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> |
14605409 | 19 | #include <linux/fsverity.h> |
39279cc3 CM |
20 | #include "ctree.h" |
21 | #include "disk-io.h" | |
22 | #include "transaction.h" | |
23 | #include "btrfs_inode.h" | |
39279cc3 | 24 | #include "print-tree.h" |
e02119d5 CM |
25 | #include "tree-log.h" |
26 | #include "locking.h" | |
2aaa6655 | 27 | #include "volumes.h" |
fcebe456 | 28 | #include "qgroup.h" |
ebb8765b | 29 | #include "compression.h" |
86736342 | 30 | #include "delalloc-space.h" |
6a177381 | 31 | #include "reflink.h" |
f02a85d2 | 32 | #include "subpage.h" |
39279cc3 | 33 | |
9247f317 | 34 | static struct kmem_cache *btrfs_inode_defrag_cachep; |
4cb5300b CM |
35 | /* |
36 | * when auto defrag is enabled we | |
37 | * queue up these defrag structs to remember which | |
38 | * inodes need defragging passes | |
39 | */ | |
40 | struct inode_defrag { | |
41 | struct rb_node rb_node; | |
42 | /* objectid */ | |
43 | u64 ino; | |
44 | /* | |
45 | * transid where the defrag was added, we search for | |
46 | * extents newer than this | |
47 | */ | |
48 | u64 transid; | |
49 | ||
50 | /* root objectid */ | |
51 | u64 root; | |
558732df QW |
52 | |
53 | /* | |
54 | * The extent size threshold for autodefrag. | |
55 | * | |
56 | * This value is different for compressed/non-compressed extents, | |
57 | * thus needs to be passed from higher layer. | |
58 | * (aka, inode_should_defrag()) | |
59 | */ | |
60 | u32 extent_thresh; | |
4cb5300b CM |
61 | }; |
62 | ||
762f2263 MX |
63 | static int __compare_inode_defrag(struct inode_defrag *defrag1, |
64 | struct inode_defrag *defrag2) | |
65 | { | |
66 | if (defrag1->root > defrag2->root) | |
67 | return 1; | |
68 | else if (defrag1->root < defrag2->root) | |
69 | return -1; | |
70 | else if (defrag1->ino > defrag2->ino) | |
71 | return 1; | |
72 | else if (defrag1->ino < defrag2->ino) | |
73 | return -1; | |
74 | else | |
75 | return 0; | |
76 | } | |
77 | ||
4cb5300b CM |
78 | /* pop a record for an inode into the defrag tree. The lock |
79 | * must be held already | |
80 | * | |
81 | * If you're inserting a record for an older transid than an | |
82 | * existing record, the transid already in the tree is lowered | |
83 | * | |
84 | * If an existing record is found the defrag item you | |
85 | * pass in is freed | |
86 | */ | |
6158e1ce | 87 | static int __btrfs_add_inode_defrag(struct btrfs_inode *inode, |
4cb5300b CM |
88 | struct inode_defrag *defrag) |
89 | { | |
3ffbd68c | 90 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
4cb5300b CM |
91 | struct inode_defrag *entry; |
92 | struct rb_node **p; | |
93 | struct rb_node *parent = NULL; | |
762f2263 | 94 | int ret; |
4cb5300b | 95 | |
0b246afa | 96 | p = &fs_info->defrag_inodes.rb_node; |
4cb5300b CM |
97 | while (*p) { |
98 | parent = *p; | |
99 | entry = rb_entry(parent, struct inode_defrag, rb_node); | |
100 | ||
762f2263 MX |
101 | ret = __compare_inode_defrag(defrag, entry); |
102 | if (ret < 0) | |
4cb5300b | 103 | p = &parent->rb_left; |
762f2263 | 104 | else if (ret > 0) |
4cb5300b CM |
105 | p = &parent->rb_right; |
106 | else { | |
107 | /* if we're reinserting an entry for | |
108 | * an old defrag run, make sure to | |
109 | * lower the transid of our existing record | |
110 | */ | |
111 | if (defrag->transid < entry->transid) | |
112 | entry->transid = defrag->transid; | |
558732df QW |
113 | entry->extent_thresh = min(defrag->extent_thresh, |
114 | entry->extent_thresh); | |
8ddc4734 | 115 | return -EEXIST; |
4cb5300b CM |
116 | } |
117 | } | |
6158e1ce | 118 | set_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags); |
4cb5300b | 119 | rb_link_node(&defrag->rb_node, parent, p); |
0b246afa | 120 | rb_insert_color(&defrag->rb_node, &fs_info->defrag_inodes); |
8ddc4734 MX |
121 | return 0; |
122 | } | |
4cb5300b | 123 | |
2ff7e61e | 124 | static inline int __need_auto_defrag(struct btrfs_fs_info *fs_info) |
8ddc4734 | 125 | { |
0b246afa | 126 | if (!btrfs_test_opt(fs_info, AUTO_DEFRAG)) |
8ddc4734 MX |
127 | return 0; |
128 | ||
0b246afa | 129 | if (btrfs_fs_closing(fs_info)) |
8ddc4734 | 130 | return 0; |
4cb5300b | 131 | |
8ddc4734 | 132 | return 1; |
4cb5300b CM |
133 | } |
134 | ||
135 | /* | |
136 | * insert a defrag record for this inode if auto defrag is | |
137 | * enabled | |
138 | */ | |
139 | int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans, | |
558732df | 140 | struct btrfs_inode *inode, u32 extent_thresh) |
4cb5300b | 141 | { |
6158e1ce | 142 | struct btrfs_root *root = inode->root; |
3ffbd68c | 143 | struct btrfs_fs_info *fs_info = root->fs_info; |
4cb5300b | 144 | struct inode_defrag *defrag; |
4cb5300b | 145 | u64 transid; |
8ddc4734 | 146 | int ret; |
4cb5300b | 147 | |
2ff7e61e | 148 | if (!__need_auto_defrag(fs_info)) |
4cb5300b CM |
149 | return 0; |
150 | ||
6158e1ce | 151 | if (test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags)) |
4cb5300b CM |
152 | return 0; |
153 | ||
154 | if (trans) | |
155 | transid = trans->transid; | |
156 | else | |
6158e1ce | 157 | transid = inode->root->last_trans; |
4cb5300b | 158 | |
9247f317 | 159 | defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS); |
4cb5300b CM |
160 | if (!defrag) |
161 | return -ENOMEM; | |
162 | ||
6158e1ce | 163 | defrag->ino = btrfs_ino(inode); |
4cb5300b CM |
164 | defrag->transid = transid; |
165 | defrag->root = root->root_key.objectid; | |
558732df | 166 | defrag->extent_thresh = extent_thresh; |
4cb5300b | 167 | |
0b246afa | 168 | spin_lock(&fs_info->defrag_inodes_lock); |
6158e1ce | 169 | if (!test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags)) { |
8ddc4734 MX |
170 | /* |
171 | * If we set IN_DEFRAG flag and evict the inode from memory, | |
172 | * and then re-read this inode, this new inode doesn't have | |
173 | * IN_DEFRAG flag. At the case, we may find the existed defrag. | |
174 | */ | |
175 | ret = __btrfs_add_inode_defrag(inode, defrag); | |
176 | if (ret) | |
177 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
178 | } else { | |
9247f317 | 179 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); |
8ddc4734 | 180 | } |
0b246afa | 181 | spin_unlock(&fs_info->defrag_inodes_lock); |
a0f98dde | 182 | return 0; |
4cb5300b CM |
183 | } |
184 | ||
185 | /* | |
26176e7c MX |
186 | * pick the defragable inode that we want, if it doesn't exist, we will get |
187 | * the next one. | |
4cb5300b | 188 | */ |
26176e7c MX |
189 | static struct inode_defrag * |
190 | btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino) | |
4cb5300b CM |
191 | { |
192 | struct inode_defrag *entry = NULL; | |
762f2263 | 193 | struct inode_defrag tmp; |
4cb5300b CM |
194 | struct rb_node *p; |
195 | struct rb_node *parent = NULL; | |
762f2263 MX |
196 | int ret; |
197 | ||
198 | tmp.ino = ino; | |
199 | tmp.root = root; | |
4cb5300b | 200 | |
26176e7c MX |
201 | spin_lock(&fs_info->defrag_inodes_lock); |
202 | p = fs_info->defrag_inodes.rb_node; | |
4cb5300b CM |
203 | while (p) { |
204 | parent = p; | |
205 | entry = rb_entry(parent, struct inode_defrag, rb_node); | |
206 | ||
762f2263 MX |
207 | ret = __compare_inode_defrag(&tmp, entry); |
208 | if (ret < 0) | |
4cb5300b | 209 | p = parent->rb_left; |
762f2263 | 210 | else if (ret > 0) |
4cb5300b CM |
211 | p = parent->rb_right; |
212 | else | |
26176e7c | 213 | goto out; |
4cb5300b CM |
214 | } |
215 | ||
26176e7c MX |
216 | if (parent && __compare_inode_defrag(&tmp, entry) > 0) { |
217 | parent = rb_next(parent); | |
218 | if (parent) | |
4cb5300b | 219 | entry = rb_entry(parent, struct inode_defrag, rb_node); |
26176e7c MX |
220 | else |
221 | entry = NULL; | |
4cb5300b | 222 | } |
26176e7c MX |
223 | out: |
224 | if (entry) | |
225 | rb_erase(parent, &fs_info->defrag_inodes); | |
226 | spin_unlock(&fs_info->defrag_inodes_lock); | |
227 | return entry; | |
4cb5300b CM |
228 | } |
229 | ||
26176e7c | 230 | void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info) |
4cb5300b CM |
231 | { |
232 | struct inode_defrag *defrag; | |
26176e7c MX |
233 | struct rb_node *node; |
234 | ||
235 | spin_lock(&fs_info->defrag_inodes_lock); | |
236 | node = rb_first(&fs_info->defrag_inodes); | |
237 | while (node) { | |
238 | rb_erase(node, &fs_info->defrag_inodes); | |
239 | defrag = rb_entry(node, struct inode_defrag, rb_node); | |
240 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
241 | ||
351810c1 | 242 | cond_resched_lock(&fs_info->defrag_inodes_lock); |
26176e7c MX |
243 | |
244 | node = rb_first(&fs_info->defrag_inodes); | |
245 | } | |
246 | spin_unlock(&fs_info->defrag_inodes_lock); | |
247 | } | |
248 | ||
249 | #define BTRFS_DEFRAG_BATCH 1024 | |
250 | ||
251 | static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info, | |
252 | struct inode_defrag *defrag) | |
253 | { | |
4cb5300b CM |
254 | struct btrfs_root *inode_root; |
255 | struct inode *inode; | |
4cb5300b | 256 | struct btrfs_ioctl_defrag_range_args range; |
26fbac25 QW |
257 | int ret = 0; |
258 | u64 cur = 0; | |
259 | ||
260 | again: | |
261 | if (test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state)) | |
262 | goto cleanup; | |
263 | if (!__need_auto_defrag(fs_info)) | |
264 | goto cleanup; | |
4cb5300b | 265 | |
26176e7c | 266 | /* get the inode */ |
56e9357a | 267 | inode_root = btrfs_get_fs_root(fs_info, defrag->root, true); |
26176e7c | 268 | if (IS_ERR(inode_root)) { |
6f1c3605 LB |
269 | ret = PTR_ERR(inode_root); |
270 | goto cleanup; | |
271 | } | |
26176e7c | 272 | |
0202e83f | 273 | inode = btrfs_iget(fs_info->sb, defrag->ino, inode_root); |
00246528 | 274 | btrfs_put_root(inode_root); |
26176e7c | 275 | if (IS_ERR(inode)) { |
6f1c3605 LB |
276 | ret = PTR_ERR(inode); |
277 | goto cleanup; | |
26176e7c MX |
278 | } |
279 | ||
26fbac25 QW |
280 | if (cur >= i_size_read(inode)) { |
281 | iput(inode); | |
282 | goto cleanup; | |
283 | } | |
284 | ||
26176e7c MX |
285 | /* do a chunk of defrag */ |
286 | clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags); | |
4cb5300b CM |
287 | memset(&range, 0, sizeof(range)); |
288 | range.len = (u64)-1; | |
26fbac25 | 289 | range.start = cur; |
558732df | 290 | range.extent_thresh = defrag->extent_thresh; |
b66f00da MX |
291 | |
292 | sb_start_write(fs_info->sb); | |
26fbac25 | 293 | ret = btrfs_defrag_file(inode, NULL, &range, defrag->transid, |
26176e7c | 294 | BTRFS_DEFRAG_BATCH); |
b66f00da | 295 | sb_end_write(fs_info->sb); |
26176e7c | 296 | iput(inode); |
26fbac25 QW |
297 | |
298 | if (ret < 0) | |
299 | goto cleanup; | |
300 | ||
301 | cur = max(cur + fs_info->sectorsize, range.start); | |
302 | goto again; | |
303 | ||
6f1c3605 | 304 | cleanup: |
6f1c3605 LB |
305 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); |
306 | return ret; | |
26176e7c MX |
307 | } |
308 | ||
309 | /* | |
310 | * run through the list of inodes in the FS that need | |
311 | * defragging | |
312 | */ | |
313 | int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info) | |
314 | { | |
315 | struct inode_defrag *defrag; | |
316 | u64 first_ino = 0; | |
317 | u64 root_objectid = 0; | |
4cb5300b CM |
318 | |
319 | atomic_inc(&fs_info->defrag_running); | |
67871254 | 320 | while (1) { |
dc81cdc5 MX |
321 | /* Pause the auto defragger. */ |
322 | if (test_bit(BTRFS_FS_STATE_REMOUNTING, | |
323 | &fs_info->fs_state)) | |
324 | break; | |
325 | ||
2ff7e61e | 326 | if (!__need_auto_defrag(fs_info)) |
26176e7c | 327 | break; |
4cb5300b CM |
328 | |
329 | /* find an inode to defrag */ | |
26176e7c MX |
330 | defrag = btrfs_pick_defrag_inode(fs_info, root_objectid, |
331 | first_ino); | |
4cb5300b | 332 | if (!defrag) { |
26176e7c | 333 | if (root_objectid || first_ino) { |
762f2263 | 334 | root_objectid = 0; |
4cb5300b CM |
335 | first_ino = 0; |
336 | continue; | |
337 | } else { | |
338 | break; | |
339 | } | |
340 | } | |
341 | ||
4cb5300b | 342 | first_ino = defrag->ino + 1; |
762f2263 | 343 | root_objectid = defrag->root; |
4cb5300b | 344 | |
26176e7c | 345 | __btrfs_run_defrag_inode(fs_info, defrag); |
4cb5300b | 346 | } |
4cb5300b CM |
347 | atomic_dec(&fs_info->defrag_running); |
348 | ||
349 | /* | |
350 | * during unmount, we use the transaction_wait queue to | |
351 | * wait for the defragger to stop | |
352 | */ | |
353 | wake_up(&fs_info->transaction_wait); | |
354 | return 0; | |
355 | } | |
39279cc3 | 356 | |
d352ac68 CM |
357 | /* simple helper to fault in pages and copy. This should go away |
358 | * and be replaced with calls into generic code. | |
359 | */ | |
ee22f0c4 | 360 | static noinline int btrfs_copy_from_user(loff_t pos, size_t write_bytes, |
a1b32a59 | 361 | struct page **prepared_pages, |
11c65dcc | 362 | struct iov_iter *i) |
39279cc3 | 363 | { |
914ee295 | 364 | size_t copied = 0; |
d0215f3e | 365 | size_t total_copied = 0; |
11c65dcc | 366 | int pg = 0; |
7073017a | 367 | int offset = offset_in_page(pos); |
39279cc3 | 368 | |
11c65dcc | 369 | while (write_bytes > 0) { |
39279cc3 | 370 | size_t count = min_t(size_t, |
09cbfeaf | 371 | PAGE_SIZE - offset, write_bytes); |
11c65dcc | 372 | struct page *page = prepared_pages[pg]; |
914ee295 XZ |
373 | /* |
374 | * Copy data from userspace to the current page | |
914ee295 | 375 | */ |
f0b65f39 | 376 | copied = copy_page_from_iter_atomic(page, offset, count, i); |
11c65dcc | 377 | |
39279cc3 CM |
378 | /* Flush processor's dcache for this page */ |
379 | flush_dcache_page(page); | |
31339acd CM |
380 | |
381 | /* | |
382 | * if we get a partial write, we can end up with | |
383 | * partially up to date pages. These add | |
384 | * a lot of complexity, so make sure they don't | |
385 | * happen by forcing this copy to be retried. | |
386 | * | |
387 | * The rest of the btrfs_file_write code will fall | |
388 | * back to page at a time copies after we return 0. | |
389 | */ | |
f0b65f39 AV |
390 | if (unlikely(copied < count)) { |
391 | if (!PageUptodate(page)) { | |
392 | iov_iter_revert(i, copied); | |
393 | copied = 0; | |
394 | } | |
395 | if (!copied) | |
396 | break; | |
397 | } | |
31339acd | 398 | |
11c65dcc | 399 | write_bytes -= copied; |
914ee295 | 400 | total_copied += copied; |
f0b65f39 AV |
401 | offset += copied; |
402 | if (offset == PAGE_SIZE) { | |
11c65dcc JB |
403 | pg++; |
404 | offset = 0; | |
405 | } | |
39279cc3 | 406 | } |
914ee295 | 407 | return total_copied; |
39279cc3 CM |
408 | } |
409 | ||
d352ac68 CM |
410 | /* |
411 | * unlocks pages after btrfs_file_write is done with them | |
412 | */ | |
e4f94347 QW |
413 | static void btrfs_drop_pages(struct btrfs_fs_info *fs_info, |
414 | struct page **pages, size_t num_pages, | |
415 | u64 pos, u64 copied) | |
39279cc3 CM |
416 | { |
417 | size_t i; | |
e4f94347 QW |
418 | u64 block_start = round_down(pos, fs_info->sectorsize); |
419 | u64 block_len = round_up(pos + copied, fs_info->sectorsize) - block_start; | |
420 | ||
421 | ASSERT(block_len <= U32_MAX); | |
39279cc3 | 422 | for (i = 0; i < num_pages; i++) { |
d352ac68 CM |
423 | /* page checked is some magic around finding pages that |
424 | * have been modified without going through btrfs_set_page_dirty | |
2457aec6 MG |
425 | * clear it here. There should be no need to mark the pages |
426 | * accessed as prepare_pages should have marked them accessed | |
427 | * in prepare_pages via find_or_create_page() | |
d352ac68 | 428 | */ |
e4f94347 QW |
429 | btrfs_page_clamp_clear_checked(fs_info, pages[i], block_start, |
430 | block_len); | |
39279cc3 | 431 | unlock_page(pages[i]); |
09cbfeaf | 432 | put_page(pages[i]); |
39279cc3 CM |
433 | } |
434 | } | |
435 | ||
d352ac68 | 436 | /* |
c0fab480 QW |
437 | * After btrfs_copy_from_user(), update the following things for delalloc: |
438 | * - Mark newly dirtied pages as DELALLOC in the io tree. | |
439 | * Used to advise which range is to be written back. | |
440 | * - Mark modified pages as Uptodate/Dirty and not needing COW fixup | |
441 | * - Update inode size for past EOF write | |
d352ac68 | 442 | */ |
088545f6 | 443 | int btrfs_dirty_pages(struct btrfs_inode *inode, struct page **pages, |
2ff7e61e | 444 | size_t num_pages, loff_t pos, size_t write_bytes, |
aa8c1a41 | 445 | struct extent_state **cached, bool noreserve) |
39279cc3 | 446 | { |
088545f6 | 447 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
39279cc3 | 448 | int err = 0; |
a52d9a80 | 449 | int i; |
db94535d | 450 | u64 num_bytes; |
a52d9a80 CM |
451 | u64 start_pos; |
452 | u64 end_of_last_block; | |
453 | u64 end_pos = pos + write_bytes; | |
088545f6 | 454 | loff_t isize = i_size_read(&inode->vfs_inode); |
e3b8a485 | 455 | unsigned int extra_bits = 0; |
39279cc3 | 456 | |
aa8c1a41 GR |
457 | if (write_bytes == 0) |
458 | return 0; | |
459 | ||
460 | if (noreserve) | |
461 | extra_bits |= EXTENT_NORESERVE; | |
462 | ||
13f0dd8f | 463 | start_pos = round_down(pos, fs_info->sectorsize); |
da17066c | 464 | num_bytes = round_up(write_bytes + pos - start_pos, |
0b246afa | 465 | fs_info->sectorsize); |
f02a85d2 | 466 | ASSERT(num_bytes <= U32_MAX); |
39279cc3 | 467 | |
db94535d | 468 | end_of_last_block = start_pos + num_bytes - 1; |
e3b8a485 | 469 | |
7703bdd8 CM |
470 | /* |
471 | * The pages may have already been dirty, clear out old accounting so | |
472 | * we can set things up properly | |
473 | */ | |
088545f6 | 474 | clear_extent_bit(&inode->io_tree, start_pos, end_of_last_block, |
e182163d | 475 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, |
bd015294 | 476 | cached); |
7703bdd8 | 477 | |
088545f6 | 478 | err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block, |
330a5827 | 479 | extra_bits, cached); |
d0215f3e JB |
480 | if (err) |
481 | return err; | |
9ed74f2d | 482 | |
c8b97818 CM |
483 | for (i = 0; i < num_pages; i++) { |
484 | struct page *p = pages[i]; | |
f02a85d2 QW |
485 | |
486 | btrfs_page_clamp_set_uptodate(fs_info, p, start_pos, num_bytes); | |
e4f94347 | 487 | btrfs_page_clamp_clear_checked(fs_info, p, start_pos, num_bytes); |
f02a85d2 | 488 | btrfs_page_clamp_set_dirty(fs_info, p, start_pos, num_bytes); |
a52d9a80 | 489 | } |
9f570b8d JB |
490 | |
491 | /* | |
492 | * we've only changed i_size in ram, and we haven't updated | |
493 | * the disk i_size. There is no need to log the inode | |
494 | * at this time. | |
495 | */ | |
496 | if (end_pos > isize) | |
088545f6 | 497 | i_size_write(&inode->vfs_inode, end_pos); |
a22285a6 | 498 | return 0; |
39279cc3 CM |
499 | } |
500 | ||
d352ac68 CM |
501 | /* |
502 | * this drops all the extents in the cache that intersect the range | |
503 | * [start, end]. Existing extents are split as required. | |
504 | */ | |
dcdbc059 | 505 | void btrfs_drop_extent_cache(struct btrfs_inode *inode, u64 start, u64 end, |
7014cdb4 | 506 | int skip_pinned) |
a52d9a80 CM |
507 | { |
508 | struct extent_map *em; | |
3b951516 CM |
509 | struct extent_map *split = NULL; |
510 | struct extent_map *split2 = NULL; | |
dcdbc059 | 511 | struct extent_map_tree *em_tree = &inode->extent_tree; |
39b5637f | 512 | u64 len = end - start + 1; |
5dc562c5 | 513 | u64 gen; |
3b951516 CM |
514 | int ret; |
515 | int testend = 1; | |
5b21f2ed | 516 | unsigned long flags; |
c8b97818 | 517 | int compressed = 0; |
09a2a8f9 | 518 | bool modified; |
a52d9a80 | 519 | |
e6dcd2dc | 520 | WARN_ON(end < start); |
3b951516 | 521 | if (end == (u64)-1) { |
39b5637f | 522 | len = (u64)-1; |
3b951516 CM |
523 | testend = 0; |
524 | } | |
d397712b | 525 | while (1) { |
7014cdb4 JB |
526 | int no_splits = 0; |
527 | ||
09a2a8f9 | 528 | modified = false; |
3b951516 | 529 | if (!split) |
172ddd60 | 530 | split = alloc_extent_map(); |
3b951516 | 531 | if (!split2) |
172ddd60 | 532 | split2 = alloc_extent_map(); |
7014cdb4 JB |
533 | if (!split || !split2) |
534 | no_splits = 1; | |
3b951516 | 535 | |
890871be | 536 | write_lock(&em_tree->lock); |
39b5637f | 537 | em = lookup_extent_mapping(em_tree, start, len); |
d1310b2e | 538 | if (!em) { |
890871be | 539 | write_unlock(&em_tree->lock); |
a52d9a80 | 540 | break; |
d1310b2e | 541 | } |
5b21f2ed | 542 | flags = em->flags; |
5dc562c5 | 543 | gen = em->generation; |
5b21f2ed | 544 | if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) { |
55ef6899 | 545 | if (testend && em->start + em->len >= start + len) { |
5b21f2ed | 546 | free_extent_map(em); |
a1ed835e | 547 | write_unlock(&em_tree->lock); |
5b21f2ed ZY |
548 | break; |
549 | } | |
55ef6899 YZ |
550 | start = em->start + em->len; |
551 | if (testend) | |
5b21f2ed | 552 | len = start + len - (em->start + em->len); |
5b21f2ed | 553 | free_extent_map(em); |
a1ed835e | 554 | write_unlock(&em_tree->lock); |
5b21f2ed ZY |
555 | continue; |
556 | } | |
c8b97818 | 557 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
3ce7e67a | 558 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
3b277594 | 559 | clear_bit(EXTENT_FLAG_LOGGING, &flags); |
09a2a8f9 | 560 | modified = !list_empty(&em->list); |
7014cdb4 JB |
561 | if (no_splits) |
562 | goto next; | |
3b951516 | 563 | |
ee20a983 | 564 | if (em->start < start) { |
3b951516 CM |
565 | split->start = em->start; |
566 | split->len = start - em->start; | |
ee20a983 JB |
567 | |
568 | if (em->block_start < EXTENT_MAP_LAST_BYTE) { | |
569 | split->orig_start = em->orig_start; | |
570 | split->block_start = em->block_start; | |
571 | ||
572 | if (compressed) | |
573 | split->block_len = em->block_len; | |
574 | else | |
575 | split->block_len = split->len; | |
576 | split->orig_block_len = max(split->block_len, | |
577 | em->orig_block_len); | |
578 | split->ram_bytes = em->ram_bytes; | |
579 | } else { | |
580 | split->orig_start = split->start; | |
581 | split->block_len = 0; | |
582 | split->block_start = em->block_start; | |
583 | split->orig_block_len = 0; | |
584 | split->ram_bytes = split->len; | |
585 | } | |
586 | ||
5dc562c5 | 587 | split->generation = gen; |
5b21f2ed | 588 | split->flags = flags; |
261507a0 | 589 | split->compress_type = em->compress_type; |
176840b3 | 590 | replace_extent_mapping(em_tree, em, split, modified); |
3b951516 CM |
591 | free_extent_map(split); |
592 | split = split2; | |
593 | split2 = NULL; | |
594 | } | |
ee20a983 | 595 | if (testend && em->start + em->len > start + len) { |
3b951516 CM |
596 | u64 diff = start + len - em->start; |
597 | ||
598 | split->start = start + len; | |
599 | split->len = em->start + em->len - (start + len); | |
5b21f2ed | 600 | split->flags = flags; |
261507a0 | 601 | split->compress_type = em->compress_type; |
5dc562c5 | 602 | split->generation = gen; |
ee20a983 JB |
603 | |
604 | if (em->block_start < EXTENT_MAP_LAST_BYTE) { | |
605 | split->orig_block_len = max(em->block_len, | |
b4939680 | 606 | em->orig_block_len); |
3b951516 | 607 | |
ee20a983 JB |
608 | split->ram_bytes = em->ram_bytes; |
609 | if (compressed) { | |
610 | split->block_len = em->block_len; | |
611 | split->block_start = em->block_start; | |
612 | split->orig_start = em->orig_start; | |
613 | } else { | |
614 | split->block_len = split->len; | |
615 | split->block_start = em->block_start | |
616 | + diff; | |
617 | split->orig_start = em->orig_start; | |
618 | } | |
c8b97818 | 619 | } else { |
ee20a983 JB |
620 | split->ram_bytes = split->len; |
621 | split->orig_start = split->start; | |
622 | split->block_len = 0; | |
623 | split->block_start = em->block_start; | |
624 | split->orig_block_len = 0; | |
c8b97818 | 625 | } |
3b951516 | 626 | |
176840b3 FM |
627 | if (extent_map_in_tree(em)) { |
628 | replace_extent_mapping(em_tree, em, split, | |
629 | modified); | |
630 | } else { | |
631 | ret = add_extent_mapping(em_tree, split, | |
632 | modified); | |
633 | ASSERT(ret == 0); /* Logic error */ | |
634 | } | |
3b951516 CM |
635 | free_extent_map(split); |
636 | split = NULL; | |
637 | } | |
7014cdb4 | 638 | next: |
176840b3 FM |
639 | if (extent_map_in_tree(em)) |
640 | remove_extent_mapping(em_tree, em); | |
890871be | 641 | write_unlock(&em_tree->lock); |
d1310b2e | 642 | |
a52d9a80 CM |
643 | /* once for us */ |
644 | free_extent_map(em); | |
645 | /* once for the tree*/ | |
646 | free_extent_map(em); | |
647 | } | |
3b951516 CM |
648 | if (split) |
649 | free_extent_map(split); | |
650 | if (split2) | |
651 | free_extent_map(split2); | |
a52d9a80 CM |
652 | } |
653 | ||
39279cc3 CM |
654 | /* |
655 | * this is very complex, but the basic idea is to drop all extents | |
656 | * in the range start - end. hint_block is filled in with a block number | |
657 | * that would be a good hint to the block allocator for this file. | |
658 | * | |
659 | * If an extent intersects the range but is not entirely inside the range | |
660 | * it is either truncated or split. Anything entirely inside the range | |
661 | * is deleted from the tree. | |
2766ff61 FM |
662 | * |
663 | * Note: the VFS' inode number of bytes is not updated, it's up to the caller | |
664 | * to deal with that. We set the field 'bytes_found' of the arguments structure | |
665 | * with the number of allocated bytes found in the target range, so that the | |
666 | * caller can update the inode's number of bytes in an atomic way when | |
667 | * replacing extents in a range to avoid races with stat(2). | |
39279cc3 | 668 | */ |
5893dfb9 FM |
669 | int btrfs_drop_extents(struct btrfs_trans_handle *trans, |
670 | struct btrfs_root *root, struct btrfs_inode *inode, | |
671 | struct btrfs_drop_extents_args *args) | |
39279cc3 | 672 | { |
0b246afa | 673 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 674 | struct extent_buffer *leaf; |
920bbbfb | 675 | struct btrfs_file_extent_item *fi; |
82fa113f | 676 | struct btrfs_ref ref = { 0 }; |
00f5c795 | 677 | struct btrfs_key key; |
920bbbfb | 678 | struct btrfs_key new_key; |
906c448c | 679 | u64 ino = btrfs_ino(inode); |
5893dfb9 | 680 | u64 search_start = args->start; |
920bbbfb YZ |
681 | u64 disk_bytenr = 0; |
682 | u64 num_bytes = 0; | |
683 | u64 extent_offset = 0; | |
684 | u64 extent_end = 0; | |
5893dfb9 | 685 | u64 last_end = args->start; |
920bbbfb YZ |
686 | int del_nr = 0; |
687 | int del_slot = 0; | |
688 | int extent_type; | |
ccd467d6 | 689 | int recow; |
00f5c795 | 690 | int ret; |
dc7fdde3 | 691 | int modify_tree = -1; |
27cdeb70 | 692 | int update_refs; |
c3308f84 | 693 | int found = 0; |
5893dfb9 FM |
694 | struct btrfs_path *path = args->path; |
695 | ||
2766ff61 | 696 | args->bytes_found = 0; |
5893dfb9 FM |
697 | args->extent_inserted = false; |
698 | ||
699 | /* Must always have a path if ->replace_extent is true */ | |
700 | ASSERT(!(args->replace_extent && !args->path)); | |
701 | ||
702 | if (!path) { | |
703 | path = btrfs_alloc_path(); | |
704 | if (!path) { | |
705 | ret = -ENOMEM; | |
706 | goto out; | |
707 | } | |
708 | } | |
39279cc3 | 709 | |
5893dfb9 FM |
710 | if (args->drop_cache) |
711 | btrfs_drop_extent_cache(inode, args->start, args->end - 1, 0); | |
a52d9a80 | 712 | |
5893dfb9 | 713 | if (args->start >= inode->disk_i_size && !args->replace_extent) |
dc7fdde3 CM |
714 | modify_tree = 0; |
715 | ||
d175209b | 716 | update_refs = (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID); |
d397712b | 717 | while (1) { |
ccd467d6 | 718 | recow = 0; |
33345d01 | 719 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
dc7fdde3 | 720 | search_start, modify_tree); |
39279cc3 | 721 | if (ret < 0) |
920bbbfb | 722 | break; |
5893dfb9 | 723 | if (ret > 0 && path->slots[0] > 0 && search_start == args->start) { |
920bbbfb YZ |
724 | leaf = path->nodes[0]; |
725 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1); | |
33345d01 | 726 | if (key.objectid == ino && |
920bbbfb YZ |
727 | key.type == BTRFS_EXTENT_DATA_KEY) |
728 | path->slots[0]--; | |
39279cc3 | 729 | } |
920bbbfb | 730 | ret = 0; |
8c2383c3 | 731 | next_slot: |
5f39d397 | 732 | leaf = path->nodes[0]; |
920bbbfb YZ |
733 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { |
734 | BUG_ON(del_nr > 0); | |
735 | ret = btrfs_next_leaf(root, path); | |
736 | if (ret < 0) | |
737 | break; | |
738 | if (ret > 0) { | |
739 | ret = 0; | |
740 | break; | |
8c2383c3 | 741 | } |
920bbbfb YZ |
742 | leaf = path->nodes[0]; |
743 | recow = 1; | |
744 | } | |
745 | ||
746 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
aeafbf84 FM |
747 | |
748 | if (key.objectid > ino) | |
749 | break; | |
750 | if (WARN_ON_ONCE(key.objectid < ino) || | |
751 | key.type < BTRFS_EXTENT_DATA_KEY) { | |
752 | ASSERT(del_nr == 0); | |
753 | path->slots[0]++; | |
754 | goto next_slot; | |
755 | } | |
5893dfb9 | 756 | if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= args->end) |
920bbbfb YZ |
757 | break; |
758 | ||
759 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
760 | struct btrfs_file_extent_item); | |
761 | extent_type = btrfs_file_extent_type(leaf, fi); | |
762 | ||
763 | if (extent_type == BTRFS_FILE_EXTENT_REG || | |
764 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
765 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
766 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
767 | extent_offset = btrfs_file_extent_offset(leaf, fi); | |
768 | extent_end = key.offset + | |
769 | btrfs_file_extent_num_bytes(leaf, fi); | |
770 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
771 | extent_end = key.offset + | |
e41ca589 | 772 | btrfs_file_extent_ram_bytes(leaf, fi); |
8c2383c3 | 773 | } else { |
aeafbf84 FM |
774 | /* can't happen */ |
775 | BUG(); | |
39279cc3 CM |
776 | } |
777 | ||
fc19c5e7 FM |
778 | /* |
779 | * Don't skip extent items representing 0 byte lengths. They | |
780 | * used to be created (bug) if while punching holes we hit | |
781 | * -ENOSPC condition. So if we find one here, just ensure we | |
782 | * delete it, otherwise we would insert a new file extent item | |
783 | * with the same key (offset) as that 0 bytes length file | |
784 | * extent item in the call to setup_items_for_insert() later | |
785 | * in this function. | |
786 | */ | |
62fe51c1 JB |
787 | if (extent_end == key.offset && extent_end >= search_start) { |
788 | last_end = extent_end; | |
fc19c5e7 | 789 | goto delete_extent_item; |
62fe51c1 | 790 | } |
fc19c5e7 | 791 | |
920bbbfb YZ |
792 | if (extent_end <= search_start) { |
793 | path->slots[0]++; | |
8c2383c3 | 794 | goto next_slot; |
39279cc3 CM |
795 | } |
796 | ||
c3308f84 | 797 | found = 1; |
5893dfb9 | 798 | search_start = max(key.offset, args->start); |
dc7fdde3 CM |
799 | if (recow || !modify_tree) { |
800 | modify_tree = -1; | |
b3b4aa74 | 801 | btrfs_release_path(path); |
920bbbfb | 802 | continue; |
39279cc3 | 803 | } |
6643558d | 804 | |
920bbbfb YZ |
805 | /* |
806 | * | - range to drop - | | |
807 | * | -------- extent -------- | | |
808 | */ | |
5893dfb9 | 809 | if (args->start > key.offset && args->end < extent_end) { |
920bbbfb | 810 | BUG_ON(del_nr > 0); |
00fdf13a | 811 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
3f9e3df8 | 812 | ret = -EOPNOTSUPP; |
00fdf13a LB |
813 | break; |
814 | } | |
920bbbfb YZ |
815 | |
816 | memcpy(&new_key, &key, sizeof(new_key)); | |
5893dfb9 | 817 | new_key.offset = args->start; |
920bbbfb YZ |
818 | ret = btrfs_duplicate_item(trans, root, path, |
819 | &new_key); | |
820 | if (ret == -EAGAIN) { | |
b3b4aa74 | 821 | btrfs_release_path(path); |
920bbbfb | 822 | continue; |
6643558d | 823 | } |
920bbbfb YZ |
824 | if (ret < 0) |
825 | break; | |
826 | ||
827 | leaf = path->nodes[0]; | |
828 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
829 | struct btrfs_file_extent_item); | |
830 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
5893dfb9 | 831 | args->start - key.offset); |
920bbbfb YZ |
832 | |
833 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
834 | struct btrfs_file_extent_item); | |
835 | ||
5893dfb9 | 836 | extent_offset += args->start - key.offset; |
920bbbfb YZ |
837 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); |
838 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
5893dfb9 | 839 | extent_end - args->start); |
920bbbfb YZ |
840 | btrfs_mark_buffer_dirty(leaf); |
841 | ||
5dc562c5 | 842 | if (update_refs && disk_bytenr > 0) { |
82fa113f QW |
843 | btrfs_init_generic_ref(&ref, |
844 | BTRFS_ADD_DELAYED_REF, | |
845 | disk_bytenr, num_bytes, 0); | |
846 | btrfs_init_data_ref(&ref, | |
920bbbfb YZ |
847 | root->root_key.objectid, |
848 | new_key.objectid, | |
f42c5da6 NB |
849 | args->start - extent_offset, |
850 | 0, false); | |
82fa113f | 851 | ret = btrfs_inc_extent_ref(trans, &ref); |
79787eaa | 852 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 853 | } |
5893dfb9 | 854 | key.offset = args->start; |
6643558d | 855 | } |
62fe51c1 JB |
856 | /* |
857 | * From here on out we will have actually dropped something, so | |
858 | * last_end can be updated. | |
859 | */ | |
860 | last_end = extent_end; | |
861 | ||
920bbbfb YZ |
862 | /* |
863 | * | ---- range to drop ----- | | |
864 | * | -------- extent -------- | | |
865 | */ | |
5893dfb9 | 866 | if (args->start <= key.offset && args->end < extent_end) { |
00fdf13a | 867 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
3f9e3df8 | 868 | ret = -EOPNOTSUPP; |
00fdf13a LB |
869 | break; |
870 | } | |
6643558d | 871 | |
920bbbfb | 872 | memcpy(&new_key, &key, sizeof(new_key)); |
5893dfb9 | 873 | new_key.offset = args->end; |
0b246afa | 874 | btrfs_set_item_key_safe(fs_info, path, &new_key); |
6643558d | 875 | |
5893dfb9 | 876 | extent_offset += args->end - key.offset; |
920bbbfb YZ |
877 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); |
878 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
5893dfb9 | 879 | extent_end - args->end); |
920bbbfb | 880 | btrfs_mark_buffer_dirty(leaf); |
2671485d | 881 | if (update_refs && disk_bytenr > 0) |
2766ff61 | 882 | args->bytes_found += args->end - key.offset; |
920bbbfb | 883 | break; |
39279cc3 | 884 | } |
771ed689 | 885 | |
920bbbfb YZ |
886 | search_start = extent_end; |
887 | /* | |
888 | * | ---- range to drop ----- | | |
889 | * | -------- extent -------- | | |
890 | */ | |
5893dfb9 | 891 | if (args->start > key.offset && args->end >= extent_end) { |
920bbbfb | 892 | BUG_ON(del_nr > 0); |
00fdf13a | 893 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
3f9e3df8 | 894 | ret = -EOPNOTSUPP; |
00fdf13a LB |
895 | break; |
896 | } | |
8c2383c3 | 897 | |
920bbbfb | 898 | btrfs_set_file_extent_num_bytes(leaf, fi, |
5893dfb9 | 899 | args->start - key.offset); |
920bbbfb | 900 | btrfs_mark_buffer_dirty(leaf); |
2671485d | 901 | if (update_refs && disk_bytenr > 0) |
2766ff61 | 902 | args->bytes_found += extent_end - args->start; |
5893dfb9 | 903 | if (args->end == extent_end) |
920bbbfb | 904 | break; |
c8b97818 | 905 | |
920bbbfb YZ |
906 | path->slots[0]++; |
907 | goto next_slot; | |
31840ae1 ZY |
908 | } |
909 | ||
920bbbfb YZ |
910 | /* |
911 | * | ---- range to drop ----- | | |
912 | * | ------ extent ------ | | |
913 | */ | |
5893dfb9 | 914 | if (args->start <= key.offset && args->end >= extent_end) { |
fc19c5e7 | 915 | delete_extent_item: |
920bbbfb YZ |
916 | if (del_nr == 0) { |
917 | del_slot = path->slots[0]; | |
918 | del_nr = 1; | |
919 | } else { | |
920 | BUG_ON(del_slot + del_nr != path->slots[0]); | |
921 | del_nr++; | |
922 | } | |
31840ae1 | 923 | |
5dc562c5 JB |
924 | if (update_refs && |
925 | extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
2766ff61 | 926 | args->bytes_found += extent_end - key.offset; |
920bbbfb | 927 | extent_end = ALIGN(extent_end, |
0b246afa | 928 | fs_info->sectorsize); |
5dc562c5 | 929 | } else if (update_refs && disk_bytenr > 0) { |
ffd4bb2a QW |
930 | btrfs_init_generic_ref(&ref, |
931 | BTRFS_DROP_DELAYED_REF, | |
932 | disk_bytenr, num_bytes, 0); | |
933 | btrfs_init_data_ref(&ref, | |
920bbbfb | 934 | root->root_key.objectid, |
ffd4bb2a | 935 | key.objectid, |
f42c5da6 NB |
936 | key.offset - extent_offset, 0, |
937 | false); | |
ffd4bb2a | 938 | ret = btrfs_free_extent(trans, &ref); |
79787eaa | 939 | BUG_ON(ret); /* -ENOMEM */ |
2766ff61 | 940 | args->bytes_found += extent_end - key.offset; |
31840ae1 | 941 | } |
31840ae1 | 942 | |
5893dfb9 | 943 | if (args->end == extent_end) |
920bbbfb YZ |
944 | break; |
945 | ||
946 | if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) { | |
947 | path->slots[0]++; | |
948 | goto next_slot; | |
949 | } | |
950 | ||
951 | ret = btrfs_del_items(trans, root, path, del_slot, | |
952 | del_nr); | |
79787eaa | 953 | if (ret) { |
66642832 | 954 | btrfs_abort_transaction(trans, ret); |
5dc562c5 | 955 | break; |
79787eaa | 956 | } |
920bbbfb YZ |
957 | |
958 | del_nr = 0; | |
959 | del_slot = 0; | |
960 | ||
b3b4aa74 | 961 | btrfs_release_path(path); |
920bbbfb | 962 | continue; |
39279cc3 | 963 | } |
920bbbfb | 964 | |
290342f6 | 965 | BUG(); |
39279cc3 | 966 | } |
920bbbfb | 967 | |
79787eaa | 968 | if (!ret && del_nr > 0) { |
1acae57b FDBM |
969 | /* |
970 | * Set path->slots[0] to first slot, so that after the delete | |
971 | * if items are move off from our leaf to its immediate left or | |
972 | * right neighbor leafs, we end up with a correct and adjusted | |
5893dfb9 | 973 | * path->slots[0] for our insertion (if args->replace_extent). |
1acae57b FDBM |
974 | */ |
975 | path->slots[0] = del_slot; | |
920bbbfb | 976 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
79787eaa | 977 | if (ret) |
66642832 | 978 | btrfs_abort_transaction(trans, ret); |
d5f37527 | 979 | } |
1acae57b | 980 | |
d5f37527 FDBM |
981 | leaf = path->nodes[0]; |
982 | /* | |
983 | * If btrfs_del_items() was called, it might have deleted a leaf, in | |
984 | * which case it unlocked our path, so check path->locks[0] matches a | |
985 | * write lock. | |
986 | */ | |
7ecb4c31 | 987 | if (!ret && args->replace_extent && |
ac5887c8 | 988 | path->locks[0] == BTRFS_WRITE_LOCK && |
e902baac | 989 | btrfs_leaf_free_space(leaf) >= |
5893dfb9 | 990 | sizeof(struct btrfs_item) + args->extent_item_size) { |
d5f37527 FDBM |
991 | |
992 | key.objectid = ino; | |
993 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5893dfb9 | 994 | key.offset = args->start; |
d5f37527 FDBM |
995 | if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) { |
996 | struct btrfs_key slot_key; | |
997 | ||
998 | btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]); | |
999 | if (btrfs_comp_cpu_keys(&key, &slot_key) > 0) | |
1000 | path->slots[0]++; | |
1acae57b | 1001 | } |
f0641656 | 1002 | btrfs_setup_item_for_insert(root, path, &key, args->extent_item_size); |
5893dfb9 | 1003 | args->extent_inserted = true; |
6643558d | 1004 | } |
920bbbfb | 1005 | |
5893dfb9 FM |
1006 | if (!args->path) |
1007 | btrfs_free_path(path); | |
1008 | else if (!args->extent_inserted) | |
1acae57b | 1009 | btrfs_release_path(path); |
5893dfb9 FM |
1010 | out: |
1011 | args->drop_end = found ? min(args->end, last_end) : args->end; | |
5dc562c5 | 1012 | |
39279cc3 CM |
1013 | return ret; |
1014 | } | |
1015 | ||
d899e052 | 1016 | static int extent_mergeable(struct extent_buffer *leaf, int slot, |
6c7d54ac YZ |
1017 | u64 objectid, u64 bytenr, u64 orig_offset, |
1018 | u64 *start, u64 *end) | |
d899e052 YZ |
1019 | { |
1020 | struct btrfs_file_extent_item *fi; | |
1021 | struct btrfs_key key; | |
1022 | u64 extent_end; | |
1023 | ||
1024 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) | |
1025 | return 0; | |
1026 | ||
1027 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
1028 | if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY) | |
1029 | return 0; | |
1030 | ||
1031 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
1032 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG || | |
1033 | btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr || | |
6c7d54ac | 1034 | btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset || |
d899e052 YZ |
1035 | btrfs_file_extent_compression(leaf, fi) || |
1036 | btrfs_file_extent_encryption(leaf, fi) || | |
1037 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1038 | return 0; | |
1039 | ||
1040 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
1041 | if ((*start && *start != key.offset) || (*end && *end != extent_end)) | |
1042 | return 0; | |
1043 | ||
1044 | *start = key.offset; | |
1045 | *end = extent_end; | |
1046 | return 1; | |
1047 | } | |
1048 | ||
1049 | /* | |
1050 | * Mark extent in the range start - end as written. | |
1051 | * | |
1052 | * This changes extent type from 'pre-allocated' to 'regular'. If only | |
1053 | * part of extent is marked as written, the extent will be split into | |
1054 | * two or three. | |
1055 | */ | |
1056 | int btrfs_mark_extent_written(struct btrfs_trans_handle *trans, | |
7a6d7067 | 1057 | struct btrfs_inode *inode, u64 start, u64 end) |
d899e052 | 1058 | { |
3ffbd68c | 1059 | struct btrfs_fs_info *fs_info = trans->fs_info; |
7a6d7067 | 1060 | struct btrfs_root *root = inode->root; |
d899e052 YZ |
1061 | struct extent_buffer *leaf; |
1062 | struct btrfs_path *path; | |
1063 | struct btrfs_file_extent_item *fi; | |
82fa113f | 1064 | struct btrfs_ref ref = { 0 }; |
d899e052 | 1065 | struct btrfs_key key; |
920bbbfb | 1066 | struct btrfs_key new_key; |
d899e052 YZ |
1067 | u64 bytenr; |
1068 | u64 num_bytes; | |
1069 | u64 extent_end; | |
5d4f98a2 | 1070 | u64 orig_offset; |
d899e052 YZ |
1071 | u64 other_start; |
1072 | u64 other_end; | |
920bbbfb YZ |
1073 | u64 split; |
1074 | int del_nr = 0; | |
1075 | int del_slot = 0; | |
6c7d54ac | 1076 | int recow; |
e7b2ec3d | 1077 | int ret = 0; |
7a6d7067 | 1078 | u64 ino = btrfs_ino(inode); |
d899e052 | 1079 | |
d899e052 | 1080 | path = btrfs_alloc_path(); |
d8926bb3 MF |
1081 | if (!path) |
1082 | return -ENOMEM; | |
d899e052 | 1083 | again: |
6c7d54ac | 1084 | recow = 0; |
920bbbfb | 1085 | split = start; |
33345d01 | 1086 | key.objectid = ino; |
d899e052 | 1087 | key.type = BTRFS_EXTENT_DATA_KEY; |
920bbbfb | 1088 | key.offset = split; |
d899e052 YZ |
1089 | |
1090 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
41415730 JB |
1091 | if (ret < 0) |
1092 | goto out; | |
d899e052 YZ |
1093 | if (ret > 0 && path->slots[0] > 0) |
1094 | path->slots[0]--; | |
1095 | ||
1096 | leaf = path->nodes[0]; | |
1097 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
9c8e63db JB |
1098 | if (key.objectid != ino || |
1099 | key.type != BTRFS_EXTENT_DATA_KEY) { | |
1100 | ret = -EINVAL; | |
1101 | btrfs_abort_transaction(trans, ret); | |
1102 | goto out; | |
1103 | } | |
d899e052 YZ |
1104 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1105 | struct btrfs_file_extent_item); | |
9c8e63db JB |
1106 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_PREALLOC) { |
1107 | ret = -EINVAL; | |
1108 | btrfs_abort_transaction(trans, ret); | |
1109 | goto out; | |
1110 | } | |
d899e052 | 1111 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
9c8e63db JB |
1112 | if (key.offset > start || extent_end < end) { |
1113 | ret = -EINVAL; | |
1114 | btrfs_abort_transaction(trans, ret); | |
1115 | goto out; | |
1116 | } | |
d899e052 YZ |
1117 | |
1118 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
1119 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
5d4f98a2 | 1120 | orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi); |
6c7d54ac YZ |
1121 | memcpy(&new_key, &key, sizeof(new_key)); |
1122 | ||
1123 | if (start == key.offset && end < extent_end) { | |
1124 | other_start = 0; | |
1125 | other_end = start; | |
1126 | if (extent_mergeable(leaf, path->slots[0] - 1, | |
33345d01 | 1127 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1128 | &other_start, &other_end)) { |
1129 | new_key.offset = end; | |
0b246afa | 1130 | btrfs_set_item_key_safe(fs_info, path, &new_key); |
6c7d54ac YZ |
1131 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1132 | struct btrfs_file_extent_item); | |
224ecce5 JB |
1133 | btrfs_set_file_extent_generation(leaf, fi, |
1134 | trans->transid); | |
6c7d54ac YZ |
1135 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1136 | extent_end - end); | |
1137 | btrfs_set_file_extent_offset(leaf, fi, | |
1138 | end - orig_offset); | |
1139 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
1140 | struct btrfs_file_extent_item); | |
224ecce5 JB |
1141 | btrfs_set_file_extent_generation(leaf, fi, |
1142 | trans->transid); | |
6c7d54ac YZ |
1143 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1144 | end - other_start); | |
1145 | btrfs_mark_buffer_dirty(leaf); | |
1146 | goto out; | |
1147 | } | |
1148 | } | |
1149 | ||
1150 | if (start > key.offset && end == extent_end) { | |
1151 | other_start = end; | |
1152 | other_end = 0; | |
1153 | if (extent_mergeable(leaf, path->slots[0] + 1, | |
33345d01 | 1154 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1155 | &other_start, &other_end)) { |
1156 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1157 | struct btrfs_file_extent_item); | |
1158 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
1159 | start - key.offset); | |
224ecce5 JB |
1160 | btrfs_set_file_extent_generation(leaf, fi, |
1161 | trans->transid); | |
6c7d54ac YZ |
1162 | path->slots[0]++; |
1163 | new_key.offset = start; | |
0b246afa | 1164 | btrfs_set_item_key_safe(fs_info, path, &new_key); |
6c7d54ac YZ |
1165 | |
1166 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1167 | struct btrfs_file_extent_item); | |
224ecce5 JB |
1168 | btrfs_set_file_extent_generation(leaf, fi, |
1169 | trans->transid); | |
6c7d54ac YZ |
1170 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1171 | other_end - start); | |
1172 | btrfs_set_file_extent_offset(leaf, fi, | |
1173 | start - orig_offset); | |
1174 | btrfs_mark_buffer_dirty(leaf); | |
1175 | goto out; | |
1176 | } | |
1177 | } | |
d899e052 | 1178 | |
920bbbfb YZ |
1179 | while (start > key.offset || end < extent_end) { |
1180 | if (key.offset == start) | |
1181 | split = end; | |
1182 | ||
920bbbfb YZ |
1183 | new_key.offset = split; |
1184 | ret = btrfs_duplicate_item(trans, root, path, &new_key); | |
1185 | if (ret == -EAGAIN) { | |
b3b4aa74 | 1186 | btrfs_release_path(path); |
920bbbfb | 1187 | goto again; |
d899e052 | 1188 | } |
79787eaa | 1189 | if (ret < 0) { |
66642832 | 1190 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
1191 | goto out; |
1192 | } | |
d899e052 | 1193 | |
920bbbfb YZ |
1194 | leaf = path->nodes[0]; |
1195 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
d899e052 | 1196 | struct btrfs_file_extent_item); |
224ecce5 | 1197 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
d899e052 | 1198 | btrfs_set_file_extent_num_bytes(leaf, fi, |
920bbbfb YZ |
1199 | split - key.offset); |
1200 | ||
1201 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1202 | struct btrfs_file_extent_item); | |
1203 | ||
224ecce5 | 1204 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
920bbbfb YZ |
1205 | btrfs_set_file_extent_offset(leaf, fi, split - orig_offset); |
1206 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
1207 | extent_end - split); | |
d899e052 YZ |
1208 | btrfs_mark_buffer_dirty(leaf); |
1209 | ||
82fa113f QW |
1210 | btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, bytenr, |
1211 | num_bytes, 0); | |
1212 | btrfs_init_data_ref(&ref, root->root_key.objectid, ino, | |
f42c5da6 | 1213 | orig_offset, 0, false); |
82fa113f | 1214 | ret = btrfs_inc_extent_ref(trans, &ref); |
9c8e63db JB |
1215 | if (ret) { |
1216 | btrfs_abort_transaction(trans, ret); | |
1217 | goto out; | |
1218 | } | |
d899e052 | 1219 | |
920bbbfb YZ |
1220 | if (split == start) { |
1221 | key.offset = start; | |
1222 | } else { | |
9c8e63db JB |
1223 | if (start != key.offset) { |
1224 | ret = -EINVAL; | |
1225 | btrfs_abort_transaction(trans, ret); | |
1226 | goto out; | |
1227 | } | |
d899e052 | 1228 | path->slots[0]--; |
920bbbfb | 1229 | extent_end = end; |
d899e052 | 1230 | } |
6c7d54ac | 1231 | recow = 1; |
d899e052 YZ |
1232 | } |
1233 | ||
920bbbfb YZ |
1234 | other_start = end; |
1235 | other_end = 0; | |
ffd4bb2a QW |
1236 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, |
1237 | num_bytes, 0); | |
f42c5da6 NB |
1238 | btrfs_init_data_ref(&ref, root->root_key.objectid, ino, orig_offset, |
1239 | 0, false); | |
6c7d54ac | 1240 | if (extent_mergeable(leaf, path->slots[0] + 1, |
33345d01 | 1241 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1242 | &other_start, &other_end)) { |
1243 | if (recow) { | |
b3b4aa74 | 1244 | btrfs_release_path(path); |
6c7d54ac YZ |
1245 | goto again; |
1246 | } | |
920bbbfb YZ |
1247 | extent_end = other_end; |
1248 | del_slot = path->slots[0] + 1; | |
1249 | del_nr++; | |
ffd4bb2a | 1250 | ret = btrfs_free_extent(trans, &ref); |
9c8e63db JB |
1251 | if (ret) { |
1252 | btrfs_abort_transaction(trans, ret); | |
1253 | goto out; | |
1254 | } | |
d899e052 | 1255 | } |
920bbbfb YZ |
1256 | other_start = 0; |
1257 | other_end = start; | |
6c7d54ac | 1258 | if (extent_mergeable(leaf, path->slots[0] - 1, |
33345d01 | 1259 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1260 | &other_start, &other_end)) { |
1261 | if (recow) { | |
b3b4aa74 | 1262 | btrfs_release_path(path); |
6c7d54ac YZ |
1263 | goto again; |
1264 | } | |
920bbbfb YZ |
1265 | key.offset = other_start; |
1266 | del_slot = path->slots[0]; | |
1267 | del_nr++; | |
ffd4bb2a | 1268 | ret = btrfs_free_extent(trans, &ref); |
9c8e63db JB |
1269 | if (ret) { |
1270 | btrfs_abort_transaction(trans, ret); | |
1271 | goto out; | |
1272 | } | |
920bbbfb YZ |
1273 | } |
1274 | if (del_nr == 0) { | |
3f6fae95 SL |
1275 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1276 | struct btrfs_file_extent_item); | |
920bbbfb YZ |
1277 | btrfs_set_file_extent_type(leaf, fi, |
1278 | BTRFS_FILE_EXTENT_REG); | |
224ecce5 | 1279 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
920bbbfb | 1280 | btrfs_mark_buffer_dirty(leaf); |
6c7d54ac | 1281 | } else { |
3f6fae95 SL |
1282 | fi = btrfs_item_ptr(leaf, del_slot - 1, |
1283 | struct btrfs_file_extent_item); | |
6c7d54ac YZ |
1284 | btrfs_set_file_extent_type(leaf, fi, |
1285 | BTRFS_FILE_EXTENT_REG); | |
224ecce5 | 1286 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
6c7d54ac YZ |
1287 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1288 | extent_end - key.offset); | |
1289 | btrfs_mark_buffer_dirty(leaf); | |
920bbbfb | 1290 | |
6c7d54ac | 1291 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
79787eaa | 1292 | if (ret < 0) { |
66642832 | 1293 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
1294 | goto out; |
1295 | } | |
6c7d54ac | 1296 | } |
920bbbfb | 1297 | out: |
d899e052 | 1298 | btrfs_free_path(path); |
e7b2ec3d | 1299 | return ret; |
d899e052 YZ |
1300 | } |
1301 | ||
b1bf862e CM |
1302 | /* |
1303 | * on error we return an unlocked page and the error value | |
1304 | * on success we return a locked page and 0 | |
1305 | */ | |
bb1591b4 CM |
1306 | static int prepare_uptodate_page(struct inode *inode, |
1307 | struct page *page, u64 pos, | |
b6316429 | 1308 | bool force_uptodate) |
b1bf862e | 1309 | { |
fb12489b | 1310 | struct folio *folio = page_folio(page); |
b1bf862e CM |
1311 | int ret = 0; |
1312 | ||
09cbfeaf | 1313 | if (((pos & (PAGE_SIZE - 1)) || force_uptodate) && |
b6316429 | 1314 | !PageUptodate(page)) { |
fb12489b | 1315 | ret = btrfs_read_folio(NULL, folio); |
b1bf862e CM |
1316 | if (ret) |
1317 | return ret; | |
1318 | lock_page(page); | |
1319 | if (!PageUptodate(page)) { | |
1320 | unlock_page(page); | |
1321 | return -EIO; | |
1322 | } | |
e0467866 QW |
1323 | |
1324 | /* | |
fb12489b | 1325 | * Since btrfs_read_folio() will unlock the folio before it |
f913cff3 | 1326 | * returns, there is a window where btrfs_release_folio() can be |
7c11d0ae QW |
1327 | * called to release the page. Here we check both inode |
1328 | * mapping and PagePrivate() to make sure the page was not | |
1329 | * released. | |
e0467866 QW |
1330 | * |
1331 | * The private flag check is essential for subpage as we need | |
1332 | * to store extra bitmap using page->private. | |
1333 | */ | |
1334 | if (page->mapping != inode->i_mapping || !PagePrivate(page)) { | |
bb1591b4 CM |
1335 | unlock_page(page); |
1336 | return -EAGAIN; | |
1337 | } | |
b1bf862e CM |
1338 | } |
1339 | return 0; | |
1340 | } | |
1341 | ||
fc226000 SR |
1342 | static unsigned int get_prepare_fgp_flags(bool nowait) |
1343 | { | |
1344 | unsigned int fgp_flags = FGP_LOCK | FGP_ACCESSED | FGP_CREAT; | |
1345 | ||
1346 | if (nowait) | |
1347 | fgp_flags |= FGP_NOWAIT; | |
1348 | ||
1349 | return fgp_flags; | |
1350 | } | |
1351 | ||
1352 | static gfp_t get_prepare_gfp_flags(struct inode *inode, bool nowait) | |
1353 | { | |
1354 | gfp_t gfp; | |
1355 | ||
1356 | gfp = btrfs_alloc_write_mask(inode->i_mapping); | |
1357 | if (nowait) { | |
1358 | gfp &= ~__GFP_DIRECT_RECLAIM; | |
1359 | gfp |= GFP_NOWAIT; | |
1360 | } | |
1361 | ||
1362 | return gfp; | |
1363 | } | |
1364 | ||
39279cc3 | 1365 | /* |
376cc685 | 1366 | * this just gets pages into the page cache and locks them down. |
39279cc3 | 1367 | */ |
b37392ea MX |
1368 | static noinline int prepare_pages(struct inode *inode, struct page **pages, |
1369 | size_t num_pages, loff_t pos, | |
fc226000 SR |
1370 | size_t write_bytes, bool force_uptodate, |
1371 | bool nowait) | |
39279cc3 CM |
1372 | { |
1373 | int i; | |
09cbfeaf | 1374 | unsigned long index = pos >> PAGE_SHIFT; |
fc226000 SR |
1375 | gfp_t mask = get_prepare_gfp_flags(inode, nowait); |
1376 | unsigned int fgp_flags = get_prepare_fgp_flags(nowait); | |
fc28b62d | 1377 | int err = 0; |
376cc685 | 1378 | int faili; |
8c2383c3 | 1379 | |
39279cc3 | 1380 | for (i = 0; i < num_pages; i++) { |
bb1591b4 | 1381 | again: |
fc226000 SR |
1382 | pages[i] = pagecache_get_page(inode->i_mapping, index + i, |
1383 | fgp_flags, mask | __GFP_WRITE); | |
39279cc3 | 1384 | if (!pages[i]) { |
b1bf862e | 1385 | faili = i - 1; |
fc226000 SR |
1386 | if (nowait) |
1387 | err = -EAGAIN; | |
1388 | else | |
1389 | err = -ENOMEM; | |
b1bf862e CM |
1390 | goto fail; |
1391 | } | |
1392 | ||
32443de3 QW |
1393 | err = set_page_extent_mapped(pages[i]); |
1394 | if (err < 0) { | |
1395 | faili = i; | |
1396 | goto fail; | |
1397 | } | |
1398 | ||
b1bf862e | 1399 | if (i == 0) |
bb1591b4 | 1400 | err = prepare_uptodate_page(inode, pages[i], pos, |
b6316429 | 1401 | force_uptodate); |
bb1591b4 CM |
1402 | if (!err && i == num_pages - 1) |
1403 | err = prepare_uptodate_page(inode, pages[i], | |
b6316429 | 1404 | pos + write_bytes, false); |
b1bf862e | 1405 | if (err) { |
09cbfeaf | 1406 | put_page(pages[i]); |
fc226000 | 1407 | if (!nowait && err == -EAGAIN) { |
bb1591b4 CM |
1408 | err = 0; |
1409 | goto again; | |
1410 | } | |
b1bf862e CM |
1411 | faili = i - 1; |
1412 | goto fail; | |
39279cc3 | 1413 | } |
ccd467d6 | 1414 | wait_on_page_writeback(pages[i]); |
39279cc3 | 1415 | } |
376cc685 MX |
1416 | |
1417 | return 0; | |
1418 | fail: | |
1419 | while (faili >= 0) { | |
1420 | unlock_page(pages[faili]); | |
09cbfeaf | 1421 | put_page(pages[faili]); |
376cc685 MX |
1422 | faili--; |
1423 | } | |
1424 | return err; | |
1425 | ||
1426 | } | |
1427 | ||
1428 | /* | |
1429 | * This function locks the extent and properly waits for data=ordered extents | |
1430 | * to finish before allowing the pages to be modified if need. | |
1431 | * | |
1432 | * The return value: | |
1433 | * 1 - the extent is locked | |
1434 | * 0 - the extent is not locked, and everything is OK | |
1435 | * -EAGAIN - need re-prepare the pages | |
1436 | * the other < 0 number - Something wrong happens | |
1437 | */ | |
1438 | static noinline int | |
2cff578c | 1439 | lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages, |
376cc685 | 1440 | size_t num_pages, loff_t pos, |
2e78c927 | 1441 | size_t write_bytes, |
2fcab928 | 1442 | u64 *lockstart, u64 *lockend, bool nowait, |
376cc685 MX |
1443 | struct extent_state **cached_state) |
1444 | { | |
3ffbd68c | 1445 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
376cc685 MX |
1446 | u64 start_pos; |
1447 | u64 last_pos; | |
1448 | int i; | |
1449 | int ret = 0; | |
1450 | ||
0b246afa | 1451 | start_pos = round_down(pos, fs_info->sectorsize); |
e21139c6 | 1452 | last_pos = round_up(pos + write_bytes, fs_info->sectorsize) - 1; |
376cc685 | 1453 | |
e3b8a485 | 1454 | if (start_pos < inode->vfs_inode.i_size) { |
e6dcd2dc | 1455 | struct btrfs_ordered_extent *ordered; |
a7e3b975 | 1456 | |
2fcab928 SR |
1457 | if (nowait) { |
1458 | if (!try_lock_extent(&inode->io_tree, start_pos, last_pos)) { | |
1459 | for (i = 0; i < num_pages; i++) { | |
1460 | unlock_page(pages[i]); | |
1461 | put_page(pages[i]); | |
1462 | pages[i] = NULL; | |
1463 | } | |
1464 | ||
1465 | return -EAGAIN; | |
1466 | } | |
1467 | } else { | |
1468 | lock_extent(&inode->io_tree, start_pos, last_pos, cached_state); | |
1469 | } | |
1470 | ||
b88935bf MX |
1471 | ordered = btrfs_lookup_ordered_range(inode, start_pos, |
1472 | last_pos - start_pos + 1); | |
e6dcd2dc | 1473 | if (ordered && |
bffe633e | 1474 | ordered->file_offset + ordered->num_bytes > start_pos && |
376cc685 | 1475 | ordered->file_offset <= last_pos) { |
570eb97b JB |
1476 | unlock_extent(&inode->io_tree, start_pos, last_pos, |
1477 | cached_state); | |
e6dcd2dc CM |
1478 | for (i = 0; i < num_pages; i++) { |
1479 | unlock_page(pages[i]); | |
09cbfeaf | 1480 | put_page(pages[i]); |
e6dcd2dc | 1481 | } |
c0a43603 | 1482 | btrfs_start_ordered_extent(ordered, 1); |
b88935bf MX |
1483 | btrfs_put_ordered_extent(ordered); |
1484 | return -EAGAIN; | |
e6dcd2dc CM |
1485 | } |
1486 | if (ordered) | |
1487 | btrfs_put_ordered_extent(ordered); | |
7703bdd8 | 1488 | |
376cc685 MX |
1489 | *lockstart = start_pos; |
1490 | *lockend = last_pos; | |
1491 | ret = 1; | |
0762704b | 1492 | } |
376cc685 | 1493 | |
7703bdd8 | 1494 | /* |
32443de3 QW |
1495 | * We should be called after prepare_pages() which should have locked |
1496 | * all pages in the range. | |
7703bdd8 | 1497 | */ |
32443de3 | 1498 | for (i = 0; i < num_pages; i++) |
e6dcd2dc | 1499 | WARN_ON(!PageLocked(pages[i])); |
b1bf862e | 1500 | |
376cc685 | 1501 | return ret; |
39279cc3 CM |
1502 | } |
1503 | ||
d7a8ab4e FM |
1504 | /* |
1505 | * Check if we can do nocow write into the range [@pos, @pos + @write_bytes) | |
1506 | * | |
1507 | * @pos: File offset. | |
1508 | * @write_bytes: The length to write, will be updated to the nocow writeable | |
1509 | * range. | |
1510 | * | |
1511 | * This function will flush ordered extents in the range to ensure proper | |
1512 | * nocow checks. | |
1513 | * | |
1514 | * Return: | |
1515 | * > 0 If we can nocow, and updates @write_bytes. | |
1516 | * 0 If we can't do a nocow write. | |
1517 | * -EAGAIN If we can't do a nocow write because snapshoting of the inode's | |
1518 | * root is in progress. | |
1519 | * < 0 If an error happened. | |
1520 | * | |
1521 | * NOTE: Callers need to call btrfs_check_nocow_unlock() if we return > 0. | |
1522 | */ | |
1523 | int btrfs_check_nocow_lock(struct btrfs_inode *inode, loff_t pos, | |
80f9d241 | 1524 | size_t *write_bytes, bool nowait) |
7ee9e440 | 1525 | { |
3ffbd68c | 1526 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
85b7ab67 | 1527 | struct btrfs_root *root = inode->root; |
7ee9e440 JB |
1528 | u64 lockstart, lockend; |
1529 | u64 num_bytes; | |
1530 | int ret; | |
1531 | ||
38d37aa9 QW |
1532 | if (!(inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC))) |
1533 | return 0; | |
1534 | ||
d7a8ab4e | 1535 | if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) |
5f791ec3 | 1536 | return -EAGAIN; |
8257b2dc | 1537 | |
0b246afa | 1538 | lockstart = round_down(pos, fs_info->sectorsize); |
da17066c | 1539 | lockend = round_up(pos + *write_bytes, |
0b246afa | 1540 | fs_info->sectorsize) - 1; |
5dbb75ed | 1541 | num_bytes = lockend - lockstart + 1; |
7ee9e440 | 1542 | |
80f9d241 JB |
1543 | if (nowait) { |
1544 | if (!btrfs_try_lock_ordered_range(inode, lockstart, lockend)) { | |
1545 | btrfs_drew_write_unlock(&root->snapshot_lock); | |
1546 | return -EAGAIN; | |
1547 | } | |
1548 | } else { | |
1549 | btrfs_lock_and_flush_ordered_range(inode, lockstart, lockend, NULL); | |
1550 | } | |
85b7ab67 | 1551 | ret = can_nocow_extent(&inode->vfs_inode, lockstart, &num_bytes, |
80f9d241 JB |
1552 | NULL, NULL, NULL, nowait, false); |
1553 | if (ret <= 0) | |
d7a8ab4e | 1554 | btrfs_drew_write_unlock(&root->snapshot_lock); |
80f9d241 | 1555 | else |
c933956d MX |
1556 | *write_bytes = min_t(size_t, *write_bytes , |
1557 | num_bytes - pos + lockstart); | |
570eb97b | 1558 | unlock_extent(&inode->io_tree, lockstart, lockend, NULL); |
7ee9e440 JB |
1559 | |
1560 | return ret; | |
1561 | } | |
1562 | ||
38d37aa9 QW |
1563 | void btrfs_check_nocow_unlock(struct btrfs_inode *inode) |
1564 | { | |
1565 | btrfs_drew_write_unlock(&inode->root->snapshot_lock); | |
1566 | } | |
1567 | ||
b8d8e1fd GR |
1568 | static void update_time_for_write(struct inode *inode) |
1569 | { | |
1570 | struct timespec64 now; | |
1571 | ||
1572 | if (IS_NOCMTIME(inode)) | |
1573 | return; | |
1574 | ||
1575 | now = current_time(inode); | |
1576 | if (!timespec64_equal(&inode->i_mtime, &now)) | |
1577 | inode->i_mtime = now; | |
1578 | ||
1579 | if (!timespec64_equal(&inode->i_ctime, &now)) | |
1580 | inode->i_ctime = now; | |
1581 | ||
1582 | if (IS_I_VERSION(inode)) | |
1583 | inode_inc_iversion(inode); | |
1584 | } | |
1585 | ||
1586 | static int btrfs_write_check(struct kiocb *iocb, struct iov_iter *from, | |
1587 | size_t count) | |
1588 | { | |
1589 | struct file *file = iocb->ki_filp; | |
1590 | struct inode *inode = file_inode(file); | |
1591 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
1592 | loff_t pos = iocb->ki_pos; | |
1593 | int ret; | |
1594 | loff_t oldsize; | |
1595 | loff_t start_pos; | |
1596 | ||
d7a8ab4e FM |
1597 | /* |
1598 | * Quickly bail out on NOWAIT writes if we don't have the nodatacow or | |
1599 | * prealloc flags, as without those flags we always have to COW. We will | |
1600 | * later check if we can really COW into the target range (using | |
1601 | * can_nocow_extent() at btrfs_get_blocks_direct_write()). | |
1602 | */ | |
1603 | if ((iocb->ki_flags & IOCB_NOWAIT) && | |
1604 | !(BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC))) | |
1605 | return -EAGAIN; | |
b8d8e1fd GR |
1606 | |
1607 | current->backing_dev_info = inode_to_bdi(inode); | |
1608 | ret = file_remove_privs(file); | |
1609 | if (ret) | |
1610 | return ret; | |
1611 | ||
1612 | /* | |
1613 | * We reserve space for updating the inode when we reserve space for the | |
1614 | * extent we are going to write, so we will enospc out there. We don't | |
1615 | * need to start yet another transaction to update the inode as we will | |
1616 | * update the inode when we finish writing whatever data we write. | |
1617 | */ | |
1618 | update_time_for_write(inode); | |
1619 | ||
1620 | start_pos = round_down(pos, fs_info->sectorsize); | |
1621 | oldsize = i_size_read(inode); | |
1622 | if (start_pos > oldsize) { | |
1623 | /* Expand hole size to cover write data, preventing empty gap */ | |
1624 | loff_t end_pos = round_up(pos + count, fs_info->sectorsize); | |
1625 | ||
b06359a3 | 1626 | ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, end_pos); |
b8d8e1fd GR |
1627 | if (ret) { |
1628 | current->backing_dev_info = NULL; | |
1629 | return ret; | |
1630 | } | |
1631 | } | |
1632 | ||
1633 | return 0; | |
1634 | } | |
1635 | ||
e4af400a GR |
1636 | static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb, |
1637 | struct iov_iter *i) | |
4b46fce2 | 1638 | { |
e4af400a | 1639 | struct file *file = iocb->ki_filp; |
c3523706 | 1640 | loff_t pos; |
496ad9aa | 1641 | struct inode *inode = file_inode(file); |
0b246afa | 1642 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
11c65dcc | 1643 | struct page **pages = NULL; |
364ecf36 | 1644 | struct extent_changeset *data_reserved = NULL; |
7ee9e440 | 1645 | u64 release_bytes = 0; |
376cc685 MX |
1646 | u64 lockstart; |
1647 | u64 lockend; | |
d0215f3e JB |
1648 | size_t num_written = 0; |
1649 | int nrptrs; | |
c3523706 | 1650 | ssize_t ret; |
7ee9e440 | 1651 | bool only_release_metadata = false; |
b6316429 | 1652 | bool force_page_uptodate = false; |
5e8b9ef3 | 1653 | loff_t old_isize = i_size_read(inode); |
c3523706 | 1654 | unsigned int ilock_flags = 0; |
304e45ac | 1655 | const bool nowait = (iocb->ki_flags & IOCB_NOWAIT); |
965f47ae | 1656 | unsigned int bdp_flags = (nowait ? BDP_ASYNC : 0); |
c3523706 | 1657 | |
304e45ac | 1658 | if (nowait) |
c3523706 GR |
1659 | ilock_flags |= BTRFS_ILOCK_TRY; |
1660 | ||
1661 | ret = btrfs_inode_lock(inode, ilock_flags); | |
1662 | if (ret < 0) | |
1663 | return ret; | |
4b46fce2 | 1664 | |
c3523706 GR |
1665 | ret = generic_write_checks(iocb, i); |
1666 | if (ret <= 0) | |
1667 | goto out; | |
1668 | ||
1669 | ret = btrfs_write_check(iocb, i, ret); | |
1670 | if (ret < 0) | |
1671 | goto out; | |
1672 | ||
1673 | pos = iocb->ki_pos; | |
09cbfeaf KS |
1674 | nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_SIZE), |
1675 | PAGE_SIZE / (sizeof(struct page *))); | |
142349f5 WF |
1676 | nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied); |
1677 | nrptrs = max(nrptrs, 8); | |
31e818fe | 1678 | pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL); |
c3523706 GR |
1679 | if (!pages) { |
1680 | ret = -ENOMEM; | |
1681 | goto out; | |
1682 | } | |
ab93dbec | 1683 | |
d0215f3e | 1684 | while (iov_iter_count(i) > 0) { |
c67d970f | 1685 | struct extent_state *cached_state = NULL; |
7073017a | 1686 | size_t offset = offset_in_page(pos); |
2e78c927 | 1687 | size_t sector_offset; |
d0215f3e | 1688 | size_t write_bytes = min(iov_iter_count(i), |
09cbfeaf | 1689 | nrptrs * (size_t)PAGE_SIZE - |
8c2383c3 | 1690 | offset); |
eefa45f5 | 1691 | size_t num_pages; |
7ee9e440 | 1692 | size_t reserve_bytes; |
d0215f3e JB |
1693 | size_t dirty_pages; |
1694 | size_t copied; | |
2e78c927 CR |
1695 | size_t dirty_sectors; |
1696 | size_t num_sectors; | |
79f015f2 | 1697 | int extents_locked; |
39279cc3 | 1698 | |
914ee295 XZ |
1699 | /* |
1700 | * Fault pages before locking them in prepare_pages | |
1701 | * to avoid recursive lock | |
1702 | */ | |
a6294593 | 1703 | if (unlikely(fault_in_iov_iter_readable(i, write_bytes))) { |
914ee295 | 1704 | ret = -EFAULT; |
d0215f3e | 1705 | break; |
914ee295 XZ |
1706 | } |
1707 | ||
a0e248bb | 1708 | only_release_metadata = false; |
da17066c | 1709 | sector_offset = pos & (fs_info->sectorsize - 1); |
d9d8b2a5 | 1710 | |
364ecf36 | 1711 | extent_changeset_release(data_reserved); |
36ea6f3e NB |
1712 | ret = btrfs_check_data_free_space(BTRFS_I(inode), |
1713 | &data_reserved, pos, | |
304e45ac | 1714 | write_bytes, nowait); |
c6887cd1 | 1715 | if (ret < 0) { |
80f9d241 JB |
1716 | int can_nocow; |
1717 | ||
304e45ac SR |
1718 | if (nowait && (ret == -ENOSPC || ret == -EAGAIN)) { |
1719 | ret = -EAGAIN; | |
1720 | break; | |
1721 | } | |
1722 | ||
eefa45f5 GR |
1723 | /* |
1724 | * If we don't have to COW at the offset, reserve | |
1725 | * metadata only. write_bytes may get smaller than | |
1726 | * requested here. | |
1727 | */ | |
80f9d241 | 1728 | can_nocow = btrfs_check_nocow_lock(BTRFS_I(inode), pos, |
304e45ac | 1729 | &write_bytes, nowait); |
80f9d241 JB |
1730 | if (can_nocow < 0) |
1731 | ret = can_nocow; | |
1732 | if (can_nocow > 0) | |
1733 | ret = 0; | |
1734 | if (ret) | |
c6887cd1 | 1735 | break; |
80f9d241 | 1736 | only_release_metadata = true; |
c6887cd1 | 1737 | } |
1832a6d5 | 1738 | |
eefa45f5 GR |
1739 | num_pages = DIV_ROUND_UP(write_bytes + offset, PAGE_SIZE); |
1740 | WARN_ON(num_pages > nrptrs); | |
1741 | reserve_bytes = round_up(write_bytes + sector_offset, | |
1742 | fs_info->sectorsize); | |
8b62f87b | 1743 | WARN_ON(reserve_bytes == 0); |
9f3db423 | 1744 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), |
28c9b1e7 | 1745 | reserve_bytes, |
304e45ac | 1746 | reserve_bytes, nowait); |
7ee9e440 JB |
1747 | if (ret) { |
1748 | if (!only_release_metadata) | |
25ce28ca | 1749 | btrfs_free_reserved_data_space(BTRFS_I(inode), |
bc42bda2 QW |
1750 | data_reserved, pos, |
1751 | write_bytes); | |
8257b2dc | 1752 | else |
38d37aa9 | 1753 | btrfs_check_nocow_unlock(BTRFS_I(inode)); |
7ee9e440 JB |
1754 | break; |
1755 | } | |
1756 | ||
1757 | release_bytes = reserve_bytes; | |
376cc685 | 1758 | again: |
965f47ae SR |
1759 | ret = balance_dirty_pages_ratelimited_flags(inode->i_mapping, bdp_flags); |
1760 | if (ret) | |
1761 | break; | |
1762 | ||
4a64001f JB |
1763 | /* |
1764 | * This is going to setup the pages array with the number of | |
1765 | * pages we want, so we don't really need to worry about the | |
1766 | * contents of pages from loop to loop | |
1767 | */ | |
b37392ea | 1768 | ret = prepare_pages(inode, pages, num_pages, |
fc226000 | 1769 | pos, write_bytes, force_page_uptodate, false); |
8b62f87b JB |
1770 | if (ret) { |
1771 | btrfs_delalloc_release_extents(BTRFS_I(inode), | |
8702ba93 | 1772 | reserve_bytes); |
d0215f3e | 1773 | break; |
8b62f87b | 1774 | } |
39279cc3 | 1775 | |
79f015f2 GR |
1776 | extents_locked = lock_and_cleanup_extent_if_need( |
1777 | BTRFS_I(inode), pages, | |
2cff578c | 1778 | num_pages, pos, write_bytes, &lockstart, |
304e45ac | 1779 | &lockend, nowait, &cached_state); |
79f015f2 | 1780 | if (extents_locked < 0) { |
304e45ac | 1781 | if (!nowait && extents_locked == -EAGAIN) |
376cc685 | 1782 | goto again; |
304e45ac | 1783 | |
8b62f87b | 1784 | btrfs_delalloc_release_extents(BTRFS_I(inode), |
8702ba93 | 1785 | reserve_bytes); |
79f015f2 | 1786 | ret = extents_locked; |
376cc685 | 1787 | break; |
376cc685 MX |
1788 | } |
1789 | ||
ee22f0c4 | 1790 | copied = btrfs_copy_from_user(pos, write_bytes, pages, i); |
b1bf862e | 1791 | |
0b246afa | 1792 | num_sectors = BTRFS_BYTES_TO_BLKS(fs_info, reserve_bytes); |
56244ef1 | 1793 | dirty_sectors = round_up(copied + sector_offset, |
0b246afa JM |
1794 | fs_info->sectorsize); |
1795 | dirty_sectors = BTRFS_BYTES_TO_BLKS(fs_info, dirty_sectors); | |
56244ef1 | 1796 | |
b1bf862e CM |
1797 | /* |
1798 | * if we have trouble faulting in the pages, fall | |
1799 | * back to one page at a time | |
1800 | */ | |
1801 | if (copied < write_bytes) | |
1802 | nrptrs = 1; | |
1803 | ||
b6316429 JB |
1804 | if (copied == 0) { |
1805 | force_page_uptodate = true; | |
56244ef1 | 1806 | dirty_sectors = 0; |
b1bf862e | 1807 | dirty_pages = 0; |
b6316429 JB |
1808 | } else { |
1809 | force_page_uptodate = false; | |
ed6078f7 | 1810 | dirty_pages = DIV_ROUND_UP(copied + offset, |
09cbfeaf | 1811 | PAGE_SIZE); |
b6316429 | 1812 | } |
914ee295 | 1813 | |
2e78c927 | 1814 | if (num_sectors > dirty_sectors) { |
8b8b08cb | 1815 | /* release everything except the sectors we dirtied */ |
265fdfa6 | 1816 | release_bytes -= dirty_sectors << fs_info->sectorsize_bits; |
485290a7 | 1817 | if (only_release_metadata) { |
691fa059 | 1818 | btrfs_delalloc_release_metadata(BTRFS_I(inode), |
43b18595 | 1819 | release_bytes, true); |
485290a7 QW |
1820 | } else { |
1821 | u64 __pos; | |
1822 | ||
da17066c | 1823 | __pos = round_down(pos, |
0b246afa | 1824 | fs_info->sectorsize) + |
09cbfeaf | 1825 | (dirty_pages << PAGE_SHIFT); |
86d52921 | 1826 | btrfs_delalloc_release_space(BTRFS_I(inode), |
bc42bda2 | 1827 | data_reserved, __pos, |
43b18595 | 1828 | release_bytes, true); |
485290a7 | 1829 | } |
914ee295 XZ |
1830 | } |
1831 | ||
2e78c927 | 1832 | release_bytes = round_up(copied + sector_offset, |
0b246afa | 1833 | fs_info->sectorsize); |
376cc685 | 1834 | |
aa8c1a41 GR |
1835 | ret = btrfs_dirty_pages(BTRFS_I(inode), pages, |
1836 | dirty_pages, pos, copied, | |
1837 | &cached_state, only_release_metadata); | |
c67d970f FM |
1838 | |
1839 | /* | |
1840 | * If we have not locked the extent range, because the range's | |
1841 | * start offset is >= i_size, we might still have a non-NULL | |
1842 | * cached extent state, acquired while marking the extent range | |
1843 | * as delalloc through btrfs_dirty_pages(). Therefore free any | |
1844 | * possible cached extent state to avoid a memory leak. | |
1845 | */ | |
79f015f2 | 1846 | if (extents_locked) |
570eb97b JB |
1847 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, |
1848 | lockend, &cached_state); | |
c67d970f FM |
1849 | else |
1850 | free_extent_state(cached_state); | |
1851 | ||
8702ba93 | 1852 | btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes); |
f1de9683 | 1853 | if (ret) { |
e4f94347 | 1854 | btrfs_drop_pages(fs_info, pages, num_pages, pos, copied); |
376cc685 | 1855 | break; |
f1de9683 | 1856 | } |
39279cc3 | 1857 | |
376cc685 | 1858 | release_bytes = 0; |
8257b2dc | 1859 | if (only_release_metadata) |
38d37aa9 | 1860 | btrfs_check_nocow_unlock(BTRFS_I(inode)); |
8257b2dc | 1861 | |
e4f94347 | 1862 | btrfs_drop_pages(fs_info, pages, num_pages, pos, copied); |
f1de9683 | 1863 | |
d0215f3e JB |
1864 | cond_resched(); |
1865 | ||
914ee295 XZ |
1866 | pos += copied; |
1867 | num_written += copied; | |
d0215f3e | 1868 | } |
39279cc3 | 1869 | |
d0215f3e JB |
1870 | kfree(pages); |
1871 | ||
7ee9e440 | 1872 | if (release_bytes) { |
8257b2dc | 1873 | if (only_release_metadata) { |
38d37aa9 | 1874 | btrfs_check_nocow_unlock(BTRFS_I(inode)); |
691fa059 | 1875 | btrfs_delalloc_release_metadata(BTRFS_I(inode), |
43b18595 | 1876 | release_bytes, true); |
8257b2dc | 1877 | } else { |
86d52921 NB |
1878 | btrfs_delalloc_release_space(BTRFS_I(inode), |
1879 | data_reserved, | |
bc42bda2 | 1880 | round_down(pos, fs_info->sectorsize), |
43b18595 | 1881 | release_bytes, true); |
8257b2dc | 1882 | } |
7ee9e440 JB |
1883 | } |
1884 | ||
364ecf36 | 1885 | extent_changeset_free(data_reserved); |
5e8b9ef3 GR |
1886 | if (num_written > 0) { |
1887 | pagecache_isize_extended(inode, old_isize, iocb->ki_pos); | |
1888 | iocb->ki_pos += num_written; | |
1889 | } | |
c3523706 GR |
1890 | out: |
1891 | btrfs_inode_unlock(inode, ilock_flags); | |
d0215f3e JB |
1892 | return num_written ? num_written : ret; |
1893 | } | |
1894 | ||
4e4cabec GR |
1895 | static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info, |
1896 | const struct iov_iter *iter, loff_t offset) | |
1897 | { | |
1898 | const u32 blocksize_mask = fs_info->sectorsize - 1; | |
1899 | ||
1900 | if (offset & blocksize_mask) | |
1901 | return -EINVAL; | |
1902 | ||
1903 | if (iov_iter_alignment(iter) & blocksize_mask) | |
1904 | return -EINVAL; | |
1905 | ||
1906 | return 0; | |
1907 | } | |
1908 | ||
1909 | static ssize_t btrfs_direct_write(struct kiocb *iocb, struct iov_iter *from) | |
d0215f3e JB |
1910 | { |
1911 | struct file *file = iocb->ki_filp; | |
728404da | 1912 | struct inode *inode = file_inode(file); |
4e4cabec | 1913 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
c3523706 | 1914 | loff_t pos; |
4e4cabec | 1915 | ssize_t written = 0; |
d0215f3e | 1916 | ssize_t written_buffered; |
51bd9563 | 1917 | size_t prev_left = 0; |
d0215f3e | 1918 | loff_t endbyte; |
c3523706 GR |
1919 | ssize_t err; |
1920 | unsigned int ilock_flags = 0; | |
1921 | ||
1922 | if (iocb->ki_flags & IOCB_NOWAIT) | |
1923 | ilock_flags |= BTRFS_ILOCK_TRY; | |
1924 | ||
e9adabb9 GR |
1925 | /* If the write DIO is within EOF, use a shared lock */ |
1926 | if (iocb->ki_pos + iov_iter_count(from) <= i_size_read(inode)) | |
1927 | ilock_flags |= BTRFS_ILOCK_SHARED; | |
1928 | ||
1929 | relock: | |
c3523706 GR |
1930 | err = btrfs_inode_lock(inode, ilock_flags); |
1931 | if (err < 0) | |
1932 | return err; | |
1933 | ||
1934 | err = generic_write_checks(iocb, from); | |
1935 | if (err <= 0) { | |
1936 | btrfs_inode_unlock(inode, ilock_flags); | |
1937 | return err; | |
1938 | } | |
d0215f3e | 1939 | |
c3523706 GR |
1940 | err = btrfs_write_check(iocb, from, err); |
1941 | if (err < 0) { | |
1942 | btrfs_inode_unlock(inode, ilock_flags); | |
1943 | goto out; | |
1944 | } | |
1945 | ||
1946 | pos = iocb->ki_pos; | |
e9adabb9 GR |
1947 | /* |
1948 | * Re-check since file size may have changed just before taking the | |
1949 | * lock or pos may have changed because of O_APPEND in generic_write_check() | |
1950 | */ | |
1951 | if ((ilock_flags & BTRFS_ILOCK_SHARED) && | |
1952 | pos + iov_iter_count(from) > i_size_read(inode)) { | |
1953 | btrfs_inode_unlock(inode, ilock_flags); | |
1954 | ilock_flags &= ~BTRFS_ILOCK_SHARED; | |
1955 | goto relock; | |
1956 | } | |
c3523706 GR |
1957 | |
1958 | if (check_direct_IO(fs_info, from, pos)) { | |
1959 | btrfs_inode_unlock(inode, ilock_flags); | |
4e4cabec | 1960 | goto buffered; |
c3523706 | 1961 | } |
4e4cabec | 1962 | |
51bd9563 FM |
1963 | /* |
1964 | * The iov_iter can be mapped to the same file range we are writing to. | |
1965 | * If that's the case, then we will deadlock in the iomap code, because | |
1966 | * it first calls our callback btrfs_dio_iomap_begin(), which will create | |
1967 | * an ordered extent, and after that it will fault in the pages that the | |
1968 | * iov_iter refers to. During the fault in we end up in the readahead | |
1969 | * pages code (starting at btrfs_readahead()), which will lock the range, | |
1970 | * find that ordered extent and then wait for it to complete (at | |
1971 | * btrfs_lock_and_flush_ordered_range()), resulting in a deadlock since | |
1972 | * obviously the ordered extent can never complete as we didn't submit | |
1973 | * yet the respective bio(s). This always happens when the buffer is | |
1974 | * memory mapped to the same file range, since the iomap DIO code always | |
1975 | * invalidates pages in the target file range (after starting and waiting | |
1976 | * for any writeback). | |
1977 | * | |
1978 | * So here we disable page faults in the iov_iter and then retry if we | |
1979 | * got -EFAULT, faulting in the pages before the retry. | |
1980 | */ | |
1981 | again: | |
1982 | from->nofault = true; | |
36e8c622 | 1983 | err = btrfs_dio_rw(iocb, from, written); |
51bd9563 | 1984 | from->nofault = false; |
d0215f3e | 1985 | |
51bd9563 FM |
1986 | /* No increment (+=) because iomap returns a cumulative value. */ |
1987 | if (err > 0) | |
1988 | written = err; | |
1989 | ||
1990 | if (iov_iter_count(from) > 0 && (err == -EFAULT || err > 0)) { | |
1991 | const size_t left = iov_iter_count(from); | |
1992 | /* | |
1993 | * We have more data left to write. Try to fault in as many as | |
1994 | * possible of the remainder pages and retry. We do this without | |
1995 | * releasing and locking again the inode, to prevent races with | |
1996 | * truncate. | |
1997 | * | |
1998 | * Also, in case the iov refers to pages in the file range of the | |
1999 | * file we want to write to (due to a mmap), we could enter an | |
2000 | * infinite loop if we retry after faulting the pages in, since | |
2001 | * iomap will invalidate any pages in the range early on, before | |
2002 | * it tries to fault in the pages of the iov. So we keep track of | |
2003 | * how much was left of iov in the previous EFAULT and fallback | |
2004 | * to buffered IO in case we haven't made any progress. | |
2005 | */ | |
2006 | if (left == prev_left) { | |
2007 | err = -ENOTBLK; | |
2008 | } else { | |
2009 | fault_in_iov_iter_readable(from, left); | |
2010 | prev_left = left; | |
2011 | goto again; | |
2012 | } | |
a42fa643 GR |
2013 | } |
2014 | ||
51bd9563 FM |
2015 | btrfs_inode_unlock(inode, ilock_flags); |
2016 | ||
ac5e6669 FM |
2017 | /* |
2018 | * If 'err' is -ENOTBLK or we have not written all data, then it means | |
2019 | * we must fallback to buffered IO. | |
2020 | */ | |
51bd9563 | 2021 | if ((err < 0 && err != -ENOTBLK) || !iov_iter_count(from)) |
c3523706 | 2022 | goto out; |
d0215f3e | 2023 | |
4e4cabec | 2024 | buffered: |
ac5e6669 FM |
2025 | /* |
2026 | * If we are in a NOWAIT context, then return -EAGAIN to signal the caller | |
2027 | * it must retry the operation in a context where blocking is acceptable, | |
2028 | * since we currently don't have NOWAIT semantics support for buffered IO | |
2029 | * and may block there for many reasons (reserving space for example). | |
2030 | */ | |
2031 | if (iocb->ki_flags & IOCB_NOWAIT) { | |
2032 | err = -EAGAIN; | |
2033 | goto out; | |
2034 | } | |
2035 | ||
e4af400a GR |
2036 | pos = iocb->ki_pos; |
2037 | written_buffered = btrfs_buffered_write(iocb, from); | |
d0215f3e JB |
2038 | if (written_buffered < 0) { |
2039 | err = written_buffered; | |
2040 | goto out; | |
39279cc3 | 2041 | } |
075bdbdb FM |
2042 | /* |
2043 | * Ensure all data is persisted. We want the next direct IO read to be | |
2044 | * able to read what was just written. | |
2045 | */ | |
d0215f3e | 2046 | endbyte = pos + written_buffered - 1; |
728404da | 2047 | err = btrfs_fdatawrite_range(inode, pos, endbyte); |
075bdbdb FM |
2048 | if (err) |
2049 | goto out; | |
728404da | 2050 | err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte); |
d0215f3e JB |
2051 | if (err) |
2052 | goto out; | |
2053 | written += written_buffered; | |
867c4f93 | 2054 | iocb->ki_pos = pos + written_buffered; |
09cbfeaf KS |
2055 | invalidate_mapping_pages(file->f_mapping, pos >> PAGE_SHIFT, |
2056 | endbyte >> PAGE_SHIFT); | |
39279cc3 | 2057 | out: |
51bd9563 | 2058 | return err < 0 ? err : written; |
d0215f3e | 2059 | } |
5b92ee72 | 2060 | |
7c0c7269 OS |
2061 | static ssize_t btrfs_encoded_write(struct kiocb *iocb, struct iov_iter *from, |
2062 | const struct btrfs_ioctl_encoded_io_args *encoded) | |
2063 | { | |
2064 | struct file *file = iocb->ki_filp; | |
2065 | struct inode *inode = file_inode(file); | |
2066 | loff_t count; | |
2067 | ssize_t ret; | |
2068 | ||
2069 | btrfs_inode_lock(inode, 0); | |
2070 | count = encoded->len; | |
2071 | ret = generic_write_checks_count(iocb, &count); | |
2072 | if (ret == 0 && count != encoded->len) { | |
2073 | /* | |
2074 | * The write got truncated by generic_write_checks_count(). We | |
2075 | * can't do a partial encoded write. | |
2076 | */ | |
2077 | ret = -EFBIG; | |
2078 | } | |
2079 | if (ret || encoded->len == 0) | |
2080 | goto out; | |
2081 | ||
2082 | ret = btrfs_write_check(iocb, from, encoded->len); | |
2083 | if (ret < 0) | |
2084 | goto out; | |
2085 | ||
2086 | ret = btrfs_do_encoded_write(iocb, from, encoded); | |
2087 | out: | |
2088 | btrfs_inode_unlock(inode, 0); | |
2089 | return ret; | |
2090 | } | |
2091 | ||
2092 | ssize_t btrfs_do_write_iter(struct kiocb *iocb, struct iov_iter *from, | |
2093 | const struct btrfs_ioctl_encoded_io_args *encoded) | |
d0215f3e JB |
2094 | { |
2095 | struct file *file = iocb->ki_filp; | |
14971657 | 2096 | struct btrfs_inode *inode = BTRFS_I(file_inode(file)); |
7c0c7269 | 2097 | ssize_t num_written, num_sync; |
91b94c5d | 2098 | const bool sync = iocb_is_dsync(iocb); |
d0215f3e | 2099 | |
c86537a4 GR |
2100 | /* |
2101 | * If the fs flips readonly due to some impossible error, although we | |
2102 | * have opened a file as writable, we have to stop this write operation | |
2103 | * to ensure consistency. | |
2104 | */ | |
84961539 | 2105 | if (BTRFS_FS_ERROR(inode->root->fs_info)) |
c86537a4 GR |
2106 | return -EROFS; |
2107 | ||
926078b2 | 2108 | if (encoded && (iocb->ki_flags & IOCB_NOWAIT)) |
91f9943e CH |
2109 | return -EOPNOTSUPP; |
2110 | ||
b812ce28 | 2111 | if (sync) |
14971657 | 2112 | atomic_inc(&inode->sync_writers); |
b812ce28 | 2113 | |
7c0c7269 OS |
2114 | if (encoded) { |
2115 | num_written = btrfs_encoded_write(iocb, from, encoded); | |
2116 | num_sync = encoded->len; | |
2117 | } else if (iocb->ki_flags & IOCB_DIRECT) { | |
c1867eb3 DS |
2118 | num_written = btrfs_direct_write(iocb, from); |
2119 | num_sync = num_written; | |
7c0c7269 | 2120 | } else { |
c1867eb3 DS |
2121 | num_written = btrfs_buffered_write(iocb, from); |
2122 | num_sync = num_written; | |
7c0c7269 | 2123 | } |
d0215f3e | 2124 | |
bc0939fc FM |
2125 | btrfs_set_inode_last_sub_trans(inode); |
2126 | ||
7c0c7269 OS |
2127 | if (num_sync > 0) { |
2128 | num_sync = generic_write_sync(iocb, num_sync); | |
2129 | if (num_sync < 0) | |
2130 | num_written = num_sync; | |
2131 | } | |
0a3404dc | 2132 | |
b812ce28 | 2133 | if (sync) |
14971657 | 2134 | atomic_dec(&inode->sync_writers); |
b8d8e1fd | 2135 | |
39279cc3 | 2136 | current->backing_dev_info = NULL; |
c3523706 | 2137 | return num_written; |
39279cc3 CM |
2138 | } |
2139 | ||
7c0c7269 OS |
2140 | static ssize_t btrfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
2141 | { | |
2142 | return btrfs_do_write_iter(iocb, from, NULL); | |
2143 | } | |
2144 | ||
d397712b | 2145 | int btrfs_release_file(struct inode *inode, struct file *filp) |
e1b81e67 | 2146 | { |
23b5ec74 JB |
2147 | struct btrfs_file_private *private = filp->private_data; |
2148 | ||
23b5ec74 JB |
2149 | if (private && private->filldir_buf) |
2150 | kfree(private->filldir_buf); | |
2151 | kfree(private); | |
2152 | filp->private_data = NULL; | |
2153 | ||
f6dc45c7 | 2154 | /* |
1fd4033d NB |
2155 | * Set by setattr when we are about to truncate a file from a non-zero |
2156 | * size to a zero size. This tries to flush down new bytes that may | |
2157 | * have been written if the application were using truncate to replace | |
2158 | * a file in place. | |
f6dc45c7 | 2159 | */ |
1fd4033d | 2160 | if (test_and_clear_bit(BTRFS_INODE_FLUSH_ON_CLOSE, |
f6dc45c7 CM |
2161 | &BTRFS_I(inode)->runtime_flags)) |
2162 | filemap_flush(inode->i_mapping); | |
e1b81e67 M |
2163 | return 0; |
2164 | } | |
2165 | ||
669249ee FM |
2166 | static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end) |
2167 | { | |
2168 | int ret; | |
343e4fc1 | 2169 | struct blk_plug plug; |
669249ee | 2170 | |
343e4fc1 LB |
2171 | /* |
2172 | * This is only called in fsync, which would do synchronous writes, so | |
2173 | * a plug can merge adjacent IOs as much as possible. Esp. in case of | |
2174 | * multiple disks using raid profile, a large IO can be split to | |
2175 | * several segments of stripe length (currently 64K). | |
2176 | */ | |
2177 | blk_start_plug(&plug); | |
669249ee | 2178 | atomic_inc(&BTRFS_I(inode)->sync_writers); |
728404da | 2179 | ret = btrfs_fdatawrite_range(inode, start, end); |
669249ee | 2180 | atomic_dec(&BTRFS_I(inode)->sync_writers); |
343e4fc1 | 2181 | blk_finish_plug(&plug); |
669249ee FM |
2182 | |
2183 | return ret; | |
2184 | } | |
2185 | ||
626e9f41 FM |
2186 | static inline bool skip_inode_logging(const struct btrfs_log_ctx *ctx) |
2187 | { | |
2188 | struct btrfs_inode *inode = BTRFS_I(ctx->inode); | |
2189 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
2190 | ||
2191 | if (btrfs_inode_in_log(inode, fs_info->generation) && | |
2192 | list_empty(&ctx->ordered_extents)) | |
2193 | return true; | |
2194 | ||
2195 | /* | |
2196 | * If we are doing a fast fsync we can not bail out if the inode's | |
2197 | * last_trans is <= then the last committed transaction, because we only | |
2198 | * update the last_trans of the inode during ordered extent completion, | |
2199 | * and for a fast fsync we don't wait for that, we only wait for the | |
2200 | * writeback to complete. | |
2201 | */ | |
2202 | if (inode->last_trans <= fs_info->last_trans_committed && | |
2203 | (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) || | |
2204 | list_empty(&ctx->ordered_extents))) | |
2205 | return true; | |
2206 | ||
2207 | return false; | |
2208 | } | |
2209 | ||
d352ac68 CM |
2210 | /* |
2211 | * fsync call for both files and directories. This logs the inode into | |
2212 | * the tree log instead of forcing full commits whenever possible. | |
2213 | * | |
2214 | * It needs to call filemap_fdatawait so that all ordered extent updates are | |
2215 | * in the metadata btree are up to date for copying to the log. | |
2216 | * | |
2217 | * It drops the inode mutex before doing the tree log commit. This is an | |
2218 | * important optimization for directories because holding the mutex prevents | |
2219 | * new operations on the dir while we write to disk. | |
2220 | */ | |
02c24a82 | 2221 | int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync) |
39279cc3 | 2222 | { |
de17e793 | 2223 | struct dentry *dentry = file_dentry(file); |
2b0143b5 | 2224 | struct inode *inode = d_inode(dentry); |
0b246afa | 2225 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 2226 | struct btrfs_root *root = BTRFS_I(inode)->root; |
39279cc3 | 2227 | struct btrfs_trans_handle *trans; |
8b050d35 | 2228 | struct btrfs_log_ctx ctx; |
333427a5 | 2229 | int ret = 0, err; |
48778179 FM |
2230 | u64 len; |
2231 | bool full_sync; | |
39279cc3 | 2232 | |
1abe9b8a | 2233 | trace_btrfs_sync_file(file, datasync); |
257c62e1 | 2234 | |
ebb70442 LB |
2235 | btrfs_init_log_ctx(&ctx, inode); |
2236 | ||
95418ed1 | 2237 | /* |
48778179 FM |
2238 | * Always set the range to a full range, otherwise we can get into |
2239 | * several problems, from missing file extent items to represent holes | |
2240 | * when not using the NO_HOLES feature, to log tree corruption due to | |
2241 | * races between hole detection during logging and completion of ordered | |
2242 | * extents outside the range, to missing checksums due to ordered extents | |
2243 | * for which we flushed only a subset of their pages. | |
95418ed1 | 2244 | */ |
48778179 FM |
2245 | start = 0; |
2246 | end = LLONG_MAX; | |
2247 | len = (u64)LLONG_MAX + 1; | |
95418ed1 | 2248 | |
90abccf2 MX |
2249 | /* |
2250 | * We write the dirty pages in the range and wait until they complete | |
2251 | * out of the ->i_mutex. If so, we can flush the dirty pages by | |
2ab28f32 JB |
2252 | * multi-task, and make the performance up. See |
2253 | * btrfs_wait_ordered_range for an explanation of the ASYNC check. | |
90abccf2 | 2254 | */ |
669249ee | 2255 | ret = start_ordered_ops(inode, start, end); |
90abccf2 | 2256 | if (ret) |
333427a5 | 2257 | goto out; |
90abccf2 | 2258 | |
885f46d8 | 2259 | btrfs_inode_lock(inode, BTRFS_ILOCK_MMAP); |
c495144b | 2260 | |
2ecb7923 | 2261 | atomic_inc(&root->log_batch); |
b5e6c3e1 | 2262 | |
7af59743 | 2263 | /* |
48778179 FM |
2264 | * Always check for the full sync flag while holding the inode's lock, |
2265 | * to avoid races with other tasks. The flag must be either set all the | |
2266 | * time during logging or always off all the time while logging. | |
7af59743 | 2267 | */ |
48778179 FM |
2268 | full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
2269 | &BTRFS_I(inode)->runtime_flags); | |
7af59743 | 2270 | |
aab15e8e | 2271 | /* |
885f46d8 FM |
2272 | * Before we acquired the inode's lock and the mmap lock, someone may |
2273 | * have dirtied more pages in the target range. We need to make sure | |
2274 | * that writeback for any such pages does not start while we are logging | |
2275 | * the inode, because if it does, any of the following might happen when | |
2276 | * we are not doing a full inode sync: | |
aab15e8e FM |
2277 | * |
2278 | * 1) We log an extent after its writeback finishes but before its | |
2279 | * checksums are added to the csum tree, leading to -EIO errors | |
2280 | * when attempting to read the extent after a log replay. | |
2281 | * | |
2282 | * 2) We can end up logging an extent before its writeback finishes. | |
2283 | * Therefore after the log replay we will have a file extent item | |
2284 | * pointing to an unwritten extent (and no data checksums as well). | |
2285 | * | |
2286 | * So trigger writeback for any eventual new dirty pages and then we | |
2287 | * wait for all ordered extents to complete below. | |
2288 | */ | |
2289 | ret = start_ordered_ops(inode, start, end); | |
2290 | if (ret) { | |
885f46d8 | 2291 | btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP); |
aab15e8e FM |
2292 | goto out; |
2293 | } | |
2294 | ||
669249ee | 2295 | /* |
b5e6c3e1 | 2296 | * We have to do this here to avoid the priority inversion of waiting on |
52042d8e | 2297 | * IO of a lower priority task while holding a transaction open. |
ba0b084a | 2298 | * |
48778179 FM |
2299 | * For a full fsync we wait for the ordered extents to complete while |
2300 | * for a fast fsync we wait just for writeback to complete, and then | |
2301 | * attach the ordered extents to the transaction so that a transaction | |
2302 | * commit waits for their completion, to avoid data loss if we fsync, | |
2303 | * the current transaction commits before the ordered extents complete | |
2304 | * and a power failure happens right after that. | |
d8e3fb10 NA |
2305 | * |
2306 | * For zoned filesystem, if a write IO uses a ZONE_APPEND command, the | |
2307 | * logical address recorded in the ordered extent may change. We need | |
2308 | * to wait for the IO to stabilize the logical address. | |
669249ee | 2309 | */ |
d8e3fb10 | 2310 | if (full_sync || btrfs_is_zoned(fs_info)) { |
48778179 FM |
2311 | ret = btrfs_wait_ordered_range(inode, start, len); |
2312 | } else { | |
2313 | /* | |
2314 | * Get our ordered extents as soon as possible to avoid doing | |
2315 | * checksum lookups in the csum tree, and use instead the | |
2316 | * checksums attached to the ordered extents. | |
2317 | */ | |
2318 | btrfs_get_ordered_extents_for_logging(BTRFS_I(inode), | |
2319 | &ctx.ordered_extents); | |
2320 | ret = filemap_fdatawait_range(inode->i_mapping, start, end); | |
0ef8b726 | 2321 | } |
48778179 FM |
2322 | |
2323 | if (ret) | |
2324 | goto out_release_extents; | |
2325 | ||
2ecb7923 | 2326 | atomic_inc(&root->log_batch); |
257c62e1 | 2327 | |
a4abeea4 | 2328 | smp_mb(); |
626e9f41 | 2329 | if (skip_inode_logging(&ctx)) { |
5dc562c5 | 2330 | /* |
01327610 | 2331 | * We've had everything committed since the last time we were |
5dc562c5 JB |
2332 | * modified so clear this flag in case it was set for whatever |
2333 | * reason, it's no longer relevant. | |
2334 | */ | |
2335 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
2336 | &BTRFS_I(inode)->runtime_flags); | |
0596a904 FM |
2337 | /* |
2338 | * An ordered extent might have started before and completed | |
2339 | * already with io errors, in which case the inode was not | |
2340 | * updated and we end up here. So check the inode's mapping | |
333427a5 JL |
2341 | * for any errors that might have happened since we last |
2342 | * checked called fsync. | |
0596a904 | 2343 | */ |
333427a5 | 2344 | ret = filemap_check_wb_err(inode->i_mapping, file->f_wb_err); |
48778179 | 2345 | goto out_release_extents; |
15ee9bc7 | 2346 | } |
15ee9bc7 | 2347 | |
5039eddc JB |
2348 | /* |
2349 | * We use start here because we will need to wait on the IO to complete | |
2350 | * in btrfs_sync_log, which could require joining a transaction (for | |
2351 | * example checking cross references in the nocow path). If we use join | |
2352 | * here we could get into a situation where we're waiting on IO to | |
2353 | * happen that is blocked on a transaction trying to commit. With start | |
2354 | * we inc the extwriter counter, so we wait for all extwriters to exit | |
52042d8e | 2355 | * before we start blocking joiners. This comment is to keep somebody |
5039eddc JB |
2356 | * from thinking they are super smart and changing this to |
2357 | * btrfs_join_transaction *cough*Josef*cough*. | |
2358 | */ | |
a22285a6 YZ |
2359 | trans = btrfs_start_transaction(root, 0); |
2360 | if (IS_ERR(trans)) { | |
2361 | ret = PTR_ERR(trans); | |
48778179 | 2362 | goto out_release_extents; |
39279cc3 | 2363 | } |
d0c2f4fa | 2364 | trans->in_fsync = true; |
e02119d5 | 2365 | |
48778179 FM |
2366 | ret = btrfs_log_dentry_safe(trans, dentry, &ctx); |
2367 | btrfs_release_log_ctx_extents(&ctx); | |
02c24a82 | 2368 | if (ret < 0) { |
a0634be5 | 2369 | /* Fallthrough and commit/free transaction. */ |
f31f09f6 | 2370 | ret = BTRFS_LOG_FORCE_COMMIT; |
02c24a82 | 2371 | } |
49eb7e46 CM |
2372 | |
2373 | /* we've logged all the items and now have a consistent | |
2374 | * version of the file in the log. It is possible that | |
2375 | * someone will come in and modify the file, but that's | |
2376 | * fine because the log is consistent on disk, and we | |
2377 | * have references to all of the file's extents | |
2378 | * | |
2379 | * It is possible that someone will come in and log the | |
2380 | * file again, but that will end up using the synchronization | |
2381 | * inside btrfs_sync_log to keep things safe. | |
2382 | */ | |
885f46d8 | 2383 | btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP); |
49eb7e46 | 2384 | |
bf7ba8ee JB |
2385 | if (ret == BTRFS_NO_LOG_SYNC) { |
2386 | ret = btrfs_end_transaction(trans); | |
2387 | goto out; | |
2388 | } | |
2389 | ||
2390 | /* We successfully logged the inode, attempt to sync the log. */ | |
2391 | if (!ret) { | |
2392 | ret = btrfs_sync_log(trans, root, &ctx); | |
0ef8b726 | 2393 | if (!ret) { |
bf7ba8ee JB |
2394 | ret = btrfs_end_transaction(trans); |
2395 | goto out; | |
48778179 | 2396 | } |
bf7ba8ee JB |
2397 | } |
2398 | ||
2399 | /* | |
2400 | * At this point we need to commit the transaction because we had | |
2401 | * btrfs_need_log_full_commit() or some other error. | |
2402 | * | |
2403 | * If we didn't do a full sync we have to stop the trans handle, wait on | |
2404 | * the ordered extents, start it again and commit the transaction. If | |
2405 | * we attempt to wait on the ordered extents here we could deadlock with | |
2406 | * something like fallocate() that is holding the extent lock trying to | |
2407 | * start a transaction while some other thread is trying to commit the | |
2408 | * transaction while we (fsync) are currently holding the transaction | |
2409 | * open. | |
2410 | */ | |
2411 | if (!full_sync) { | |
3a45bb20 | 2412 | ret = btrfs_end_transaction(trans); |
bf7ba8ee JB |
2413 | if (ret) |
2414 | goto out; | |
2415 | ret = btrfs_wait_ordered_range(inode, start, len); | |
2416 | if (ret) | |
2417 | goto out; | |
2418 | ||
2419 | /* | |
2420 | * This is safe to use here because we're only interested in | |
2421 | * making sure the transaction that had the ordered extents is | |
2422 | * committed. We aren't waiting on anything past this point, | |
2423 | * we're purely getting the transaction and committing it. | |
2424 | */ | |
2425 | trans = btrfs_attach_transaction_barrier(root); | |
2426 | if (IS_ERR(trans)) { | |
2427 | ret = PTR_ERR(trans); | |
2428 | ||
2429 | /* | |
2430 | * We committed the transaction and there's no currently | |
2431 | * running transaction, this means everything we care | |
2432 | * about made it to disk and we are done. | |
2433 | */ | |
2434 | if (ret == -ENOENT) | |
2435 | ret = 0; | |
2436 | goto out; | |
2437 | } | |
e02119d5 | 2438 | } |
bf7ba8ee JB |
2439 | |
2440 | ret = btrfs_commit_transaction(trans); | |
39279cc3 | 2441 | out: |
ebb70442 | 2442 | ASSERT(list_empty(&ctx.list)); |
e09d94c9 | 2443 | ASSERT(list_empty(&ctx.conflict_inodes)); |
333427a5 JL |
2444 | err = file_check_and_advance_wb_err(file); |
2445 | if (!ret) | |
2446 | ret = err; | |
014e4ac4 | 2447 | return ret > 0 ? -EIO : ret; |
48778179 FM |
2448 | |
2449 | out_release_extents: | |
2450 | btrfs_release_log_ctx_extents(&ctx); | |
885f46d8 | 2451 | btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP); |
48778179 | 2452 | goto out; |
39279cc3 CM |
2453 | } |
2454 | ||
f0f37e2f | 2455 | static const struct vm_operations_struct btrfs_file_vm_ops = { |
92fee66d | 2456 | .fault = filemap_fault, |
f1820361 | 2457 | .map_pages = filemap_map_pages, |
9ebefb18 CM |
2458 | .page_mkwrite = btrfs_page_mkwrite, |
2459 | }; | |
2460 | ||
2461 | static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma) | |
2462 | { | |
058a457e MX |
2463 | struct address_space *mapping = filp->f_mapping; |
2464 | ||
7e0a1265 | 2465 | if (!mapping->a_ops->read_folio) |
058a457e MX |
2466 | return -ENOEXEC; |
2467 | ||
9ebefb18 | 2468 | file_accessed(filp); |
058a457e | 2469 | vma->vm_ops = &btrfs_file_vm_ops; |
058a457e | 2470 | |
9ebefb18 CM |
2471 | return 0; |
2472 | } | |
2473 | ||
35339c24 | 2474 | static int hole_mergeable(struct btrfs_inode *inode, struct extent_buffer *leaf, |
2aaa6655 JB |
2475 | int slot, u64 start, u64 end) |
2476 | { | |
2477 | struct btrfs_file_extent_item *fi; | |
2478 | struct btrfs_key key; | |
2479 | ||
2480 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) | |
2481 | return 0; | |
2482 | ||
2483 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
35339c24 | 2484 | if (key.objectid != btrfs_ino(inode) || |
2aaa6655 JB |
2485 | key.type != BTRFS_EXTENT_DATA_KEY) |
2486 | return 0; | |
2487 | ||
2488 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
2489 | ||
2490 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2491 | return 0; | |
2492 | ||
2493 | if (btrfs_file_extent_disk_bytenr(leaf, fi)) | |
2494 | return 0; | |
2495 | ||
2496 | if (key.offset == end) | |
2497 | return 1; | |
2498 | if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start) | |
2499 | return 1; | |
2500 | return 0; | |
2501 | } | |
2502 | ||
a012a74e NB |
2503 | static int fill_holes(struct btrfs_trans_handle *trans, |
2504 | struct btrfs_inode *inode, | |
2505 | struct btrfs_path *path, u64 offset, u64 end) | |
2aaa6655 | 2506 | { |
3ffbd68c | 2507 | struct btrfs_fs_info *fs_info = trans->fs_info; |
a012a74e | 2508 | struct btrfs_root *root = inode->root; |
2aaa6655 JB |
2509 | struct extent_buffer *leaf; |
2510 | struct btrfs_file_extent_item *fi; | |
2511 | struct extent_map *hole_em; | |
a012a74e | 2512 | struct extent_map_tree *em_tree = &inode->extent_tree; |
2aaa6655 JB |
2513 | struct btrfs_key key; |
2514 | int ret; | |
2515 | ||
0b246afa | 2516 | if (btrfs_fs_incompat(fs_info, NO_HOLES)) |
16e7549f JB |
2517 | goto out; |
2518 | ||
a012a74e | 2519 | key.objectid = btrfs_ino(inode); |
2aaa6655 JB |
2520 | key.type = BTRFS_EXTENT_DATA_KEY; |
2521 | key.offset = offset; | |
2522 | ||
2aaa6655 | 2523 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
f94480bd JB |
2524 | if (ret <= 0) { |
2525 | /* | |
2526 | * We should have dropped this offset, so if we find it then | |
2527 | * something has gone horribly wrong. | |
2528 | */ | |
2529 | if (ret == 0) | |
2530 | ret = -EINVAL; | |
2aaa6655 | 2531 | return ret; |
f94480bd | 2532 | } |
2aaa6655 JB |
2533 | |
2534 | leaf = path->nodes[0]; | |
a012a74e | 2535 | if (hole_mergeable(inode, leaf, path->slots[0] - 1, offset, end)) { |
2aaa6655 JB |
2536 | u64 num_bytes; |
2537 | ||
2538 | path->slots[0]--; | |
2539 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2540 | struct btrfs_file_extent_item); | |
2541 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + | |
2542 | end - offset; | |
2543 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2544 | btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes); | |
2545 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
e6e3dec6 | 2546 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
2aaa6655 JB |
2547 | btrfs_mark_buffer_dirty(leaf); |
2548 | goto out; | |
2549 | } | |
2550 | ||
1707e26d | 2551 | if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) { |
2aaa6655 JB |
2552 | u64 num_bytes; |
2553 | ||
2aaa6655 | 2554 | key.offset = offset; |
0b246afa | 2555 | btrfs_set_item_key_safe(fs_info, path, &key); |
2aaa6655 JB |
2556 | fi = btrfs_item_ptr(leaf, path->slots[0], |
2557 | struct btrfs_file_extent_item); | |
2558 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end - | |
2559 | offset; | |
2560 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2561 | btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes); | |
2562 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
e6e3dec6 | 2563 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
2aaa6655 JB |
2564 | btrfs_mark_buffer_dirty(leaf); |
2565 | goto out; | |
2566 | } | |
2567 | btrfs_release_path(path); | |
2568 | ||
d1f68ba0 OS |
2569 | ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, |
2570 | end - offset); | |
2aaa6655 JB |
2571 | if (ret) |
2572 | return ret; | |
2573 | ||
2574 | out: | |
2575 | btrfs_release_path(path); | |
2576 | ||
2577 | hole_em = alloc_extent_map(); | |
2578 | if (!hole_em) { | |
2579 | btrfs_drop_extent_cache(inode, offset, end - 1, 0); | |
23e3337f | 2580 | btrfs_set_inode_full_sync(inode); |
2aaa6655 JB |
2581 | } else { |
2582 | hole_em->start = offset; | |
2583 | hole_em->len = end - offset; | |
cc95bef6 | 2584 | hole_em->ram_bytes = hole_em->len; |
2aaa6655 JB |
2585 | hole_em->orig_start = offset; |
2586 | ||
2587 | hole_em->block_start = EXTENT_MAP_HOLE; | |
2588 | hole_em->block_len = 0; | |
b4939680 | 2589 | hole_em->orig_block_len = 0; |
2aaa6655 JB |
2590 | hole_em->compress_type = BTRFS_COMPRESS_NONE; |
2591 | hole_em->generation = trans->transid; | |
2592 | ||
2593 | do { | |
2594 | btrfs_drop_extent_cache(inode, offset, end - 1, 0); | |
2595 | write_lock(&em_tree->lock); | |
09a2a8f9 | 2596 | ret = add_extent_mapping(em_tree, hole_em, 1); |
2aaa6655 JB |
2597 | write_unlock(&em_tree->lock); |
2598 | } while (ret == -EEXIST); | |
2599 | free_extent_map(hole_em); | |
2600 | if (ret) | |
23e3337f | 2601 | btrfs_set_inode_full_sync(inode); |
2aaa6655 JB |
2602 | } |
2603 | ||
2604 | return 0; | |
2605 | } | |
2606 | ||
d7781546 QW |
2607 | /* |
2608 | * Find a hole extent on given inode and change start/len to the end of hole | |
2609 | * extent.(hole/vacuum extent whose em->start <= start && | |
2610 | * em->start + em->len > start) | |
2611 | * When a hole extent is found, return 1 and modify start/len. | |
2612 | */ | |
dea46d84 | 2613 | static int find_first_non_hole(struct btrfs_inode *inode, u64 *start, u64 *len) |
d7781546 | 2614 | { |
dea46d84 | 2615 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
d7781546 QW |
2616 | struct extent_map *em; |
2617 | int ret = 0; | |
2618 | ||
dea46d84 | 2619 | em = btrfs_get_extent(inode, NULL, 0, |
609805d8 | 2620 | round_down(*start, fs_info->sectorsize), |
39b07b5d | 2621 | round_up(*len, fs_info->sectorsize)); |
9986277e DC |
2622 | if (IS_ERR(em)) |
2623 | return PTR_ERR(em); | |
d7781546 QW |
2624 | |
2625 | /* Hole or vacuum extent(only exists in no-hole mode) */ | |
2626 | if (em->block_start == EXTENT_MAP_HOLE) { | |
2627 | ret = 1; | |
2628 | *len = em->start + em->len > *start + *len ? | |
2629 | 0 : *start + *len - em->start - em->len; | |
2630 | *start = em->start + em->len; | |
2631 | } | |
2632 | free_extent_map(em); | |
2633 | return ret; | |
2634 | } | |
2635 | ||
55961c8a FM |
2636 | static void btrfs_punch_hole_lock_range(struct inode *inode, |
2637 | const u64 lockstart, | |
2638 | const u64 lockend, | |
2639 | struct extent_state **cached_state) | |
f27451f2 | 2640 | { |
0528476b QW |
2641 | /* |
2642 | * For subpage case, if the range is not at page boundary, we could | |
2643 | * have pages at the leading/tailing part of the range. | |
2644 | * This could lead to dead loop since filemap_range_has_page() | |
2645 | * will always return true. | |
2646 | * So here we need to do extra page alignment for | |
2647 | * filemap_range_has_page(). | |
2648 | */ | |
2649 | const u64 page_lockstart = round_up(lockstart, PAGE_SIZE); | |
2650 | const u64 page_lockend = round_down(lockend + 1, PAGE_SIZE) - 1; | |
2651 | ||
f27451f2 | 2652 | while (1) { |
f27451f2 FM |
2653 | truncate_pagecache_range(inode, lockstart, lockend); |
2654 | ||
570eb97b JB |
2655 | lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
2656 | cached_state); | |
f27451f2 | 2657 | /* |
55961c8a FM |
2658 | * We can't have ordered extents in the range, nor dirty/writeback |
2659 | * pages, because we have locked the inode's VFS lock in exclusive | |
2660 | * mode, we have locked the inode's i_mmap_lock in exclusive mode, | |
2661 | * we have flushed all delalloc in the range and we have waited | |
2662 | * for any ordered extents in the range to complete. | |
2663 | * We can race with anyone reading pages from this range, so after | |
2664 | * locking the range check if we have pages in the range, and if | |
2665 | * we do, unlock the range and retry. | |
f27451f2 | 2666 | */ |
55961c8a FM |
2667 | if (!filemap_range_has_page(inode->i_mapping, page_lockstart, |
2668 | page_lockend)) | |
f27451f2 | 2669 | break; |
55961c8a | 2670 | |
570eb97b JB |
2671 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
2672 | cached_state); | |
f27451f2 | 2673 | } |
63c34cb4 FM |
2674 | |
2675 | btrfs_assert_inode_range_clean(BTRFS_I(inode), lockstart, lockend); | |
f27451f2 FM |
2676 | } |
2677 | ||
0cbb5bdf | 2678 | static int btrfs_insert_replace_extent(struct btrfs_trans_handle *trans, |
03fcb1ab | 2679 | struct btrfs_inode *inode, |
690a5dbf | 2680 | struct btrfs_path *path, |
bf385648 | 2681 | struct btrfs_replace_extent_info *extent_info, |
2766ff61 FM |
2682 | const u64 replace_len, |
2683 | const u64 bytes_to_drop) | |
690a5dbf | 2684 | { |
03fcb1ab NB |
2685 | struct btrfs_fs_info *fs_info = trans->fs_info; |
2686 | struct btrfs_root *root = inode->root; | |
690a5dbf FM |
2687 | struct btrfs_file_extent_item *extent; |
2688 | struct extent_buffer *leaf; | |
2689 | struct btrfs_key key; | |
2690 | int slot; | |
2691 | struct btrfs_ref ref = { 0 }; | |
690a5dbf FM |
2692 | int ret; |
2693 | ||
bf385648 | 2694 | if (replace_len == 0) |
690a5dbf FM |
2695 | return 0; |
2696 | ||
bf385648 | 2697 | if (extent_info->disk_offset == 0 && |
2766ff61 | 2698 | btrfs_fs_incompat(fs_info, NO_HOLES)) { |
03fcb1ab | 2699 | btrfs_update_inode_bytes(inode, 0, bytes_to_drop); |
690a5dbf | 2700 | return 0; |
2766ff61 | 2701 | } |
690a5dbf | 2702 | |
03fcb1ab | 2703 | key.objectid = btrfs_ino(inode); |
690a5dbf | 2704 | key.type = BTRFS_EXTENT_DATA_KEY; |
bf385648 | 2705 | key.offset = extent_info->file_offset; |
690a5dbf | 2706 | ret = btrfs_insert_empty_item(trans, root, path, &key, |
fb870f6c | 2707 | sizeof(struct btrfs_file_extent_item)); |
690a5dbf FM |
2708 | if (ret) |
2709 | return ret; | |
2710 | leaf = path->nodes[0]; | |
2711 | slot = path->slots[0]; | |
bf385648 | 2712 | write_extent_buffer(leaf, extent_info->extent_buf, |
690a5dbf | 2713 | btrfs_item_ptr_offset(leaf, slot), |
fb870f6c | 2714 | sizeof(struct btrfs_file_extent_item)); |
690a5dbf | 2715 | extent = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); |
fb870f6c | 2716 | ASSERT(btrfs_file_extent_type(leaf, extent) != BTRFS_FILE_EXTENT_INLINE); |
bf385648 FM |
2717 | btrfs_set_file_extent_offset(leaf, extent, extent_info->data_offset); |
2718 | btrfs_set_file_extent_num_bytes(leaf, extent, replace_len); | |
2719 | if (extent_info->is_new_extent) | |
8fccebfa | 2720 | btrfs_set_file_extent_generation(leaf, extent, trans->transid); |
690a5dbf FM |
2721 | btrfs_mark_buffer_dirty(leaf); |
2722 | btrfs_release_path(path); | |
2723 | ||
03fcb1ab NB |
2724 | ret = btrfs_inode_set_file_extent_range(inode, extent_info->file_offset, |
2725 | replace_len); | |
9ddc959e JB |
2726 | if (ret) |
2727 | return ret; | |
2728 | ||
690a5dbf | 2729 | /* If it's a hole, nothing more needs to be done. */ |
2766ff61 | 2730 | if (extent_info->disk_offset == 0) { |
03fcb1ab | 2731 | btrfs_update_inode_bytes(inode, 0, bytes_to_drop); |
690a5dbf | 2732 | return 0; |
2766ff61 | 2733 | } |
690a5dbf | 2734 | |
03fcb1ab | 2735 | btrfs_update_inode_bytes(inode, replace_len, bytes_to_drop); |
8fccebfa | 2736 | |
bf385648 FM |
2737 | if (extent_info->is_new_extent && extent_info->insertions == 0) { |
2738 | key.objectid = extent_info->disk_offset; | |
8fccebfa | 2739 | key.type = BTRFS_EXTENT_ITEM_KEY; |
bf385648 | 2740 | key.offset = extent_info->disk_len; |
8fccebfa | 2741 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
03fcb1ab | 2742 | btrfs_ino(inode), |
bf385648 FM |
2743 | extent_info->file_offset, |
2744 | extent_info->qgroup_reserved, | |
8fccebfa FM |
2745 | &key); |
2746 | } else { | |
2747 | u64 ref_offset; | |
2748 | ||
2749 | btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, | |
bf385648 FM |
2750 | extent_info->disk_offset, |
2751 | extent_info->disk_len, 0); | |
2752 | ref_offset = extent_info->file_offset - extent_info->data_offset; | |
8fccebfa | 2753 | btrfs_init_data_ref(&ref, root->root_key.objectid, |
f42c5da6 | 2754 | btrfs_ino(inode), ref_offset, 0, false); |
8fccebfa FM |
2755 | ret = btrfs_inc_extent_ref(trans, &ref); |
2756 | } | |
2757 | ||
bf385648 | 2758 | extent_info->insertions++; |
690a5dbf FM |
2759 | |
2760 | return ret; | |
2761 | } | |
2762 | ||
9cba40a6 FM |
2763 | /* |
2764 | * The respective range must have been previously locked, as well as the inode. | |
2765 | * The end offset is inclusive (last byte of the range). | |
bf385648 FM |
2766 | * @extent_info is NULL for fallocate's hole punching and non-NULL when replacing |
2767 | * the file range with an extent. | |
2768 | * When not punching a hole, we don't want to end up in a state where we dropped | |
2769 | * extents without inserting a new one, so we must abort the transaction to avoid | |
2770 | * a corruption. | |
9cba40a6 | 2771 | */ |
bfc78479 NB |
2772 | int btrfs_replace_file_extents(struct btrfs_inode *inode, |
2773 | struct btrfs_path *path, const u64 start, | |
2774 | const u64 end, | |
2775 | struct btrfs_replace_extent_info *extent_info, | |
2776 | struct btrfs_trans_handle **trans_out) | |
9cba40a6 | 2777 | { |
5893dfb9 | 2778 | struct btrfs_drop_extents_args drop_args = { 0 }; |
bfc78479 NB |
2779 | struct btrfs_root *root = inode->root; |
2780 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2bd36e7b | 2781 | u64 min_size = btrfs_calc_insert_metadata_size(fs_info, 1); |
bfc78479 | 2782 | u64 ino_size = round_up(inode->vfs_inode.i_size, fs_info->sectorsize); |
9cba40a6 FM |
2783 | struct btrfs_trans_handle *trans = NULL; |
2784 | struct btrfs_block_rsv *rsv; | |
2785 | unsigned int rsv_count; | |
2786 | u64 cur_offset; | |
9cba40a6 FM |
2787 | u64 len = end - start; |
2788 | int ret = 0; | |
2789 | ||
2790 | if (end <= start) | |
2791 | return -EINVAL; | |
2792 | ||
2793 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); | |
2794 | if (!rsv) { | |
2795 | ret = -ENOMEM; | |
2796 | goto out; | |
2797 | } | |
2bd36e7b | 2798 | rsv->size = btrfs_calc_insert_metadata_size(fs_info, 1); |
710d5921 | 2799 | rsv->failfast = true; |
9cba40a6 FM |
2800 | |
2801 | /* | |
2802 | * 1 - update the inode | |
2803 | * 1 - removing the extents in the range | |
bf385648 FM |
2804 | * 1 - adding the hole extent if no_holes isn't set or if we are |
2805 | * replacing the range with a new extent | |
9cba40a6 | 2806 | */ |
bf385648 | 2807 | if (!btrfs_fs_incompat(fs_info, NO_HOLES) || extent_info) |
690a5dbf FM |
2808 | rsv_count = 3; |
2809 | else | |
2810 | rsv_count = 2; | |
2811 | ||
9cba40a6 FM |
2812 | trans = btrfs_start_transaction(root, rsv_count); |
2813 | if (IS_ERR(trans)) { | |
2814 | ret = PTR_ERR(trans); | |
2815 | trans = NULL; | |
2816 | goto out_free; | |
2817 | } | |
2818 | ||
2819 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv, | |
2820 | min_size, false); | |
650c9cab FM |
2821 | if (WARN_ON(ret)) |
2822 | goto out_trans; | |
9cba40a6 FM |
2823 | trans->block_rsv = rsv; |
2824 | ||
2825 | cur_offset = start; | |
5893dfb9 FM |
2826 | drop_args.path = path; |
2827 | drop_args.end = end + 1; | |
2828 | drop_args.drop_cache = true; | |
9cba40a6 | 2829 | while (cur_offset < end) { |
5893dfb9 | 2830 | drop_args.start = cur_offset; |
bfc78479 | 2831 | ret = btrfs_drop_extents(trans, root, inode, &drop_args); |
2766ff61 FM |
2832 | /* If we are punching a hole decrement the inode's byte count */ |
2833 | if (!extent_info) | |
bfc78479 | 2834 | btrfs_update_inode_bytes(inode, 0, |
2766ff61 | 2835 | drop_args.bytes_found); |
690a5dbf FM |
2836 | if (ret != -ENOSPC) { |
2837 | /* | |
4afb912f JB |
2838 | * The only time we don't want to abort is if we are |
2839 | * attempting to clone a partial inline extent, in which | |
2840 | * case we'll get EOPNOTSUPP. However if we aren't | |
2841 | * clone we need to abort no matter what, because if we | |
2842 | * got EOPNOTSUPP via prealloc then we messed up and | |
2843 | * need to abort. | |
690a5dbf | 2844 | */ |
4afb912f JB |
2845 | if (ret && |
2846 | (ret != -EOPNOTSUPP || | |
2847 | (extent_info && extent_info->is_new_extent))) | |
690a5dbf | 2848 | btrfs_abort_transaction(trans, ret); |
9cba40a6 | 2849 | break; |
690a5dbf | 2850 | } |
9cba40a6 FM |
2851 | |
2852 | trans->block_rsv = &fs_info->trans_block_rsv; | |
2853 | ||
5893dfb9 | 2854 | if (!extent_info && cur_offset < drop_args.drop_end && |
690a5dbf | 2855 | cur_offset < ino_size) { |
bfc78479 NB |
2856 | ret = fill_holes(trans, inode, path, cur_offset, |
2857 | drop_args.drop_end); | |
9cba40a6 FM |
2858 | if (ret) { |
2859 | /* | |
2860 | * If we failed then we didn't insert our hole | |
2861 | * entries for the area we dropped, so now the | |
2862 | * fs is corrupted, so we must abort the | |
2863 | * transaction. | |
2864 | */ | |
2865 | btrfs_abort_transaction(trans, ret); | |
2866 | break; | |
2867 | } | |
5893dfb9 | 2868 | } else if (!extent_info && cur_offset < drop_args.drop_end) { |
9ddc959e JB |
2869 | /* |
2870 | * We are past the i_size here, but since we didn't | |
2871 | * insert holes we need to clear the mapped area so we | |
2872 | * know to not set disk_i_size in this area until a new | |
2873 | * file extent is inserted here. | |
2874 | */ | |
bfc78479 | 2875 | ret = btrfs_inode_clear_file_extent_range(inode, |
5893dfb9 FM |
2876 | cur_offset, |
2877 | drop_args.drop_end - cur_offset); | |
9ddc959e JB |
2878 | if (ret) { |
2879 | /* | |
2880 | * We couldn't clear our area, so we could | |
2881 | * presumably adjust up and corrupt the fs, so | |
2882 | * we need to abort. | |
2883 | */ | |
2884 | btrfs_abort_transaction(trans, ret); | |
2885 | break; | |
2886 | } | |
9cba40a6 FM |
2887 | } |
2888 | ||
5893dfb9 FM |
2889 | if (extent_info && |
2890 | drop_args.drop_end > extent_info->file_offset) { | |
2891 | u64 replace_len = drop_args.drop_end - | |
2892 | extent_info->file_offset; | |
690a5dbf | 2893 | |
bfc78479 NB |
2894 | ret = btrfs_insert_replace_extent(trans, inode, path, |
2895 | extent_info, replace_len, | |
03fcb1ab | 2896 | drop_args.bytes_found); |
690a5dbf FM |
2897 | if (ret) { |
2898 | btrfs_abort_transaction(trans, ret); | |
2899 | break; | |
2900 | } | |
bf385648 FM |
2901 | extent_info->data_len -= replace_len; |
2902 | extent_info->data_offset += replace_len; | |
2903 | extent_info->file_offset += replace_len; | |
690a5dbf FM |
2904 | } |
2905 | ||
983d8209 FM |
2906 | /* |
2907 | * We are releasing our handle on the transaction, balance the | |
2908 | * dirty pages of the btree inode and flush delayed items, and | |
2909 | * then get a new transaction handle, which may now point to a | |
2910 | * new transaction in case someone else may have committed the | |
2911 | * transaction we used to replace/drop file extent items. So | |
2912 | * bump the inode's iversion and update mtime and ctime except | |
2913 | * if we are called from a dedupe context. This is because a | |
2914 | * power failure/crash may happen after the transaction is | |
2915 | * committed and before we finish replacing/dropping all the | |
2916 | * file extent items we need. | |
2917 | */ | |
2918 | inode_inc_iversion(&inode->vfs_inode); | |
2919 | ||
2920 | if (!extent_info || extent_info->update_times) { | |
2921 | inode->vfs_inode.i_mtime = current_time(&inode->vfs_inode); | |
2922 | inode->vfs_inode.i_ctime = inode->vfs_inode.i_mtime; | |
2923 | } | |
2924 | ||
bfc78479 | 2925 | ret = btrfs_update_inode(trans, root, inode); |
9cba40a6 FM |
2926 | if (ret) |
2927 | break; | |
2928 | ||
2929 | btrfs_end_transaction(trans); | |
2930 | btrfs_btree_balance_dirty(fs_info); | |
2931 | ||
2932 | trans = btrfs_start_transaction(root, rsv_count); | |
2933 | if (IS_ERR(trans)) { | |
2934 | ret = PTR_ERR(trans); | |
2935 | trans = NULL; | |
2936 | break; | |
2937 | } | |
2938 | ||
2939 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, | |
2940 | rsv, min_size, false); | |
650c9cab FM |
2941 | if (WARN_ON(ret)) |
2942 | break; | |
9cba40a6 FM |
2943 | trans->block_rsv = rsv; |
2944 | ||
3227788c BC |
2945 | cur_offset = drop_args.drop_end; |
2946 | len = end - cur_offset; | |
2947 | if (!extent_info && len) { | |
bfc78479 | 2948 | ret = find_first_non_hole(inode, &cur_offset, &len); |
690a5dbf FM |
2949 | if (unlikely(ret < 0)) |
2950 | break; | |
2951 | if (ret && !len) { | |
2952 | ret = 0; | |
2953 | break; | |
2954 | } | |
9cba40a6 FM |
2955 | } |
2956 | } | |
2957 | ||
690a5dbf FM |
2958 | /* |
2959 | * If we were cloning, force the next fsync to be a full one since we | |
2960 | * we replaced (or just dropped in the case of cloning holes when | |
e2b84217 FM |
2961 | * NO_HOLES is enabled) file extent items and did not setup new extent |
2962 | * maps for the replacement extents (or holes). | |
690a5dbf | 2963 | */ |
bf385648 | 2964 | if (extent_info && !extent_info->is_new_extent) |
23e3337f | 2965 | btrfs_set_inode_full_sync(inode); |
690a5dbf | 2966 | |
9cba40a6 FM |
2967 | if (ret) |
2968 | goto out_trans; | |
2969 | ||
2970 | trans->block_rsv = &fs_info->trans_block_rsv; | |
2971 | /* | |
2972 | * If we are using the NO_HOLES feature we might have had already an | |
2973 | * hole that overlaps a part of the region [lockstart, lockend] and | |
2974 | * ends at (or beyond) lockend. Since we have no file extent items to | |
2975 | * represent holes, drop_end can be less than lockend and so we must | |
2976 | * make sure we have an extent map representing the existing hole (the | |
2977 | * call to __btrfs_drop_extents() might have dropped the existing extent | |
2978 | * map representing the existing hole), otherwise the fast fsync path | |
2979 | * will not record the existence of the hole region | |
2980 | * [existing_hole_start, lockend]. | |
2981 | */ | |
5893dfb9 FM |
2982 | if (drop_args.drop_end <= end) |
2983 | drop_args.drop_end = end + 1; | |
9cba40a6 FM |
2984 | /* |
2985 | * Don't insert file hole extent item if it's for a range beyond eof | |
2986 | * (because it's useless) or if it represents a 0 bytes range (when | |
2987 | * cur_offset == drop_end). | |
2988 | */ | |
5893dfb9 FM |
2989 | if (!extent_info && cur_offset < ino_size && |
2990 | cur_offset < drop_args.drop_end) { | |
bfc78479 NB |
2991 | ret = fill_holes(trans, inode, path, cur_offset, |
2992 | drop_args.drop_end); | |
9cba40a6 FM |
2993 | if (ret) { |
2994 | /* Same comment as above. */ | |
2995 | btrfs_abort_transaction(trans, ret); | |
2996 | goto out_trans; | |
2997 | } | |
5893dfb9 | 2998 | } else if (!extent_info && cur_offset < drop_args.drop_end) { |
9ddc959e | 2999 | /* See the comment in the loop above for the reasoning here. */ |
bfc78479 NB |
3000 | ret = btrfs_inode_clear_file_extent_range(inode, cur_offset, |
3001 | drop_args.drop_end - cur_offset); | |
9ddc959e JB |
3002 | if (ret) { |
3003 | btrfs_abort_transaction(trans, ret); | |
3004 | goto out_trans; | |
3005 | } | |
3006 | ||
9cba40a6 | 3007 | } |
bf385648 | 3008 | if (extent_info) { |
bfc78479 | 3009 | ret = btrfs_insert_replace_extent(trans, inode, path, |
03fcb1ab NB |
3010 | extent_info, extent_info->data_len, |
3011 | drop_args.bytes_found); | |
690a5dbf FM |
3012 | if (ret) { |
3013 | btrfs_abort_transaction(trans, ret); | |
3014 | goto out_trans; | |
3015 | } | |
3016 | } | |
9cba40a6 FM |
3017 | |
3018 | out_trans: | |
3019 | if (!trans) | |
3020 | goto out_free; | |
3021 | ||
3022 | trans->block_rsv = &fs_info->trans_block_rsv; | |
3023 | if (ret) | |
3024 | btrfs_end_transaction(trans); | |
3025 | else | |
3026 | *trans_out = trans; | |
3027 | out_free: | |
3028 | btrfs_free_block_rsv(fs_info, rsv); | |
3029 | out: | |
3030 | return ret; | |
3031 | } | |
3032 | ||
05fd9564 | 3033 | static int btrfs_punch_hole(struct file *file, loff_t offset, loff_t len) |
2aaa6655 | 3034 | { |
05fd9564 | 3035 | struct inode *inode = file_inode(file); |
0b246afa | 3036 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2aaa6655 JB |
3037 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3038 | struct extent_state *cached_state = NULL; | |
3039 | struct btrfs_path *path; | |
9cba40a6 | 3040 | struct btrfs_trans_handle *trans = NULL; |
d7781546 QW |
3041 | u64 lockstart; |
3042 | u64 lockend; | |
3043 | u64 tail_start; | |
3044 | u64 tail_len; | |
3045 | u64 orig_start = offset; | |
2aaa6655 | 3046 | int ret = 0; |
9703fefe | 3047 | bool same_block; |
a1a50f60 | 3048 | u64 ino_size; |
9703fefe | 3049 | bool truncated_block = false; |
e8c1c76e | 3050 | bool updated_inode = false; |
2aaa6655 | 3051 | |
bd6526d0 FM |
3052 | btrfs_inode_lock(inode, BTRFS_ILOCK_MMAP); |
3053 | ||
0ef8b726 JB |
3054 | ret = btrfs_wait_ordered_range(inode, offset, len); |
3055 | if (ret) | |
bd6526d0 | 3056 | goto out_only_mutex; |
2aaa6655 | 3057 | |
0b246afa | 3058 | ino_size = round_up(inode->i_size, fs_info->sectorsize); |
dea46d84 | 3059 | ret = find_first_non_hole(BTRFS_I(inode), &offset, &len); |
d7781546 QW |
3060 | if (ret < 0) |
3061 | goto out_only_mutex; | |
3062 | if (ret && !len) { | |
3063 | /* Already in a large hole */ | |
3064 | ret = 0; | |
3065 | goto out_only_mutex; | |
3066 | } | |
3067 | ||
05fd9564 DW |
3068 | ret = file_modified(file); |
3069 | if (ret) | |
3070 | goto out_only_mutex; | |
3071 | ||
ee8ba05c JB |
3072 | lockstart = round_up(offset, fs_info->sectorsize); |
3073 | lockend = round_down(offset + len, fs_info->sectorsize) - 1; | |
0b246afa JM |
3074 | same_block = (BTRFS_BYTES_TO_BLKS(fs_info, offset)) |
3075 | == (BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)); | |
7426cc04 | 3076 | /* |
9703fefe | 3077 | * We needn't truncate any block which is beyond the end of the file |
7426cc04 MX |
3078 | * because we are sure there is no data there. |
3079 | */ | |
2aaa6655 | 3080 | /* |
9703fefe CR |
3081 | * Only do this if we are in the same block and we aren't doing the |
3082 | * entire block. | |
2aaa6655 | 3083 | */ |
0b246afa | 3084 | if (same_block && len < fs_info->sectorsize) { |
e8c1c76e | 3085 | if (offset < ino_size) { |
9703fefe | 3086 | truncated_block = true; |
217f42eb NB |
3087 | ret = btrfs_truncate_block(BTRFS_I(inode), offset, len, |
3088 | 0); | |
e8c1c76e FM |
3089 | } else { |
3090 | ret = 0; | |
3091 | } | |
d7781546 | 3092 | goto out_only_mutex; |
2aaa6655 JB |
3093 | } |
3094 | ||
9703fefe | 3095 | /* zero back part of the first block */ |
12870f1c | 3096 | if (offset < ino_size) { |
9703fefe | 3097 | truncated_block = true; |
217f42eb | 3098 | ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0); |
7426cc04 | 3099 | if (ret) { |
8d9b4a16 | 3100 | btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP); |
7426cc04 MX |
3101 | return ret; |
3102 | } | |
2aaa6655 JB |
3103 | } |
3104 | ||
d7781546 QW |
3105 | /* Check the aligned pages after the first unaligned page, |
3106 | * if offset != orig_start, which means the first unaligned page | |
01327610 | 3107 | * including several following pages are already in holes, |
d7781546 QW |
3108 | * the extra check can be skipped */ |
3109 | if (offset == orig_start) { | |
3110 | /* after truncate page, check hole again */ | |
3111 | len = offset + len - lockstart; | |
3112 | offset = lockstart; | |
dea46d84 | 3113 | ret = find_first_non_hole(BTRFS_I(inode), &offset, &len); |
d7781546 QW |
3114 | if (ret < 0) |
3115 | goto out_only_mutex; | |
3116 | if (ret && !len) { | |
3117 | ret = 0; | |
3118 | goto out_only_mutex; | |
3119 | } | |
3120 | lockstart = offset; | |
3121 | } | |
3122 | ||
3123 | /* Check the tail unaligned part is in a hole */ | |
3124 | tail_start = lockend + 1; | |
3125 | tail_len = offset + len - tail_start; | |
3126 | if (tail_len) { | |
dea46d84 | 3127 | ret = find_first_non_hole(BTRFS_I(inode), &tail_start, &tail_len); |
d7781546 QW |
3128 | if (unlikely(ret < 0)) |
3129 | goto out_only_mutex; | |
3130 | if (!ret) { | |
3131 | /* zero the front end of the last page */ | |
3132 | if (tail_start + tail_len < ino_size) { | |
9703fefe | 3133 | truncated_block = true; |
217f42eb | 3134 | ret = btrfs_truncate_block(BTRFS_I(inode), |
9703fefe CR |
3135 | tail_start + tail_len, |
3136 | 0, 1); | |
d7781546 QW |
3137 | if (ret) |
3138 | goto out_only_mutex; | |
51f395ad | 3139 | } |
0061280d | 3140 | } |
2aaa6655 JB |
3141 | } |
3142 | ||
3143 | if (lockend < lockstart) { | |
e8c1c76e FM |
3144 | ret = 0; |
3145 | goto out_only_mutex; | |
2aaa6655 JB |
3146 | } |
3147 | ||
55961c8a | 3148 | btrfs_punch_hole_lock_range(inode, lockstart, lockend, &cached_state); |
2aaa6655 JB |
3149 | |
3150 | path = btrfs_alloc_path(); | |
3151 | if (!path) { | |
3152 | ret = -ENOMEM; | |
3153 | goto out; | |
3154 | } | |
3155 | ||
bfc78479 NB |
3156 | ret = btrfs_replace_file_extents(BTRFS_I(inode), path, lockstart, |
3157 | lockend, NULL, &trans); | |
9cba40a6 FM |
3158 | btrfs_free_path(path); |
3159 | if (ret) | |
3160 | goto out; | |
2aaa6655 | 3161 | |
9cba40a6 | 3162 | ASSERT(trans != NULL); |
e1f5790e | 3163 | inode_inc_iversion(inode); |
c1867eb3 DS |
3164 | inode->i_mtime = current_time(inode); |
3165 | inode->i_ctime = inode->i_mtime; | |
9a56fcd1 | 3166 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
e8c1c76e | 3167 | updated_inode = true; |
3a45bb20 | 3168 | btrfs_end_transaction(trans); |
2ff7e61e | 3169 | btrfs_btree_balance_dirty(fs_info); |
2aaa6655 | 3170 | out: |
570eb97b JB |
3171 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
3172 | &cached_state); | |
d7781546 | 3173 | out_only_mutex: |
9cba40a6 | 3174 | if (!updated_inode && truncated_block && !ret) { |
e8c1c76e FM |
3175 | /* |
3176 | * If we only end up zeroing part of a page, we still need to | |
3177 | * update the inode item, so that all the time fields are | |
3178 | * updated as well as the necessary btrfs inode in memory fields | |
3179 | * for detecting, at fsync time, if the inode isn't yet in the | |
3180 | * log tree or it's there but not up to date. | |
3181 | */ | |
17900668 FM |
3182 | struct timespec64 now = current_time(inode); |
3183 | ||
3184 | inode_inc_iversion(inode); | |
3185 | inode->i_mtime = now; | |
3186 | inode->i_ctime = now; | |
e8c1c76e FM |
3187 | trans = btrfs_start_transaction(root, 1); |
3188 | if (IS_ERR(trans)) { | |
9cba40a6 | 3189 | ret = PTR_ERR(trans); |
e8c1c76e | 3190 | } else { |
9cba40a6 FM |
3191 | int ret2; |
3192 | ||
9a56fcd1 | 3193 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
9cba40a6 FM |
3194 | ret2 = btrfs_end_transaction(trans); |
3195 | if (!ret) | |
3196 | ret = ret2; | |
e8c1c76e FM |
3197 | } |
3198 | } | |
8d9b4a16 | 3199 | btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP); |
9cba40a6 | 3200 | return ret; |
2aaa6655 JB |
3201 | } |
3202 | ||
14524a84 QW |
3203 | /* Helper structure to record which range is already reserved */ |
3204 | struct falloc_range { | |
3205 | struct list_head list; | |
3206 | u64 start; | |
3207 | u64 len; | |
3208 | }; | |
3209 | ||
3210 | /* | |
3211 | * Helper function to add falloc range | |
3212 | * | |
3213 | * Caller should have locked the larger range of extent containing | |
3214 | * [start, len) | |
3215 | */ | |
3216 | static int add_falloc_range(struct list_head *head, u64 start, u64 len) | |
3217 | { | |
14524a84 QW |
3218 | struct falloc_range *range = NULL; |
3219 | ||
77d25534 NB |
3220 | if (!list_empty(head)) { |
3221 | /* | |
3222 | * As fallocate iterates by bytenr order, we only need to check | |
3223 | * the last range. | |
3224 | */ | |
3225 | range = list_last_entry(head, struct falloc_range, list); | |
3226 | if (range->start + range->len == start) { | |
3227 | range->len += len; | |
3228 | return 0; | |
3229 | } | |
14524a84 | 3230 | } |
77d25534 | 3231 | |
32fc932e | 3232 | range = kmalloc(sizeof(*range), GFP_KERNEL); |
14524a84 QW |
3233 | if (!range) |
3234 | return -ENOMEM; | |
3235 | range->start = start; | |
3236 | range->len = len; | |
3237 | list_add_tail(&range->list, head); | |
3238 | return 0; | |
3239 | } | |
3240 | ||
f27451f2 FM |
3241 | static int btrfs_fallocate_update_isize(struct inode *inode, |
3242 | const u64 end, | |
3243 | const int mode) | |
3244 | { | |
3245 | struct btrfs_trans_handle *trans; | |
3246 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3247 | int ret; | |
3248 | int ret2; | |
3249 | ||
3250 | if (mode & FALLOC_FL_KEEP_SIZE || end <= i_size_read(inode)) | |
3251 | return 0; | |
3252 | ||
3253 | trans = btrfs_start_transaction(root, 1); | |
3254 | if (IS_ERR(trans)) | |
3255 | return PTR_ERR(trans); | |
3256 | ||
3257 | inode->i_ctime = current_time(inode); | |
3258 | i_size_write(inode, end); | |
76aea537 | 3259 | btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0); |
9a56fcd1 | 3260 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
f27451f2 FM |
3261 | ret2 = btrfs_end_transaction(trans); |
3262 | ||
3263 | return ret ? ret : ret2; | |
3264 | } | |
3265 | ||
81fdf638 | 3266 | enum { |
f262fa8d DS |
3267 | RANGE_BOUNDARY_WRITTEN_EXTENT, |
3268 | RANGE_BOUNDARY_PREALLOC_EXTENT, | |
3269 | RANGE_BOUNDARY_HOLE, | |
81fdf638 FM |
3270 | }; |
3271 | ||
948dfeb8 | 3272 | static int btrfs_zero_range_check_range_boundary(struct btrfs_inode *inode, |
f27451f2 FM |
3273 | u64 offset) |
3274 | { | |
ee8ba05c | 3275 | const u64 sectorsize = inode->root->fs_info->sectorsize; |
f27451f2 | 3276 | struct extent_map *em; |
81fdf638 | 3277 | int ret; |
f27451f2 FM |
3278 | |
3279 | offset = round_down(offset, sectorsize); | |
948dfeb8 | 3280 | em = btrfs_get_extent(inode, NULL, 0, offset, sectorsize); |
f27451f2 FM |
3281 | if (IS_ERR(em)) |
3282 | return PTR_ERR(em); | |
3283 | ||
3284 | if (em->block_start == EXTENT_MAP_HOLE) | |
81fdf638 FM |
3285 | ret = RANGE_BOUNDARY_HOLE; |
3286 | else if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
3287 | ret = RANGE_BOUNDARY_PREALLOC_EXTENT; | |
3288 | else | |
3289 | ret = RANGE_BOUNDARY_WRITTEN_EXTENT; | |
f27451f2 FM |
3290 | |
3291 | free_extent_map(em); | |
3292 | return ret; | |
3293 | } | |
3294 | ||
3295 | static int btrfs_zero_range(struct inode *inode, | |
3296 | loff_t offset, | |
3297 | loff_t len, | |
3298 | const int mode) | |
3299 | { | |
3300 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
3301 | struct extent_map *em; | |
3302 | struct extent_changeset *data_reserved = NULL; | |
3303 | int ret; | |
3304 | u64 alloc_hint = 0; | |
ee8ba05c | 3305 | const u64 sectorsize = fs_info->sectorsize; |
f27451f2 FM |
3306 | u64 alloc_start = round_down(offset, sectorsize); |
3307 | u64 alloc_end = round_up(offset + len, sectorsize); | |
3308 | u64 bytes_to_reserve = 0; | |
3309 | bool space_reserved = false; | |
3310 | ||
39b07b5d OS |
3311 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start, |
3312 | alloc_end - alloc_start); | |
f27451f2 FM |
3313 | if (IS_ERR(em)) { |
3314 | ret = PTR_ERR(em); | |
3315 | goto out; | |
3316 | } | |
3317 | ||
3318 | /* | |
3319 | * Avoid hole punching and extent allocation for some cases. More cases | |
3320 | * could be considered, but these are unlikely common and we keep things | |
3321 | * as simple as possible for now. Also, intentionally, if the target | |
3322 | * range contains one or more prealloc extents together with regular | |
3323 | * extents and holes, we drop all the existing extents and allocate a | |
3324 | * new prealloc extent, so that we get a larger contiguous disk extent. | |
3325 | */ | |
3326 | if (em->start <= alloc_start && | |
3327 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { | |
3328 | const u64 em_end = em->start + em->len; | |
3329 | ||
3330 | if (em_end >= offset + len) { | |
3331 | /* | |
3332 | * The whole range is already a prealloc extent, | |
3333 | * do nothing except updating the inode's i_size if | |
3334 | * needed. | |
3335 | */ | |
3336 | free_extent_map(em); | |
3337 | ret = btrfs_fallocate_update_isize(inode, offset + len, | |
3338 | mode); | |
3339 | goto out; | |
3340 | } | |
3341 | /* | |
3342 | * Part of the range is already a prealloc extent, so operate | |
3343 | * only on the remaining part of the range. | |
3344 | */ | |
3345 | alloc_start = em_end; | |
3346 | ASSERT(IS_ALIGNED(alloc_start, sectorsize)); | |
3347 | len = offset + len - alloc_start; | |
3348 | offset = alloc_start; | |
3349 | alloc_hint = em->block_start + em->len; | |
3350 | } | |
3351 | free_extent_map(em); | |
3352 | ||
3353 | if (BTRFS_BYTES_TO_BLKS(fs_info, offset) == | |
3354 | BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)) { | |
39b07b5d OS |
3355 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start, |
3356 | sectorsize); | |
f27451f2 FM |
3357 | if (IS_ERR(em)) { |
3358 | ret = PTR_ERR(em); | |
3359 | goto out; | |
3360 | } | |
3361 | ||
3362 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { | |
3363 | free_extent_map(em); | |
3364 | ret = btrfs_fallocate_update_isize(inode, offset + len, | |
3365 | mode); | |
3366 | goto out; | |
3367 | } | |
3368 | if (len < sectorsize && em->block_start != EXTENT_MAP_HOLE) { | |
3369 | free_extent_map(em); | |
217f42eb NB |
3370 | ret = btrfs_truncate_block(BTRFS_I(inode), offset, len, |
3371 | 0); | |
f27451f2 FM |
3372 | if (!ret) |
3373 | ret = btrfs_fallocate_update_isize(inode, | |
3374 | offset + len, | |
3375 | mode); | |
3376 | return ret; | |
3377 | } | |
3378 | free_extent_map(em); | |
3379 | alloc_start = round_down(offset, sectorsize); | |
3380 | alloc_end = alloc_start + sectorsize; | |
3381 | goto reserve_space; | |
3382 | } | |
3383 | ||
3384 | alloc_start = round_up(offset, sectorsize); | |
3385 | alloc_end = round_down(offset + len, sectorsize); | |
3386 | ||
3387 | /* | |
3388 | * For unaligned ranges, check the pages at the boundaries, they might | |
3389 | * map to an extent, in which case we need to partially zero them, or | |
3390 | * they might map to a hole, in which case we need our allocation range | |
3391 | * to cover them. | |
3392 | */ | |
3393 | if (!IS_ALIGNED(offset, sectorsize)) { | |
948dfeb8 NB |
3394 | ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode), |
3395 | offset); | |
f27451f2 FM |
3396 | if (ret < 0) |
3397 | goto out; | |
81fdf638 | 3398 | if (ret == RANGE_BOUNDARY_HOLE) { |
f27451f2 FM |
3399 | alloc_start = round_down(offset, sectorsize); |
3400 | ret = 0; | |
81fdf638 | 3401 | } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) { |
217f42eb | 3402 | ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0); |
f27451f2 FM |
3403 | if (ret) |
3404 | goto out; | |
81fdf638 FM |
3405 | } else { |
3406 | ret = 0; | |
f27451f2 FM |
3407 | } |
3408 | } | |
3409 | ||
3410 | if (!IS_ALIGNED(offset + len, sectorsize)) { | |
948dfeb8 | 3411 | ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode), |
f27451f2 FM |
3412 | offset + len); |
3413 | if (ret < 0) | |
3414 | goto out; | |
81fdf638 | 3415 | if (ret == RANGE_BOUNDARY_HOLE) { |
f27451f2 FM |
3416 | alloc_end = round_up(offset + len, sectorsize); |
3417 | ret = 0; | |
81fdf638 | 3418 | } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) { |
217f42eb NB |
3419 | ret = btrfs_truncate_block(BTRFS_I(inode), offset + len, |
3420 | 0, 1); | |
f27451f2 FM |
3421 | if (ret) |
3422 | goto out; | |
81fdf638 FM |
3423 | } else { |
3424 | ret = 0; | |
f27451f2 FM |
3425 | } |
3426 | } | |
3427 | ||
3428 | reserve_space: | |
3429 | if (alloc_start < alloc_end) { | |
3430 | struct extent_state *cached_state = NULL; | |
3431 | const u64 lockstart = alloc_start; | |
3432 | const u64 lockend = alloc_end - 1; | |
3433 | ||
3434 | bytes_to_reserve = alloc_end - alloc_start; | |
3435 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), | |
3436 | bytes_to_reserve); | |
3437 | if (ret < 0) | |
3438 | goto out; | |
3439 | space_reserved = true; | |
55961c8a FM |
3440 | btrfs_punch_hole_lock_range(inode, lockstart, lockend, |
3441 | &cached_state); | |
7661a3e0 | 3442 | ret = btrfs_qgroup_reserve_data(BTRFS_I(inode), &data_reserved, |
a7f8b1c2 | 3443 | alloc_start, bytes_to_reserve); |
4f6a49de | 3444 | if (ret) { |
570eb97b JB |
3445 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, |
3446 | lockend, &cached_state); | |
a7f8b1c2 | 3447 | goto out; |
4f6a49de | 3448 | } |
f27451f2 FM |
3449 | ret = btrfs_prealloc_file_range(inode, mode, alloc_start, |
3450 | alloc_end - alloc_start, | |
3451 | i_blocksize(inode), | |
3452 | offset + len, &alloc_hint); | |
570eb97b JB |
3453 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
3454 | &cached_state); | |
f27451f2 | 3455 | /* btrfs_prealloc_file_range releases reserved space on error */ |
9f13ce74 | 3456 | if (ret) { |
f27451f2 | 3457 | space_reserved = false; |
9f13ce74 FM |
3458 | goto out; |
3459 | } | |
f27451f2 | 3460 | } |
9f13ce74 | 3461 | ret = btrfs_fallocate_update_isize(inode, offset + len, mode); |
f27451f2 FM |
3462 | out: |
3463 | if (ret && space_reserved) | |
25ce28ca | 3464 | btrfs_free_reserved_data_space(BTRFS_I(inode), data_reserved, |
f27451f2 FM |
3465 | alloc_start, bytes_to_reserve); |
3466 | extent_changeset_free(data_reserved); | |
3467 | ||
3468 | return ret; | |
3469 | } | |
3470 | ||
2fe17c10 CH |
3471 | static long btrfs_fallocate(struct file *file, int mode, |
3472 | loff_t offset, loff_t len) | |
3473 | { | |
496ad9aa | 3474 | struct inode *inode = file_inode(file); |
2fe17c10 | 3475 | struct extent_state *cached_state = NULL; |
364ecf36 | 3476 | struct extent_changeset *data_reserved = NULL; |
14524a84 QW |
3477 | struct falloc_range *range; |
3478 | struct falloc_range *tmp; | |
3479 | struct list_head reserve_list; | |
2fe17c10 CH |
3480 | u64 cur_offset; |
3481 | u64 last_byte; | |
3482 | u64 alloc_start; | |
3483 | u64 alloc_end; | |
3484 | u64 alloc_hint = 0; | |
3485 | u64 locked_end; | |
14524a84 | 3486 | u64 actual_end = 0; |
47e1d1c7 FM |
3487 | u64 data_space_needed = 0; |
3488 | u64 data_space_reserved = 0; | |
3489 | u64 qgroup_reserved = 0; | |
2fe17c10 | 3490 | struct extent_map *em; |
ee8ba05c | 3491 | int blocksize = BTRFS_I(inode)->root->fs_info->sectorsize; |
2fe17c10 CH |
3492 | int ret; |
3493 | ||
f1569c4c NA |
3494 | /* Do not allow fallocate in ZONED mode */ |
3495 | if (btrfs_is_zoned(btrfs_sb(inode->i_sb))) | |
3496 | return -EOPNOTSUPP; | |
3497 | ||
797f4277 MX |
3498 | alloc_start = round_down(offset, blocksize); |
3499 | alloc_end = round_up(offset + len, blocksize); | |
18513091 | 3500 | cur_offset = alloc_start; |
2fe17c10 | 3501 | |
2aaa6655 | 3502 | /* Make sure we aren't being give some crap mode */ |
f27451f2 FM |
3503 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | |
3504 | FALLOC_FL_ZERO_RANGE)) | |
2fe17c10 CH |
3505 | return -EOPNOTSUPP; |
3506 | ||
2aaa6655 | 3507 | if (mode & FALLOC_FL_PUNCH_HOLE) |
05fd9564 | 3508 | return btrfs_punch_hole(file, offset, len); |
2aaa6655 | 3509 | |
8d9b4a16 | 3510 | btrfs_inode_lock(inode, BTRFS_ILOCK_MMAP); |
2a162ce9 DI |
3511 | |
3512 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) { | |
3513 | ret = inode_newsize_ok(inode, offset + len); | |
3514 | if (ret) | |
3515 | goto out; | |
3516 | } | |
2fe17c10 | 3517 | |
05fd9564 DW |
3518 | ret = file_modified(file); |
3519 | if (ret) | |
3520 | goto out; | |
3521 | ||
14524a84 QW |
3522 | /* |
3523 | * TODO: Move these two operations after we have checked | |
3524 | * accurate reserved space, or fallocate can still fail but | |
3525 | * with page truncated or size expanded. | |
3526 | * | |
3527 | * But that's a minor problem and won't do much harm BTW. | |
3528 | */ | |
2fe17c10 | 3529 | if (alloc_start > inode->i_size) { |
b06359a3 | 3530 | ret = btrfs_cont_expand(BTRFS_I(inode), i_size_read(inode), |
a41ad394 | 3531 | alloc_start); |
2fe17c10 CH |
3532 | if (ret) |
3533 | goto out; | |
0f6925fa | 3534 | } else if (offset + len > inode->i_size) { |
a71754fc JB |
3535 | /* |
3536 | * If we are fallocating from the end of the file onward we | |
9703fefe CR |
3537 | * need to zero out the end of the block if i_size lands in the |
3538 | * middle of a block. | |
a71754fc | 3539 | */ |
217f42eb | 3540 | ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0); |
a71754fc JB |
3541 | if (ret) |
3542 | goto out; | |
2fe17c10 CH |
3543 | } |
3544 | ||
a71754fc | 3545 | /* |
ffa8fc60 FM |
3546 | * We have locked the inode at the VFS level (in exclusive mode) and we |
3547 | * have locked the i_mmap_lock lock (in exclusive mode). Now before | |
3548 | * locking the file range, flush all dealloc in the range and wait for | |
3549 | * all ordered extents in the range to complete. After this we can lock | |
3550 | * the file range and, due to the previous locking we did, we know there | |
3551 | * can't be more delalloc or ordered extents in the range. | |
a71754fc | 3552 | */ |
0ef8b726 JB |
3553 | ret = btrfs_wait_ordered_range(inode, alloc_start, |
3554 | alloc_end - alloc_start); | |
3555 | if (ret) | |
3556 | goto out; | |
a71754fc | 3557 | |
f27451f2 FM |
3558 | if (mode & FALLOC_FL_ZERO_RANGE) { |
3559 | ret = btrfs_zero_range(inode, offset, len, mode); | |
8d9b4a16 | 3560 | btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP); |
f27451f2 FM |
3561 | return ret; |
3562 | } | |
3563 | ||
2fe17c10 | 3564 | locked_end = alloc_end - 1; |
570eb97b JB |
3565 | lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
3566 | &cached_state); | |
2fe17c10 | 3567 | |
63c34cb4 FM |
3568 | btrfs_assert_inode_range_clean(BTRFS_I(inode), alloc_start, locked_end); |
3569 | ||
14524a84 QW |
3570 | /* First, check if we exceed the qgroup limit */ |
3571 | INIT_LIST_HEAD(&reserve_list); | |
6b7d6e93 | 3572 | while (cur_offset < alloc_end) { |
fc4f21b1 | 3573 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset, |
39b07b5d | 3574 | alloc_end - cur_offset); |
9986277e DC |
3575 | if (IS_ERR(em)) { |
3576 | ret = PTR_ERR(em); | |
79787eaa JM |
3577 | break; |
3578 | } | |
2fe17c10 | 3579 | last_byte = min(extent_map_end(em), alloc_end); |
f1e490a7 | 3580 | actual_end = min_t(u64, extent_map_end(em), offset + len); |
797f4277 | 3581 | last_byte = ALIGN(last_byte, blocksize); |
2fe17c10 CH |
3582 | if (em->block_start == EXTENT_MAP_HOLE || |
3583 | (cur_offset >= inode->i_size && | |
3584 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
47e1d1c7 FM |
3585 | const u64 range_len = last_byte - cur_offset; |
3586 | ||
3587 | ret = add_falloc_range(&reserve_list, cur_offset, range_len); | |
14524a84 QW |
3588 | if (ret < 0) { |
3589 | free_extent_map(em); | |
3590 | break; | |
3d850dd4 | 3591 | } |
7661a3e0 | 3592 | ret = btrfs_qgroup_reserve_data(BTRFS_I(inode), |
47e1d1c7 | 3593 | &data_reserved, cur_offset, range_len); |
be2d253c FM |
3594 | if (ret < 0) { |
3595 | free_extent_map(em); | |
14524a84 | 3596 | break; |
be2d253c | 3597 | } |
47e1d1c7 FM |
3598 | qgroup_reserved += range_len; |
3599 | data_space_needed += range_len; | |
2fe17c10 CH |
3600 | } |
3601 | free_extent_map(em); | |
2fe17c10 | 3602 | cur_offset = last_byte; |
14524a84 QW |
3603 | } |
3604 | ||
47e1d1c7 FM |
3605 | if (!ret && data_space_needed > 0) { |
3606 | /* | |
3607 | * We are safe to reserve space here as we can't have delalloc | |
3608 | * in the range, see above. | |
3609 | */ | |
3610 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), | |
3611 | data_space_needed); | |
3612 | if (!ret) | |
3613 | data_space_reserved = data_space_needed; | |
3614 | } | |
3615 | ||
14524a84 QW |
3616 | /* |
3617 | * If ret is still 0, means we're OK to fallocate. | |
3618 | * Or just cleanup the list and exit. | |
3619 | */ | |
3620 | list_for_each_entry_safe(range, tmp, &reserve_list, list) { | |
47e1d1c7 | 3621 | if (!ret) { |
14524a84 QW |
3622 | ret = btrfs_prealloc_file_range(inode, mode, |
3623 | range->start, | |
93407472 | 3624 | range->len, i_blocksize(inode), |
14524a84 | 3625 | offset + len, &alloc_hint); |
47e1d1c7 FM |
3626 | /* |
3627 | * btrfs_prealloc_file_range() releases space even | |
3628 | * if it returns an error. | |
3629 | */ | |
3630 | data_space_reserved -= range->len; | |
3631 | qgroup_reserved -= range->len; | |
3632 | } else if (data_space_reserved > 0) { | |
25ce28ca | 3633 | btrfs_free_reserved_data_space(BTRFS_I(inode), |
47e1d1c7 FM |
3634 | data_reserved, range->start, |
3635 | range->len); | |
3636 | data_space_reserved -= range->len; | |
3637 | qgroup_reserved -= range->len; | |
3638 | } else if (qgroup_reserved > 0) { | |
3639 | btrfs_qgroup_free_data(BTRFS_I(inode), data_reserved, | |
3640 | range->start, range->len); | |
3641 | qgroup_reserved -= range->len; | |
3642 | } | |
14524a84 QW |
3643 | list_del(&range->list); |
3644 | kfree(range); | |
3645 | } | |
3646 | if (ret < 0) | |
3647 | goto out_unlock; | |
3648 | ||
f27451f2 FM |
3649 | /* |
3650 | * We didn't need to allocate any more space, but we still extended the | |
3651 | * size of the file so we need to update i_size and the inode item. | |
3652 | */ | |
3653 | ret = btrfs_fallocate_update_isize(inode, actual_end, mode); | |
14524a84 | 3654 | out_unlock: |
570eb97b JB |
3655 | unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
3656 | &cached_state); | |
2fe17c10 | 3657 | out: |
8d9b4a16 | 3658 | btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP); |
364ecf36 | 3659 | extent_changeset_free(data_reserved); |
2fe17c10 CH |
3660 | return ret; |
3661 | } | |
3662 | ||
b6e83356 | 3663 | /* |
ac3c0d36 FM |
3664 | * Helper for btrfs_find_delalloc_in_range(). Find a subrange in a given range |
3665 | * that has unflushed and/or flushing delalloc. There might be other adjacent | |
3666 | * subranges after the one it found, so btrfs_find_delalloc_in_range() keeps | |
3667 | * looping while it gets adjacent subranges, and merging them together. | |
b6e83356 FM |
3668 | */ |
3669 | static bool find_delalloc_subrange(struct btrfs_inode *inode, u64 start, u64 end, | |
3670 | u64 *delalloc_start_ret, u64 *delalloc_end_ret) | |
3671 | { | |
3672 | const u64 len = end + 1 - start; | |
3673 | struct extent_map_tree *em_tree = &inode->extent_tree; | |
3674 | struct extent_map *em; | |
3675 | u64 em_end; | |
3676 | u64 delalloc_len; | |
3677 | ||
3678 | /* | |
3679 | * Search the io tree first for EXTENT_DELALLOC. If we find any, it | |
3680 | * means we have delalloc (dirty pages) for which writeback has not | |
3681 | * started yet. | |
3682 | */ | |
3683 | *delalloc_start_ret = start; | |
3684 | delalloc_len = count_range_bits(&inode->io_tree, delalloc_start_ret, end, | |
3685 | len, EXTENT_DELALLOC, 1); | |
3686 | /* | |
3687 | * If delalloc was found then *delalloc_start_ret has a sector size | |
3688 | * aligned value (rounded down). | |
3689 | */ | |
3690 | if (delalloc_len > 0) | |
3691 | *delalloc_end_ret = *delalloc_start_ret + delalloc_len - 1; | |
3692 | ||
3693 | /* | |
3694 | * Now also check if there's any extent map in the range that does not | |
3695 | * map to a hole or prealloc extent. We do this because: | |
3696 | * | |
3697 | * 1) When delalloc is flushed, the file range is locked, we clear the | |
3698 | * EXTENT_DELALLOC bit from the io tree and create an extent map for | |
3699 | * an allocated extent. So we might just have been called after | |
3700 | * delalloc is flushed and before the ordered extent completes and | |
3701 | * inserts the new file extent item in the subvolume's btree; | |
3702 | * | |
3703 | * 2) We may have an extent map created by flushing delalloc for a | |
3704 | * subrange that starts before the subrange we found marked with | |
3705 | * EXTENT_DELALLOC in the io tree. | |
3706 | */ | |
3707 | read_lock(&em_tree->lock); | |
3708 | em = lookup_extent_mapping(em_tree, start, len); | |
3709 | read_unlock(&em_tree->lock); | |
3710 | ||
3711 | /* extent_map_end() returns a non-inclusive end offset. */ | |
3712 | em_end = em ? extent_map_end(em) : 0; | |
3713 | ||
3714 | /* | |
3715 | * If we have a hole/prealloc extent map, check the next one if this one | |
3716 | * ends before our range's end. | |
3717 | */ | |
3718 | if (em && (em->block_start == EXTENT_MAP_HOLE || | |
3719 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) && em_end < end) { | |
3720 | struct extent_map *next_em; | |
3721 | ||
3722 | read_lock(&em_tree->lock); | |
3723 | next_em = lookup_extent_mapping(em_tree, em_end, len - em_end); | |
3724 | read_unlock(&em_tree->lock); | |
3725 | ||
3726 | free_extent_map(em); | |
3727 | em_end = next_em ? extent_map_end(next_em) : 0; | |
3728 | em = next_em; | |
3729 | } | |
3730 | ||
3731 | if (em && (em->block_start == EXTENT_MAP_HOLE || | |
3732 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
3733 | free_extent_map(em); | |
3734 | em = NULL; | |
3735 | } | |
3736 | ||
3737 | /* | |
3738 | * No extent map or one for a hole or prealloc extent. Use the delalloc | |
3739 | * range we found in the io tree if we have one. | |
3740 | */ | |
3741 | if (!em) | |
3742 | return (delalloc_len > 0); | |
3743 | ||
3744 | /* | |
3745 | * We don't have any range as EXTENT_DELALLOC in the io tree, so the | |
3746 | * extent map is the only subrange representing delalloc. | |
3747 | */ | |
3748 | if (delalloc_len == 0) { | |
3749 | *delalloc_start_ret = em->start; | |
3750 | *delalloc_end_ret = min(end, em_end - 1); | |
3751 | free_extent_map(em); | |
3752 | return true; | |
3753 | } | |
3754 | ||
3755 | /* | |
3756 | * The extent map represents a delalloc range that starts before the | |
3757 | * delalloc range we found in the io tree. | |
3758 | */ | |
3759 | if (em->start < *delalloc_start_ret) { | |
3760 | *delalloc_start_ret = em->start; | |
3761 | /* | |
3762 | * If the ranges are adjacent, return a combined range. | |
3763 | * Otherwise return the extent map's range. | |
3764 | */ | |
3765 | if (em_end < *delalloc_start_ret) | |
3766 | *delalloc_end_ret = min(end, em_end - 1); | |
3767 | ||
3768 | free_extent_map(em); | |
3769 | return true; | |
3770 | } | |
3771 | ||
3772 | /* | |
3773 | * The extent map starts after the delalloc range we found in the io | |
3774 | * tree. If it's adjacent, return a combined range, otherwise return | |
3775 | * the range found in the io tree. | |
3776 | */ | |
3777 | if (*delalloc_end_ret + 1 == em->start) | |
3778 | *delalloc_end_ret = min(end, em_end - 1); | |
3779 | ||
3780 | free_extent_map(em); | |
3781 | return true; | |
3782 | } | |
3783 | ||
3784 | /* | |
3785 | * Check if there's delalloc in a given range. | |
3786 | * | |
3787 | * @inode: The inode. | |
3788 | * @start: The start offset of the range. It does not need to be | |
3789 | * sector size aligned. | |
3790 | * @end: The end offset (inclusive value) of the search range. | |
3791 | * It does not need to be sector size aligned. | |
3792 | * @delalloc_start_ret: Output argument, set to the start offset of the | |
3793 | * subrange found with delalloc (may not be sector size | |
3794 | * aligned). | |
3795 | * @delalloc_end_ret: Output argument, set to he end offset (inclusive value) | |
3796 | * of the subrange found with delalloc. | |
3797 | * | |
3798 | * Returns true if a subrange with delalloc is found within the given range, and | |
3799 | * if so it sets @delalloc_start_ret and @delalloc_end_ret with the start and | |
3800 | * end offsets of the subrange. | |
3801 | */ | |
ac3c0d36 FM |
3802 | bool btrfs_find_delalloc_in_range(struct btrfs_inode *inode, u64 start, u64 end, |
3803 | u64 *delalloc_start_ret, u64 *delalloc_end_ret) | |
b6e83356 FM |
3804 | { |
3805 | u64 cur_offset = round_down(start, inode->root->fs_info->sectorsize); | |
3806 | u64 prev_delalloc_end = 0; | |
3807 | bool ret = false; | |
3808 | ||
3809 | while (cur_offset < end) { | |
3810 | u64 delalloc_start; | |
3811 | u64 delalloc_end; | |
3812 | bool delalloc; | |
3813 | ||
3814 | delalloc = find_delalloc_subrange(inode, cur_offset, end, | |
3815 | &delalloc_start, | |
3816 | &delalloc_end); | |
3817 | if (!delalloc) | |
3818 | break; | |
3819 | ||
3820 | if (prev_delalloc_end == 0) { | |
3821 | /* First subrange found. */ | |
3822 | *delalloc_start_ret = max(delalloc_start, start); | |
3823 | *delalloc_end_ret = delalloc_end; | |
3824 | ret = true; | |
3825 | } else if (delalloc_start == prev_delalloc_end + 1) { | |
3826 | /* Subrange adjacent to the previous one, merge them. */ | |
3827 | *delalloc_end_ret = delalloc_end; | |
3828 | } else { | |
3829 | /* Subrange not adjacent to the previous one, exit. */ | |
3830 | break; | |
3831 | } | |
3832 | ||
3833 | prev_delalloc_end = delalloc_end; | |
3834 | cur_offset = delalloc_end + 1; | |
3835 | cond_resched(); | |
3836 | } | |
3837 | ||
3838 | return ret; | |
3839 | } | |
3840 | ||
3841 | /* | |
3842 | * Check if there's a hole or delalloc range in a range representing a hole (or | |
3843 | * prealloc extent) found in the inode's subvolume btree. | |
3844 | * | |
3845 | * @inode: The inode. | |
3846 | * @whence: Seek mode (SEEK_DATA or SEEK_HOLE). | |
3847 | * @start: Start offset of the hole region. It does not need to be sector | |
3848 | * size aligned. | |
3849 | * @end: End offset (inclusive value) of the hole region. It does not | |
3850 | * need to be sector size aligned. | |
3851 | * @start_ret: Return parameter, used to set the start of the subrange in the | |
3852 | * hole that matches the search criteria (seek mode), if such | |
3853 | * subrange is found (return value of the function is true). | |
3854 | * The value returned here may not be sector size aligned. | |
3855 | * | |
3856 | * Returns true if a subrange matching the given seek mode is found, and if one | |
3857 | * is found, it updates @start_ret with the start of the subrange. | |
3858 | */ | |
3859 | static bool find_desired_extent_in_hole(struct btrfs_inode *inode, int whence, | |
3860 | u64 start, u64 end, u64 *start_ret) | |
3861 | { | |
3862 | u64 delalloc_start; | |
3863 | u64 delalloc_end; | |
3864 | bool delalloc; | |
3865 | ||
ac3c0d36 FM |
3866 | delalloc = btrfs_find_delalloc_in_range(inode, start, end, |
3867 | &delalloc_start, &delalloc_end); | |
b6e83356 FM |
3868 | if (delalloc && whence == SEEK_DATA) { |
3869 | *start_ret = delalloc_start; | |
3870 | return true; | |
3871 | } | |
3872 | ||
3873 | if (delalloc && whence == SEEK_HOLE) { | |
3874 | /* | |
3875 | * We found delalloc but it starts after out start offset. So we | |
3876 | * have a hole between our start offset and the delalloc start. | |
3877 | */ | |
3878 | if (start < delalloc_start) { | |
3879 | *start_ret = start; | |
3880 | return true; | |
3881 | } | |
3882 | /* | |
3883 | * Delalloc range starts at our start offset. | |
3884 | * If the delalloc range's length is smaller than our range, | |
3885 | * then it means we have a hole that starts where the delalloc | |
3886 | * subrange ends. | |
3887 | */ | |
3888 | if (delalloc_end < end) { | |
3889 | *start_ret = delalloc_end + 1; | |
3890 | return true; | |
3891 | } | |
3892 | ||
3893 | /* There's delalloc for the whole range. */ | |
3894 | return false; | |
3895 | } | |
3896 | ||
3897 | if (!delalloc && whence == SEEK_HOLE) { | |
3898 | *start_ret = start; | |
3899 | return true; | |
3900 | } | |
3901 | ||
3902 | /* | |
3903 | * No delalloc in the range and we are seeking for data. The caller has | |
3904 | * to iterate to the next extent item in the subvolume btree. | |
3905 | */ | |
3906 | return false; | |
3907 | } | |
3908 | ||
cca5de97 | 3909 | static loff_t find_desired_extent(struct btrfs_inode *inode, loff_t offset, |
bc80230e | 3910 | int whence) |
b2675157 | 3911 | { |
cca5de97 | 3912 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
b2675157 | 3913 | struct extent_state *cached_state = NULL; |
b6e83356 FM |
3914 | const loff_t i_size = i_size_read(&inode->vfs_inode); |
3915 | const u64 ino = btrfs_ino(inode); | |
3916 | struct btrfs_root *root = inode->root; | |
3917 | struct btrfs_path *path; | |
3918 | struct btrfs_key key; | |
3919 | u64 last_extent_end; | |
4d1a40c6 LB |
3920 | u64 lockstart; |
3921 | u64 lockend; | |
3922 | u64 start; | |
b6e83356 FM |
3923 | int ret; |
3924 | bool found = false; | |
b2675157 | 3925 | |
bc80230e | 3926 | if (i_size == 0 || offset >= i_size) |
4d1a40c6 LB |
3927 | return -ENXIO; |
3928 | ||
b6e83356 FM |
3929 | /* |
3930 | * Quick path. If the inode has no prealloc extents and its number of | |
3931 | * bytes used matches its i_size, then it can not have holes. | |
3932 | */ | |
3933 | if (whence == SEEK_HOLE && | |
3934 | !(inode->flags & BTRFS_INODE_PREALLOC) && | |
3935 | inode_get_bytes(&inode->vfs_inode) == i_size) | |
3936 | return i_size; | |
3937 | ||
4d1a40c6 | 3938 | /* |
bc80230e | 3939 | * offset can be negative, in this case we start finding DATA/HOLE from |
4d1a40c6 LB |
3940 | * the very start of the file. |
3941 | */ | |
bc80230e | 3942 | start = max_t(loff_t, 0, offset); |
4d1a40c6 | 3943 | |
0b246afa | 3944 | lockstart = round_down(start, fs_info->sectorsize); |
d79b7c26 | 3945 | lockend = round_up(i_size, fs_info->sectorsize); |
b2675157 | 3946 | if (lockend <= lockstart) |
0b246afa | 3947 | lockend = lockstart + fs_info->sectorsize; |
1214b53f | 3948 | lockend--; |
b6e83356 FM |
3949 | |
3950 | path = btrfs_alloc_path(); | |
3951 | if (!path) | |
3952 | return -ENOMEM; | |
3953 | path->reada = READA_FORWARD; | |
3954 | ||
3955 | key.objectid = ino; | |
3956 | key.type = BTRFS_EXTENT_DATA_KEY; | |
3957 | key.offset = start; | |
3958 | ||
3959 | last_extent_end = lockstart; | |
b2675157 | 3960 | |
570eb97b | 3961 | lock_extent(&inode->io_tree, lockstart, lockend, &cached_state); |
b2675157 | 3962 | |
b6e83356 FM |
3963 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3964 | if (ret < 0) { | |
3965 | goto out; | |
3966 | } else if (ret > 0 && path->slots[0] > 0) { | |
3967 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1); | |
3968 | if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY) | |
3969 | path->slots[0]--; | |
3970 | } | |
3971 | ||
d79b7c26 | 3972 | while (start < i_size) { |
b6e83356 FM |
3973 | struct extent_buffer *leaf = path->nodes[0]; |
3974 | struct btrfs_file_extent_item *extent; | |
3975 | u64 extent_end; | |
3976 | ||
3977 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
3978 | ret = btrfs_next_leaf(root, path); | |
3979 | if (ret < 0) | |
3980 | goto out; | |
3981 | else if (ret > 0) | |
3982 | break; | |
3983 | ||
3984 | leaf = path->nodes[0]; | |
b2675157 JB |
3985 | } |
3986 | ||
b6e83356 FM |
3987 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
3988 | if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) | |
7f4ca37c | 3989 | break; |
b2675157 | 3990 | |
b6e83356 FM |
3991 | extent_end = btrfs_file_extent_end(path); |
3992 | ||
3993 | /* | |
3994 | * In the first iteration we may have a slot that points to an | |
3995 | * extent that ends before our start offset, so skip it. | |
3996 | */ | |
3997 | if (extent_end <= start) { | |
3998 | path->slots[0]++; | |
3999 | continue; | |
4000 | } | |
4001 | ||
4002 | /* We have an implicit hole, NO_HOLES feature is likely set. */ | |
4003 | if (last_extent_end < key.offset) { | |
4004 | u64 search_start = last_extent_end; | |
4005 | u64 found_start; | |
4006 | ||
4007 | /* | |
4008 | * First iteration, @start matches @offset and it's | |
4009 | * within the hole. | |
4010 | */ | |
4011 | if (start == offset) | |
4012 | search_start = offset; | |
4013 | ||
4014 | found = find_desired_extent_in_hole(inode, whence, | |
4015 | search_start, | |
4016 | key.offset - 1, | |
4017 | &found_start); | |
4018 | if (found) { | |
4019 | start = found_start; | |
4020 | break; | |
4021 | } | |
4022 | /* | |
4023 | * Didn't find data or a hole (due to delalloc) in the | |
4024 | * implicit hole range, so need to analyze the extent. | |
4025 | */ | |
4026 | } | |
4027 | ||
4028 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
4029 | struct btrfs_file_extent_item); | |
4030 | ||
4031 | if (btrfs_file_extent_disk_bytenr(leaf, extent) == 0 || | |
4032 | btrfs_file_extent_type(leaf, extent) == | |
4033 | BTRFS_FILE_EXTENT_PREALLOC) { | |
4034 | /* | |
4035 | * Explicit hole or prealloc extent, search for delalloc. | |
4036 | * A prealloc extent is treated like a hole. | |
4037 | */ | |
4038 | u64 search_start = key.offset; | |
4039 | u64 found_start; | |
4040 | ||
4041 | /* | |
4042 | * First iteration, @start matches @offset and it's | |
4043 | * within the hole. | |
4044 | */ | |
4045 | if (start == offset) | |
4046 | search_start = offset; | |
4047 | ||
4048 | found = find_desired_extent_in_hole(inode, whence, | |
4049 | search_start, | |
4050 | extent_end - 1, | |
4051 | &found_start); | |
4052 | if (found) { | |
4053 | start = found_start; | |
4054 | break; | |
4055 | } | |
4056 | /* | |
4057 | * Didn't find data or a hole (due to delalloc) in the | |
4058 | * implicit hole range, so need to analyze the next | |
4059 | * extent item. | |
4060 | */ | |
4061 | } else { | |
4062 | /* | |
4063 | * Found a regular or inline extent. | |
4064 | * If we are seeking for data, adjust the start offset | |
4065 | * and stop, we're done. | |
4066 | */ | |
4067 | if (whence == SEEK_DATA) { | |
4068 | start = max_t(u64, key.offset, offset); | |
4069 | found = true; | |
4070 | break; | |
4071 | } | |
4072 | /* | |
4073 | * Else, we are seeking for a hole, check the next file | |
4074 | * extent item. | |
4075 | */ | |
4076 | } | |
4077 | ||
4078 | start = extent_end; | |
4079 | last_extent_end = extent_end; | |
4080 | path->slots[0]++; | |
aed0ca18 FM |
4081 | if (fatal_signal_pending(current)) { |
4082 | ret = -EINTR; | |
b6e83356 | 4083 | goto out; |
aed0ca18 | 4084 | } |
b2675157 JB |
4085 | cond_resched(); |
4086 | } | |
b6e83356 FM |
4087 | |
4088 | /* We have an implicit hole from the last extent found up to i_size. */ | |
4089 | if (!found && start < i_size) { | |
4090 | found = find_desired_extent_in_hole(inode, whence, start, | |
4091 | i_size - 1, &start); | |
4092 | if (!found) | |
4093 | start = i_size; | |
4094 | } | |
4095 | ||
4096 | out: | |
570eb97b | 4097 | unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state); |
b6e83356 FM |
4098 | btrfs_free_path(path); |
4099 | ||
4100 | if (ret < 0) | |
4101 | return ret; | |
4102 | ||
4103 | if (whence == SEEK_DATA && start >= i_size) | |
4104 | return -ENXIO; | |
bc80230e | 4105 | |
b6e83356 | 4106 | return min_t(loff_t, start, i_size); |
b2675157 JB |
4107 | } |
4108 | ||
965c8e59 | 4109 | static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence) |
b2675157 JB |
4110 | { |
4111 | struct inode *inode = file->f_mapping->host; | |
b2675157 | 4112 | |
965c8e59 | 4113 | switch (whence) { |
2034f3b4 NB |
4114 | default: |
4115 | return generic_file_llseek(file, offset, whence); | |
b2675157 JB |
4116 | case SEEK_DATA: |
4117 | case SEEK_HOLE: | |
a14b78ad | 4118 | btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED); |
cca5de97 | 4119 | offset = find_desired_extent(BTRFS_I(inode), offset, whence); |
a14b78ad | 4120 | btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED); |
bc80230e | 4121 | break; |
b2675157 JB |
4122 | } |
4123 | ||
bc80230e NB |
4124 | if (offset < 0) |
4125 | return offset; | |
4126 | ||
2034f3b4 | 4127 | return vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
b2675157 JB |
4128 | } |
4129 | ||
edf064e7 GR |
4130 | static int btrfs_file_open(struct inode *inode, struct file *filp) |
4131 | { | |
14605409 BB |
4132 | int ret; |
4133 | ||
926078b2 | 4134 | filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC | FMODE_BUF_WASYNC; |
14605409 BB |
4135 | |
4136 | ret = fsverity_file_open(inode, filp); | |
4137 | if (ret) | |
4138 | return ret; | |
edf064e7 GR |
4139 | return generic_file_open(inode, filp); |
4140 | } | |
4141 | ||
4e4cabec GR |
4142 | static int check_direct_read(struct btrfs_fs_info *fs_info, |
4143 | const struct iov_iter *iter, loff_t offset) | |
4144 | { | |
4145 | int ret; | |
4146 | int i, seg; | |
4147 | ||
4148 | ret = check_direct_IO(fs_info, iter, offset); | |
4149 | if (ret < 0) | |
4150 | return ret; | |
4151 | ||
4152 | if (!iter_is_iovec(iter)) | |
4153 | return 0; | |
4154 | ||
4155 | for (seg = 0; seg < iter->nr_segs; seg++) | |
4156 | for (i = seg + 1; i < iter->nr_segs; i++) | |
4157 | if (iter->iov[seg].iov_base == iter->iov[i].iov_base) | |
4158 | return -EINVAL; | |
4159 | return 0; | |
4160 | } | |
4161 | ||
4162 | static ssize_t btrfs_direct_read(struct kiocb *iocb, struct iov_iter *to) | |
4163 | { | |
4164 | struct inode *inode = file_inode(iocb->ki_filp); | |
51bd9563 FM |
4165 | size_t prev_left = 0; |
4166 | ssize_t read = 0; | |
4e4cabec GR |
4167 | ssize_t ret; |
4168 | ||
14605409 BB |
4169 | if (fsverity_active(inode)) |
4170 | return 0; | |
4171 | ||
4e4cabec GR |
4172 | if (check_direct_read(btrfs_sb(inode->i_sb), to, iocb->ki_pos)) |
4173 | return 0; | |
4174 | ||
a14b78ad | 4175 | btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED); |
51bd9563 FM |
4176 | again: |
4177 | /* | |
4178 | * This is similar to what we do for direct IO writes, see the comment | |
4179 | * at btrfs_direct_write(), but we also disable page faults in addition | |
4180 | * to disabling them only at the iov_iter level. This is because when | |
4181 | * reading from a hole or prealloc extent, iomap calls iov_iter_zero(), | |
4182 | * which can still trigger page fault ins despite having set ->nofault | |
4183 | * to true of our 'to' iov_iter. | |
4184 | * | |
4185 | * The difference to direct IO writes is that we deadlock when trying | |
4186 | * to lock the extent range in the inode's tree during he page reads | |
4187 | * triggered by the fault in (while for writes it is due to waiting for | |
4188 | * our own ordered extent). This is because for direct IO reads, | |
4189 | * btrfs_dio_iomap_begin() returns with the extent range locked, which | |
4190 | * is only unlocked in the endio callback (end_bio_extent_readpage()). | |
4191 | */ | |
4192 | pagefault_disable(); | |
4193 | to->nofault = true; | |
36e8c622 | 4194 | ret = btrfs_dio_rw(iocb, to, read); |
51bd9563 FM |
4195 | to->nofault = false; |
4196 | pagefault_enable(); | |
4197 | ||
4198 | /* No increment (+=) because iomap returns a cumulative value. */ | |
4199 | if (ret > 0) | |
4200 | read = ret; | |
4201 | ||
4202 | if (iov_iter_count(to) > 0 && (ret == -EFAULT || ret > 0)) { | |
4203 | const size_t left = iov_iter_count(to); | |
4204 | ||
4205 | if (left == prev_left) { | |
4206 | /* | |
4207 | * We didn't make any progress since the last attempt, | |
4208 | * fallback to a buffered read for the remainder of the | |
4209 | * range. This is just to avoid any possibility of looping | |
4210 | * for too long. | |
4211 | */ | |
4212 | ret = read; | |
4213 | } else { | |
4214 | /* | |
4215 | * We made some progress since the last retry or this is | |
4216 | * the first time we are retrying. Fault in as many pages | |
4217 | * as possible and retry. | |
4218 | */ | |
4219 | fault_in_iov_iter_writeable(to, left); | |
4220 | prev_left = left; | |
4221 | goto again; | |
4222 | } | |
4223 | } | |
a14b78ad | 4224 | btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED); |
51bd9563 | 4225 | return ret < 0 ? ret : read; |
4e4cabec GR |
4226 | } |
4227 | ||
f85781fb GR |
4228 | static ssize_t btrfs_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
4229 | { | |
4230 | ssize_t ret = 0; | |
4231 | ||
4232 | if (iocb->ki_flags & IOCB_DIRECT) { | |
4e4cabec | 4233 | ret = btrfs_direct_read(iocb, to); |
0425e7ba JT |
4234 | if (ret < 0 || !iov_iter_count(to) || |
4235 | iocb->ki_pos >= i_size_read(file_inode(iocb->ki_filp))) | |
f85781fb GR |
4236 | return ret; |
4237 | } | |
4238 | ||
87fa0f3e | 4239 | return filemap_read(iocb, to, ret); |
f85781fb GR |
4240 | } |
4241 | ||
828c0950 | 4242 | const struct file_operations btrfs_file_operations = { |
b2675157 | 4243 | .llseek = btrfs_file_llseek, |
f85781fb | 4244 | .read_iter = btrfs_file_read_iter, |
e9906a98 | 4245 | .splice_read = generic_file_splice_read, |
b30ac0fc | 4246 | .write_iter = btrfs_file_write_iter, |
d7776591 | 4247 | .splice_write = iter_file_splice_write, |
9ebefb18 | 4248 | .mmap = btrfs_file_mmap, |
edf064e7 | 4249 | .open = btrfs_file_open, |
e1b81e67 | 4250 | .release = btrfs_release_file, |
b0c58223 | 4251 | .get_unmapped_area = thp_get_unmapped_area, |
39279cc3 | 4252 | .fsync = btrfs_sync_file, |
2fe17c10 | 4253 | .fallocate = btrfs_fallocate, |
34287aa3 | 4254 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 4255 | #ifdef CONFIG_COMPAT |
4c63c245 | 4256 | .compat_ioctl = btrfs_compat_ioctl, |
39279cc3 | 4257 | #endif |
2e5dfc99 | 4258 | .remap_file_range = btrfs_remap_file_range, |
39279cc3 | 4259 | }; |
9247f317 | 4260 | |
e67c718b | 4261 | void __cold btrfs_auto_defrag_exit(void) |
9247f317 | 4262 | { |
5598e900 | 4263 | kmem_cache_destroy(btrfs_inode_defrag_cachep); |
9247f317 MX |
4264 | } |
4265 | ||
f5c29bd9 | 4266 | int __init btrfs_auto_defrag_init(void) |
9247f317 MX |
4267 | { |
4268 | btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag", | |
4269 | sizeof(struct inode_defrag), 0, | |
fba4b697 | 4270 | SLAB_MEM_SPREAD, |
9247f317 MX |
4271 | NULL); |
4272 | if (!btrfs_inode_defrag_cachep) | |
4273 | return -ENOMEM; | |
4274 | ||
4275 | return 0; | |
4276 | } | |
728404da FM |
4277 | |
4278 | int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end) | |
4279 | { | |
4280 | int ret; | |
4281 | ||
4282 | /* | |
4283 | * So with compression we will find and lock a dirty page and clear the | |
4284 | * first one as dirty, setup an async extent, and immediately return | |
4285 | * with the entire range locked but with nobody actually marked with | |
4286 | * writeback. So we can't just filemap_write_and_wait_range() and | |
4287 | * expect it to work since it will just kick off a thread to do the | |
4288 | * actual work. So we need to call filemap_fdatawrite_range _again_ | |
4289 | * since it will wait on the page lock, which won't be unlocked until | |
4290 | * after the pages have been marked as writeback and so we're good to go | |
4291 | * from there. We have to do this otherwise we'll miss the ordered | |
4292 | * extents and that results in badness. Please Josef, do not think you | |
4293 | * know better and pull this out at some point in the future, it is | |
4294 | * right and you are wrong. | |
4295 | */ | |
4296 | ret = filemap_fdatawrite_range(inode->i_mapping, start, end); | |
4297 | if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
4298 | &BTRFS_I(inode)->runtime_flags)) | |
4299 | ret = filemap_fdatawrite_range(inode->i_mapping, start, end); | |
4300 | ||
4301 | return ret; | |
4302 | } |