Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
39279cc3 CM |
19 | #include <linux/fs.h> |
20 | #include <linux/pagemap.h> | |
21 | #include <linux/highmem.h> | |
22 | #include <linux/time.h> | |
23 | #include <linux/init.h> | |
24 | #include <linux/string.h> | |
39279cc3 CM |
25 | #include <linux/backing-dev.h> |
26 | #include <linux/mpage.h> | |
2fe17c10 | 27 | #include <linux/falloc.h> |
39279cc3 CM |
28 | #include <linux/swap.h> |
29 | #include <linux/writeback.h> | |
30 | #include <linux/statfs.h> | |
31 | #include <linux/compat.h> | |
5a0e3ad6 | 32 | #include <linux/slab.h> |
55e301fd | 33 | #include <linux/btrfs.h> |
e2e40f2c | 34 | #include <linux/uio.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" | |
2aaa6655 | 42 | #include "volumes.h" |
fcebe456 | 43 | #include "qgroup.h" |
ebb8765b | 44 | #include "compression.h" |
39279cc3 | 45 | |
9247f317 | 46 | static struct kmem_cache *btrfs_inode_defrag_cachep; |
4cb5300b CM |
47 | /* |
48 | * when auto defrag is enabled we | |
49 | * queue up these defrag structs to remember which | |
50 | * inodes need defragging passes | |
51 | */ | |
52 | struct inode_defrag { | |
53 | struct rb_node rb_node; | |
54 | /* objectid */ | |
55 | u64 ino; | |
56 | /* | |
57 | * transid where the defrag was added, we search for | |
58 | * extents newer than this | |
59 | */ | |
60 | u64 transid; | |
61 | ||
62 | /* root objectid */ | |
63 | u64 root; | |
64 | ||
65 | /* last offset we were able to defrag */ | |
66 | u64 last_offset; | |
67 | ||
68 | /* if we've wrapped around back to zero once already */ | |
69 | int cycled; | |
70 | }; | |
71 | ||
762f2263 MX |
72 | static int __compare_inode_defrag(struct inode_defrag *defrag1, |
73 | struct inode_defrag *defrag2) | |
74 | { | |
75 | if (defrag1->root > defrag2->root) | |
76 | return 1; | |
77 | else if (defrag1->root < defrag2->root) | |
78 | return -1; | |
79 | else if (defrag1->ino > defrag2->ino) | |
80 | return 1; | |
81 | else if (defrag1->ino < defrag2->ino) | |
82 | return -1; | |
83 | else | |
84 | return 0; | |
85 | } | |
86 | ||
4cb5300b CM |
87 | /* pop a record for an inode into the defrag tree. The lock |
88 | * must be held already | |
89 | * | |
90 | * If you're inserting a record for an older transid than an | |
91 | * existing record, the transid already in the tree is lowered | |
92 | * | |
93 | * If an existing record is found the defrag item you | |
94 | * pass in is freed | |
95 | */ | |
8ddc4734 | 96 | static int __btrfs_add_inode_defrag(struct inode *inode, |
4cb5300b CM |
97 | struct inode_defrag *defrag) |
98 | { | |
99 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
100 | struct inode_defrag *entry; | |
101 | struct rb_node **p; | |
102 | struct rb_node *parent = NULL; | |
762f2263 | 103 | int ret; |
4cb5300b CM |
104 | |
105 | p = &root->fs_info->defrag_inodes.rb_node; | |
106 | while (*p) { | |
107 | parent = *p; | |
108 | entry = rb_entry(parent, struct inode_defrag, rb_node); | |
109 | ||
762f2263 MX |
110 | ret = __compare_inode_defrag(defrag, entry); |
111 | if (ret < 0) | |
4cb5300b | 112 | p = &parent->rb_left; |
762f2263 | 113 | else if (ret > 0) |
4cb5300b CM |
114 | p = &parent->rb_right; |
115 | else { | |
116 | /* if we're reinserting an entry for | |
117 | * an old defrag run, make sure to | |
118 | * lower the transid of our existing record | |
119 | */ | |
120 | if (defrag->transid < entry->transid) | |
121 | entry->transid = defrag->transid; | |
122 | if (defrag->last_offset > entry->last_offset) | |
123 | entry->last_offset = defrag->last_offset; | |
8ddc4734 | 124 | return -EEXIST; |
4cb5300b CM |
125 | } |
126 | } | |
72ac3c0d | 127 | set_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags); |
4cb5300b CM |
128 | rb_link_node(&defrag->rb_node, parent, p); |
129 | rb_insert_color(&defrag->rb_node, &root->fs_info->defrag_inodes); | |
8ddc4734 MX |
130 | return 0; |
131 | } | |
4cb5300b | 132 | |
8ddc4734 MX |
133 | static inline int __need_auto_defrag(struct btrfs_root *root) |
134 | { | |
135 | if (!btrfs_test_opt(root, AUTO_DEFRAG)) | |
136 | return 0; | |
137 | ||
138 | if (btrfs_fs_closing(root->fs_info)) | |
139 | return 0; | |
4cb5300b | 140 | |
8ddc4734 | 141 | return 1; |
4cb5300b CM |
142 | } |
143 | ||
144 | /* | |
145 | * insert a defrag record for this inode if auto defrag is | |
146 | * enabled | |
147 | */ | |
148 | int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans, | |
149 | struct inode *inode) | |
150 | { | |
151 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
152 | struct inode_defrag *defrag; | |
4cb5300b | 153 | u64 transid; |
8ddc4734 | 154 | int ret; |
4cb5300b | 155 | |
8ddc4734 | 156 | if (!__need_auto_defrag(root)) |
4cb5300b CM |
157 | return 0; |
158 | ||
72ac3c0d | 159 | if (test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags)) |
4cb5300b CM |
160 | return 0; |
161 | ||
162 | if (trans) | |
163 | transid = trans->transid; | |
164 | else | |
165 | transid = BTRFS_I(inode)->root->last_trans; | |
166 | ||
9247f317 | 167 | defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS); |
4cb5300b CM |
168 | if (!defrag) |
169 | return -ENOMEM; | |
170 | ||
a4689d2b | 171 | defrag->ino = btrfs_ino(inode); |
4cb5300b CM |
172 | defrag->transid = transid; |
173 | defrag->root = root->root_key.objectid; | |
174 | ||
175 | spin_lock(&root->fs_info->defrag_inodes_lock); | |
8ddc4734 MX |
176 | if (!test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags)) { |
177 | /* | |
178 | * If we set IN_DEFRAG flag and evict the inode from memory, | |
179 | * and then re-read this inode, this new inode doesn't have | |
180 | * IN_DEFRAG flag. At the case, we may find the existed defrag. | |
181 | */ | |
182 | ret = __btrfs_add_inode_defrag(inode, defrag); | |
183 | if (ret) | |
184 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
185 | } else { | |
9247f317 | 186 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); |
8ddc4734 | 187 | } |
4cb5300b | 188 | spin_unlock(&root->fs_info->defrag_inodes_lock); |
a0f98dde | 189 | return 0; |
4cb5300b CM |
190 | } |
191 | ||
192 | /* | |
8ddc4734 MX |
193 | * Requeue the defrag object. If there is a defrag object that points to |
194 | * the same inode in the tree, we will merge them together (by | |
195 | * __btrfs_add_inode_defrag()) and free the one that we want to requeue. | |
4cb5300b | 196 | */ |
48a3b636 ES |
197 | static void btrfs_requeue_inode_defrag(struct inode *inode, |
198 | struct inode_defrag *defrag) | |
8ddc4734 MX |
199 | { |
200 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
201 | int ret; | |
202 | ||
203 | if (!__need_auto_defrag(root)) | |
204 | goto out; | |
205 | ||
206 | /* | |
207 | * Here we don't check the IN_DEFRAG flag, because we need merge | |
208 | * them together. | |
209 | */ | |
210 | spin_lock(&root->fs_info->defrag_inodes_lock); | |
211 | ret = __btrfs_add_inode_defrag(inode, defrag); | |
212 | spin_unlock(&root->fs_info->defrag_inodes_lock); | |
213 | if (ret) | |
214 | goto out; | |
215 | return; | |
216 | out: | |
217 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
218 | } | |
219 | ||
4cb5300b | 220 | /* |
26176e7c MX |
221 | * pick the defragable inode that we want, if it doesn't exist, we will get |
222 | * the next one. | |
4cb5300b | 223 | */ |
26176e7c MX |
224 | static struct inode_defrag * |
225 | btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino) | |
4cb5300b CM |
226 | { |
227 | struct inode_defrag *entry = NULL; | |
762f2263 | 228 | struct inode_defrag tmp; |
4cb5300b CM |
229 | struct rb_node *p; |
230 | struct rb_node *parent = NULL; | |
762f2263 MX |
231 | int ret; |
232 | ||
233 | tmp.ino = ino; | |
234 | tmp.root = root; | |
4cb5300b | 235 | |
26176e7c MX |
236 | spin_lock(&fs_info->defrag_inodes_lock); |
237 | p = fs_info->defrag_inodes.rb_node; | |
4cb5300b CM |
238 | while (p) { |
239 | parent = p; | |
240 | entry = rb_entry(parent, struct inode_defrag, rb_node); | |
241 | ||
762f2263 MX |
242 | ret = __compare_inode_defrag(&tmp, entry); |
243 | if (ret < 0) | |
4cb5300b | 244 | p = parent->rb_left; |
762f2263 | 245 | else if (ret > 0) |
4cb5300b CM |
246 | p = parent->rb_right; |
247 | else | |
26176e7c | 248 | goto out; |
4cb5300b CM |
249 | } |
250 | ||
26176e7c MX |
251 | if (parent && __compare_inode_defrag(&tmp, entry) > 0) { |
252 | parent = rb_next(parent); | |
253 | if (parent) | |
4cb5300b | 254 | entry = rb_entry(parent, struct inode_defrag, rb_node); |
26176e7c MX |
255 | else |
256 | entry = NULL; | |
4cb5300b | 257 | } |
26176e7c MX |
258 | out: |
259 | if (entry) | |
260 | rb_erase(parent, &fs_info->defrag_inodes); | |
261 | spin_unlock(&fs_info->defrag_inodes_lock); | |
262 | return entry; | |
4cb5300b CM |
263 | } |
264 | ||
26176e7c | 265 | void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info) |
4cb5300b CM |
266 | { |
267 | struct inode_defrag *defrag; | |
26176e7c MX |
268 | struct rb_node *node; |
269 | ||
270 | spin_lock(&fs_info->defrag_inodes_lock); | |
271 | node = rb_first(&fs_info->defrag_inodes); | |
272 | while (node) { | |
273 | rb_erase(node, &fs_info->defrag_inodes); | |
274 | defrag = rb_entry(node, struct inode_defrag, rb_node); | |
275 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
276 | ||
351810c1 | 277 | cond_resched_lock(&fs_info->defrag_inodes_lock); |
26176e7c MX |
278 | |
279 | node = rb_first(&fs_info->defrag_inodes); | |
280 | } | |
281 | spin_unlock(&fs_info->defrag_inodes_lock); | |
282 | } | |
283 | ||
284 | #define BTRFS_DEFRAG_BATCH 1024 | |
285 | ||
286 | static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info, | |
287 | struct inode_defrag *defrag) | |
288 | { | |
4cb5300b CM |
289 | struct btrfs_root *inode_root; |
290 | struct inode *inode; | |
4cb5300b CM |
291 | struct btrfs_key key; |
292 | struct btrfs_ioctl_defrag_range_args range; | |
4cb5300b | 293 | int num_defrag; |
6f1c3605 LB |
294 | int index; |
295 | int ret; | |
4cb5300b | 296 | |
26176e7c MX |
297 | /* get the inode */ |
298 | key.objectid = defrag->root; | |
962a298f | 299 | key.type = BTRFS_ROOT_ITEM_KEY; |
26176e7c | 300 | key.offset = (u64)-1; |
6f1c3605 LB |
301 | |
302 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
303 | ||
26176e7c MX |
304 | inode_root = btrfs_read_fs_root_no_name(fs_info, &key); |
305 | if (IS_ERR(inode_root)) { | |
6f1c3605 LB |
306 | ret = PTR_ERR(inode_root); |
307 | goto cleanup; | |
308 | } | |
26176e7c MX |
309 | |
310 | key.objectid = defrag->ino; | |
962a298f | 311 | key.type = BTRFS_INODE_ITEM_KEY; |
26176e7c MX |
312 | key.offset = 0; |
313 | inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL); | |
314 | if (IS_ERR(inode)) { | |
6f1c3605 LB |
315 | ret = PTR_ERR(inode); |
316 | goto cleanup; | |
26176e7c | 317 | } |
6f1c3605 | 318 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
26176e7c MX |
319 | |
320 | /* do a chunk of defrag */ | |
321 | clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags); | |
4cb5300b CM |
322 | memset(&range, 0, sizeof(range)); |
323 | range.len = (u64)-1; | |
26176e7c | 324 | range.start = defrag->last_offset; |
b66f00da MX |
325 | |
326 | sb_start_write(fs_info->sb); | |
26176e7c MX |
327 | num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid, |
328 | BTRFS_DEFRAG_BATCH); | |
b66f00da | 329 | sb_end_write(fs_info->sb); |
26176e7c MX |
330 | /* |
331 | * if we filled the whole defrag batch, there | |
332 | * must be more work to do. Queue this defrag | |
333 | * again | |
334 | */ | |
335 | if (num_defrag == BTRFS_DEFRAG_BATCH) { | |
336 | defrag->last_offset = range.start; | |
337 | btrfs_requeue_inode_defrag(inode, defrag); | |
338 | } else if (defrag->last_offset && !defrag->cycled) { | |
339 | /* | |
340 | * we didn't fill our defrag batch, but | |
341 | * we didn't start at zero. Make sure we loop | |
342 | * around to the start of the file. | |
343 | */ | |
344 | defrag->last_offset = 0; | |
345 | defrag->cycled = 1; | |
346 | btrfs_requeue_inode_defrag(inode, defrag); | |
347 | } else { | |
348 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
349 | } | |
350 | ||
351 | iput(inode); | |
352 | return 0; | |
6f1c3605 LB |
353 | cleanup: |
354 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
355 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
356 | return ret; | |
26176e7c MX |
357 | } |
358 | ||
359 | /* | |
360 | * run through the list of inodes in the FS that need | |
361 | * defragging | |
362 | */ | |
363 | int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info) | |
364 | { | |
365 | struct inode_defrag *defrag; | |
366 | u64 first_ino = 0; | |
367 | u64 root_objectid = 0; | |
4cb5300b CM |
368 | |
369 | atomic_inc(&fs_info->defrag_running); | |
67871254 | 370 | while (1) { |
dc81cdc5 MX |
371 | /* Pause the auto defragger. */ |
372 | if (test_bit(BTRFS_FS_STATE_REMOUNTING, | |
373 | &fs_info->fs_state)) | |
374 | break; | |
375 | ||
26176e7c MX |
376 | if (!__need_auto_defrag(fs_info->tree_root)) |
377 | break; | |
4cb5300b CM |
378 | |
379 | /* find an inode to defrag */ | |
26176e7c MX |
380 | defrag = btrfs_pick_defrag_inode(fs_info, root_objectid, |
381 | first_ino); | |
4cb5300b | 382 | if (!defrag) { |
26176e7c | 383 | if (root_objectid || first_ino) { |
762f2263 | 384 | root_objectid = 0; |
4cb5300b CM |
385 | first_ino = 0; |
386 | continue; | |
387 | } else { | |
388 | break; | |
389 | } | |
390 | } | |
391 | ||
4cb5300b | 392 | first_ino = defrag->ino + 1; |
762f2263 | 393 | root_objectid = defrag->root; |
4cb5300b | 394 | |
26176e7c | 395 | __btrfs_run_defrag_inode(fs_info, defrag); |
4cb5300b | 396 | } |
4cb5300b CM |
397 | atomic_dec(&fs_info->defrag_running); |
398 | ||
399 | /* | |
400 | * during unmount, we use the transaction_wait queue to | |
401 | * wait for the defragger to stop | |
402 | */ | |
403 | wake_up(&fs_info->transaction_wait); | |
404 | return 0; | |
405 | } | |
39279cc3 | 406 | |
d352ac68 CM |
407 | /* simple helper to fault in pages and copy. This should go away |
408 | * and be replaced with calls into generic code. | |
409 | */ | |
ee22f0c4 | 410 | static noinline int btrfs_copy_from_user(loff_t pos, size_t write_bytes, |
a1b32a59 | 411 | struct page **prepared_pages, |
11c65dcc | 412 | struct iov_iter *i) |
39279cc3 | 413 | { |
914ee295 | 414 | size_t copied = 0; |
d0215f3e | 415 | size_t total_copied = 0; |
11c65dcc | 416 | int pg = 0; |
39279cc3 CM |
417 | int offset = pos & (PAGE_CACHE_SIZE - 1); |
418 | ||
11c65dcc | 419 | while (write_bytes > 0) { |
39279cc3 CM |
420 | size_t count = min_t(size_t, |
421 | PAGE_CACHE_SIZE - offset, write_bytes); | |
11c65dcc | 422 | struct page *page = prepared_pages[pg]; |
914ee295 XZ |
423 | /* |
424 | * Copy data from userspace to the current page | |
914ee295 | 425 | */ |
914ee295 | 426 | copied = iov_iter_copy_from_user_atomic(page, i, offset, count); |
11c65dcc | 427 | |
39279cc3 CM |
428 | /* Flush processor's dcache for this page */ |
429 | flush_dcache_page(page); | |
31339acd CM |
430 | |
431 | /* | |
432 | * if we get a partial write, we can end up with | |
433 | * partially up to date pages. These add | |
434 | * a lot of complexity, so make sure they don't | |
435 | * happen by forcing this copy to be retried. | |
436 | * | |
437 | * The rest of the btrfs_file_write code will fall | |
438 | * back to page at a time copies after we return 0. | |
439 | */ | |
440 | if (!PageUptodate(page) && copied < count) | |
441 | copied = 0; | |
442 | ||
11c65dcc JB |
443 | iov_iter_advance(i, copied); |
444 | write_bytes -= copied; | |
914ee295 | 445 | total_copied += copied; |
39279cc3 | 446 | |
b30ac0fc | 447 | /* Return to btrfs_file_write_iter to fault page */ |
9f570b8d | 448 | if (unlikely(copied == 0)) |
914ee295 | 449 | break; |
11c65dcc | 450 | |
ee39b432 | 451 | if (copied < PAGE_CACHE_SIZE - offset) { |
11c65dcc JB |
452 | offset += copied; |
453 | } else { | |
454 | pg++; | |
455 | offset = 0; | |
456 | } | |
39279cc3 | 457 | } |
914ee295 | 458 | return total_copied; |
39279cc3 CM |
459 | } |
460 | ||
d352ac68 CM |
461 | /* |
462 | * unlocks pages after btrfs_file_write is done with them | |
463 | */ | |
48a3b636 | 464 | static void btrfs_drop_pages(struct page **pages, size_t num_pages) |
39279cc3 CM |
465 | { |
466 | size_t i; | |
467 | for (i = 0; i < num_pages; i++) { | |
d352ac68 CM |
468 | /* page checked is some magic around finding pages that |
469 | * have been modified without going through btrfs_set_page_dirty | |
2457aec6 MG |
470 | * clear it here. There should be no need to mark the pages |
471 | * accessed as prepare_pages should have marked them accessed | |
472 | * in prepare_pages via find_or_create_page() | |
d352ac68 | 473 | */ |
4a096752 | 474 | ClearPageChecked(pages[i]); |
39279cc3 | 475 | unlock_page(pages[i]); |
39279cc3 CM |
476 | page_cache_release(pages[i]); |
477 | } | |
478 | } | |
479 | ||
d352ac68 CM |
480 | /* |
481 | * after copy_from_user, pages need to be dirtied and we need to make | |
482 | * sure holes are created between the current EOF and the start of | |
483 | * any next extents (if required). | |
484 | * | |
485 | * this also makes the decision about creating an inline extent vs | |
486 | * doing real data extents, marking pages dirty and delalloc as required. | |
487 | */ | |
be1a12a0 | 488 | int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode, |
48a3b636 ES |
489 | struct page **pages, size_t num_pages, |
490 | loff_t pos, size_t write_bytes, | |
491 | struct extent_state **cached) | |
39279cc3 | 492 | { |
39279cc3 | 493 | int err = 0; |
a52d9a80 | 494 | int i; |
db94535d | 495 | u64 num_bytes; |
a52d9a80 CM |
496 | u64 start_pos; |
497 | u64 end_of_last_block; | |
498 | u64 end_pos = pos + write_bytes; | |
499 | loff_t isize = i_size_read(inode); | |
39279cc3 | 500 | |
5f39d397 | 501 | start_pos = pos & ~((u64)root->sectorsize - 1); |
2e78c927 | 502 | num_bytes = round_up(write_bytes + pos - start_pos, root->sectorsize); |
39279cc3 | 503 | |
db94535d | 504 | end_of_last_block = start_pos + num_bytes - 1; |
2ac55d41 | 505 | err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block, |
be1a12a0 | 506 | cached); |
d0215f3e JB |
507 | if (err) |
508 | return err; | |
9ed74f2d | 509 | |
c8b97818 CM |
510 | for (i = 0; i < num_pages; i++) { |
511 | struct page *p = pages[i]; | |
512 | SetPageUptodate(p); | |
513 | ClearPageChecked(p); | |
514 | set_page_dirty(p); | |
a52d9a80 | 515 | } |
9f570b8d JB |
516 | |
517 | /* | |
518 | * we've only changed i_size in ram, and we haven't updated | |
519 | * the disk i_size. There is no need to log the inode | |
520 | * at this time. | |
521 | */ | |
522 | if (end_pos > isize) | |
a52d9a80 | 523 | i_size_write(inode, end_pos); |
a22285a6 | 524 | return 0; |
39279cc3 CM |
525 | } |
526 | ||
d352ac68 CM |
527 | /* |
528 | * this drops all the extents in the cache that intersect the range | |
529 | * [start, end]. Existing extents are split as required. | |
530 | */ | |
7014cdb4 JB |
531 | void btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end, |
532 | int skip_pinned) | |
a52d9a80 CM |
533 | { |
534 | struct extent_map *em; | |
3b951516 CM |
535 | struct extent_map *split = NULL; |
536 | struct extent_map *split2 = NULL; | |
a52d9a80 | 537 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
39b5637f | 538 | u64 len = end - start + 1; |
5dc562c5 | 539 | u64 gen; |
3b951516 CM |
540 | int ret; |
541 | int testend = 1; | |
5b21f2ed | 542 | unsigned long flags; |
c8b97818 | 543 | int compressed = 0; |
09a2a8f9 | 544 | bool modified; |
a52d9a80 | 545 | |
e6dcd2dc | 546 | WARN_ON(end < start); |
3b951516 | 547 | if (end == (u64)-1) { |
39b5637f | 548 | len = (u64)-1; |
3b951516 CM |
549 | testend = 0; |
550 | } | |
d397712b | 551 | while (1) { |
7014cdb4 JB |
552 | int no_splits = 0; |
553 | ||
09a2a8f9 | 554 | modified = false; |
3b951516 | 555 | if (!split) |
172ddd60 | 556 | split = alloc_extent_map(); |
3b951516 | 557 | if (!split2) |
172ddd60 | 558 | split2 = alloc_extent_map(); |
7014cdb4 JB |
559 | if (!split || !split2) |
560 | no_splits = 1; | |
3b951516 | 561 | |
890871be | 562 | write_lock(&em_tree->lock); |
39b5637f | 563 | em = lookup_extent_mapping(em_tree, start, len); |
d1310b2e | 564 | if (!em) { |
890871be | 565 | write_unlock(&em_tree->lock); |
a52d9a80 | 566 | break; |
d1310b2e | 567 | } |
5b21f2ed | 568 | flags = em->flags; |
5dc562c5 | 569 | gen = em->generation; |
5b21f2ed | 570 | if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) { |
55ef6899 | 571 | if (testend && em->start + em->len >= start + len) { |
5b21f2ed | 572 | free_extent_map(em); |
a1ed835e | 573 | write_unlock(&em_tree->lock); |
5b21f2ed ZY |
574 | break; |
575 | } | |
55ef6899 YZ |
576 | start = em->start + em->len; |
577 | if (testend) | |
5b21f2ed | 578 | len = start + len - (em->start + em->len); |
5b21f2ed | 579 | free_extent_map(em); |
a1ed835e | 580 | write_unlock(&em_tree->lock); |
5b21f2ed ZY |
581 | continue; |
582 | } | |
c8b97818 | 583 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
3ce7e67a | 584 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
3b277594 | 585 | clear_bit(EXTENT_FLAG_LOGGING, &flags); |
09a2a8f9 | 586 | modified = !list_empty(&em->list); |
7014cdb4 JB |
587 | if (no_splits) |
588 | goto next; | |
3b951516 | 589 | |
ee20a983 | 590 | if (em->start < start) { |
3b951516 CM |
591 | split->start = em->start; |
592 | split->len = start - em->start; | |
ee20a983 JB |
593 | |
594 | if (em->block_start < EXTENT_MAP_LAST_BYTE) { | |
595 | split->orig_start = em->orig_start; | |
596 | split->block_start = em->block_start; | |
597 | ||
598 | if (compressed) | |
599 | split->block_len = em->block_len; | |
600 | else | |
601 | split->block_len = split->len; | |
602 | split->orig_block_len = max(split->block_len, | |
603 | em->orig_block_len); | |
604 | split->ram_bytes = em->ram_bytes; | |
605 | } else { | |
606 | split->orig_start = split->start; | |
607 | split->block_len = 0; | |
608 | split->block_start = em->block_start; | |
609 | split->orig_block_len = 0; | |
610 | split->ram_bytes = split->len; | |
611 | } | |
612 | ||
5dc562c5 | 613 | split->generation = gen; |
3b951516 | 614 | split->bdev = em->bdev; |
5b21f2ed | 615 | split->flags = flags; |
261507a0 | 616 | split->compress_type = em->compress_type; |
176840b3 | 617 | replace_extent_mapping(em_tree, em, split, modified); |
3b951516 CM |
618 | free_extent_map(split); |
619 | split = split2; | |
620 | split2 = NULL; | |
621 | } | |
ee20a983 | 622 | if (testend && em->start + em->len > start + len) { |
3b951516 CM |
623 | u64 diff = start + len - em->start; |
624 | ||
625 | split->start = start + len; | |
626 | split->len = em->start + em->len - (start + len); | |
627 | split->bdev = em->bdev; | |
5b21f2ed | 628 | split->flags = flags; |
261507a0 | 629 | split->compress_type = em->compress_type; |
5dc562c5 | 630 | split->generation = gen; |
ee20a983 JB |
631 | |
632 | if (em->block_start < EXTENT_MAP_LAST_BYTE) { | |
633 | split->orig_block_len = max(em->block_len, | |
b4939680 | 634 | em->orig_block_len); |
3b951516 | 635 | |
ee20a983 JB |
636 | split->ram_bytes = em->ram_bytes; |
637 | if (compressed) { | |
638 | split->block_len = em->block_len; | |
639 | split->block_start = em->block_start; | |
640 | split->orig_start = em->orig_start; | |
641 | } else { | |
642 | split->block_len = split->len; | |
643 | split->block_start = em->block_start | |
644 | + diff; | |
645 | split->orig_start = em->orig_start; | |
646 | } | |
c8b97818 | 647 | } else { |
ee20a983 JB |
648 | split->ram_bytes = split->len; |
649 | split->orig_start = split->start; | |
650 | split->block_len = 0; | |
651 | split->block_start = em->block_start; | |
652 | split->orig_block_len = 0; | |
c8b97818 | 653 | } |
3b951516 | 654 | |
176840b3 FM |
655 | if (extent_map_in_tree(em)) { |
656 | replace_extent_mapping(em_tree, em, split, | |
657 | modified); | |
658 | } else { | |
659 | ret = add_extent_mapping(em_tree, split, | |
660 | modified); | |
661 | ASSERT(ret == 0); /* Logic error */ | |
662 | } | |
3b951516 CM |
663 | free_extent_map(split); |
664 | split = NULL; | |
665 | } | |
7014cdb4 | 666 | next: |
176840b3 FM |
667 | if (extent_map_in_tree(em)) |
668 | remove_extent_mapping(em_tree, em); | |
890871be | 669 | write_unlock(&em_tree->lock); |
d1310b2e | 670 | |
a52d9a80 CM |
671 | /* once for us */ |
672 | free_extent_map(em); | |
673 | /* once for the tree*/ | |
674 | free_extent_map(em); | |
675 | } | |
3b951516 CM |
676 | if (split) |
677 | free_extent_map(split); | |
678 | if (split2) | |
679 | free_extent_map(split2); | |
a52d9a80 CM |
680 | } |
681 | ||
39279cc3 CM |
682 | /* |
683 | * this is very complex, but the basic idea is to drop all extents | |
684 | * in the range start - end. hint_block is filled in with a block number | |
685 | * that would be a good hint to the block allocator for this file. | |
686 | * | |
687 | * If an extent intersects the range but is not entirely inside the range | |
688 | * it is either truncated or split. Anything entirely inside the range | |
689 | * is deleted from the tree. | |
690 | */ | |
5dc562c5 JB |
691 | int __btrfs_drop_extents(struct btrfs_trans_handle *trans, |
692 | struct btrfs_root *root, struct inode *inode, | |
693 | struct btrfs_path *path, u64 start, u64 end, | |
1acae57b FDBM |
694 | u64 *drop_end, int drop_cache, |
695 | int replace_extent, | |
696 | u32 extent_item_size, | |
697 | int *key_inserted) | |
39279cc3 | 698 | { |
5f39d397 | 699 | struct extent_buffer *leaf; |
920bbbfb | 700 | struct btrfs_file_extent_item *fi; |
00f5c795 | 701 | struct btrfs_key key; |
920bbbfb | 702 | struct btrfs_key new_key; |
33345d01 | 703 | u64 ino = btrfs_ino(inode); |
920bbbfb YZ |
704 | u64 search_start = start; |
705 | u64 disk_bytenr = 0; | |
706 | u64 num_bytes = 0; | |
707 | u64 extent_offset = 0; | |
708 | u64 extent_end = 0; | |
709 | int del_nr = 0; | |
710 | int del_slot = 0; | |
711 | int extent_type; | |
ccd467d6 | 712 | int recow; |
00f5c795 | 713 | int ret; |
dc7fdde3 | 714 | int modify_tree = -1; |
27cdeb70 | 715 | int update_refs; |
c3308f84 | 716 | int found = 0; |
1acae57b | 717 | int leafs_visited = 0; |
39279cc3 | 718 | |
a1ed835e CM |
719 | if (drop_cache) |
720 | btrfs_drop_extent_cache(inode, start, end - 1, 0); | |
a52d9a80 | 721 | |
d5f37527 | 722 | if (start >= BTRFS_I(inode)->disk_i_size && !replace_extent) |
dc7fdde3 CM |
723 | modify_tree = 0; |
724 | ||
27cdeb70 MX |
725 | update_refs = (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
726 | root == root->fs_info->tree_root); | |
d397712b | 727 | while (1) { |
ccd467d6 | 728 | recow = 0; |
33345d01 | 729 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
dc7fdde3 | 730 | search_start, modify_tree); |
39279cc3 | 731 | if (ret < 0) |
920bbbfb YZ |
732 | break; |
733 | if (ret > 0 && path->slots[0] > 0 && search_start == start) { | |
734 | leaf = path->nodes[0]; | |
735 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1); | |
33345d01 | 736 | if (key.objectid == ino && |
920bbbfb YZ |
737 | key.type == BTRFS_EXTENT_DATA_KEY) |
738 | path->slots[0]--; | |
39279cc3 | 739 | } |
920bbbfb | 740 | ret = 0; |
1acae57b | 741 | leafs_visited++; |
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 | } |
1acae57b | 753 | leafs_visited++; |
920bbbfb YZ |
754 | leaf = path->nodes[0]; |
755 | recow = 1; | |
756 | } | |
757 | ||
758 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
aeafbf84 FM |
759 | |
760 | if (key.objectid > ino) | |
761 | break; | |
762 | if (WARN_ON_ONCE(key.objectid < ino) || | |
763 | key.type < BTRFS_EXTENT_DATA_KEY) { | |
764 | ASSERT(del_nr == 0); | |
765 | path->slots[0]++; | |
766 | goto next_slot; | |
767 | } | |
768 | if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end) | |
920bbbfb YZ |
769 | break; |
770 | ||
771 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
772 | struct btrfs_file_extent_item); | |
773 | extent_type = btrfs_file_extent_type(leaf, fi); | |
774 | ||
775 | if (extent_type == BTRFS_FILE_EXTENT_REG || | |
776 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
777 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
778 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
779 | extent_offset = btrfs_file_extent_offset(leaf, fi); | |
780 | extent_end = key.offset + | |
781 | btrfs_file_extent_num_bytes(leaf, fi); | |
782 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
783 | extent_end = key.offset + | |
514ac8ad CM |
784 | btrfs_file_extent_inline_len(leaf, |
785 | path->slots[0], fi); | |
8c2383c3 | 786 | } else { |
aeafbf84 FM |
787 | /* can't happen */ |
788 | BUG(); | |
39279cc3 CM |
789 | } |
790 | ||
fc19c5e7 FM |
791 | /* |
792 | * Don't skip extent items representing 0 byte lengths. They | |
793 | * used to be created (bug) if while punching holes we hit | |
794 | * -ENOSPC condition. So if we find one here, just ensure we | |
795 | * delete it, otherwise we would insert a new file extent item | |
796 | * with the same key (offset) as that 0 bytes length file | |
797 | * extent item in the call to setup_items_for_insert() later | |
798 | * in this function. | |
799 | */ | |
800 | if (extent_end == key.offset && extent_end >= search_start) | |
801 | goto delete_extent_item; | |
802 | ||
920bbbfb YZ |
803 | if (extent_end <= search_start) { |
804 | path->slots[0]++; | |
8c2383c3 | 805 | goto next_slot; |
39279cc3 CM |
806 | } |
807 | ||
c3308f84 | 808 | found = 1; |
920bbbfb | 809 | search_start = max(key.offset, start); |
dc7fdde3 CM |
810 | if (recow || !modify_tree) { |
811 | modify_tree = -1; | |
b3b4aa74 | 812 | btrfs_release_path(path); |
920bbbfb | 813 | continue; |
39279cc3 | 814 | } |
6643558d | 815 | |
920bbbfb YZ |
816 | /* |
817 | * | - range to drop - | | |
818 | * | -------- extent -------- | | |
819 | */ | |
820 | if (start > key.offset && end < extent_end) { | |
821 | BUG_ON(del_nr > 0); | |
00fdf13a | 822 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
3f9e3df8 | 823 | ret = -EOPNOTSUPP; |
00fdf13a LB |
824 | break; |
825 | } | |
920bbbfb YZ |
826 | |
827 | memcpy(&new_key, &key, sizeof(new_key)); | |
828 | new_key.offset = start; | |
829 | ret = btrfs_duplicate_item(trans, root, path, | |
830 | &new_key); | |
831 | if (ret == -EAGAIN) { | |
b3b4aa74 | 832 | btrfs_release_path(path); |
920bbbfb | 833 | continue; |
6643558d | 834 | } |
920bbbfb YZ |
835 | if (ret < 0) |
836 | break; | |
837 | ||
838 | leaf = path->nodes[0]; | |
839 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
840 | struct btrfs_file_extent_item); | |
841 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
842 | start - key.offset); | |
843 | ||
844 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
845 | struct btrfs_file_extent_item); | |
846 | ||
847 | extent_offset += start - key.offset; | |
848 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); | |
849 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
850 | extent_end - start); | |
851 | btrfs_mark_buffer_dirty(leaf); | |
852 | ||
5dc562c5 | 853 | if (update_refs && disk_bytenr > 0) { |
771ed689 | 854 | ret = btrfs_inc_extent_ref(trans, root, |
920bbbfb YZ |
855 | disk_bytenr, num_bytes, 0, |
856 | root->root_key.objectid, | |
857 | new_key.objectid, | |
b06c4bf5 | 858 | start - extent_offset); |
79787eaa | 859 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 860 | } |
920bbbfb | 861 | key.offset = start; |
6643558d | 862 | } |
920bbbfb YZ |
863 | /* |
864 | * | ---- range to drop ----- | | |
865 | * | -------- extent -------- | | |
866 | */ | |
867 | if (start <= key.offset && end < extent_end) { | |
00fdf13a | 868 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
3f9e3df8 | 869 | ret = -EOPNOTSUPP; |
00fdf13a LB |
870 | break; |
871 | } | |
6643558d | 872 | |
920bbbfb YZ |
873 | memcpy(&new_key, &key, sizeof(new_key)); |
874 | new_key.offset = end; | |
b7a0365e | 875 | btrfs_set_item_key_safe(root->fs_info, path, &new_key); |
6643558d | 876 | |
920bbbfb YZ |
877 | extent_offset += end - key.offset; |
878 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); | |
879 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
880 | extent_end - end); | |
881 | btrfs_mark_buffer_dirty(leaf); | |
2671485d | 882 | if (update_refs && disk_bytenr > 0) |
920bbbfb | 883 | inode_sub_bytes(inode, end - key.offset); |
920bbbfb | 884 | break; |
39279cc3 | 885 | } |
771ed689 | 886 | |
920bbbfb YZ |
887 | search_start = extent_end; |
888 | /* | |
889 | * | ---- range to drop ----- | | |
890 | * | -------- extent -------- | | |
891 | */ | |
892 | if (start > key.offset && end >= extent_end) { | |
893 | BUG_ON(del_nr > 0); | |
00fdf13a | 894 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
3f9e3df8 | 895 | ret = -EOPNOTSUPP; |
00fdf13a LB |
896 | break; |
897 | } | |
8c2383c3 | 898 | |
920bbbfb YZ |
899 | btrfs_set_file_extent_num_bytes(leaf, fi, |
900 | start - key.offset); | |
901 | btrfs_mark_buffer_dirty(leaf); | |
2671485d | 902 | if (update_refs && disk_bytenr > 0) |
920bbbfb | 903 | inode_sub_bytes(inode, extent_end - start); |
920bbbfb YZ |
904 | if (end == extent_end) |
905 | break; | |
c8b97818 | 906 | |
920bbbfb YZ |
907 | path->slots[0]++; |
908 | goto next_slot; | |
31840ae1 ZY |
909 | } |
910 | ||
920bbbfb YZ |
911 | /* |
912 | * | ---- range to drop ----- | | |
913 | * | ------ extent ------ | | |
914 | */ | |
915 | if (start <= key.offset && end >= extent_end) { | |
fc19c5e7 | 916 | delete_extent_item: |
920bbbfb YZ |
917 | if (del_nr == 0) { |
918 | del_slot = path->slots[0]; | |
919 | del_nr = 1; | |
920 | } else { | |
921 | BUG_ON(del_slot + del_nr != path->slots[0]); | |
922 | del_nr++; | |
923 | } | |
31840ae1 | 924 | |
5dc562c5 JB |
925 | if (update_refs && |
926 | extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
a76a3cd4 | 927 | inode_sub_bytes(inode, |
920bbbfb YZ |
928 | extent_end - key.offset); |
929 | extent_end = ALIGN(extent_end, | |
930 | root->sectorsize); | |
5dc562c5 | 931 | } else if (update_refs && disk_bytenr > 0) { |
31840ae1 | 932 | ret = btrfs_free_extent(trans, root, |
920bbbfb YZ |
933 | disk_bytenr, num_bytes, 0, |
934 | root->root_key.objectid, | |
5d4f98a2 | 935 | key.objectid, key.offset - |
b06c4bf5 | 936 | extent_offset); |
79787eaa | 937 | BUG_ON(ret); /* -ENOMEM */ |
920bbbfb YZ |
938 | inode_sub_bytes(inode, |
939 | extent_end - key.offset); | |
31840ae1 | 940 | } |
31840ae1 | 941 | |
920bbbfb YZ |
942 | if (end == extent_end) |
943 | break; | |
944 | ||
945 | if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) { | |
946 | path->slots[0]++; | |
947 | goto next_slot; | |
948 | } | |
949 | ||
950 | ret = btrfs_del_items(trans, root, path, del_slot, | |
951 | del_nr); | |
79787eaa JM |
952 | if (ret) { |
953 | btrfs_abort_transaction(trans, root, ret); | |
5dc562c5 | 954 | break; |
79787eaa | 955 | } |
920bbbfb YZ |
956 | |
957 | del_nr = 0; | |
958 | del_slot = 0; | |
959 | ||
b3b4aa74 | 960 | btrfs_release_path(path); |
920bbbfb | 961 | continue; |
39279cc3 | 962 | } |
920bbbfb YZ |
963 | |
964 | BUG_ON(1); | |
39279cc3 | 965 | } |
920bbbfb | 966 | |
79787eaa | 967 | if (!ret && del_nr > 0) { |
1acae57b FDBM |
968 | /* |
969 | * Set path->slots[0] to first slot, so that after the delete | |
970 | * if items are move off from our leaf to its immediate left or | |
971 | * right neighbor leafs, we end up with a correct and adjusted | |
d5f37527 | 972 | * path->slots[0] for our insertion (if replace_extent != 0). |
1acae57b FDBM |
973 | */ |
974 | path->slots[0] = del_slot; | |
920bbbfb | 975 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
79787eaa JM |
976 | if (ret) |
977 | btrfs_abort_transaction(trans, root, ret); | |
d5f37527 | 978 | } |
1acae57b | 979 | |
d5f37527 FDBM |
980 | leaf = path->nodes[0]; |
981 | /* | |
982 | * If btrfs_del_items() was called, it might have deleted a leaf, in | |
983 | * which case it unlocked our path, so check path->locks[0] matches a | |
984 | * write lock. | |
985 | */ | |
986 | if (!ret && replace_extent && leafs_visited == 1 && | |
987 | (path->locks[0] == BTRFS_WRITE_LOCK_BLOCKING || | |
988 | path->locks[0] == BTRFS_WRITE_LOCK) && | |
989 | btrfs_leaf_free_space(root, leaf) >= | |
990 | sizeof(struct btrfs_item) + extent_item_size) { | |
991 | ||
992 | key.objectid = ino; | |
993 | key.type = BTRFS_EXTENT_DATA_KEY; | |
994 | key.offset = start; | |
995 | if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) { | |
996 | struct btrfs_key slot_key; | |
997 | ||
998 | btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]); | |
999 | if (btrfs_comp_cpu_keys(&key, &slot_key) > 0) | |
1000 | path->slots[0]++; | |
1acae57b | 1001 | } |
d5f37527 FDBM |
1002 | setup_items_for_insert(root, path, &key, |
1003 | &extent_item_size, | |
1004 | extent_item_size, | |
1005 | sizeof(struct btrfs_item) + | |
1006 | extent_item_size, 1); | |
1007 | *key_inserted = 1; | |
6643558d | 1008 | } |
920bbbfb | 1009 | |
1acae57b FDBM |
1010 | if (!replace_extent || !(*key_inserted)) |
1011 | btrfs_release_path(path); | |
2aaa6655 | 1012 | if (drop_end) |
c3308f84 | 1013 | *drop_end = found ? min(end, extent_end) : end; |
5dc562c5 JB |
1014 | return ret; |
1015 | } | |
1016 | ||
1017 | int btrfs_drop_extents(struct btrfs_trans_handle *trans, | |
1018 | struct btrfs_root *root, struct inode *inode, u64 start, | |
2671485d | 1019 | u64 end, int drop_cache) |
5dc562c5 JB |
1020 | { |
1021 | struct btrfs_path *path; | |
1022 | int ret; | |
1023 | ||
1024 | path = btrfs_alloc_path(); | |
1025 | if (!path) | |
1026 | return -ENOMEM; | |
2aaa6655 | 1027 | ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL, |
1acae57b | 1028 | drop_cache, 0, 0, NULL); |
920bbbfb | 1029 | btrfs_free_path(path); |
39279cc3 CM |
1030 | return ret; |
1031 | } | |
1032 | ||
d899e052 | 1033 | static int extent_mergeable(struct extent_buffer *leaf, int slot, |
6c7d54ac YZ |
1034 | u64 objectid, u64 bytenr, u64 orig_offset, |
1035 | u64 *start, u64 *end) | |
d899e052 YZ |
1036 | { |
1037 | struct btrfs_file_extent_item *fi; | |
1038 | struct btrfs_key key; | |
1039 | u64 extent_end; | |
1040 | ||
1041 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) | |
1042 | return 0; | |
1043 | ||
1044 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
1045 | if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY) | |
1046 | return 0; | |
1047 | ||
1048 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
1049 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG || | |
1050 | btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr || | |
6c7d54ac | 1051 | btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset || |
d899e052 YZ |
1052 | btrfs_file_extent_compression(leaf, fi) || |
1053 | btrfs_file_extent_encryption(leaf, fi) || | |
1054 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1055 | return 0; | |
1056 | ||
1057 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
1058 | if ((*start && *start != key.offset) || (*end && *end != extent_end)) | |
1059 | return 0; | |
1060 | ||
1061 | *start = key.offset; | |
1062 | *end = extent_end; | |
1063 | return 1; | |
1064 | } | |
1065 | ||
1066 | /* | |
1067 | * Mark extent in the range start - end as written. | |
1068 | * | |
1069 | * This changes extent type from 'pre-allocated' to 'regular'. If only | |
1070 | * part of extent is marked as written, the extent will be split into | |
1071 | * two or three. | |
1072 | */ | |
1073 | int btrfs_mark_extent_written(struct btrfs_trans_handle *trans, | |
d899e052 YZ |
1074 | struct inode *inode, u64 start, u64 end) |
1075 | { | |
920bbbfb | 1076 | struct btrfs_root *root = BTRFS_I(inode)->root; |
d899e052 YZ |
1077 | struct extent_buffer *leaf; |
1078 | struct btrfs_path *path; | |
1079 | struct btrfs_file_extent_item *fi; | |
1080 | struct btrfs_key key; | |
920bbbfb | 1081 | struct btrfs_key new_key; |
d899e052 YZ |
1082 | u64 bytenr; |
1083 | u64 num_bytes; | |
1084 | u64 extent_end; | |
5d4f98a2 | 1085 | u64 orig_offset; |
d899e052 YZ |
1086 | u64 other_start; |
1087 | u64 other_end; | |
920bbbfb YZ |
1088 | u64 split; |
1089 | int del_nr = 0; | |
1090 | int del_slot = 0; | |
6c7d54ac | 1091 | int recow; |
d899e052 | 1092 | int ret; |
33345d01 | 1093 | u64 ino = btrfs_ino(inode); |
d899e052 | 1094 | |
d899e052 | 1095 | path = btrfs_alloc_path(); |
d8926bb3 MF |
1096 | if (!path) |
1097 | return -ENOMEM; | |
d899e052 | 1098 | again: |
6c7d54ac | 1099 | recow = 0; |
920bbbfb | 1100 | split = start; |
33345d01 | 1101 | key.objectid = ino; |
d899e052 | 1102 | key.type = BTRFS_EXTENT_DATA_KEY; |
920bbbfb | 1103 | key.offset = split; |
d899e052 YZ |
1104 | |
1105 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
41415730 JB |
1106 | if (ret < 0) |
1107 | goto out; | |
d899e052 YZ |
1108 | if (ret > 0 && path->slots[0] > 0) |
1109 | path->slots[0]--; | |
1110 | ||
1111 | leaf = path->nodes[0]; | |
1112 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
33345d01 | 1113 | BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY); |
d899e052 YZ |
1114 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1115 | struct btrfs_file_extent_item); | |
920bbbfb YZ |
1116 | BUG_ON(btrfs_file_extent_type(leaf, fi) != |
1117 | BTRFS_FILE_EXTENT_PREALLOC); | |
d899e052 YZ |
1118 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
1119 | BUG_ON(key.offset > start || extent_end < end); | |
1120 | ||
1121 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
1122 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
5d4f98a2 | 1123 | orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi); |
6c7d54ac YZ |
1124 | memcpy(&new_key, &key, sizeof(new_key)); |
1125 | ||
1126 | if (start == key.offset && end < extent_end) { | |
1127 | other_start = 0; | |
1128 | other_end = start; | |
1129 | if (extent_mergeable(leaf, path->slots[0] - 1, | |
33345d01 | 1130 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1131 | &other_start, &other_end)) { |
1132 | new_key.offset = end; | |
b7a0365e | 1133 | btrfs_set_item_key_safe(root->fs_info, path, &new_key); |
6c7d54ac YZ |
1134 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1135 | struct btrfs_file_extent_item); | |
224ecce5 JB |
1136 | btrfs_set_file_extent_generation(leaf, fi, |
1137 | trans->transid); | |
6c7d54ac YZ |
1138 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1139 | extent_end - end); | |
1140 | btrfs_set_file_extent_offset(leaf, fi, | |
1141 | end - orig_offset); | |
1142 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
1143 | struct btrfs_file_extent_item); | |
224ecce5 JB |
1144 | btrfs_set_file_extent_generation(leaf, fi, |
1145 | trans->transid); | |
6c7d54ac YZ |
1146 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1147 | end - other_start); | |
1148 | btrfs_mark_buffer_dirty(leaf); | |
1149 | goto out; | |
1150 | } | |
1151 | } | |
1152 | ||
1153 | if (start > key.offset && end == extent_end) { | |
1154 | other_start = end; | |
1155 | other_end = 0; | |
1156 | if (extent_mergeable(leaf, path->slots[0] + 1, | |
33345d01 | 1157 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1158 | &other_start, &other_end)) { |
1159 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1160 | struct btrfs_file_extent_item); | |
1161 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
1162 | start - key.offset); | |
224ecce5 JB |
1163 | btrfs_set_file_extent_generation(leaf, fi, |
1164 | trans->transid); | |
6c7d54ac YZ |
1165 | path->slots[0]++; |
1166 | new_key.offset = start; | |
b7a0365e | 1167 | btrfs_set_item_key_safe(root->fs_info, path, &new_key); |
6c7d54ac YZ |
1168 | |
1169 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1170 | struct btrfs_file_extent_item); | |
224ecce5 JB |
1171 | btrfs_set_file_extent_generation(leaf, fi, |
1172 | trans->transid); | |
6c7d54ac YZ |
1173 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1174 | other_end - start); | |
1175 | btrfs_set_file_extent_offset(leaf, fi, | |
1176 | start - orig_offset); | |
1177 | btrfs_mark_buffer_dirty(leaf); | |
1178 | goto out; | |
1179 | } | |
1180 | } | |
d899e052 | 1181 | |
920bbbfb YZ |
1182 | while (start > key.offset || end < extent_end) { |
1183 | if (key.offset == start) | |
1184 | split = end; | |
1185 | ||
920bbbfb YZ |
1186 | new_key.offset = split; |
1187 | ret = btrfs_duplicate_item(trans, root, path, &new_key); | |
1188 | if (ret == -EAGAIN) { | |
b3b4aa74 | 1189 | btrfs_release_path(path); |
920bbbfb | 1190 | goto again; |
d899e052 | 1191 | } |
79787eaa JM |
1192 | if (ret < 0) { |
1193 | btrfs_abort_transaction(trans, root, ret); | |
1194 | goto out; | |
1195 | } | |
d899e052 | 1196 | |
920bbbfb YZ |
1197 | leaf = path->nodes[0]; |
1198 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
d899e052 | 1199 | struct btrfs_file_extent_item); |
224ecce5 | 1200 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
d899e052 | 1201 | btrfs_set_file_extent_num_bytes(leaf, fi, |
920bbbfb YZ |
1202 | split - key.offset); |
1203 | ||
1204 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1205 | struct btrfs_file_extent_item); | |
1206 | ||
224ecce5 | 1207 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
920bbbfb YZ |
1208 | btrfs_set_file_extent_offset(leaf, fi, split - orig_offset); |
1209 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
1210 | extent_end - split); | |
d899e052 YZ |
1211 | btrfs_mark_buffer_dirty(leaf); |
1212 | ||
920bbbfb YZ |
1213 | ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0, |
1214 | root->root_key.objectid, | |
b06c4bf5 | 1215 | ino, orig_offset); |
79787eaa | 1216 | BUG_ON(ret); /* -ENOMEM */ |
d899e052 | 1217 | |
920bbbfb YZ |
1218 | if (split == start) { |
1219 | key.offset = start; | |
1220 | } else { | |
1221 | BUG_ON(start != key.offset); | |
d899e052 | 1222 | path->slots[0]--; |
920bbbfb | 1223 | extent_end = end; |
d899e052 | 1224 | } |
6c7d54ac | 1225 | recow = 1; |
d899e052 YZ |
1226 | } |
1227 | ||
920bbbfb YZ |
1228 | other_start = end; |
1229 | other_end = 0; | |
6c7d54ac | 1230 | if (extent_mergeable(leaf, path->slots[0] + 1, |
33345d01 | 1231 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1232 | &other_start, &other_end)) { |
1233 | if (recow) { | |
b3b4aa74 | 1234 | btrfs_release_path(path); |
6c7d54ac YZ |
1235 | goto again; |
1236 | } | |
920bbbfb YZ |
1237 | extent_end = other_end; |
1238 | del_slot = path->slots[0] + 1; | |
1239 | del_nr++; | |
1240 | ret = btrfs_free_extent(trans, root, bytenr, num_bytes, | |
1241 | 0, root->root_key.objectid, | |
b06c4bf5 | 1242 | ino, orig_offset); |
79787eaa | 1243 | BUG_ON(ret); /* -ENOMEM */ |
d899e052 | 1244 | } |
920bbbfb YZ |
1245 | other_start = 0; |
1246 | other_end = start; | |
6c7d54ac | 1247 | if (extent_mergeable(leaf, path->slots[0] - 1, |
33345d01 | 1248 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1249 | &other_start, &other_end)) { |
1250 | if (recow) { | |
b3b4aa74 | 1251 | btrfs_release_path(path); |
6c7d54ac YZ |
1252 | goto again; |
1253 | } | |
920bbbfb YZ |
1254 | key.offset = other_start; |
1255 | del_slot = path->slots[0]; | |
1256 | del_nr++; | |
1257 | ret = btrfs_free_extent(trans, root, bytenr, num_bytes, | |
1258 | 0, root->root_key.objectid, | |
b06c4bf5 | 1259 | ino, orig_offset); |
79787eaa | 1260 | BUG_ON(ret); /* -ENOMEM */ |
920bbbfb YZ |
1261 | } |
1262 | if (del_nr == 0) { | |
3f6fae95 SL |
1263 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1264 | struct btrfs_file_extent_item); | |
920bbbfb YZ |
1265 | btrfs_set_file_extent_type(leaf, fi, |
1266 | BTRFS_FILE_EXTENT_REG); | |
224ecce5 | 1267 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
920bbbfb | 1268 | btrfs_mark_buffer_dirty(leaf); |
6c7d54ac | 1269 | } else { |
3f6fae95 SL |
1270 | fi = btrfs_item_ptr(leaf, del_slot - 1, |
1271 | struct btrfs_file_extent_item); | |
6c7d54ac YZ |
1272 | btrfs_set_file_extent_type(leaf, fi, |
1273 | BTRFS_FILE_EXTENT_REG); | |
224ecce5 | 1274 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
6c7d54ac YZ |
1275 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1276 | extent_end - key.offset); | |
1277 | btrfs_mark_buffer_dirty(leaf); | |
920bbbfb | 1278 | |
6c7d54ac | 1279 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
79787eaa JM |
1280 | if (ret < 0) { |
1281 | btrfs_abort_transaction(trans, root, ret); | |
1282 | goto out; | |
1283 | } | |
6c7d54ac | 1284 | } |
920bbbfb | 1285 | out: |
d899e052 YZ |
1286 | btrfs_free_path(path); |
1287 | return 0; | |
1288 | } | |
1289 | ||
b1bf862e CM |
1290 | /* |
1291 | * on error we return an unlocked page and the error value | |
1292 | * on success we return a locked page and 0 | |
1293 | */ | |
bb1591b4 CM |
1294 | static int prepare_uptodate_page(struct inode *inode, |
1295 | struct page *page, u64 pos, | |
b6316429 | 1296 | bool force_uptodate) |
b1bf862e CM |
1297 | { |
1298 | int ret = 0; | |
1299 | ||
b6316429 JB |
1300 | if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) && |
1301 | !PageUptodate(page)) { | |
b1bf862e CM |
1302 | ret = btrfs_readpage(NULL, page); |
1303 | if (ret) | |
1304 | return ret; | |
1305 | lock_page(page); | |
1306 | if (!PageUptodate(page)) { | |
1307 | unlock_page(page); | |
1308 | return -EIO; | |
1309 | } | |
bb1591b4 CM |
1310 | if (page->mapping != inode->i_mapping) { |
1311 | unlock_page(page); | |
1312 | return -EAGAIN; | |
1313 | } | |
b1bf862e CM |
1314 | } |
1315 | return 0; | |
1316 | } | |
1317 | ||
39279cc3 | 1318 | /* |
376cc685 | 1319 | * this just gets pages into the page cache and locks them down. |
39279cc3 | 1320 | */ |
b37392ea MX |
1321 | static noinline int prepare_pages(struct inode *inode, struct page **pages, |
1322 | size_t num_pages, loff_t pos, | |
1323 | size_t write_bytes, bool force_uptodate) | |
39279cc3 CM |
1324 | { |
1325 | int i; | |
1326 | unsigned long index = pos >> PAGE_CACHE_SHIFT; | |
3b16a4e3 | 1327 | gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); |
fc28b62d | 1328 | int err = 0; |
376cc685 | 1329 | int faili; |
8c2383c3 | 1330 | |
39279cc3 | 1331 | for (i = 0; i < num_pages; i++) { |
bb1591b4 | 1332 | again: |
a94733d0 | 1333 | pages[i] = find_or_create_page(inode->i_mapping, index + i, |
e3a41a5b | 1334 | mask | __GFP_WRITE); |
39279cc3 | 1335 | if (!pages[i]) { |
b1bf862e CM |
1336 | faili = i - 1; |
1337 | err = -ENOMEM; | |
1338 | goto fail; | |
1339 | } | |
1340 | ||
1341 | if (i == 0) | |
bb1591b4 | 1342 | err = prepare_uptodate_page(inode, pages[i], pos, |
b6316429 | 1343 | force_uptodate); |
bb1591b4 CM |
1344 | if (!err && i == num_pages - 1) |
1345 | err = prepare_uptodate_page(inode, pages[i], | |
b6316429 | 1346 | pos + write_bytes, false); |
b1bf862e CM |
1347 | if (err) { |
1348 | page_cache_release(pages[i]); | |
bb1591b4 CM |
1349 | if (err == -EAGAIN) { |
1350 | err = 0; | |
1351 | goto again; | |
1352 | } | |
b1bf862e CM |
1353 | faili = i - 1; |
1354 | goto fail; | |
39279cc3 | 1355 | } |
ccd467d6 | 1356 | wait_on_page_writeback(pages[i]); |
39279cc3 | 1357 | } |
376cc685 MX |
1358 | |
1359 | return 0; | |
1360 | fail: | |
1361 | while (faili >= 0) { | |
1362 | unlock_page(pages[faili]); | |
1363 | page_cache_release(pages[faili]); | |
1364 | faili--; | |
1365 | } | |
1366 | return err; | |
1367 | ||
1368 | } | |
1369 | ||
1370 | /* | |
1371 | * This function locks the extent and properly waits for data=ordered extents | |
1372 | * to finish before allowing the pages to be modified if need. | |
1373 | * | |
1374 | * The return value: | |
1375 | * 1 - the extent is locked | |
1376 | * 0 - the extent is not locked, and everything is OK | |
1377 | * -EAGAIN - need re-prepare the pages | |
1378 | * the other < 0 number - Something wrong happens | |
1379 | */ | |
1380 | static noinline int | |
1381 | lock_and_cleanup_extent_if_need(struct inode *inode, struct page **pages, | |
1382 | size_t num_pages, loff_t pos, | |
2e78c927 | 1383 | size_t write_bytes, |
376cc685 MX |
1384 | u64 *lockstart, u64 *lockend, |
1385 | struct extent_state **cached_state) | |
1386 | { | |
2e78c927 | 1387 | struct btrfs_root *root = BTRFS_I(inode)->root; |
376cc685 MX |
1388 | u64 start_pos; |
1389 | u64 last_pos; | |
1390 | int i; | |
1391 | int ret = 0; | |
1392 | ||
2e78c927 CR |
1393 | start_pos = round_down(pos, root->sectorsize); |
1394 | last_pos = start_pos | |
1395 | + round_up(pos + write_bytes - start_pos, root->sectorsize) - 1; | |
376cc685 | 1396 | |
0762704b | 1397 | if (start_pos < inode->i_size) { |
e6dcd2dc | 1398 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 1399 | lock_extent_bits(&BTRFS_I(inode)->io_tree, |
ff13db41 | 1400 | start_pos, last_pos, cached_state); |
b88935bf MX |
1401 | ordered = btrfs_lookup_ordered_range(inode, start_pos, |
1402 | last_pos - start_pos + 1); | |
e6dcd2dc CM |
1403 | if (ordered && |
1404 | ordered->file_offset + ordered->len > start_pos && | |
376cc685 | 1405 | ordered->file_offset <= last_pos) { |
2ac55d41 | 1406 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, |
376cc685 MX |
1407 | start_pos, last_pos, |
1408 | cached_state, GFP_NOFS); | |
e6dcd2dc CM |
1409 | for (i = 0; i < num_pages; i++) { |
1410 | unlock_page(pages[i]); | |
1411 | page_cache_release(pages[i]); | |
1412 | } | |
b88935bf MX |
1413 | btrfs_start_ordered_extent(inode, ordered, 1); |
1414 | btrfs_put_ordered_extent(ordered); | |
1415 | return -EAGAIN; | |
e6dcd2dc CM |
1416 | } |
1417 | if (ordered) | |
1418 | btrfs_put_ordered_extent(ordered); | |
1419 | ||
2ac55d41 | 1420 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos, |
376cc685 | 1421 | last_pos, EXTENT_DIRTY | EXTENT_DELALLOC | |
9e8a4a8b | 1422 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, |
376cc685 MX |
1423 | 0, 0, cached_state, GFP_NOFS); |
1424 | *lockstart = start_pos; | |
1425 | *lockend = last_pos; | |
1426 | ret = 1; | |
0762704b | 1427 | } |
376cc685 | 1428 | |
e6dcd2dc | 1429 | for (i = 0; i < num_pages; i++) { |
32c7f202 WF |
1430 | if (clear_page_dirty_for_io(pages[i])) |
1431 | account_page_redirty(pages[i]); | |
e6dcd2dc CM |
1432 | set_page_extent_mapped(pages[i]); |
1433 | WARN_ON(!PageLocked(pages[i])); | |
1434 | } | |
b1bf862e | 1435 | |
376cc685 | 1436 | return ret; |
39279cc3 CM |
1437 | } |
1438 | ||
7ee9e440 JB |
1439 | static noinline int check_can_nocow(struct inode *inode, loff_t pos, |
1440 | size_t *write_bytes) | |
1441 | { | |
7ee9e440 JB |
1442 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1443 | struct btrfs_ordered_extent *ordered; | |
1444 | u64 lockstart, lockend; | |
1445 | u64 num_bytes; | |
1446 | int ret; | |
1447 | ||
9ea24bbe | 1448 | ret = btrfs_start_write_no_snapshoting(root); |
8257b2dc MX |
1449 | if (!ret) |
1450 | return -ENOSPC; | |
1451 | ||
7ee9e440 | 1452 | lockstart = round_down(pos, root->sectorsize); |
c933956d | 1453 | lockend = round_up(pos + *write_bytes, root->sectorsize) - 1; |
7ee9e440 JB |
1454 | |
1455 | while (1) { | |
1456 | lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend); | |
1457 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
1458 | lockend - lockstart + 1); | |
1459 | if (!ordered) { | |
1460 | break; | |
1461 | } | |
1462 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend); | |
1463 | btrfs_start_ordered_extent(inode, ordered, 1); | |
1464 | btrfs_put_ordered_extent(ordered); | |
1465 | } | |
1466 | ||
7ee9e440 | 1467 | num_bytes = lockend - lockstart + 1; |
00361589 | 1468 | ret = can_nocow_extent(inode, lockstart, &num_bytes, NULL, NULL, NULL); |
7ee9e440 JB |
1469 | if (ret <= 0) { |
1470 | ret = 0; | |
9ea24bbe | 1471 | btrfs_end_write_no_snapshoting(root); |
7ee9e440 | 1472 | } else { |
c933956d MX |
1473 | *write_bytes = min_t(size_t, *write_bytes , |
1474 | num_bytes - pos + lockstart); | |
7ee9e440 JB |
1475 | } |
1476 | ||
1477 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend); | |
1478 | ||
1479 | return ret; | |
1480 | } | |
1481 | ||
d0215f3e JB |
1482 | static noinline ssize_t __btrfs_buffered_write(struct file *file, |
1483 | struct iov_iter *i, | |
1484 | loff_t pos) | |
4b46fce2 | 1485 | { |
496ad9aa | 1486 | struct inode *inode = file_inode(file); |
11c65dcc | 1487 | struct btrfs_root *root = BTRFS_I(inode)->root; |
11c65dcc | 1488 | struct page **pages = NULL; |
376cc685 | 1489 | struct extent_state *cached_state = NULL; |
7ee9e440 | 1490 | u64 release_bytes = 0; |
376cc685 MX |
1491 | u64 lockstart; |
1492 | u64 lockend; | |
d0215f3e JB |
1493 | size_t num_written = 0; |
1494 | int nrptrs; | |
c9149235 | 1495 | int ret = 0; |
7ee9e440 | 1496 | bool only_release_metadata = false; |
b6316429 | 1497 | bool force_page_uptodate = false; |
376cc685 | 1498 | bool need_unlock; |
4b46fce2 | 1499 | |
ed6078f7 DS |
1500 | nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_CACHE_SIZE), |
1501 | PAGE_CACHE_SIZE / (sizeof(struct page *))); | |
142349f5 WF |
1502 | nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied); |
1503 | nrptrs = max(nrptrs, 8); | |
31e818fe | 1504 | pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL); |
d0215f3e JB |
1505 | if (!pages) |
1506 | return -ENOMEM; | |
ab93dbec | 1507 | |
d0215f3e | 1508 | while (iov_iter_count(i) > 0) { |
39279cc3 | 1509 | size_t offset = pos & (PAGE_CACHE_SIZE - 1); |
2e78c927 | 1510 | size_t sector_offset; |
d0215f3e | 1511 | size_t write_bytes = min(iov_iter_count(i), |
11c65dcc | 1512 | nrptrs * (size_t)PAGE_CACHE_SIZE - |
8c2383c3 | 1513 | offset); |
ed6078f7 DS |
1514 | size_t num_pages = DIV_ROUND_UP(write_bytes + offset, |
1515 | PAGE_CACHE_SIZE); | |
7ee9e440 | 1516 | size_t reserve_bytes; |
d0215f3e JB |
1517 | size_t dirty_pages; |
1518 | size_t copied; | |
2e78c927 CR |
1519 | size_t dirty_sectors; |
1520 | size_t num_sectors; | |
39279cc3 | 1521 | |
8c2383c3 | 1522 | WARN_ON(num_pages > nrptrs); |
1832a6d5 | 1523 | |
914ee295 XZ |
1524 | /* |
1525 | * Fault pages before locking them in prepare_pages | |
1526 | * to avoid recursive lock | |
1527 | */ | |
d0215f3e | 1528 | if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) { |
914ee295 | 1529 | ret = -EFAULT; |
d0215f3e | 1530 | break; |
914ee295 XZ |
1531 | } |
1532 | ||
2e78c927 CR |
1533 | sector_offset = pos & (root->sectorsize - 1); |
1534 | reserve_bytes = round_up(write_bytes + sector_offset, | |
1535 | root->sectorsize); | |
d9d8b2a5 | 1536 | |
4da2e26a ZL |
1537 | if ((BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW | |
1538 | BTRFS_INODE_PREALLOC)) && | |
1539 | check_can_nocow(inode, pos, &write_bytes) > 0) { | |
1540 | /* | |
1541 | * For nodata cow case, no need to reserve | |
1542 | * data space. | |
1543 | */ | |
1544 | only_release_metadata = true; | |
1545 | /* | |
1546 | * our prealloc extent may be smaller than | |
1547 | * write_bytes, so scale down. | |
1548 | */ | |
1549 | num_pages = DIV_ROUND_UP(write_bytes + offset, | |
1550 | PAGE_CACHE_SIZE); | |
fa695b01 DS |
1551 | reserve_bytes = round_up(write_bytes + sector_offset, |
1552 | root->sectorsize); | |
4da2e26a | 1553 | goto reserve_metadata; |
7ee9e440 | 1554 | } |
4da2e26a | 1555 | |
7cf5b976 | 1556 | ret = btrfs_check_data_free_space(inode, pos, write_bytes); |
d9d8b2a5 | 1557 | if (ret < 0) |
d0215f3e | 1558 | break; |
1832a6d5 | 1559 | |
d9d8b2a5 | 1560 | reserve_metadata: |
7ee9e440 JB |
1561 | ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes); |
1562 | if (ret) { | |
1563 | if (!only_release_metadata) | |
7cf5b976 QW |
1564 | btrfs_free_reserved_data_space(inode, pos, |
1565 | write_bytes); | |
8257b2dc | 1566 | else |
9ea24bbe | 1567 | btrfs_end_write_no_snapshoting(root); |
7ee9e440 JB |
1568 | break; |
1569 | } | |
1570 | ||
1571 | release_bytes = reserve_bytes; | |
376cc685 MX |
1572 | need_unlock = false; |
1573 | again: | |
4a64001f JB |
1574 | /* |
1575 | * This is going to setup the pages array with the number of | |
1576 | * pages we want, so we don't really need to worry about the | |
1577 | * contents of pages from loop to loop | |
1578 | */ | |
b37392ea MX |
1579 | ret = prepare_pages(inode, pages, num_pages, |
1580 | pos, write_bytes, | |
b6316429 | 1581 | force_page_uptodate); |
7ee9e440 | 1582 | if (ret) |
d0215f3e | 1583 | break; |
39279cc3 | 1584 | |
376cc685 | 1585 | ret = lock_and_cleanup_extent_if_need(inode, pages, num_pages, |
2e78c927 CR |
1586 | pos, write_bytes, &lockstart, |
1587 | &lockend, &cached_state); | |
376cc685 MX |
1588 | if (ret < 0) { |
1589 | if (ret == -EAGAIN) | |
1590 | goto again; | |
1591 | break; | |
1592 | } else if (ret > 0) { | |
1593 | need_unlock = true; | |
1594 | ret = 0; | |
1595 | } | |
1596 | ||
ee22f0c4 | 1597 | copied = btrfs_copy_from_user(pos, write_bytes, pages, i); |
b1bf862e CM |
1598 | |
1599 | /* | |
1600 | * if we have trouble faulting in the pages, fall | |
1601 | * back to one page at a time | |
1602 | */ | |
1603 | if (copied < write_bytes) | |
1604 | nrptrs = 1; | |
1605 | ||
b6316429 JB |
1606 | if (copied == 0) { |
1607 | force_page_uptodate = true; | |
b1bf862e | 1608 | dirty_pages = 0; |
b6316429 JB |
1609 | } else { |
1610 | force_page_uptodate = false; | |
ed6078f7 DS |
1611 | dirty_pages = DIV_ROUND_UP(copied + offset, |
1612 | PAGE_CACHE_SIZE); | |
b6316429 | 1613 | } |
914ee295 | 1614 | |
d0215f3e JB |
1615 | /* |
1616 | * If we had a short copy we need to release the excess delaloc | |
1617 | * bytes we reserved. We need to increment outstanding_extents | |
1618 | * because btrfs_delalloc_release_space will decrement it, but | |
1619 | * we still have an outstanding extent for the chunk we actually | |
1620 | * managed to copy. | |
1621 | */ | |
2e78c927 CR |
1622 | num_sectors = BTRFS_BYTES_TO_BLKS(root->fs_info, |
1623 | reserve_bytes); | |
1624 | dirty_sectors = round_up(copied + sector_offset, | |
1625 | root->sectorsize); | |
1626 | dirty_sectors = BTRFS_BYTES_TO_BLKS(root->fs_info, | |
1627 | dirty_sectors); | |
1628 | ||
1629 | if (num_sectors > dirty_sectors) { | |
1630 | release_bytes = (write_bytes - copied) | |
1631 | & ~((u64)root->sectorsize - 1); | |
9e0baf60 JB |
1632 | if (copied > 0) { |
1633 | spin_lock(&BTRFS_I(inode)->lock); | |
1634 | BTRFS_I(inode)->outstanding_extents++; | |
1635 | spin_unlock(&BTRFS_I(inode)->lock); | |
1636 | } | |
485290a7 | 1637 | if (only_release_metadata) { |
7ee9e440 JB |
1638 | btrfs_delalloc_release_metadata(inode, |
1639 | release_bytes); | |
485290a7 QW |
1640 | } else { |
1641 | u64 __pos; | |
1642 | ||
1643 | __pos = round_down(pos, root->sectorsize) + | |
1644 | (dirty_pages << PAGE_CACHE_SHIFT); | |
1645 | btrfs_delalloc_release_space(inode, __pos, | |
7ee9e440 | 1646 | release_bytes); |
485290a7 | 1647 | } |
914ee295 XZ |
1648 | } |
1649 | ||
2e78c927 CR |
1650 | release_bytes = round_up(copied + sector_offset, |
1651 | root->sectorsize); | |
376cc685 MX |
1652 | |
1653 | if (copied > 0) | |
be1a12a0 JB |
1654 | ret = btrfs_dirty_pages(root, inode, pages, |
1655 | dirty_pages, pos, copied, | |
1656 | NULL); | |
376cc685 MX |
1657 | if (need_unlock) |
1658 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, | |
1659 | lockstart, lockend, &cached_state, | |
1660 | GFP_NOFS); | |
f1de9683 MX |
1661 | if (ret) { |
1662 | btrfs_drop_pages(pages, num_pages); | |
376cc685 | 1663 | break; |
f1de9683 | 1664 | } |
39279cc3 | 1665 | |
376cc685 | 1666 | release_bytes = 0; |
8257b2dc | 1667 | if (only_release_metadata) |
9ea24bbe | 1668 | btrfs_end_write_no_snapshoting(root); |
8257b2dc | 1669 | |
7ee9e440 | 1670 | if (only_release_metadata && copied > 0) { |
f64c7b12 | 1671 | lockstart = round_down(pos, root->sectorsize); |
2e78c927 | 1672 | lockend = round_up(pos + copied, root->sectorsize) - 1; |
7ee9e440 JB |
1673 | |
1674 | set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
1675 | lockend, EXTENT_NORESERVE, NULL, | |
1676 | NULL, GFP_NOFS); | |
1677 | only_release_metadata = false; | |
1678 | } | |
1679 | ||
f1de9683 MX |
1680 | btrfs_drop_pages(pages, num_pages); |
1681 | ||
d0215f3e JB |
1682 | cond_resched(); |
1683 | ||
d0e1d66b | 1684 | balance_dirty_pages_ratelimited(inode->i_mapping); |
707e8a07 | 1685 | if (dirty_pages < (root->nodesize >> PAGE_CACHE_SHIFT) + 1) |
b53d3f5d | 1686 | btrfs_btree_balance_dirty(root); |
cb843a6f | 1687 | |
914ee295 XZ |
1688 | pos += copied; |
1689 | num_written += copied; | |
d0215f3e | 1690 | } |
39279cc3 | 1691 | |
d0215f3e JB |
1692 | kfree(pages); |
1693 | ||
7ee9e440 | 1694 | if (release_bytes) { |
8257b2dc | 1695 | if (only_release_metadata) { |
9ea24bbe | 1696 | btrfs_end_write_no_snapshoting(root); |
7ee9e440 | 1697 | btrfs_delalloc_release_metadata(inode, release_bytes); |
8257b2dc | 1698 | } else { |
7cf5b976 | 1699 | btrfs_delalloc_release_space(inode, pos, release_bytes); |
8257b2dc | 1700 | } |
7ee9e440 JB |
1701 | } |
1702 | ||
d0215f3e JB |
1703 | return num_written ? num_written : ret; |
1704 | } | |
1705 | ||
1706 | static ssize_t __btrfs_direct_write(struct kiocb *iocb, | |
0ae5e4d3 | 1707 | struct iov_iter *from, |
0c949334 | 1708 | loff_t pos) |
d0215f3e JB |
1709 | { |
1710 | struct file *file = iocb->ki_filp; | |
728404da | 1711 | struct inode *inode = file_inode(file); |
d0215f3e JB |
1712 | ssize_t written; |
1713 | ssize_t written_buffered; | |
1714 | loff_t endbyte; | |
1715 | int err; | |
1716 | ||
0c949334 | 1717 | written = generic_file_direct_write(iocb, from, pos); |
d0215f3e | 1718 | |
0c949334 | 1719 | if (written < 0 || !iov_iter_count(from)) |
d0215f3e JB |
1720 | return written; |
1721 | ||
1722 | pos += written; | |
0ae5e4d3 | 1723 | written_buffered = __btrfs_buffered_write(file, from, pos); |
d0215f3e JB |
1724 | if (written_buffered < 0) { |
1725 | err = written_buffered; | |
1726 | goto out; | |
39279cc3 | 1727 | } |
075bdbdb FM |
1728 | /* |
1729 | * Ensure all data is persisted. We want the next direct IO read to be | |
1730 | * able to read what was just written. | |
1731 | */ | |
d0215f3e | 1732 | endbyte = pos + written_buffered - 1; |
728404da | 1733 | err = btrfs_fdatawrite_range(inode, pos, endbyte); |
075bdbdb FM |
1734 | if (err) |
1735 | goto out; | |
728404da | 1736 | err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte); |
d0215f3e JB |
1737 | if (err) |
1738 | goto out; | |
1739 | written += written_buffered; | |
867c4f93 | 1740 | iocb->ki_pos = pos + written_buffered; |
d0215f3e JB |
1741 | invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT, |
1742 | endbyte >> PAGE_CACHE_SHIFT); | |
39279cc3 | 1743 | out: |
d0215f3e JB |
1744 | return written ? written : err; |
1745 | } | |
5b92ee72 | 1746 | |
6c760c07 JB |
1747 | static void update_time_for_write(struct inode *inode) |
1748 | { | |
1749 | struct timespec now; | |
1750 | ||
1751 | if (IS_NOCMTIME(inode)) | |
1752 | return; | |
1753 | ||
1754 | now = current_fs_time(inode->i_sb); | |
1755 | if (!timespec_equal(&inode->i_mtime, &now)) | |
1756 | inode->i_mtime = now; | |
1757 | ||
1758 | if (!timespec_equal(&inode->i_ctime, &now)) | |
1759 | inode->i_ctime = now; | |
1760 | ||
1761 | if (IS_I_VERSION(inode)) | |
1762 | inode_inc_iversion(inode); | |
1763 | } | |
1764 | ||
b30ac0fc AV |
1765 | static ssize_t btrfs_file_write_iter(struct kiocb *iocb, |
1766 | struct iov_iter *from) | |
d0215f3e JB |
1767 | { |
1768 | struct file *file = iocb->ki_filp; | |
496ad9aa | 1769 | struct inode *inode = file_inode(file); |
d0215f3e | 1770 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0c1a98c8 | 1771 | u64 start_pos; |
3ac0d7b9 | 1772 | u64 end_pos; |
d0215f3e | 1773 | ssize_t num_written = 0; |
b812ce28 | 1774 | bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host); |
3309dd04 AV |
1775 | ssize_t err; |
1776 | loff_t pos; | |
1777 | size_t count; | |
27772b68 CR |
1778 | loff_t oldsize; |
1779 | int clean_page = 0; | |
d0215f3e | 1780 | |
5955102c | 1781 | inode_lock(inode); |
3309dd04 AV |
1782 | err = generic_write_checks(iocb, from); |
1783 | if (err <= 0) { | |
5955102c | 1784 | inode_unlock(inode); |
3309dd04 | 1785 | return err; |
d0215f3e JB |
1786 | } |
1787 | ||
3309dd04 | 1788 | current->backing_dev_info = inode_to_bdi(inode); |
5fa8e0a1 | 1789 | err = file_remove_privs(file); |
d0215f3e | 1790 | if (err) { |
5955102c | 1791 | inode_unlock(inode); |
d0215f3e JB |
1792 | goto out; |
1793 | } | |
1794 | ||
1795 | /* | |
1796 | * If BTRFS flips readonly due to some impossible error | |
1797 | * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR), | |
1798 | * although we have opened a file as writable, we have | |
1799 | * to stop this write operation to ensure FS consistency. | |
1800 | */ | |
87533c47 | 1801 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) { |
5955102c | 1802 | inode_unlock(inode); |
d0215f3e JB |
1803 | err = -EROFS; |
1804 | goto out; | |
1805 | } | |
1806 | ||
6c760c07 JB |
1807 | /* |
1808 | * We reserve space for updating the inode when we reserve space for the | |
1809 | * extent we are going to write, so we will enospc out there. We don't | |
1810 | * need to start yet another transaction to update the inode as we will | |
1811 | * update the inode when we finish writing whatever data we write. | |
1812 | */ | |
1813 | update_time_for_write(inode); | |
d0215f3e | 1814 | |
3309dd04 AV |
1815 | pos = iocb->ki_pos; |
1816 | count = iov_iter_count(from); | |
0c1a98c8 | 1817 | start_pos = round_down(pos, root->sectorsize); |
27772b68 CR |
1818 | oldsize = i_size_read(inode); |
1819 | if (start_pos > oldsize) { | |
3ac0d7b9 | 1820 | /* Expand hole size to cover write data, preventing empty gap */ |
c5f7d0bb | 1821 | end_pos = round_up(pos + count, root->sectorsize); |
27772b68 | 1822 | err = btrfs_cont_expand(inode, oldsize, end_pos); |
0c1a98c8 | 1823 | if (err) { |
5955102c | 1824 | inode_unlock(inode); |
0c1a98c8 MX |
1825 | goto out; |
1826 | } | |
27772b68 CR |
1827 | if (start_pos > round_up(oldsize, root->sectorsize)) |
1828 | clean_page = 1; | |
0c1a98c8 MX |
1829 | } |
1830 | ||
b812ce28 JB |
1831 | if (sync) |
1832 | atomic_inc(&BTRFS_I(inode)->sync_writers); | |
1833 | ||
2ba48ce5 | 1834 | if (iocb->ki_flags & IOCB_DIRECT) { |
b30ac0fc | 1835 | num_written = __btrfs_direct_write(iocb, from, pos); |
d0215f3e | 1836 | } else { |
b30ac0fc | 1837 | num_written = __btrfs_buffered_write(file, from, pos); |
d0215f3e | 1838 | if (num_written > 0) |
867c4f93 | 1839 | iocb->ki_pos = pos + num_written; |
27772b68 CR |
1840 | if (clean_page) |
1841 | pagecache_isize_extended(inode, oldsize, | |
1842 | i_size_read(inode)); | |
d0215f3e JB |
1843 | } |
1844 | ||
5955102c | 1845 | inode_unlock(inode); |
2ff3e9b6 | 1846 | |
5a3f23d5 | 1847 | /* |
6c760c07 JB |
1848 | * We also have to set last_sub_trans to the current log transid, |
1849 | * otherwise subsequent syncs to a file that's been synced in this | |
bb7ab3b9 | 1850 | * transaction will appear to have already occurred. |
5a3f23d5 | 1851 | */ |
2f2ff0ee | 1852 | spin_lock(&BTRFS_I(inode)->lock); |
6c760c07 | 1853 | BTRFS_I(inode)->last_sub_trans = root->log_transid; |
2f2ff0ee | 1854 | spin_unlock(&BTRFS_I(inode)->lock); |
02afc27f | 1855 | if (num_written > 0) { |
d0215f3e | 1856 | err = generic_write_sync(file, pos, num_written); |
45d4f855 | 1857 | if (err < 0) |
2ff3e9b6 CM |
1858 | num_written = err; |
1859 | } | |
0a3404dc | 1860 | |
b812ce28 JB |
1861 | if (sync) |
1862 | atomic_dec(&BTRFS_I(inode)->sync_writers); | |
0a3404dc | 1863 | out: |
39279cc3 | 1864 | current->backing_dev_info = NULL; |
39279cc3 CM |
1865 | return num_written ? num_written : err; |
1866 | } | |
1867 | ||
d397712b | 1868 | int btrfs_release_file(struct inode *inode, struct file *filp) |
e1b81e67 | 1869 | { |
6bf13c0c SW |
1870 | if (filp->private_data) |
1871 | btrfs_ioctl_trans_end(filp); | |
f6dc45c7 CM |
1872 | /* |
1873 | * ordered_data_close is set by settattr when we are about to truncate | |
1874 | * a file from a non-zero size to a zero size. This tries to | |
1875 | * flush down new bytes that may have been written if the | |
1876 | * application were using truncate to replace a file in place. | |
1877 | */ | |
1878 | if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, | |
1879 | &BTRFS_I(inode)->runtime_flags)) | |
1880 | filemap_flush(inode->i_mapping); | |
e1b81e67 M |
1881 | return 0; |
1882 | } | |
1883 | ||
669249ee FM |
1884 | static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end) |
1885 | { | |
1886 | int ret; | |
1887 | ||
1888 | atomic_inc(&BTRFS_I(inode)->sync_writers); | |
728404da | 1889 | ret = btrfs_fdatawrite_range(inode, start, end); |
669249ee FM |
1890 | atomic_dec(&BTRFS_I(inode)->sync_writers); |
1891 | ||
1892 | return ret; | |
1893 | } | |
1894 | ||
d352ac68 CM |
1895 | /* |
1896 | * fsync call for both files and directories. This logs the inode into | |
1897 | * the tree log instead of forcing full commits whenever possible. | |
1898 | * | |
1899 | * It needs to call filemap_fdatawait so that all ordered extent updates are | |
1900 | * in the metadata btree are up to date for copying to the log. | |
1901 | * | |
1902 | * It drops the inode mutex before doing the tree log commit. This is an | |
1903 | * important optimization for directories because holding the mutex prevents | |
1904 | * new operations on the dir while we write to disk. | |
1905 | */ | |
02c24a82 | 1906 | int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync) |
39279cc3 | 1907 | { |
7ea80859 | 1908 | struct dentry *dentry = file->f_path.dentry; |
2b0143b5 | 1909 | struct inode *inode = d_inode(dentry); |
39279cc3 | 1910 | struct btrfs_root *root = BTRFS_I(inode)->root; |
39279cc3 | 1911 | struct btrfs_trans_handle *trans; |
8b050d35 MX |
1912 | struct btrfs_log_ctx ctx; |
1913 | int ret = 0; | |
2ab28f32 | 1914 | bool full_sync = 0; |
9dcbeed4 | 1915 | u64 len; |
39279cc3 | 1916 | |
9dcbeed4 DS |
1917 | /* |
1918 | * The range length can be represented by u64, we have to do the typecasts | |
1919 | * to avoid signed overflow if it's [0, LLONG_MAX] eg. from fsync() | |
1920 | */ | |
1921 | len = (u64)end - (u64)start + 1; | |
1abe9b8a | 1922 | trace_btrfs_sync_file(file, datasync); |
257c62e1 | 1923 | |
90abccf2 MX |
1924 | /* |
1925 | * We write the dirty pages in the range and wait until they complete | |
1926 | * out of the ->i_mutex. If so, we can flush the dirty pages by | |
2ab28f32 JB |
1927 | * multi-task, and make the performance up. See |
1928 | * btrfs_wait_ordered_range for an explanation of the ASYNC check. | |
90abccf2 | 1929 | */ |
669249ee | 1930 | ret = start_ordered_ops(inode, start, end); |
90abccf2 MX |
1931 | if (ret) |
1932 | return ret; | |
1933 | ||
5955102c | 1934 | inode_lock(inode); |
2ecb7923 | 1935 | atomic_inc(&root->log_batch); |
2ab28f32 JB |
1936 | full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
1937 | &BTRFS_I(inode)->runtime_flags); | |
669249ee FM |
1938 | /* |
1939 | * We might have have had more pages made dirty after calling | |
1940 | * start_ordered_ops and before acquiring the inode's i_mutex. | |
1941 | */ | |
0ef8b726 | 1942 | if (full_sync) { |
669249ee FM |
1943 | /* |
1944 | * For a full sync, we need to make sure any ordered operations | |
1945 | * start and finish before we start logging the inode, so that | |
1946 | * all extents are persisted and the respective file extent | |
1947 | * items are in the fs/subvol btree. | |
1948 | */ | |
b659ef02 | 1949 | ret = btrfs_wait_ordered_range(inode, start, len); |
669249ee FM |
1950 | } else { |
1951 | /* | |
1952 | * Start any new ordered operations before starting to log the | |
1953 | * inode. We will wait for them to finish in btrfs_sync_log(). | |
1954 | * | |
1955 | * Right before acquiring the inode's mutex, we might have new | |
1956 | * writes dirtying pages, which won't immediately start the | |
1957 | * respective ordered operations - that is done through the | |
1958 | * fill_delalloc callbacks invoked from the writepage and | |
1959 | * writepages address space operations. So make sure we start | |
1960 | * all ordered operations before starting to log our inode. Not | |
1961 | * doing this means that while logging the inode, writeback | |
1962 | * could start and invoke writepage/writepages, which would call | |
1963 | * the fill_delalloc callbacks (cow_file_range, | |
1964 | * submit_compressed_extents). These callbacks add first an | |
1965 | * extent map to the modified list of extents and then create | |
1966 | * the respective ordered operation, which means in | |
1967 | * tree-log.c:btrfs_log_inode() we might capture all existing | |
1968 | * ordered operations (with btrfs_get_logged_extents()) before | |
1969 | * the fill_delalloc callback adds its ordered operation, and by | |
1970 | * the time we visit the modified list of extent maps (with | |
1971 | * btrfs_log_changed_extents()), we see and process the extent | |
1972 | * map they created. We then use the extent map to construct a | |
1973 | * file extent item for logging without waiting for the | |
1974 | * respective ordered operation to finish - this file extent | |
1975 | * item points to a disk location that might not have yet been | |
1976 | * written to, containing random data - so after a crash a log | |
1977 | * replay will make our inode have file extent items that point | |
1978 | * to disk locations containing invalid data, as we returned | |
1979 | * success to userspace without waiting for the respective | |
1980 | * ordered operation to finish, because it wasn't captured by | |
1981 | * btrfs_get_logged_extents(). | |
1982 | */ | |
1983 | ret = start_ordered_ops(inode, start, end); | |
1984 | } | |
1985 | if (ret) { | |
5955102c | 1986 | inode_unlock(inode); |
669249ee | 1987 | goto out; |
0ef8b726 | 1988 | } |
2ecb7923 | 1989 | atomic_inc(&root->log_batch); |
257c62e1 | 1990 | |
39279cc3 | 1991 | /* |
3a8b36f3 FM |
1992 | * If the last transaction that changed this file was before the current |
1993 | * transaction and we have the full sync flag set in our inode, we can | |
1994 | * bail out now without any syncing. | |
1995 | * | |
1996 | * Note that we can't bail out if the full sync flag isn't set. This is | |
1997 | * because when the full sync flag is set we start all ordered extents | |
1998 | * and wait for them to fully complete - when they complete they update | |
1999 | * the inode's last_trans field through: | |
2000 | * | |
2001 | * btrfs_finish_ordered_io() -> | |
2002 | * btrfs_update_inode_fallback() -> | |
2003 | * btrfs_update_inode() -> | |
2004 | * btrfs_set_inode_last_trans() | |
2005 | * | |
2006 | * So we are sure that last_trans is up to date and can do this check to | |
2007 | * bail out safely. For the fast path, when the full sync flag is not | |
2008 | * set in our inode, we can not do it because we start only our ordered | |
2009 | * extents and don't wait for them to complete (that is when | |
2010 | * btrfs_finish_ordered_io runs), so here at this point their last_trans | |
2011 | * value might be less than or equals to fs_info->last_trans_committed, | |
2012 | * and setting a speculative last_trans for an inode when a buffered | |
2013 | * write is made (such as fs_info->generation + 1 for example) would not | |
2014 | * be reliable since after setting the value and before fsync is called | |
2015 | * any number of transactions can start and commit (transaction kthread | |
2016 | * commits the current transaction periodically), and a transaction | |
2017 | * commit does not start nor waits for ordered extents to complete. | |
257c62e1 | 2018 | */ |
a4abeea4 | 2019 | smp_mb(); |
22ee6985 | 2020 | if (btrfs_inode_in_log(inode, root->fs_info->generation) || |
affc0ff9 FM |
2021 | (full_sync && BTRFS_I(inode)->last_trans <= |
2022 | root->fs_info->last_trans_committed) || | |
2023 | (!btrfs_have_ordered_extents_in_range(inode, start, len) && | |
2024 | BTRFS_I(inode)->last_trans | |
2025 | <= root->fs_info->last_trans_committed)) { | |
5dc562c5 JB |
2026 | /* |
2027 | * We'v had everything committed since the last time we were | |
2028 | * modified so clear this flag in case it was set for whatever | |
2029 | * reason, it's no longer relevant. | |
2030 | */ | |
2031 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
2032 | &BTRFS_I(inode)->runtime_flags); | |
5955102c | 2033 | inode_unlock(inode); |
15ee9bc7 JB |
2034 | goto out; |
2035 | } | |
15ee9bc7 JB |
2036 | |
2037 | /* | |
a52d9a80 CM |
2038 | * ok we haven't committed the transaction yet, lets do a commit |
2039 | */ | |
6f902af4 | 2040 | if (file->private_data) |
6bf13c0c SW |
2041 | btrfs_ioctl_trans_end(file); |
2042 | ||
5039eddc JB |
2043 | /* |
2044 | * We use start here because we will need to wait on the IO to complete | |
2045 | * in btrfs_sync_log, which could require joining a transaction (for | |
2046 | * example checking cross references in the nocow path). If we use join | |
2047 | * here we could get into a situation where we're waiting on IO to | |
2048 | * happen that is blocked on a transaction trying to commit. With start | |
2049 | * we inc the extwriter counter, so we wait for all extwriters to exit | |
2050 | * before we start blocking join'ers. This comment is to keep somebody | |
2051 | * from thinking they are super smart and changing this to | |
2052 | * btrfs_join_transaction *cough*Josef*cough*. | |
2053 | */ | |
a22285a6 YZ |
2054 | trans = btrfs_start_transaction(root, 0); |
2055 | if (IS_ERR(trans)) { | |
2056 | ret = PTR_ERR(trans); | |
5955102c | 2057 | inode_unlock(inode); |
39279cc3 CM |
2058 | goto out; |
2059 | } | |
5039eddc | 2060 | trans->sync = true; |
e02119d5 | 2061 | |
8b050d35 MX |
2062 | btrfs_init_log_ctx(&ctx); |
2063 | ||
49dae1bc | 2064 | ret = btrfs_log_dentry_safe(trans, root, dentry, start, end, &ctx); |
02c24a82 | 2065 | if (ret < 0) { |
a0634be5 FDBM |
2066 | /* Fallthrough and commit/free transaction. */ |
2067 | ret = 1; | |
02c24a82 | 2068 | } |
49eb7e46 CM |
2069 | |
2070 | /* we've logged all the items and now have a consistent | |
2071 | * version of the file in the log. It is possible that | |
2072 | * someone will come in and modify the file, but that's | |
2073 | * fine because the log is consistent on disk, and we | |
2074 | * have references to all of the file's extents | |
2075 | * | |
2076 | * It is possible that someone will come in and log the | |
2077 | * file again, but that will end up using the synchronization | |
2078 | * inside btrfs_sync_log to keep things safe. | |
2079 | */ | |
5955102c | 2080 | inode_unlock(inode); |
49eb7e46 | 2081 | |
8407f553 FM |
2082 | /* |
2083 | * If any of the ordered extents had an error, just return it to user | |
2084 | * space, so that the application knows some writes didn't succeed and | |
2085 | * can take proper action (retry for e.g.). Blindly committing the | |
2086 | * transaction in this case, would fool userspace that everything was | |
2087 | * successful. And we also want to make sure our log doesn't contain | |
2088 | * file extent items pointing to extents that weren't fully written to - | |
2089 | * just like in the non fast fsync path, where we check for the ordered | |
2090 | * operation's error flag before writing to the log tree and return -EIO | |
2091 | * if any of them had this flag set (btrfs_wait_ordered_range) - | |
2092 | * therefore we need to check for errors in the ordered operations, | |
2093 | * which are indicated by ctx.io_err. | |
2094 | */ | |
2095 | if (ctx.io_err) { | |
2096 | btrfs_end_transaction(trans, root); | |
2097 | ret = ctx.io_err; | |
2098 | goto out; | |
2099 | } | |
2100 | ||
257c62e1 | 2101 | if (ret != BTRFS_NO_LOG_SYNC) { |
0ef8b726 | 2102 | if (!ret) { |
8b050d35 | 2103 | ret = btrfs_sync_log(trans, root, &ctx); |
0ef8b726 | 2104 | if (!ret) { |
257c62e1 | 2105 | ret = btrfs_end_transaction(trans, root); |
0ef8b726 | 2106 | goto out; |
2ab28f32 | 2107 | } |
257c62e1 | 2108 | } |
0ef8b726 | 2109 | if (!full_sync) { |
9dcbeed4 | 2110 | ret = btrfs_wait_ordered_range(inode, start, len); |
b05fd874 FM |
2111 | if (ret) { |
2112 | btrfs_end_transaction(trans, root); | |
0ef8b726 | 2113 | goto out; |
b05fd874 | 2114 | } |
0ef8b726 JB |
2115 | } |
2116 | ret = btrfs_commit_transaction(trans, root); | |
257c62e1 CM |
2117 | } else { |
2118 | ret = btrfs_end_transaction(trans, root); | |
e02119d5 | 2119 | } |
39279cc3 | 2120 | out: |
014e4ac4 | 2121 | return ret > 0 ? -EIO : ret; |
39279cc3 CM |
2122 | } |
2123 | ||
f0f37e2f | 2124 | static const struct vm_operations_struct btrfs_file_vm_ops = { |
92fee66d | 2125 | .fault = filemap_fault, |
f1820361 | 2126 | .map_pages = filemap_map_pages, |
9ebefb18 CM |
2127 | .page_mkwrite = btrfs_page_mkwrite, |
2128 | }; | |
2129 | ||
2130 | static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma) | |
2131 | { | |
058a457e MX |
2132 | struct address_space *mapping = filp->f_mapping; |
2133 | ||
2134 | if (!mapping->a_ops->readpage) | |
2135 | return -ENOEXEC; | |
2136 | ||
9ebefb18 | 2137 | file_accessed(filp); |
058a457e | 2138 | vma->vm_ops = &btrfs_file_vm_ops; |
058a457e | 2139 | |
9ebefb18 CM |
2140 | return 0; |
2141 | } | |
2142 | ||
2aaa6655 JB |
2143 | static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf, |
2144 | int slot, u64 start, u64 end) | |
2145 | { | |
2146 | struct btrfs_file_extent_item *fi; | |
2147 | struct btrfs_key key; | |
2148 | ||
2149 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) | |
2150 | return 0; | |
2151 | ||
2152 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
2153 | if (key.objectid != btrfs_ino(inode) || | |
2154 | key.type != BTRFS_EXTENT_DATA_KEY) | |
2155 | return 0; | |
2156 | ||
2157 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
2158 | ||
2159 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2160 | return 0; | |
2161 | ||
2162 | if (btrfs_file_extent_disk_bytenr(leaf, fi)) | |
2163 | return 0; | |
2164 | ||
2165 | if (key.offset == end) | |
2166 | return 1; | |
2167 | if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start) | |
2168 | return 1; | |
2169 | return 0; | |
2170 | } | |
2171 | ||
2172 | static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode, | |
2173 | struct btrfs_path *path, u64 offset, u64 end) | |
2174 | { | |
2175 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2176 | struct extent_buffer *leaf; | |
2177 | struct btrfs_file_extent_item *fi; | |
2178 | struct extent_map *hole_em; | |
2179 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
2180 | struct btrfs_key key; | |
2181 | int ret; | |
2182 | ||
16e7549f JB |
2183 | if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) |
2184 | goto out; | |
2185 | ||
2aaa6655 JB |
2186 | key.objectid = btrfs_ino(inode); |
2187 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2188 | key.offset = offset; | |
2189 | ||
2aaa6655 JB |
2190 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
2191 | if (ret < 0) | |
2192 | return ret; | |
2193 | BUG_ON(!ret); | |
2194 | ||
2195 | leaf = path->nodes[0]; | |
2196 | if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) { | |
2197 | u64 num_bytes; | |
2198 | ||
2199 | path->slots[0]--; | |
2200 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2201 | struct btrfs_file_extent_item); | |
2202 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + | |
2203 | end - offset; | |
2204 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2205 | btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes); | |
2206 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
2207 | btrfs_mark_buffer_dirty(leaf); | |
2208 | goto out; | |
2209 | } | |
2210 | ||
1707e26d | 2211 | if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) { |
2aaa6655 JB |
2212 | u64 num_bytes; |
2213 | ||
2aaa6655 | 2214 | key.offset = offset; |
b7a0365e | 2215 | btrfs_set_item_key_safe(root->fs_info, path, &key); |
2aaa6655 JB |
2216 | fi = btrfs_item_ptr(leaf, path->slots[0], |
2217 | struct btrfs_file_extent_item); | |
2218 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end - | |
2219 | offset; | |
2220 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2221 | btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes); | |
2222 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
2223 | btrfs_mark_buffer_dirty(leaf); | |
2224 | goto out; | |
2225 | } | |
2226 | btrfs_release_path(path); | |
2227 | ||
2228 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset, | |
2229 | 0, 0, end - offset, 0, end - offset, | |
2230 | 0, 0, 0); | |
2231 | if (ret) | |
2232 | return ret; | |
2233 | ||
2234 | out: | |
2235 | btrfs_release_path(path); | |
2236 | ||
2237 | hole_em = alloc_extent_map(); | |
2238 | if (!hole_em) { | |
2239 | btrfs_drop_extent_cache(inode, offset, end - 1, 0); | |
2240 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
2241 | &BTRFS_I(inode)->runtime_flags); | |
2242 | } else { | |
2243 | hole_em->start = offset; | |
2244 | hole_em->len = end - offset; | |
cc95bef6 | 2245 | hole_em->ram_bytes = hole_em->len; |
2aaa6655 JB |
2246 | hole_em->orig_start = offset; |
2247 | ||
2248 | hole_em->block_start = EXTENT_MAP_HOLE; | |
2249 | hole_em->block_len = 0; | |
b4939680 | 2250 | hole_em->orig_block_len = 0; |
2aaa6655 JB |
2251 | hole_em->bdev = root->fs_info->fs_devices->latest_bdev; |
2252 | hole_em->compress_type = BTRFS_COMPRESS_NONE; | |
2253 | hole_em->generation = trans->transid; | |
2254 | ||
2255 | do { | |
2256 | btrfs_drop_extent_cache(inode, offset, end - 1, 0); | |
2257 | write_lock(&em_tree->lock); | |
09a2a8f9 | 2258 | ret = add_extent_mapping(em_tree, hole_em, 1); |
2aaa6655 JB |
2259 | write_unlock(&em_tree->lock); |
2260 | } while (ret == -EEXIST); | |
2261 | free_extent_map(hole_em); | |
2262 | if (ret) | |
2263 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
2264 | &BTRFS_I(inode)->runtime_flags); | |
2265 | } | |
2266 | ||
2267 | return 0; | |
2268 | } | |
2269 | ||
d7781546 QW |
2270 | /* |
2271 | * Find a hole extent on given inode and change start/len to the end of hole | |
2272 | * extent.(hole/vacuum extent whose em->start <= start && | |
2273 | * em->start + em->len > start) | |
2274 | * When a hole extent is found, return 1 and modify start/len. | |
2275 | */ | |
2276 | static int find_first_non_hole(struct inode *inode, u64 *start, u64 *len) | |
2277 | { | |
2278 | struct extent_map *em; | |
2279 | int ret = 0; | |
2280 | ||
2281 | em = btrfs_get_extent(inode, NULL, 0, *start, *len, 0); | |
2282 | if (IS_ERR_OR_NULL(em)) { | |
2283 | if (!em) | |
2284 | ret = -ENOMEM; | |
2285 | else | |
2286 | ret = PTR_ERR(em); | |
2287 | return ret; | |
2288 | } | |
2289 | ||
2290 | /* Hole or vacuum extent(only exists in no-hole mode) */ | |
2291 | if (em->block_start == EXTENT_MAP_HOLE) { | |
2292 | ret = 1; | |
2293 | *len = em->start + em->len > *start + *len ? | |
2294 | 0 : *start + *len - em->start - em->len; | |
2295 | *start = em->start + em->len; | |
2296 | } | |
2297 | free_extent_map(em); | |
2298 | return ret; | |
2299 | } | |
2300 | ||
2aaa6655 JB |
2301 | static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len) |
2302 | { | |
2303 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2304 | struct extent_state *cached_state = NULL; | |
2305 | struct btrfs_path *path; | |
2306 | struct btrfs_block_rsv *rsv; | |
2307 | struct btrfs_trans_handle *trans; | |
d7781546 QW |
2308 | u64 lockstart; |
2309 | u64 lockend; | |
2310 | u64 tail_start; | |
2311 | u64 tail_len; | |
2312 | u64 orig_start = offset; | |
2313 | u64 cur_offset; | |
2aaa6655 JB |
2314 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
2315 | u64 drop_end; | |
2aaa6655 JB |
2316 | int ret = 0; |
2317 | int err = 0; | |
6e4d6fa1 | 2318 | unsigned int rsv_count; |
9703fefe | 2319 | bool same_block; |
16e7549f | 2320 | bool no_holes = btrfs_fs_incompat(root->fs_info, NO_HOLES); |
a1a50f60 | 2321 | u64 ino_size; |
9703fefe | 2322 | bool truncated_block = false; |
e8c1c76e | 2323 | bool updated_inode = false; |
2aaa6655 | 2324 | |
0ef8b726 JB |
2325 | ret = btrfs_wait_ordered_range(inode, offset, len); |
2326 | if (ret) | |
2327 | return ret; | |
2aaa6655 | 2328 | |
5955102c | 2329 | inode_lock(inode); |
9703fefe | 2330 | ino_size = round_up(inode->i_size, root->sectorsize); |
d7781546 QW |
2331 | ret = find_first_non_hole(inode, &offset, &len); |
2332 | if (ret < 0) | |
2333 | goto out_only_mutex; | |
2334 | if (ret && !len) { | |
2335 | /* Already in a large hole */ | |
2336 | ret = 0; | |
2337 | goto out_only_mutex; | |
2338 | } | |
2339 | ||
51f395ad | 2340 | lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize); |
d7781546 QW |
2341 | lockend = round_down(offset + len, |
2342 | BTRFS_I(inode)->root->sectorsize) - 1; | |
9703fefe CR |
2343 | same_block = (BTRFS_BYTES_TO_BLKS(root->fs_info, offset)) |
2344 | == (BTRFS_BYTES_TO_BLKS(root->fs_info, offset + len - 1)); | |
7426cc04 | 2345 | /* |
9703fefe | 2346 | * We needn't truncate any block which is beyond the end of the file |
7426cc04 MX |
2347 | * because we are sure there is no data there. |
2348 | */ | |
2aaa6655 | 2349 | /* |
9703fefe CR |
2350 | * Only do this if we are in the same block and we aren't doing the |
2351 | * entire block. | |
2aaa6655 | 2352 | */ |
9703fefe | 2353 | if (same_block && len < root->sectorsize) { |
e8c1c76e | 2354 | if (offset < ino_size) { |
9703fefe CR |
2355 | truncated_block = true; |
2356 | ret = btrfs_truncate_block(inode, offset, len, 0); | |
e8c1c76e FM |
2357 | } else { |
2358 | ret = 0; | |
2359 | } | |
d7781546 | 2360 | goto out_only_mutex; |
2aaa6655 JB |
2361 | } |
2362 | ||
9703fefe | 2363 | /* zero back part of the first block */ |
12870f1c | 2364 | if (offset < ino_size) { |
9703fefe CR |
2365 | truncated_block = true; |
2366 | ret = btrfs_truncate_block(inode, offset, 0, 0); | |
7426cc04 | 2367 | if (ret) { |
5955102c | 2368 | inode_unlock(inode); |
7426cc04 MX |
2369 | return ret; |
2370 | } | |
2aaa6655 JB |
2371 | } |
2372 | ||
d7781546 QW |
2373 | /* Check the aligned pages after the first unaligned page, |
2374 | * if offset != orig_start, which means the first unaligned page | |
2375 | * including serveral following pages are already in holes, | |
2376 | * the extra check can be skipped */ | |
2377 | if (offset == orig_start) { | |
2378 | /* after truncate page, check hole again */ | |
2379 | len = offset + len - lockstart; | |
2380 | offset = lockstart; | |
2381 | ret = find_first_non_hole(inode, &offset, &len); | |
2382 | if (ret < 0) | |
2383 | goto out_only_mutex; | |
2384 | if (ret && !len) { | |
2385 | ret = 0; | |
2386 | goto out_only_mutex; | |
2387 | } | |
2388 | lockstart = offset; | |
2389 | } | |
2390 | ||
2391 | /* Check the tail unaligned part is in a hole */ | |
2392 | tail_start = lockend + 1; | |
2393 | tail_len = offset + len - tail_start; | |
2394 | if (tail_len) { | |
2395 | ret = find_first_non_hole(inode, &tail_start, &tail_len); | |
2396 | if (unlikely(ret < 0)) | |
2397 | goto out_only_mutex; | |
2398 | if (!ret) { | |
2399 | /* zero the front end of the last page */ | |
2400 | if (tail_start + tail_len < ino_size) { | |
9703fefe CR |
2401 | truncated_block = true; |
2402 | ret = btrfs_truncate_block(inode, | |
2403 | tail_start + tail_len, | |
2404 | 0, 1); | |
d7781546 QW |
2405 | if (ret) |
2406 | goto out_only_mutex; | |
51f395ad | 2407 | } |
0061280d | 2408 | } |
2aaa6655 JB |
2409 | } |
2410 | ||
2411 | if (lockend < lockstart) { | |
e8c1c76e FM |
2412 | ret = 0; |
2413 | goto out_only_mutex; | |
2aaa6655 JB |
2414 | } |
2415 | ||
2416 | while (1) { | |
2417 | struct btrfs_ordered_extent *ordered; | |
2418 | ||
2419 | truncate_pagecache_range(inode, lockstart, lockend); | |
2420 | ||
2421 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
ff13db41 | 2422 | &cached_state); |
2aaa6655 JB |
2423 | ordered = btrfs_lookup_first_ordered_extent(inode, lockend); |
2424 | ||
2425 | /* | |
2426 | * We need to make sure we have no ordered extents in this range | |
2427 | * and nobody raced in and read a page in this range, if we did | |
2428 | * we need to try again. | |
2429 | */ | |
2430 | if ((!ordered || | |
6126e3ca | 2431 | (ordered->file_offset + ordered->len <= lockstart || |
2aaa6655 | 2432 | ordered->file_offset > lockend)) && |
fc4adbff | 2433 | !btrfs_page_exists_in_range(inode, lockstart, lockend)) { |
2aaa6655 JB |
2434 | if (ordered) |
2435 | btrfs_put_ordered_extent(ordered); | |
2436 | break; | |
2437 | } | |
2438 | if (ordered) | |
2439 | btrfs_put_ordered_extent(ordered); | |
2440 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, | |
2441 | lockend, &cached_state, GFP_NOFS); | |
0ef8b726 JB |
2442 | ret = btrfs_wait_ordered_range(inode, lockstart, |
2443 | lockend - lockstart + 1); | |
2444 | if (ret) { | |
5955102c | 2445 | inode_unlock(inode); |
0ef8b726 JB |
2446 | return ret; |
2447 | } | |
2aaa6655 JB |
2448 | } |
2449 | ||
2450 | path = btrfs_alloc_path(); | |
2451 | if (!path) { | |
2452 | ret = -ENOMEM; | |
2453 | goto out; | |
2454 | } | |
2455 | ||
66d8f3dd | 2456 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
2aaa6655 JB |
2457 | if (!rsv) { |
2458 | ret = -ENOMEM; | |
2459 | goto out_free; | |
2460 | } | |
2461 | rsv->size = btrfs_calc_trunc_metadata_size(root, 1); | |
2462 | rsv->failfast = 1; | |
2463 | ||
2464 | /* | |
2465 | * 1 - update the inode | |
2466 | * 1 - removing the extents in the range | |
16e7549f | 2467 | * 1 - adding the hole extent if no_holes isn't set |
2aaa6655 | 2468 | */ |
16e7549f JB |
2469 | rsv_count = no_holes ? 2 : 3; |
2470 | trans = btrfs_start_transaction(root, rsv_count); | |
2aaa6655 JB |
2471 | if (IS_ERR(trans)) { |
2472 | err = PTR_ERR(trans); | |
2473 | goto out_free; | |
2474 | } | |
2475 | ||
2476 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv, | |
2477 | min_size); | |
2478 | BUG_ON(ret); | |
2479 | trans->block_rsv = rsv; | |
2480 | ||
d7781546 QW |
2481 | cur_offset = lockstart; |
2482 | len = lockend - cur_offset; | |
2aaa6655 JB |
2483 | while (cur_offset < lockend) { |
2484 | ret = __btrfs_drop_extents(trans, root, inode, path, | |
2485 | cur_offset, lockend + 1, | |
1acae57b | 2486 | &drop_end, 1, 0, 0, NULL); |
2aaa6655 JB |
2487 | if (ret != -ENOSPC) |
2488 | break; | |
2489 | ||
2490 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
2491 | ||
12870f1c FM |
2492 | if (cur_offset < ino_size) { |
2493 | ret = fill_holes(trans, inode, path, cur_offset, | |
2494 | drop_end); | |
2495 | if (ret) { | |
2496 | err = ret; | |
2497 | break; | |
2498 | } | |
2aaa6655 JB |
2499 | } |
2500 | ||
2501 | cur_offset = drop_end; | |
2502 | ||
2503 | ret = btrfs_update_inode(trans, root, inode); | |
2504 | if (ret) { | |
2505 | err = ret; | |
2506 | break; | |
2507 | } | |
2508 | ||
2aaa6655 | 2509 | btrfs_end_transaction(trans, root); |
b53d3f5d | 2510 | btrfs_btree_balance_dirty(root); |
2aaa6655 | 2511 | |
16e7549f | 2512 | trans = btrfs_start_transaction(root, rsv_count); |
2aaa6655 JB |
2513 | if (IS_ERR(trans)) { |
2514 | ret = PTR_ERR(trans); | |
2515 | trans = NULL; | |
2516 | break; | |
2517 | } | |
2518 | ||
2519 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, | |
2520 | rsv, min_size); | |
2521 | BUG_ON(ret); /* shouldn't happen */ | |
2522 | trans->block_rsv = rsv; | |
d7781546 QW |
2523 | |
2524 | ret = find_first_non_hole(inode, &cur_offset, &len); | |
2525 | if (unlikely(ret < 0)) | |
2526 | break; | |
2527 | if (ret && !len) { | |
2528 | ret = 0; | |
2529 | break; | |
2530 | } | |
2aaa6655 JB |
2531 | } |
2532 | ||
2533 | if (ret) { | |
2534 | err = ret; | |
2535 | goto out_trans; | |
2536 | } | |
2537 | ||
2538 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
2959a32a FM |
2539 | /* |
2540 | * If we are using the NO_HOLES feature we might have had already an | |
2541 | * hole that overlaps a part of the region [lockstart, lockend] and | |
2542 | * ends at (or beyond) lockend. Since we have no file extent items to | |
2543 | * represent holes, drop_end can be less than lockend and so we must | |
2544 | * make sure we have an extent map representing the existing hole (the | |
2545 | * call to __btrfs_drop_extents() might have dropped the existing extent | |
2546 | * map representing the existing hole), otherwise the fast fsync path | |
2547 | * will not record the existence of the hole region | |
2548 | * [existing_hole_start, lockend]. | |
2549 | */ | |
2550 | if (drop_end <= lockend) | |
2551 | drop_end = lockend + 1; | |
fc19c5e7 FM |
2552 | /* |
2553 | * Don't insert file hole extent item if it's for a range beyond eof | |
2554 | * (because it's useless) or if it represents a 0 bytes range (when | |
2555 | * cur_offset == drop_end). | |
2556 | */ | |
2557 | if (cur_offset < ino_size && cur_offset < drop_end) { | |
12870f1c FM |
2558 | ret = fill_holes(trans, inode, path, cur_offset, drop_end); |
2559 | if (ret) { | |
2560 | err = ret; | |
2561 | goto out_trans; | |
2562 | } | |
2aaa6655 JB |
2563 | } |
2564 | ||
2565 | out_trans: | |
2566 | if (!trans) | |
2567 | goto out_free; | |
2568 | ||
e1f5790e | 2569 | inode_inc_iversion(inode); |
04b285f3 | 2570 | inode->i_mtime = inode->i_ctime = current_fs_time(inode->i_sb); |
e1f5790e | 2571 | |
2aaa6655 JB |
2572 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
2573 | ret = btrfs_update_inode(trans, root, inode); | |
e8c1c76e | 2574 | updated_inode = true; |
2aaa6655 | 2575 | btrfs_end_transaction(trans, root); |
b53d3f5d | 2576 | btrfs_btree_balance_dirty(root); |
2aaa6655 JB |
2577 | out_free: |
2578 | btrfs_free_path(path); | |
2579 | btrfs_free_block_rsv(root, rsv); | |
2580 | out: | |
2581 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
2582 | &cached_state, GFP_NOFS); | |
d7781546 | 2583 | out_only_mutex: |
9703fefe | 2584 | if (!updated_inode && truncated_block && !ret && !err) { |
e8c1c76e FM |
2585 | /* |
2586 | * If we only end up zeroing part of a page, we still need to | |
2587 | * update the inode item, so that all the time fields are | |
2588 | * updated as well as the necessary btrfs inode in memory fields | |
2589 | * for detecting, at fsync time, if the inode isn't yet in the | |
2590 | * log tree or it's there but not up to date. | |
2591 | */ | |
2592 | trans = btrfs_start_transaction(root, 1); | |
2593 | if (IS_ERR(trans)) { | |
2594 | err = PTR_ERR(trans); | |
2595 | } else { | |
2596 | err = btrfs_update_inode(trans, root, inode); | |
2597 | ret = btrfs_end_transaction(trans, root); | |
2598 | } | |
2599 | } | |
5955102c | 2600 | inode_unlock(inode); |
2aaa6655 JB |
2601 | if (ret && !err) |
2602 | err = ret; | |
2603 | return err; | |
2604 | } | |
2605 | ||
14524a84 QW |
2606 | /* Helper structure to record which range is already reserved */ |
2607 | struct falloc_range { | |
2608 | struct list_head list; | |
2609 | u64 start; | |
2610 | u64 len; | |
2611 | }; | |
2612 | ||
2613 | /* | |
2614 | * Helper function to add falloc range | |
2615 | * | |
2616 | * Caller should have locked the larger range of extent containing | |
2617 | * [start, len) | |
2618 | */ | |
2619 | static int add_falloc_range(struct list_head *head, u64 start, u64 len) | |
2620 | { | |
2621 | struct falloc_range *prev = NULL; | |
2622 | struct falloc_range *range = NULL; | |
2623 | ||
2624 | if (list_empty(head)) | |
2625 | goto insert; | |
2626 | ||
2627 | /* | |
2628 | * As fallocate iterate by bytenr order, we only need to check | |
2629 | * the last range. | |
2630 | */ | |
2631 | prev = list_entry(head->prev, struct falloc_range, list); | |
2632 | if (prev->start + prev->len == start) { | |
2633 | prev->len += len; | |
2634 | return 0; | |
2635 | } | |
2636 | insert: | |
32fc932e | 2637 | range = kmalloc(sizeof(*range), GFP_KERNEL); |
14524a84 QW |
2638 | if (!range) |
2639 | return -ENOMEM; | |
2640 | range->start = start; | |
2641 | range->len = len; | |
2642 | list_add_tail(&range->list, head); | |
2643 | return 0; | |
2644 | } | |
2645 | ||
2fe17c10 CH |
2646 | static long btrfs_fallocate(struct file *file, int mode, |
2647 | loff_t offset, loff_t len) | |
2648 | { | |
496ad9aa | 2649 | struct inode *inode = file_inode(file); |
2fe17c10 | 2650 | struct extent_state *cached_state = NULL; |
14524a84 QW |
2651 | struct falloc_range *range; |
2652 | struct falloc_range *tmp; | |
2653 | struct list_head reserve_list; | |
2fe17c10 CH |
2654 | u64 cur_offset; |
2655 | u64 last_byte; | |
2656 | u64 alloc_start; | |
2657 | u64 alloc_end; | |
2658 | u64 alloc_hint = 0; | |
2659 | u64 locked_end; | |
14524a84 | 2660 | u64 actual_end = 0; |
2fe17c10 | 2661 | struct extent_map *em; |
797f4277 | 2662 | int blocksize = BTRFS_I(inode)->root->sectorsize; |
2fe17c10 CH |
2663 | int ret; |
2664 | ||
797f4277 MX |
2665 | alloc_start = round_down(offset, blocksize); |
2666 | alloc_end = round_up(offset + len, blocksize); | |
2fe17c10 | 2667 | |
2aaa6655 JB |
2668 | /* Make sure we aren't being give some crap mode */ |
2669 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
2fe17c10 CH |
2670 | return -EOPNOTSUPP; |
2671 | ||
2aaa6655 JB |
2672 | if (mode & FALLOC_FL_PUNCH_HOLE) |
2673 | return btrfs_punch_hole(inode, offset, len); | |
2674 | ||
d98456fc | 2675 | /* |
14524a84 QW |
2676 | * Only trigger disk allocation, don't trigger qgroup reserve |
2677 | * | |
2678 | * For qgroup space, it will be checked later. | |
d98456fc | 2679 | */ |
14524a84 QW |
2680 | ret = btrfs_alloc_data_chunk_ondemand(inode, alloc_end - alloc_start); |
2681 | if (ret < 0) | |
d98456fc CM |
2682 | return ret; |
2683 | ||
5955102c | 2684 | inode_lock(inode); |
2fe17c10 CH |
2685 | ret = inode_newsize_ok(inode, alloc_end); |
2686 | if (ret) | |
2687 | goto out; | |
2688 | ||
14524a84 QW |
2689 | /* |
2690 | * TODO: Move these two operations after we have checked | |
2691 | * accurate reserved space, or fallocate can still fail but | |
2692 | * with page truncated or size expanded. | |
2693 | * | |
2694 | * But that's a minor problem and won't do much harm BTW. | |
2695 | */ | |
2fe17c10 | 2696 | if (alloc_start > inode->i_size) { |
a41ad394 JB |
2697 | ret = btrfs_cont_expand(inode, i_size_read(inode), |
2698 | alloc_start); | |
2fe17c10 CH |
2699 | if (ret) |
2700 | goto out; | |
0f6925fa | 2701 | } else if (offset + len > inode->i_size) { |
a71754fc JB |
2702 | /* |
2703 | * If we are fallocating from the end of the file onward we | |
9703fefe CR |
2704 | * need to zero out the end of the block if i_size lands in the |
2705 | * middle of a block. | |
a71754fc | 2706 | */ |
9703fefe | 2707 | ret = btrfs_truncate_block(inode, inode->i_size, 0, 0); |
a71754fc JB |
2708 | if (ret) |
2709 | goto out; | |
2fe17c10 CH |
2710 | } |
2711 | ||
a71754fc JB |
2712 | /* |
2713 | * wait for ordered IO before we have any locks. We'll loop again | |
2714 | * below with the locks held. | |
2715 | */ | |
0ef8b726 JB |
2716 | ret = btrfs_wait_ordered_range(inode, alloc_start, |
2717 | alloc_end - alloc_start); | |
2718 | if (ret) | |
2719 | goto out; | |
a71754fc | 2720 | |
2fe17c10 CH |
2721 | locked_end = alloc_end - 1; |
2722 | while (1) { | |
2723 | struct btrfs_ordered_extent *ordered; | |
2724 | ||
2725 | /* the extent lock is ordered inside the running | |
2726 | * transaction | |
2727 | */ | |
2728 | lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start, | |
ff13db41 | 2729 | locked_end, &cached_state); |
2fe17c10 CH |
2730 | ordered = btrfs_lookup_first_ordered_extent(inode, |
2731 | alloc_end - 1); | |
2732 | if (ordered && | |
2733 | ordered->file_offset + ordered->len > alloc_start && | |
2734 | ordered->file_offset < alloc_end) { | |
2735 | btrfs_put_ordered_extent(ordered); | |
2736 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, | |
2737 | alloc_start, locked_end, | |
32fc932e | 2738 | &cached_state, GFP_KERNEL); |
2fe17c10 CH |
2739 | /* |
2740 | * we can't wait on the range with the transaction | |
2741 | * running or with the extent lock held | |
2742 | */ | |
0ef8b726 JB |
2743 | ret = btrfs_wait_ordered_range(inode, alloc_start, |
2744 | alloc_end - alloc_start); | |
2745 | if (ret) | |
2746 | goto out; | |
2fe17c10 CH |
2747 | } else { |
2748 | if (ordered) | |
2749 | btrfs_put_ordered_extent(ordered); | |
2750 | break; | |
2751 | } | |
2752 | } | |
2753 | ||
14524a84 QW |
2754 | /* First, check if we exceed the qgroup limit */ |
2755 | INIT_LIST_HEAD(&reserve_list); | |
2fe17c10 CH |
2756 | cur_offset = alloc_start; |
2757 | while (1) { | |
2758 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
2759 | alloc_end - cur_offset, 0); | |
79787eaa JM |
2760 | if (IS_ERR_OR_NULL(em)) { |
2761 | if (!em) | |
2762 | ret = -ENOMEM; | |
2763 | else | |
2764 | ret = PTR_ERR(em); | |
2765 | break; | |
2766 | } | |
2fe17c10 | 2767 | last_byte = min(extent_map_end(em), alloc_end); |
f1e490a7 | 2768 | actual_end = min_t(u64, extent_map_end(em), offset + len); |
797f4277 | 2769 | last_byte = ALIGN(last_byte, blocksize); |
2fe17c10 CH |
2770 | if (em->block_start == EXTENT_MAP_HOLE || |
2771 | (cur_offset >= inode->i_size && | |
2772 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
14524a84 QW |
2773 | ret = add_falloc_range(&reserve_list, cur_offset, |
2774 | last_byte - cur_offset); | |
2775 | if (ret < 0) { | |
2776 | free_extent_map(em); | |
2777 | break; | |
3d850dd4 | 2778 | } |
14524a84 QW |
2779 | ret = btrfs_qgroup_reserve_data(inode, cur_offset, |
2780 | last_byte - cur_offset); | |
2781 | if (ret < 0) | |
2782 | break; | |
2fe17c10 CH |
2783 | } |
2784 | free_extent_map(em); | |
2fe17c10 | 2785 | cur_offset = last_byte; |
14524a84 | 2786 | if (cur_offset >= alloc_end) |
2fe17c10 | 2787 | break; |
14524a84 QW |
2788 | } |
2789 | ||
2790 | /* | |
2791 | * If ret is still 0, means we're OK to fallocate. | |
2792 | * Or just cleanup the list and exit. | |
2793 | */ | |
2794 | list_for_each_entry_safe(range, tmp, &reserve_list, list) { | |
2795 | if (!ret) | |
2796 | ret = btrfs_prealloc_file_range(inode, mode, | |
2797 | range->start, | |
2798 | range->len, 1 << inode->i_blkbits, | |
2799 | offset + len, &alloc_hint); | |
2800 | list_del(&range->list); | |
2801 | kfree(range); | |
2802 | } | |
2803 | if (ret < 0) | |
2804 | goto out_unlock; | |
2805 | ||
2806 | if (actual_end > inode->i_size && | |
2807 | !(mode & FALLOC_FL_KEEP_SIZE)) { | |
2808 | struct btrfs_trans_handle *trans; | |
2809 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2810 | ||
2811 | /* | |
2812 | * We didn't need to allocate any more space, but we | |
2813 | * still extended the size of the file so we need to | |
2814 | * update i_size and the inode item. | |
2815 | */ | |
2816 | trans = btrfs_start_transaction(root, 1); | |
2817 | if (IS_ERR(trans)) { | |
2818 | ret = PTR_ERR(trans); | |
2819 | } else { | |
04b285f3 | 2820 | inode->i_ctime = current_fs_time(inode->i_sb); |
14524a84 QW |
2821 | i_size_write(inode, actual_end); |
2822 | btrfs_ordered_update_i_size(inode, actual_end, NULL); | |
2823 | ret = btrfs_update_inode(trans, root, inode); | |
2824 | if (ret) | |
2825 | btrfs_end_transaction(trans, root); | |
2826 | else | |
2827 | ret = btrfs_end_transaction(trans, root); | |
2fe17c10 CH |
2828 | } |
2829 | } | |
14524a84 | 2830 | out_unlock: |
2fe17c10 | 2831 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
32fc932e | 2832 | &cached_state, GFP_KERNEL); |
2fe17c10 | 2833 | out: |
14524a84 QW |
2834 | /* |
2835 | * As we waited the extent range, the data_rsv_map must be empty | |
2836 | * in the range, as written data range will be released from it. | |
2837 | * And for prealloacted extent, it will also be released when | |
2838 | * its metadata is written. | |
2839 | * So this is completely used as cleanup. | |
2840 | */ | |
2841 | btrfs_qgroup_free_data(inode, alloc_start, alloc_end - alloc_start); | |
5955102c | 2842 | inode_unlock(inode); |
d98456fc | 2843 | /* Let go of our reservation. */ |
7cf5b976 QW |
2844 | btrfs_free_reserved_data_space(inode, alloc_start, |
2845 | alloc_end - alloc_start); | |
2fe17c10 CH |
2846 | return ret; |
2847 | } | |
2848 | ||
965c8e59 | 2849 | static int find_desired_extent(struct inode *inode, loff_t *offset, int whence) |
b2675157 JB |
2850 | { |
2851 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7f4ca37c | 2852 | struct extent_map *em = NULL; |
b2675157 | 2853 | struct extent_state *cached_state = NULL; |
4d1a40c6 LB |
2854 | u64 lockstart; |
2855 | u64 lockend; | |
2856 | u64 start; | |
2857 | u64 len; | |
b2675157 JB |
2858 | int ret = 0; |
2859 | ||
4d1a40c6 LB |
2860 | if (inode->i_size == 0) |
2861 | return -ENXIO; | |
2862 | ||
2863 | /* | |
2864 | * *offset can be negative, in this case we start finding DATA/HOLE from | |
2865 | * the very start of the file. | |
2866 | */ | |
2867 | start = max_t(loff_t, 0, *offset); | |
2868 | ||
2869 | lockstart = round_down(start, root->sectorsize); | |
2870 | lockend = round_up(i_size_read(inode), root->sectorsize); | |
b2675157 JB |
2871 | if (lockend <= lockstart) |
2872 | lockend = lockstart + root->sectorsize; | |
1214b53f | 2873 | lockend--; |
b2675157 JB |
2874 | len = lockend - lockstart + 1; |
2875 | ||
ff13db41 | 2876 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
d0082371 | 2877 | &cached_state); |
b2675157 | 2878 | |
7f4ca37c | 2879 | while (start < inode->i_size) { |
b2675157 JB |
2880 | em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0); |
2881 | if (IS_ERR(em)) { | |
6af021d8 | 2882 | ret = PTR_ERR(em); |
7f4ca37c | 2883 | em = NULL; |
b2675157 JB |
2884 | break; |
2885 | } | |
2886 | ||
7f4ca37c JB |
2887 | if (whence == SEEK_HOLE && |
2888 | (em->block_start == EXTENT_MAP_HOLE || | |
2889 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) | |
2890 | break; | |
2891 | else if (whence == SEEK_DATA && | |
2892 | (em->block_start != EXTENT_MAP_HOLE && | |
2893 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) | |
2894 | break; | |
b2675157 JB |
2895 | |
2896 | start = em->start + em->len; | |
b2675157 | 2897 | free_extent_map(em); |
7f4ca37c | 2898 | em = NULL; |
b2675157 JB |
2899 | cond_resched(); |
2900 | } | |
7f4ca37c JB |
2901 | free_extent_map(em); |
2902 | if (!ret) { | |
2903 | if (whence == SEEK_DATA && start >= inode->i_size) | |
2904 | ret = -ENXIO; | |
2905 | else | |
2906 | *offset = min_t(loff_t, start, inode->i_size); | |
2907 | } | |
b2675157 JB |
2908 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
2909 | &cached_state, GFP_NOFS); | |
2910 | return ret; | |
2911 | } | |
2912 | ||
965c8e59 | 2913 | static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence) |
b2675157 JB |
2914 | { |
2915 | struct inode *inode = file->f_mapping->host; | |
2916 | int ret; | |
2917 | ||
5955102c | 2918 | inode_lock(inode); |
965c8e59 | 2919 | switch (whence) { |
b2675157 JB |
2920 | case SEEK_END: |
2921 | case SEEK_CUR: | |
965c8e59 | 2922 | offset = generic_file_llseek(file, offset, whence); |
b2675157 JB |
2923 | goto out; |
2924 | case SEEK_DATA: | |
2925 | case SEEK_HOLE: | |
48802c8a | 2926 | if (offset >= i_size_read(inode)) { |
5955102c | 2927 | inode_unlock(inode); |
48802c8a JL |
2928 | return -ENXIO; |
2929 | } | |
2930 | ||
965c8e59 | 2931 | ret = find_desired_extent(inode, &offset, whence); |
b2675157 | 2932 | if (ret) { |
5955102c | 2933 | inode_unlock(inode); |
b2675157 JB |
2934 | return ret; |
2935 | } | |
2936 | } | |
2937 | ||
46a1c2c7 | 2938 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
b2675157 | 2939 | out: |
5955102c | 2940 | inode_unlock(inode); |
b2675157 JB |
2941 | return offset; |
2942 | } | |
2943 | ||
828c0950 | 2944 | const struct file_operations btrfs_file_operations = { |
b2675157 | 2945 | .llseek = btrfs_file_llseek, |
aad4f8bb | 2946 | .read_iter = generic_file_read_iter, |
e9906a98 | 2947 | .splice_read = generic_file_splice_read, |
b30ac0fc | 2948 | .write_iter = btrfs_file_write_iter, |
9ebefb18 | 2949 | .mmap = btrfs_file_mmap, |
39279cc3 | 2950 | .open = generic_file_open, |
e1b81e67 | 2951 | .release = btrfs_release_file, |
39279cc3 | 2952 | .fsync = btrfs_sync_file, |
2fe17c10 | 2953 | .fallocate = btrfs_fallocate, |
34287aa3 | 2954 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 2955 | #ifdef CONFIG_COMPAT |
34287aa3 | 2956 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 2957 | #endif |
3db11b2e | 2958 | .copy_file_range = btrfs_copy_file_range, |
04b38d60 | 2959 | .clone_file_range = btrfs_clone_file_range, |
2b3909f8 | 2960 | .dedupe_file_range = btrfs_dedupe_file_range, |
39279cc3 | 2961 | }; |
9247f317 MX |
2962 | |
2963 | void btrfs_auto_defrag_exit(void) | |
2964 | { | |
5598e900 | 2965 | kmem_cache_destroy(btrfs_inode_defrag_cachep); |
9247f317 MX |
2966 | } |
2967 | ||
2968 | int btrfs_auto_defrag_init(void) | |
2969 | { | |
2970 | btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag", | |
2971 | sizeof(struct inode_defrag), 0, | |
2972 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, | |
2973 | NULL); | |
2974 | if (!btrfs_inode_defrag_cachep) | |
2975 | return -ENOMEM; | |
2976 | ||
2977 | return 0; | |
2978 | } | |
728404da FM |
2979 | |
2980 | int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end) | |
2981 | { | |
2982 | int ret; | |
2983 | ||
2984 | /* | |
2985 | * So with compression we will find and lock a dirty page and clear the | |
2986 | * first one as dirty, setup an async extent, and immediately return | |
2987 | * with the entire range locked but with nobody actually marked with | |
2988 | * writeback. So we can't just filemap_write_and_wait_range() and | |
2989 | * expect it to work since it will just kick off a thread to do the | |
2990 | * actual work. So we need to call filemap_fdatawrite_range _again_ | |
2991 | * since it will wait on the page lock, which won't be unlocked until | |
2992 | * after the pages have been marked as writeback and so we're good to go | |
2993 | * from there. We have to do this otherwise we'll miss the ordered | |
2994 | * extents and that results in badness. Please Josef, do not think you | |
2995 | * know better and pull this out at some point in the future, it is | |
2996 | * right and you are wrong. | |
2997 | */ | |
2998 | ret = filemap_fdatawrite_range(inode->i_mapping, start, end); | |
2999 | if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
3000 | &BTRFS_I(inode)->runtime_flags)) | |
3001 | ret = filemap_fdatawrite_range(inode->i_mapping, start, end); | |
3002 | ||
3003 | return ret; | |
3004 | } |