]>
Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
6cbd5570 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
6cbd5570 CM |
4 | */ |
5 | ||
39279cc3 CM |
6 | #include <linux/fs.h> |
7 | #include <linux/pagemap.h> | |
39279cc3 CM |
8 | #include <linux/time.h> |
9 | #include <linux/init.h> | |
10 | #include <linux/string.h> | |
39279cc3 | 11 | #include <linux/backing-dev.h> |
2fe17c10 | 12 | #include <linux/falloc.h> |
39279cc3 | 13 | #include <linux/writeback.h> |
39279cc3 | 14 | #include <linux/compat.h> |
5a0e3ad6 | 15 | #include <linux/slab.h> |
55e301fd | 16 | #include <linux/btrfs.h> |
e2e40f2c | 17 | #include <linux/uio.h> |
ae5e165d | 18 | #include <linux/iversion.h> |
39279cc3 CM |
19 | #include "ctree.h" |
20 | #include "disk-io.h" | |
21 | #include "transaction.h" | |
22 | #include "btrfs_inode.h" | |
39279cc3 | 23 | #include "print-tree.h" |
e02119d5 CM |
24 | #include "tree-log.h" |
25 | #include "locking.h" | |
2aaa6655 | 26 | #include "volumes.h" |
fcebe456 | 27 | #include "qgroup.h" |
ebb8765b | 28 | #include "compression.h" |
86736342 | 29 | #include "delalloc-space.h" |
39279cc3 | 30 | |
9247f317 | 31 | static struct kmem_cache *btrfs_inode_defrag_cachep; |
4cb5300b CM |
32 | /* |
33 | * when auto defrag is enabled we | |
34 | * queue up these defrag structs to remember which | |
35 | * inodes need defragging passes | |
36 | */ | |
37 | struct inode_defrag { | |
38 | struct rb_node rb_node; | |
39 | /* objectid */ | |
40 | u64 ino; | |
41 | /* | |
42 | * transid where the defrag was added, we search for | |
43 | * extents newer than this | |
44 | */ | |
45 | u64 transid; | |
46 | ||
47 | /* root objectid */ | |
48 | u64 root; | |
49 | ||
50 | /* last offset we were able to defrag */ | |
51 | u64 last_offset; | |
52 | ||
53 | /* if we've wrapped around back to zero once already */ | |
54 | int cycled; | |
55 | }; | |
56 | ||
762f2263 MX |
57 | static int __compare_inode_defrag(struct inode_defrag *defrag1, |
58 | struct inode_defrag *defrag2) | |
59 | { | |
60 | if (defrag1->root > defrag2->root) | |
61 | return 1; | |
62 | else if (defrag1->root < defrag2->root) | |
63 | return -1; | |
64 | else if (defrag1->ino > defrag2->ino) | |
65 | return 1; | |
66 | else if (defrag1->ino < defrag2->ino) | |
67 | return -1; | |
68 | else | |
69 | return 0; | |
70 | } | |
71 | ||
4cb5300b CM |
72 | /* pop a record for an inode into the defrag tree. The lock |
73 | * must be held already | |
74 | * | |
75 | * If you're inserting a record for an older transid than an | |
76 | * existing record, the transid already in the tree is lowered | |
77 | * | |
78 | * If an existing record is found the defrag item you | |
79 | * pass in is freed | |
80 | */ | |
6158e1ce | 81 | static int __btrfs_add_inode_defrag(struct btrfs_inode *inode, |
4cb5300b CM |
82 | struct inode_defrag *defrag) |
83 | { | |
3ffbd68c | 84 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
4cb5300b CM |
85 | struct inode_defrag *entry; |
86 | struct rb_node **p; | |
87 | struct rb_node *parent = NULL; | |
762f2263 | 88 | int ret; |
4cb5300b | 89 | |
0b246afa | 90 | p = &fs_info->defrag_inodes.rb_node; |
4cb5300b CM |
91 | while (*p) { |
92 | parent = *p; | |
93 | entry = rb_entry(parent, struct inode_defrag, rb_node); | |
94 | ||
762f2263 MX |
95 | ret = __compare_inode_defrag(defrag, entry); |
96 | if (ret < 0) | |
4cb5300b | 97 | p = &parent->rb_left; |
762f2263 | 98 | else if (ret > 0) |
4cb5300b CM |
99 | p = &parent->rb_right; |
100 | else { | |
101 | /* if we're reinserting an entry for | |
102 | * an old defrag run, make sure to | |
103 | * lower the transid of our existing record | |
104 | */ | |
105 | if (defrag->transid < entry->transid) | |
106 | entry->transid = defrag->transid; | |
107 | if (defrag->last_offset > entry->last_offset) | |
108 | entry->last_offset = defrag->last_offset; | |
8ddc4734 | 109 | return -EEXIST; |
4cb5300b CM |
110 | } |
111 | } | |
6158e1ce | 112 | set_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags); |
4cb5300b | 113 | rb_link_node(&defrag->rb_node, parent, p); |
0b246afa | 114 | rb_insert_color(&defrag->rb_node, &fs_info->defrag_inodes); |
8ddc4734 MX |
115 | return 0; |
116 | } | |
4cb5300b | 117 | |
2ff7e61e | 118 | static inline int __need_auto_defrag(struct btrfs_fs_info *fs_info) |
8ddc4734 | 119 | { |
0b246afa | 120 | if (!btrfs_test_opt(fs_info, AUTO_DEFRAG)) |
8ddc4734 MX |
121 | return 0; |
122 | ||
0b246afa | 123 | if (btrfs_fs_closing(fs_info)) |
8ddc4734 | 124 | return 0; |
4cb5300b | 125 | |
8ddc4734 | 126 | return 1; |
4cb5300b CM |
127 | } |
128 | ||
129 | /* | |
130 | * insert a defrag record for this inode if auto defrag is | |
131 | * enabled | |
132 | */ | |
133 | int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans, | |
6158e1ce | 134 | struct btrfs_inode *inode) |
4cb5300b | 135 | { |
6158e1ce | 136 | struct btrfs_root *root = inode->root; |
3ffbd68c | 137 | struct btrfs_fs_info *fs_info = root->fs_info; |
4cb5300b | 138 | struct inode_defrag *defrag; |
4cb5300b | 139 | u64 transid; |
8ddc4734 | 140 | int ret; |
4cb5300b | 141 | |
2ff7e61e | 142 | if (!__need_auto_defrag(fs_info)) |
4cb5300b CM |
143 | return 0; |
144 | ||
6158e1ce | 145 | if (test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags)) |
4cb5300b CM |
146 | return 0; |
147 | ||
148 | if (trans) | |
149 | transid = trans->transid; | |
150 | else | |
6158e1ce | 151 | transid = inode->root->last_trans; |
4cb5300b | 152 | |
9247f317 | 153 | defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS); |
4cb5300b CM |
154 | if (!defrag) |
155 | return -ENOMEM; | |
156 | ||
6158e1ce | 157 | defrag->ino = btrfs_ino(inode); |
4cb5300b CM |
158 | defrag->transid = transid; |
159 | defrag->root = root->root_key.objectid; | |
160 | ||
0b246afa | 161 | spin_lock(&fs_info->defrag_inodes_lock); |
6158e1ce | 162 | if (!test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags)) { |
8ddc4734 MX |
163 | /* |
164 | * If we set IN_DEFRAG flag and evict the inode from memory, | |
165 | * and then re-read this inode, this new inode doesn't have | |
166 | * IN_DEFRAG flag. At the case, we may find the existed defrag. | |
167 | */ | |
168 | ret = __btrfs_add_inode_defrag(inode, defrag); | |
169 | if (ret) | |
170 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
171 | } else { | |
9247f317 | 172 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); |
8ddc4734 | 173 | } |
0b246afa | 174 | spin_unlock(&fs_info->defrag_inodes_lock); |
a0f98dde | 175 | return 0; |
4cb5300b CM |
176 | } |
177 | ||
178 | /* | |
8ddc4734 MX |
179 | * Requeue the defrag object. If there is a defrag object that points to |
180 | * the same inode in the tree, we will merge them together (by | |
181 | * __btrfs_add_inode_defrag()) and free the one that we want to requeue. | |
4cb5300b | 182 | */ |
46e59791 | 183 | static void btrfs_requeue_inode_defrag(struct btrfs_inode *inode, |
48a3b636 | 184 | struct inode_defrag *defrag) |
8ddc4734 | 185 | { |
3ffbd68c | 186 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
8ddc4734 MX |
187 | int ret; |
188 | ||
2ff7e61e | 189 | if (!__need_auto_defrag(fs_info)) |
8ddc4734 MX |
190 | goto out; |
191 | ||
192 | /* | |
193 | * Here we don't check the IN_DEFRAG flag, because we need merge | |
194 | * them together. | |
195 | */ | |
0b246afa | 196 | spin_lock(&fs_info->defrag_inodes_lock); |
8ddc4734 | 197 | ret = __btrfs_add_inode_defrag(inode, defrag); |
0b246afa | 198 | spin_unlock(&fs_info->defrag_inodes_lock); |
8ddc4734 MX |
199 | if (ret) |
200 | goto out; | |
201 | return; | |
202 | out: | |
203 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
204 | } | |
205 | ||
4cb5300b | 206 | /* |
26176e7c MX |
207 | * pick the defragable inode that we want, if it doesn't exist, we will get |
208 | * the next one. | |
4cb5300b | 209 | */ |
26176e7c MX |
210 | static struct inode_defrag * |
211 | btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino) | |
4cb5300b CM |
212 | { |
213 | struct inode_defrag *entry = NULL; | |
762f2263 | 214 | struct inode_defrag tmp; |
4cb5300b CM |
215 | struct rb_node *p; |
216 | struct rb_node *parent = NULL; | |
762f2263 MX |
217 | int ret; |
218 | ||
219 | tmp.ino = ino; | |
220 | tmp.root = root; | |
4cb5300b | 221 | |
26176e7c MX |
222 | spin_lock(&fs_info->defrag_inodes_lock); |
223 | p = fs_info->defrag_inodes.rb_node; | |
4cb5300b CM |
224 | while (p) { |
225 | parent = p; | |
226 | entry = rb_entry(parent, struct inode_defrag, rb_node); | |
227 | ||
762f2263 MX |
228 | ret = __compare_inode_defrag(&tmp, entry); |
229 | if (ret < 0) | |
4cb5300b | 230 | p = parent->rb_left; |
762f2263 | 231 | else if (ret > 0) |
4cb5300b CM |
232 | p = parent->rb_right; |
233 | else | |
26176e7c | 234 | goto out; |
4cb5300b CM |
235 | } |
236 | ||
26176e7c MX |
237 | if (parent && __compare_inode_defrag(&tmp, entry) > 0) { |
238 | parent = rb_next(parent); | |
239 | if (parent) | |
4cb5300b | 240 | entry = rb_entry(parent, struct inode_defrag, rb_node); |
26176e7c MX |
241 | else |
242 | entry = NULL; | |
4cb5300b | 243 | } |
26176e7c MX |
244 | out: |
245 | if (entry) | |
246 | rb_erase(parent, &fs_info->defrag_inodes); | |
247 | spin_unlock(&fs_info->defrag_inodes_lock); | |
248 | return entry; | |
4cb5300b CM |
249 | } |
250 | ||
26176e7c | 251 | void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info) |
4cb5300b CM |
252 | { |
253 | struct inode_defrag *defrag; | |
26176e7c MX |
254 | struct rb_node *node; |
255 | ||
256 | spin_lock(&fs_info->defrag_inodes_lock); | |
257 | node = rb_first(&fs_info->defrag_inodes); | |
258 | while (node) { | |
259 | rb_erase(node, &fs_info->defrag_inodes); | |
260 | defrag = rb_entry(node, struct inode_defrag, rb_node); | |
261 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
262 | ||
351810c1 | 263 | cond_resched_lock(&fs_info->defrag_inodes_lock); |
26176e7c MX |
264 | |
265 | node = rb_first(&fs_info->defrag_inodes); | |
266 | } | |
267 | spin_unlock(&fs_info->defrag_inodes_lock); | |
268 | } | |
269 | ||
270 | #define BTRFS_DEFRAG_BATCH 1024 | |
271 | ||
272 | static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info, | |
273 | struct inode_defrag *defrag) | |
274 | { | |
4cb5300b CM |
275 | struct btrfs_root *inode_root; |
276 | struct inode *inode; | |
4cb5300b CM |
277 | struct btrfs_key key; |
278 | struct btrfs_ioctl_defrag_range_args range; | |
4cb5300b | 279 | int num_defrag; |
6f1c3605 LB |
280 | int index; |
281 | int ret; | |
4cb5300b | 282 | |
26176e7c MX |
283 | /* get the inode */ |
284 | key.objectid = defrag->root; | |
962a298f | 285 | key.type = BTRFS_ROOT_ITEM_KEY; |
26176e7c | 286 | key.offset = (u64)-1; |
6f1c3605 LB |
287 | |
288 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
289 | ||
26176e7c MX |
290 | inode_root = btrfs_read_fs_root_no_name(fs_info, &key); |
291 | if (IS_ERR(inode_root)) { | |
6f1c3605 LB |
292 | ret = PTR_ERR(inode_root); |
293 | goto cleanup; | |
294 | } | |
26176e7c MX |
295 | |
296 | key.objectid = defrag->ino; | |
962a298f | 297 | key.type = BTRFS_INODE_ITEM_KEY; |
26176e7c MX |
298 | key.offset = 0; |
299 | inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL); | |
300 | if (IS_ERR(inode)) { | |
6f1c3605 LB |
301 | ret = PTR_ERR(inode); |
302 | goto cleanup; | |
26176e7c | 303 | } |
6f1c3605 | 304 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
26176e7c MX |
305 | |
306 | /* do a chunk of defrag */ | |
307 | clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags); | |
4cb5300b CM |
308 | memset(&range, 0, sizeof(range)); |
309 | range.len = (u64)-1; | |
26176e7c | 310 | range.start = defrag->last_offset; |
b66f00da MX |
311 | |
312 | sb_start_write(fs_info->sb); | |
26176e7c MX |
313 | num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid, |
314 | BTRFS_DEFRAG_BATCH); | |
b66f00da | 315 | sb_end_write(fs_info->sb); |
26176e7c MX |
316 | /* |
317 | * if we filled the whole defrag batch, there | |
318 | * must be more work to do. Queue this defrag | |
319 | * again | |
320 | */ | |
321 | if (num_defrag == BTRFS_DEFRAG_BATCH) { | |
322 | defrag->last_offset = range.start; | |
46e59791 | 323 | btrfs_requeue_inode_defrag(BTRFS_I(inode), defrag); |
26176e7c MX |
324 | } else if (defrag->last_offset && !defrag->cycled) { |
325 | /* | |
326 | * we didn't fill our defrag batch, but | |
327 | * we didn't start at zero. Make sure we loop | |
328 | * around to the start of the file. | |
329 | */ | |
330 | defrag->last_offset = 0; | |
331 | defrag->cycled = 1; | |
46e59791 | 332 | btrfs_requeue_inode_defrag(BTRFS_I(inode), defrag); |
26176e7c MX |
333 | } else { |
334 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
335 | } | |
336 | ||
337 | iput(inode); | |
338 | return 0; | |
6f1c3605 LB |
339 | cleanup: |
340 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
341 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
342 | return ret; | |
26176e7c MX |
343 | } |
344 | ||
345 | /* | |
346 | * run through the list of inodes in the FS that need | |
347 | * defragging | |
348 | */ | |
349 | int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info) | |
350 | { | |
351 | struct inode_defrag *defrag; | |
352 | u64 first_ino = 0; | |
353 | u64 root_objectid = 0; | |
4cb5300b CM |
354 | |
355 | atomic_inc(&fs_info->defrag_running); | |
67871254 | 356 | while (1) { |
dc81cdc5 MX |
357 | /* Pause the auto defragger. */ |
358 | if (test_bit(BTRFS_FS_STATE_REMOUNTING, | |
359 | &fs_info->fs_state)) | |
360 | break; | |
361 | ||
2ff7e61e | 362 | if (!__need_auto_defrag(fs_info)) |
26176e7c | 363 | break; |
4cb5300b CM |
364 | |
365 | /* find an inode to defrag */ | |
26176e7c MX |
366 | defrag = btrfs_pick_defrag_inode(fs_info, root_objectid, |
367 | first_ino); | |
4cb5300b | 368 | if (!defrag) { |
26176e7c | 369 | if (root_objectid || first_ino) { |
762f2263 | 370 | root_objectid = 0; |
4cb5300b CM |
371 | first_ino = 0; |
372 | continue; | |
373 | } else { | |
374 | break; | |
375 | } | |
376 | } | |
377 | ||
4cb5300b | 378 | first_ino = defrag->ino + 1; |
762f2263 | 379 | root_objectid = defrag->root; |
4cb5300b | 380 | |
26176e7c | 381 | __btrfs_run_defrag_inode(fs_info, defrag); |
4cb5300b | 382 | } |
4cb5300b CM |
383 | atomic_dec(&fs_info->defrag_running); |
384 | ||
385 | /* | |
386 | * during unmount, we use the transaction_wait queue to | |
387 | * wait for the defragger to stop | |
388 | */ | |
389 | wake_up(&fs_info->transaction_wait); | |
390 | return 0; | |
391 | } | |
39279cc3 | 392 | |
d352ac68 CM |
393 | /* simple helper to fault in pages and copy. This should go away |
394 | * and be replaced with calls into generic code. | |
395 | */ | |
ee22f0c4 | 396 | static noinline int btrfs_copy_from_user(loff_t pos, size_t write_bytes, |
a1b32a59 | 397 | struct page **prepared_pages, |
11c65dcc | 398 | struct iov_iter *i) |
39279cc3 | 399 | { |
914ee295 | 400 | size_t copied = 0; |
d0215f3e | 401 | size_t total_copied = 0; |
11c65dcc | 402 | int pg = 0; |
7073017a | 403 | int offset = offset_in_page(pos); |
39279cc3 | 404 | |
11c65dcc | 405 | while (write_bytes > 0) { |
39279cc3 | 406 | size_t count = min_t(size_t, |
09cbfeaf | 407 | PAGE_SIZE - offset, write_bytes); |
11c65dcc | 408 | struct page *page = prepared_pages[pg]; |
914ee295 XZ |
409 | /* |
410 | * Copy data from userspace to the current page | |
914ee295 | 411 | */ |
914ee295 | 412 | copied = iov_iter_copy_from_user_atomic(page, i, offset, count); |
11c65dcc | 413 | |
39279cc3 CM |
414 | /* Flush processor's dcache for this page */ |
415 | flush_dcache_page(page); | |
31339acd CM |
416 | |
417 | /* | |
418 | * if we get a partial write, we can end up with | |
419 | * partially up to date pages. These add | |
420 | * a lot of complexity, so make sure they don't | |
421 | * happen by forcing this copy to be retried. | |
422 | * | |
423 | * The rest of the btrfs_file_write code will fall | |
424 | * back to page at a time copies after we return 0. | |
425 | */ | |
426 | if (!PageUptodate(page) && copied < count) | |
427 | copied = 0; | |
428 | ||
11c65dcc JB |
429 | iov_iter_advance(i, copied); |
430 | write_bytes -= copied; | |
914ee295 | 431 | total_copied += copied; |
39279cc3 | 432 | |
b30ac0fc | 433 | /* Return to btrfs_file_write_iter to fault page */ |
9f570b8d | 434 | if (unlikely(copied == 0)) |
914ee295 | 435 | break; |
11c65dcc | 436 | |
09cbfeaf | 437 | if (copied < PAGE_SIZE - offset) { |
11c65dcc JB |
438 | offset += copied; |
439 | } else { | |
440 | pg++; | |
441 | offset = 0; | |
442 | } | |
39279cc3 | 443 | } |
914ee295 | 444 | return total_copied; |
39279cc3 CM |
445 | } |
446 | ||
d352ac68 CM |
447 | /* |
448 | * unlocks pages after btrfs_file_write is done with them | |
449 | */ | |
48a3b636 | 450 | static void btrfs_drop_pages(struct page **pages, size_t num_pages) |
39279cc3 CM |
451 | { |
452 | size_t i; | |
453 | for (i = 0; i < num_pages; i++) { | |
d352ac68 CM |
454 | /* page checked is some magic around finding pages that |
455 | * have been modified without going through btrfs_set_page_dirty | |
2457aec6 MG |
456 | * clear it here. There should be no need to mark the pages |
457 | * accessed as prepare_pages should have marked them accessed | |
458 | * in prepare_pages via find_or_create_page() | |
d352ac68 | 459 | */ |
4a096752 | 460 | ClearPageChecked(pages[i]); |
39279cc3 | 461 | unlock_page(pages[i]); |
09cbfeaf | 462 | put_page(pages[i]); |
39279cc3 CM |
463 | } |
464 | } | |
465 | ||
f48bf66b FM |
466 | static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode, |
467 | const u64 start, | |
468 | const u64 len, | |
469 | struct extent_state **cached_state) | |
470 | { | |
471 | u64 search_start = start; | |
472 | const u64 end = start + len - 1; | |
473 | ||
474 | while (search_start < end) { | |
475 | const u64 search_len = end - search_start + 1; | |
476 | struct extent_map *em; | |
477 | u64 em_len; | |
478 | int ret = 0; | |
479 | ||
480 | em = btrfs_get_extent(inode, NULL, 0, search_start, | |
481 | search_len, 0); | |
482 | if (IS_ERR(em)) | |
483 | return PTR_ERR(em); | |
484 | ||
485 | if (em->block_start != EXTENT_MAP_HOLE) | |
486 | goto next; | |
487 | ||
488 | em_len = em->len; | |
489 | if (em->start < search_start) | |
490 | em_len -= search_start - em->start; | |
491 | if (em_len > search_len) | |
492 | em_len = search_len; | |
493 | ||
494 | ret = set_extent_bit(&inode->io_tree, search_start, | |
495 | search_start + em_len - 1, | |
496 | EXTENT_DELALLOC_NEW, | |
497 | NULL, cached_state, GFP_NOFS); | |
498 | next: | |
499 | search_start = extent_map_end(em); | |
500 | free_extent_map(em); | |
501 | if (ret) | |
502 | return ret; | |
503 | } | |
504 | return 0; | |
505 | } | |
506 | ||
d352ac68 CM |
507 | /* |
508 | * after copy_from_user, pages need to be dirtied and we need to make | |
509 | * sure holes are created between the current EOF and the start of | |
510 | * any next extents (if required). | |
511 | * | |
512 | * this also makes the decision about creating an inline extent vs | |
513 | * doing real data extents, marking pages dirty and delalloc as required. | |
514 | */ | |
2ff7e61e JM |
515 | int btrfs_dirty_pages(struct inode *inode, struct page **pages, |
516 | size_t num_pages, loff_t pos, size_t write_bytes, | |
517 | struct extent_state **cached) | |
39279cc3 | 518 | { |
0b246afa | 519 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 520 | int err = 0; |
a52d9a80 | 521 | int i; |
db94535d | 522 | u64 num_bytes; |
a52d9a80 CM |
523 | u64 start_pos; |
524 | u64 end_of_last_block; | |
525 | u64 end_pos = pos + write_bytes; | |
526 | loff_t isize = i_size_read(inode); | |
e3b8a485 | 527 | unsigned int extra_bits = 0; |
39279cc3 | 528 | |
0b246afa | 529 | start_pos = pos & ~((u64) fs_info->sectorsize - 1); |
da17066c | 530 | num_bytes = round_up(write_bytes + pos - start_pos, |
0b246afa | 531 | fs_info->sectorsize); |
39279cc3 | 532 | |
db94535d | 533 | end_of_last_block = start_pos + num_bytes - 1; |
e3b8a485 | 534 | |
7703bdd8 CM |
535 | /* |
536 | * The pages may have already been dirty, clear out old accounting so | |
537 | * we can set things up properly | |
538 | */ | |
539 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos, end_of_last_block, | |
540 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
541 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0, cached); | |
542 | ||
e3b8a485 FM |
543 | if (!btrfs_is_free_space_inode(BTRFS_I(inode))) { |
544 | if (start_pos >= isize && | |
545 | !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)) { | |
546 | /* | |
547 | * There can't be any extents following eof in this case | |
548 | * so just set the delalloc new bit for the range | |
549 | * directly. | |
550 | */ | |
551 | extra_bits |= EXTENT_DELALLOC_NEW; | |
552 | } else { | |
553 | err = btrfs_find_new_delalloc_bytes(BTRFS_I(inode), | |
554 | start_pos, | |
555 | num_bytes, cached); | |
556 | if (err) | |
557 | return err; | |
558 | } | |
559 | } | |
560 | ||
2ac55d41 | 561 | err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block, |
e3b8a485 | 562 | extra_bits, cached, 0); |
d0215f3e JB |
563 | if (err) |
564 | return err; | |
9ed74f2d | 565 | |
c8b97818 CM |
566 | for (i = 0; i < num_pages; i++) { |
567 | struct page *p = pages[i]; | |
568 | SetPageUptodate(p); | |
569 | ClearPageChecked(p); | |
570 | set_page_dirty(p); | |
a52d9a80 | 571 | } |
9f570b8d JB |
572 | |
573 | /* | |
574 | * we've only changed i_size in ram, and we haven't updated | |
575 | * the disk i_size. There is no need to log the inode | |
576 | * at this time. | |
577 | */ | |
578 | if (end_pos > isize) | |
a52d9a80 | 579 | i_size_write(inode, end_pos); |
a22285a6 | 580 | return 0; |
39279cc3 CM |
581 | } |
582 | ||
d352ac68 CM |
583 | /* |
584 | * this drops all the extents in the cache that intersect the range | |
585 | * [start, end]. Existing extents are split as required. | |
586 | */ | |
dcdbc059 | 587 | void btrfs_drop_extent_cache(struct btrfs_inode *inode, u64 start, u64 end, |
7014cdb4 | 588 | int skip_pinned) |
a52d9a80 CM |
589 | { |
590 | struct extent_map *em; | |
3b951516 CM |
591 | struct extent_map *split = NULL; |
592 | struct extent_map *split2 = NULL; | |
dcdbc059 | 593 | struct extent_map_tree *em_tree = &inode->extent_tree; |
39b5637f | 594 | u64 len = end - start + 1; |
5dc562c5 | 595 | u64 gen; |
3b951516 CM |
596 | int ret; |
597 | int testend = 1; | |
5b21f2ed | 598 | unsigned long flags; |
c8b97818 | 599 | int compressed = 0; |
09a2a8f9 | 600 | bool modified; |
a52d9a80 | 601 | |
e6dcd2dc | 602 | WARN_ON(end < start); |
3b951516 | 603 | if (end == (u64)-1) { |
39b5637f | 604 | len = (u64)-1; |
3b951516 CM |
605 | testend = 0; |
606 | } | |
d397712b | 607 | while (1) { |
7014cdb4 JB |
608 | int no_splits = 0; |
609 | ||
09a2a8f9 | 610 | modified = false; |
3b951516 | 611 | if (!split) |
172ddd60 | 612 | split = alloc_extent_map(); |
3b951516 | 613 | if (!split2) |
172ddd60 | 614 | split2 = alloc_extent_map(); |
7014cdb4 JB |
615 | if (!split || !split2) |
616 | no_splits = 1; | |
3b951516 | 617 | |
890871be | 618 | write_lock(&em_tree->lock); |
39b5637f | 619 | em = lookup_extent_mapping(em_tree, start, len); |
d1310b2e | 620 | if (!em) { |
890871be | 621 | write_unlock(&em_tree->lock); |
a52d9a80 | 622 | break; |
d1310b2e | 623 | } |
5b21f2ed | 624 | flags = em->flags; |
5dc562c5 | 625 | gen = em->generation; |
5b21f2ed | 626 | if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) { |
55ef6899 | 627 | if (testend && em->start + em->len >= start + len) { |
5b21f2ed | 628 | free_extent_map(em); |
a1ed835e | 629 | write_unlock(&em_tree->lock); |
5b21f2ed ZY |
630 | break; |
631 | } | |
55ef6899 YZ |
632 | start = em->start + em->len; |
633 | if (testend) | |
5b21f2ed | 634 | len = start + len - (em->start + em->len); |
5b21f2ed | 635 | free_extent_map(em); |
a1ed835e | 636 | write_unlock(&em_tree->lock); |
5b21f2ed ZY |
637 | continue; |
638 | } | |
c8b97818 | 639 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
3ce7e67a | 640 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
3b277594 | 641 | clear_bit(EXTENT_FLAG_LOGGING, &flags); |
09a2a8f9 | 642 | modified = !list_empty(&em->list); |
7014cdb4 JB |
643 | if (no_splits) |
644 | goto next; | |
3b951516 | 645 | |
ee20a983 | 646 | if (em->start < start) { |
3b951516 CM |
647 | split->start = em->start; |
648 | split->len = start - em->start; | |
ee20a983 JB |
649 | |
650 | if (em->block_start < EXTENT_MAP_LAST_BYTE) { | |
651 | split->orig_start = em->orig_start; | |
652 | split->block_start = em->block_start; | |
653 | ||
654 | if (compressed) | |
655 | split->block_len = em->block_len; | |
656 | else | |
657 | split->block_len = split->len; | |
658 | split->orig_block_len = max(split->block_len, | |
659 | em->orig_block_len); | |
660 | split->ram_bytes = em->ram_bytes; | |
661 | } else { | |
662 | split->orig_start = split->start; | |
663 | split->block_len = 0; | |
664 | split->block_start = em->block_start; | |
665 | split->orig_block_len = 0; | |
666 | split->ram_bytes = split->len; | |
667 | } | |
668 | ||
5dc562c5 | 669 | split->generation = gen; |
3b951516 | 670 | split->bdev = em->bdev; |
5b21f2ed | 671 | split->flags = flags; |
261507a0 | 672 | split->compress_type = em->compress_type; |
176840b3 | 673 | replace_extent_mapping(em_tree, em, split, modified); |
3b951516 CM |
674 | free_extent_map(split); |
675 | split = split2; | |
676 | split2 = NULL; | |
677 | } | |
ee20a983 | 678 | if (testend && em->start + em->len > start + len) { |
3b951516 CM |
679 | u64 diff = start + len - em->start; |
680 | ||
681 | split->start = start + len; | |
682 | split->len = em->start + em->len - (start + len); | |
683 | split->bdev = em->bdev; | |
5b21f2ed | 684 | split->flags = flags; |
261507a0 | 685 | split->compress_type = em->compress_type; |
5dc562c5 | 686 | split->generation = gen; |
ee20a983 JB |
687 | |
688 | if (em->block_start < EXTENT_MAP_LAST_BYTE) { | |
689 | split->orig_block_len = max(em->block_len, | |
b4939680 | 690 | em->orig_block_len); |
3b951516 | 691 | |
ee20a983 JB |
692 | split->ram_bytes = em->ram_bytes; |
693 | if (compressed) { | |
694 | split->block_len = em->block_len; | |
695 | split->block_start = em->block_start; | |
696 | split->orig_start = em->orig_start; | |
697 | } else { | |
698 | split->block_len = split->len; | |
699 | split->block_start = em->block_start | |
700 | + diff; | |
701 | split->orig_start = em->orig_start; | |
702 | } | |
c8b97818 | 703 | } else { |
ee20a983 JB |
704 | split->ram_bytes = split->len; |
705 | split->orig_start = split->start; | |
706 | split->block_len = 0; | |
707 | split->block_start = em->block_start; | |
708 | split->orig_block_len = 0; | |
c8b97818 | 709 | } |
3b951516 | 710 | |
176840b3 FM |
711 | if (extent_map_in_tree(em)) { |
712 | replace_extent_mapping(em_tree, em, split, | |
713 | modified); | |
714 | } else { | |
715 | ret = add_extent_mapping(em_tree, split, | |
716 | modified); | |
717 | ASSERT(ret == 0); /* Logic error */ | |
718 | } | |
3b951516 CM |
719 | free_extent_map(split); |
720 | split = NULL; | |
721 | } | |
7014cdb4 | 722 | next: |
176840b3 FM |
723 | if (extent_map_in_tree(em)) |
724 | remove_extent_mapping(em_tree, em); | |
890871be | 725 | write_unlock(&em_tree->lock); |
d1310b2e | 726 | |
a52d9a80 CM |
727 | /* once for us */ |
728 | free_extent_map(em); | |
729 | /* once for the tree*/ | |
730 | free_extent_map(em); | |
731 | } | |
3b951516 CM |
732 | if (split) |
733 | free_extent_map(split); | |
734 | if (split2) | |
735 | free_extent_map(split2); | |
a52d9a80 CM |
736 | } |
737 | ||
39279cc3 CM |
738 | /* |
739 | * this is very complex, but the basic idea is to drop all extents | |
740 | * in the range start - end. hint_block is filled in with a block number | |
741 | * that would be a good hint to the block allocator for this file. | |
742 | * | |
743 | * If an extent intersects the range but is not entirely inside the range | |
744 | * it is either truncated or split. Anything entirely inside the range | |
745 | * is deleted from the tree. | |
746 | */ | |
5dc562c5 JB |
747 | int __btrfs_drop_extents(struct btrfs_trans_handle *trans, |
748 | struct btrfs_root *root, struct inode *inode, | |
749 | struct btrfs_path *path, u64 start, u64 end, | |
1acae57b FDBM |
750 | u64 *drop_end, int drop_cache, |
751 | int replace_extent, | |
752 | u32 extent_item_size, | |
753 | int *key_inserted) | |
39279cc3 | 754 | { |
0b246afa | 755 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 756 | struct extent_buffer *leaf; |
920bbbfb | 757 | struct btrfs_file_extent_item *fi; |
82fa113f | 758 | struct btrfs_ref ref = { 0 }; |
00f5c795 | 759 | struct btrfs_key key; |
920bbbfb | 760 | struct btrfs_key new_key; |
4a0cc7ca | 761 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
920bbbfb YZ |
762 | u64 search_start = start; |
763 | u64 disk_bytenr = 0; | |
764 | u64 num_bytes = 0; | |
765 | u64 extent_offset = 0; | |
766 | u64 extent_end = 0; | |
62fe51c1 | 767 | u64 last_end = start; |
920bbbfb YZ |
768 | int del_nr = 0; |
769 | int del_slot = 0; | |
770 | int extent_type; | |
ccd467d6 | 771 | int recow; |
00f5c795 | 772 | int ret; |
dc7fdde3 | 773 | int modify_tree = -1; |
27cdeb70 | 774 | int update_refs; |
c3308f84 | 775 | int found = 0; |
1acae57b | 776 | int leafs_visited = 0; |
39279cc3 | 777 | |
a1ed835e | 778 | if (drop_cache) |
dcdbc059 | 779 | btrfs_drop_extent_cache(BTRFS_I(inode), start, end - 1, 0); |
a52d9a80 | 780 | |
d5f37527 | 781 | if (start >= BTRFS_I(inode)->disk_i_size && !replace_extent) |
dc7fdde3 CM |
782 | modify_tree = 0; |
783 | ||
27cdeb70 | 784 | update_refs = (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa | 785 | root == fs_info->tree_root); |
d397712b | 786 | while (1) { |
ccd467d6 | 787 | recow = 0; |
33345d01 | 788 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
dc7fdde3 | 789 | search_start, modify_tree); |
39279cc3 | 790 | if (ret < 0) |
920bbbfb YZ |
791 | break; |
792 | if (ret > 0 && path->slots[0] > 0 && search_start == start) { | |
793 | leaf = path->nodes[0]; | |
794 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1); | |
33345d01 | 795 | if (key.objectid == ino && |
920bbbfb YZ |
796 | key.type == BTRFS_EXTENT_DATA_KEY) |
797 | path->slots[0]--; | |
39279cc3 | 798 | } |
920bbbfb | 799 | ret = 0; |
1acae57b | 800 | leafs_visited++; |
8c2383c3 | 801 | next_slot: |
5f39d397 | 802 | leaf = path->nodes[0]; |
920bbbfb YZ |
803 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { |
804 | BUG_ON(del_nr > 0); | |
805 | ret = btrfs_next_leaf(root, path); | |
806 | if (ret < 0) | |
807 | break; | |
808 | if (ret > 0) { | |
809 | ret = 0; | |
810 | break; | |
8c2383c3 | 811 | } |
1acae57b | 812 | leafs_visited++; |
920bbbfb YZ |
813 | leaf = path->nodes[0]; |
814 | recow = 1; | |
815 | } | |
816 | ||
817 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
aeafbf84 FM |
818 | |
819 | if (key.objectid > ino) | |
820 | break; | |
821 | if (WARN_ON_ONCE(key.objectid < ino) || | |
822 | key.type < BTRFS_EXTENT_DATA_KEY) { | |
823 | ASSERT(del_nr == 0); | |
824 | path->slots[0]++; | |
825 | goto next_slot; | |
826 | } | |
827 | if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end) | |
920bbbfb YZ |
828 | break; |
829 | ||
830 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
831 | struct btrfs_file_extent_item); | |
832 | extent_type = btrfs_file_extent_type(leaf, fi); | |
833 | ||
834 | if (extent_type == BTRFS_FILE_EXTENT_REG || | |
835 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
836 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
837 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
838 | extent_offset = btrfs_file_extent_offset(leaf, fi); | |
839 | extent_end = key.offset + | |
840 | btrfs_file_extent_num_bytes(leaf, fi); | |
841 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
842 | extent_end = key.offset + | |
e41ca589 | 843 | btrfs_file_extent_ram_bytes(leaf, fi); |
8c2383c3 | 844 | } else { |
aeafbf84 FM |
845 | /* can't happen */ |
846 | BUG(); | |
39279cc3 CM |
847 | } |
848 | ||
fc19c5e7 FM |
849 | /* |
850 | * Don't skip extent items representing 0 byte lengths. They | |
851 | * used to be created (bug) if while punching holes we hit | |
852 | * -ENOSPC condition. So if we find one here, just ensure we | |
853 | * delete it, otherwise we would insert a new file extent item | |
854 | * with the same key (offset) as that 0 bytes length file | |
855 | * extent item in the call to setup_items_for_insert() later | |
856 | * in this function. | |
857 | */ | |
62fe51c1 JB |
858 | if (extent_end == key.offset && extent_end >= search_start) { |
859 | last_end = extent_end; | |
fc19c5e7 | 860 | goto delete_extent_item; |
62fe51c1 | 861 | } |
fc19c5e7 | 862 | |
920bbbfb YZ |
863 | if (extent_end <= search_start) { |
864 | path->slots[0]++; | |
8c2383c3 | 865 | goto next_slot; |
39279cc3 CM |
866 | } |
867 | ||
c3308f84 | 868 | found = 1; |
920bbbfb | 869 | search_start = max(key.offset, start); |
dc7fdde3 CM |
870 | if (recow || !modify_tree) { |
871 | modify_tree = -1; | |
b3b4aa74 | 872 | btrfs_release_path(path); |
920bbbfb | 873 | continue; |
39279cc3 | 874 | } |
6643558d | 875 | |
920bbbfb YZ |
876 | /* |
877 | * | - range to drop - | | |
878 | * | -------- extent -------- | | |
879 | */ | |
880 | if (start > key.offset && end < extent_end) { | |
881 | BUG_ON(del_nr > 0); | |
00fdf13a | 882 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
3f9e3df8 | 883 | ret = -EOPNOTSUPP; |
00fdf13a LB |
884 | break; |
885 | } | |
920bbbfb YZ |
886 | |
887 | memcpy(&new_key, &key, sizeof(new_key)); | |
888 | new_key.offset = start; | |
889 | ret = btrfs_duplicate_item(trans, root, path, | |
890 | &new_key); | |
891 | if (ret == -EAGAIN) { | |
b3b4aa74 | 892 | btrfs_release_path(path); |
920bbbfb | 893 | continue; |
6643558d | 894 | } |
920bbbfb YZ |
895 | if (ret < 0) |
896 | break; | |
897 | ||
898 | leaf = path->nodes[0]; | |
899 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
900 | struct btrfs_file_extent_item); | |
901 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
902 | start - key.offset); | |
903 | ||
904 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
905 | struct btrfs_file_extent_item); | |
906 | ||
907 | extent_offset += start - key.offset; | |
908 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); | |
909 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
910 | extent_end - start); | |
911 | btrfs_mark_buffer_dirty(leaf); | |
912 | ||
5dc562c5 | 913 | if (update_refs && disk_bytenr > 0) { |
82fa113f QW |
914 | btrfs_init_generic_ref(&ref, |
915 | BTRFS_ADD_DELAYED_REF, | |
916 | disk_bytenr, num_bytes, 0); | |
917 | btrfs_init_data_ref(&ref, | |
920bbbfb YZ |
918 | root->root_key.objectid, |
919 | new_key.objectid, | |
b06c4bf5 | 920 | start - extent_offset); |
82fa113f | 921 | ret = btrfs_inc_extent_ref(trans, &ref); |
79787eaa | 922 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 923 | } |
920bbbfb | 924 | key.offset = start; |
6643558d | 925 | } |
62fe51c1 JB |
926 | /* |
927 | * From here on out we will have actually dropped something, so | |
928 | * last_end can be updated. | |
929 | */ | |
930 | last_end = extent_end; | |
931 | ||
920bbbfb YZ |
932 | /* |
933 | * | ---- range to drop ----- | | |
934 | * | -------- extent -------- | | |
935 | */ | |
936 | if (start <= key.offset && end < extent_end) { | |
00fdf13a | 937 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
3f9e3df8 | 938 | ret = -EOPNOTSUPP; |
00fdf13a LB |
939 | break; |
940 | } | |
6643558d | 941 | |
920bbbfb YZ |
942 | memcpy(&new_key, &key, sizeof(new_key)); |
943 | new_key.offset = end; | |
0b246afa | 944 | btrfs_set_item_key_safe(fs_info, path, &new_key); |
6643558d | 945 | |
920bbbfb YZ |
946 | extent_offset += end - key.offset; |
947 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); | |
948 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
949 | extent_end - end); | |
950 | btrfs_mark_buffer_dirty(leaf); | |
2671485d | 951 | if (update_refs && disk_bytenr > 0) |
920bbbfb | 952 | inode_sub_bytes(inode, end - key.offset); |
920bbbfb | 953 | break; |
39279cc3 | 954 | } |
771ed689 | 955 | |
920bbbfb YZ |
956 | search_start = extent_end; |
957 | /* | |
958 | * | ---- range to drop ----- | | |
959 | * | -------- extent -------- | | |
960 | */ | |
961 | if (start > key.offset && end >= extent_end) { | |
962 | BUG_ON(del_nr > 0); | |
00fdf13a | 963 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
3f9e3df8 | 964 | ret = -EOPNOTSUPP; |
00fdf13a LB |
965 | break; |
966 | } | |
8c2383c3 | 967 | |
920bbbfb YZ |
968 | btrfs_set_file_extent_num_bytes(leaf, fi, |
969 | start - key.offset); | |
970 | btrfs_mark_buffer_dirty(leaf); | |
2671485d | 971 | if (update_refs && disk_bytenr > 0) |
920bbbfb | 972 | inode_sub_bytes(inode, extent_end - start); |
920bbbfb YZ |
973 | if (end == extent_end) |
974 | break; | |
c8b97818 | 975 | |
920bbbfb YZ |
976 | path->slots[0]++; |
977 | goto next_slot; | |
31840ae1 ZY |
978 | } |
979 | ||
920bbbfb YZ |
980 | /* |
981 | * | ---- range to drop ----- | | |
982 | * | ------ extent ------ | | |
983 | */ | |
984 | if (start <= key.offset && end >= extent_end) { | |
fc19c5e7 | 985 | delete_extent_item: |
920bbbfb YZ |
986 | if (del_nr == 0) { |
987 | del_slot = path->slots[0]; | |
988 | del_nr = 1; | |
989 | } else { | |
990 | BUG_ON(del_slot + del_nr != path->slots[0]); | |
991 | del_nr++; | |
992 | } | |
31840ae1 | 993 | |
5dc562c5 JB |
994 | if (update_refs && |
995 | extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
a76a3cd4 | 996 | inode_sub_bytes(inode, |
920bbbfb YZ |
997 | extent_end - key.offset); |
998 | extent_end = ALIGN(extent_end, | |
0b246afa | 999 | fs_info->sectorsize); |
5dc562c5 | 1000 | } else if (update_refs && disk_bytenr > 0) { |
ffd4bb2a QW |
1001 | btrfs_init_generic_ref(&ref, |
1002 | BTRFS_DROP_DELAYED_REF, | |
1003 | disk_bytenr, num_bytes, 0); | |
1004 | btrfs_init_data_ref(&ref, | |
920bbbfb | 1005 | root->root_key.objectid, |
ffd4bb2a QW |
1006 | key.objectid, |
1007 | key.offset - extent_offset); | |
1008 | ret = btrfs_free_extent(trans, &ref); | |
79787eaa | 1009 | BUG_ON(ret); /* -ENOMEM */ |
920bbbfb YZ |
1010 | inode_sub_bytes(inode, |
1011 | extent_end - key.offset); | |
31840ae1 | 1012 | } |
31840ae1 | 1013 | |
920bbbfb YZ |
1014 | if (end == extent_end) |
1015 | break; | |
1016 | ||
1017 | if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) { | |
1018 | path->slots[0]++; | |
1019 | goto next_slot; | |
1020 | } | |
1021 | ||
1022 | ret = btrfs_del_items(trans, root, path, del_slot, | |
1023 | del_nr); | |
79787eaa | 1024 | if (ret) { |
66642832 | 1025 | btrfs_abort_transaction(trans, ret); |
5dc562c5 | 1026 | break; |
79787eaa | 1027 | } |
920bbbfb YZ |
1028 | |
1029 | del_nr = 0; | |
1030 | del_slot = 0; | |
1031 | ||
b3b4aa74 | 1032 | btrfs_release_path(path); |
920bbbfb | 1033 | continue; |
39279cc3 | 1034 | } |
920bbbfb | 1035 | |
290342f6 | 1036 | BUG(); |
39279cc3 | 1037 | } |
920bbbfb | 1038 | |
79787eaa | 1039 | if (!ret && del_nr > 0) { |
1acae57b FDBM |
1040 | /* |
1041 | * Set path->slots[0] to first slot, so that after the delete | |
1042 | * if items are move off from our leaf to its immediate left or | |
1043 | * right neighbor leafs, we end up with a correct and adjusted | |
d5f37527 | 1044 | * path->slots[0] for our insertion (if replace_extent != 0). |
1acae57b FDBM |
1045 | */ |
1046 | path->slots[0] = del_slot; | |
920bbbfb | 1047 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
79787eaa | 1048 | if (ret) |
66642832 | 1049 | btrfs_abort_transaction(trans, ret); |
d5f37527 | 1050 | } |
1acae57b | 1051 | |
d5f37527 FDBM |
1052 | leaf = path->nodes[0]; |
1053 | /* | |
1054 | * If btrfs_del_items() was called, it might have deleted a leaf, in | |
1055 | * which case it unlocked our path, so check path->locks[0] matches a | |
1056 | * write lock. | |
1057 | */ | |
1058 | if (!ret && replace_extent && leafs_visited == 1 && | |
1059 | (path->locks[0] == BTRFS_WRITE_LOCK_BLOCKING || | |
1060 | path->locks[0] == BTRFS_WRITE_LOCK) && | |
e902baac | 1061 | btrfs_leaf_free_space(leaf) >= |
d5f37527 FDBM |
1062 | sizeof(struct btrfs_item) + extent_item_size) { |
1063 | ||
1064 | key.objectid = ino; | |
1065 | key.type = BTRFS_EXTENT_DATA_KEY; | |
1066 | key.offset = start; | |
1067 | if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) { | |
1068 | struct btrfs_key slot_key; | |
1069 | ||
1070 | btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]); | |
1071 | if (btrfs_comp_cpu_keys(&key, &slot_key) > 0) | |
1072 | path->slots[0]++; | |
1acae57b | 1073 | } |
d5f37527 FDBM |
1074 | setup_items_for_insert(root, path, &key, |
1075 | &extent_item_size, | |
1076 | extent_item_size, | |
1077 | sizeof(struct btrfs_item) + | |
1078 | extent_item_size, 1); | |
1079 | *key_inserted = 1; | |
6643558d | 1080 | } |
920bbbfb | 1081 | |
1acae57b FDBM |
1082 | if (!replace_extent || !(*key_inserted)) |
1083 | btrfs_release_path(path); | |
2aaa6655 | 1084 | if (drop_end) |
62fe51c1 | 1085 | *drop_end = found ? min(end, last_end) : end; |
5dc562c5 JB |
1086 | return ret; |
1087 | } | |
1088 | ||
1089 | int btrfs_drop_extents(struct btrfs_trans_handle *trans, | |
1090 | struct btrfs_root *root, struct inode *inode, u64 start, | |
2671485d | 1091 | u64 end, int drop_cache) |
5dc562c5 JB |
1092 | { |
1093 | struct btrfs_path *path; | |
1094 | int ret; | |
1095 | ||
1096 | path = btrfs_alloc_path(); | |
1097 | if (!path) | |
1098 | return -ENOMEM; | |
2aaa6655 | 1099 | ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL, |
1acae57b | 1100 | drop_cache, 0, 0, NULL); |
920bbbfb | 1101 | btrfs_free_path(path); |
39279cc3 CM |
1102 | return ret; |
1103 | } | |
1104 | ||
d899e052 | 1105 | static int extent_mergeable(struct extent_buffer *leaf, int slot, |
6c7d54ac YZ |
1106 | u64 objectid, u64 bytenr, u64 orig_offset, |
1107 | u64 *start, u64 *end) | |
d899e052 YZ |
1108 | { |
1109 | struct btrfs_file_extent_item *fi; | |
1110 | struct btrfs_key key; | |
1111 | u64 extent_end; | |
1112 | ||
1113 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) | |
1114 | return 0; | |
1115 | ||
1116 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
1117 | if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY) | |
1118 | return 0; | |
1119 | ||
1120 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
1121 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG || | |
1122 | btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr || | |
6c7d54ac | 1123 | btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset || |
d899e052 YZ |
1124 | btrfs_file_extent_compression(leaf, fi) || |
1125 | btrfs_file_extent_encryption(leaf, fi) || | |
1126 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1127 | return 0; | |
1128 | ||
1129 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
1130 | if ((*start && *start != key.offset) || (*end && *end != extent_end)) | |
1131 | return 0; | |
1132 | ||
1133 | *start = key.offset; | |
1134 | *end = extent_end; | |
1135 | return 1; | |
1136 | } | |
1137 | ||
1138 | /* | |
1139 | * Mark extent in the range start - end as written. | |
1140 | * | |
1141 | * This changes extent type from 'pre-allocated' to 'regular'. If only | |
1142 | * part of extent is marked as written, the extent will be split into | |
1143 | * two or three. | |
1144 | */ | |
1145 | int btrfs_mark_extent_written(struct btrfs_trans_handle *trans, | |
7a6d7067 | 1146 | struct btrfs_inode *inode, u64 start, u64 end) |
d899e052 | 1147 | { |
3ffbd68c | 1148 | struct btrfs_fs_info *fs_info = trans->fs_info; |
7a6d7067 | 1149 | struct btrfs_root *root = inode->root; |
d899e052 YZ |
1150 | struct extent_buffer *leaf; |
1151 | struct btrfs_path *path; | |
1152 | struct btrfs_file_extent_item *fi; | |
82fa113f | 1153 | struct btrfs_ref ref = { 0 }; |
d899e052 | 1154 | struct btrfs_key key; |
920bbbfb | 1155 | struct btrfs_key new_key; |
d899e052 YZ |
1156 | u64 bytenr; |
1157 | u64 num_bytes; | |
1158 | u64 extent_end; | |
5d4f98a2 | 1159 | u64 orig_offset; |
d899e052 YZ |
1160 | u64 other_start; |
1161 | u64 other_end; | |
920bbbfb YZ |
1162 | u64 split; |
1163 | int del_nr = 0; | |
1164 | int del_slot = 0; | |
6c7d54ac | 1165 | int recow; |
d899e052 | 1166 | int ret; |
7a6d7067 | 1167 | u64 ino = btrfs_ino(inode); |
d899e052 | 1168 | |
d899e052 | 1169 | path = btrfs_alloc_path(); |
d8926bb3 MF |
1170 | if (!path) |
1171 | return -ENOMEM; | |
d899e052 | 1172 | again: |
6c7d54ac | 1173 | recow = 0; |
920bbbfb | 1174 | split = start; |
33345d01 | 1175 | key.objectid = ino; |
d899e052 | 1176 | key.type = BTRFS_EXTENT_DATA_KEY; |
920bbbfb | 1177 | key.offset = split; |
d899e052 YZ |
1178 | |
1179 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
41415730 JB |
1180 | if (ret < 0) |
1181 | goto out; | |
d899e052 YZ |
1182 | if (ret > 0 && path->slots[0] > 0) |
1183 | path->slots[0]--; | |
1184 | ||
1185 | leaf = path->nodes[0]; | |
1186 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
9c8e63db JB |
1187 | if (key.objectid != ino || |
1188 | key.type != BTRFS_EXTENT_DATA_KEY) { | |
1189 | ret = -EINVAL; | |
1190 | btrfs_abort_transaction(trans, ret); | |
1191 | goto out; | |
1192 | } | |
d899e052 YZ |
1193 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1194 | struct btrfs_file_extent_item); | |
9c8e63db JB |
1195 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_PREALLOC) { |
1196 | ret = -EINVAL; | |
1197 | btrfs_abort_transaction(trans, ret); | |
1198 | goto out; | |
1199 | } | |
d899e052 | 1200 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
9c8e63db JB |
1201 | if (key.offset > start || extent_end < end) { |
1202 | ret = -EINVAL; | |
1203 | btrfs_abort_transaction(trans, ret); | |
1204 | goto out; | |
1205 | } | |
d899e052 YZ |
1206 | |
1207 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
1208 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
5d4f98a2 | 1209 | orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi); |
6c7d54ac YZ |
1210 | memcpy(&new_key, &key, sizeof(new_key)); |
1211 | ||
1212 | if (start == key.offset && end < extent_end) { | |
1213 | other_start = 0; | |
1214 | other_end = start; | |
1215 | if (extent_mergeable(leaf, path->slots[0] - 1, | |
33345d01 | 1216 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1217 | &other_start, &other_end)) { |
1218 | new_key.offset = end; | |
0b246afa | 1219 | btrfs_set_item_key_safe(fs_info, path, &new_key); |
6c7d54ac YZ |
1220 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1221 | struct btrfs_file_extent_item); | |
224ecce5 JB |
1222 | btrfs_set_file_extent_generation(leaf, fi, |
1223 | trans->transid); | |
6c7d54ac YZ |
1224 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1225 | extent_end - end); | |
1226 | btrfs_set_file_extent_offset(leaf, fi, | |
1227 | end - orig_offset); | |
1228 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
1229 | struct btrfs_file_extent_item); | |
224ecce5 JB |
1230 | btrfs_set_file_extent_generation(leaf, fi, |
1231 | trans->transid); | |
6c7d54ac YZ |
1232 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1233 | end - other_start); | |
1234 | btrfs_mark_buffer_dirty(leaf); | |
1235 | goto out; | |
1236 | } | |
1237 | } | |
1238 | ||
1239 | if (start > key.offset && end == extent_end) { | |
1240 | other_start = end; | |
1241 | other_end = 0; | |
1242 | if (extent_mergeable(leaf, path->slots[0] + 1, | |
33345d01 | 1243 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1244 | &other_start, &other_end)) { |
1245 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1246 | struct btrfs_file_extent_item); | |
1247 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
1248 | start - key.offset); | |
224ecce5 JB |
1249 | btrfs_set_file_extent_generation(leaf, fi, |
1250 | trans->transid); | |
6c7d54ac YZ |
1251 | path->slots[0]++; |
1252 | new_key.offset = start; | |
0b246afa | 1253 | btrfs_set_item_key_safe(fs_info, path, &new_key); |
6c7d54ac YZ |
1254 | |
1255 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1256 | struct btrfs_file_extent_item); | |
224ecce5 JB |
1257 | btrfs_set_file_extent_generation(leaf, fi, |
1258 | trans->transid); | |
6c7d54ac YZ |
1259 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1260 | other_end - start); | |
1261 | btrfs_set_file_extent_offset(leaf, fi, | |
1262 | start - orig_offset); | |
1263 | btrfs_mark_buffer_dirty(leaf); | |
1264 | goto out; | |
1265 | } | |
1266 | } | |
d899e052 | 1267 | |
920bbbfb YZ |
1268 | while (start > key.offset || end < extent_end) { |
1269 | if (key.offset == start) | |
1270 | split = end; | |
1271 | ||
920bbbfb YZ |
1272 | new_key.offset = split; |
1273 | ret = btrfs_duplicate_item(trans, root, path, &new_key); | |
1274 | if (ret == -EAGAIN) { | |
b3b4aa74 | 1275 | btrfs_release_path(path); |
920bbbfb | 1276 | goto again; |
d899e052 | 1277 | } |
79787eaa | 1278 | if (ret < 0) { |
66642832 | 1279 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
1280 | goto out; |
1281 | } | |
d899e052 | 1282 | |
920bbbfb YZ |
1283 | leaf = path->nodes[0]; |
1284 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
d899e052 | 1285 | struct btrfs_file_extent_item); |
224ecce5 | 1286 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
d899e052 | 1287 | btrfs_set_file_extent_num_bytes(leaf, fi, |
920bbbfb YZ |
1288 | split - key.offset); |
1289 | ||
1290 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1291 | struct btrfs_file_extent_item); | |
1292 | ||
224ecce5 | 1293 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
920bbbfb YZ |
1294 | btrfs_set_file_extent_offset(leaf, fi, split - orig_offset); |
1295 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
1296 | extent_end - split); | |
d899e052 YZ |
1297 | btrfs_mark_buffer_dirty(leaf); |
1298 | ||
82fa113f QW |
1299 | btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, bytenr, |
1300 | num_bytes, 0); | |
1301 | btrfs_init_data_ref(&ref, root->root_key.objectid, ino, | |
1302 | orig_offset); | |
1303 | ret = btrfs_inc_extent_ref(trans, &ref); | |
9c8e63db JB |
1304 | if (ret) { |
1305 | btrfs_abort_transaction(trans, ret); | |
1306 | goto out; | |
1307 | } | |
d899e052 | 1308 | |
920bbbfb YZ |
1309 | if (split == start) { |
1310 | key.offset = start; | |
1311 | } else { | |
9c8e63db JB |
1312 | if (start != key.offset) { |
1313 | ret = -EINVAL; | |
1314 | btrfs_abort_transaction(trans, ret); | |
1315 | goto out; | |
1316 | } | |
d899e052 | 1317 | path->slots[0]--; |
920bbbfb | 1318 | extent_end = end; |
d899e052 | 1319 | } |
6c7d54ac | 1320 | recow = 1; |
d899e052 YZ |
1321 | } |
1322 | ||
920bbbfb YZ |
1323 | other_start = end; |
1324 | other_end = 0; | |
ffd4bb2a QW |
1325 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, |
1326 | num_bytes, 0); | |
1327 | btrfs_init_data_ref(&ref, root->root_key.objectid, ino, orig_offset); | |
6c7d54ac | 1328 | if (extent_mergeable(leaf, path->slots[0] + 1, |
33345d01 | 1329 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1330 | &other_start, &other_end)) { |
1331 | if (recow) { | |
b3b4aa74 | 1332 | btrfs_release_path(path); |
6c7d54ac YZ |
1333 | goto again; |
1334 | } | |
920bbbfb YZ |
1335 | extent_end = other_end; |
1336 | del_slot = path->slots[0] + 1; | |
1337 | del_nr++; | |
ffd4bb2a | 1338 | ret = btrfs_free_extent(trans, &ref); |
9c8e63db JB |
1339 | if (ret) { |
1340 | btrfs_abort_transaction(trans, ret); | |
1341 | goto out; | |
1342 | } | |
d899e052 | 1343 | } |
920bbbfb YZ |
1344 | other_start = 0; |
1345 | other_end = start; | |
6c7d54ac | 1346 | if (extent_mergeable(leaf, path->slots[0] - 1, |
33345d01 | 1347 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1348 | &other_start, &other_end)) { |
1349 | if (recow) { | |
b3b4aa74 | 1350 | btrfs_release_path(path); |
6c7d54ac YZ |
1351 | goto again; |
1352 | } | |
920bbbfb YZ |
1353 | key.offset = other_start; |
1354 | del_slot = path->slots[0]; | |
1355 | del_nr++; | |
ffd4bb2a | 1356 | ret = btrfs_free_extent(trans, &ref); |
9c8e63db JB |
1357 | if (ret) { |
1358 | btrfs_abort_transaction(trans, ret); | |
1359 | goto out; | |
1360 | } | |
920bbbfb YZ |
1361 | } |
1362 | if (del_nr == 0) { | |
3f6fae95 SL |
1363 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1364 | struct btrfs_file_extent_item); | |
920bbbfb YZ |
1365 | btrfs_set_file_extent_type(leaf, fi, |
1366 | BTRFS_FILE_EXTENT_REG); | |
224ecce5 | 1367 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
920bbbfb | 1368 | btrfs_mark_buffer_dirty(leaf); |
6c7d54ac | 1369 | } else { |
3f6fae95 SL |
1370 | fi = btrfs_item_ptr(leaf, del_slot - 1, |
1371 | struct btrfs_file_extent_item); | |
6c7d54ac YZ |
1372 | btrfs_set_file_extent_type(leaf, fi, |
1373 | BTRFS_FILE_EXTENT_REG); | |
224ecce5 | 1374 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
6c7d54ac YZ |
1375 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1376 | extent_end - key.offset); | |
1377 | btrfs_mark_buffer_dirty(leaf); | |
920bbbfb | 1378 | |
6c7d54ac | 1379 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
79787eaa | 1380 | if (ret < 0) { |
66642832 | 1381 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
1382 | goto out; |
1383 | } | |
6c7d54ac | 1384 | } |
920bbbfb | 1385 | out: |
d899e052 YZ |
1386 | btrfs_free_path(path); |
1387 | return 0; | |
1388 | } | |
1389 | ||
b1bf862e CM |
1390 | /* |
1391 | * on error we return an unlocked page and the error value | |
1392 | * on success we return a locked page and 0 | |
1393 | */ | |
bb1591b4 CM |
1394 | static int prepare_uptodate_page(struct inode *inode, |
1395 | struct page *page, u64 pos, | |
b6316429 | 1396 | bool force_uptodate) |
b1bf862e CM |
1397 | { |
1398 | int ret = 0; | |
1399 | ||
09cbfeaf | 1400 | if (((pos & (PAGE_SIZE - 1)) || force_uptodate) && |
b6316429 | 1401 | !PageUptodate(page)) { |
b1bf862e CM |
1402 | ret = btrfs_readpage(NULL, page); |
1403 | if (ret) | |
1404 | return ret; | |
1405 | lock_page(page); | |
1406 | if (!PageUptodate(page)) { | |
1407 | unlock_page(page); | |
1408 | return -EIO; | |
1409 | } | |
bb1591b4 CM |
1410 | if (page->mapping != inode->i_mapping) { |
1411 | unlock_page(page); | |
1412 | return -EAGAIN; | |
1413 | } | |
b1bf862e CM |
1414 | } |
1415 | return 0; | |
1416 | } | |
1417 | ||
39279cc3 | 1418 | /* |
376cc685 | 1419 | * this just gets pages into the page cache and locks them down. |
39279cc3 | 1420 | */ |
b37392ea MX |
1421 | static noinline int prepare_pages(struct inode *inode, struct page **pages, |
1422 | size_t num_pages, loff_t pos, | |
1423 | size_t write_bytes, bool force_uptodate) | |
39279cc3 CM |
1424 | { |
1425 | int i; | |
09cbfeaf | 1426 | unsigned long index = pos >> PAGE_SHIFT; |
3b16a4e3 | 1427 | gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); |
fc28b62d | 1428 | int err = 0; |
376cc685 | 1429 | int faili; |
8c2383c3 | 1430 | |
39279cc3 | 1431 | for (i = 0; i < num_pages; i++) { |
bb1591b4 | 1432 | again: |
a94733d0 | 1433 | pages[i] = find_or_create_page(inode->i_mapping, index + i, |
e3a41a5b | 1434 | mask | __GFP_WRITE); |
39279cc3 | 1435 | if (!pages[i]) { |
b1bf862e CM |
1436 | faili = i - 1; |
1437 | err = -ENOMEM; | |
1438 | goto fail; | |
1439 | } | |
1440 | ||
1441 | if (i == 0) | |
bb1591b4 | 1442 | err = prepare_uptodate_page(inode, pages[i], pos, |
b6316429 | 1443 | force_uptodate); |
bb1591b4 CM |
1444 | if (!err && i == num_pages - 1) |
1445 | err = prepare_uptodate_page(inode, pages[i], | |
b6316429 | 1446 | pos + write_bytes, false); |
b1bf862e | 1447 | if (err) { |
09cbfeaf | 1448 | put_page(pages[i]); |
bb1591b4 CM |
1449 | if (err == -EAGAIN) { |
1450 | err = 0; | |
1451 | goto again; | |
1452 | } | |
b1bf862e CM |
1453 | faili = i - 1; |
1454 | goto fail; | |
39279cc3 | 1455 | } |
ccd467d6 | 1456 | wait_on_page_writeback(pages[i]); |
39279cc3 | 1457 | } |
376cc685 MX |
1458 | |
1459 | return 0; | |
1460 | fail: | |
1461 | while (faili >= 0) { | |
1462 | unlock_page(pages[faili]); | |
09cbfeaf | 1463 | put_page(pages[faili]); |
376cc685 MX |
1464 | faili--; |
1465 | } | |
1466 | return err; | |
1467 | ||
1468 | } | |
1469 | ||
1470 | /* | |
1471 | * This function locks the extent and properly waits for data=ordered extents | |
1472 | * to finish before allowing the pages to be modified if need. | |
1473 | * | |
1474 | * The return value: | |
1475 | * 1 - the extent is locked | |
1476 | * 0 - the extent is not locked, and everything is OK | |
1477 | * -EAGAIN - need re-prepare the pages | |
1478 | * the other < 0 number - Something wrong happens | |
1479 | */ | |
1480 | static noinline int | |
2cff578c | 1481 | lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages, |
376cc685 | 1482 | size_t num_pages, loff_t pos, |
2e78c927 | 1483 | size_t write_bytes, |
376cc685 MX |
1484 | u64 *lockstart, u64 *lockend, |
1485 | struct extent_state **cached_state) | |
1486 | { | |
3ffbd68c | 1487 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
376cc685 MX |
1488 | u64 start_pos; |
1489 | u64 last_pos; | |
1490 | int i; | |
1491 | int ret = 0; | |
1492 | ||
0b246afa | 1493 | start_pos = round_down(pos, fs_info->sectorsize); |
2e78c927 | 1494 | last_pos = start_pos |
da17066c | 1495 | + round_up(pos + write_bytes - start_pos, |
0b246afa | 1496 | fs_info->sectorsize) - 1; |
376cc685 | 1497 | |
e3b8a485 | 1498 | if (start_pos < inode->vfs_inode.i_size) { |
e6dcd2dc | 1499 | struct btrfs_ordered_extent *ordered; |
a7e3b975 | 1500 | |
2cff578c NB |
1501 | lock_extent_bits(&inode->io_tree, start_pos, last_pos, |
1502 | cached_state); | |
b88935bf MX |
1503 | ordered = btrfs_lookup_ordered_range(inode, start_pos, |
1504 | last_pos - start_pos + 1); | |
e6dcd2dc CM |
1505 | if (ordered && |
1506 | ordered->file_offset + ordered->len > start_pos && | |
376cc685 | 1507 | ordered->file_offset <= last_pos) { |
2cff578c | 1508 | unlock_extent_cached(&inode->io_tree, start_pos, |
e43bbe5e | 1509 | last_pos, cached_state); |
e6dcd2dc CM |
1510 | for (i = 0; i < num_pages; i++) { |
1511 | unlock_page(pages[i]); | |
09cbfeaf | 1512 | put_page(pages[i]); |
e6dcd2dc | 1513 | } |
2cff578c NB |
1514 | btrfs_start_ordered_extent(&inode->vfs_inode, |
1515 | ordered, 1); | |
b88935bf MX |
1516 | btrfs_put_ordered_extent(ordered); |
1517 | return -EAGAIN; | |
e6dcd2dc CM |
1518 | } |
1519 | if (ordered) | |
1520 | btrfs_put_ordered_extent(ordered); | |
7703bdd8 | 1521 | |
376cc685 MX |
1522 | *lockstart = start_pos; |
1523 | *lockend = last_pos; | |
1524 | ret = 1; | |
0762704b | 1525 | } |
376cc685 | 1526 | |
7703bdd8 CM |
1527 | /* |
1528 | * It's possible the pages are dirty right now, but we don't want | |
1529 | * to clean them yet because copy_from_user may catch a page fault | |
1530 | * and we might have to fall back to one page at a time. If that | |
1531 | * happens, we'll unlock these pages and we'd have a window where | |
1532 | * reclaim could sneak in and drop the once-dirty page on the floor | |
1533 | * without writing it. | |
1534 | * | |
1535 | * We have the pages locked and the extent range locked, so there's | |
1536 | * no way someone can start IO on any dirty pages in this range. | |
1537 | * | |
1538 | * We'll call btrfs_dirty_pages() later on, and that will flip around | |
1539 | * delalloc bits and dirty the pages as required. | |
1540 | */ | |
e6dcd2dc | 1541 | for (i = 0; i < num_pages; i++) { |
e6dcd2dc CM |
1542 | set_page_extent_mapped(pages[i]); |
1543 | WARN_ON(!PageLocked(pages[i])); | |
1544 | } | |
b1bf862e | 1545 | |
376cc685 | 1546 | return ret; |
39279cc3 CM |
1547 | } |
1548 | ||
85b7ab67 | 1549 | static noinline int check_can_nocow(struct btrfs_inode *inode, loff_t pos, |
7ee9e440 JB |
1550 | size_t *write_bytes) |
1551 | { | |
3ffbd68c | 1552 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
85b7ab67 | 1553 | struct btrfs_root *root = inode->root; |
7ee9e440 JB |
1554 | u64 lockstart, lockend; |
1555 | u64 num_bytes; | |
1556 | int ret; | |
1557 | ||
ea14b57f | 1558 | ret = btrfs_start_write_no_snapshotting(root); |
8257b2dc | 1559 | if (!ret) |
5f791ec3 | 1560 | return -EAGAIN; |
8257b2dc | 1561 | |
0b246afa | 1562 | lockstart = round_down(pos, fs_info->sectorsize); |
da17066c | 1563 | lockend = round_up(pos + *write_bytes, |
0b246afa | 1564 | fs_info->sectorsize) - 1; |
7ee9e440 | 1565 | |
23d31bd4 NB |
1566 | btrfs_lock_and_flush_ordered_range(&inode->io_tree, inode, lockstart, |
1567 | lockend, NULL); | |
7ee9e440 | 1568 | |
7ee9e440 | 1569 | num_bytes = lockend - lockstart + 1; |
85b7ab67 NB |
1570 | ret = can_nocow_extent(&inode->vfs_inode, lockstart, &num_bytes, |
1571 | NULL, NULL, NULL); | |
7ee9e440 JB |
1572 | if (ret <= 0) { |
1573 | ret = 0; | |
ea14b57f | 1574 | btrfs_end_write_no_snapshotting(root); |
7ee9e440 | 1575 | } else { |
c933956d MX |
1576 | *write_bytes = min_t(size_t, *write_bytes , |
1577 | num_bytes - pos + lockstart); | |
7ee9e440 JB |
1578 | } |
1579 | ||
85b7ab67 | 1580 | unlock_extent(&inode->io_tree, lockstart, lockend); |
7ee9e440 JB |
1581 | |
1582 | return ret; | |
1583 | } | |
1584 | ||
e4af400a GR |
1585 | static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb, |
1586 | struct iov_iter *i) | |
4b46fce2 | 1587 | { |
e4af400a GR |
1588 | struct file *file = iocb->ki_filp; |
1589 | loff_t pos = iocb->ki_pos; | |
496ad9aa | 1590 | struct inode *inode = file_inode(file); |
0b246afa | 1591 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
11c65dcc | 1592 | struct btrfs_root *root = BTRFS_I(inode)->root; |
11c65dcc | 1593 | struct page **pages = NULL; |
364ecf36 | 1594 | struct extent_changeset *data_reserved = NULL; |
7ee9e440 | 1595 | u64 release_bytes = 0; |
376cc685 MX |
1596 | u64 lockstart; |
1597 | u64 lockend; | |
d0215f3e JB |
1598 | size_t num_written = 0; |
1599 | int nrptrs; | |
c9149235 | 1600 | int ret = 0; |
7ee9e440 | 1601 | bool only_release_metadata = false; |
b6316429 | 1602 | bool force_page_uptodate = false; |
4b46fce2 | 1603 | |
09cbfeaf KS |
1604 | nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_SIZE), |
1605 | PAGE_SIZE / (sizeof(struct page *))); | |
142349f5 WF |
1606 | nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied); |
1607 | nrptrs = max(nrptrs, 8); | |
31e818fe | 1608 | pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL); |
d0215f3e JB |
1609 | if (!pages) |
1610 | return -ENOMEM; | |
ab93dbec | 1611 | |
d0215f3e | 1612 | while (iov_iter_count(i) > 0) { |
91a7ad84 | 1613 | struct extent_state *cached_state = NULL; |
7073017a | 1614 | size_t offset = offset_in_page(pos); |
2e78c927 | 1615 | size_t sector_offset; |
d0215f3e | 1616 | size_t write_bytes = min(iov_iter_count(i), |
09cbfeaf | 1617 | nrptrs * (size_t)PAGE_SIZE - |
8c2383c3 | 1618 | offset); |
ed6078f7 | 1619 | size_t num_pages = DIV_ROUND_UP(write_bytes + offset, |
09cbfeaf | 1620 | PAGE_SIZE); |
7ee9e440 | 1621 | size_t reserve_bytes; |
d0215f3e JB |
1622 | size_t dirty_pages; |
1623 | size_t copied; | |
2e78c927 CR |
1624 | size_t dirty_sectors; |
1625 | size_t num_sectors; | |
79f015f2 | 1626 | int extents_locked; |
39279cc3 | 1627 | |
8c2383c3 | 1628 | WARN_ON(num_pages > nrptrs); |
1832a6d5 | 1629 | |
914ee295 XZ |
1630 | /* |
1631 | * Fault pages before locking them in prepare_pages | |
1632 | * to avoid recursive lock | |
1633 | */ | |
d0215f3e | 1634 | if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) { |
914ee295 | 1635 | ret = -EFAULT; |
d0215f3e | 1636 | break; |
914ee295 XZ |
1637 | } |
1638 | ||
da17066c | 1639 | sector_offset = pos & (fs_info->sectorsize - 1); |
2e78c927 | 1640 | reserve_bytes = round_up(write_bytes + sector_offset, |
da17066c | 1641 | fs_info->sectorsize); |
d9d8b2a5 | 1642 | |
364ecf36 QW |
1643 | extent_changeset_release(data_reserved); |
1644 | ret = btrfs_check_data_free_space(inode, &data_reserved, pos, | |
1645 | write_bytes); | |
c6887cd1 JB |
1646 | if (ret < 0) { |
1647 | if ((BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW | | |
1648 | BTRFS_INODE_PREALLOC)) && | |
85b7ab67 NB |
1649 | check_can_nocow(BTRFS_I(inode), pos, |
1650 | &write_bytes) > 0) { | |
c6887cd1 JB |
1651 | /* |
1652 | * For nodata cow case, no need to reserve | |
1653 | * data space. | |
1654 | */ | |
1655 | only_release_metadata = true; | |
1656 | /* | |
1657 | * our prealloc extent may be smaller than | |
1658 | * write_bytes, so scale down. | |
1659 | */ | |
1660 | num_pages = DIV_ROUND_UP(write_bytes + offset, | |
1661 | PAGE_SIZE); | |
1662 | reserve_bytes = round_up(write_bytes + | |
1663 | sector_offset, | |
da17066c | 1664 | fs_info->sectorsize); |
c6887cd1 JB |
1665 | } else { |
1666 | break; | |
1667 | } | |
1668 | } | |
1832a6d5 | 1669 | |
8b62f87b | 1670 | WARN_ON(reserve_bytes == 0); |
9f3db423 NB |
1671 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), |
1672 | reserve_bytes); | |
7ee9e440 JB |
1673 | if (ret) { |
1674 | if (!only_release_metadata) | |
bc42bda2 QW |
1675 | btrfs_free_reserved_data_space(inode, |
1676 | data_reserved, pos, | |
1677 | write_bytes); | |
8257b2dc | 1678 | else |
ea14b57f | 1679 | btrfs_end_write_no_snapshotting(root); |
7ee9e440 JB |
1680 | break; |
1681 | } | |
1682 | ||
1683 | release_bytes = reserve_bytes; | |
376cc685 | 1684 | again: |
4a64001f JB |
1685 | /* |
1686 | * This is going to setup the pages array with the number of | |
1687 | * pages we want, so we don't really need to worry about the | |
1688 | * contents of pages from loop to loop | |
1689 | */ | |
b37392ea MX |
1690 | ret = prepare_pages(inode, pages, num_pages, |
1691 | pos, write_bytes, | |
b6316429 | 1692 | force_page_uptodate); |
8b62f87b JB |
1693 | if (ret) { |
1694 | btrfs_delalloc_release_extents(BTRFS_I(inode), | |
ccfa4269 | 1695 | reserve_bytes); |
d0215f3e | 1696 | break; |
8b62f87b | 1697 | } |
39279cc3 | 1698 | |
79f015f2 GR |
1699 | extents_locked = lock_and_cleanup_extent_if_need( |
1700 | BTRFS_I(inode), pages, | |
2cff578c NB |
1701 | num_pages, pos, write_bytes, &lockstart, |
1702 | &lockend, &cached_state); | |
79f015f2 GR |
1703 | if (extents_locked < 0) { |
1704 | if (extents_locked == -EAGAIN) | |
376cc685 | 1705 | goto again; |
8b62f87b | 1706 | btrfs_delalloc_release_extents(BTRFS_I(inode), |
ccfa4269 | 1707 | reserve_bytes); |
79f015f2 | 1708 | ret = extents_locked; |
376cc685 | 1709 | break; |
376cc685 MX |
1710 | } |
1711 | ||
ee22f0c4 | 1712 | copied = btrfs_copy_from_user(pos, write_bytes, pages, i); |
b1bf862e | 1713 | |
0b246afa | 1714 | num_sectors = BTRFS_BYTES_TO_BLKS(fs_info, reserve_bytes); |
56244ef1 | 1715 | dirty_sectors = round_up(copied + sector_offset, |
0b246afa JM |
1716 | fs_info->sectorsize); |
1717 | dirty_sectors = BTRFS_BYTES_TO_BLKS(fs_info, dirty_sectors); | |
56244ef1 | 1718 | |
b1bf862e CM |
1719 | /* |
1720 | * if we have trouble faulting in the pages, fall | |
1721 | * back to one page at a time | |
1722 | */ | |
1723 | if (copied < write_bytes) | |
1724 | nrptrs = 1; | |
1725 | ||
b6316429 JB |
1726 | if (copied == 0) { |
1727 | force_page_uptodate = true; | |
56244ef1 | 1728 | dirty_sectors = 0; |
b1bf862e | 1729 | dirty_pages = 0; |
b6316429 JB |
1730 | } else { |
1731 | force_page_uptodate = false; | |
ed6078f7 | 1732 | dirty_pages = DIV_ROUND_UP(copied + offset, |
09cbfeaf | 1733 | PAGE_SIZE); |
b6316429 | 1734 | } |
914ee295 | 1735 | |
2e78c927 | 1736 | if (num_sectors > dirty_sectors) { |
8b8b08cb CM |
1737 | /* release everything except the sectors we dirtied */ |
1738 | release_bytes -= dirty_sectors << | |
0b246afa | 1739 | fs_info->sb->s_blocksize_bits; |
485290a7 | 1740 | if (only_release_metadata) { |
691fa059 | 1741 | btrfs_delalloc_release_metadata(BTRFS_I(inode), |
43b18595 | 1742 | release_bytes, true); |
485290a7 QW |
1743 | } else { |
1744 | u64 __pos; | |
1745 | ||
da17066c | 1746 | __pos = round_down(pos, |
0b246afa | 1747 | fs_info->sectorsize) + |
09cbfeaf | 1748 | (dirty_pages << PAGE_SHIFT); |
bc42bda2 QW |
1749 | btrfs_delalloc_release_space(inode, |
1750 | data_reserved, __pos, | |
43b18595 | 1751 | release_bytes, true); |
485290a7 | 1752 | } |
914ee295 XZ |
1753 | } |
1754 | ||
2e78c927 | 1755 | release_bytes = round_up(copied + sector_offset, |
0b246afa | 1756 | fs_info->sectorsize); |
376cc685 MX |
1757 | |
1758 | if (copied > 0) | |
2ff7e61e | 1759 | ret = btrfs_dirty_pages(inode, pages, dirty_pages, |
94f45071 | 1760 | pos, copied, &cached_state); |
91a7ad84 FM |
1761 | |
1762 | /* | |
1763 | * If we have not locked the extent range, because the range's | |
1764 | * start offset is >= i_size, we might still have a non-NULL | |
1765 | * cached extent state, acquired while marking the extent range | |
1766 | * as delalloc through btrfs_dirty_pages(). Therefore free any | |
1767 | * possible cached extent state to avoid a memory leak. | |
1768 | */ | |
79f015f2 | 1769 | if (extents_locked) |
376cc685 | 1770 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, |
e43bbe5e | 1771 | lockstart, lockend, &cached_state); |
91a7ad84 FM |
1772 | else |
1773 | free_extent_state(cached_state); | |
1774 | ||
ccfa4269 | 1775 | btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes); |
f1de9683 MX |
1776 | if (ret) { |
1777 | btrfs_drop_pages(pages, num_pages); | |
376cc685 | 1778 | break; |
f1de9683 | 1779 | } |
39279cc3 | 1780 | |
376cc685 | 1781 | release_bytes = 0; |
8257b2dc | 1782 | if (only_release_metadata) |
ea14b57f | 1783 | btrfs_end_write_no_snapshotting(root); |
8257b2dc | 1784 | |
7ee9e440 | 1785 | if (only_release_metadata && copied > 0) { |
da17066c | 1786 | lockstart = round_down(pos, |
0b246afa | 1787 | fs_info->sectorsize); |
da17066c | 1788 | lockend = round_up(pos + copied, |
0b246afa | 1789 | fs_info->sectorsize) - 1; |
7ee9e440 JB |
1790 | |
1791 | set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
1792 | lockend, EXTENT_NORESERVE, NULL, | |
1793 | NULL, GFP_NOFS); | |
1794 | only_release_metadata = false; | |
1795 | } | |
1796 | ||
f1de9683 MX |
1797 | btrfs_drop_pages(pages, num_pages); |
1798 | ||
d0215f3e JB |
1799 | cond_resched(); |
1800 | ||
d0e1d66b | 1801 | balance_dirty_pages_ratelimited(inode->i_mapping); |
0b246afa | 1802 | if (dirty_pages < (fs_info->nodesize >> PAGE_SHIFT) + 1) |
2ff7e61e | 1803 | btrfs_btree_balance_dirty(fs_info); |
cb843a6f | 1804 | |
914ee295 XZ |
1805 | pos += copied; |
1806 | num_written += copied; | |
d0215f3e | 1807 | } |
39279cc3 | 1808 | |
d0215f3e JB |
1809 | kfree(pages); |
1810 | ||
7ee9e440 | 1811 | if (release_bytes) { |
8257b2dc | 1812 | if (only_release_metadata) { |
ea14b57f | 1813 | btrfs_end_write_no_snapshotting(root); |
691fa059 | 1814 | btrfs_delalloc_release_metadata(BTRFS_I(inode), |
43b18595 | 1815 | release_bytes, true); |
8257b2dc | 1816 | } else { |
bc42bda2 QW |
1817 | btrfs_delalloc_release_space(inode, data_reserved, |
1818 | round_down(pos, fs_info->sectorsize), | |
43b18595 | 1819 | release_bytes, true); |
8257b2dc | 1820 | } |
7ee9e440 JB |
1821 | } |
1822 | ||
364ecf36 | 1823 | extent_changeset_free(data_reserved); |
d0215f3e JB |
1824 | return num_written ? num_written : ret; |
1825 | } | |
1826 | ||
1af5bb49 | 1827 | static ssize_t __btrfs_direct_write(struct kiocb *iocb, struct iov_iter *from) |
d0215f3e JB |
1828 | { |
1829 | struct file *file = iocb->ki_filp; | |
728404da | 1830 | struct inode *inode = file_inode(file); |
e4af400a | 1831 | loff_t pos; |
d0215f3e JB |
1832 | ssize_t written; |
1833 | ssize_t written_buffered; | |
1834 | loff_t endbyte; | |
1835 | int err; | |
1836 | ||
1af5bb49 | 1837 | written = generic_file_direct_write(iocb, from); |
d0215f3e | 1838 | |
0c949334 | 1839 | if (written < 0 || !iov_iter_count(from)) |
d0215f3e JB |
1840 | return written; |
1841 | ||
e4af400a GR |
1842 | pos = iocb->ki_pos; |
1843 | written_buffered = btrfs_buffered_write(iocb, from); | |
d0215f3e JB |
1844 | if (written_buffered < 0) { |
1845 | err = written_buffered; | |
1846 | goto out; | |
39279cc3 | 1847 | } |
075bdbdb FM |
1848 | /* |
1849 | * Ensure all data is persisted. We want the next direct IO read to be | |
1850 | * able to read what was just written. | |
1851 | */ | |
d0215f3e | 1852 | endbyte = pos + written_buffered - 1; |
728404da | 1853 | err = btrfs_fdatawrite_range(inode, pos, endbyte); |
075bdbdb FM |
1854 | if (err) |
1855 | goto out; | |
728404da | 1856 | err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte); |
d0215f3e JB |
1857 | if (err) |
1858 | goto out; | |
1859 | written += written_buffered; | |
867c4f93 | 1860 | iocb->ki_pos = pos + written_buffered; |
09cbfeaf KS |
1861 | invalidate_mapping_pages(file->f_mapping, pos >> PAGE_SHIFT, |
1862 | endbyte >> PAGE_SHIFT); | |
39279cc3 | 1863 | out: |
d0215f3e JB |
1864 | return written ? written : err; |
1865 | } | |
5b92ee72 | 1866 | |
6c760c07 JB |
1867 | static void update_time_for_write(struct inode *inode) |
1868 | { | |
95582b00 | 1869 | struct timespec64 now; |
6c760c07 JB |
1870 | |
1871 | if (IS_NOCMTIME(inode)) | |
1872 | return; | |
1873 | ||
c2050a45 | 1874 | now = current_time(inode); |
95582b00 | 1875 | if (!timespec64_equal(&inode->i_mtime, &now)) |
6c760c07 JB |
1876 | inode->i_mtime = now; |
1877 | ||
95582b00 | 1878 | if (!timespec64_equal(&inode->i_ctime, &now)) |
6c760c07 JB |
1879 | inode->i_ctime = now; |
1880 | ||
1881 | if (IS_I_VERSION(inode)) | |
1882 | inode_inc_iversion(inode); | |
1883 | } | |
1884 | ||
b30ac0fc AV |
1885 | static ssize_t btrfs_file_write_iter(struct kiocb *iocb, |
1886 | struct iov_iter *from) | |
d0215f3e JB |
1887 | { |
1888 | struct file *file = iocb->ki_filp; | |
496ad9aa | 1889 | struct inode *inode = file_inode(file); |
0b246afa | 1890 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
d0215f3e | 1891 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0c1a98c8 | 1892 | u64 start_pos; |
3ac0d7b9 | 1893 | u64 end_pos; |
d0215f3e | 1894 | ssize_t num_written = 0; |
b812ce28 | 1895 | bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host); |
3309dd04 | 1896 | ssize_t err; |
ff0fa732 | 1897 | loff_t pos; |
edf064e7 | 1898 | size_t count = iov_iter_count(from); |
27772b68 CR |
1899 | loff_t oldsize; |
1900 | int clean_page = 0; | |
d0215f3e | 1901 | |
91f9943e CH |
1902 | if (!(iocb->ki_flags & IOCB_DIRECT) && |
1903 | (iocb->ki_flags & IOCB_NOWAIT)) | |
1904 | return -EOPNOTSUPP; | |
1905 | ||
ff0fa732 GR |
1906 | if (!inode_trylock(inode)) { |
1907 | if (iocb->ki_flags & IOCB_NOWAIT) | |
edf064e7 | 1908 | return -EAGAIN; |
ff0fa732 GR |
1909 | inode_lock(inode); |
1910 | } | |
1911 | ||
1912 | err = generic_write_checks(iocb, from); | |
1913 | if (err <= 0) { | |
1914 | inode_unlock(inode); | |
1915 | return err; | |
1916 | } | |
1917 | ||
1918 | pos = iocb->ki_pos; | |
1919 | if (iocb->ki_flags & IOCB_NOWAIT) { | |
edf064e7 GR |
1920 | /* |
1921 | * We will allocate space in case nodatacow is not set, | |
1922 | * so bail | |
1923 | */ | |
1924 | if (!(BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW | | |
1925 | BTRFS_INODE_PREALLOC)) || | |
1926 | check_can_nocow(BTRFS_I(inode), pos, &count) <= 0) { | |
1927 | inode_unlock(inode); | |
1928 | return -EAGAIN; | |
1929 | } | |
d0215f3e JB |
1930 | } |
1931 | ||
3309dd04 | 1932 | current->backing_dev_info = inode_to_bdi(inode); |
5fa8e0a1 | 1933 | err = file_remove_privs(file); |
d0215f3e | 1934 | if (err) { |
5955102c | 1935 | inode_unlock(inode); |
d0215f3e JB |
1936 | goto out; |
1937 | } | |
1938 | ||
1939 | /* | |
1940 | * If BTRFS flips readonly due to some impossible error | |
1941 | * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR), | |
1942 | * although we have opened a file as writable, we have | |
1943 | * to stop this write operation to ensure FS consistency. | |
1944 | */ | |
0b246afa | 1945 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
5955102c | 1946 | inode_unlock(inode); |
d0215f3e JB |
1947 | err = -EROFS; |
1948 | goto out; | |
1949 | } | |
1950 | ||
6c760c07 JB |
1951 | /* |
1952 | * We reserve space for updating the inode when we reserve space for the | |
1953 | * extent we are going to write, so we will enospc out there. We don't | |
1954 | * need to start yet another transaction to update the inode as we will | |
1955 | * update the inode when we finish writing whatever data we write. | |
1956 | */ | |
1957 | update_time_for_write(inode); | |
d0215f3e | 1958 | |
0b246afa | 1959 | start_pos = round_down(pos, fs_info->sectorsize); |
27772b68 CR |
1960 | oldsize = i_size_read(inode); |
1961 | if (start_pos > oldsize) { | |
3ac0d7b9 | 1962 | /* Expand hole size to cover write data, preventing empty gap */ |
da17066c | 1963 | end_pos = round_up(pos + count, |
0b246afa | 1964 | fs_info->sectorsize); |
27772b68 | 1965 | err = btrfs_cont_expand(inode, oldsize, end_pos); |
0c1a98c8 | 1966 | if (err) { |
5955102c | 1967 | inode_unlock(inode); |
0c1a98c8 MX |
1968 | goto out; |
1969 | } | |
0b246afa | 1970 | if (start_pos > round_up(oldsize, fs_info->sectorsize)) |
27772b68 | 1971 | clean_page = 1; |
0c1a98c8 MX |
1972 | } |
1973 | ||
b812ce28 JB |
1974 | if (sync) |
1975 | atomic_inc(&BTRFS_I(inode)->sync_writers); | |
1976 | ||
2ba48ce5 | 1977 | if (iocb->ki_flags & IOCB_DIRECT) { |
1af5bb49 | 1978 | num_written = __btrfs_direct_write(iocb, from); |
d0215f3e | 1979 | } else { |
e4af400a | 1980 | num_written = btrfs_buffered_write(iocb, from); |
d0215f3e | 1981 | if (num_written > 0) |
867c4f93 | 1982 | iocb->ki_pos = pos + num_written; |
27772b68 CR |
1983 | if (clean_page) |
1984 | pagecache_isize_extended(inode, oldsize, | |
1985 | i_size_read(inode)); | |
d0215f3e JB |
1986 | } |
1987 | ||
5955102c | 1988 | inode_unlock(inode); |
2ff3e9b6 | 1989 | |
5a3f23d5 | 1990 | /* |
6c760c07 JB |
1991 | * We also have to set last_sub_trans to the current log transid, |
1992 | * otherwise subsequent syncs to a file that's been synced in this | |
bb7ab3b9 | 1993 | * transaction will appear to have already occurred. |
5a3f23d5 | 1994 | */ |
2f2ff0ee | 1995 | spin_lock(&BTRFS_I(inode)->lock); |
6c760c07 | 1996 | BTRFS_I(inode)->last_sub_trans = root->log_transid; |
2f2ff0ee | 1997 | spin_unlock(&BTRFS_I(inode)->lock); |
e2592217 CH |
1998 | if (num_written > 0) |
1999 | num_written = generic_write_sync(iocb, num_written); | |
0a3404dc | 2000 | |
b812ce28 JB |
2001 | if (sync) |
2002 | atomic_dec(&BTRFS_I(inode)->sync_writers); | |
0a3404dc | 2003 | out: |
39279cc3 | 2004 | current->backing_dev_info = NULL; |
39279cc3 CM |
2005 | return num_written ? num_written : err; |
2006 | } | |
2007 | ||
d397712b | 2008 | int btrfs_release_file(struct inode *inode, struct file *filp) |
e1b81e67 | 2009 | { |
23b5ec74 JB |
2010 | struct btrfs_file_private *private = filp->private_data; |
2011 | ||
23b5ec74 JB |
2012 | if (private && private->filldir_buf) |
2013 | kfree(private->filldir_buf); | |
2014 | kfree(private); | |
2015 | filp->private_data = NULL; | |
2016 | ||
f6dc45c7 | 2017 | /* |
52042d8e | 2018 | * ordered_data_close is set by setattr when we are about to truncate |
f6dc45c7 CM |
2019 | * a file from a non-zero size to a zero size. This tries to |
2020 | * flush down new bytes that may have been written if the | |
2021 | * application were using truncate to replace a file in place. | |
2022 | */ | |
2023 | if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, | |
2024 | &BTRFS_I(inode)->runtime_flags)) | |
2025 | filemap_flush(inode->i_mapping); | |
e1b81e67 M |
2026 | return 0; |
2027 | } | |
2028 | ||
669249ee FM |
2029 | static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end) |
2030 | { | |
2031 | int ret; | |
343e4fc1 | 2032 | struct blk_plug plug; |
669249ee | 2033 | |
343e4fc1 LB |
2034 | /* |
2035 | * This is only called in fsync, which would do synchronous writes, so | |
2036 | * a plug can merge adjacent IOs as much as possible. Esp. in case of | |
2037 | * multiple disks using raid profile, a large IO can be split to | |
2038 | * several segments of stripe length (currently 64K). | |
2039 | */ | |
2040 | blk_start_plug(&plug); | |
669249ee | 2041 | atomic_inc(&BTRFS_I(inode)->sync_writers); |
728404da | 2042 | ret = btrfs_fdatawrite_range(inode, start, end); |
669249ee | 2043 | atomic_dec(&BTRFS_I(inode)->sync_writers); |
343e4fc1 | 2044 | blk_finish_plug(&plug); |
669249ee FM |
2045 | |
2046 | return ret; | |
2047 | } | |
2048 | ||
d352ac68 CM |
2049 | /* |
2050 | * fsync call for both files and directories. This logs the inode into | |
2051 | * the tree log instead of forcing full commits whenever possible. | |
2052 | * | |
2053 | * It needs to call filemap_fdatawait so that all ordered extent updates are | |
2054 | * in the metadata btree are up to date for copying to the log. | |
2055 | * | |
2056 | * It drops the inode mutex before doing the tree log commit. This is an | |
2057 | * important optimization for directories because holding the mutex prevents | |
2058 | * new operations on the dir while we write to disk. | |
2059 | */ | |
02c24a82 | 2060 | int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync) |
39279cc3 | 2061 | { |
de17e793 | 2062 | struct dentry *dentry = file_dentry(file); |
2b0143b5 | 2063 | struct inode *inode = d_inode(dentry); |
0b246afa | 2064 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 2065 | struct btrfs_root *root = BTRFS_I(inode)->root; |
39279cc3 | 2066 | struct btrfs_trans_handle *trans; |
8b050d35 | 2067 | struct btrfs_log_ctx ctx; |
333427a5 | 2068 | int ret = 0, err; |
39279cc3 | 2069 | |
1abe9b8a | 2070 | trace_btrfs_sync_file(file, datasync); |
257c62e1 | 2071 | |
ebb70442 LB |
2072 | btrfs_init_log_ctx(&ctx, inode); |
2073 | ||
90abccf2 MX |
2074 | /* |
2075 | * We write the dirty pages in the range and wait until they complete | |
2076 | * out of the ->i_mutex. If so, we can flush the dirty pages by | |
2ab28f32 JB |
2077 | * multi-task, and make the performance up. See |
2078 | * btrfs_wait_ordered_range for an explanation of the ASYNC check. | |
90abccf2 | 2079 | */ |
669249ee | 2080 | ret = start_ordered_ops(inode, start, end); |
90abccf2 | 2081 | if (ret) |
333427a5 | 2082 | goto out; |
90abccf2 | 2083 | |
5955102c | 2084 | inode_lock(inode); |
c495144b JB |
2085 | |
2086 | /* | |
2087 | * We take the dio_sem here because the tree log stuff can race with | |
2088 | * lockless dio writes and get an extent map logged for an extent we | |
2089 | * never waited on. We need it this high up for lockdep reasons. | |
2090 | */ | |
2091 | down_write(&BTRFS_I(inode)->dio_sem); | |
2092 | ||
2ecb7923 | 2093 | atomic_inc(&root->log_batch); |
b5e6c3e1 | 2094 | |
d2c5a3c3 FM |
2095 | /* |
2096 | * If the inode needs a full sync, make sure we use a full range to | |
2097 | * avoid log tree corruption, due to hole detection racing with ordered | |
2098 | * extent completion for adjacent ranges, and assertion failures during | |
2099 | * hole detection. Do this while holding the inode lock, to avoid races | |
2100 | * with other tasks. | |
2101 | */ | |
2102 | if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
2103 | &BTRFS_I(inode)->runtime_flags)) { | |
2104 | start = 0; | |
2105 | end = LLONG_MAX; | |
2106 | } | |
2107 | ||
aab15e8e FM |
2108 | /* |
2109 | * Before we acquired the inode's lock, someone may have dirtied more | |
2110 | * pages in the target range. We need to make sure that writeback for | |
2111 | * any such pages does not start while we are logging the inode, because | |
2112 | * if it does, any of the following might happen when we are not doing a | |
2113 | * full inode sync: | |
2114 | * | |
2115 | * 1) We log an extent after its writeback finishes but before its | |
2116 | * checksums are added to the csum tree, leading to -EIO errors | |
2117 | * when attempting to read the extent after a log replay. | |
2118 | * | |
2119 | * 2) We can end up logging an extent before its writeback finishes. | |
2120 | * Therefore after the log replay we will have a file extent item | |
2121 | * pointing to an unwritten extent (and no data checksums as well). | |
2122 | * | |
2123 | * So trigger writeback for any eventual new dirty pages and then we | |
2124 | * wait for all ordered extents to complete below. | |
2125 | */ | |
2126 | ret = start_ordered_ops(inode, start, end); | |
2127 | if (ret) { | |
2128 | inode_unlock(inode); | |
2129 | goto out; | |
2130 | } | |
2131 | ||
669249ee | 2132 | /* |
b5e6c3e1 | 2133 | * We have to do this here to avoid the priority inversion of waiting on |
52042d8e | 2134 | * IO of a lower priority task while holding a transaction open. |
d2c5a3c3 FM |
2135 | * |
2136 | * Also, the range length can be represented by u64, we have to do the | |
2137 | * typecasts to avoid signed overflow if it's [0, LLONG_MAX]. | |
669249ee | 2138 | */ |
d2c5a3c3 | 2139 | ret = btrfs_wait_ordered_range(inode, start, (u64)end - (u64)start + 1); |
669249ee | 2140 | if (ret) { |
c495144b | 2141 | up_write(&BTRFS_I(inode)->dio_sem); |
5955102c | 2142 | inode_unlock(inode); |
669249ee | 2143 | goto out; |
0ef8b726 | 2144 | } |
2ecb7923 | 2145 | atomic_inc(&root->log_batch); |
257c62e1 | 2146 | |
a4abeea4 | 2147 | smp_mb(); |
0f8939b8 | 2148 | if (btrfs_inode_in_log(BTRFS_I(inode), fs_info->generation) || |
ca5788ab | 2149 | BTRFS_I(inode)->last_trans <= fs_info->last_trans_committed) { |
5dc562c5 | 2150 | /* |
01327610 | 2151 | * We've had everything committed since the last time we were |
5dc562c5 JB |
2152 | * modified so clear this flag in case it was set for whatever |
2153 | * reason, it's no longer relevant. | |
2154 | */ | |
2155 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
2156 | &BTRFS_I(inode)->runtime_flags); | |
0596a904 FM |
2157 | /* |
2158 | * An ordered extent might have started before and completed | |
2159 | * already with io errors, in which case the inode was not | |
2160 | * updated and we end up here. So check the inode's mapping | |
333427a5 JL |
2161 | * for any errors that might have happened since we last |
2162 | * checked called fsync. | |
0596a904 | 2163 | */ |
333427a5 | 2164 | ret = filemap_check_wb_err(inode->i_mapping, file->f_wb_err); |
c495144b | 2165 | up_write(&BTRFS_I(inode)->dio_sem); |
5955102c | 2166 | inode_unlock(inode); |
15ee9bc7 JB |
2167 | goto out; |
2168 | } | |
15ee9bc7 | 2169 | |
5039eddc JB |
2170 | /* |
2171 | * We use start here because we will need to wait on the IO to complete | |
2172 | * in btrfs_sync_log, which could require joining a transaction (for | |
2173 | * example checking cross references in the nocow path). If we use join | |
2174 | * here we could get into a situation where we're waiting on IO to | |
2175 | * happen that is blocked on a transaction trying to commit. With start | |
2176 | * we inc the extwriter counter, so we wait for all extwriters to exit | |
52042d8e | 2177 | * before we start blocking joiners. This comment is to keep somebody |
5039eddc JB |
2178 | * from thinking they are super smart and changing this to |
2179 | * btrfs_join_transaction *cough*Josef*cough*. | |
2180 | */ | |
a22285a6 YZ |
2181 | trans = btrfs_start_transaction(root, 0); |
2182 | if (IS_ERR(trans)) { | |
2183 | ret = PTR_ERR(trans); | |
c495144b | 2184 | up_write(&BTRFS_I(inode)->dio_sem); |
5955102c | 2185 | inode_unlock(inode); |
39279cc3 CM |
2186 | goto out; |
2187 | } | |
e02119d5 | 2188 | |
e5b84f7a | 2189 | ret = btrfs_log_dentry_safe(trans, dentry, start, end, &ctx); |
02c24a82 | 2190 | if (ret < 0) { |
a0634be5 FDBM |
2191 | /* Fallthrough and commit/free transaction. */ |
2192 | ret = 1; | |
02c24a82 | 2193 | } |
49eb7e46 CM |
2194 | |
2195 | /* we've logged all the items and now have a consistent | |
2196 | * version of the file in the log. It is possible that | |
2197 | * someone will come in and modify the file, but that's | |
2198 | * fine because the log is consistent on disk, and we | |
2199 | * have references to all of the file's extents | |
2200 | * | |
2201 | * It is possible that someone will come in and log the | |
2202 | * file again, but that will end up using the synchronization | |
2203 | * inside btrfs_sync_log to keep things safe. | |
2204 | */ | |
c495144b | 2205 | up_write(&BTRFS_I(inode)->dio_sem); |
5955102c | 2206 | inode_unlock(inode); |
49eb7e46 | 2207 | |
257c62e1 | 2208 | if (ret != BTRFS_NO_LOG_SYNC) { |
0ef8b726 | 2209 | if (!ret) { |
8b050d35 | 2210 | ret = btrfs_sync_log(trans, root, &ctx); |
0ef8b726 | 2211 | if (!ret) { |
3a45bb20 | 2212 | ret = btrfs_end_transaction(trans); |
0ef8b726 | 2213 | goto out; |
2ab28f32 | 2214 | } |
257c62e1 | 2215 | } |
3a45bb20 | 2216 | ret = btrfs_commit_transaction(trans); |
257c62e1 | 2217 | } else { |
3a45bb20 | 2218 | ret = btrfs_end_transaction(trans); |
e02119d5 | 2219 | } |
39279cc3 | 2220 | out: |
ebb70442 | 2221 | ASSERT(list_empty(&ctx.list)); |
333427a5 JL |
2222 | err = file_check_and_advance_wb_err(file); |
2223 | if (!ret) | |
2224 | ret = err; | |
014e4ac4 | 2225 | return ret > 0 ? -EIO : ret; |
39279cc3 CM |
2226 | } |
2227 | ||
f0f37e2f | 2228 | static const struct vm_operations_struct btrfs_file_vm_ops = { |
92fee66d | 2229 | .fault = filemap_fault, |
f1820361 | 2230 | .map_pages = filemap_map_pages, |
9ebefb18 CM |
2231 | .page_mkwrite = btrfs_page_mkwrite, |
2232 | }; | |
2233 | ||
2234 | static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma) | |
2235 | { | |
058a457e MX |
2236 | struct address_space *mapping = filp->f_mapping; |
2237 | ||
2238 | if (!mapping->a_ops->readpage) | |
2239 | return -ENOEXEC; | |
2240 | ||
9ebefb18 | 2241 | file_accessed(filp); |
058a457e | 2242 | vma->vm_ops = &btrfs_file_vm_ops; |
058a457e | 2243 | |
9ebefb18 CM |
2244 | return 0; |
2245 | } | |
2246 | ||
35339c24 | 2247 | static int hole_mergeable(struct btrfs_inode *inode, struct extent_buffer *leaf, |
2aaa6655 JB |
2248 | int slot, u64 start, u64 end) |
2249 | { | |
2250 | struct btrfs_file_extent_item *fi; | |
2251 | struct btrfs_key key; | |
2252 | ||
2253 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) | |
2254 | return 0; | |
2255 | ||
2256 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
35339c24 | 2257 | if (key.objectid != btrfs_ino(inode) || |
2aaa6655 JB |
2258 | key.type != BTRFS_EXTENT_DATA_KEY) |
2259 | return 0; | |
2260 | ||
2261 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
2262 | ||
2263 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2264 | return 0; | |
2265 | ||
2266 | if (btrfs_file_extent_disk_bytenr(leaf, fi)) | |
2267 | return 0; | |
2268 | ||
2269 | if (key.offset == end) | |
2270 | return 1; | |
2271 | if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start) | |
2272 | return 1; | |
2273 | return 0; | |
2274 | } | |
2275 | ||
a012a74e NB |
2276 | static int fill_holes(struct btrfs_trans_handle *trans, |
2277 | struct btrfs_inode *inode, | |
2278 | struct btrfs_path *path, u64 offset, u64 end) | |
2aaa6655 | 2279 | { |
3ffbd68c | 2280 | struct btrfs_fs_info *fs_info = trans->fs_info; |
a012a74e | 2281 | struct btrfs_root *root = inode->root; |
2aaa6655 JB |
2282 | struct extent_buffer *leaf; |
2283 | struct btrfs_file_extent_item *fi; | |
2284 | struct extent_map *hole_em; | |
a012a74e | 2285 | struct extent_map_tree *em_tree = &inode->extent_tree; |
2aaa6655 JB |
2286 | struct btrfs_key key; |
2287 | int ret; | |
2288 | ||
0b246afa | 2289 | if (btrfs_fs_incompat(fs_info, NO_HOLES)) |
16e7549f JB |
2290 | goto out; |
2291 | ||
a012a74e | 2292 | key.objectid = btrfs_ino(inode); |
2aaa6655 JB |
2293 | key.type = BTRFS_EXTENT_DATA_KEY; |
2294 | key.offset = offset; | |
2295 | ||
2aaa6655 | 2296 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
f94480bd JB |
2297 | if (ret <= 0) { |
2298 | /* | |
2299 | * We should have dropped this offset, so if we find it then | |
2300 | * something has gone horribly wrong. | |
2301 | */ | |
2302 | if (ret == 0) | |
2303 | ret = -EINVAL; | |
2aaa6655 | 2304 | return ret; |
f94480bd | 2305 | } |
2aaa6655 JB |
2306 | |
2307 | leaf = path->nodes[0]; | |
a012a74e | 2308 | if (hole_mergeable(inode, leaf, path->slots[0] - 1, offset, end)) { |
2aaa6655 JB |
2309 | u64 num_bytes; |
2310 | ||
2311 | path->slots[0]--; | |
2312 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2313 | struct btrfs_file_extent_item); | |
2314 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + | |
2315 | end - offset; | |
2316 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2317 | btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes); | |
2318 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
2319 | btrfs_mark_buffer_dirty(leaf); | |
2320 | goto out; | |
2321 | } | |
2322 | ||
1707e26d | 2323 | if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) { |
2aaa6655 JB |
2324 | u64 num_bytes; |
2325 | ||
2aaa6655 | 2326 | key.offset = offset; |
0b246afa | 2327 | btrfs_set_item_key_safe(fs_info, path, &key); |
2aaa6655 JB |
2328 | fi = btrfs_item_ptr(leaf, path->slots[0], |
2329 | struct btrfs_file_extent_item); | |
2330 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end - | |
2331 | offset; | |
2332 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2333 | btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes); | |
2334 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
2335 | btrfs_mark_buffer_dirty(leaf); | |
2336 | goto out; | |
2337 | } | |
2338 | btrfs_release_path(path); | |
2339 | ||
a012a74e | 2340 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), |
f85b7379 | 2341 | offset, 0, 0, end - offset, 0, end - offset, 0, 0, 0); |
2aaa6655 JB |
2342 | if (ret) |
2343 | return ret; | |
2344 | ||
2345 | out: | |
2346 | btrfs_release_path(path); | |
2347 | ||
2348 | hole_em = alloc_extent_map(); | |
2349 | if (!hole_em) { | |
2350 | btrfs_drop_extent_cache(inode, offset, end - 1, 0); | |
a012a74e | 2351 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags); |
2aaa6655 JB |
2352 | } else { |
2353 | hole_em->start = offset; | |
2354 | hole_em->len = end - offset; | |
cc95bef6 | 2355 | hole_em->ram_bytes = hole_em->len; |
2aaa6655 JB |
2356 | hole_em->orig_start = offset; |
2357 | ||
2358 | hole_em->block_start = EXTENT_MAP_HOLE; | |
2359 | hole_em->block_len = 0; | |
b4939680 | 2360 | hole_em->orig_block_len = 0; |
0b246afa | 2361 | hole_em->bdev = fs_info->fs_devices->latest_bdev; |
2aaa6655 JB |
2362 | hole_em->compress_type = BTRFS_COMPRESS_NONE; |
2363 | hole_em->generation = trans->transid; | |
2364 | ||
2365 | do { | |
2366 | btrfs_drop_extent_cache(inode, offset, end - 1, 0); | |
2367 | write_lock(&em_tree->lock); | |
09a2a8f9 | 2368 | ret = add_extent_mapping(em_tree, hole_em, 1); |
2aaa6655 JB |
2369 | write_unlock(&em_tree->lock); |
2370 | } while (ret == -EEXIST); | |
2371 | free_extent_map(hole_em); | |
2372 | if (ret) | |
2373 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
a012a74e | 2374 | &inode->runtime_flags); |
2aaa6655 JB |
2375 | } |
2376 | ||
2377 | return 0; | |
2378 | } | |
2379 | ||
d7781546 QW |
2380 | /* |
2381 | * Find a hole extent on given inode and change start/len to the end of hole | |
2382 | * extent.(hole/vacuum extent whose em->start <= start && | |
2383 | * em->start + em->len > start) | |
2384 | * When a hole extent is found, return 1 and modify start/len. | |
2385 | */ | |
2386 | static int find_first_non_hole(struct inode *inode, u64 *start, u64 *len) | |
2387 | { | |
609805d8 | 2388 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
d7781546 QW |
2389 | struct extent_map *em; |
2390 | int ret = 0; | |
2391 | ||
609805d8 FM |
2392 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, |
2393 | round_down(*start, fs_info->sectorsize), | |
2394 | round_up(*len, fs_info->sectorsize), 0); | |
9986277e DC |
2395 | if (IS_ERR(em)) |
2396 | return PTR_ERR(em); | |
d7781546 QW |
2397 | |
2398 | /* Hole or vacuum extent(only exists in no-hole mode) */ | |
2399 | if (em->block_start == EXTENT_MAP_HOLE) { | |
2400 | ret = 1; | |
2401 | *len = em->start + em->len > *start + *len ? | |
2402 | 0 : *start + *len - em->start - em->len; | |
2403 | *start = em->start + em->len; | |
2404 | } | |
2405 | free_extent_map(em); | |
2406 | return ret; | |
2407 | } | |
2408 | ||
f27451f2 FM |
2409 | static int btrfs_punch_hole_lock_range(struct inode *inode, |
2410 | const u64 lockstart, | |
2411 | const u64 lockend, | |
2412 | struct extent_state **cached_state) | |
2413 | { | |
2414 | while (1) { | |
2415 | struct btrfs_ordered_extent *ordered; | |
2416 | int ret; | |
2417 | ||
2418 | truncate_pagecache_range(inode, lockstart, lockend); | |
2419 | ||
2420 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
2421 | cached_state); | |
2422 | ordered = btrfs_lookup_first_ordered_extent(inode, lockend); | |
2423 | ||
2424 | /* | |
2425 | * We need to make sure we have no ordered extents in this range | |
2426 | * and nobody raced in and read a page in this range, if we did | |
2427 | * we need to try again. | |
2428 | */ | |
2429 | if ((!ordered || | |
2430 | (ordered->file_offset + ordered->len <= lockstart || | |
2431 | ordered->file_offset > lockend)) && | |
051c98eb DS |
2432 | !filemap_range_has_page(inode->i_mapping, |
2433 | lockstart, lockend)) { | |
f27451f2 FM |
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); | |
2442 | ret = btrfs_wait_ordered_range(inode, lockstart, | |
2443 | lockend - lockstart + 1); | |
2444 | if (ret) | |
2445 | return ret; | |
2446 | } | |
2447 | return 0; | |
2448 | } | |
2449 | ||
2aaa6655 JB |
2450 | static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len) |
2451 | { | |
0b246afa | 2452 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2aaa6655 JB |
2453 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2454 | struct extent_state *cached_state = NULL; | |
2455 | struct btrfs_path *path; | |
2456 | struct btrfs_block_rsv *rsv; | |
2457 | struct btrfs_trans_handle *trans; | |
d7781546 QW |
2458 | u64 lockstart; |
2459 | u64 lockend; | |
2460 | u64 tail_start; | |
2461 | u64 tail_len; | |
2462 | u64 orig_start = offset; | |
2463 | u64 cur_offset; | |
5f52a2c5 | 2464 | u64 min_size = btrfs_calc_trans_metadata_size(fs_info, 1); |
2aaa6655 | 2465 | u64 drop_end; |
2aaa6655 JB |
2466 | int ret = 0; |
2467 | int err = 0; | |
6e4d6fa1 | 2468 | unsigned int rsv_count; |
9703fefe | 2469 | bool same_block; |
0b246afa | 2470 | bool no_holes = btrfs_fs_incompat(fs_info, NO_HOLES); |
a1a50f60 | 2471 | u64 ino_size; |
9703fefe | 2472 | bool truncated_block = false; |
e8c1c76e | 2473 | bool updated_inode = false; |
2aaa6655 | 2474 | |
0ef8b726 JB |
2475 | ret = btrfs_wait_ordered_range(inode, offset, len); |
2476 | if (ret) | |
2477 | return ret; | |
2aaa6655 | 2478 | |
5955102c | 2479 | inode_lock(inode); |
0b246afa | 2480 | ino_size = round_up(inode->i_size, fs_info->sectorsize); |
d7781546 QW |
2481 | ret = find_first_non_hole(inode, &offset, &len); |
2482 | if (ret < 0) | |
2483 | goto out_only_mutex; | |
2484 | if (ret && !len) { | |
2485 | /* Already in a large hole */ | |
2486 | ret = 0; | |
2487 | goto out_only_mutex; | |
2488 | } | |
2489 | ||
da17066c | 2490 | lockstart = round_up(offset, btrfs_inode_sectorsize(inode)); |
d7781546 | 2491 | lockend = round_down(offset + len, |
da17066c | 2492 | btrfs_inode_sectorsize(inode)) - 1; |
0b246afa JM |
2493 | same_block = (BTRFS_BYTES_TO_BLKS(fs_info, offset)) |
2494 | == (BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)); | |
7426cc04 | 2495 | /* |
9703fefe | 2496 | * We needn't truncate any block which is beyond the end of the file |
7426cc04 MX |
2497 | * because we are sure there is no data there. |
2498 | */ | |
2aaa6655 | 2499 | /* |
9703fefe CR |
2500 | * Only do this if we are in the same block and we aren't doing the |
2501 | * entire block. | |
2aaa6655 | 2502 | */ |
0b246afa | 2503 | if (same_block && len < fs_info->sectorsize) { |
e8c1c76e | 2504 | if (offset < ino_size) { |
9703fefe CR |
2505 | truncated_block = true; |
2506 | ret = btrfs_truncate_block(inode, offset, len, 0); | |
e8c1c76e FM |
2507 | } else { |
2508 | ret = 0; | |
2509 | } | |
d7781546 | 2510 | goto out_only_mutex; |
2aaa6655 JB |
2511 | } |
2512 | ||
9703fefe | 2513 | /* zero back part of the first block */ |
12870f1c | 2514 | if (offset < ino_size) { |
9703fefe CR |
2515 | truncated_block = true; |
2516 | ret = btrfs_truncate_block(inode, offset, 0, 0); | |
7426cc04 | 2517 | if (ret) { |
5955102c | 2518 | inode_unlock(inode); |
7426cc04 MX |
2519 | return ret; |
2520 | } | |
2aaa6655 JB |
2521 | } |
2522 | ||
d7781546 QW |
2523 | /* Check the aligned pages after the first unaligned page, |
2524 | * if offset != orig_start, which means the first unaligned page | |
01327610 | 2525 | * including several following pages are already in holes, |
d7781546 QW |
2526 | * the extra check can be skipped */ |
2527 | if (offset == orig_start) { | |
2528 | /* after truncate page, check hole again */ | |
2529 | len = offset + len - lockstart; | |
2530 | offset = lockstart; | |
2531 | ret = find_first_non_hole(inode, &offset, &len); | |
2532 | if (ret < 0) | |
2533 | goto out_only_mutex; | |
2534 | if (ret && !len) { | |
2535 | ret = 0; | |
2536 | goto out_only_mutex; | |
2537 | } | |
2538 | lockstart = offset; | |
2539 | } | |
2540 | ||
2541 | /* Check the tail unaligned part is in a hole */ | |
2542 | tail_start = lockend + 1; | |
2543 | tail_len = offset + len - tail_start; | |
2544 | if (tail_len) { | |
2545 | ret = find_first_non_hole(inode, &tail_start, &tail_len); | |
2546 | if (unlikely(ret < 0)) | |
2547 | goto out_only_mutex; | |
2548 | if (!ret) { | |
2549 | /* zero the front end of the last page */ | |
2550 | if (tail_start + tail_len < ino_size) { | |
9703fefe CR |
2551 | truncated_block = true; |
2552 | ret = btrfs_truncate_block(inode, | |
2553 | tail_start + tail_len, | |
2554 | 0, 1); | |
d7781546 QW |
2555 | if (ret) |
2556 | goto out_only_mutex; | |
51f395ad | 2557 | } |
0061280d | 2558 | } |
2aaa6655 JB |
2559 | } |
2560 | ||
2561 | if (lockend < lockstart) { | |
e8c1c76e FM |
2562 | ret = 0; |
2563 | goto out_only_mutex; | |
2aaa6655 JB |
2564 | } |
2565 | ||
f27451f2 FM |
2566 | ret = btrfs_punch_hole_lock_range(inode, lockstart, lockend, |
2567 | &cached_state); | |
8fca9550 | 2568 | if (ret) |
f27451f2 | 2569 | goto out_only_mutex; |
2aaa6655 JB |
2570 | |
2571 | path = btrfs_alloc_path(); | |
2572 | if (!path) { | |
2573 | ret = -ENOMEM; | |
2574 | goto out; | |
2575 | } | |
2576 | ||
2ff7e61e | 2577 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
2aaa6655 JB |
2578 | if (!rsv) { |
2579 | ret = -ENOMEM; | |
2580 | goto out_free; | |
2581 | } | |
5f52a2c5 | 2582 | rsv->size = btrfs_calc_trans_metadata_size(fs_info, 1); |
2aaa6655 JB |
2583 | rsv->failfast = 1; |
2584 | ||
2585 | /* | |
2586 | * 1 - update the inode | |
2587 | * 1 - removing the extents in the range | |
16e7549f | 2588 | * 1 - adding the hole extent if no_holes isn't set |
2aaa6655 | 2589 | */ |
16e7549f JB |
2590 | rsv_count = no_holes ? 2 : 3; |
2591 | trans = btrfs_start_transaction(root, rsv_count); | |
2aaa6655 JB |
2592 | if (IS_ERR(trans)) { |
2593 | err = PTR_ERR(trans); | |
2594 | goto out_free; | |
2595 | } | |
2596 | ||
0b246afa | 2597 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv, |
3a584174 | 2598 | min_size, false); |
2aaa6655 JB |
2599 | BUG_ON(ret); |
2600 | trans->block_rsv = rsv; | |
2601 | ||
d7781546 QW |
2602 | cur_offset = lockstart; |
2603 | len = lockend - cur_offset; | |
2aaa6655 JB |
2604 | while (cur_offset < lockend) { |
2605 | ret = __btrfs_drop_extents(trans, root, inode, path, | |
2606 | cur_offset, lockend + 1, | |
1acae57b | 2607 | &drop_end, 1, 0, 0, NULL); |
2aaa6655 JB |
2608 | if (ret != -ENOSPC) |
2609 | break; | |
2610 | ||
0b246afa | 2611 | trans->block_rsv = &fs_info->trans_block_rsv; |
2aaa6655 | 2612 | |
62fe51c1 | 2613 | if (cur_offset < drop_end && cur_offset < ino_size) { |
a012a74e NB |
2614 | ret = fill_holes(trans, BTRFS_I(inode), path, |
2615 | cur_offset, drop_end); | |
12870f1c | 2616 | if (ret) { |
f94480bd JB |
2617 | /* |
2618 | * If we failed then we didn't insert our hole | |
2619 | * entries for the area we dropped, so now the | |
2620 | * fs is corrupted, so we must abort the | |
2621 | * transaction. | |
2622 | */ | |
2623 | btrfs_abort_transaction(trans, ret); | |
12870f1c FM |
2624 | err = ret; |
2625 | break; | |
2626 | } | |
2aaa6655 JB |
2627 | } |
2628 | ||
2629 | cur_offset = drop_end; | |
2630 | ||
2631 | ret = btrfs_update_inode(trans, root, inode); | |
2632 | if (ret) { | |
2633 | err = ret; | |
2634 | break; | |
2635 | } | |
2636 | ||
3a45bb20 | 2637 | btrfs_end_transaction(trans); |
2ff7e61e | 2638 | btrfs_btree_balance_dirty(fs_info); |
2aaa6655 | 2639 | |
16e7549f | 2640 | trans = btrfs_start_transaction(root, rsv_count); |
2aaa6655 JB |
2641 | if (IS_ERR(trans)) { |
2642 | ret = PTR_ERR(trans); | |
2643 | trans = NULL; | |
2644 | break; | |
2645 | } | |
2646 | ||
0b246afa | 2647 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, |
3a584174 | 2648 | rsv, min_size, false); |
2aaa6655 JB |
2649 | BUG_ON(ret); /* shouldn't happen */ |
2650 | trans->block_rsv = rsv; | |
d7781546 QW |
2651 | |
2652 | ret = find_first_non_hole(inode, &cur_offset, &len); | |
2653 | if (unlikely(ret < 0)) | |
2654 | break; | |
2655 | if (ret && !len) { | |
2656 | ret = 0; | |
2657 | break; | |
2658 | } | |
2aaa6655 JB |
2659 | } |
2660 | ||
2661 | if (ret) { | |
2662 | err = ret; | |
2663 | goto out_trans; | |
2664 | } | |
2665 | ||
0b246afa | 2666 | trans->block_rsv = &fs_info->trans_block_rsv; |
2959a32a FM |
2667 | /* |
2668 | * If we are using the NO_HOLES feature we might have had already an | |
2669 | * hole that overlaps a part of the region [lockstart, lockend] and | |
2670 | * ends at (or beyond) lockend. Since we have no file extent items to | |
2671 | * represent holes, drop_end can be less than lockend and so we must | |
2672 | * make sure we have an extent map representing the existing hole (the | |
2673 | * call to __btrfs_drop_extents() might have dropped the existing extent | |
2674 | * map representing the existing hole), otherwise the fast fsync path | |
2675 | * will not record the existence of the hole region | |
2676 | * [existing_hole_start, lockend]. | |
2677 | */ | |
2678 | if (drop_end <= lockend) | |
2679 | drop_end = lockend + 1; | |
fc19c5e7 FM |
2680 | /* |
2681 | * Don't insert file hole extent item if it's for a range beyond eof | |
2682 | * (because it's useless) or if it represents a 0 bytes range (when | |
2683 | * cur_offset == drop_end). | |
2684 | */ | |
2685 | if (cur_offset < ino_size && cur_offset < drop_end) { | |
a012a74e NB |
2686 | ret = fill_holes(trans, BTRFS_I(inode), path, |
2687 | cur_offset, drop_end); | |
12870f1c | 2688 | if (ret) { |
f94480bd JB |
2689 | /* Same comment as above. */ |
2690 | btrfs_abort_transaction(trans, ret); | |
12870f1c FM |
2691 | err = ret; |
2692 | goto out_trans; | |
2693 | } | |
2aaa6655 JB |
2694 | } |
2695 | ||
2696 | out_trans: | |
2697 | if (!trans) | |
2698 | goto out_free; | |
2699 | ||
e1f5790e | 2700 | inode_inc_iversion(inode); |
c2050a45 | 2701 | inode->i_mtime = inode->i_ctime = current_time(inode); |
e1f5790e | 2702 | |
0b246afa | 2703 | trans->block_rsv = &fs_info->trans_block_rsv; |
2aaa6655 | 2704 | ret = btrfs_update_inode(trans, root, inode); |
e8c1c76e | 2705 | updated_inode = true; |
3a45bb20 | 2706 | btrfs_end_transaction(trans); |
2ff7e61e | 2707 | btrfs_btree_balance_dirty(fs_info); |
2aaa6655 JB |
2708 | out_free: |
2709 | btrfs_free_path(path); | |
2ff7e61e | 2710 | btrfs_free_block_rsv(fs_info, rsv); |
2aaa6655 JB |
2711 | out: |
2712 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
e43bbe5e | 2713 | &cached_state); |
d7781546 | 2714 | out_only_mutex: |
9703fefe | 2715 | if (!updated_inode && truncated_block && !ret && !err) { |
e8c1c76e FM |
2716 | /* |
2717 | * If we only end up zeroing part of a page, we still need to | |
2718 | * update the inode item, so that all the time fields are | |
2719 | * updated as well as the necessary btrfs inode in memory fields | |
2720 | * for detecting, at fsync time, if the inode isn't yet in the | |
2721 | * log tree or it's there but not up to date. | |
2722 | */ | |
17900668 FM |
2723 | struct timespec64 now = current_time(inode); |
2724 | ||
2725 | inode_inc_iversion(inode); | |
2726 | inode->i_mtime = now; | |
2727 | inode->i_ctime = now; | |
e8c1c76e FM |
2728 | trans = btrfs_start_transaction(root, 1); |
2729 | if (IS_ERR(trans)) { | |
2730 | err = PTR_ERR(trans); | |
2731 | } else { | |
2732 | err = btrfs_update_inode(trans, root, inode); | |
3a45bb20 | 2733 | ret = btrfs_end_transaction(trans); |
e8c1c76e FM |
2734 | } |
2735 | } | |
5955102c | 2736 | inode_unlock(inode); |
2aaa6655 JB |
2737 | if (ret && !err) |
2738 | err = ret; | |
2739 | return err; | |
2740 | } | |
2741 | ||
14524a84 QW |
2742 | /* Helper structure to record which range is already reserved */ |
2743 | struct falloc_range { | |
2744 | struct list_head list; | |
2745 | u64 start; | |
2746 | u64 len; | |
2747 | }; | |
2748 | ||
2749 | /* | |
2750 | * Helper function to add falloc range | |
2751 | * | |
2752 | * Caller should have locked the larger range of extent containing | |
2753 | * [start, len) | |
2754 | */ | |
2755 | static int add_falloc_range(struct list_head *head, u64 start, u64 len) | |
2756 | { | |
2757 | struct falloc_range *prev = NULL; | |
2758 | struct falloc_range *range = NULL; | |
2759 | ||
2760 | if (list_empty(head)) | |
2761 | goto insert; | |
2762 | ||
2763 | /* | |
2764 | * As fallocate iterate by bytenr order, we only need to check | |
2765 | * the last range. | |
2766 | */ | |
2767 | prev = list_entry(head->prev, struct falloc_range, list); | |
2768 | if (prev->start + prev->len == start) { | |
2769 | prev->len += len; | |
2770 | return 0; | |
2771 | } | |
2772 | insert: | |
32fc932e | 2773 | range = kmalloc(sizeof(*range), GFP_KERNEL); |
14524a84 QW |
2774 | if (!range) |
2775 | return -ENOMEM; | |
2776 | range->start = start; | |
2777 | range->len = len; | |
2778 | list_add_tail(&range->list, head); | |
2779 | return 0; | |
2780 | } | |
2781 | ||
f27451f2 FM |
2782 | static int btrfs_fallocate_update_isize(struct inode *inode, |
2783 | const u64 end, | |
2784 | const int mode) | |
2785 | { | |
2786 | struct btrfs_trans_handle *trans; | |
2787 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2788 | int ret; | |
2789 | int ret2; | |
2790 | ||
2791 | if (mode & FALLOC_FL_KEEP_SIZE || end <= i_size_read(inode)) | |
2792 | return 0; | |
2793 | ||
2794 | trans = btrfs_start_transaction(root, 1); | |
2795 | if (IS_ERR(trans)) | |
2796 | return PTR_ERR(trans); | |
2797 | ||
2798 | inode->i_ctime = current_time(inode); | |
2799 | i_size_write(inode, end); | |
2800 | btrfs_ordered_update_i_size(inode, end, NULL); | |
2801 | ret = btrfs_update_inode(trans, root, inode); | |
2802 | ret2 = btrfs_end_transaction(trans); | |
2803 | ||
2804 | return ret ? ret : ret2; | |
2805 | } | |
2806 | ||
81fdf638 | 2807 | enum { |
f262fa8d DS |
2808 | RANGE_BOUNDARY_WRITTEN_EXTENT, |
2809 | RANGE_BOUNDARY_PREALLOC_EXTENT, | |
2810 | RANGE_BOUNDARY_HOLE, | |
81fdf638 FM |
2811 | }; |
2812 | ||
f27451f2 FM |
2813 | static int btrfs_zero_range_check_range_boundary(struct inode *inode, |
2814 | u64 offset) | |
2815 | { | |
2816 | const u64 sectorsize = btrfs_inode_sectorsize(inode); | |
2817 | struct extent_map *em; | |
81fdf638 | 2818 | int ret; |
f27451f2 FM |
2819 | |
2820 | offset = round_down(offset, sectorsize); | |
2821 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, offset, sectorsize, 0); | |
2822 | if (IS_ERR(em)) | |
2823 | return PTR_ERR(em); | |
2824 | ||
2825 | if (em->block_start == EXTENT_MAP_HOLE) | |
81fdf638 FM |
2826 | ret = RANGE_BOUNDARY_HOLE; |
2827 | else if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
2828 | ret = RANGE_BOUNDARY_PREALLOC_EXTENT; | |
2829 | else | |
2830 | ret = RANGE_BOUNDARY_WRITTEN_EXTENT; | |
f27451f2 FM |
2831 | |
2832 | free_extent_map(em); | |
2833 | return ret; | |
2834 | } | |
2835 | ||
2836 | static int btrfs_zero_range(struct inode *inode, | |
2837 | loff_t offset, | |
2838 | loff_t len, | |
2839 | const int mode) | |
2840 | { | |
2841 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2842 | struct extent_map *em; | |
2843 | struct extent_changeset *data_reserved = NULL; | |
2844 | int ret; | |
2845 | u64 alloc_hint = 0; | |
2846 | const u64 sectorsize = btrfs_inode_sectorsize(inode); | |
2847 | u64 alloc_start = round_down(offset, sectorsize); | |
2848 | u64 alloc_end = round_up(offset + len, sectorsize); | |
2849 | u64 bytes_to_reserve = 0; | |
2850 | bool space_reserved = false; | |
2851 | ||
2852 | inode_dio_wait(inode); | |
2853 | ||
2854 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, | |
2855 | alloc_start, alloc_end - alloc_start, 0); | |
2856 | if (IS_ERR(em)) { | |
2857 | ret = PTR_ERR(em); | |
2858 | goto out; | |
2859 | } | |
2860 | ||
2861 | /* | |
2862 | * Avoid hole punching and extent allocation for some cases. More cases | |
2863 | * could be considered, but these are unlikely common and we keep things | |
2864 | * as simple as possible for now. Also, intentionally, if the target | |
2865 | * range contains one or more prealloc extents together with regular | |
2866 | * extents and holes, we drop all the existing extents and allocate a | |
2867 | * new prealloc extent, so that we get a larger contiguous disk extent. | |
2868 | */ | |
2869 | if (em->start <= alloc_start && | |
2870 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { | |
2871 | const u64 em_end = em->start + em->len; | |
2872 | ||
2873 | if (em_end >= offset + len) { | |
2874 | /* | |
2875 | * The whole range is already a prealloc extent, | |
2876 | * do nothing except updating the inode's i_size if | |
2877 | * needed. | |
2878 | */ | |
2879 | free_extent_map(em); | |
2880 | ret = btrfs_fallocate_update_isize(inode, offset + len, | |
2881 | mode); | |
2882 | goto out; | |
2883 | } | |
2884 | /* | |
2885 | * Part of the range is already a prealloc extent, so operate | |
2886 | * only on the remaining part of the range. | |
2887 | */ | |
2888 | alloc_start = em_end; | |
2889 | ASSERT(IS_ALIGNED(alloc_start, sectorsize)); | |
2890 | len = offset + len - alloc_start; | |
2891 | offset = alloc_start; | |
2892 | alloc_hint = em->block_start + em->len; | |
2893 | } | |
2894 | free_extent_map(em); | |
2895 | ||
2896 | if (BTRFS_BYTES_TO_BLKS(fs_info, offset) == | |
2897 | BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)) { | |
2898 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, | |
2899 | alloc_start, sectorsize, 0); | |
2900 | if (IS_ERR(em)) { | |
2901 | ret = PTR_ERR(em); | |
2902 | goto out; | |
2903 | } | |
2904 | ||
2905 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { | |
2906 | free_extent_map(em); | |
2907 | ret = btrfs_fallocate_update_isize(inode, offset + len, | |
2908 | mode); | |
2909 | goto out; | |
2910 | } | |
2911 | if (len < sectorsize && em->block_start != EXTENT_MAP_HOLE) { | |
2912 | free_extent_map(em); | |
2913 | ret = btrfs_truncate_block(inode, offset, len, 0); | |
2914 | if (!ret) | |
2915 | ret = btrfs_fallocate_update_isize(inode, | |
2916 | offset + len, | |
2917 | mode); | |
2918 | return ret; | |
2919 | } | |
2920 | free_extent_map(em); | |
2921 | alloc_start = round_down(offset, sectorsize); | |
2922 | alloc_end = alloc_start + sectorsize; | |
2923 | goto reserve_space; | |
2924 | } | |
2925 | ||
2926 | alloc_start = round_up(offset, sectorsize); | |
2927 | alloc_end = round_down(offset + len, sectorsize); | |
2928 | ||
2929 | /* | |
2930 | * For unaligned ranges, check the pages at the boundaries, they might | |
2931 | * map to an extent, in which case we need to partially zero them, or | |
2932 | * they might map to a hole, in which case we need our allocation range | |
2933 | * to cover them. | |
2934 | */ | |
2935 | if (!IS_ALIGNED(offset, sectorsize)) { | |
2936 | ret = btrfs_zero_range_check_range_boundary(inode, offset); | |
2937 | if (ret < 0) | |
2938 | goto out; | |
81fdf638 | 2939 | if (ret == RANGE_BOUNDARY_HOLE) { |
f27451f2 FM |
2940 | alloc_start = round_down(offset, sectorsize); |
2941 | ret = 0; | |
81fdf638 | 2942 | } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) { |
f27451f2 FM |
2943 | ret = btrfs_truncate_block(inode, offset, 0, 0); |
2944 | if (ret) | |
2945 | goto out; | |
81fdf638 FM |
2946 | } else { |
2947 | ret = 0; | |
f27451f2 FM |
2948 | } |
2949 | } | |
2950 | ||
2951 | if (!IS_ALIGNED(offset + len, sectorsize)) { | |
2952 | ret = btrfs_zero_range_check_range_boundary(inode, | |
2953 | offset + len); | |
2954 | if (ret < 0) | |
2955 | goto out; | |
81fdf638 | 2956 | if (ret == RANGE_BOUNDARY_HOLE) { |
f27451f2 FM |
2957 | alloc_end = round_up(offset + len, sectorsize); |
2958 | ret = 0; | |
81fdf638 | 2959 | } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) { |
f27451f2 FM |
2960 | ret = btrfs_truncate_block(inode, offset + len, 0, 1); |
2961 | if (ret) | |
2962 | goto out; | |
81fdf638 FM |
2963 | } else { |
2964 | ret = 0; | |
f27451f2 FM |
2965 | } |
2966 | } | |
2967 | ||
2968 | reserve_space: | |
2969 | if (alloc_start < alloc_end) { | |
2970 | struct extent_state *cached_state = NULL; | |
2971 | const u64 lockstart = alloc_start; | |
2972 | const u64 lockend = alloc_end - 1; | |
2973 | ||
2974 | bytes_to_reserve = alloc_end - alloc_start; | |
2975 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), | |
2976 | bytes_to_reserve); | |
2977 | if (ret < 0) | |
2978 | goto out; | |
2979 | space_reserved = true; | |
2980 | ret = btrfs_qgroup_reserve_data(inode, &data_reserved, | |
2981 | alloc_start, bytes_to_reserve); | |
2982 | if (ret) | |
2983 | goto out; | |
2984 | ret = btrfs_punch_hole_lock_range(inode, lockstart, lockend, | |
2985 | &cached_state); | |
2986 | if (ret) | |
2987 | goto out; | |
2988 | ret = btrfs_prealloc_file_range(inode, mode, alloc_start, | |
2989 | alloc_end - alloc_start, | |
2990 | i_blocksize(inode), | |
2991 | offset + len, &alloc_hint); | |
2992 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, | |
2993 | lockend, &cached_state); | |
2994 | /* btrfs_prealloc_file_range releases reserved space on error */ | |
9f13ce74 | 2995 | if (ret) { |
f27451f2 | 2996 | space_reserved = false; |
9f13ce74 FM |
2997 | goto out; |
2998 | } | |
f27451f2 | 2999 | } |
9f13ce74 | 3000 | ret = btrfs_fallocate_update_isize(inode, offset + len, mode); |
f27451f2 FM |
3001 | out: |
3002 | if (ret && space_reserved) | |
3003 | btrfs_free_reserved_data_space(inode, data_reserved, | |
3004 | alloc_start, bytes_to_reserve); | |
3005 | extent_changeset_free(data_reserved); | |
3006 | ||
3007 | return ret; | |
3008 | } | |
3009 | ||
2fe17c10 CH |
3010 | static long btrfs_fallocate(struct file *file, int mode, |
3011 | loff_t offset, loff_t len) | |
3012 | { | |
496ad9aa | 3013 | struct inode *inode = file_inode(file); |
2fe17c10 | 3014 | struct extent_state *cached_state = NULL; |
364ecf36 | 3015 | struct extent_changeset *data_reserved = NULL; |
14524a84 QW |
3016 | struct falloc_range *range; |
3017 | struct falloc_range *tmp; | |
3018 | struct list_head reserve_list; | |
2fe17c10 CH |
3019 | u64 cur_offset; |
3020 | u64 last_byte; | |
3021 | u64 alloc_start; | |
3022 | u64 alloc_end; | |
3023 | u64 alloc_hint = 0; | |
3024 | u64 locked_end; | |
14524a84 | 3025 | u64 actual_end = 0; |
2fe17c10 | 3026 | struct extent_map *em; |
da17066c | 3027 | int blocksize = btrfs_inode_sectorsize(inode); |
2fe17c10 CH |
3028 | int ret; |
3029 | ||
797f4277 MX |
3030 | alloc_start = round_down(offset, blocksize); |
3031 | alloc_end = round_up(offset + len, blocksize); | |
18513091 | 3032 | cur_offset = alloc_start; |
2fe17c10 | 3033 | |
2aaa6655 | 3034 | /* Make sure we aren't being give some crap mode */ |
f27451f2 FM |
3035 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | |
3036 | FALLOC_FL_ZERO_RANGE)) | |
2fe17c10 CH |
3037 | return -EOPNOTSUPP; |
3038 | ||
2aaa6655 JB |
3039 | if (mode & FALLOC_FL_PUNCH_HOLE) |
3040 | return btrfs_punch_hole(inode, offset, len); | |
3041 | ||
d98456fc | 3042 | /* |
14524a84 QW |
3043 | * Only trigger disk allocation, don't trigger qgroup reserve |
3044 | * | |
3045 | * For qgroup space, it will be checked later. | |
d98456fc | 3046 | */ |
f27451f2 FM |
3047 | if (!(mode & FALLOC_FL_ZERO_RANGE)) { |
3048 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), | |
3049 | alloc_end - alloc_start); | |
3050 | if (ret < 0) | |
3051 | return ret; | |
3052 | } | |
d98456fc | 3053 | |
5955102c | 3054 | inode_lock(inode); |
2a162ce9 DI |
3055 | |
3056 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) { | |
3057 | ret = inode_newsize_ok(inode, offset + len); | |
3058 | if (ret) | |
3059 | goto out; | |
3060 | } | |
2fe17c10 | 3061 | |
14524a84 QW |
3062 | /* |
3063 | * TODO: Move these two operations after we have checked | |
3064 | * accurate reserved space, or fallocate can still fail but | |
3065 | * with page truncated or size expanded. | |
3066 | * | |
3067 | * But that's a minor problem and won't do much harm BTW. | |
3068 | */ | |
2fe17c10 | 3069 | if (alloc_start > inode->i_size) { |
a41ad394 JB |
3070 | ret = btrfs_cont_expand(inode, i_size_read(inode), |
3071 | alloc_start); | |
2fe17c10 CH |
3072 | if (ret) |
3073 | goto out; | |
0f6925fa | 3074 | } else if (offset + len > inode->i_size) { |
a71754fc JB |
3075 | /* |
3076 | * If we are fallocating from the end of the file onward we | |
9703fefe CR |
3077 | * need to zero out the end of the block if i_size lands in the |
3078 | * middle of a block. | |
a71754fc | 3079 | */ |
9703fefe | 3080 | ret = btrfs_truncate_block(inode, inode->i_size, 0, 0); |
a71754fc JB |
3081 | if (ret) |
3082 | goto out; | |
2fe17c10 CH |
3083 | } |
3084 | ||
a71754fc JB |
3085 | /* |
3086 | * wait for ordered IO before we have any locks. We'll loop again | |
3087 | * below with the locks held. | |
3088 | */ | |
0ef8b726 JB |
3089 | ret = btrfs_wait_ordered_range(inode, alloc_start, |
3090 | alloc_end - alloc_start); | |
3091 | if (ret) | |
3092 | goto out; | |
a71754fc | 3093 | |
f27451f2 FM |
3094 | if (mode & FALLOC_FL_ZERO_RANGE) { |
3095 | ret = btrfs_zero_range(inode, offset, len, mode); | |
3096 | inode_unlock(inode); | |
3097 | return ret; | |
3098 | } | |
3099 | ||
2fe17c10 CH |
3100 | locked_end = alloc_end - 1; |
3101 | while (1) { | |
3102 | struct btrfs_ordered_extent *ordered; | |
3103 | ||
3104 | /* the extent lock is ordered inside the running | |
3105 | * transaction | |
3106 | */ | |
3107 | lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start, | |
ff13db41 | 3108 | locked_end, &cached_state); |
96b09dde NB |
3109 | ordered = btrfs_lookup_first_ordered_extent(inode, locked_end); |
3110 | ||
2fe17c10 CH |
3111 | if (ordered && |
3112 | ordered->file_offset + ordered->len > alloc_start && | |
3113 | ordered->file_offset < alloc_end) { | |
3114 | btrfs_put_ordered_extent(ordered); | |
3115 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, | |
3116 | alloc_start, locked_end, | |
e43bbe5e | 3117 | &cached_state); |
2fe17c10 CH |
3118 | /* |
3119 | * we can't wait on the range with the transaction | |
3120 | * running or with the extent lock held | |
3121 | */ | |
0ef8b726 JB |
3122 | ret = btrfs_wait_ordered_range(inode, alloc_start, |
3123 | alloc_end - alloc_start); | |
3124 | if (ret) | |
3125 | goto out; | |
2fe17c10 CH |
3126 | } else { |
3127 | if (ordered) | |
3128 | btrfs_put_ordered_extent(ordered); | |
3129 | break; | |
3130 | } | |
3131 | } | |
3132 | ||
14524a84 QW |
3133 | /* First, check if we exceed the qgroup limit */ |
3134 | INIT_LIST_HEAD(&reserve_list); | |
6b7d6e93 | 3135 | while (cur_offset < alloc_end) { |
fc4f21b1 | 3136 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset, |
2fe17c10 | 3137 | alloc_end - cur_offset, 0); |
9986277e DC |
3138 | if (IS_ERR(em)) { |
3139 | ret = PTR_ERR(em); | |
79787eaa JM |
3140 | break; |
3141 | } | |
2fe17c10 | 3142 | last_byte = min(extent_map_end(em), alloc_end); |
f1e490a7 | 3143 | actual_end = min_t(u64, extent_map_end(em), offset + len); |
797f4277 | 3144 | last_byte = ALIGN(last_byte, blocksize); |
2fe17c10 CH |
3145 | if (em->block_start == EXTENT_MAP_HOLE || |
3146 | (cur_offset >= inode->i_size && | |
3147 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
14524a84 QW |
3148 | ret = add_falloc_range(&reserve_list, cur_offset, |
3149 | last_byte - cur_offset); | |
3150 | if (ret < 0) { | |
3151 | free_extent_map(em); | |
3152 | break; | |
3d850dd4 | 3153 | } |
364ecf36 QW |
3154 | ret = btrfs_qgroup_reserve_data(inode, &data_reserved, |
3155 | cur_offset, last_byte - cur_offset); | |
be2d253c | 3156 | if (ret < 0) { |
39ad3173 | 3157 | cur_offset = last_byte; |
be2d253c | 3158 | free_extent_map(em); |
14524a84 | 3159 | break; |
be2d253c | 3160 | } |
18513091 WX |
3161 | } else { |
3162 | /* | |
3163 | * Do not need to reserve unwritten extent for this | |
3164 | * range, free reserved data space first, otherwise | |
3165 | * it'll result in false ENOSPC error. | |
3166 | */ | |
bc42bda2 QW |
3167 | btrfs_free_reserved_data_space(inode, data_reserved, |
3168 | cur_offset, last_byte - cur_offset); | |
2fe17c10 CH |
3169 | } |
3170 | free_extent_map(em); | |
2fe17c10 | 3171 | cur_offset = last_byte; |
14524a84 QW |
3172 | } |
3173 | ||
3174 | /* | |
3175 | * If ret is still 0, means we're OK to fallocate. | |
3176 | * Or just cleanup the list and exit. | |
3177 | */ | |
3178 | list_for_each_entry_safe(range, tmp, &reserve_list, list) { | |
3179 | if (!ret) | |
3180 | ret = btrfs_prealloc_file_range(inode, mode, | |
3181 | range->start, | |
93407472 | 3182 | range->len, i_blocksize(inode), |
14524a84 | 3183 | offset + len, &alloc_hint); |
18513091 | 3184 | else |
bc42bda2 QW |
3185 | btrfs_free_reserved_data_space(inode, |
3186 | data_reserved, range->start, | |
3187 | range->len); | |
14524a84 QW |
3188 | list_del(&range->list); |
3189 | kfree(range); | |
3190 | } | |
3191 | if (ret < 0) | |
3192 | goto out_unlock; | |
3193 | ||
f27451f2 FM |
3194 | /* |
3195 | * We didn't need to allocate any more space, but we still extended the | |
3196 | * size of the file so we need to update i_size and the inode item. | |
3197 | */ | |
3198 | ret = btrfs_fallocate_update_isize(inode, actual_end, mode); | |
14524a84 | 3199 | out_unlock: |
2fe17c10 | 3200 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
e43bbe5e | 3201 | &cached_state); |
2fe17c10 | 3202 | out: |
5955102c | 3203 | inode_unlock(inode); |
d98456fc | 3204 | /* Let go of our reservation. */ |
f27451f2 | 3205 | if (ret != 0 && !(mode & FALLOC_FL_ZERO_RANGE)) |
bc42bda2 | 3206 | btrfs_free_reserved_data_space(inode, data_reserved, |
39ad3173 | 3207 | cur_offset, alloc_end - cur_offset); |
364ecf36 | 3208 | extent_changeset_free(data_reserved); |
2fe17c10 CH |
3209 | return ret; |
3210 | } | |
3211 | ||
965c8e59 | 3212 | static int find_desired_extent(struct inode *inode, loff_t *offset, int whence) |
b2675157 | 3213 | { |
0b246afa | 3214 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
7f4ca37c | 3215 | struct extent_map *em = NULL; |
b2675157 | 3216 | struct extent_state *cached_state = NULL; |
4d1a40c6 LB |
3217 | u64 lockstart; |
3218 | u64 lockend; | |
3219 | u64 start; | |
3220 | u64 len; | |
b2675157 JB |
3221 | int ret = 0; |
3222 | ||
4d1a40c6 LB |
3223 | if (inode->i_size == 0) |
3224 | return -ENXIO; | |
3225 | ||
3226 | /* | |
3227 | * *offset can be negative, in this case we start finding DATA/HOLE from | |
3228 | * the very start of the file. | |
3229 | */ | |
3230 | start = max_t(loff_t, 0, *offset); | |
3231 | ||
0b246afa | 3232 | lockstart = round_down(start, fs_info->sectorsize); |
da17066c | 3233 | lockend = round_up(i_size_read(inode), |
0b246afa | 3234 | fs_info->sectorsize); |
b2675157 | 3235 | if (lockend <= lockstart) |
0b246afa | 3236 | lockend = lockstart + fs_info->sectorsize; |
1214b53f | 3237 | lockend--; |
b2675157 JB |
3238 | len = lockend - lockstart + 1; |
3239 | ||
ff13db41 | 3240 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
d0082371 | 3241 | &cached_state); |
b2675157 | 3242 | |
7f4ca37c | 3243 | while (start < inode->i_size) { |
4ab47a8d | 3244 | em = btrfs_get_extent_fiemap(BTRFS_I(inode), start, len); |
b2675157 | 3245 | if (IS_ERR(em)) { |
6af021d8 | 3246 | ret = PTR_ERR(em); |
7f4ca37c | 3247 | em = NULL; |
b2675157 JB |
3248 | break; |
3249 | } | |
3250 | ||
7f4ca37c JB |
3251 | if (whence == SEEK_HOLE && |
3252 | (em->block_start == EXTENT_MAP_HOLE || | |
3253 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) | |
3254 | break; | |
3255 | else if (whence == SEEK_DATA && | |
3256 | (em->block_start != EXTENT_MAP_HOLE && | |
3257 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) | |
3258 | break; | |
b2675157 JB |
3259 | |
3260 | start = em->start + em->len; | |
b2675157 | 3261 | free_extent_map(em); |
7f4ca37c | 3262 | em = NULL; |
b2675157 JB |
3263 | cond_resched(); |
3264 | } | |
7f4ca37c JB |
3265 | free_extent_map(em); |
3266 | if (!ret) { | |
3267 | if (whence == SEEK_DATA && start >= inode->i_size) | |
3268 | ret = -ENXIO; | |
3269 | else | |
3270 | *offset = min_t(loff_t, start, inode->i_size); | |
3271 | } | |
b2675157 | 3272 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
e43bbe5e | 3273 | &cached_state); |
b2675157 JB |
3274 | return ret; |
3275 | } | |
3276 | ||
965c8e59 | 3277 | static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence) |
b2675157 JB |
3278 | { |
3279 | struct inode *inode = file->f_mapping->host; | |
3280 | int ret; | |
3281 | ||
5955102c | 3282 | inode_lock(inode); |
965c8e59 | 3283 | switch (whence) { |
b2675157 JB |
3284 | case SEEK_END: |
3285 | case SEEK_CUR: | |
965c8e59 | 3286 | offset = generic_file_llseek(file, offset, whence); |
b2675157 JB |
3287 | goto out; |
3288 | case SEEK_DATA: | |
3289 | case SEEK_HOLE: | |
48802c8a | 3290 | if (offset >= i_size_read(inode)) { |
5955102c | 3291 | inode_unlock(inode); |
48802c8a JL |
3292 | return -ENXIO; |
3293 | } | |
3294 | ||
965c8e59 | 3295 | ret = find_desired_extent(inode, &offset, whence); |
b2675157 | 3296 | if (ret) { |
5955102c | 3297 | inode_unlock(inode); |
b2675157 JB |
3298 | return ret; |
3299 | } | |
3300 | } | |
3301 | ||
46a1c2c7 | 3302 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
b2675157 | 3303 | out: |
5955102c | 3304 | inode_unlock(inode); |
b2675157 JB |
3305 | return offset; |
3306 | } | |
3307 | ||
edf064e7 GR |
3308 | static int btrfs_file_open(struct inode *inode, struct file *filp) |
3309 | { | |
91f9943e | 3310 | filp->f_mode |= FMODE_NOWAIT; |
edf064e7 GR |
3311 | return generic_file_open(inode, filp); |
3312 | } | |
3313 | ||
828c0950 | 3314 | const struct file_operations btrfs_file_operations = { |
b2675157 | 3315 | .llseek = btrfs_file_llseek, |
aad4f8bb | 3316 | .read_iter = generic_file_read_iter, |
e9906a98 | 3317 | .splice_read = generic_file_splice_read, |
b30ac0fc | 3318 | .write_iter = btrfs_file_write_iter, |
9ebefb18 | 3319 | .mmap = btrfs_file_mmap, |
edf064e7 | 3320 | .open = btrfs_file_open, |
e1b81e67 | 3321 | .release = btrfs_release_file, |
39279cc3 | 3322 | .fsync = btrfs_sync_file, |
2fe17c10 | 3323 | .fallocate = btrfs_fallocate, |
34287aa3 | 3324 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 3325 | #ifdef CONFIG_COMPAT |
4c63c245 | 3326 | .compat_ioctl = btrfs_compat_ioctl, |
39279cc3 | 3327 | #endif |
2e5dfc99 | 3328 | .remap_file_range = btrfs_remap_file_range, |
39279cc3 | 3329 | }; |
9247f317 | 3330 | |
e67c718b | 3331 | void __cold btrfs_auto_defrag_exit(void) |
9247f317 | 3332 | { |
5598e900 | 3333 | kmem_cache_destroy(btrfs_inode_defrag_cachep); |
9247f317 MX |
3334 | } |
3335 | ||
f5c29bd9 | 3336 | int __init btrfs_auto_defrag_init(void) |
9247f317 MX |
3337 | { |
3338 | btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag", | |
3339 | sizeof(struct inode_defrag), 0, | |
fba4b697 | 3340 | SLAB_MEM_SPREAD, |
9247f317 MX |
3341 | NULL); |
3342 | if (!btrfs_inode_defrag_cachep) | |
3343 | return -ENOMEM; | |
3344 | ||
3345 | return 0; | |
3346 | } | |
728404da FM |
3347 | |
3348 | int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end) | |
3349 | { | |
3350 | int ret; | |
3351 | ||
3352 | /* | |
3353 | * So with compression we will find and lock a dirty page and clear the | |
3354 | * first one as dirty, setup an async extent, and immediately return | |
3355 | * with the entire range locked but with nobody actually marked with | |
3356 | * writeback. So we can't just filemap_write_and_wait_range() and | |
3357 | * expect it to work since it will just kick off a thread to do the | |
3358 | * actual work. So we need to call filemap_fdatawrite_range _again_ | |
3359 | * since it will wait on the page lock, which won't be unlocked until | |
3360 | * after the pages have been marked as writeback and so we're good to go | |
3361 | * from there. We have to do this otherwise we'll miss the ordered | |
3362 | * extents and that results in badness. Please Josef, do not think you | |
3363 | * know better and pull this out at some point in the future, it is | |
3364 | * right and you are wrong. | |
3365 | */ | |
3366 | ret = filemap_fdatawrite_range(inode->i_mapping, start, end); | |
3367 | if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
3368 | &BTRFS_I(inode)->runtime_flags)) | |
3369 | ret = filemap_fdatawrite_range(inode->i_mapping, start, end); | |
3370 | ||
3371 | return ret; | |
3372 | } |