN_DEFRAG flag, because we need merge | |
206 | * them together. | |
207 | */ | |
208 | spin_lock(&root->fs_info->defrag_inodes_lock); | |
209 | ret = __btrfs_add_inode_defrag(inode, defrag); | |
210 | spin_unlock(&root->fs_info->defrag_inodes_lock); | |
211 | if (ret) | |
212 | goto out; | |
213 | return; | |
214 | out: | |
215 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
216 | } | |
217 | ||
4cb5300b | 218 | /* |
26176e7c MX |
219 | * pick the defragable inode that we want, if it doesn't exist, we will get |
220 | * the next one. | |
4cb5300b | 221 | */ |
26176e7c MX |
222 | static struct inode_defrag * |
223 | btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino) | |
4cb5300b CM |
224 | { |
225 | struct inode_defrag *entry = NULL; | |
762f2263 | 226 | struct inode_defrag tmp; |
4cb5300b CM |
227 | struct rb_node *p; |
228 | struct rb_node *parent = NULL; | |
762f2263 MX |
229 | int ret; |
230 | ||
231 | tmp.ino = ino; | |
232 | tmp.root = root; | |
4cb5300b | 233 | |
26176e7c MX |
234 | spin_lock(&fs_info->defrag_inodes_lock); |
235 | p = fs_info->defrag_inodes.rb_node; | |
4cb5300b CM |
236 | while (p) { |
237 | parent = p; | |
238 | entry = rb_entry(parent, struct inode_defrag, rb_node); | |
239 | ||
762f2263 MX |
240 | ret = __compare_inode_defrag(&tmp, entry); |
241 | if (ret < 0) | |
4cb5300b | 242 | p = parent->rb_left; |
762f2263 | 243 | else if (ret > 0) |
4cb5300b CM |
244 | p = parent->rb_right; |
245 | else | |
26176e7c | 246 | goto out; |
4cb5300b CM |
247 | } |
248 | ||
26176e7c MX |
249 | if (parent && __compare_inode_defrag(&tmp, entry) > 0) { |
250 | parent = rb_next(parent); | |
251 | if (parent) | |
4cb5300b | 252 | entry = rb_entry(parent, struct inode_defrag, rb_node); |
26176e7c MX |
253 | else |
254 | entry = NULL; | |
4cb5300b | 255 | } |
26176e7c MX |
256 | out: |
257 | if (entry) | |
258 | rb_erase(parent, &fs_info->defrag_inodes); | |
259 | spin_unlock(&fs_info->defrag_inodes_lock); | |
260 | return entry; | |
4cb5300b CM |
261 | } |
262 | ||
26176e7c | 263 | void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info) |
4cb5300b CM |
264 | { |
265 | struct inode_defrag *defrag; | |
26176e7c MX |
266 | struct rb_node *node; |
267 | ||
268 | spin_lock(&fs_info->defrag_inodes_lock); | |
269 | node = rb_first(&fs_info->defrag_inodes); | |
270 | while (node) { | |
271 | rb_erase(node, &fs_info->defrag_inodes); | |
272 | defrag = rb_entry(node, struct inode_defrag, rb_node); | |
273 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
274 | ||
275 | if (need_resched()) { | |
276 | spin_unlock(&fs_info->defrag_inodes_lock); | |
277 | cond_resched(); | |
278 | spin_lock(&fs_info->defrag_inodes_lock); | |
279 | } | |
280 | ||
281 | node = rb_first(&fs_info->defrag_inodes); | |
282 | } | |
283 | spin_unlock(&fs_info->defrag_inodes_lock); | |
284 | } | |
285 | ||
286 | #define BTRFS_DEFRAG_BATCH 1024 | |
287 | ||
288 | static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info, | |
289 | struct inode_defrag *defrag) | |
290 | { | |
4cb5300b CM |
291 | struct btrfs_root *inode_root; |
292 | struct inode *inode; | |
4cb5300b CM |
293 | struct btrfs_key key; |
294 | struct btrfs_ioctl_defrag_range_args range; | |
4cb5300b | 295 | int num_defrag; |
6f1c3605 LB |
296 | int index; |
297 | int ret; | |
4cb5300b | 298 | |
26176e7c MX |
299 | /* get the inode */ |
300 | key.objectid = defrag->root; | |
301 | btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY); | |
302 | key.offset = (u64)-1; | |
6f1c3605 LB |
303 | |
304 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
305 | ||
26176e7c MX |
306 | inode_root = btrfs_read_fs_root_no_name(fs_info, &key); |
307 | if (IS_ERR(inode_root)) { | |
6f1c3605 LB |
308 | ret = PTR_ERR(inode_root); |
309 | goto cleanup; | |
310 | } | |
311 | if (btrfs_root_refs(&inode_root->root_item) == 0) { | |
312 | ret = -ENOENT; | |
313 | goto cleanup; | |
26176e7c MX |
314 | } |
315 | ||
316 | key.objectid = defrag->ino; | |
317 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); | |
318 | key.offset = 0; | |
319 | inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL); | |
320 | if (IS_ERR(inode)) { | |
6f1c3605 LB |
321 | ret = PTR_ERR(inode); |
322 | goto cleanup; | |
26176e7c | 323 | } |
6f1c3605 | 324 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
26176e7c MX |
325 | |
326 | /* do a chunk of defrag */ | |
327 | clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags); | |
4cb5300b CM |
328 | memset(&range, 0, sizeof(range)); |
329 | range.len = (u64)-1; | |
26176e7c | 330 | range.start = defrag->last_offset; |
b66f00da MX |
331 | |
332 | sb_start_write(fs_info->sb); | |
26176e7c MX |
333 | num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid, |
334 | BTRFS_DEFRAG_BATCH); | |
b66f00da | 335 | sb_end_write(fs_info->sb); |
26176e7c MX |
336 | /* |
337 | * if we filled the whole defrag batch, there | |
338 | * must be more work to do. Queue this defrag | |
339 | * again | |
340 | */ | |
341 | if (num_defrag == BTRFS_DEFRAG_BATCH) { | |
342 | defrag->last_offset = range.start; | |
343 | btrfs_requeue_inode_defrag(inode, defrag); | |
344 | } else if (defrag->last_offset && !defrag->cycled) { | |
345 | /* | |
346 | * we didn't fill our defrag batch, but | |
347 | * we didn't start at zero. Make sure we loop | |
348 | * around to the start of the file. | |
349 | */ | |
350 | defrag->last_offset = 0; | |
351 | defrag->cycled = 1; | |
352 | btrfs_requeue_inode_defrag(inode, defrag); | |
353 | } else { | |
354 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
355 | } | |
356 | ||
357 | iput(inode); | |
358 | return 0; | |
6f1c3605 LB |
359 | cleanup: |
360 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
361 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
362 | return ret; | |
26176e7c MX |
363 | } |
364 | ||
365 | /* | |
366 | * run through the list of inodes in the FS that need | |
367 | * defragging | |
368 | */ | |
369 | int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info) | |
370 | { | |
371 | struct inode_defrag *defrag; | |
372 | u64 first_ino = 0; | |
373 | u64 root_objectid = 0; | |
4cb5300b CM |
374 | |
375 | atomic_inc(&fs_info->defrag_running); | |
4cb5300b | 376 | while(1) { |
26176e7c MX |
377 | if (!__need_auto_defrag(fs_info->tree_root)) |
378 | break; | |
4cb5300b CM |
379 | |
380 | /* find an inode to defrag */ | |
26176e7c MX |
381 | defrag = btrfs_pick_defrag_inode(fs_info, root_objectid, |
382 | first_ino); | |
4cb5300b | 383 | if (!defrag) { |
26176e7c | 384 | if (root_objectid || first_ino) { |
762f2263 | 385 | root_objectid = 0; |
4cb5300b CM |
386 | first_ino = 0; |
387 | continue; | |
388 | } else { | |
389 | break; | |
390 | } | |
391 | } | |
392 | ||
4cb5300b | 393 | first_ino = defrag->ino + 1; |
762f2263 | 394 | root_objectid = defrag->root; |
4cb5300b | 395 | |
26176e7c | 396 | __btrfs_run_defrag_inode(fs_info, defrag); |
4cb5300b | 397 | } |
4cb5300b CM |
398 | atomic_dec(&fs_info->defrag_running); |
399 | ||
400 | /* | |
401 | * during unmount, we use the transaction_wait queue to | |
402 | * wait for the defragger to stop | |
403 | */ | |
404 | wake_up(&fs_info->transaction_wait); | |
405 | return 0; | |
406 | } | |
39279cc3 | 407 | |
d352ac68 CM |
408 | /* simple helper to fault in pages and copy. This should go away |
409 | * and be replaced with calls into generic code. | |
410 | */ | |
d397712b | 411 | static noinline int btrfs_copy_from_user(loff_t pos, int num_pages, |
d0215f3e | 412 | size_t write_bytes, |
a1b32a59 | 413 | struct page **prepared_pages, |
11c65dcc | 414 | struct iov_iter *i) |
39279cc3 | 415 | { |
914ee295 | 416 | size_t copied = 0; |
d0215f3e | 417 | size_t total_copied = 0; |
11c65dcc | 418 | int pg = 0; |
39279cc3 CM |
419 | int offset = pos & (PAGE_CACHE_SIZE - 1); |
420 | ||
11c65dcc | 421 | while (write_bytes > 0) { |
39279cc3 CM |
422 | size_t count = min_t(size_t, |
423 | PAGE_CACHE_SIZE - offset, write_bytes); | |
11c65dcc | 424 | struct page *page = prepared_pages[pg]; |
914ee295 XZ |
425 | /* |
426 | * Copy data from userspace to the current page | |
427 | * | |
428 | * Disable pagefault to avoid recursive lock since | |
429 | * the pages are already locked | |
430 | */ | |
431 | pagefault_disable(); | |
432 | copied = iov_iter_copy_from_user_atomic(page, i, offset, count); | |
433 | pagefault_enable(); | |
11c65dcc | 434 | |
39279cc3 CM |
435 | /* Flush processor's dcache for this page */ |
436 | flush_dcache_page(page); | |
31339acd CM |
437 | |
438 | /* | |
439 | * if we get a partial write, we can end up with | |
440 | * partially up to date pages. These add | |
441 | * a lot of complexity, so make sure they don't | |
442 | * happen by forcing this copy to be retried. | |
443 | * | |
444 | * The rest of the btrfs_file_write code will fall | |
445 | * back to page at a time copies after we return 0. | |
446 | */ | |
447 | if (!PageUptodate(page) && copied < count) | |
448 | copied = 0; | |
449 | ||
11c65dcc JB |
450 | iov_iter_advance(i, copied); |
451 | write_bytes -= copied; | |
914ee295 | 452 | total_copied += copied; |
39279cc3 | 453 | |
914ee295 | 454 | /* Return to btrfs_file_aio_write to fault page */ |
9f570b8d | 455 | if (unlikely(copied == 0)) |
914ee295 | 456 | break; |
11c65dcc JB |
457 | |
458 | if (unlikely(copied < PAGE_CACHE_SIZE - offset)) { | |
459 | offset += copied; | |
460 | } else { | |
461 | pg++; | |
462 | offset = 0; | |
463 | } | |
39279cc3 | 464 | } |
914ee295 | 465 | return total_copied; |
39279cc3 CM |
466 | } |
467 | ||
d352ac68 CM |
468 | /* |
469 | * unlocks pages after btrfs_file_write is done with them | |
470 | */ | |
be1a12a0 | 471 | void btrfs_drop_pages(struct page **pages, size_t num_pages) |
39279cc3 CM |
472 | { |
473 | size_t i; | |
474 | for (i = 0; i < num_pages; i++) { | |
d352ac68 CM |
475 | /* page checked is some magic around finding pages that |
476 | * have been modified without going through btrfs_set_page_dirty | |
477 | * clear it here | |
478 | */ | |
4a096752 | 479 | ClearPageChecked(pages[i]); |
39279cc3 CM |
480 | unlock_page(pages[i]); |
481 | mark_page_accessed(pages[i]); | |
482 | page_cache_release(pages[i]); | |
483 | } | |
484 | } | |
485 | ||
d352ac68 CM |
486 | /* |
487 | * after copy_from_user, pages need to be dirtied and we need to make | |
488 | * sure holes are created between the current EOF and the start of | |
489 | * any next extents (if required). | |
490 | * | |
491 | * this also makes the decision about creating an inline extent vs | |
492 | * doing real data extents, marking pages dirty and delalloc as required. | |
493 | */ | |
be1a12a0 JB |
494 | int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode, |
495 | struct page **pages, size_t num_pages, | |
496 | loff_t pos, size_t write_bytes, | |
497 | struct extent_state **cached) | |
39279cc3 | 498 | { |
39279cc3 | 499 | int err = 0; |
a52d9a80 | 500 | int i; |
db94535d | 501 | u64 num_bytes; |
a52d9a80 CM |
502 | u64 start_pos; |
503 | u64 end_of_last_block; | |
504 | u64 end_pos = pos + write_bytes; | |
505 | loff_t isize = i_size_read(inode); | |
39279cc3 | 506 | |
5f39d397 | 507 | start_pos = pos & ~((u64)root->sectorsize - 1); |
db94535d CM |
508 | num_bytes = (write_bytes + pos - start_pos + |
509 | root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
39279cc3 | 510 | |
db94535d | 511 | end_of_last_block = start_pos + num_bytes - 1; |
2ac55d41 | 512 | err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block, |
be1a12a0 | 513 | cached); |
d0215f3e JB |
514 | if (err) |
515 | return err; | |
9ed74f2d | 516 | |
c8b97818 CM |
517 | for (i = 0; i < num_pages; i++) { |
518 | struct page *p = pages[i]; | |
519 | SetPageUptodate(p); | |
520 | ClearPageChecked(p); | |
521 | set_page_dirty(p); | |
a52d9a80 | 522 | } |
9f570b8d JB |
523 | |
524 | /* | |
525 | * we've only changed i_size in ram, and we haven't updated | |
526 | * the disk i_size. There is no need to log the inode | |
527 | * at this time. | |
528 | */ | |
529 | if (end_pos > isize) | |
a52d9a80 | 530 | i_size_write(inode, end_pos); |
a22285a6 | 531 | return 0; |
39279cc3 CM |
532 | } |
533 | ||
d352ac68 CM |
534 | /* |
535 | * this drops all the extents in the cache that intersect the range | |
536 | * [start, end]. Existing extents are split as required. | |
537 | */ | |
7014cdb4 JB |
538 | void btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end, |
539 | int skip_pinned) | |
a52d9a80 CM |
540 | { |
541 | struct extent_map *em; | |
3b951516 CM |
542 | struct extent_map *split = NULL; |
543 | struct extent_map *split2 = NULL; | |
a52d9a80 | 544 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
39b5637f | 545 | u64 len = end - start + 1; |
5dc562c5 | 546 | u64 gen; |
3b951516 CM |
547 | int ret; |
548 | int testend = 1; | |
5b21f2ed | 549 | unsigned long flags; |
c8b97818 | 550 | int compressed = 0; |
a52d9a80 | 551 | |
e6dcd2dc | 552 | WARN_ON(end < start); |
3b951516 | 553 | if (end == (u64)-1) { |
39b5637f | 554 | len = (u64)-1; |
3b951516 CM |
555 | testend = 0; |
556 | } | |
d397712b | 557 | while (1) { |
7014cdb4 JB |
558 | int no_splits = 0; |
559 | ||
3b951516 | 560 | if (!split) |
172ddd60 | 561 | split = alloc_extent_map(); |
3b951516 | 562 | if (!split2) |
172ddd60 | 563 | split2 = alloc_extent_map(); |
7014cdb4 JB |
564 | if (!split || !split2) |
565 | no_splits = 1; | |
3b951516 | 566 | |
890871be | 567 | write_lock(&em_tree->lock); |
39b5637f | 568 | em = lookup_extent_mapping(em_tree, start, len); |
d1310b2e | 569 | if (!em) { |
890871be | 570 | write_unlock(&em_tree->lock); |
a52d9a80 | 571 | break; |
d1310b2e | 572 | } |
5b21f2ed | 573 | flags = em->flags; |
5dc562c5 | 574 | gen = em->generation; |
5b21f2ed | 575 | if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) { |
55ef6899 | 576 | if (testend && em->start + em->len >= start + len) { |
5b21f2ed | 577 | free_extent_map(em); |
a1ed835e | 578 | write_unlock(&em_tree->lock); |
5b21f2ed ZY |
579 | break; |
580 | } | |
55ef6899 YZ |
581 | start = em->start + em->len; |
582 | if (testend) | |
5b21f2ed | 583 | len = start + len - (em->start + em->len); |
5b21f2ed | 584 | free_extent_map(em); |
a1ed835e | 585 | write_unlock(&em_tree->lock); |
5b21f2ed ZY |
586 | continue; |
587 | } | |
c8b97818 | 588 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
3ce7e67a | 589 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
a52d9a80 | 590 | remove_extent_mapping(em_tree, em); |
7014cdb4 JB |
591 | if (no_splits) |
592 | goto next; | |
3b951516 CM |
593 | |
594 | if (em->block_start < EXTENT_MAP_LAST_BYTE && | |
595 | em->start < start) { | |
596 | split->start = em->start; | |
597 | split->len = start - em->start; | |
ff5b7ee3 | 598 | split->orig_start = em->orig_start; |
3b951516 | 599 | split->block_start = em->block_start; |
c8b97818 CM |
600 | |
601 | if (compressed) | |
602 | split->block_len = em->block_len; | |
603 | else | |
604 | split->block_len = split->len; | |
b4939680 JB |
605 | split->orig_block_len = max(split->block_len, |
606 | em->orig_block_len); | |
5dc562c5 | 607 | split->generation = gen; |
3b951516 | 608 | split->bdev = em->bdev; |
5b21f2ed | 609 | split->flags = flags; |
261507a0 | 610 | split->compress_type = em->compress_type; |
3b951516 | 611 | ret = add_extent_mapping(em_tree, split); |
79787eaa | 612 | BUG_ON(ret); /* Logic error */ |
5dc562c5 | 613 | list_move(&split->list, &em_tree->modified_extents); |
3b951516 CM |
614 | free_extent_map(split); |
615 | split = split2; | |
616 | split2 = NULL; | |
617 | } | |
618 | if (em->block_start < EXTENT_MAP_LAST_BYTE && | |
619 | testend && em->start + em->len > start + len) { | |
620 | u64 diff = start + len - em->start; | |
621 | ||
622 | split->start = start + len; | |
623 | split->len = em->start + em->len - (start + len); | |
624 | split->bdev = em->bdev; | |
5b21f2ed | 625 | split->flags = flags; |
261507a0 | 626 | split->compress_type = em->compress_type; |
5dc562c5 | 627 | split->generation = gen; |
b4939680 JB |
628 | split->orig_block_len = max(em->block_len, |
629 | em->orig_block_len); | |
3b951516 | 630 | |
c8b97818 CM |
631 | if (compressed) { |
632 | split->block_len = em->block_len; | |
633 | split->block_start = em->block_start; | |
445a6944 | 634 | split->orig_start = em->orig_start; |
c8b97818 CM |
635 | } else { |
636 | split->block_len = split->len; | |
637 | split->block_start = em->block_start + diff; | |
70c8a91c | 638 | split->orig_start = em->orig_start; |
c8b97818 | 639 | } |
3b951516 CM |
640 | |
641 | ret = add_extent_mapping(em_tree, split); | |
79787eaa | 642 | BUG_ON(ret); /* Logic error */ |
5dc562c5 | 643 | list_move(&split->list, &em_tree->modified_extents); |
3b951516 CM |
644 | free_extent_map(split); |
645 | split = NULL; | |
646 | } | |
7014cdb4 | 647 | next: |
890871be | 648 | write_unlock(&em_tree->lock); |
d1310b2e | 649 | |
a52d9a80 CM |
650 | /* once for us */ |
651 | free_extent_map(em); | |
652 | /* once for the tree*/ | |
653 | free_extent_map(em); | |
654 | } | |
3b951516 CM |
655 | if (split) |
656 | free_extent_map(split); | |
657 | if (split2) | |
658 | free_extent_map(split2); | |
a52d9a80 CM |
659 | } |
660 | ||
39279cc3 CM |
661 | /* |
662 | * this is very complex, but the basic idea is to drop all extents | |
663 | * in the range start - end. hint_block is filled in with a block number | |
664 | * that would be a good hint to the block allocator for this file. | |
665 | * | |
666 | * If an extent intersects the range but is not entirely inside the range | |
667 | * it is either truncated or split. Anything entirely inside the range | |
668 | * is deleted from the tree. | |
669 | */ | |
5dc562c5 JB |
670 | int __btrfs_drop_extents(struct btrfs_trans_handle *trans, |
671 | struct btrfs_root *root, struct inode *inode, | |
672 | struct btrfs_path *path, u64 start, u64 end, | |
2aaa6655 | 673 | u64 *drop_end, int drop_cache) |
39279cc3 | 674 | { |
5f39d397 | 675 | struct extent_buffer *leaf; |
920bbbfb | 676 | struct btrfs_file_extent_item *fi; |
00f5c795 | 677 | struct btrfs_key key; |
920bbbfb | 678 | struct btrfs_key new_key; |
33345d01 | 679 | u64 ino = btrfs_ino(inode); |
920bbbfb YZ |
680 | u64 search_start = start; |
681 | u64 disk_bytenr = 0; | |
682 | u64 num_bytes = 0; | |
683 | u64 extent_offset = 0; | |
684 | u64 extent_end = 0; | |
685 | int del_nr = 0; | |
686 | int del_slot = 0; | |
687 | int extent_type; | |
ccd467d6 | 688 | int recow; |
00f5c795 | 689 | int ret; |
dc7fdde3 | 690 | int modify_tree = -1; |
5dc562c5 | 691 | int update_refs = (root->ref_cows || root == root->fs_info->tree_root); |
c3308f84 | 692 | int found = 0; |
39279cc3 | 693 | |
a1ed835e CM |
694 | if (drop_cache) |
695 | btrfs_drop_extent_cache(inode, start, end - 1, 0); | |
a52d9a80 | 696 | |
dc7fdde3 CM |
697 | if (start >= BTRFS_I(inode)->disk_i_size) |
698 | modify_tree = 0; | |
699 | ||
d397712b | 700 | while (1) { |
ccd467d6 | 701 | recow = 0; |
33345d01 | 702 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
dc7fdde3 | 703 | search_start, modify_tree); |
39279cc3 | 704 | if (ret < 0) |
920bbbfb YZ |
705 | break; |
706 | if (ret > 0 && path->slots[0] > 0 && search_start == start) { | |
707 | leaf = path->nodes[0]; | |
708 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1); | |
33345d01 | 709 | if (key.objectid == ino && |
920bbbfb YZ |
710 | key.type == BTRFS_EXTENT_DATA_KEY) |
711 | path->slots[0]--; | |
39279cc3 | 712 | } |
920bbbfb | 713 | ret = 0; |
8c2383c3 | 714 | next_slot: |
5f39d397 | 715 | leaf = path->nodes[0]; |
920bbbfb YZ |
716 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { |
717 | BUG_ON(del_nr > 0); | |
718 | ret = btrfs_next_leaf(root, path); | |
719 | if (ret < 0) | |
720 | break; | |
721 | if (ret > 0) { | |
722 | ret = 0; | |
723 | break; | |
8c2383c3 | 724 | } |
920bbbfb YZ |
725 | leaf = path->nodes[0]; |
726 | recow = 1; | |
727 | } | |
728 | ||
729 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
33345d01 | 730 | if (key.objectid > ino || |
920bbbfb YZ |
731 | key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end) |
732 | break; | |
733 | ||
734 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
735 | struct btrfs_file_extent_item); | |
736 | extent_type = btrfs_file_extent_type(leaf, fi); | |
737 | ||
738 | if (extent_type == BTRFS_FILE_EXTENT_REG || | |
739 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
740 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
741 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
742 | extent_offset = btrfs_file_extent_offset(leaf, fi); | |
743 | extent_end = key.offset + | |
744 | btrfs_file_extent_num_bytes(leaf, fi); | |
745 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
746 | extent_end = key.offset + | |
747 | btrfs_file_extent_inline_len(leaf, fi); | |
8c2383c3 | 748 | } else { |
920bbbfb | 749 | WARN_ON(1); |
8c2383c3 | 750 | extent_end = search_start; |
39279cc3 CM |
751 | } |
752 | ||
920bbbfb YZ |
753 | if (extent_end <= search_start) { |
754 | path->slots[0]++; | |
8c2383c3 | 755 | goto next_slot; |
39279cc3 CM |
756 | } |
757 | ||
c3308f84 | 758 | found = 1; |
920bbbfb | 759 | search_start = max(key.offset, start); |
dc7fdde3 CM |
760 | if (recow || !modify_tree) { |
761 | modify_tree = -1; | |
b3b4aa74 | 762 | btrfs_release_path(path); |
920bbbfb | 763 | continue; |
39279cc3 | 764 | } |
6643558d | 765 | |
920bbbfb YZ |
766 | /* |
767 | * | - range to drop - | | |
768 | * | -------- extent -------- | | |
769 | */ | |
770 | if (start > key.offset && end < extent_end) { | |
771 | BUG_ON(del_nr > 0); | |
772 | BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE); | |
773 | ||
774 | memcpy(&new_key, &key, sizeof(new_key)); | |
775 | new_key.offset = start; | |
776 | ret = btrfs_duplicate_item(trans, root, path, | |
777 | &new_key); | |
778 | if (ret == -EAGAIN) { | |
b3b4aa74 | 779 | btrfs_release_path(path); |
920bbbfb | 780 | continue; |
6643558d | 781 | } |
920bbbfb YZ |
782 | if (ret < 0) |
783 | break; | |
784 | ||
785 | leaf = path->nodes[0]; | |
786 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
787 | struct btrfs_file_extent_item); | |
788 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
789 | start - key.offset); | |
790 | ||
791 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
792 | struct btrfs_file_extent_item); | |
793 | ||
794 | extent_offset += start - key.offset; | |
795 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); | |
796 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
797 | extent_end - start); | |
798 | btrfs_mark_buffer_dirty(leaf); | |
799 | ||
5dc562c5 | 800 | if (update_refs && disk_bytenr > 0) { |
771ed689 | 801 | ret = btrfs_inc_extent_ref(trans, root, |
920bbbfb YZ |
802 | disk_bytenr, num_bytes, 0, |
803 | root->root_key.objectid, | |
804 | new_key.objectid, | |
66d7e7f0 | 805 | start - extent_offset, 0); |
79787eaa | 806 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 807 | } |
920bbbfb | 808 | key.offset = start; |
6643558d | 809 | } |
920bbbfb YZ |
810 | /* |
811 | * | ---- range to drop ----- | | |
812 | * | -------- extent -------- | | |
813 | */ | |
814 | if (start <= key.offset && end < extent_end) { | |
815 | BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE); | |
6643558d | 816 | |
920bbbfb YZ |
817 | memcpy(&new_key, &key, sizeof(new_key)); |
818 | new_key.offset = end; | |
819 | btrfs_set_item_key_safe(trans, root, path, &new_key); | |
6643558d | 820 | |
920bbbfb YZ |
821 | extent_offset += end - key.offset; |
822 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); | |
823 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
824 | extent_end - end); | |
825 | btrfs_mark_buffer_dirty(leaf); | |
2671485d | 826 | if (update_refs && disk_bytenr > 0) |
920bbbfb | 827 | inode_sub_bytes(inode, end - key.offset); |
920bbbfb | 828 | break; |
39279cc3 | 829 | } |
771ed689 | 830 | |
920bbbfb YZ |
831 | search_start = extent_end; |
832 | /* | |
833 | * | ---- range to drop ----- | | |
834 | * | -------- extent -------- | | |
835 | */ | |
836 | if (start > key.offset && end >= extent_end) { | |
837 | BUG_ON(del_nr > 0); | |
838 | BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE); | |
8c2383c3 | 839 | |
920bbbfb YZ |
840 | btrfs_set_file_extent_num_bytes(leaf, fi, |
841 | start - key.offset); | |
842 | btrfs_mark_buffer_dirty(leaf); | |
2671485d | 843 | if (update_refs && disk_bytenr > 0) |
920bbbfb | 844 | inode_sub_bytes(inode, extent_end - start); |
920bbbfb YZ |
845 | if (end == extent_end) |
846 | break; | |
c8b97818 | 847 | |
920bbbfb YZ |
848 | path->slots[0]++; |
849 | goto next_slot; | |
31840ae1 ZY |
850 | } |
851 | ||
920bbbfb YZ |
852 | /* |
853 | * | ---- range to drop ----- | | |
854 | * | ------ extent ------ | | |
855 | */ | |
856 | if (start <= key.offset && end >= extent_end) { | |
857 | if (del_nr == 0) { | |
858 | del_slot = path->slots[0]; | |
859 | del_nr = 1; | |
860 | } else { | |
861 | BUG_ON(del_slot + del_nr != path->slots[0]); | |
862 | del_nr++; | |
863 | } | |
31840ae1 | 864 | |
5dc562c5 JB |
865 | if (update_refs && |
866 | extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
a76a3cd4 | 867 | inode_sub_bytes(inode, |
920bbbfb YZ |
868 | extent_end - key.offset); |
869 | extent_end = ALIGN(extent_end, | |
870 | root->sectorsize); | |
5dc562c5 | 871 | } else if (update_refs && disk_bytenr > 0) { |
31840ae1 | 872 | ret = btrfs_free_extent(trans, root, |
920bbbfb YZ |
873 | disk_bytenr, num_bytes, 0, |
874 | root->root_key.objectid, | |
5d4f98a2 | 875 | key.objectid, key.offset - |
66d7e7f0 | 876 | extent_offset, 0); |
79787eaa | 877 | BUG_ON(ret); /* -ENOMEM */ |
920bbbfb YZ |
878 | inode_sub_bytes(inode, |
879 | extent_end - key.offset); | |
31840ae1 | 880 | } |
31840ae1 | 881 | |
920bbbfb YZ |
882 | if (end == extent_end) |
883 | break; | |
884 | ||
885 | if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) { | |
886 | path->slots[0]++; | |
887 | goto next_slot; | |
888 | } | |
889 | ||
890 | ret = btrfs_del_items(trans, root, path, del_slot, | |
891 | del_nr); | |
79787eaa JM |
892 | if (ret) { |
893 | btrfs_abort_transaction(trans, root, ret); | |
5dc562c5 | 894 | break; |
79787eaa | 895 | } |
920bbbfb YZ |
896 | |
897 | del_nr = 0; | |
898 | del_slot = 0; | |
899 | ||
b3b4aa74 | 900 | btrfs_release_path(path); |
920bbbfb | 901 | continue; |
39279cc3 | 902 | } |
920bbbfb YZ |
903 | |
904 | BUG_ON(1); | |
39279cc3 | 905 | } |
920bbbfb | 906 | |
79787eaa | 907 | if (!ret && del_nr > 0) { |
920bbbfb | 908 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
79787eaa JM |
909 | if (ret) |
910 | btrfs_abort_transaction(trans, root, ret); | |
6643558d | 911 | } |
920bbbfb | 912 | |
2aaa6655 | 913 | if (drop_end) |
c3308f84 | 914 | *drop_end = found ? min(end, extent_end) : end; |
5dc562c5 JB |
915 | btrfs_release_path(path); |
916 | return ret; | |
917 | } | |
918 | ||
919 | int btrfs_drop_extents(struct btrfs_trans_handle *trans, | |
920 | struct btrfs_root *root, struct inode *inode, u64 start, | |
2671485d | 921 | u64 end, int drop_cache) |
5dc562c5 JB |
922 | { |
923 | struct btrfs_path *path; | |
924 | int ret; | |
925 | ||
926 | path = btrfs_alloc_path(); | |
927 | if (!path) | |
928 | return -ENOMEM; | |
2aaa6655 | 929 | ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL, |
2671485d | 930 | drop_cache); |
920bbbfb | 931 | btrfs_free_path(path); |
39279cc3 CM |
932 | return ret; |
933 | } | |
934 | ||
d899e052 | 935 | static int extent_mergeable(struct extent_buffer *leaf, int slot, |
6c7d54ac YZ |
936 | u64 objectid, u64 bytenr, u64 orig_offset, |
937 | u64 *start, u64 *end) | |
d899e052 YZ |
938 | { |
939 | struct btrfs_file_extent_item *fi; | |
940 | struct btrfs_key key; | |
941 | u64 extent_end; | |
942 | ||
943 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) | |
944 | return 0; | |
945 | ||
946 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
947 | if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY) | |
948 | return 0; | |
949 | ||
950 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
951 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG || | |
952 | btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr || | |
6c7d54ac | 953 | btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset || |
d899e052 YZ |
954 | btrfs_file_extent_compression(leaf, fi) || |
955 | btrfs_file_extent_encryption(leaf, fi) || | |
956 | btrfs_file_extent_other_encoding(leaf, fi)) | |
957 | return 0; | |
958 | ||
959 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
960 | if ((*start && *start != key.offset) || (*end && *end != extent_end)) | |
961 | return 0; | |
962 | ||
963 | *start = key.offset; | |
964 | *end = extent_end; | |
965 | return 1; | |
966 | } | |
967 | ||
968 | /* | |
969 | * Mark extent in the range start - end as written. | |
970 | * | |
971 | * This changes extent type from 'pre-allocated' to 'regular'. If only | |
972 | * part of extent is marked as written, the extent will be split into | |
973 | * two or three. | |
974 | */ | |
975 | int btrfs_mark_extent_written(struct btrfs_trans_handle *trans, | |
d899e052 YZ |
976 | struct inode *inode, u64 start, u64 end) |
977 | { | |
920bbbfb | 978 | struct btrfs_root *root = BTRFS_I(inode)->root; |
d899e052 YZ |
979 | struct extent_buffer *leaf; |
980 | struct btrfs_path *path; | |
981 | struct btrfs_file_extent_item *fi; | |
982 | struct btrfs_key key; | |
920bbbfb | 983 | struct btrfs_key new_key; |
d899e052 YZ |
984 | u64 bytenr; |
985 | u64 num_bytes; | |
986 | u64 extent_end; | |
5d4f98a2 | 987 | u64 orig_offset; |
d899e052 YZ |
988 | u64 other_start; |
989 | u64 other_end; | |
920bbbfb YZ |
990 | u64 split; |
991 | int del_nr = 0; | |
992 | int del_slot = 0; | |
6c7d54ac | 993 | int recow; |
d899e052 | 994 | int ret; |
33345d01 | 995 | u64 ino = btrfs_ino(inode); |
d899e052 | 996 | |
d899e052 | 997 | path = btrfs_alloc_path(); |
d8926bb3 MF |
998 | if (!path) |
999 | return -ENOMEM; | |
d899e052 | 1000 | again: |
6c7d54ac | 1001 | recow = 0; |
920bbbfb | 1002 | split = start; |
33345d01 | 1003 | key.objectid = ino; |
d899e052 | 1004 | key.type = BTRFS_EXTENT_DATA_KEY; |
920bbbfb | 1005 | key.offset = split; |
d899e052 YZ |
1006 | |
1007 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
41415730 JB |
1008 | if (ret < 0) |
1009 | goto out; | |
d899e052 YZ |
1010 | if (ret > 0 && path->slots[0] > 0) |
1011 | path->slots[0]--; | |
1012 | ||
1013 | leaf = path->nodes[0]; | |
1014 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
33345d01 | 1015 | BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY); |
d899e052 YZ |
1016 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1017 | struct btrfs_file_extent_item); | |
920bbbfb YZ |
1018 | BUG_ON(btrfs_file_extent_type(leaf, fi) != |
1019 | BTRFS_FILE_EXTENT_PREALLOC); | |
d899e052 YZ |
1020 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
1021 | BUG_ON(key.offset > start || extent_end < end); | |
1022 | ||
1023 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
1024 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
5d4f98a2 | 1025 | orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi); |
6c7d54ac YZ |
1026 | memcpy(&new_key, &key, sizeof(new_key)); |
1027 | ||
1028 | if (start == key.offset && end < extent_end) { | |
1029 | other_start = 0; | |
1030 | other_end = start; | |
1031 | if (extent_mergeable(leaf, path->slots[0] - 1, | |
33345d01 | 1032 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1033 | &other_start, &other_end)) { |
1034 | new_key.offset = end; | |
1035 | btrfs_set_item_key_safe(trans, root, path, &new_key); | |
1036 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1037 | struct btrfs_file_extent_item); | |
224ecce5 JB |
1038 | btrfs_set_file_extent_generation(leaf, fi, |
1039 | trans->transid); | |
6c7d54ac YZ |
1040 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1041 | extent_end - end); | |
1042 | btrfs_set_file_extent_offset(leaf, fi, | |
1043 | end - orig_offset); | |
1044 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
1045 | struct btrfs_file_extent_item); | |
224ecce5 JB |
1046 | btrfs_set_file_extent_generation(leaf, fi, |
1047 | trans->transid); | |
6c7d54ac YZ |
1048 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1049 | end - other_start); | |
1050 | btrfs_mark_buffer_dirty(leaf); | |
1051 | goto out; | |
1052 | } | |
1053 | } | |
1054 | ||
1055 | if (start > key.offset && end == extent_end) { | |
1056 | other_start = end; | |
1057 | other_end = 0; | |
1058 | if (extent_mergeable(leaf, path->slots[0] + 1, | |
33345d01 | 1059 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1060 | &other_start, &other_end)) { |
1061 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1062 | struct btrfs_file_extent_item); | |
1063 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
1064 | start - key.offset); | |
224ecce5 JB |
1065 | btrfs_set_file_extent_generation(leaf, fi, |
1066 | trans->transid); | |
6c7d54ac YZ |
1067 | path->slots[0]++; |
1068 | new_key.offset = start; | |
1069 | btrfs_set_item_key_safe(trans, root, path, &new_key); | |
1070 | ||
1071 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1072 | struct btrfs_file_extent_item); | |
224ecce5 JB |
1073 | btrfs_set_file_extent_generation(leaf, fi, |
1074 | trans->transid); | |
6c7d54ac YZ |
1075 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1076 | other_end - start); | |
1077 | btrfs_set_file_extent_offset(leaf, fi, | |
1078 | start - orig_offset); | |
1079 | btrfs_mark_buffer_dirty(leaf); | |
1080 | goto out; | |
1081 | } | |
1082 | } | |
d899e052 | 1083 | |
920bbbfb YZ |
1084 | while (start > key.offset || end < extent_end) { |
1085 | if (key.offset == start) | |
1086 | split = end; | |
1087 | ||
920bbbfb YZ |
1088 | new_key.offset = split; |
1089 | ret = btrfs_duplicate_item(trans, root, path, &new_key); | |
1090 | if (ret == -EAGAIN) { | |
b3b4aa74 | 1091 | btrfs_release_path(path); |
920bbbfb | 1092 | goto again; |
d899e052 | 1093 | } |
79787eaa JM |
1094 | if (ret < 0) { |
1095 | btrfs_abort_transaction(trans, root, ret); | |
1096 | goto out; | |
1097 | } | |
d899e052 | 1098 | |
920bbbfb YZ |
1099 | leaf = path->nodes[0]; |
1100 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
d899e052 | 1101 | struct btrfs_file_extent_item); |
224ecce5 | 1102 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
d899e052 | 1103 | btrfs_set_file_extent_num_bytes(leaf, fi, |
920bbbfb YZ |
1104 | split - key.offset); |
1105 | ||
1106 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1107 | struct btrfs_file_extent_item); | |
1108 | ||
224ecce5 | 1109 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
920bbbfb YZ |
1110 | btrfs_set_file_extent_offset(leaf, fi, split - orig_offset); |
1111 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
1112 | extent_end - split); | |
d899e052 YZ |
1113 | btrfs_mark_buffer_dirty(leaf); |
1114 | ||
920bbbfb YZ |
1115 | ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0, |
1116 | root->root_key.objectid, | |
66d7e7f0 | 1117 | ino, orig_offset, 0); |
79787eaa | 1118 | BUG_ON(ret); /* -ENOMEM */ |
d899e052 | 1119 | |
920bbbfb YZ |
1120 | if (split == start) { |
1121 | key.offset = start; | |
1122 | } else { | |
1123 | BUG_ON(start != key.offset); | |
d899e052 | 1124 | path->slots[0]--; |
920bbbfb | 1125 | extent_end = end; |
d899e052 | 1126 | } |
6c7d54ac | 1127 | recow = 1; |
d899e052 YZ |
1128 | } |
1129 | ||
920bbbfb YZ |
1130 | other_start = end; |
1131 | other_end = 0; | |
6c7d54ac | 1132 | if (extent_mergeable(leaf, path->slots[0] + 1, |
33345d01 | 1133 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1134 | &other_start, &other_end)) { |
1135 | if (recow) { | |
b3b4aa74 | 1136 | btrfs_release_path(path); |
6c7d54ac YZ |
1137 | goto again; |
1138 | } | |
920bbbfb YZ |
1139 | extent_end = other_end; |
1140 | del_slot = path->slots[0] + 1; | |
1141 | del_nr++; | |
1142 | ret = btrfs_free_extent(trans, root, bytenr, num_bytes, | |
1143 | 0, root->root_key.objectid, | |
66d7e7f0 | 1144 | ino, orig_offset, 0); |
79787eaa | 1145 | BUG_ON(ret); /* -ENOMEM */ |
d899e052 | 1146 | } |
920bbbfb YZ |
1147 | other_start = 0; |
1148 | other_end = start; | |
6c7d54ac | 1149 | if (extent_mergeable(leaf, path->slots[0] - 1, |
33345d01 | 1150 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1151 | &other_start, &other_end)) { |
1152 | if (recow) { | |
b3b4aa74 | 1153 | btrfs_release_path(path); |
6c7d54ac YZ |
1154 | goto again; |
1155 | } | |
920bbbfb YZ |
1156 | key.offset = other_start; |
1157 | del_slot = path->slots[0]; | |
1158 | del_nr++; | |
1159 | ret = btrfs_free_extent(trans, root, bytenr, num_bytes, | |
1160 | 0, root->root_key.objectid, | |
66d7e7f0 | 1161 | ino, orig_offset, 0); |
79787eaa | 1162 | BUG_ON(ret); /* -ENOMEM */ |
920bbbfb YZ |
1163 | } |
1164 | if (del_nr == 0) { | |
3f6fae95 SL |
1165 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1166 | struct btrfs_file_extent_item); | |
920bbbfb YZ |
1167 | btrfs_set_file_extent_type(leaf, fi, |
1168 | BTRFS_FILE_EXTENT_REG); | |
224ecce5 | 1169 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
920bbbfb | 1170 | btrfs_mark_buffer_dirty(leaf); |
6c7d54ac | 1171 | } else { |
3f6fae95 SL |
1172 | fi = btrfs_item_ptr(leaf, del_slot - 1, |
1173 | struct btrfs_file_extent_item); | |
6c7d54ac YZ |
1174 | btrfs_set_file_extent_type(leaf, fi, |
1175 | BTRFS_FILE_EXTENT_REG); | |
224ecce5 | 1176 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
6c7d54ac YZ |
1177 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1178 | extent_end - key.offset); | |
1179 | btrfs_mark_buffer_dirty(leaf); | |
920bbbfb | 1180 | |
6c7d54ac | 1181 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
79787eaa JM |
1182 | if (ret < 0) { |
1183 | btrfs_abort_transaction(trans, root, ret); | |
1184 | goto out; | |
1185 | } | |
6c7d54ac | 1186 | } |
920bbbfb | 1187 | out: |
d899e052 YZ |
1188 | btrfs_free_path(path); |
1189 | return 0; | |
1190 | } | |
1191 | ||
b1bf862e CM |
1192 | /* |
1193 | * on error we return an unlocked page and the error value | |
1194 | * on success we return a locked page and 0 | |
1195 | */ | |
b6316429 JB |
1196 | static int prepare_uptodate_page(struct page *page, u64 pos, |
1197 | bool force_uptodate) | |
b1bf862e CM |
1198 | { |
1199 | int ret = 0; | |
1200 | ||
b6316429 JB |
1201 | if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) && |
1202 | !PageUptodate(page)) { | |
b1bf862e CM |
1203 | ret = btrfs_readpage(NULL, page); |
1204 | if (ret) | |
1205 | return ret; | |
1206 | lock_page(page); | |
1207 | if (!PageUptodate(page)) { | |
1208 | unlock_page(page); | |
1209 | return -EIO; | |
1210 | } | |
1211 | } | |
1212 | return 0; | |
1213 | } | |
1214 | ||
39279cc3 | 1215 | /* |
d352ac68 CM |
1216 | * this gets pages into the page cache and locks them down, it also properly |
1217 | * waits for data=ordered extents to finish before allowing the pages to be | |
1218 | * modified. | |
39279cc3 | 1219 | */ |
d397712b | 1220 | static noinline int prepare_pages(struct btrfs_root *root, struct file *file, |
98ed5174 CM |
1221 | struct page **pages, size_t num_pages, |
1222 | loff_t pos, unsigned long first_index, | |
b6316429 | 1223 | size_t write_bytes, bool force_uptodate) |
39279cc3 | 1224 | { |
2ac55d41 | 1225 | struct extent_state *cached_state = NULL; |
39279cc3 CM |
1226 | int i; |
1227 | unsigned long index = pos >> PAGE_CACHE_SHIFT; | |
6da6abae | 1228 | struct inode *inode = fdentry(file)->d_inode; |
3b16a4e3 | 1229 | gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); |
39279cc3 | 1230 | int err = 0; |
b1bf862e | 1231 | int faili = 0; |
8c2383c3 | 1232 | u64 start_pos; |
e6dcd2dc | 1233 | u64 last_pos; |
8c2383c3 | 1234 | |
5f39d397 | 1235 | start_pos = pos & ~((u64)root->sectorsize - 1); |
e6dcd2dc | 1236 | last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT; |
39279cc3 | 1237 | |
e6dcd2dc | 1238 | again: |
39279cc3 | 1239 | for (i = 0; i < num_pages; i++) { |
a94733d0 | 1240 | pages[i] = find_or_create_page(inode->i_mapping, index + i, |
e3a41a5b | 1241 | mask | __GFP_WRITE); |
39279cc3 | 1242 | if (!pages[i]) { |
b1bf862e CM |
1243 | faili = i - 1; |
1244 | err = -ENOMEM; | |
1245 | goto fail; | |
1246 | } | |
1247 | ||
1248 | if (i == 0) | |
b6316429 JB |
1249 | err = prepare_uptodate_page(pages[i], pos, |
1250 | force_uptodate); | |
b1bf862e CM |
1251 | if (i == num_pages - 1) |
1252 | err = prepare_uptodate_page(pages[i], | |
b6316429 | 1253 | pos + write_bytes, false); |
b1bf862e CM |
1254 | if (err) { |
1255 | page_cache_release(pages[i]); | |
1256 | faili = i - 1; | |
1257 | goto fail; | |
39279cc3 | 1258 | } |
ccd467d6 | 1259 | wait_on_page_writeback(pages[i]); |
39279cc3 | 1260 | } |
b1bf862e | 1261 | err = 0; |
0762704b | 1262 | if (start_pos < inode->i_size) { |
e6dcd2dc | 1263 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 1264 | lock_extent_bits(&BTRFS_I(inode)->io_tree, |
d0082371 | 1265 | start_pos, last_pos - 1, 0, &cached_state); |
d397712b CM |
1266 | ordered = btrfs_lookup_first_ordered_extent(inode, |
1267 | last_pos - 1); | |
e6dcd2dc CM |
1268 | if (ordered && |
1269 | ordered->file_offset + ordered->len > start_pos && | |
1270 | ordered->file_offset < last_pos) { | |
1271 | btrfs_put_ordered_extent(ordered); | |
2ac55d41 JB |
1272 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, |
1273 | start_pos, last_pos - 1, | |
1274 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
1275 | for (i = 0; i < num_pages; i++) { |
1276 | unlock_page(pages[i]); | |
1277 | page_cache_release(pages[i]); | |
1278 | } | |
1279 | btrfs_wait_ordered_range(inode, start_pos, | |
1280 | last_pos - start_pos); | |
1281 | goto again; | |
1282 | } | |
1283 | if (ordered) | |
1284 | btrfs_put_ordered_extent(ordered); | |
1285 | ||
2ac55d41 | 1286 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos, |
32c00aff | 1287 | last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC | |
9e8a4a8b LB |
1288 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, |
1289 | 0, 0, &cached_state, GFP_NOFS); | |
2ac55d41 JB |
1290 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, |
1291 | start_pos, last_pos - 1, &cached_state, | |
1292 | GFP_NOFS); | |
0762704b | 1293 | } |
e6dcd2dc | 1294 | for (i = 0; i < num_pages; i++) { |
32c7f202 WF |
1295 | if (clear_page_dirty_for_io(pages[i])) |
1296 | account_page_redirty(pages[i]); | |
e6dcd2dc CM |
1297 | set_page_extent_mapped(pages[i]); |
1298 | WARN_ON(!PageLocked(pages[i])); | |
1299 | } | |
39279cc3 | 1300 | return 0; |
b1bf862e CM |
1301 | fail: |
1302 | while (faili >= 0) { | |
1303 | unlock_page(pages[faili]); | |
1304 | page_cache_release(pages[faili]); | |
1305 | faili--; | |
1306 | } | |
1307 | return err; | |
1308 | ||
39279cc3 CM |
1309 | } |
1310 | ||
d0215f3e JB |
1311 | static noinline ssize_t __btrfs_buffered_write(struct file *file, |
1312 | struct iov_iter *i, | |
1313 | loff_t pos) | |
4b46fce2 | 1314 | { |
11c65dcc JB |
1315 | struct inode *inode = fdentry(file)->d_inode; |
1316 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
11c65dcc | 1317 | struct page **pages = NULL; |
39279cc3 | 1318 | unsigned long first_index; |
d0215f3e JB |
1319 | size_t num_written = 0; |
1320 | int nrptrs; | |
c9149235 | 1321 | int ret = 0; |
b6316429 | 1322 | bool force_page_uptodate = false; |
4b46fce2 | 1323 | |
d0215f3e | 1324 | nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) / |
11c65dcc JB |
1325 | PAGE_CACHE_SIZE, PAGE_CACHE_SIZE / |
1326 | (sizeof(struct page *))); | |
142349f5 WF |
1327 | nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied); |
1328 | nrptrs = max(nrptrs, 8); | |
8c2383c3 | 1329 | pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL); |
d0215f3e JB |
1330 | if (!pages) |
1331 | return -ENOMEM; | |
ab93dbec | 1332 | |
39279cc3 | 1333 | first_index = pos >> PAGE_CACHE_SHIFT; |
39279cc3 | 1334 | |
d0215f3e | 1335 | while (iov_iter_count(i) > 0) { |
39279cc3 | 1336 | size_t offset = pos & (PAGE_CACHE_SIZE - 1); |
d0215f3e | 1337 | size_t write_bytes = min(iov_iter_count(i), |
11c65dcc | 1338 | nrptrs * (size_t)PAGE_CACHE_SIZE - |
8c2383c3 | 1339 | offset); |
3a90983d YZ |
1340 | size_t num_pages = (write_bytes + offset + |
1341 | PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
d0215f3e JB |
1342 | size_t dirty_pages; |
1343 | size_t copied; | |
39279cc3 | 1344 | |
8c2383c3 | 1345 | WARN_ON(num_pages > nrptrs); |
1832a6d5 | 1346 | |
914ee295 XZ |
1347 | /* |
1348 | * Fault pages before locking them in prepare_pages | |
1349 | * to avoid recursive lock | |
1350 | */ | |
d0215f3e | 1351 | if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) { |
914ee295 | 1352 | ret = -EFAULT; |
d0215f3e | 1353 | break; |
914ee295 XZ |
1354 | } |
1355 | ||
1356 | ret = btrfs_delalloc_reserve_space(inode, | |
1357 | num_pages << PAGE_CACHE_SHIFT); | |
1832a6d5 | 1358 | if (ret) |
d0215f3e | 1359 | break; |
1832a6d5 | 1360 | |
4a64001f JB |
1361 | /* |
1362 | * This is going to setup the pages array with the number of | |
1363 | * pages we want, so we don't really need to worry about the | |
1364 | * contents of pages from loop to loop | |
1365 | */ | |
39279cc3 | 1366 | ret = prepare_pages(root, file, pages, num_pages, |
b6316429 JB |
1367 | pos, first_index, write_bytes, |
1368 | force_page_uptodate); | |
6a63209f | 1369 | if (ret) { |
914ee295 XZ |
1370 | btrfs_delalloc_release_space(inode, |
1371 | num_pages << PAGE_CACHE_SHIFT); | |
d0215f3e | 1372 | break; |
6a63209f | 1373 | } |
39279cc3 | 1374 | |
914ee295 | 1375 | copied = btrfs_copy_from_user(pos, num_pages, |
d0215f3e | 1376 | write_bytes, pages, i); |
b1bf862e CM |
1377 | |
1378 | /* | |
1379 | * if we have trouble faulting in the pages, fall | |
1380 | * back to one page at a time | |
1381 | */ | |
1382 | if (copied < write_bytes) | |
1383 | nrptrs = 1; | |
1384 | ||
b6316429 JB |
1385 | if (copied == 0) { |
1386 | force_page_uptodate = true; | |
b1bf862e | 1387 | dirty_pages = 0; |
b6316429 JB |
1388 | } else { |
1389 | force_page_uptodate = false; | |
b1bf862e CM |
1390 | dirty_pages = (copied + offset + |
1391 | PAGE_CACHE_SIZE - 1) >> | |
1392 | PAGE_CACHE_SHIFT; | |
b6316429 | 1393 | } |
914ee295 | 1394 | |
d0215f3e JB |
1395 | /* |
1396 | * If we had a short copy we need to release the excess delaloc | |
1397 | * bytes we reserved. We need to increment outstanding_extents | |
1398 | * because btrfs_delalloc_release_space will decrement it, but | |
1399 | * we still have an outstanding extent for the chunk we actually | |
1400 | * managed to copy. | |
1401 | */ | |
914ee295 | 1402 | if (num_pages > dirty_pages) { |
9e0baf60 JB |
1403 | if (copied > 0) { |
1404 | spin_lock(&BTRFS_I(inode)->lock); | |
1405 | BTRFS_I(inode)->outstanding_extents++; | |
1406 | spin_unlock(&BTRFS_I(inode)->lock); | |
1407 | } | |
914ee295 XZ |
1408 | btrfs_delalloc_release_space(inode, |
1409 | (num_pages - dirty_pages) << | |
1410 | PAGE_CACHE_SHIFT); | |
1411 | } | |
1412 | ||
1413 | if (copied > 0) { | |
be1a12a0 JB |
1414 | ret = btrfs_dirty_pages(root, inode, pages, |
1415 | dirty_pages, pos, copied, | |
1416 | NULL); | |
d0215f3e JB |
1417 | if (ret) { |
1418 | btrfs_delalloc_release_space(inode, | |
1419 | dirty_pages << PAGE_CACHE_SHIFT); | |
1420 | btrfs_drop_pages(pages, num_pages); | |
1421 | break; | |
1422 | } | |
54aa1f4d | 1423 | } |
39279cc3 | 1424 | |
39279cc3 CM |
1425 | btrfs_drop_pages(pages, num_pages); |
1426 | ||
d0215f3e JB |
1427 | cond_resched(); |
1428 | ||
1429 | balance_dirty_pages_ratelimited_nr(inode->i_mapping, | |
1430 | dirty_pages); | |
1431 | if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1) | |
b53d3f5d | 1432 | btrfs_btree_balance_dirty(root); |
cb843a6f | 1433 | |
914ee295 XZ |
1434 | pos += copied; |
1435 | num_written += copied; | |
d0215f3e | 1436 | } |
39279cc3 | 1437 | |
d0215f3e JB |
1438 | kfree(pages); |
1439 | ||
1440 | return num_written ? num_written : ret; | |
1441 | } | |
1442 | ||
1443 | static ssize_t __btrfs_direct_write(struct kiocb *iocb, | |
1444 | const struct iovec *iov, | |
1445 | unsigned long nr_segs, loff_t pos, | |
1446 | loff_t *ppos, size_t count, size_t ocount) | |
1447 | { | |
1448 | struct file *file = iocb->ki_filp; | |
d0215f3e JB |
1449 | struct iov_iter i; |
1450 | ssize_t written; | |
1451 | ssize_t written_buffered; | |
1452 | loff_t endbyte; | |
1453 | int err; | |
1454 | ||
1455 | written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos, | |
1456 | count, ocount); | |
1457 | ||
d0215f3e JB |
1458 | if (written < 0 || written == count) |
1459 | return written; | |
1460 | ||
1461 | pos += written; | |
1462 | count -= written; | |
1463 | iov_iter_init(&i, iov, nr_segs, count, written); | |
1464 | written_buffered = __btrfs_buffered_write(file, &i, pos); | |
1465 | if (written_buffered < 0) { | |
1466 | err = written_buffered; | |
1467 | goto out; | |
39279cc3 | 1468 | } |
d0215f3e JB |
1469 | endbyte = pos + written_buffered - 1; |
1470 | err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte); | |
1471 | if (err) | |
1472 | goto out; | |
1473 | written += written_buffered; | |
1474 | *ppos = pos + written_buffered; | |
1475 | invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT, | |
1476 | endbyte >> PAGE_CACHE_SHIFT); | |
39279cc3 | 1477 | out: |
d0215f3e JB |
1478 | return written ? written : err; |
1479 | } | |
5b92ee72 | 1480 | |
6c760c07 JB |
1481 | static void update_time_for_write(struct inode *inode) |
1482 | { | |
1483 | struct timespec now; | |
1484 | ||
1485 | if (IS_NOCMTIME(inode)) | |
1486 | return; | |
1487 | ||
1488 | now = current_fs_time(inode->i_sb); | |
1489 | if (!timespec_equal(&inode->i_mtime, &now)) | |
1490 | inode->i_mtime = now; | |
1491 | ||
1492 | if (!timespec_equal(&inode->i_ctime, &now)) | |
1493 | inode->i_ctime = now; | |
1494 | ||
1495 | if (IS_I_VERSION(inode)) | |
1496 | inode_inc_iversion(inode); | |
1497 | } | |
1498 | ||
d0215f3e JB |
1499 | static ssize_t btrfs_file_aio_write(struct kiocb *iocb, |
1500 | const struct iovec *iov, | |
1501 | unsigned long nr_segs, loff_t pos) | |
1502 | { | |
1503 | struct file *file = iocb->ki_filp; | |
1504 | struct inode *inode = fdentry(file)->d_inode; | |
1505 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1506 | loff_t *ppos = &iocb->ki_pos; | |
0c1a98c8 | 1507 | u64 start_pos; |
d0215f3e JB |
1508 | ssize_t num_written = 0; |
1509 | ssize_t err = 0; | |
1510 | size_t count, ocount; | |
b812ce28 | 1511 | bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host); |
d0215f3e | 1512 | |
b2b5ef5c | 1513 | sb_start_write(inode->i_sb); |
d0215f3e JB |
1514 | |
1515 | mutex_lock(&inode->i_mutex); | |
1516 | ||
1517 | err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ); | |
1518 | if (err) { | |
1519 | mutex_unlock(&inode->i_mutex); | |
1520 | goto out; | |
1521 | } | |
1522 | count = ocount; | |
1523 | ||
1524 | current->backing_dev_info = inode->i_mapping->backing_dev_info; | |
1525 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); | |
1526 | if (err) { | |
1527 | mutex_unlock(&inode->i_mutex); | |
1528 | goto out; | |
1529 | } | |
1530 | ||
1531 | if (count == 0) { | |
1532 | mutex_unlock(&inode->i_mutex); | |
1533 | goto out; | |
1534 | } | |
1535 | ||
1536 | err = file_remove_suid(file); | |
1537 | if (err) { | |
1538 | mutex_unlock(&inode->i_mutex); | |
1539 | goto out; | |
1540 | } | |
1541 | ||
1542 | /* | |
1543 | * If BTRFS flips readonly due to some impossible error | |
1544 | * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR), | |
1545 | * although we have opened a file as writable, we have | |
1546 | * to stop this write operation to ensure FS consistency. | |
1547 | */ | |
1548 | if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) { | |
1549 | mutex_unlock(&inode->i_mutex); | |
1550 | err = -EROFS; | |
1551 | goto out; | |
1552 | } | |
1553 | ||
6c760c07 JB |
1554 | /* |
1555 | * We reserve space for updating the inode when we reserve space for the | |
1556 | * extent we are going to write, so we will enospc out there. We don't | |
1557 | * need to start yet another transaction to update the inode as we will | |
1558 | * update the inode when we finish writing whatever data we write. | |
1559 | */ | |
1560 | update_time_for_write(inode); | |
d0215f3e | 1561 | |
0c1a98c8 MX |
1562 | start_pos = round_down(pos, root->sectorsize); |
1563 | if (start_pos > i_size_read(inode)) { | |
1564 | err = btrfs_cont_expand(inode, i_size_read(inode), start_pos); | |
1565 | if (err) { | |
1566 | mutex_unlock(&inode->i_mutex); | |
1567 | goto out; | |
1568 | } | |
1569 | } | |
1570 | ||
b812ce28 JB |
1571 | if (sync) |
1572 | atomic_inc(&BTRFS_I(inode)->sync_writers); | |
1573 | ||
d0215f3e JB |
1574 | if (unlikely(file->f_flags & O_DIRECT)) { |
1575 | num_written = __btrfs_direct_write(iocb, iov, nr_segs, | |
1576 | pos, ppos, count, ocount); | |
1577 | } else { | |
1578 | struct iov_iter i; | |
1579 | ||
1580 | iov_iter_init(&i, iov, nr_segs, count, num_written); | |
1581 | ||
1582 | num_written = __btrfs_buffered_write(file, &i, pos); | |
1583 | if (num_written > 0) | |
1584 | *ppos = pos + num_written; | |
1585 | } | |
1586 | ||
1587 | mutex_unlock(&inode->i_mutex); | |
2ff3e9b6 | 1588 | |
5a3f23d5 CM |
1589 | /* |
1590 | * we want to make sure fsync finds this change | |
1591 | * but we haven't joined a transaction running right now. | |
1592 | * | |
1593 | * Later on, someone is sure to update the inode and get the | |
1594 | * real transid recorded. | |
1595 | * | |
1596 | * We set last_trans now to the fs_info generation + 1, | |
1597 | * this will either be one more than the running transaction | |
1598 | * or the generation used for the next transaction if there isn't | |
1599 | * one running right now. | |
6c760c07 JB |
1600 | * |
1601 | * We also have to set last_sub_trans to the current log transid, | |
1602 | * otherwise subsequent syncs to a file that's been synced in this | |
1603 | * transaction will appear to have already occured. | |
5a3f23d5 CM |
1604 | */ |
1605 | BTRFS_I(inode)->last_trans = root->fs_info->generation + 1; | |
6c760c07 | 1606 | BTRFS_I(inode)->last_sub_trans = root->log_transid; |
d0215f3e JB |
1607 | if (num_written > 0 || num_written == -EIOCBQUEUED) { |
1608 | err = generic_write_sync(file, pos, num_written); | |
1609 | if (err < 0 && num_written > 0) | |
2ff3e9b6 CM |
1610 | num_written = err; |
1611 | } | |
0a3404dc | 1612 | |
b812ce28 JB |
1613 | if (sync) |
1614 | atomic_dec(&BTRFS_I(inode)->sync_writers); | |
0a3404dc | 1615 | out: |
b2b5ef5c | 1616 | sb_end_write(inode->i_sb); |
39279cc3 | 1617 | current->backing_dev_info = NULL; |
39279cc3 CM |
1618 | return num_written ? num_written : err; |
1619 | } | |
1620 | ||
d397712b | 1621 | int btrfs_release_file(struct inode *inode, struct file *filp) |
e1b81e67 | 1622 | { |
5a3f23d5 CM |
1623 | /* |
1624 | * ordered_data_close is set by settattr when we are about to truncate | |
1625 | * a file from a non-zero size to a zero size. This tries to | |
1626 | * flush down new bytes that may have been written if the | |
1627 | * application were using truncate to replace a file in place. | |
1628 | */ | |
72ac3c0d JB |
1629 | if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
1630 | &BTRFS_I(inode)->runtime_flags)) { | |
5a3f23d5 CM |
1631 | btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode); |
1632 | if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) | |
1633 | filemap_flush(inode->i_mapping); | |
1634 | } | |
6bf13c0c SW |
1635 | if (filp->private_data) |
1636 | btrfs_ioctl_trans_end(filp); | |
e1b81e67 M |
1637 | return 0; |
1638 | } | |
1639 | ||
d352ac68 CM |
1640 | /* |
1641 | * fsync call for both files and directories. This logs the inode into | |
1642 | * the tree log instead of forcing full commits whenever possible. | |
1643 | * | |
1644 | * It needs to call filemap_fdatawait so that all ordered extent updates are | |
1645 | * in the metadata btree are up to date for copying to the log. | |
1646 | * | |
1647 | * It drops the inode mutex before doing the tree log commit. This is an | |
1648 | * important optimization for directories because holding the mutex prevents | |
1649 | * new operations on the dir while we write to disk. | |
1650 | */ | |
02c24a82 | 1651 | int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync) |
39279cc3 | 1652 | { |
7ea80859 | 1653 | struct dentry *dentry = file->f_path.dentry; |
39279cc3 CM |
1654 | struct inode *inode = dentry->d_inode; |
1655 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
15ee9bc7 | 1656 | int ret = 0; |
39279cc3 | 1657 | struct btrfs_trans_handle *trans; |
2ab28f32 | 1658 | bool full_sync = 0; |
39279cc3 | 1659 | |
1abe9b8a | 1660 | trace_btrfs_sync_file(file, datasync); |
257c62e1 | 1661 | |
90abccf2 MX |
1662 | /* |
1663 | * We write the dirty pages in the range and wait until they complete | |
1664 | * out of the ->i_mutex. If so, we can flush the dirty pages by | |
2ab28f32 JB |
1665 | * multi-task, and make the performance up. See |
1666 | * btrfs_wait_ordered_range for an explanation of the ASYNC check. | |
90abccf2 | 1667 | */ |
b812ce28 | 1668 | atomic_inc(&BTRFS_I(inode)->sync_writers); |
2ab28f32 JB |
1669 | ret = filemap_fdatawrite_range(inode->i_mapping, start, end); |
1670 | if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
1671 | &BTRFS_I(inode)->runtime_flags)) | |
1672 | ret = filemap_fdatawrite_range(inode->i_mapping, start, end); | |
b812ce28 | 1673 | atomic_dec(&BTRFS_I(inode)->sync_writers); |
90abccf2 MX |
1674 | if (ret) |
1675 | return ret; | |
1676 | ||
02c24a82 JB |
1677 | mutex_lock(&inode->i_mutex); |
1678 | ||
0885ef5b | 1679 | /* |
90abccf2 MX |
1680 | * We flush the dirty pages again to avoid some dirty pages in the |
1681 | * range being left. | |
0885ef5b | 1682 | */ |
2ecb7923 | 1683 | atomic_inc(&root->log_batch); |
2ab28f32 JB |
1684 | full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
1685 | &BTRFS_I(inode)->runtime_flags); | |
1686 | if (full_sync) | |
1687 | btrfs_wait_ordered_range(inode, start, end - start + 1); | |
2ecb7923 | 1688 | atomic_inc(&root->log_batch); |
257c62e1 | 1689 | |
39279cc3 | 1690 | /* |
15ee9bc7 JB |
1691 | * check the transaction that last modified this inode |
1692 | * and see if its already been committed | |
39279cc3 | 1693 | */ |
02c24a82 JB |
1694 | if (!BTRFS_I(inode)->last_trans) { |
1695 | mutex_unlock(&inode->i_mutex); | |
15ee9bc7 | 1696 | goto out; |
02c24a82 | 1697 | } |
a2135011 | 1698 | |
257c62e1 CM |
1699 | /* |
1700 | * if the last transaction that changed this file was before | |
1701 | * the current transaction, we can bail out now without any | |
1702 | * syncing | |
1703 | */ | |
a4abeea4 | 1704 | smp_mb(); |
22ee6985 JB |
1705 | if (btrfs_inode_in_log(inode, root->fs_info->generation) || |
1706 | BTRFS_I(inode)->last_trans <= | |
15ee9bc7 JB |
1707 | root->fs_info->last_trans_committed) { |
1708 | BTRFS_I(inode)->last_trans = 0; | |
5dc562c5 JB |
1709 | |
1710 | /* | |
1711 | * We'v had everything committed since the last time we were | |
1712 | * modified so clear this flag in case it was set for whatever | |
1713 | * reason, it's no longer relevant. | |
1714 | */ | |
1715 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
1716 | &BTRFS_I(inode)->runtime_flags); | |
02c24a82 | 1717 | mutex_unlock(&inode->i_mutex); |
15ee9bc7 JB |
1718 | goto out; |
1719 | } | |
15ee9bc7 JB |
1720 | |
1721 | /* | |
a52d9a80 CM |
1722 | * ok we haven't committed the transaction yet, lets do a commit |
1723 | */ | |
6f902af4 | 1724 | if (file->private_data) |
6bf13c0c SW |
1725 | btrfs_ioctl_trans_end(file); |
1726 | ||
a22285a6 YZ |
1727 | trans = btrfs_start_transaction(root, 0); |
1728 | if (IS_ERR(trans)) { | |
1729 | ret = PTR_ERR(trans); | |
02c24a82 | 1730 | mutex_unlock(&inode->i_mutex); |
39279cc3 CM |
1731 | goto out; |
1732 | } | |
e02119d5 | 1733 | |
2cfbd50b | 1734 | ret = btrfs_log_dentry_safe(trans, root, dentry); |
02c24a82 JB |
1735 | if (ret < 0) { |
1736 | mutex_unlock(&inode->i_mutex); | |
e02119d5 | 1737 | goto out; |
02c24a82 | 1738 | } |
49eb7e46 CM |
1739 | |
1740 | /* we've logged all the items and now have a consistent | |
1741 | * version of the file in the log. It is possible that | |
1742 | * someone will come in and modify the file, but that's | |
1743 | * fine because the log is consistent on disk, and we | |
1744 | * have references to all of the file's extents | |
1745 | * | |
1746 | * It is possible that someone will come in and log the | |
1747 | * file again, but that will end up using the synchronization | |
1748 | * inside btrfs_sync_log to keep things safe. | |
1749 | */ | |
02c24a82 | 1750 | mutex_unlock(&inode->i_mutex); |
49eb7e46 | 1751 | |
257c62e1 CM |
1752 | if (ret != BTRFS_NO_LOG_SYNC) { |
1753 | if (ret > 0) { | |
2ab28f32 JB |
1754 | /* |
1755 | * If we didn't already wait for ordered extents we need | |
1756 | * to do that now. | |
1757 | */ | |
1758 | if (!full_sync) | |
1759 | btrfs_wait_ordered_range(inode, start, | |
1760 | end - start + 1); | |
12fcfd22 | 1761 | ret = btrfs_commit_transaction(trans, root); |
257c62e1 CM |
1762 | } else { |
1763 | ret = btrfs_sync_log(trans, root); | |
2ab28f32 | 1764 | if (ret == 0) { |
257c62e1 | 1765 | ret = btrfs_end_transaction(trans, root); |
2ab28f32 JB |
1766 | } else { |
1767 | if (!full_sync) | |
1768 | btrfs_wait_ordered_range(inode, start, | |
1769 | end - | |
1770 | start + 1); | |
257c62e1 | 1771 | ret = btrfs_commit_transaction(trans, root); |
2ab28f32 | 1772 | } |
257c62e1 CM |
1773 | } |
1774 | } else { | |
1775 | ret = btrfs_end_transaction(trans, root); | |
e02119d5 | 1776 | } |
39279cc3 | 1777 | out: |
014e4ac4 | 1778 | return ret > 0 ? -EIO : ret; |
39279cc3 CM |
1779 | } |
1780 | ||
f0f37e2f | 1781 | static const struct vm_operations_struct btrfs_file_vm_ops = { |
92fee66d | 1782 | .fault = filemap_fault, |
9ebefb18 | 1783 | .page_mkwrite = btrfs_page_mkwrite, |
0b173bc4 | 1784 | .remap_pages = generic_file_remap_pages, |
9ebefb18 CM |
1785 | }; |
1786 | ||
1787 | static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma) | |
1788 | { | |
058a457e MX |
1789 | struct address_space *mapping = filp->f_mapping; |
1790 | ||
1791 | if (!mapping->a_ops->readpage) | |
1792 | return -ENOEXEC; | |
1793 | ||
9ebefb18 | 1794 | file_accessed(filp); |
058a457e | 1795 | vma->vm_ops = &btrfs_file_vm_ops; |
058a457e | 1796 | |
9ebefb18 CM |
1797 | return 0; |
1798 | } | |
1799 | ||
2aaa6655 JB |
1800 | static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf, |
1801 | int slot, u64 start, u64 end) | |
1802 | { | |
1803 | struct btrfs_file_extent_item *fi; | |
1804 | struct btrfs_key key; | |
1805 | ||
1806 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) | |
1807 | return 0; | |
1808 | ||
1809 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
1810 | if (key.objectid != btrfs_ino(inode) || | |
1811 | key.type != BTRFS_EXTENT_DATA_KEY) | |
1812 | return 0; | |
1813 | ||
1814 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
1815 | ||
1816 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
1817 | return 0; | |
1818 | ||
1819 | if (btrfs_file_extent_disk_bytenr(leaf, fi)) | |
1820 | return 0; | |
1821 | ||
1822 | if (key.offset == end) | |
1823 | return 1; | |
1824 | if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start) | |
1825 | return 1; | |
1826 | return 0; | |
1827 | } | |
1828 | ||
1829 | static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode, | |
1830 | struct btrfs_path *path, u64 offset, u64 end) | |
1831 | { | |
1832 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1833 | struct extent_buffer *leaf; | |
1834 | struct btrfs_file_extent_item *fi; | |
1835 | struct extent_map *hole_em; | |
1836 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
1837 | struct btrfs_key key; | |
1838 | int ret; | |
1839 | ||
1840 | key.objectid = btrfs_ino(inode); | |
1841 | key.type = BTRFS_EXTENT_DATA_KEY; | |
1842 | key.offset = offset; | |
1843 | ||
1844 | ||
1845 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
1846 | if (ret < 0) | |
1847 | return ret; | |
1848 | BUG_ON(!ret); | |
1849 | ||
1850 | leaf = path->nodes[0]; | |
1851 | if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) { | |
1852 | u64 num_bytes; | |
1853 | ||
1854 | path->slots[0]--; | |
1855 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1856 | struct btrfs_file_extent_item); | |
1857 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + | |
1858 | end - offset; | |
1859 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1860 | btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes); | |
1861 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1862 | btrfs_mark_buffer_dirty(leaf); | |
1863 | goto out; | |
1864 | } | |
1865 | ||
1866 | if (hole_mergeable(inode, leaf, path->slots[0]+1, offset, end)) { | |
1867 | u64 num_bytes; | |
1868 | ||
1869 | path->slots[0]++; | |
1870 | key.offset = offset; | |
1871 | btrfs_set_item_key_safe(trans, root, path, &key); | |
1872 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1873 | struct btrfs_file_extent_item); | |
1874 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end - | |
1875 | offset; | |
1876 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1877 | btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes); | |
1878 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1879 | btrfs_mark_buffer_dirty(leaf); | |
1880 | goto out; | |
1881 | } | |
1882 | btrfs_release_path(path); | |
1883 | ||
1884 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset, | |
1885 | 0, 0, end - offset, 0, end - offset, | |
1886 | 0, 0, 0); | |
1887 | if (ret) | |
1888 | return ret; | |
1889 | ||
1890 | out: | |
1891 | btrfs_release_path(path); | |
1892 | ||
1893 | hole_em = alloc_extent_map(); | |
1894 | if (!hole_em) { | |
1895 | btrfs_drop_extent_cache(inode, offset, end - 1, 0); | |
1896 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
1897 | &BTRFS_I(inode)->runtime_flags); | |
1898 | } else { | |
1899 | hole_em->start = offset; | |
1900 | hole_em->len = end - offset; | |
1901 | hole_em->orig_start = offset; | |
1902 | ||
1903 | hole_em->block_start = EXTENT_MAP_HOLE; | |
1904 | hole_em->block_len = 0; | |
b4939680 | 1905 | hole_em->orig_block_len = 0; |
2aaa6655 JB |
1906 | hole_em->bdev = root->fs_info->fs_devices->latest_bdev; |
1907 | hole_em->compress_type = BTRFS_COMPRESS_NONE; | |
1908 | hole_em->generation = trans->transid; | |
1909 | ||
1910 | do { | |
1911 | btrfs_drop_extent_cache(inode, offset, end - 1, 0); | |
1912 | write_lock(&em_tree->lock); | |
1913 | ret = add_extent_mapping(em_tree, hole_em); | |
1914 | if (!ret) | |
1915 | list_move(&hole_em->list, | |
1916 | &em_tree->modified_extents); | |
1917 | write_unlock(&em_tree->lock); | |
1918 | } while (ret == -EEXIST); | |
1919 | free_extent_map(hole_em); | |
1920 | if (ret) | |
1921 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
1922 | &BTRFS_I(inode)->runtime_flags); | |
1923 | } | |
1924 | ||
1925 | return 0; | |
1926 | } | |
1927 | ||
1928 | static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len) | |
1929 | { | |
1930 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1931 | struct extent_state *cached_state = NULL; | |
1932 | struct btrfs_path *path; | |
1933 | struct btrfs_block_rsv *rsv; | |
1934 | struct btrfs_trans_handle *trans; | |
0061280d MX |
1935 | u64 lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize); |
1936 | u64 lockend = round_down(offset + len, | |
1937 | BTRFS_I(inode)->root->sectorsize) - 1; | |
2aaa6655 JB |
1938 | u64 cur_offset = lockstart; |
1939 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); | |
1940 | u64 drop_end; | |
2aaa6655 JB |
1941 | int ret = 0; |
1942 | int err = 0; | |
6347b3c4 MX |
1943 | bool same_page = ((offset >> PAGE_CACHE_SHIFT) == |
1944 | ((offset + len - 1) >> PAGE_CACHE_SHIFT)); | |
2aaa6655 JB |
1945 | |
1946 | btrfs_wait_ordered_range(inode, offset, len); | |
1947 | ||
1948 | mutex_lock(&inode->i_mutex); | |
7426cc04 MX |
1949 | /* |
1950 | * We needn't truncate any page which is beyond the end of the file | |
1951 | * because we are sure there is no data there. | |
1952 | */ | |
2aaa6655 JB |
1953 | /* |
1954 | * Only do this if we are in the same page and we aren't doing the | |
1955 | * entire page. | |
1956 | */ | |
1957 | if (same_page && len < PAGE_CACHE_SIZE) { | |
7426cc04 MX |
1958 | if (offset < round_up(inode->i_size, PAGE_CACHE_SIZE)) |
1959 | ret = btrfs_truncate_page(inode, offset, len, 0); | |
2aaa6655 JB |
1960 | mutex_unlock(&inode->i_mutex); |
1961 | return ret; | |
1962 | } | |
1963 | ||
1964 | /* zero back part of the first page */ | |
7426cc04 MX |
1965 | if (offset < round_up(inode->i_size, PAGE_CACHE_SIZE)) { |
1966 | ret = btrfs_truncate_page(inode, offset, 0, 0); | |
1967 | if (ret) { | |
1968 | mutex_unlock(&inode->i_mutex); | |
1969 | return ret; | |
1970 | } | |
2aaa6655 JB |
1971 | } |
1972 | ||
1973 | /* zero the front end of the last page */ | |
0061280d MX |
1974 | if (offset + len < round_up(inode->i_size, PAGE_CACHE_SIZE)) { |
1975 | ret = btrfs_truncate_page(inode, offset + len, 0, 1); | |
1976 | if (ret) { | |
1977 | mutex_unlock(&inode->i_mutex); | |
1978 | return ret; | |
1979 | } | |
2aaa6655 JB |
1980 | } |
1981 | ||
1982 | if (lockend < lockstart) { | |
1983 | mutex_unlock(&inode->i_mutex); | |
1984 | return 0; | |
1985 | } | |
1986 | ||
1987 | while (1) { | |
1988 | struct btrfs_ordered_extent *ordered; | |
1989 | ||
1990 | truncate_pagecache_range(inode, lockstart, lockend); | |
1991 | ||
1992 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
1993 | 0, &cached_state); | |
1994 | ordered = btrfs_lookup_first_ordered_extent(inode, lockend); | |
1995 | ||
1996 | /* | |
1997 | * We need to make sure we have no ordered extents in this range | |
1998 | * and nobody raced in and read a page in this range, if we did | |
1999 | * we need to try again. | |
2000 | */ | |
2001 | if ((!ordered || | |
2002 | (ordered->file_offset + ordered->len < lockstart || | |
2003 | ordered->file_offset > lockend)) && | |
2004 | !test_range_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
2005 | lockend, EXTENT_UPTODATE, 0, | |
2006 | cached_state)) { | |
2007 | if (ordered) | |
2008 | btrfs_put_ordered_extent(ordered); | |
2009 | break; | |
2010 | } | |
2011 | if (ordered) | |
2012 | btrfs_put_ordered_extent(ordered); | |
2013 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, | |
2014 | lockend, &cached_state, GFP_NOFS); | |
2015 | btrfs_wait_ordered_range(inode, lockstart, | |
2016 | lockend - lockstart + 1); | |
2017 | } | |
2018 | ||
2019 | path = btrfs_alloc_path(); | |
2020 | if (!path) { | |
2021 | ret = -ENOMEM; | |
2022 | goto out; | |
2023 | } | |
2024 | ||
66d8f3dd | 2025 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
2aaa6655 JB |
2026 | if (!rsv) { |
2027 | ret = -ENOMEM; | |
2028 | goto out_free; | |
2029 | } | |
2030 | rsv->size = btrfs_calc_trunc_metadata_size(root, 1); | |
2031 | rsv->failfast = 1; | |
2032 | ||
2033 | /* | |
2034 | * 1 - update the inode | |
2035 | * 1 - removing the extents in the range | |
2036 | * 1 - adding the hole extent | |
2037 | */ | |
2038 | trans = btrfs_start_transaction(root, 3); | |
2039 | if (IS_ERR(trans)) { | |
2040 | err = PTR_ERR(trans); | |
2041 | goto out_free; | |
2042 | } | |
2043 | ||
2044 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv, | |
2045 | min_size); | |
2046 | BUG_ON(ret); | |
2047 | trans->block_rsv = rsv; | |
2048 | ||
2049 | while (cur_offset < lockend) { | |
2050 | ret = __btrfs_drop_extents(trans, root, inode, path, | |
2051 | cur_offset, lockend + 1, | |
2052 | &drop_end, 1); | |
2053 | if (ret != -ENOSPC) | |
2054 | break; | |
2055 | ||
2056 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
2057 | ||
2058 | ret = fill_holes(trans, inode, path, cur_offset, drop_end); | |
2059 | if (ret) { | |
2060 | err = ret; | |
2061 | break; | |
2062 | } | |
2063 | ||
2064 | cur_offset = drop_end; | |
2065 | ||
2066 | ret = btrfs_update_inode(trans, root, inode); | |
2067 | if (ret) { | |
2068 | err = ret; | |
2069 | break; | |
2070 | } | |
2071 | ||
2aaa6655 | 2072 | btrfs_end_transaction(trans, root); |
b53d3f5d | 2073 | btrfs_btree_balance_dirty(root); |
2aaa6655 JB |
2074 | |
2075 | trans = btrfs_start_transaction(root, 3); | |
2076 | if (IS_ERR(trans)) { | |
2077 | ret = PTR_ERR(trans); | |
2078 | trans = NULL; | |
2079 | break; | |
2080 | } | |
2081 | ||
2082 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, | |
2083 | rsv, min_size); | |
2084 | BUG_ON(ret); /* shouldn't happen */ | |
2085 | trans->block_rsv = rsv; | |
2086 | } | |
2087 | ||
2088 | if (ret) { | |
2089 | err = ret; | |
2090 | goto out_trans; | |
2091 | } | |
2092 | ||
2093 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
2094 | ret = fill_holes(trans, inode, path, cur_offset, drop_end); | |
2095 | if (ret) { | |
2096 | err = ret; | |
2097 | goto out_trans; | |
2098 | } | |
2099 | ||
2100 | out_trans: | |
2101 | if (!trans) | |
2102 | goto out_free; | |
2103 | ||
e1f5790e TI |
2104 | inode_inc_iversion(inode); |
2105 | inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
2106 | ||
2aaa6655 JB |
2107 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
2108 | ret = btrfs_update_inode(trans, root, inode); | |
2aaa6655 | 2109 | btrfs_end_transaction(trans, root); |
b53d3f5d | 2110 | btrfs_btree_balance_dirty(root); |
2aaa6655 JB |
2111 | out_free: |
2112 | btrfs_free_path(path); | |
2113 | btrfs_free_block_rsv(root, rsv); | |
2114 | out: | |
2115 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
2116 | &cached_state, GFP_NOFS); | |
2117 | mutex_unlock(&inode->i_mutex); | |
2118 | if (ret && !err) | |
2119 | err = ret; | |
2120 | return err; | |
2121 | } | |
2122 | ||
2fe17c10 CH |
2123 | static long btrfs_fallocate(struct file *file, int mode, |
2124 | loff_t offset, loff_t len) | |
2125 | { | |
2126 | struct inode *inode = file->f_path.dentry->d_inode; | |
2127 | struct extent_state *cached_state = NULL; | |
2128 | u64 cur_offset; | |
2129 | u64 last_byte; | |
2130 | u64 alloc_start; | |
2131 | u64 alloc_end; | |
2132 | u64 alloc_hint = 0; | |
2133 | u64 locked_end; | |
2fe17c10 | 2134 | struct extent_map *em; |
797f4277 | 2135 | int blocksize = BTRFS_I(inode)->root->sectorsize; |
2fe17c10 CH |
2136 | int ret; |
2137 | ||
797f4277 MX |
2138 | alloc_start = round_down(offset, blocksize); |
2139 | alloc_end = round_up(offset + len, blocksize); | |
2fe17c10 | 2140 | |
2aaa6655 JB |
2141 | /* Make sure we aren't being give some crap mode */ |
2142 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
2fe17c10 CH |
2143 | return -EOPNOTSUPP; |
2144 | ||
2aaa6655 JB |
2145 | if (mode & FALLOC_FL_PUNCH_HOLE) |
2146 | return btrfs_punch_hole(inode, offset, len); | |
2147 | ||
d98456fc CM |
2148 | /* |
2149 | * Make sure we have enough space before we do the | |
2150 | * allocation. | |
2151 | */ | |
0ff6fabd | 2152 | ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start); |
d98456fc CM |
2153 | if (ret) |
2154 | return ret; | |
2155 | ||
2fe17c10 CH |
2156 | /* |
2157 | * wait for ordered IO before we have any locks. We'll loop again | |
2158 | * below with the locks held. | |
2159 | */ | |
2160 | btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start); | |
2161 | ||
2162 | mutex_lock(&inode->i_mutex); | |
2163 | ret = inode_newsize_ok(inode, alloc_end); | |
2164 | if (ret) | |
2165 | goto out; | |
2166 | ||
2167 | if (alloc_start > inode->i_size) { | |
a41ad394 JB |
2168 | ret = btrfs_cont_expand(inode, i_size_read(inode), |
2169 | alloc_start); | |
2fe17c10 CH |
2170 | if (ret) |
2171 | goto out; | |
2172 | } | |
2173 | ||
2fe17c10 CH |
2174 | locked_end = alloc_end - 1; |
2175 | while (1) { | |
2176 | struct btrfs_ordered_extent *ordered; | |
2177 | ||
2178 | /* the extent lock is ordered inside the running | |
2179 | * transaction | |
2180 | */ | |
2181 | lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start, | |
d0082371 | 2182 | locked_end, 0, &cached_state); |
2fe17c10 CH |
2183 | ordered = btrfs_lookup_first_ordered_extent(inode, |
2184 | alloc_end - 1); | |
2185 | if (ordered && | |
2186 | ordered->file_offset + ordered->len > alloc_start && | |
2187 | ordered->file_offset < alloc_end) { | |
2188 | btrfs_put_ordered_extent(ordered); | |
2189 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, | |
2190 | alloc_start, locked_end, | |
2191 | &cached_state, GFP_NOFS); | |
2192 | /* | |
2193 | * we can't wait on the range with the transaction | |
2194 | * running or with the extent lock held | |
2195 | */ | |
2196 | btrfs_wait_ordered_range(inode, alloc_start, | |
2197 | alloc_end - alloc_start); | |
2198 | } else { | |
2199 | if (ordered) | |
2200 | btrfs_put_ordered_extent(ordered); | |
2201 | break; | |
2202 | } | |
2203 | } | |
2204 | ||
2205 | cur_offset = alloc_start; | |
2206 | while (1) { | |
f1e490a7 JB |
2207 | u64 actual_end; |
2208 | ||
2fe17c10 CH |
2209 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, |
2210 | alloc_end - cur_offset, 0); | |
79787eaa JM |
2211 | if (IS_ERR_OR_NULL(em)) { |
2212 | if (!em) | |
2213 | ret = -ENOMEM; | |
2214 | else | |
2215 | ret = PTR_ERR(em); | |
2216 | break; | |
2217 | } | |
2fe17c10 | 2218 | last_byte = min(extent_map_end(em), alloc_end); |
f1e490a7 | 2219 | actual_end = min_t(u64, extent_map_end(em), offset + len); |
797f4277 | 2220 | last_byte = ALIGN(last_byte, blocksize); |
f1e490a7 | 2221 | |
2fe17c10 CH |
2222 | if (em->block_start == EXTENT_MAP_HOLE || |
2223 | (cur_offset >= inode->i_size && | |
2224 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
2225 | ret = btrfs_prealloc_file_range(inode, mode, cur_offset, | |
2226 | last_byte - cur_offset, | |
2227 | 1 << inode->i_blkbits, | |
2228 | offset + len, | |
2229 | &alloc_hint); | |
1b9c332b | 2230 | |
2fe17c10 CH |
2231 | if (ret < 0) { |
2232 | free_extent_map(em); | |
2233 | break; | |
2234 | } | |
f1e490a7 JB |
2235 | } else if (actual_end > inode->i_size && |
2236 | !(mode & FALLOC_FL_KEEP_SIZE)) { | |
2237 | /* | |
2238 | * We didn't need to allocate any more space, but we | |
2239 | * still extended the size of the file so we need to | |
2240 | * update i_size. | |
2241 | */ | |
2242 | inode->i_ctime = CURRENT_TIME; | |
2243 | i_size_write(inode, actual_end); | |
2244 | btrfs_ordered_update_i_size(inode, actual_end, NULL); | |
2fe17c10 CH |
2245 | } |
2246 | free_extent_map(em); | |
2247 | ||
2248 | cur_offset = last_byte; | |
2249 | if (cur_offset >= alloc_end) { | |
2250 | ret = 0; | |
2251 | break; | |
2252 | } | |
2253 | } | |
2254 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, | |
2255 | &cached_state, GFP_NOFS); | |
2fe17c10 CH |
2256 | out: |
2257 | mutex_unlock(&inode->i_mutex); | |
d98456fc | 2258 | /* Let go of our reservation. */ |
0ff6fabd | 2259 | btrfs_free_reserved_data_space(inode, alloc_end - alloc_start); |
2fe17c10 CH |
2260 | return ret; |
2261 | } | |
2262 | ||
b2675157 JB |
2263 | static int find_desired_extent(struct inode *inode, loff_t *offset, int origin) |
2264 | { | |
2265 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2266 | struct extent_map *em; | |
2267 | struct extent_state *cached_state = NULL; | |
2268 | u64 lockstart = *offset; | |
2269 | u64 lockend = i_size_read(inode); | |
2270 | u64 start = *offset; | |
2271 | u64 orig_start = *offset; | |
2272 | u64 len = i_size_read(inode); | |
2273 | u64 last_end = 0; | |
2274 | int ret = 0; | |
2275 | ||
2276 | lockend = max_t(u64, root->sectorsize, lockend); | |
2277 | if (lockend <= lockstart) | |
2278 | lockend = lockstart + root->sectorsize; | |
2279 | ||
1214b53f | 2280 | lockend--; |
b2675157 JB |
2281 | len = lockend - lockstart + 1; |
2282 | ||
2283 | len = max_t(u64, len, root->sectorsize); | |
2284 | if (inode->i_size == 0) | |
2285 | return -ENXIO; | |
2286 | ||
2287 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0, | |
d0082371 | 2288 | &cached_state); |
b2675157 JB |
2289 | |
2290 | /* | |
2291 | * Delalloc is such a pain. If we have a hole and we have pending | |
2292 | * delalloc for a portion of the hole we will get back a hole that | |
2293 | * exists for the entire range since it hasn't been actually written | |
2294 | * yet. So to take care of this case we need to look for an extent just | |
2295 | * before the position we want in case there is outstanding delalloc | |
2296 | * going on here. | |
2297 | */ | |
2298 | if (origin == SEEK_HOLE && start != 0) { | |
2299 | if (start <= root->sectorsize) | |
2300 | em = btrfs_get_extent_fiemap(inode, NULL, 0, 0, | |
2301 | root->sectorsize, 0); | |
2302 | else | |
2303 | em = btrfs_get_extent_fiemap(inode, NULL, 0, | |
2304 | start - root->sectorsize, | |
2305 | root->sectorsize, 0); | |
2306 | if (IS_ERR(em)) { | |
6af021d8 | 2307 | ret = PTR_ERR(em); |
b2675157 JB |
2308 | goto out; |
2309 | } | |
2310 | last_end = em->start + em->len; | |
2311 | if (em->block_start == EXTENT_MAP_DELALLOC) | |
2312 | last_end = min_t(u64, last_end, inode->i_size); | |
2313 | free_extent_map(em); | |
2314 | } | |
2315 | ||
2316 | while (1) { | |
2317 | em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0); | |
2318 | if (IS_ERR(em)) { | |
6af021d8 | 2319 | ret = PTR_ERR(em); |
b2675157 JB |
2320 | break; |
2321 | } | |
2322 | ||
2323 | if (em->block_start == EXTENT_MAP_HOLE) { | |
2324 | if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) { | |
2325 | if (last_end <= orig_start) { | |
2326 | free_extent_map(em); | |
2327 | ret = -ENXIO; | |
2328 | break; | |
2329 | } | |
2330 | } | |
2331 | ||
2332 | if (origin == SEEK_HOLE) { | |
2333 | *offset = start; | |
2334 | free_extent_map(em); | |
2335 | break; | |
2336 | } | |
2337 | } else { | |
2338 | if (origin == SEEK_DATA) { | |
2339 | if (em->block_start == EXTENT_MAP_DELALLOC) { | |
2340 | if (start >= inode->i_size) { | |
2341 | free_extent_map(em); | |
2342 | ret = -ENXIO; | |
2343 | break; | |
2344 | } | |
2345 | } | |
2346 | ||
f9e4fb53 LB |
2347 | if (!test_bit(EXTENT_FLAG_PREALLOC, |
2348 | &em->flags)) { | |
2349 | *offset = start; | |
2350 | free_extent_map(em); | |
2351 | break; | |
2352 | } | |
b2675157 JB |
2353 | } |
2354 | } | |
2355 | ||
2356 | start = em->start + em->len; | |
2357 | last_end = em->start + em->len; | |
2358 | ||
2359 | if (em->block_start == EXTENT_MAP_DELALLOC) | |
2360 | last_end = min_t(u64, last_end, inode->i_size); | |
2361 | ||
2362 | if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) { | |
2363 | free_extent_map(em); | |
2364 | ret = -ENXIO; | |
2365 | break; | |
2366 | } | |
2367 | free_extent_map(em); | |
2368 | cond_resched(); | |
2369 | } | |
2370 | if (!ret) | |
2371 | *offset = min(*offset, inode->i_size); | |
2372 | out: | |
2373 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
2374 | &cached_state, GFP_NOFS); | |
2375 | return ret; | |
2376 | } | |
2377 | ||
2378 | static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int origin) | |
2379 | { | |
2380 | struct inode *inode = file->f_mapping->host; | |
2381 | int ret; | |
2382 | ||
2383 | mutex_lock(&inode->i_mutex); | |
2384 | switch (origin) { | |
2385 | case SEEK_END: | |
2386 | case SEEK_CUR: | |
ef3d0fd2 | 2387 | offset = generic_file_llseek(file, offset, origin); |
b2675157 JB |
2388 | goto out; |
2389 | case SEEK_DATA: | |
2390 | case SEEK_HOLE: | |
48802c8a JL |
2391 | if (offset >= i_size_read(inode)) { |
2392 | mutex_unlock(&inode->i_mutex); | |
2393 | return -ENXIO; | |
2394 | } | |
2395 | ||
b2675157 JB |
2396 | ret = find_desired_extent(inode, &offset, origin); |
2397 | if (ret) { | |
2398 | mutex_unlock(&inode->i_mutex); | |
2399 | return ret; | |
2400 | } | |
2401 | } | |
2402 | ||
9a4327ca | 2403 | if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET)) { |
48802c8a | 2404 | offset = -EINVAL; |
9a4327ca DC |
2405 | goto out; |
2406 | } | |
2407 | if (offset > inode->i_sb->s_maxbytes) { | |
48802c8a | 2408 | offset = -EINVAL; |
9a4327ca DC |
2409 | goto out; |
2410 | } | |
b2675157 JB |
2411 | |
2412 | /* Special lock needed here? */ | |
2413 | if (offset != file->f_pos) { | |
2414 | file->f_pos = offset; | |
2415 | file->f_version = 0; | |
2416 | } | |
2417 | out: | |
2418 | mutex_unlock(&inode->i_mutex); | |
2419 | return offset; | |
2420 | } | |
2421 | ||
828c0950 | 2422 | const struct file_operations btrfs_file_operations = { |
b2675157 | 2423 | .llseek = btrfs_file_llseek, |
39279cc3 | 2424 | .read = do_sync_read, |
4a001071 | 2425 | .write = do_sync_write, |
9ebefb18 | 2426 | .aio_read = generic_file_aio_read, |
e9906a98 | 2427 | .splice_read = generic_file_splice_read, |
11c65dcc | 2428 | .aio_write = btrfs_file_aio_write, |
9ebefb18 | 2429 | .mmap = btrfs_file_mmap, |
39279cc3 | 2430 | .open = generic_file_open, |
e1b81e67 | 2431 | .release = btrfs_release_file, |
39279cc3 | 2432 | .fsync = btrfs_sync_file, |
2fe17c10 | 2433 | .fallocate = btrfs_fallocate, |
34287aa3 | 2434 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 2435 | #ifdef CONFIG_COMPAT |
34287aa3 | 2436 | .compat_ioctl = btrfs_ioctl, |
39279cc3 CM |
2437 | #endif |
2438 | }; | |
9247f317 MX |
2439 | |
2440 | void btrfs_auto_defrag_exit(void) | |
2441 | { | |
2442 | if (btrfs_inode_defrag_cachep) | |
2443 | kmem_cache_destroy(btrfs_inode_defrag_cachep); | |
2444 | } | |
2445 | ||
2446 | int btrfs_auto_defrag_init(void) | |
2447 | { | |
2448 | btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag", | |
2449 | sizeof(struct inode_defrag), 0, | |
2450 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, | |
2451 | NULL); | |
2452 | if (!btrfs_inode_defrag_cachep) | |
2453 | return -ENOMEM; | |
2454 | ||
2455 | return 0; | |
2456 | } |