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> | |
a27bb332 | 27 | #include <linux/aio.h> |
2fe17c10 | 28 | #include <linux/falloc.h> |
39279cc3 CM |
29 | #include <linux/swap.h> |
30 | #include <linux/writeback.h> | |
31 | #include <linux/statfs.h> | |
32 | #include <linux/compat.h> | |
5a0e3ad6 | 33 | #include <linux/slab.h> |
55e301fd | 34 | #include <linux/btrfs.h> |
39279cc3 CM |
35 | #include "ctree.h" |
36 | #include "disk-io.h" | |
37 | #include "transaction.h" | |
38 | #include "btrfs_inode.h" | |
39279cc3 | 39 | #include "print-tree.h" |
e02119d5 CM |
40 | #include "tree-log.h" |
41 | #include "locking.h" | |
12fa8ec6 | 42 | #include "compat.h" |
2aaa6655 | 43 | #include "volumes.h" |
39279cc3 | 44 | |
9247f317 | 45 | static struct kmem_cache *btrfs_inode_defrag_cachep; |
4cb5300b CM |
46 | /* |
47 | * when auto defrag is enabled we | |
48 | * queue up these defrag structs to remember which | |
49 | * inodes need defragging passes | |
50 | */ | |
51 | struct inode_defrag { | |
52 | struct rb_node rb_node; | |
53 | /* objectid */ | |
54 | u64 ino; | |
55 | /* | |
56 | * transid where the defrag was added, we search for | |
57 | * extents newer than this | |
58 | */ | |
59 | u64 transid; | |
60 | ||
61 | /* root objectid */ | |
62 | u64 root; | |
63 | ||
64 | /* last offset we were able to defrag */ | |
65 | u64 last_offset; | |
66 | ||
67 | /* if we've wrapped around back to zero once already */ | |
68 | int cycled; | |
69 | }; | |
70 | ||
762f2263 MX |
71 | static int __compare_inode_defrag(struct inode_defrag *defrag1, |
72 | struct inode_defrag *defrag2) | |
73 | { | |
74 | if (defrag1->root > defrag2->root) | |
75 | return 1; | |
76 | else if (defrag1->root < defrag2->root) | |
77 | return -1; | |
78 | else if (defrag1->ino > defrag2->ino) | |
79 | return 1; | |
80 | else if (defrag1->ino < defrag2->ino) | |
81 | return -1; | |
82 | else | |
83 | return 0; | |
84 | } | |
85 | ||
4cb5300b CM |
86 | /* pop a record for an inode into the defrag tree. The lock |
87 | * must be held already | |
88 | * | |
89 | * If you're inserting a record for an older transid than an | |
90 | * existing record, the transid already in the tree is lowered | |
91 | * | |
92 | * If an existing record is found the defrag item you | |
93 | * pass in is freed | |
94 | */ | |
8ddc4734 | 95 | static int __btrfs_add_inode_defrag(struct inode *inode, |
4cb5300b CM |
96 | struct inode_defrag *defrag) |
97 | { | |
98 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
99 | struct inode_defrag *entry; | |
100 | struct rb_node **p; | |
101 | struct rb_node *parent = NULL; | |
762f2263 | 102 | int ret; |
4cb5300b CM |
103 | |
104 | p = &root->fs_info->defrag_inodes.rb_node; | |
105 | while (*p) { | |
106 | parent = *p; | |
107 | entry = rb_entry(parent, struct inode_defrag, rb_node); | |
108 | ||
762f2263 MX |
109 | ret = __compare_inode_defrag(defrag, entry); |
110 | if (ret < 0) | |
4cb5300b | 111 | p = &parent->rb_left; |
762f2263 | 112 | else if (ret > 0) |
4cb5300b CM |
113 | p = &parent->rb_right; |
114 | else { | |
115 | /* if we're reinserting an entry for | |
116 | * an old defrag run, make sure to | |
117 | * lower the transid of our existing record | |
118 | */ | |
119 | if (defrag->transid < entry->transid) | |
120 | entry->transid = defrag->transid; | |
121 | if (defrag->last_offset > entry->last_offset) | |
122 | entry->last_offset = defrag->last_offset; | |
8ddc4734 | 123 | return -EEXIST; |
4cb5300b CM |
124 | } |
125 | } | |
72ac3c0d | 126 | set_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags); |
4cb5300b CM |
127 | rb_link_node(&defrag->rb_node, parent, p); |
128 | rb_insert_color(&defrag->rb_node, &root->fs_info->defrag_inodes); | |
8ddc4734 MX |
129 | return 0; |
130 | } | |
4cb5300b | 131 | |
8ddc4734 MX |
132 | static inline int __need_auto_defrag(struct btrfs_root *root) |
133 | { | |
134 | if (!btrfs_test_opt(root, AUTO_DEFRAG)) | |
135 | return 0; | |
136 | ||
137 | if (btrfs_fs_closing(root->fs_info)) | |
138 | return 0; | |
4cb5300b | 139 | |
8ddc4734 | 140 | return 1; |
4cb5300b CM |
141 | } |
142 | ||
143 | /* | |
144 | * insert a defrag record for this inode if auto defrag is | |
145 | * enabled | |
146 | */ | |
147 | int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans, | |
148 | struct inode *inode) | |
149 | { | |
150 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
151 | struct inode_defrag *defrag; | |
4cb5300b | 152 | u64 transid; |
8ddc4734 | 153 | int ret; |
4cb5300b | 154 | |
8ddc4734 | 155 | if (!__need_auto_defrag(root)) |
4cb5300b CM |
156 | return 0; |
157 | ||
72ac3c0d | 158 | if (test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags)) |
4cb5300b CM |
159 | return 0; |
160 | ||
161 | if (trans) | |
162 | transid = trans->transid; | |
163 | else | |
164 | transid = BTRFS_I(inode)->root->last_trans; | |
165 | ||
9247f317 | 166 | defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS); |
4cb5300b CM |
167 | if (!defrag) |
168 | return -ENOMEM; | |
169 | ||
a4689d2b | 170 | defrag->ino = btrfs_ino(inode); |
4cb5300b CM |
171 | defrag->transid = transid; |
172 | defrag->root = root->root_key.objectid; | |
173 | ||
174 | spin_lock(&root->fs_info->defrag_inodes_lock); | |
8ddc4734 MX |
175 | if (!test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags)) { |
176 | /* | |
177 | * If we set IN_DEFRAG flag and evict the inode from memory, | |
178 | * and then re-read this inode, this new inode doesn't have | |
179 | * IN_DEFRAG flag. At the case, we may find the existed defrag. | |
180 | */ | |
181 | ret = __btrfs_add_inode_defrag(inode, defrag); | |
182 | if (ret) | |
183 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
184 | } else { | |
9247f317 | 185 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); |
8ddc4734 | 186 | } |
4cb5300b | 187 | spin_unlock(&root->fs_info->defrag_inodes_lock); |
a0f98dde | 188 | return 0; |
4cb5300b CM |
189 | } |
190 | ||
191 | /* | |
8ddc4734 MX |
192 | * Requeue the defrag object. If there is a defrag object that points to |
193 | * the same inode in the tree, we will merge them together (by | |
194 | * __btrfs_add_inode_defrag()) and free the one that we want to requeue. | |
4cb5300b | 195 | */ |
48a3b636 ES |
196 | static void btrfs_requeue_inode_defrag(struct inode *inode, |
197 | struct inode_defrag *defrag) | |
8ddc4734 MX |
198 | { |
199 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
200 | int ret; | |
201 | ||
202 | if (!__need_auto_defrag(root)) | |
203 | goto out; | |
204 | ||
205 | /* | |
206 | * Here we don't check the IN_DEFRAG flag, because we need merge | |
207 | * them together. | |
208 | */ | |
209 | spin_lock(&root->fs_info->defrag_inodes_lock); | |
210 | ret = __btrfs_add_inode_defrag(inode, defrag); | |
211 | spin_unlock(&root->fs_info->defrag_inodes_lock); | |
212 | if (ret) | |
213 | goto out; | |
214 | return; | |
215 | out: | |
216 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
217 | } | |
218 | ||
4cb5300b | 219 | /* |
26176e7c MX |
220 | * pick the defragable inode that we want, if it doesn't exist, we will get |
221 | * the next one. | |
4cb5300b | 222 | */ |
26176e7c MX |
223 | static struct inode_defrag * |
224 | btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino) | |
4cb5300b CM |
225 | { |
226 | struct inode_defrag *entry = NULL; | |
762f2263 | 227 | struct inode_defrag tmp; |
4cb5300b CM |
228 | struct rb_node *p; |
229 | struct rb_node *parent = NULL; | |
762f2263 MX |
230 | int ret; |
231 | ||
232 | tmp.ino = ino; | |
233 | tmp.root = root; | |
4cb5300b | 234 | |
26176e7c MX |
235 | spin_lock(&fs_info->defrag_inodes_lock); |
236 | p = fs_info->defrag_inodes.rb_node; | |
4cb5300b CM |
237 | while (p) { |
238 | parent = p; | |
239 | entry = rb_entry(parent, struct inode_defrag, rb_node); | |
240 | ||
762f2263 MX |
241 | ret = __compare_inode_defrag(&tmp, entry); |
242 | if (ret < 0) | |
4cb5300b | 243 | p = parent->rb_left; |
762f2263 | 244 | else if (ret > 0) |
4cb5300b CM |
245 | p = parent->rb_right; |
246 | else | |
26176e7c | 247 | goto out; |
4cb5300b CM |
248 | } |
249 | ||
26176e7c MX |
250 | if (parent && __compare_inode_defrag(&tmp, entry) > 0) { |
251 | parent = rb_next(parent); | |
252 | if (parent) | |
4cb5300b | 253 | entry = rb_entry(parent, struct inode_defrag, rb_node); |
26176e7c MX |
254 | else |
255 | entry = NULL; | |
4cb5300b | 256 | } |
26176e7c MX |
257 | out: |
258 | if (entry) | |
259 | rb_erase(parent, &fs_info->defrag_inodes); | |
260 | spin_unlock(&fs_info->defrag_inodes_lock); | |
261 | return entry; | |
4cb5300b CM |
262 | } |
263 | ||
26176e7c | 264 | void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info) |
4cb5300b CM |
265 | { |
266 | struct inode_defrag *defrag; | |
26176e7c MX |
267 | struct rb_node *node; |
268 | ||
269 | spin_lock(&fs_info->defrag_inodes_lock); | |
270 | node = rb_first(&fs_info->defrag_inodes); | |
271 | while (node) { | |
272 | rb_erase(node, &fs_info->defrag_inodes); | |
273 | defrag = rb_entry(node, struct inode_defrag, rb_node); | |
274 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
275 | ||
276 | if (need_resched()) { | |
277 | spin_unlock(&fs_info->defrag_inodes_lock); | |
278 | cond_resched(); | |
279 | spin_lock(&fs_info->defrag_inodes_lock); | |
280 | } | |
281 | ||
282 | node = rb_first(&fs_info->defrag_inodes); | |
283 | } | |
284 | spin_unlock(&fs_info->defrag_inodes_lock); | |
285 | } | |
286 | ||
287 | #define BTRFS_DEFRAG_BATCH 1024 | |
288 | ||
289 | static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info, | |
290 | struct inode_defrag *defrag) | |
291 | { | |
4cb5300b CM |
292 | struct btrfs_root *inode_root; |
293 | struct inode *inode; | |
4cb5300b CM |
294 | struct btrfs_key key; |
295 | struct btrfs_ioctl_defrag_range_args range; | |
4cb5300b | 296 | int num_defrag; |
6f1c3605 LB |
297 | int index; |
298 | int ret; | |
4cb5300b | 299 | |
26176e7c MX |
300 | /* get the inode */ |
301 | key.objectid = defrag->root; | |
302 | btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY); | |
303 | key.offset = (u64)-1; | |
6f1c3605 LB |
304 | |
305 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
306 | ||
26176e7c MX |
307 | inode_root = btrfs_read_fs_root_no_name(fs_info, &key); |
308 | if (IS_ERR(inode_root)) { | |
6f1c3605 LB |
309 | ret = PTR_ERR(inode_root); |
310 | goto cleanup; | |
311 | } | |
26176e7c | 312 | |
2112ac80 JB |
313 | if (btrfs_root_refs(&inode_root->root_item) == 0) { |
314 | ret = -ENOENT; | |
315 | goto cleanup; | |
316 | } | |
317 | ||
26176e7c MX |
318 | key.objectid = defrag->ino; |
319 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); | |
320 | key.offset = 0; | |
321 | inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL); | |
322 | if (IS_ERR(inode)) { | |
6f1c3605 LB |
323 | ret = PTR_ERR(inode); |
324 | goto cleanup; | |
26176e7c | 325 | } |
6f1c3605 | 326 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
26176e7c MX |
327 | |
328 | /* do a chunk of defrag */ | |
329 | clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags); | |
4cb5300b CM |
330 | memset(&range, 0, sizeof(range)); |
331 | range.len = (u64)-1; | |
26176e7c | 332 | range.start = defrag->last_offset; |
b66f00da MX |
333 | |
334 | sb_start_write(fs_info->sb); | |
26176e7c MX |
335 | num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid, |
336 | BTRFS_DEFRAG_BATCH); | |
b66f00da | 337 | sb_end_write(fs_info->sb); |
26176e7c MX |
338 | /* |
339 | * if we filled the whole defrag batch, there | |
340 | * must be more work to do. Queue this defrag | |
341 | * again | |
342 | */ | |
343 | if (num_defrag == BTRFS_DEFRAG_BATCH) { | |
344 | defrag->last_offset = range.start; | |
345 | btrfs_requeue_inode_defrag(inode, defrag); | |
346 | } else if (defrag->last_offset && !defrag->cycled) { | |
347 | /* | |
348 | * we didn't fill our defrag batch, but | |
349 | * we didn't start at zero. Make sure we loop | |
350 | * around to the start of the file. | |
351 | */ | |
352 | defrag->last_offset = 0; | |
353 | defrag->cycled = 1; | |
354 | btrfs_requeue_inode_defrag(inode, defrag); | |
355 | } else { | |
356 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
357 | } | |
358 | ||
359 | iput(inode); | |
360 | return 0; | |
6f1c3605 LB |
361 | cleanup: |
362 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
363 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
364 | return ret; | |
26176e7c MX |
365 | } |
366 | ||
367 | /* | |
368 | * run through the list of inodes in the FS that need | |
369 | * defragging | |
370 | */ | |
371 | int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info) | |
372 | { | |
373 | struct inode_defrag *defrag; | |
374 | u64 first_ino = 0; | |
375 | u64 root_objectid = 0; | |
4cb5300b CM |
376 | |
377 | atomic_inc(&fs_info->defrag_running); | |
4cb5300b | 378 | while(1) { |
dc81cdc5 MX |
379 | /* Pause the auto defragger. */ |
380 | if (test_bit(BTRFS_FS_STATE_REMOUNTING, | |
381 | &fs_info->fs_state)) | |
382 | break; | |
383 | ||
26176e7c MX |
384 | if (!__need_auto_defrag(fs_info->tree_root)) |
385 | break; | |
4cb5300b CM |
386 | |
387 | /* find an inode to defrag */ | |
26176e7c MX |
388 | defrag = btrfs_pick_defrag_inode(fs_info, root_objectid, |
389 | first_ino); | |
4cb5300b | 390 | if (!defrag) { |
26176e7c | 391 | if (root_objectid || first_ino) { |
762f2263 | 392 | root_objectid = 0; |
4cb5300b CM |
393 | first_ino = 0; |
394 | continue; | |
395 | } else { | |
396 | break; | |
397 | } | |
398 | } | |
399 | ||
4cb5300b | 400 | first_ino = defrag->ino + 1; |
762f2263 | 401 | root_objectid = defrag->root; |
4cb5300b | 402 | |
26176e7c | 403 | __btrfs_run_defrag_inode(fs_info, defrag); |
4cb5300b | 404 | } |
4cb5300b CM |
405 | atomic_dec(&fs_info->defrag_running); |
406 | ||
407 | /* | |
408 | * during unmount, we use the transaction_wait queue to | |
409 | * wait for the defragger to stop | |
410 | */ | |
411 | wake_up(&fs_info->transaction_wait); | |
412 | return 0; | |
413 | } | |
39279cc3 | 414 | |
d352ac68 CM |
415 | /* simple helper to fault in pages and copy. This should go away |
416 | * and be replaced with calls into generic code. | |
417 | */ | |
d397712b | 418 | static noinline int btrfs_copy_from_user(loff_t pos, int num_pages, |
d0215f3e | 419 | size_t write_bytes, |
a1b32a59 | 420 | struct page **prepared_pages, |
11c65dcc | 421 | struct iov_iter *i) |
39279cc3 | 422 | { |
914ee295 | 423 | size_t copied = 0; |
d0215f3e | 424 | size_t total_copied = 0; |
11c65dcc | 425 | int pg = 0; |
39279cc3 CM |
426 | int offset = pos & (PAGE_CACHE_SIZE - 1); |
427 | ||
11c65dcc | 428 | while (write_bytes > 0) { |
39279cc3 CM |
429 | size_t count = min_t(size_t, |
430 | PAGE_CACHE_SIZE - offset, write_bytes); | |
11c65dcc | 431 | struct page *page = prepared_pages[pg]; |
914ee295 XZ |
432 | /* |
433 | * Copy data from userspace to the current page | |
434 | * | |
435 | * Disable pagefault to avoid recursive lock since | |
436 | * the pages are already locked | |
437 | */ | |
438 | pagefault_disable(); | |
439 | copied = iov_iter_copy_from_user_atomic(page, i, offset, count); | |
440 | pagefault_enable(); | |
11c65dcc | 441 | |
39279cc3 CM |
442 | /* Flush processor's dcache for this page */ |
443 | flush_dcache_page(page); | |
31339acd CM |
444 | |
445 | /* | |
446 | * if we get a partial write, we can end up with | |
447 | * partially up to date pages. These add | |
448 | * a lot of complexity, so make sure they don't | |
449 | * happen by forcing this copy to be retried. | |
450 | * | |
451 | * The rest of the btrfs_file_write code will fall | |
452 | * back to page at a time copies after we return 0. | |
453 | */ | |
454 | if (!PageUptodate(page) && copied < count) | |
455 | copied = 0; | |
456 | ||
11c65dcc JB |
457 | iov_iter_advance(i, copied); |
458 | write_bytes -= copied; | |
914ee295 | 459 | total_copied += copied; |
39279cc3 | 460 | |
914ee295 | 461 | /* Return to btrfs_file_aio_write to fault page */ |
9f570b8d | 462 | if (unlikely(copied == 0)) |
914ee295 | 463 | break; |
11c65dcc JB |
464 | |
465 | if (unlikely(copied < PAGE_CACHE_SIZE - offset)) { | |
466 | offset += copied; | |
467 | } else { | |
468 | pg++; | |
469 | offset = 0; | |
470 | } | |
39279cc3 | 471 | } |
914ee295 | 472 | return total_copied; |
39279cc3 CM |
473 | } |
474 | ||
d352ac68 CM |
475 | /* |
476 | * unlocks pages after btrfs_file_write is done with them | |
477 | */ | |
48a3b636 | 478 | static void btrfs_drop_pages(struct page **pages, size_t num_pages) |
39279cc3 CM |
479 | { |
480 | size_t i; | |
481 | for (i = 0; i < num_pages; i++) { | |
d352ac68 CM |
482 | /* page checked is some magic around finding pages that |
483 | * have been modified without going through btrfs_set_page_dirty | |
484 | * clear it here | |
485 | */ | |
4a096752 | 486 | ClearPageChecked(pages[i]); |
39279cc3 CM |
487 | unlock_page(pages[i]); |
488 | mark_page_accessed(pages[i]); | |
489 | page_cache_release(pages[i]); | |
490 | } | |
491 | } | |
492 | ||
d352ac68 CM |
493 | /* |
494 | * after copy_from_user, pages need to be dirtied and we need to make | |
495 | * sure holes are created between the current EOF and the start of | |
496 | * any next extents (if required). | |
497 | * | |
498 | * this also makes the decision about creating an inline extent vs | |
499 | * doing real data extents, marking pages dirty and delalloc as required. | |
500 | */ | |
be1a12a0 | 501 | int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode, |
48a3b636 ES |
502 | struct page **pages, size_t num_pages, |
503 | loff_t pos, size_t write_bytes, | |
504 | struct extent_state **cached) | |
39279cc3 | 505 | { |
39279cc3 | 506 | int err = 0; |
a52d9a80 | 507 | int i; |
db94535d | 508 | u64 num_bytes; |
a52d9a80 CM |
509 | u64 start_pos; |
510 | u64 end_of_last_block; | |
511 | u64 end_pos = pos + write_bytes; | |
512 | loff_t isize = i_size_read(inode); | |
39279cc3 | 513 | |
5f39d397 | 514 | start_pos = pos & ~((u64)root->sectorsize - 1); |
fda2832f | 515 | num_bytes = ALIGN(write_bytes + pos - start_pos, root->sectorsize); |
39279cc3 | 516 | |
db94535d | 517 | end_of_last_block = start_pos + num_bytes - 1; |
2ac55d41 | 518 | err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block, |
be1a12a0 | 519 | cached); |
d0215f3e JB |
520 | if (err) |
521 | return err; | |
9ed74f2d | 522 | |
c8b97818 CM |
523 | for (i = 0; i < num_pages; i++) { |
524 | struct page *p = pages[i]; | |
525 | SetPageUptodate(p); | |
526 | ClearPageChecked(p); | |
527 | set_page_dirty(p); | |
a52d9a80 | 528 | } |
9f570b8d JB |
529 | |
530 | /* | |
531 | * we've only changed i_size in ram, and we haven't updated | |
532 | * the disk i_size. There is no need to log the inode | |
533 | * at this time. | |
534 | */ | |
535 | if (end_pos > isize) | |
a52d9a80 | 536 | i_size_write(inode, end_pos); |
a22285a6 | 537 | return 0; |
39279cc3 CM |
538 | } |
539 | ||
d352ac68 CM |
540 | /* |
541 | * this drops all the extents in the cache that intersect the range | |
542 | * [start, end]. Existing extents are split as required. | |
543 | */ | |
7014cdb4 JB |
544 | void btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end, |
545 | int skip_pinned) | |
a52d9a80 CM |
546 | { |
547 | struct extent_map *em; | |
3b951516 CM |
548 | struct extent_map *split = NULL; |
549 | struct extent_map *split2 = NULL; | |
a52d9a80 | 550 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
39b5637f | 551 | u64 len = end - start + 1; |
5dc562c5 | 552 | u64 gen; |
3b951516 CM |
553 | int ret; |
554 | int testend = 1; | |
5b21f2ed | 555 | unsigned long flags; |
c8b97818 | 556 | int compressed = 0; |
09a2a8f9 | 557 | bool modified; |
a52d9a80 | 558 | |
e6dcd2dc | 559 | WARN_ON(end < start); |
3b951516 | 560 | if (end == (u64)-1) { |
39b5637f | 561 | len = (u64)-1; |
3b951516 CM |
562 | testend = 0; |
563 | } | |
d397712b | 564 | while (1) { |
7014cdb4 JB |
565 | int no_splits = 0; |
566 | ||
09a2a8f9 | 567 | modified = false; |
3b951516 | 568 | if (!split) |
172ddd60 | 569 | split = alloc_extent_map(); |
3b951516 | 570 | if (!split2) |
172ddd60 | 571 | split2 = alloc_extent_map(); |
7014cdb4 JB |
572 | if (!split || !split2) |
573 | no_splits = 1; | |
3b951516 | 574 | |
890871be | 575 | write_lock(&em_tree->lock); |
39b5637f | 576 | em = lookup_extent_mapping(em_tree, start, len); |
d1310b2e | 577 | if (!em) { |
890871be | 578 | write_unlock(&em_tree->lock); |
a52d9a80 | 579 | break; |
d1310b2e | 580 | } |
5b21f2ed | 581 | flags = em->flags; |
5dc562c5 | 582 | gen = em->generation; |
5b21f2ed | 583 | if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) { |
55ef6899 | 584 | if (testend && em->start + em->len >= start + len) { |
5b21f2ed | 585 | free_extent_map(em); |
a1ed835e | 586 | write_unlock(&em_tree->lock); |
5b21f2ed ZY |
587 | break; |
588 | } | |
55ef6899 YZ |
589 | start = em->start + em->len; |
590 | if (testend) | |
5b21f2ed | 591 | len = start + len - (em->start + em->len); |
5b21f2ed | 592 | free_extent_map(em); |
a1ed835e | 593 | write_unlock(&em_tree->lock); |
5b21f2ed ZY |
594 | continue; |
595 | } | |
c8b97818 | 596 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
3ce7e67a | 597 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
3b277594 | 598 | clear_bit(EXTENT_FLAG_LOGGING, &flags); |
09a2a8f9 | 599 | modified = !list_empty(&em->list); |
a52d9a80 | 600 | remove_extent_mapping(em_tree, em); |
7014cdb4 JB |
601 | if (no_splits) |
602 | goto next; | |
3b951516 | 603 | |
ee20a983 | 604 | if (em->start < start) { |
3b951516 CM |
605 | split->start = em->start; |
606 | split->len = start - em->start; | |
ee20a983 JB |
607 | |
608 | if (em->block_start < EXTENT_MAP_LAST_BYTE) { | |
609 | split->orig_start = em->orig_start; | |
610 | split->block_start = em->block_start; | |
611 | ||
612 | if (compressed) | |
613 | split->block_len = em->block_len; | |
614 | else | |
615 | split->block_len = split->len; | |
616 | split->orig_block_len = max(split->block_len, | |
617 | em->orig_block_len); | |
618 | split->ram_bytes = em->ram_bytes; | |
619 | } else { | |
620 | split->orig_start = split->start; | |
621 | split->block_len = 0; | |
622 | split->block_start = em->block_start; | |
623 | split->orig_block_len = 0; | |
624 | split->ram_bytes = split->len; | |
625 | } | |
626 | ||
5dc562c5 | 627 | split->generation = gen; |
3b951516 | 628 | split->bdev = em->bdev; |
5b21f2ed | 629 | split->flags = flags; |
261507a0 | 630 | split->compress_type = em->compress_type; |
09a2a8f9 | 631 | ret = add_extent_mapping(em_tree, split, modified); |
79787eaa | 632 | BUG_ON(ret); /* Logic error */ |
3b951516 CM |
633 | free_extent_map(split); |
634 | split = split2; | |
635 | split2 = NULL; | |
636 | } | |
ee20a983 | 637 | if (testend && em->start + em->len > start + len) { |
3b951516 CM |
638 | u64 diff = start + len - em->start; |
639 | ||
640 | split->start = start + len; | |
641 | split->len = em->start + em->len - (start + len); | |
642 | split->bdev = em->bdev; | |
5b21f2ed | 643 | split->flags = flags; |
261507a0 | 644 | split->compress_type = em->compress_type; |
5dc562c5 | 645 | split->generation = gen; |
ee20a983 JB |
646 | |
647 | if (em->block_start < EXTENT_MAP_LAST_BYTE) { | |
648 | split->orig_block_len = max(em->block_len, | |
b4939680 | 649 | em->orig_block_len); |
3b951516 | 650 | |
ee20a983 JB |
651 | split->ram_bytes = em->ram_bytes; |
652 | if (compressed) { | |
653 | split->block_len = em->block_len; | |
654 | split->block_start = em->block_start; | |
655 | split->orig_start = em->orig_start; | |
656 | } else { | |
657 | split->block_len = split->len; | |
658 | split->block_start = em->block_start | |
659 | + diff; | |
660 | split->orig_start = em->orig_start; | |
661 | } | |
c8b97818 | 662 | } else { |
ee20a983 JB |
663 | split->ram_bytes = split->len; |
664 | split->orig_start = split->start; | |
665 | split->block_len = 0; | |
666 | split->block_start = em->block_start; | |
667 | split->orig_block_len = 0; | |
c8b97818 | 668 | } |
3b951516 | 669 | |
09a2a8f9 | 670 | ret = add_extent_mapping(em_tree, split, modified); |
79787eaa | 671 | BUG_ON(ret); /* Logic error */ |
3b951516 CM |
672 | free_extent_map(split); |
673 | split = NULL; | |
674 | } | |
7014cdb4 | 675 | next: |
890871be | 676 | write_unlock(&em_tree->lock); |
d1310b2e | 677 | |
a52d9a80 CM |
678 | /* once for us */ |
679 | free_extent_map(em); | |
680 | /* once for the tree*/ | |
681 | free_extent_map(em); | |
682 | } | |
3b951516 CM |
683 | if (split) |
684 | free_extent_map(split); | |
685 | if (split2) | |
686 | free_extent_map(split2); | |
a52d9a80 CM |
687 | } |
688 | ||
39279cc3 CM |
689 | /* |
690 | * this is very complex, but the basic idea is to drop all extents | |
691 | * in the range start - end. hint_block is filled in with a block number | |
692 | * that would be a good hint to the block allocator for this file. | |
693 | * | |
694 | * If an extent intersects the range but is not entirely inside the range | |
695 | * it is either truncated or split. Anything entirely inside the range | |
696 | * is deleted from the tree. | |
697 | */ | |
5dc562c5 JB |
698 | int __btrfs_drop_extents(struct btrfs_trans_handle *trans, |
699 | struct btrfs_root *root, struct inode *inode, | |
700 | struct btrfs_path *path, u64 start, u64 end, | |
2aaa6655 | 701 | u64 *drop_end, int drop_cache) |
39279cc3 | 702 | { |
5f39d397 | 703 | struct extent_buffer *leaf; |
920bbbfb | 704 | struct btrfs_file_extent_item *fi; |
00f5c795 | 705 | struct btrfs_key key; |
920bbbfb | 706 | struct btrfs_key new_key; |
33345d01 | 707 | u64 ino = btrfs_ino(inode); |
920bbbfb YZ |
708 | u64 search_start = start; |
709 | u64 disk_bytenr = 0; | |
710 | u64 num_bytes = 0; | |
711 | u64 extent_offset = 0; | |
712 | u64 extent_end = 0; | |
713 | int del_nr = 0; | |
714 | int del_slot = 0; | |
715 | int extent_type; | |
ccd467d6 | 716 | int recow; |
00f5c795 | 717 | int ret; |
dc7fdde3 | 718 | int modify_tree = -1; |
5dc562c5 | 719 | int update_refs = (root->ref_cows || root == root->fs_info->tree_root); |
c3308f84 | 720 | int found = 0; |
39279cc3 | 721 | |
a1ed835e CM |
722 | if (drop_cache) |
723 | btrfs_drop_extent_cache(inode, start, end - 1, 0); | |
a52d9a80 | 724 | |
dc7fdde3 CM |
725 | if (start >= BTRFS_I(inode)->disk_i_size) |
726 | modify_tree = 0; | |
727 | ||
d397712b | 728 | while (1) { |
ccd467d6 | 729 | recow = 0; |
33345d01 | 730 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
dc7fdde3 | 731 | search_start, modify_tree); |
39279cc3 | 732 | if (ret < 0) |
920bbbfb YZ |
733 | break; |
734 | if (ret > 0 && path->slots[0] > 0 && search_start == start) { | |
735 | leaf = path->nodes[0]; | |
736 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1); | |
33345d01 | 737 | if (key.objectid == ino && |
920bbbfb YZ |
738 | key.type == BTRFS_EXTENT_DATA_KEY) |
739 | path->slots[0]--; | |
39279cc3 | 740 | } |
920bbbfb | 741 | ret = 0; |
8c2383c3 | 742 | next_slot: |
5f39d397 | 743 | leaf = path->nodes[0]; |
920bbbfb YZ |
744 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { |
745 | BUG_ON(del_nr > 0); | |
746 | ret = btrfs_next_leaf(root, path); | |
747 | if (ret < 0) | |
748 | break; | |
749 | if (ret > 0) { | |
750 | ret = 0; | |
751 | break; | |
8c2383c3 | 752 | } |
920bbbfb YZ |
753 | leaf = path->nodes[0]; |
754 | recow = 1; | |
755 | } | |
756 | ||
757 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
33345d01 | 758 | if (key.objectid > ino || |
920bbbfb YZ |
759 | key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end) |
760 | break; | |
761 | ||
762 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
763 | struct btrfs_file_extent_item); | |
764 | extent_type = btrfs_file_extent_type(leaf, fi); | |
765 | ||
766 | if (extent_type == BTRFS_FILE_EXTENT_REG || | |
767 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
768 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
769 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
770 | extent_offset = btrfs_file_extent_offset(leaf, fi); | |
771 | extent_end = key.offset + | |
772 | btrfs_file_extent_num_bytes(leaf, fi); | |
773 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
774 | extent_end = key.offset + | |
775 | btrfs_file_extent_inline_len(leaf, fi); | |
8c2383c3 | 776 | } else { |
920bbbfb | 777 | WARN_ON(1); |
8c2383c3 | 778 | extent_end = search_start; |
39279cc3 CM |
779 | } |
780 | ||
920bbbfb YZ |
781 | if (extent_end <= search_start) { |
782 | path->slots[0]++; | |
8c2383c3 | 783 | goto next_slot; |
39279cc3 CM |
784 | } |
785 | ||
c3308f84 | 786 | found = 1; |
920bbbfb | 787 | search_start = max(key.offset, start); |
dc7fdde3 CM |
788 | if (recow || !modify_tree) { |
789 | modify_tree = -1; | |
b3b4aa74 | 790 | btrfs_release_path(path); |
920bbbfb | 791 | continue; |
39279cc3 | 792 | } |
6643558d | 793 | |
920bbbfb YZ |
794 | /* |
795 | * | - range to drop - | | |
796 | * | -------- extent -------- | | |
797 | */ | |
798 | if (start > key.offset && end < extent_end) { | |
799 | BUG_ON(del_nr > 0); | |
800 | BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE); | |
801 | ||
802 | memcpy(&new_key, &key, sizeof(new_key)); | |
803 | new_key.offset = start; | |
804 | ret = btrfs_duplicate_item(trans, root, path, | |
805 | &new_key); | |
806 | if (ret == -EAGAIN) { | |
b3b4aa74 | 807 | btrfs_release_path(path); |
920bbbfb | 808 | continue; |
6643558d | 809 | } |
920bbbfb YZ |
810 | if (ret < 0) |
811 | break; | |
812 | ||
813 | leaf = path->nodes[0]; | |
814 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
815 | struct btrfs_file_extent_item); | |
816 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
817 | start - key.offset); | |
818 | ||
819 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
820 | struct btrfs_file_extent_item); | |
821 | ||
822 | extent_offset += start - key.offset; | |
823 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); | |
824 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
825 | extent_end - start); | |
826 | btrfs_mark_buffer_dirty(leaf); | |
827 | ||
5dc562c5 | 828 | if (update_refs && disk_bytenr > 0) { |
771ed689 | 829 | ret = btrfs_inc_extent_ref(trans, root, |
920bbbfb YZ |
830 | disk_bytenr, num_bytes, 0, |
831 | root->root_key.objectid, | |
832 | new_key.objectid, | |
66d7e7f0 | 833 | start - extent_offset, 0); |
79787eaa | 834 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 835 | } |
920bbbfb | 836 | key.offset = start; |
6643558d | 837 | } |
920bbbfb YZ |
838 | /* |
839 | * | ---- range to drop ----- | | |
840 | * | -------- extent -------- | | |
841 | */ | |
842 | if (start <= key.offset && end < extent_end) { | |
843 | BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE); | |
6643558d | 844 | |
920bbbfb YZ |
845 | memcpy(&new_key, &key, sizeof(new_key)); |
846 | new_key.offset = end; | |
afe5fea7 | 847 | btrfs_set_item_key_safe(root, path, &new_key); |
6643558d | 848 | |
920bbbfb YZ |
849 | extent_offset += end - key.offset; |
850 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); | |
851 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
852 | extent_end - end); | |
853 | btrfs_mark_buffer_dirty(leaf); | |
2671485d | 854 | if (update_refs && disk_bytenr > 0) |
920bbbfb | 855 | inode_sub_bytes(inode, end - key.offset); |
920bbbfb | 856 | break; |
39279cc3 | 857 | } |
771ed689 | 858 | |
920bbbfb YZ |
859 | search_start = extent_end; |
860 | /* | |
861 | * | ---- range to drop ----- | | |
862 | * | -------- extent -------- | | |
863 | */ | |
864 | if (start > key.offset && end >= extent_end) { | |
865 | BUG_ON(del_nr > 0); | |
866 | BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE); | |
8c2383c3 | 867 | |
920bbbfb YZ |
868 | btrfs_set_file_extent_num_bytes(leaf, fi, |
869 | start - key.offset); | |
870 | btrfs_mark_buffer_dirty(leaf); | |
2671485d | 871 | if (update_refs && disk_bytenr > 0) |
920bbbfb | 872 | inode_sub_bytes(inode, extent_end - start); |
920bbbfb YZ |
873 | if (end == extent_end) |
874 | break; | |
c8b97818 | 875 | |
920bbbfb YZ |
876 | path->slots[0]++; |
877 | goto next_slot; | |
31840ae1 ZY |
878 | } |
879 | ||
920bbbfb YZ |
880 | /* |
881 | * | ---- range to drop ----- | | |
882 | * | ------ extent ------ | | |
883 | */ | |
884 | if (start <= key.offset && end >= extent_end) { | |
885 | if (del_nr == 0) { | |
886 | del_slot = path->slots[0]; | |
887 | del_nr = 1; | |
888 | } else { | |
889 | BUG_ON(del_slot + del_nr != path->slots[0]); | |
890 | del_nr++; | |
891 | } | |
31840ae1 | 892 | |
5dc562c5 JB |
893 | if (update_refs && |
894 | extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
a76a3cd4 | 895 | inode_sub_bytes(inode, |
920bbbfb YZ |
896 | extent_end - key.offset); |
897 | extent_end = ALIGN(extent_end, | |
898 | root->sectorsize); | |
5dc562c5 | 899 | } else if (update_refs && disk_bytenr > 0) { |
31840ae1 | 900 | ret = btrfs_free_extent(trans, root, |
920bbbfb YZ |
901 | disk_bytenr, num_bytes, 0, |
902 | root->root_key.objectid, | |
5d4f98a2 | 903 | key.objectid, key.offset - |
66d7e7f0 | 904 | extent_offset, 0); |
79787eaa | 905 | BUG_ON(ret); /* -ENOMEM */ |
920bbbfb YZ |
906 | inode_sub_bytes(inode, |
907 | extent_end - key.offset); | |
31840ae1 | 908 | } |
31840ae1 | 909 | |
920bbbfb YZ |
910 | if (end == extent_end) |
911 | break; | |
912 | ||
913 | if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) { | |
914 | path->slots[0]++; | |
915 | goto next_slot; | |
916 | } | |
917 | ||
918 | ret = btrfs_del_items(trans, root, path, del_slot, | |
919 | del_nr); | |
79787eaa JM |
920 | if (ret) { |
921 | btrfs_abort_transaction(trans, root, ret); | |
5dc562c5 | 922 | break; |
79787eaa | 923 | } |
920bbbfb YZ |
924 | |
925 | del_nr = 0; | |
926 | del_slot = 0; | |
927 | ||
b3b4aa74 | 928 | btrfs_release_path(path); |
920bbbfb | 929 | continue; |
39279cc3 | 930 | } |
920bbbfb YZ |
931 | |
932 | BUG_ON(1); | |
39279cc3 | 933 | } |
920bbbfb | 934 | |
79787eaa | 935 | if (!ret && del_nr > 0) { |
920bbbfb | 936 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
79787eaa JM |
937 | if (ret) |
938 | btrfs_abort_transaction(trans, root, ret); | |
6643558d | 939 | } |
920bbbfb | 940 | |
2aaa6655 | 941 | if (drop_end) |
c3308f84 | 942 | *drop_end = found ? min(end, extent_end) : end; |
5dc562c5 JB |
943 | btrfs_release_path(path); |
944 | return ret; | |
945 | } | |
946 | ||
947 | int btrfs_drop_extents(struct btrfs_trans_handle *trans, | |
948 | struct btrfs_root *root, struct inode *inode, u64 start, | |
2671485d | 949 | u64 end, int drop_cache) |
5dc562c5 JB |
950 | { |
951 | struct btrfs_path *path; | |
952 | int ret; | |
953 | ||
954 | path = btrfs_alloc_path(); | |
955 | if (!path) | |
956 | return -ENOMEM; | |
2aaa6655 | 957 | ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL, |
2671485d | 958 | drop_cache); |
920bbbfb | 959 | btrfs_free_path(path); |
39279cc3 CM |
960 | return ret; |
961 | } | |
962 | ||
d899e052 | 963 | static int extent_mergeable(struct extent_buffer *leaf, int slot, |
6c7d54ac YZ |
964 | u64 objectid, u64 bytenr, u64 orig_offset, |
965 | u64 *start, u64 *end) | |
d899e052 YZ |
966 | { |
967 | struct btrfs_file_extent_item *fi; | |
968 | struct btrfs_key key; | |
969 | u64 extent_end; | |
970 | ||
971 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) | |
972 | return 0; | |
973 | ||
974 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
975 | if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY) | |
976 | return 0; | |
977 | ||
978 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
979 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG || | |
980 | btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr || | |
6c7d54ac | 981 | btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset || |
d899e052 YZ |
982 | btrfs_file_extent_compression(leaf, fi) || |
983 | btrfs_file_extent_encryption(leaf, fi) || | |
984 | btrfs_file_extent_other_encoding(leaf, fi)) | |
985 | return 0; | |
986 | ||
987 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
988 | if ((*start && *start != key.offset) || (*end && *end != extent_end)) | |
989 | return 0; | |
990 | ||
991 | *start = key.offset; | |
992 | *end = extent_end; | |
993 | return 1; | |
994 | } | |
995 | ||
996 | /* | |
997 | * Mark extent in the range start - end as written. | |
998 | * | |
999 | * This changes extent type from 'pre-allocated' to 'regular'. If only | |
1000 | * part of extent is marked as written, the extent will be split into | |
1001 | * two or three. | |
1002 | */ | |
1003 | int btrfs_mark_extent_written(struct btrfs_trans_handle *trans, | |
d899e052 YZ |
1004 | struct inode *inode, u64 start, u64 end) |
1005 | { | |
920bbbfb | 1006 | struct btrfs_root *root = BTRFS_I(inode)->root; |
d899e052 YZ |
1007 | struct extent_buffer *leaf; |
1008 | struct btrfs_path *path; | |
1009 | struct btrfs_file_extent_item *fi; | |
1010 | struct btrfs_key key; | |
920bbbfb | 1011 | struct btrfs_key new_key; |
d899e052 YZ |
1012 | u64 bytenr; |
1013 | u64 num_bytes; | |
1014 | u64 extent_end; | |
5d4f98a2 | 1015 | u64 orig_offset; |
d899e052 YZ |
1016 | u64 other_start; |
1017 | u64 other_end; | |
920bbbfb YZ |
1018 | u64 split; |
1019 | int del_nr = 0; | |
1020 | int del_slot = 0; | |
6c7d54ac | 1021 | int recow; |
d899e052 | 1022 | int ret; |
33345d01 | 1023 | u64 ino = btrfs_ino(inode); |
d899e052 | 1024 | |
d899e052 | 1025 | path = btrfs_alloc_path(); |
d8926bb3 MF |
1026 | if (!path) |
1027 | return -ENOMEM; | |
d899e052 | 1028 | again: |
6c7d54ac | 1029 | recow = 0; |
920bbbfb | 1030 | split = start; |
33345d01 | 1031 | key.objectid = ino; |
d899e052 | 1032 | key.type = BTRFS_EXTENT_DATA_KEY; |
920bbbfb | 1033 | key.offset = split; |
d899e052 YZ |
1034 | |
1035 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
41415730 JB |
1036 | if (ret < 0) |
1037 | goto out; | |
d899e052 YZ |
1038 | if (ret > 0 && path->slots[0] > 0) |
1039 | path->slots[0]--; | |
1040 | ||
1041 | leaf = path->nodes[0]; | |
1042 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
33345d01 | 1043 | BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY); |
d899e052 YZ |
1044 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1045 | struct btrfs_file_extent_item); | |
920bbbfb YZ |
1046 | BUG_ON(btrfs_file_extent_type(leaf, fi) != |
1047 | BTRFS_FILE_EXTENT_PREALLOC); | |
d899e052 YZ |
1048 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
1049 | BUG_ON(key.offset > start || extent_end < end); | |
1050 | ||
1051 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
1052 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
5d4f98a2 | 1053 | orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi); |
6c7d54ac YZ |
1054 | memcpy(&new_key, &key, sizeof(new_key)); |
1055 | ||
1056 | if (start == key.offset && end < extent_end) { | |
1057 | other_start = 0; | |
1058 | other_end = start; | |
1059 | if (extent_mergeable(leaf, path->slots[0] - 1, | |
33345d01 | 1060 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1061 | &other_start, &other_end)) { |
1062 | new_key.offset = end; | |
afe5fea7 | 1063 | btrfs_set_item_key_safe(root, path, &new_key); |
6c7d54ac YZ |
1064 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1065 | struct btrfs_file_extent_item); | |
224ecce5 JB |
1066 | btrfs_set_file_extent_generation(leaf, fi, |
1067 | trans->transid); | |
6c7d54ac YZ |
1068 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1069 | extent_end - end); | |
1070 | btrfs_set_file_extent_offset(leaf, fi, | |
1071 | end - orig_offset); | |
1072 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
1073 | struct btrfs_file_extent_item); | |
224ecce5 JB |
1074 | btrfs_set_file_extent_generation(leaf, fi, |
1075 | trans->transid); | |
6c7d54ac YZ |
1076 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1077 | end - other_start); | |
1078 | btrfs_mark_buffer_dirty(leaf); | |
1079 | goto out; | |
1080 | } | |
1081 | } | |
1082 | ||
1083 | if (start > key.offset && end == extent_end) { | |
1084 | other_start = end; | |
1085 | other_end = 0; | |
1086 | if (extent_mergeable(leaf, path->slots[0] + 1, | |
33345d01 | 1087 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1088 | &other_start, &other_end)) { |
1089 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1090 | struct btrfs_file_extent_item); | |
1091 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
1092 | start - key.offset); | |
224ecce5 JB |
1093 | btrfs_set_file_extent_generation(leaf, fi, |
1094 | trans->transid); | |
6c7d54ac YZ |
1095 | path->slots[0]++; |
1096 | new_key.offset = start; | |
afe5fea7 | 1097 | btrfs_set_item_key_safe(root, path, &new_key); |
6c7d54ac YZ |
1098 | |
1099 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1100 | struct btrfs_file_extent_item); | |
224ecce5 JB |
1101 | btrfs_set_file_extent_generation(leaf, fi, |
1102 | trans->transid); | |
6c7d54ac YZ |
1103 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1104 | other_end - start); | |
1105 | btrfs_set_file_extent_offset(leaf, fi, | |
1106 | start - orig_offset); | |
1107 | btrfs_mark_buffer_dirty(leaf); | |
1108 | goto out; | |
1109 | } | |
1110 | } | |
d899e052 | 1111 | |
920bbbfb YZ |
1112 | while (start > key.offset || end < extent_end) { |
1113 | if (key.offset == start) | |
1114 | split = end; | |
1115 | ||
920bbbfb YZ |
1116 | new_key.offset = split; |
1117 | ret = btrfs_duplicate_item(trans, root, path, &new_key); | |
1118 | if (ret == -EAGAIN) { | |
b3b4aa74 | 1119 | btrfs_release_path(path); |
920bbbfb | 1120 | goto again; |
d899e052 | 1121 | } |
79787eaa JM |
1122 | if (ret < 0) { |
1123 | btrfs_abort_transaction(trans, root, ret); | |
1124 | goto out; | |
1125 | } | |
d899e052 | 1126 | |
920bbbfb YZ |
1127 | leaf = path->nodes[0]; |
1128 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
d899e052 | 1129 | struct btrfs_file_extent_item); |
224ecce5 | 1130 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
d899e052 | 1131 | btrfs_set_file_extent_num_bytes(leaf, fi, |
920bbbfb YZ |
1132 | split - key.offset); |
1133 | ||
1134 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1135 | struct btrfs_file_extent_item); | |
1136 | ||
224ecce5 | 1137 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
920bbbfb YZ |
1138 | btrfs_set_file_extent_offset(leaf, fi, split - orig_offset); |
1139 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
1140 | extent_end - split); | |
d899e052 YZ |
1141 | btrfs_mark_buffer_dirty(leaf); |
1142 | ||
920bbbfb YZ |
1143 | ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0, |
1144 | root->root_key.objectid, | |
66d7e7f0 | 1145 | ino, orig_offset, 0); |
79787eaa | 1146 | BUG_ON(ret); /* -ENOMEM */ |
d899e052 | 1147 | |
920bbbfb YZ |
1148 | if (split == start) { |
1149 | key.offset = start; | |
1150 | } else { | |
1151 | BUG_ON(start != key.offset); | |
d899e052 | 1152 | path->slots[0]--; |
920bbbfb | 1153 | extent_end = end; |
d899e052 | 1154 | } |
6c7d54ac | 1155 | recow = 1; |
d899e052 YZ |
1156 | } |
1157 | ||
920bbbfb YZ |
1158 | other_start = end; |
1159 | other_end = 0; | |
6c7d54ac | 1160 | if (extent_mergeable(leaf, path->slots[0] + 1, |
33345d01 | 1161 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1162 | &other_start, &other_end)) { |
1163 | if (recow) { | |
b3b4aa74 | 1164 | btrfs_release_path(path); |
6c7d54ac YZ |
1165 | goto again; |
1166 | } | |
920bbbfb YZ |
1167 | extent_end = other_end; |
1168 | del_slot = path->slots[0] + 1; | |
1169 | del_nr++; | |
1170 | ret = btrfs_free_extent(trans, root, bytenr, num_bytes, | |
1171 | 0, root->root_key.objectid, | |
66d7e7f0 | 1172 | ino, orig_offset, 0); |
79787eaa | 1173 | BUG_ON(ret); /* -ENOMEM */ |
d899e052 | 1174 | } |
920bbbfb YZ |
1175 | other_start = 0; |
1176 | other_end = start; | |
6c7d54ac | 1177 | if (extent_mergeable(leaf, path->slots[0] - 1, |
33345d01 | 1178 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1179 | &other_start, &other_end)) { |
1180 | if (recow) { | |
b3b4aa74 | 1181 | btrfs_release_path(path); |
6c7d54ac YZ |
1182 | goto again; |
1183 | } | |
920bbbfb YZ |
1184 | key.offset = other_start; |
1185 | del_slot = path->slots[0]; | |
1186 | del_nr++; | |
1187 | ret = btrfs_free_extent(trans, root, bytenr, num_bytes, | |
1188 | 0, root->root_key.objectid, | |
66d7e7f0 | 1189 | ino, orig_offset, 0); |
79787eaa | 1190 | BUG_ON(ret); /* -ENOMEM */ |
920bbbfb YZ |
1191 | } |
1192 | if (del_nr == 0) { | |
3f6fae95 SL |
1193 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1194 | struct btrfs_file_extent_item); | |
920bbbfb YZ |
1195 | btrfs_set_file_extent_type(leaf, fi, |
1196 | BTRFS_FILE_EXTENT_REG); | |
224ecce5 | 1197 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
920bbbfb | 1198 | btrfs_mark_buffer_dirty(leaf); |
6c7d54ac | 1199 | } else { |
3f6fae95 SL |
1200 | fi = btrfs_item_ptr(leaf, del_slot - 1, |
1201 | struct btrfs_file_extent_item); | |
6c7d54ac YZ |
1202 | btrfs_set_file_extent_type(leaf, fi, |
1203 | BTRFS_FILE_EXTENT_REG); | |
224ecce5 | 1204 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
6c7d54ac YZ |
1205 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1206 | extent_end - key.offset); | |
1207 | btrfs_mark_buffer_dirty(leaf); | |
920bbbfb | 1208 | |
6c7d54ac | 1209 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
79787eaa JM |
1210 | if (ret < 0) { |
1211 | btrfs_abort_transaction(trans, root, ret); | |
1212 | goto out; | |
1213 | } | |
6c7d54ac | 1214 | } |
920bbbfb | 1215 | out: |
d899e052 YZ |
1216 | btrfs_free_path(path); |
1217 | return 0; | |
1218 | } | |
1219 | ||
b1bf862e CM |
1220 | /* |
1221 | * on error we return an unlocked page and the error value | |
1222 | * on success we return a locked page and 0 | |
1223 | */ | |
b6316429 JB |
1224 | static int prepare_uptodate_page(struct page *page, u64 pos, |
1225 | bool force_uptodate) | |
b1bf862e CM |
1226 | { |
1227 | int ret = 0; | |
1228 | ||
b6316429 JB |
1229 | if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) && |
1230 | !PageUptodate(page)) { | |
b1bf862e CM |
1231 | ret = btrfs_readpage(NULL, page); |
1232 | if (ret) | |
1233 | return ret; | |
1234 | lock_page(page); | |
1235 | if (!PageUptodate(page)) { | |
1236 | unlock_page(page); | |
1237 | return -EIO; | |
1238 | } | |
1239 | } | |
1240 | return 0; | |
1241 | } | |
1242 | ||
39279cc3 | 1243 | /* |
d352ac68 CM |
1244 | * this gets pages into the page cache and locks them down, it also properly |
1245 | * waits for data=ordered extents to finish before allowing the pages to be | |
1246 | * modified. | |
39279cc3 | 1247 | */ |
d397712b | 1248 | static noinline int prepare_pages(struct btrfs_root *root, struct file *file, |
98ed5174 CM |
1249 | struct page **pages, size_t num_pages, |
1250 | loff_t pos, unsigned long first_index, | |
b6316429 | 1251 | size_t write_bytes, bool force_uptodate) |
39279cc3 | 1252 | { |
2ac55d41 | 1253 | struct extent_state *cached_state = NULL; |
39279cc3 CM |
1254 | int i; |
1255 | unsigned long index = pos >> PAGE_CACHE_SHIFT; | |
496ad9aa | 1256 | struct inode *inode = file_inode(file); |
3b16a4e3 | 1257 | gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); |
39279cc3 | 1258 | int err = 0; |
b1bf862e | 1259 | int faili = 0; |
8c2383c3 | 1260 | u64 start_pos; |
e6dcd2dc | 1261 | u64 last_pos; |
8c2383c3 | 1262 | |
5f39d397 | 1263 | start_pos = pos & ~((u64)root->sectorsize - 1); |
e6dcd2dc | 1264 | last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT; |
39279cc3 | 1265 | |
e6dcd2dc | 1266 | again: |
39279cc3 | 1267 | for (i = 0; i < num_pages; i++) { |
a94733d0 | 1268 | pages[i] = find_or_create_page(inode->i_mapping, index + i, |
e3a41a5b | 1269 | mask | __GFP_WRITE); |
39279cc3 | 1270 | if (!pages[i]) { |
b1bf862e CM |
1271 | faili = i - 1; |
1272 | err = -ENOMEM; | |
1273 | goto fail; | |
1274 | } | |
1275 | ||
1276 | if (i == 0) | |
b6316429 JB |
1277 | err = prepare_uptodate_page(pages[i], pos, |
1278 | force_uptodate); | |
b1bf862e CM |
1279 | if (i == num_pages - 1) |
1280 | err = prepare_uptodate_page(pages[i], | |
b6316429 | 1281 | pos + write_bytes, false); |
b1bf862e CM |
1282 | if (err) { |
1283 | page_cache_release(pages[i]); | |
1284 | faili = i - 1; | |
1285 | goto fail; | |
39279cc3 | 1286 | } |
ccd467d6 | 1287 | wait_on_page_writeback(pages[i]); |
39279cc3 | 1288 | } |
b1bf862e | 1289 | err = 0; |
0762704b | 1290 | if (start_pos < inode->i_size) { |
e6dcd2dc | 1291 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 1292 | lock_extent_bits(&BTRFS_I(inode)->io_tree, |
d0082371 | 1293 | start_pos, last_pos - 1, 0, &cached_state); |
d397712b CM |
1294 | ordered = btrfs_lookup_first_ordered_extent(inode, |
1295 | last_pos - 1); | |
e6dcd2dc CM |
1296 | if (ordered && |
1297 | ordered->file_offset + ordered->len > start_pos && | |
1298 | ordered->file_offset < last_pos) { | |
1299 | btrfs_put_ordered_extent(ordered); | |
2ac55d41 JB |
1300 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, |
1301 | start_pos, last_pos - 1, | |
1302 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
1303 | for (i = 0; i < num_pages; i++) { |
1304 | unlock_page(pages[i]); | |
1305 | page_cache_release(pages[i]); | |
1306 | } | |
1307 | btrfs_wait_ordered_range(inode, start_pos, | |
1308 | last_pos - start_pos); | |
1309 | goto again; | |
1310 | } | |
1311 | if (ordered) | |
1312 | btrfs_put_ordered_extent(ordered); | |
1313 | ||
2ac55d41 | 1314 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos, |
32c00aff | 1315 | last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC | |
9e8a4a8b LB |
1316 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, |
1317 | 0, 0, &cached_state, GFP_NOFS); | |
2ac55d41 JB |
1318 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, |
1319 | start_pos, last_pos - 1, &cached_state, | |
1320 | GFP_NOFS); | |
0762704b | 1321 | } |
e6dcd2dc | 1322 | for (i = 0; i < num_pages; i++) { |
32c7f202 WF |
1323 | if (clear_page_dirty_for_io(pages[i])) |
1324 | account_page_redirty(pages[i]); | |
e6dcd2dc CM |
1325 | set_page_extent_mapped(pages[i]); |
1326 | WARN_ON(!PageLocked(pages[i])); | |
1327 | } | |
39279cc3 | 1328 | return 0; |
b1bf862e CM |
1329 | fail: |
1330 | while (faili >= 0) { | |
1331 | unlock_page(pages[faili]); | |
1332 | page_cache_release(pages[faili]); | |
1333 | faili--; | |
1334 | } | |
1335 | return err; | |
1336 | ||
39279cc3 CM |
1337 | } |
1338 | ||
7ee9e440 JB |
1339 | static noinline int check_can_nocow(struct inode *inode, loff_t pos, |
1340 | size_t *write_bytes) | |
1341 | { | |
7ee9e440 JB |
1342 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1343 | struct btrfs_ordered_extent *ordered; | |
1344 | u64 lockstart, lockend; | |
1345 | u64 num_bytes; | |
1346 | int ret; | |
1347 | ||
1348 | lockstart = round_down(pos, root->sectorsize); | |
1349 | lockend = lockstart + round_up(*write_bytes, root->sectorsize) - 1; | |
1350 | ||
1351 | while (1) { | |
1352 | lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend); | |
1353 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
1354 | lockend - lockstart + 1); | |
1355 | if (!ordered) { | |
1356 | break; | |
1357 | } | |
1358 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend); | |
1359 | btrfs_start_ordered_extent(inode, ordered, 1); | |
1360 | btrfs_put_ordered_extent(ordered); | |
1361 | } | |
1362 | ||
7ee9e440 | 1363 | num_bytes = lockend - lockstart + 1; |
00361589 | 1364 | ret = can_nocow_extent(inode, lockstart, &num_bytes, NULL, NULL, NULL); |
7ee9e440 JB |
1365 | if (ret <= 0) { |
1366 | ret = 0; | |
1367 | } else { | |
1368 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
1369 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
1370 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0, | |
1371 | NULL, GFP_NOFS); | |
1372 | *write_bytes = min_t(size_t, *write_bytes, num_bytes); | |
1373 | } | |
1374 | ||
1375 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend); | |
1376 | ||
1377 | return ret; | |
1378 | } | |
1379 | ||
d0215f3e JB |
1380 | static noinline ssize_t __btrfs_buffered_write(struct file *file, |
1381 | struct iov_iter *i, | |
1382 | loff_t pos) | |
4b46fce2 | 1383 | { |
496ad9aa | 1384 | struct inode *inode = file_inode(file); |
11c65dcc | 1385 | struct btrfs_root *root = BTRFS_I(inode)->root; |
11c65dcc | 1386 | struct page **pages = NULL; |
7ee9e440 | 1387 | u64 release_bytes = 0; |
39279cc3 | 1388 | unsigned long first_index; |
d0215f3e JB |
1389 | size_t num_written = 0; |
1390 | int nrptrs; | |
c9149235 | 1391 | int ret = 0; |
7ee9e440 | 1392 | bool only_release_metadata = false; |
b6316429 | 1393 | bool force_page_uptodate = false; |
4b46fce2 | 1394 | |
d0215f3e | 1395 | nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) / |
11c65dcc JB |
1396 | PAGE_CACHE_SIZE, PAGE_CACHE_SIZE / |
1397 | (sizeof(struct page *))); | |
142349f5 WF |
1398 | nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied); |
1399 | nrptrs = max(nrptrs, 8); | |
8c2383c3 | 1400 | pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL); |
d0215f3e JB |
1401 | if (!pages) |
1402 | return -ENOMEM; | |
ab93dbec | 1403 | |
39279cc3 | 1404 | first_index = pos >> PAGE_CACHE_SHIFT; |
39279cc3 | 1405 | |
d0215f3e | 1406 | while (iov_iter_count(i) > 0) { |
39279cc3 | 1407 | size_t offset = pos & (PAGE_CACHE_SIZE - 1); |
d0215f3e | 1408 | size_t write_bytes = min(iov_iter_count(i), |
11c65dcc | 1409 | nrptrs * (size_t)PAGE_CACHE_SIZE - |
8c2383c3 | 1410 | offset); |
3a90983d YZ |
1411 | size_t num_pages = (write_bytes + offset + |
1412 | PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
7ee9e440 | 1413 | size_t reserve_bytes; |
d0215f3e JB |
1414 | size_t dirty_pages; |
1415 | size_t copied; | |
39279cc3 | 1416 | |
8c2383c3 | 1417 | WARN_ON(num_pages > nrptrs); |
1832a6d5 | 1418 | |
914ee295 XZ |
1419 | /* |
1420 | * Fault pages before locking them in prepare_pages | |
1421 | * to avoid recursive lock | |
1422 | */ | |
d0215f3e | 1423 | if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) { |
914ee295 | 1424 | ret = -EFAULT; |
d0215f3e | 1425 | break; |
914ee295 XZ |
1426 | } |
1427 | ||
7ee9e440 JB |
1428 | reserve_bytes = num_pages << PAGE_CACHE_SHIFT; |
1429 | ret = btrfs_check_data_free_space(inode, reserve_bytes); | |
1430 | if (ret == -ENOSPC && | |
1431 | (BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW | | |
1432 | BTRFS_INODE_PREALLOC))) { | |
1433 | ret = check_can_nocow(inode, pos, &write_bytes); | |
1434 | if (ret > 0) { | |
1435 | only_release_metadata = true; | |
1436 | /* | |
1437 | * our prealloc extent may be smaller than | |
1438 | * write_bytes, so scale down. | |
1439 | */ | |
1440 | num_pages = (write_bytes + offset + | |
1441 | PAGE_CACHE_SIZE - 1) >> | |
1442 | PAGE_CACHE_SHIFT; | |
1443 | reserve_bytes = num_pages << PAGE_CACHE_SHIFT; | |
1444 | ret = 0; | |
1445 | } else { | |
1446 | ret = -ENOSPC; | |
1447 | } | |
1448 | } | |
1449 | ||
1832a6d5 | 1450 | if (ret) |
d0215f3e | 1451 | break; |
1832a6d5 | 1452 | |
7ee9e440 JB |
1453 | ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes); |
1454 | if (ret) { | |
1455 | if (!only_release_metadata) | |
1456 | btrfs_free_reserved_data_space(inode, | |
1457 | reserve_bytes); | |
1458 | break; | |
1459 | } | |
1460 | ||
1461 | release_bytes = reserve_bytes; | |
1462 | ||
4a64001f JB |
1463 | /* |
1464 | * This is going to setup the pages array with the number of | |
1465 | * pages we want, so we don't really need to worry about the | |
1466 | * contents of pages from loop to loop | |
1467 | */ | |
39279cc3 | 1468 | ret = prepare_pages(root, file, pages, num_pages, |
b6316429 JB |
1469 | pos, first_index, write_bytes, |
1470 | force_page_uptodate); | |
7ee9e440 | 1471 | if (ret) |
d0215f3e | 1472 | break; |
39279cc3 | 1473 | |
914ee295 | 1474 | copied = btrfs_copy_from_user(pos, num_pages, |
d0215f3e | 1475 | write_bytes, pages, i); |
b1bf862e CM |
1476 | |
1477 | /* | |
1478 | * if we have trouble faulting in the pages, fall | |
1479 | * back to one page at a time | |
1480 | */ | |
1481 | if (copied < write_bytes) | |
1482 | nrptrs = 1; | |
1483 | ||
b6316429 JB |
1484 | if (copied == 0) { |
1485 | force_page_uptodate = true; | |
b1bf862e | 1486 | dirty_pages = 0; |
b6316429 JB |
1487 | } else { |
1488 | force_page_uptodate = false; | |
b1bf862e CM |
1489 | dirty_pages = (copied + offset + |
1490 | PAGE_CACHE_SIZE - 1) >> | |
1491 | PAGE_CACHE_SHIFT; | |
b6316429 | 1492 | } |
914ee295 | 1493 | |
d0215f3e JB |
1494 | /* |
1495 | * If we had a short copy we need to release the excess delaloc | |
1496 | * bytes we reserved. We need to increment outstanding_extents | |
1497 | * because btrfs_delalloc_release_space will decrement it, but | |
1498 | * we still have an outstanding extent for the chunk we actually | |
1499 | * managed to copy. | |
1500 | */ | |
914ee295 | 1501 | if (num_pages > dirty_pages) { |
7ee9e440 JB |
1502 | release_bytes = (num_pages - dirty_pages) << |
1503 | PAGE_CACHE_SHIFT; | |
9e0baf60 JB |
1504 | if (copied > 0) { |
1505 | spin_lock(&BTRFS_I(inode)->lock); | |
1506 | BTRFS_I(inode)->outstanding_extents++; | |
1507 | spin_unlock(&BTRFS_I(inode)->lock); | |
1508 | } | |
7ee9e440 JB |
1509 | if (only_release_metadata) |
1510 | btrfs_delalloc_release_metadata(inode, | |
1511 | release_bytes); | |
1512 | else | |
1513 | btrfs_delalloc_release_space(inode, | |
1514 | release_bytes); | |
914ee295 XZ |
1515 | } |
1516 | ||
7ee9e440 | 1517 | release_bytes = dirty_pages << PAGE_CACHE_SHIFT; |
914ee295 | 1518 | if (copied > 0) { |
be1a12a0 JB |
1519 | ret = btrfs_dirty_pages(root, inode, pages, |
1520 | dirty_pages, pos, copied, | |
1521 | NULL); | |
d0215f3e | 1522 | if (ret) { |
d0215f3e JB |
1523 | btrfs_drop_pages(pages, num_pages); |
1524 | break; | |
1525 | } | |
54aa1f4d | 1526 | } |
39279cc3 | 1527 | |
7ee9e440 | 1528 | release_bytes = 0; |
39279cc3 CM |
1529 | btrfs_drop_pages(pages, num_pages); |
1530 | ||
7ee9e440 JB |
1531 | if (only_release_metadata && copied > 0) { |
1532 | u64 lockstart = round_down(pos, root->sectorsize); | |
1533 | u64 lockend = lockstart + | |
1534 | (dirty_pages << PAGE_CACHE_SHIFT) - 1; | |
1535 | ||
1536 | set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
1537 | lockend, EXTENT_NORESERVE, NULL, | |
1538 | NULL, GFP_NOFS); | |
1539 | only_release_metadata = false; | |
1540 | } | |
1541 | ||
d0215f3e JB |
1542 | cond_resched(); |
1543 | ||
d0e1d66b | 1544 | balance_dirty_pages_ratelimited(inode->i_mapping); |
d0215f3e | 1545 | if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1) |
b53d3f5d | 1546 | btrfs_btree_balance_dirty(root); |
cb843a6f | 1547 | |
914ee295 XZ |
1548 | pos += copied; |
1549 | num_written += copied; | |
d0215f3e | 1550 | } |
39279cc3 | 1551 | |
d0215f3e JB |
1552 | kfree(pages); |
1553 | ||
7ee9e440 JB |
1554 | if (release_bytes) { |
1555 | if (only_release_metadata) | |
1556 | btrfs_delalloc_release_metadata(inode, release_bytes); | |
1557 | else | |
1558 | btrfs_delalloc_release_space(inode, release_bytes); | |
1559 | } | |
1560 | ||
d0215f3e JB |
1561 | return num_written ? num_written : ret; |
1562 | } | |
1563 | ||
1564 | static ssize_t __btrfs_direct_write(struct kiocb *iocb, | |
1565 | const struct iovec *iov, | |
1566 | unsigned long nr_segs, loff_t pos, | |
1567 | loff_t *ppos, size_t count, size_t ocount) | |
1568 | { | |
1569 | struct file *file = iocb->ki_filp; | |
d0215f3e JB |
1570 | struct iov_iter i; |
1571 | ssize_t written; | |
1572 | ssize_t written_buffered; | |
1573 | loff_t endbyte; | |
1574 | int err; | |
1575 | ||
1576 | written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos, | |
1577 | count, ocount); | |
1578 | ||
d0215f3e JB |
1579 | if (written < 0 || written == count) |
1580 | return written; | |
1581 | ||
1582 | pos += written; | |
1583 | count -= written; | |
1584 | iov_iter_init(&i, iov, nr_segs, count, written); | |
1585 | written_buffered = __btrfs_buffered_write(file, &i, pos); | |
1586 | if (written_buffered < 0) { | |
1587 | err = written_buffered; | |
1588 | goto out; | |
39279cc3 | 1589 | } |
d0215f3e JB |
1590 | endbyte = pos + written_buffered - 1; |
1591 | err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte); | |
1592 | if (err) | |
1593 | goto out; | |
1594 | written += written_buffered; | |
1595 | *ppos = pos + written_buffered; | |
1596 | invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT, | |
1597 | endbyte >> PAGE_CACHE_SHIFT); | |
39279cc3 | 1598 | out: |
d0215f3e JB |
1599 | return written ? written : err; |
1600 | } | |
5b92ee72 | 1601 | |
6c760c07 JB |
1602 | static void update_time_for_write(struct inode *inode) |
1603 | { | |
1604 | struct timespec now; | |
1605 | ||
1606 | if (IS_NOCMTIME(inode)) | |
1607 | return; | |
1608 | ||
1609 | now = current_fs_time(inode->i_sb); | |
1610 | if (!timespec_equal(&inode->i_mtime, &now)) | |
1611 | inode->i_mtime = now; | |
1612 | ||
1613 | if (!timespec_equal(&inode->i_ctime, &now)) | |
1614 | inode->i_ctime = now; | |
1615 | ||
1616 | if (IS_I_VERSION(inode)) | |
1617 | inode_inc_iversion(inode); | |
1618 | } | |
1619 | ||
d0215f3e JB |
1620 | static ssize_t btrfs_file_aio_write(struct kiocb *iocb, |
1621 | const struct iovec *iov, | |
1622 | unsigned long nr_segs, loff_t pos) | |
1623 | { | |
1624 | struct file *file = iocb->ki_filp; | |
496ad9aa | 1625 | struct inode *inode = file_inode(file); |
d0215f3e JB |
1626 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1627 | loff_t *ppos = &iocb->ki_pos; | |
0c1a98c8 | 1628 | u64 start_pos; |
d0215f3e JB |
1629 | ssize_t num_written = 0; |
1630 | ssize_t err = 0; | |
1631 | size_t count, ocount; | |
b812ce28 | 1632 | bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host); |
d0215f3e | 1633 | |
d0215f3e JB |
1634 | mutex_lock(&inode->i_mutex); |
1635 | ||
1636 | err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ); | |
1637 | if (err) { | |
1638 | mutex_unlock(&inode->i_mutex); | |
1639 | goto out; | |
1640 | } | |
1641 | count = ocount; | |
1642 | ||
1643 | current->backing_dev_info = inode->i_mapping->backing_dev_info; | |
1644 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); | |
1645 | if (err) { | |
1646 | mutex_unlock(&inode->i_mutex); | |
1647 | goto out; | |
1648 | } | |
1649 | ||
1650 | if (count == 0) { | |
1651 | mutex_unlock(&inode->i_mutex); | |
1652 | goto out; | |
1653 | } | |
1654 | ||
1655 | err = file_remove_suid(file); | |
1656 | if (err) { | |
1657 | mutex_unlock(&inode->i_mutex); | |
1658 | goto out; | |
1659 | } | |
1660 | ||
1661 | /* | |
1662 | * If BTRFS flips readonly due to some impossible error | |
1663 | * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR), | |
1664 | * although we have opened a file as writable, we have | |
1665 | * to stop this write operation to ensure FS consistency. | |
1666 | */ | |
87533c47 | 1667 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) { |
d0215f3e JB |
1668 | mutex_unlock(&inode->i_mutex); |
1669 | err = -EROFS; | |
1670 | goto out; | |
1671 | } | |
1672 | ||
6c760c07 JB |
1673 | /* |
1674 | * We reserve space for updating the inode when we reserve space for the | |
1675 | * extent we are going to write, so we will enospc out there. We don't | |
1676 | * need to start yet another transaction to update the inode as we will | |
1677 | * update the inode when we finish writing whatever data we write. | |
1678 | */ | |
1679 | update_time_for_write(inode); | |
d0215f3e | 1680 | |
0c1a98c8 MX |
1681 | start_pos = round_down(pos, root->sectorsize); |
1682 | if (start_pos > i_size_read(inode)) { | |
1683 | err = btrfs_cont_expand(inode, i_size_read(inode), start_pos); | |
1684 | if (err) { | |
1685 | mutex_unlock(&inode->i_mutex); | |
1686 | goto out; | |
1687 | } | |
1688 | } | |
1689 | ||
b812ce28 JB |
1690 | if (sync) |
1691 | atomic_inc(&BTRFS_I(inode)->sync_writers); | |
1692 | ||
d0215f3e JB |
1693 | if (unlikely(file->f_flags & O_DIRECT)) { |
1694 | num_written = __btrfs_direct_write(iocb, iov, nr_segs, | |
1695 | pos, ppos, count, ocount); | |
1696 | } else { | |
1697 | struct iov_iter i; | |
1698 | ||
1699 | iov_iter_init(&i, iov, nr_segs, count, num_written); | |
1700 | ||
1701 | num_written = __btrfs_buffered_write(file, &i, pos); | |
1702 | if (num_written > 0) | |
1703 | *ppos = pos + num_written; | |
1704 | } | |
1705 | ||
1706 | mutex_unlock(&inode->i_mutex); | |
2ff3e9b6 | 1707 | |
5a3f23d5 CM |
1708 | /* |
1709 | * we want to make sure fsync finds this change | |
1710 | * but we haven't joined a transaction running right now. | |
1711 | * | |
1712 | * Later on, someone is sure to update the inode and get the | |
1713 | * real transid recorded. | |
1714 | * | |
1715 | * We set last_trans now to the fs_info generation + 1, | |
1716 | * this will either be one more than the running transaction | |
1717 | * or the generation used for the next transaction if there isn't | |
1718 | * one running right now. | |
6c760c07 JB |
1719 | * |
1720 | * We also have to set last_sub_trans to the current log transid, | |
1721 | * otherwise subsequent syncs to a file that's been synced in this | |
1722 | * transaction will appear to have already occured. | |
5a3f23d5 CM |
1723 | */ |
1724 | BTRFS_I(inode)->last_trans = root->fs_info->generation + 1; | |
6c760c07 | 1725 | BTRFS_I(inode)->last_sub_trans = root->log_transid; |
d0215f3e JB |
1726 | if (num_written > 0 || num_written == -EIOCBQUEUED) { |
1727 | err = generic_write_sync(file, pos, num_written); | |
1728 | if (err < 0 && num_written > 0) | |
2ff3e9b6 CM |
1729 | num_written = err; |
1730 | } | |
0a3404dc | 1731 | |
b812ce28 JB |
1732 | if (sync) |
1733 | atomic_dec(&BTRFS_I(inode)->sync_writers); | |
0a3404dc | 1734 | out: |
39279cc3 | 1735 | current->backing_dev_info = NULL; |
39279cc3 CM |
1736 | return num_written ? num_written : err; |
1737 | } | |
1738 | ||
d397712b | 1739 | int btrfs_release_file(struct inode *inode, struct file *filp) |
e1b81e67 | 1740 | { |
5a3f23d5 CM |
1741 | /* |
1742 | * ordered_data_close is set by settattr when we are about to truncate | |
1743 | * a file from a non-zero size to a zero size. This tries to | |
1744 | * flush down new bytes that may have been written if the | |
1745 | * application were using truncate to replace a file in place. | |
1746 | */ | |
72ac3c0d JB |
1747 | if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
1748 | &BTRFS_I(inode)->runtime_flags)) { | |
569e0f35 JB |
1749 | struct btrfs_trans_handle *trans; |
1750 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1751 | ||
1752 | /* | |
1753 | * We need to block on a committing transaction to keep us from | |
1754 | * throwing a ordered operation on to the list and causing | |
1755 | * something like sync to deadlock trying to flush out this | |
1756 | * inode. | |
1757 | */ | |
1758 | trans = btrfs_start_transaction(root, 0); | |
1759 | if (IS_ERR(trans)) | |
1760 | return PTR_ERR(trans); | |
1761 | btrfs_add_ordered_operation(trans, BTRFS_I(inode)->root, inode); | |
1762 | btrfs_end_transaction(trans, root); | |
5a3f23d5 CM |
1763 | if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) |
1764 | filemap_flush(inode->i_mapping); | |
1765 | } | |
6bf13c0c SW |
1766 | if (filp->private_data) |
1767 | btrfs_ioctl_trans_end(filp); | |
e1b81e67 M |
1768 | return 0; |
1769 | } | |
1770 | ||
d352ac68 CM |
1771 | /* |
1772 | * fsync call for both files and directories. This logs the inode into | |
1773 | * the tree log instead of forcing full commits whenever possible. | |
1774 | * | |
1775 | * It needs to call filemap_fdatawait so that all ordered extent updates are | |
1776 | * in the metadata btree are up to date for copying to the log. | |
1777 | * | |
1778 | * It drops the inode mutex before doing the tree log commit. This is an | |
1779 | * important optimization for directories because holding the mutex prevents | |
1780 | * new operations on the dir while we write to disk. | |
1781 | */ | |
02c24a82 | 1782 | int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync) |
39279cc3 | 1783 | { |
7ea80859 | 1784 | struct dentry *dentry = file->f_path.dentry; |
39279cc3 CM |
1785 | struct inode *inode = dentry->d_inode; |
1786 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
15ee9bc7 | 1787 | int ret = 0; |
39279cc3 | 1788 | struct btrfs_trans_handle *trans; |
2ab28f32 | 1789 | bool full_sync = 0; |
39279cc3 | 1790 | |
1abe9b8a | 1791 | trace_btrfs_sync_file(file, datasync); |
257c62e1 | 1792 | |
90abccf2 MX |
1793 | /* |
1794 | * We write the dirty pages in the range and wait until they complete | |
1795 | * out of the ->i_mutex. If so, we can flush the dirty pages by | |
2ab28f32 JB |
1796 | * multi-task, and make the performance up. See |
1797 | * btrfs_wait_ordered_range for an explanation of the ASYNC check. | |
90abccf2 | 1798 | */ |
b812ce28 | 1799 | atomic_inc(&BTRFS_I(inode)->sync_writers); |
2ab28f32 JB |
1800 | ret = filemap_fdatawrite_range(inode->i_mapping, start, end); |
1801 | if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
1802 | &BTRFS_I(inode)->runtime_flags)) | |
1803 | ret = filemap_fdatawrite_range(inode->i_mapping, start, end); | |
b812ce28 | 1804 | atomic_dec(&BTRFS_I(inode)->sync_writers); |
90abccf2 MX |
1805 | if (ret) |
1806 | return ret; | |
1807 | ||
02c24a82 JB |
1808 | mutex_lock(&inode->i_mutex); |
1809 | ||
0885ef5b | 1810 | /* |
90abccf2 MX |
1811 | * We flush the dirty pages again to avoid some dirty pages in the |
1812 | * range being left. | |
0885ef5b | 1813 | */ |
2ecb7923 | 1814 | atomic_inc(&root->log_batch); |
2ab28f32 JB |
1815 | full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
1816 | &BTRFS_I(inode)->runtime_flags); | |
1817 | if (full_sync) | |
1818 | btrfs_wait_ordered_range(inode, start, end - start + 1); | |
2ecb7923 | 1819 | atomic_inc(&root->log_batch); |
257c62e1 | 1820 | |
39279cc3 | 1821 | /* |
15ee9bc7 JB |
1822 | * check the transaction that last modified this inode |
1823 | * and see if its already been committed | |
39279cc3 | 1824 | */ |
02c24a82 JB |
1825 | if (!BTRFS_I(inode)->last_trans) { |
1826 | mutex_unlock(&inode->i_mutex); | |
15ee9bc7 | 1827 | goto out; |
02c24a82 | 1828 | } |
a2135011 | 1829 | |
257c62e1 CM |
1830 | /* |
1831 | * if the last transaction that changed this file was before | |
1832 | * the current transaction, we can bail out now without any | |
1833 | * syncing | |
1834 | */ | |
a4abeea4 | 1835 | smp_mb(); |
22ee6985 JB |
1836 | if (btrfs_inode_in_log(inode, root->fs_info->generation) || |
1837 | BTRFS_I(inode)->last_trans <= | |
15ee9bc7 JB |
1838 | root->fs_info->last_trans_committed) { |
1839 | BTRFS_I(inode)->last_trans = 0; | |
5dc562c5 JB |
1840 | |
1841 | /* | |
1842 | * We'v had everything committed since the last time we were | |
1843 | * modified so clear this flag in case it was set for whatever | |
1844 | * reason, it's no longer relevant. | |
1845 | */ | |
1846 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
1847 | &BTRFS_I(inode)->runtime_flags); | |
02c24a82 | 1848 | mutex_unlock(&inode->i_mutex); |
15ee9bc7 JB |
1849 | goto out; |
1850 | } | |
15ee9bc7 JB |
1851 | |
1852 | /* | |
a52d9a80 CM |
1853 | * ok we haven't committed the transaction yet, lets do a commit |
1854 | */ | |
6f902af4 | 1855 | if (file->private_data) |
6bf13c0c SW |
1856 | btrfs_ioctl_trans_end(file); |
1857 | ||
a22285a6 YZ |
1858 | trans = btrfs_start_transaction(root, 0); |
1859 | if (IS_ERR(trans)) { | |
1860 | ret = PTR_ERR(trans); | |
02c24a82 | 1861 | mutex_unlock(&inode->i_mutex); |
39279cc3 CM |
1862 | goto out; |
1863 | } | |
e02119d5 | 1864 | |
2cfbd50b | 1865 | ret = btrfs_log_dentry_safe(trans, root, dentry); |
02c24a82 JB |
1866 | if (ret < 0) { |
1867 | mutex_unlock(&inode->i_mutex); | |
e02119d5 | 1868 | goto out; |
02c24a82 | 1869 | } |
49eb7e46 CM |
1870 | |
1871 | /* we've logged all the items and now have a consistent | |
1872 | * version of the file in the log. It is possible that | |
1873 | * someone will come in and modify the file, but that's | |
1874 | * fine because the log is consistent on disk, and we | |
1875 | * have references to all of the file's extents | |
1876 | * | |
1877 | * It is possible that someone will come in and log the | |
1878 | * file again, but that will end up using the synchronization | |
1879 | * inside btrfs_sync_log to keep things safe. | |
1880 | */ | |
02c24a82 | 1881 | mutex_unlock(&inode->i_mutex); |
49eb7e46 | 1882 | |
257c62e1 CM |
1883 | if (ret != BTRFS_NO_LOG_SYNC) { |
1884 | if (ret > 0) { | |
2ab28f32 JB |
1885 | /* |
1886 | * If we didn't already wait for ordered extents we need | |
1887 | * to do that now. | |
1888 | */ | |
1889 | if (!full_sync) | |
1890 | btrfs_wait_ordered_range(inode, start, | |
1891 | end - start + 1); | |
12fcfd22 | 1892 | ret = btrfs_commit_transaction(trans, root); |
257c62e1 CM |
1893 | } else { |
1894 | ret = btrfs_sync_log(trans, root); | |
2ab28f32 | 1895 | if (ret == 0) { |
257c62e1 | 1896 | ret = btrfs_end_transaction(trans, root); |
2ab28f32 JB |
1897 | } else { |
1898 | if (!full_sync) | |
1899 | btrfs_wait_ordered_range(inode, start, | |
1900 | end - | |
1901 | start + 1); | |
257c62e1 | 1902 | ret = btrfs_commit_transaction(trans, root); |
2ab28f32 | 1903 | } |
257c62e1 CM |
1904 | } |
1905 | } else { | |
1906 | ret = btrfs_end_transaction(trans, root); | |
e02119d5 | 1907 | } |
39279cc3 | 1908 | out: |
014e4ac4 | 1909 | return ret > 0 ? -EIO : ret; |
39279cc3 CM |
1910 | } |
1911 | ||
f0f37e2f | 1912 | static const struct vm_operations_struct btrfs_file_vm_ops = { |
92fee66d | 1913 | .fault = filemap_fault, |
9ebefb18 | 1914 | .page_mkwrite = btrfs_page_mkwrite, |
0b173bc4 | 1915 | .remap_pages = generic_file_remap_pages, |
9ebefb18 CM |
1916 | }; |
1917 | ||
1918 | static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma) | |
1919 | { | |
058a457e MX |
1920 | struct address_space *mapping = filp->f_mapping; |
1921 | ||
1922 | if (!mapping->a_ops->readpage) | |
1923 | return -ENOEXEC; | |
1924 | ||
9ebefb18 | 1925 | file_accessed(filp); |
058a457e | 1926 | vma->vm_ops = &btrfs_file_vm_ops; |
058a457e | 1927 | |
9ebefb18 CM |
1928 | return 0; |
1929 | } | |
1930 | ||
2aaa6655 JB |
1931 | static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf, |
1932 | int slot, u64 start, u64 end) | |
1933 | { | |
1934 | struct btrfs_file_extent_item *fi; | |
1935 | struct btrfs_key key; | |
1936 | ||
1937 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) | |
1938 | return 0; | |
1939 | ||
1940 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
1941 | if (key.objectid != btrfs_ino(inode) || | |
1942 | key.type != BTRFS_EXTENT_DATA_KEY) | |
1943 | return 0; | |
1944 | ||
1945 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
1946 | ||
1947 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
1948 | return 0; | |
1949 | ||
1950 | if (btrfs_file_extent_disk_bytenr(leaf, fi)) | |
1951 | return 0; | |
1952 | ||
1953 | if (key.offset == end) | |
1954 | return 1; | |
1955 | if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start) | |
1956 | return 1; | |
1957 | return 0; | |
1958 | } | |
1959 | ||
1960 | static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode, | |
1961 | struct btrfs_path *path, u64 offset, u64 end) | |
1962 | { | |
1963 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1964 | struct extent_buffer *leaf; | |
1965 | struct btrfs_file_extent_item *fi; | |
1966 | struct extent_map *hole_em; | |
1967 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
1968 | struct btrfs_key key; | |
1969 | int ret; | |
1970 | ||
1971 | key.objectid = btrfs_ino(inode); | |
1972 | key.type = BTRFS_EXTENT_DATA_KEY; | |
1973 | key.offset = offset; | |
1974 | ||
1975 | ||
1976 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
1977 | if (ret < 0) | |
1978 | return ret; | |
1979 | BUG_ON(!ret); | |
1980 | ||
1981 | leaf = path->nodes[0]; | |
1982 | if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) { | |
1983 | u64 num_bytes; | |
1984 | ||
1985 | path->slots[0]--; | |
1986 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1987 | struct btrfs_file_extent_item); | |
1988 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + | |
1989 | end - offset; | |
1990 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1991 | btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes); | |
1992 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1993 | btrfs_mark_buffer_dirty(leaf); | |
1994 | goto out; | |
1995 | } | |
1996 | ||
1997 | if (hole_mergeable(inode, leaf, path->slots[0]+1, offset, end)) { | |
1998 | u64 num_bytes; | |
1999 | ||
2000 | path->slots[0]++; | |
2001 | key.offset = offset; | |
afe5fea7 | 2002 | btrfs_set_item_key_safe(root, path, &key); |
2aaa6655 JB |
2003 | fi = btrfs_item_ptr(leaf, path->slots[0], |
2004 | struct btrfs_file_extent_item); | |
2005 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end - | |
2006 | offset; | |
2007 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2008 | btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes); | |
2009 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
2010 | btrfs_mark_buffer_dirty(leaf); | |
2011 | goto out; | |
2012 | } | |
2013 | btrfs_release_path(path); | |
2014 | ||
2015 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset, | |
2016 | 0, 0, end - offset, 0, end - offset, | |
2017 | 0, 0, 0); | |
2018 | if (ret) | |
2019 | return ret; | |
2020 | ||
2021 | out: | |
2022 | btrfs_release_path(path); | |
2023 | ||
2024 | hole_em = alloc_extent_map(); | |
2025 | if (!hole_em) { | |
2026 | btrfs_drop_extent_cache(inode, offset, end - 1, 0); | |
2027 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
2028 | &BTRFS_I(inode)->runtime_flags); | |
2029 | } else { | |
2030 | hole_em->start = offset; | |
2031 | hole_em->len = end - offset; | |
cc95bef6 | 2032 | hole_em->ram_bytes = hole_em->len; |
2aaa6655 JB |
2033 | hole_em->orig_start = offset; |
2034 | ||
2035 | hole_em->block_start = EXTENT_MAP_HOLE; | |
2036 | hole_em->block_len = 0; | |
b4939680 | 2037 | hole_em->orig_block_len = 0; |
2aaa6655 JB |
2038 | hole_em->bdev = root->fs_info->fs_devices->latest_bdev; |
2039 | hole_em->compress_type = BTRFS_COMPRESS_NONE; | |
2040 | hole_em->generation = trans->transid; | |
2041 | ||
2042 | do { | |
2043 | btrfs_drop_extent_cache(inode, offset, end - 1, 0); | |
2044 | write_lock(&em_tree->lock); | |
09a2a8f9 | 2045 | ret = add_extent_mapping(em_tree, hole_em, 1); |
2aaa6655 JB |
2046 | write_unlock(&em_tree->lock); |
2047 | } while (ret == -EEXIST); | |
2048 | free_extent_map(hole_em); | |
2049 | if (ret) | |
2050 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
2051 | &BTRFS_I(inode)->runtime_flags); | |
2052 | } | |
2053 | ||
2054 | return 0; | |
2055 | } | |
2056 | ||
2057 | static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len) | |
2058 | { | |
2059 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2060 | struct extent_state *cached_state = NULL; | |
2061 | struct btrfs_path *path; | |
2062 | struct btrfs_block_rsv *rsv; | |
2063 | struct btrfs_trans_handle *trans; | |
0061280d MX |
2064 | u64 lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize); |
2065 | u64 lockend = round_down(offset + len, | |
2066 | BTRFS_I(inode)->root->sectorsize) - 1; | |
2aaa6655 JB |
2067 | u64 cur_offset = lockstart; |
2068 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); | |
2069 | u64 drop_end; | |
2aaa6655 JB |
2070 | int ret = 0; |
2071 | int err = 0; | |
6347b3c4 MX |
2072 | bool same_page = ((offset >> PAGE_CACHE_SHIFT) == |
2073 | ((offset + len - 1) >> PAGE_CACHE_SHIFT)); | |
2aaa6655 JB |
2074 | |
2075 | btrfs_wait_ordered_range(inode, offset, len); | |
2076 | ||
2077 | mutex_lock(&inode->i_mutex); | |
7426cc04 MX |
2078 | /* |
2079 | * We needn't truncate any page which is beyond the end of the file | |
2080 | * because we are sure there is no data there. | |
2081 | */ | |
2aaa6655 JB |
2082 | /* |
2083 | * Only do this if we are in the same page and we aren't doing the | |
2084 | * entire page. | |
2085 | */ | |
2086 | if (same_page && len < PAGE_CACHE_SIZE) { | |
7426cc04 MX |
2087 | if (offset < round_up(inode->i_size, PAGE_CACHE_SIZE)) |
2088 | ret = btrfs_truncate_page(inode, offset, len, 0); | |
2aaa6655 JB |
2089 | mutex_unlock(&inode->i_mutex); |
2090 | return ret; | |
2091 | } | |
2092 | ||
2093 | /* zero back part of the first page */ | |
7426cc04 MX |
2094 | if (offset < round_up(inode->i_size, PAGE_CACHE_SIZE)) { |
2095 | ret = btrfs_truncate_page(inode, offset, 0, 0); | |
2096 | if (ret) { | |
2097 | mutex_unlock(&inode->i_mutex); | |
2098 | return ret; | |
2099 | } | |
2aaa6655 JB |
2100 | } |
2101 | ||
2102 | /* zero the front end of the last page */ | |
0061280d MX |
2103 | if (offset + len < round_up(inode->i_size, PAGE_CACHE_SIZE)) { |
2104 | ret = btrfs_truncate_page(inode, offset + len, 0, 1); | |
2105 | if (ret) { | |
2106 | mutex_unlock(&inode->i_mutex); | |
2107 | return ret; | |
2108 | } | |
2aaa6655 JB |
2109 | } |
2110 | ||
2111 | if (lockend < lockstart) { | |
2112 | mutex_unlock(&inode->i_mutex); | |
2113 | return 0; | |
2114 | } | |
2115 | ||
2116 | while (1) { | |
2117 | struct btrfs_ordered_extent *ordered; | |
2118 | ||
2119 | truncate_pagecache_range(inode, lockstart, lockend); | |
2120 | ||
2121 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
2122 | 0, &cached_state); | |
2123 | ordered = btrfs_lookup_first_ordered_extent(inode, lockend); | |
2124 | ||
2125 | /* | |
2126 | * We need to make sure we have no ordered extents in this range | |
2127 | * and nobody raced in and read a page in this range, if we did | |
2128 | * we need to try again. | |
2129 | */ | |
2130 | if ((!ordered || | |
2131 | (ordered->file_offset + ordered->len < lockstart || | |
2132 | ordered->file_offset > lockend)) && | |
2133 | !test_range_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
2134 | lockend, EXTENT_UPTODATE, 0, | |
2135 | cached_state)) { | |
2136 | if (ordered) | |
2137 | btrfs_put_ordered_extent(ordered); | |
2138 | break; | |
2139 | } | |
2140 | if (ordered) | |
2141 | btrfs_put_ordered_extent(ordered); | |
2142 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, | |
2143 | lockend, &cached_state, GFP_NOFS); | |
2144 | btrfs_wait_ordered_range(inode, lockstart, | |
2145 | lockend - lockstart + 1); | |
2146 | } | |
2147 | ||
2148 | path = btrfs_alloc_path(); | |
2149 | if (!path) { | |
2150 | ret = -ENOMEM; | |
2151 | goto out; | |
2152 | } | |
2153 | ||
66d8f3dd | 2154 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
2aaa6655 JB |
2155 | if (!rsv) { |
2156 | ret = -ENOMEM; | |
2157 | goto out_free; | |
2158 | } | |
2159 | rsv->size = btrfs_calc_trunc_metadata_size(root, 1); | |
2160 | rsv->failfast = 1; | |
2161 | ||
2162 | /* | |
2163 | * 1 - update the inode | |
2164 | * 1 - removing the extents in the range | |
2165 | * 1 - adding the hole extent | |
2166 | */ | |
2167 | trans = btrfs_start_transaction(root, 3); | |
2168 | if (IS_ERR(trans)) { | |
2169 | err = PTR_ERR(trans); | |
2170 | goto out_free; | |
2171 | } | |
2172 | ||
2173 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv, | |
2174 | min_size); | |
2175 | BUG_ON(ret); | |
2176 | trans->block_rsv = rsv; | |
2177 | ||
2178 | while (cur_offset < lockend) { | |
2179 | ret = __btrfs_drop_extents(trans, root, inode, path, | |
2180 | cur_offset, lockend + 1, | |
2181 | &drop_end, 1); | |
2182 | if (ret != -ENOSPC) | |
2183 | break; | |
2184 | ||
2185 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
2186 | ||
2187 | ret = fill_holes(trans, inode, path, cur_offset, drop_end); | |
2188 | if (ret) { | |
2189 | err = ret; | |
2190 | break; | |
2191 | } | |
2192 | ||
2193 | cur_offset = drop_end; | |
2194 | ||
2195 | ret = btrfs_update_inode(trans, root, inode); | |
2196 | if (ret) { | |
2197 | err = ret; | |
2198 | break; | |
2199 | } | |
2200 | ||
2aaa6655 | 2201 | btrfs_end_transaction(trans, root); |
b53d3f5d | 2202 | btrfs_btree_balance_dirty(root); |
2aaa6655 JB |
2203 | |
2204 | trans = btrfs_start_transaction(root, 3); | |
2205 | if (IS_ERR(trans)) { | |
2206 | ret = PTR_ERR(trans); | |
2207 | trans = NULL; | |
2208 | break; | |
2209 | } | |
2210 | ||
2211 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, | |
2212 | rsv, min_size); | |
2213 | BUG_ON(ret); /* shouldn't happen */ | |
2214 | trans->block_rsv = rsv; | |
2215 | } | |
2216 | ||
2217 | if (ret) { | |
2218 | err = ret; | |
2219 | goto out_trans; | |
2220 | } | |
2221 | ||
2222 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
2223 | ret = fill_holes(trans, inode, path, cur_offset, drop_end); | |
2224 | if (ret) { | |
2225 | err = ret; | |
2226 | goto out_trans; | |
2227 | } | |
2228 | ||
2229 | out_trans: | |
2230 | if (!trans) | |
2231 | goto out_free; | |
2232 | ||
e1f5790e TI |
2233 | inode_inc_iversion(inode); |
2234 | inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
2235 | ||
2aaa6655 JB |
2236 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
2237 | ret = btrfs_update_inode(trans, root, inode); | |
2aaa6655 | 2238 | btrfs_end_transaction(trans, root); |
b53d3f5d | 2239 | btrfs_btree_balance_dirty(root); |
2aaa6655 JB |
2240 | out_free: |
2241 | btrfs_free_path(path); | |
2242 | btrfs_free_block_rsv(root, rsv); | |
2243 | out: | |
2244 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
2245 | &cached_state, GFP_NOFS); | |
2246 | mutex_unlock(&inode->i_mutex); | |
2247 | if (ret && !err) | |
2248 | err = ret; | |
2249 | return err; | |
2250 | } | |
2251 | ||
2fe17c10 CH |
2252 | static long btrfs_fallocate(struct file *file, int mode, |
2253 | loff_t offset, loff_t len) | |
2254 | { | |
496ad9aa | 2255 | struct inode *inode = file_inode(file); |
2fe17c10 | 2256 | struct extent_state *cached_state = NULL; |
6113077c | 2257 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2fe17c10 CH |
2258 | u64 cur_offset; |
2259 | u64 last_byte; | |
2260 | u64 alloc_start; | |
2261 | u64 alloc_end; | |
2262 | u64 alloc_hint = 0; | |
2263 | u64 locked_end; | |
2fe17c10 | 2264 | struct extent_map *em; |
797f4277 | 2265 | int blocksize = BTRFS_I(inode)->root->sectorsize; |
2fe17c10 CH |
2266 | int ret; |
2267 | ||
797f4277 MX |
2268 | alloc_start = round_down(offset, blocksize); |
2269 | alloc_end = round_up(offset + len, blocksize); | |
2fe17c10 | 2270 | |
2aaa6655 JB |
2271 | /* Make sure we aren't being give some crap mode */ |
2272 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
2fe17c10 CH |
2273 | return -EOPNOTSUPP; |
2274 | ||
2aaa6655 JB |
2275 | if (mode & FALLOC_FL_PUNCH_HOLE) |
2276 | return btrfs_punch_hole(inode, offset, len); | |
2277 | ||
d98456fc CM |
2278 | /* |
2279 | * Make sure we have enough space before we do the | |
2280 | * allocation. | |
2281 | */ | |
0ff6fabd | 2282 | ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start); |
d98456fc CM |
2283 | if (ret) |
2284 | return ret; | |
6113077c WS |
2285 | if (root->fs_info->quota_enabled) { |
2286 | ret = btrfs_qgroup_reserve(root, alloc_end - alloc_start); | |
2287 | if (ret) | |
2288 | goto out_reserve_fail; | |
2289 | } | |
d98456fc | 2290 | |
2fe17c10 CH |
2291 | mutex_lock(&inode->i_mutex); |
2292 | ret = inode_newsize_ok(inode, alloc_end); | |
2293 | if (ret) | |
2294 | goto out; | |
2295 | ||
2296 | if (alloc_start > inode->i_size) { | |
a41ad394 JB |
2297 | ret = btrfs_cont_expand(inode, i_size_read(inode), |
2298 | alloc_start); | |
2fe17c10 CH |
2299 | if (ret) |
2300 | goto out; | |
a71754fc JB |
2301 | } else { |
2302 | /* | |
2303 | * If we are fallocating from the end of the file onward we | |
2304 | * need to zero out the end of the page if i_size lands in the | |
2305 | * middle of a page. | |
2306 | */ | |
2307 | ret = btrfs_truncate_page(inode, inode->i_size, 0, 0); | |
2308 | if (ret) | |
2309 | goto out; | |
2fe17c10 CH |
2310 | } |
2311 | ||
a71754fc JB |
2312 | /* |
2313 | * wait for ordered IO before we have any locks. We'll loop again | |
2314 | * below with the locks held. | |
2315 | */ | |
2316 | btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start); | |
2317 | ||
2fe17c10 CH |
2318 | locked_end = alloc_end - 1; |
2319 | while (1) { | |
2320 | struct btrfs_ordered_extent *ordered; | |
2321 | ||
2322 | /* the extent lock is ordered inside the running | |
2323 | * transaction | |
2324 | */ | |
2325 | lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start, | |
d0082371 | 2326 | locked_end, 0, &cached_state); |
2fe17c10 CH |
2327 | ordered = btrfs_lookup_first_ordered_extent(inode, |
2328 | alloc_end - 1); | |
2329 | if (ordered && | |
2330 | ordered->file_offset + ordered->len > alloc_start && | |
2331 | ordered->file_offset < alloc_end) { | |
2332 | btrfs_put_ordered_extent(ordered); | |
2333 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, | |
2334 | alloc_start, locked_end, | |
2335 | &cached_state, GFP_NOFS); | |
2336 | /* | |
2337 | * we can't wait on the range with the transaction | |
2338 | * running or with the extent lock held | |
2339 | */ | |
2340 | btrfs_wait_ordered_range(inode, alloc_start, | |
2341 | alloc_end - alloc_start); | |
2342 | } else { | |
2343 | if (ordered) | |
2344 | btrfs_put_ordered_extent(ordered); | |
2345 | break; | |
2346 | } | |
2347 | } | |
2348 | ||
2349 | cur_offset = alloc_start; | |
2350 | while (1) { | |
f1e490a7 JB |
2351 | u64 actual_end; |
2352 | ||
2fe17c10 CH |
2353 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, |
2354 | alloc_end - cur_offset, 0); | |
79787eaa JM |
2355 | if (IS_ERR_OR_NULL(em)) { |
2356 | if (!em) | |
2357 | ret = -ENOMEM; | |
2358 | else | |
2359 | ret = PTR_ERR(em); | |
2360 | break; | |
2361 | } | |
2fe17c10 | 2362 | last_byte = min(extent_map_end(em), alloc_end); |
f1e490a7 | 2363 | actual_end = min_t(u64, extent_map_end(em), offset + len); |
797f4277 | 2364 | last_byte = ALIGN(last_byte, blocksize); |
f1e490a7 | 2365 | |
2fe17c10 CH |
2366 | if (em->block_start == EXTENT_MAP_HOLE || |
2367 | (cur_offset >= inode->i_size && | |
2368 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
2369 | ret = btrfs_prealloc_file_range(inode, mode, cur_offset, | |
2370 | last_byte - cur_offset, | |
2371 | 1 << inode->i_blkbits, | |
2372 | offset + len, | |
2373 | &alloc_hint); | |
1b9c332b | 2374 | |
2fe17c10 CH |
2375 | if (ret < 0) { |
2376 | free_extent_map(em); | |
2377 | break; | |
2378 | } | |
f1e490a7 JB |
2379 | } else if (actual_end > inode->i_size && |
2380 | !(mode & FALLOC_FL_KEEP_SIZE)) { | |
2381 | /* | |
2382 | * We didn't need to allocate any more space, but we | |
2383 | * still extended the size of the file so we need to | |
2384 | * update i_size. | |
2385 | */ | |
2386 | inode->i_ctime = CURRENT_TIME; | |
2387 | i_size_write(inode, actual_end); | |
2388 | btrfs_ordered_update_i_size(inode, actual_end, NULL); | |
2fe17c10 CH |
2389 | } |
2390 | free_extent_map(em); | |
2391 | ||
2392 | cur_offset = last_byte; | |
2393 | if (cur_offset >= alloc_end) { | |
2394 | ret = 0; | |
2395 | break; | |
2396 | } | |
2397 | } | |
2398 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, | |
2399 | &cached_state, GFP_NOFS); | |
2fe17c10 CH |
2400 | out: |
2401 | mutex_unlock(&inode->i_mutex); | |
6113077c WS |
2402 | if (root->fs_info->quota_enabled) |
2403 | btrfs_qgroup_free(root, alloc_end - alloc_start); | |
2404 | out_reserve_fail: | |
d98456fc | 2405 | /* Let go of our reservation. */ |
0ff6fabd | 2406 | btrfs_free_reserved_data_space(inode, alloc_end - alloc_start); |
2fe17c10 CH |
2407 | return ret; |
2408 | } | |
2409 | ||
965c8e59 | 2410 | static int find_desired_extent(struct inode *inode, loff_t *offset, int whence) |
b2675157 JB |
2411 | { |
2412 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2413 | struct extent_map *em; | |
2414 | struct extent_state *cached_state = NULL; | |
2415 | u64 lockstart = *offset; | |
2416 | u64 lockend = i_size_read(inode); | |
2417 | u64 start = *offset; | |
2418 | u64 orig_start = *offset; | |
2419 | u64 len = i_size_read(inode); | |
2420 | u64 last_end = 0; | |
2421 | int ret = 0; | |
2422 | ||
2423 | lockend = max_t(u64, root->sectorsize, lockend); | |
2424 | if (lockend <= lockstart) | |
2425 | lockend = lockstart + root->sectorsize; | |
2426 | ||
1214b53f | 2427 | lockend--; |
b2675157 JB |
2428 | len = lockend - lockstart + 1; |
2429 | ||
2430 | len = max_t(u64, len, root->sectorsize); | |
2431 | if (inode->i_size == 0) | |
2432 | return -ENXIO; | |
2433 | ||
2434 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0, | |
d0082371 | 2435 | &cached_state); |
b2675157 JB |
2436 | |
2437 | /* | |
2438 | * Delalloc is such a pain. If we have a hole and we have pending | |
2439 | * delalloc for a portion of the hole we will get back a hole that | |
2440 | * exists for the entire range since it hasn't been actually written | |
2441 | * yet. So to take care of this case we need to look for an extent just | |
2442 | * before the position we want in case there is outstanding delalloc | |
2443 | * going on here. | |
2444 | */ | |
965c8e59 | 2445 | if (whence == SEEK_HOLE && start != 0) { |
b2675157 JB |
2446 | if (start <= root->sectorsize) |
2447 | em = btrfs_get_extent_fiemap(inode, NULL, 0, 0, | |
2448 | root->sectorsize, 0); | |
2449 | else | |
2450 | em = btrfs_get_extent_fiemap(inode, NULL, 0, | |
2451 | start - root->sectorsize, | |
2452 | root->sectorsize, 0); | |
2453 | if (IS_ERR(em)) { | |
6af021d8 | 2454 | ret = PTR_ERR(em); |
b2675157 JB |
2455 | goto out; |
2456 | } | |
2457 | last_end = em->start + em->len; | |
2458 | if (em->block_start == EXTENT_MAP_DELALLOC) | |
2459 | last_end = min_t(u64, last_end, inode->i_size); | |
2460 | free_extent_map(em); | |
2461 | } | |
2462 | ||
2463 | while (1) { | |
2464 | em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0); | |
2465 | if (IS_ERR(em)) { | |
6af021d8 | 2466 | ret = PTR_ERR(em); |
b2675157 JB |
2467 | break; |
2468 | } | |
2469 | ||
2470 | if (em->block_start == EXTENT_MAP_HOLE) { | |
2471 | if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) { | |
2472 | if (last_end <= orig_start) { | |
2473 | free_extent_map(em); | |
2474 | ret = -ENXIO; | |
2475 | break; | |
2476 | } | |
2477 | } | |
2478 | ||
965c8e59 | 2479 | if (whence == SEEK_HOLE) { |
b2675157 JB |
2480 | *offset = start; |
2481 | free_extent_map(em); | |
2482 | break; | |
2483 | } | |
2484 | } else { | |
965c8e59 | 2485 | if (whence == SEEK_DATA) { |
b2675157 JB |
2486 | if (em->block_start == EXTENT_MAP_DELALLOC) { |
2487 | if (start >= inode->i_size) { | |
2488 | free_extent_map(em); | |
2489 | ret = -ENXIO; | |
2490 | break; | |
2491 | } | |
2492 | } | |
2493 | ||
f9e4fb53 LB |
2494 | if (!test_bit(EXTENT_FLAG_PREALLOC, |
2495 | &em->flags)) { | |
2496 | *offset = start; | |
2497 | free_extent_map(em); | |
2498 | break; | |
2499 | } | |
b2675157 JB |
2500 | } |
2501 | } | |
2502 | ||
2503 | start = em->start + em->len; | |
2504 | last_end = em->start + em->len; | |
2505 | ||
2506 | if (em->block_start == EXTENT_MAP_DELALLOC) | |
2507 | last_end = min_t(u64, last_end, inode->i_size); | |
2508 | ||
2509 | if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) { | |
2510 | free_extent_map(em); | |
2511 | ret = -ENXIO; | |
2512 | break; | |
2513 | } | |
2514 | free_extent_map(em); | |
2515 | cond_resched(); | |
2516 | } | |
2517 | if (!ret) | |
2518 | *offset = min(*offset, inode->i_size); | |
2519 | out: | |
2520 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
2521 | &cached_state, GFP_NOFS); | |
2522 | return ret; | |
2523 | } | |
2524 | ||
965c8e59 | 2525 | static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence) |
b2675157 JB |
2526 | { |
2527 | struct inode *inode = file->f_mapping->host; | |
2528 | int ret; | |
2529 | ||
2530 | mutex_lock(&inode->i_mutex); | |
965c8e59 | 2531 | switch (whence) { |
b2675157 JB |
2532 | case SEEK_END: |
2533 | case SEEK_CUR: | |
965c8e59 | 2534 | offset = generic_file_llseek(file, offset, whence); |
b2675157 JB |
2535 | goto out; |
2536 | case SEEK_DATA: | |
2537 | case SEEK_HOLE: | |
48802c8a JL |
2538 | if (offset >= i_size_read(inode)) { |
2539 | mutex_unlock(&inode->i_mutex); | |
2540 | return -ENXIO; | |
2541 | } | |
2542 | ||
965c8e59 | 2543 | ret = find_desired_extent(inode, &offset, whence); |
b2675157 JB |
2544 | if (ret) { |
2545 | mutex_unlock(&inode->i_mutex); | |
2546 | return ret; | |
2547 | } | |
2548 | } | |
2549 | ||
46a1c2c7 | 2550 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
b2675157 JB |
2551 | out: |
2552 | mutex_unlock(&inode->i_mutex); | |
2553 | return offset; | |
2554 | } | |
2555 | ||
828c0950 | 2556 | const struct file_operations btrfs_file_operations = { |
b2675157 | 2557 | .llseek = btrfs_file_llseek, |
39279cc3 | 2558 | .read = do_sync_read, |
4a001071 | 2559 | .write = do_sync_write, |
9ebefb18 | 2560 | .aio_read = generic_file_aio_read, |
e9906a98 | 2561 | .splice_read = generic_file_splice_read, |
11c65dcc | 2562 | .aio_write = btrfs_file_aio_write, |
9ebefb18 | 2563 | .mmap = btrfs_file_mmap, |
39279cc3 | 2564 | .open = generic_file_open, |
e1b81e67 | 2565 | .release = btrfs_release_file, |
39279cc3 | 2566 | .fsync = btrfs_sync_file, |
2fe17c10 | 2567 | .fallocate = btrfs_fallocate, |
34287aa3 | 2568 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 2569 | #ifdef CONFIG_COMPAT |
34287aa3 | 2570 | .compat_ioctl = btrfs_ioctl, |
39279cc3 CM |
2571 | #endif |
2572 | }; | |
9247f317 MX |
2573 | |
2574 | void btrfs_auto_defrag_exit(void) | |
2575 | { | |
2576 | if (btrfs_inode_defrag_cachep) | |
2577 | kmem_cache_destroy(btrfs_inode_defrag_cachep); | |
2578 | } | |
2579 | ||
2580 | int btrfs_auto_defrag_init(void) | |
2581 | { | |
2582 | btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag", | |
2583 | sizeof(struct inode_defrag), 0, | |
2584 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, | |
2585 | NULL); | |
2586 | if (!btrfs_inode_defrag_cachep) | |
2587 | return -ENOMEM; | |
2588 | ||
2589 | return 0; | |
2590 | } |