]>
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> | |
8 | #include <linux/highmem.h> | |
9 | #include <linux/time.h> | |
10 | #include <linux/init.h> | |
11 | #include <linux/string.h> | |
39279cc3 CM |
12 | #include <linux/backing-dev.h> |
13 | #include <linux/mpage.h> | |
2fe17c10 | 14 | #include <linux/falloc.h> |
39279cc3 CM |
15 | #include <linux/swap.h> |
16 | #include <linux/writeback.h> | |
39279cc3 | 17 | #include <linux/compat.h> |
5a0e3ad6 | 18 | #include <linux/slab.h> |
55e301fd | 19 | #include <linux/btrfs.h> |
e2e40f2c | 20 | #include <linux/uio.h> |
ae5e165d | 21 | #include <linux/iversion.h> |
39279cc3 CM |
22 | #include "ctree.h" |
23 | #include "disk-io.h" | |
24 | #include "transaction.h" | |
25 | #include "btrfs_inode.h" | |
39279cc3 | 26 | #include "print-tree.h" |
e02119d5 CM |
27 | #include "tree-log.h" |
28 | #include "locking.h" | |
2aaa6655 | 29 | #include "volumes.h" |
fcebe456 | 30 | #include "qgroup.h" |
ebb8765b | 31 | #include "compression.h" |
39279cc3 | 32 | |
9247f317 | 33 | static struct kmem_cache *btrfs_inode_defrag_cachep; |
4cb5300b CM |
34 | /* |
35 | * when auto defrag is enabled we | |
36 | * queue up these defrag structs to remember which | |
37 | * inodes need defragging passes | |
38 | */ | |
39 | struct inode_defrag { | |
40 | struct rb_node rb_node; | |
41 | /* objectid */ | |
42 | u64 ino; | |
43 | /* | |
44 | * transid where the defrag was added, we search for | |
45 | * extents newer than this | |
46 | */ | |
47 | u64 transid; | |
48 | ||
49 | /* root objectid */ | |
50 | u64 root; | |
51 | ||
52 | /* last offset we were able to defrag */ | |
53 | u64 last_offset; | |
54 | ||
55 | /* if we've wrapped around back to zero once already */ | |
56 | int cycled; | |
57 | }; | |
58 | ||
762f2263 MX |
59 | static int __compare_inode_defrag(struct inode_defrag *defrag1, |
60 | struct inode_defrag *defrag2) | |
61 | { | |
62 | if (defrag1->root > defrag2->root) | |
63 | return 1; | |
64 | else if (defrag1->root < defrag2->root) | |
65 | return -1; | |
66 | else if (defrag1->ino > defrag2->ino) | |
67 | return 1; | |
68 | else if (defrag1->ino < defrag2->ino) | |
69 | return -1; | |
70 | else | |
71 | return 0; | |
72 | } | |
73 | ||
4cb5300b CM |
74 | /* pop a record for an inode into the defrag tree. The lock |
75 | * must be held already | |
76 | * | |
77 | * If you're inserting a record for an older transid than an | |
78 | * existing record, the transid already in the tree is lowered | |
79 | * | |
80 | * If an existing record is found the defrag item you | |
81 | * pass in is freed | |
82 | */ | |
6158e1ce | 83 | static int __btrfs_add_inode_defrag(struct btrfs_inode *inode, |
4cb5300b CM |
84 | struct inode_defrag *defrag) |
85 | { | |
6158e1ce | 86 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
4cb5300b CM |
87 | struct inode_defrag *entry; |
88 | struct rb_node **p; | |
89 | struct rb_node *parent = NULL; | |
762f2263 | 90 | int ret; |
4cb5300b | 91 | |
0b246afa | 92 | p = &fs_info->defrag_inodes.rb_node; |
4cb5300b CM |
93 | while (*p) { |
94 | parent = *p; | |
95 | entry = rb_entry(parent, struct inode_defrag, rb_node); | |
96 | ||
762f2263 MX |
97 | ret = __compare_inode_defrag(defrag, entry); |
98 | if (ret < 0) | |
4cb5300b | 99 | p = &parent->rb_left; |
762f2263 | 100 | else if (ret > 0) |
4cb5300b CM |
101 | p = &parent->rb_right; |
102 | else { | |
103 | /* if we're reinserting an entry for | |
104 | * an old defrag run, make sure to | |
105 | * lower the transid of our existing record | |
106 | */ | |
107 | if (defrag->transid < entry->transid) | |
108 | entry->transid = defrag->transid; | |
109 | if (defrag->last_offset > entry->last_offset) | |
110 | entry->last_offset = defrag->last_offset; | |
8ddc4734 | 111 | return -EEXIST; |
4cb5300b CM |
112 | } |
113 | } | |
6158e1ce | 114 | set_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags); |
4cb5300b | 115 | rb_link_node(&defrag->rb_node, parent, p); |
0b246afa | 116 | rb_insert_color(&defrag->rb_node, &fs_info->defrag_inodes); |
8ddc4734 MX |
117 | return 0; |
118 | } | |
4cb5300b | 119 | |
2ff7e61e | 120 | static inline int __need_auto_defrag(struct btrfs_fs_info *fs_info) |
8ddc4734 | 121 | { |
0b246afa | 122 | if (!btrfs_test_opt(fs_info, AUTO_DEFRAG)) |
8ddc4734 MX |
123 | return 0; |
124 | ||
0b246afa | 125 | if (btrfs_fs_closing(fs_info)) |
8ddc4734 | 126 | return 0; |
4cb5300b | 127 | |
8ddc4734 | 128 | return 1; |
4cb5300b CM |
129 | } |
130 | ||
131 | /* | |
132 | * insert a defrag record for this inode if auto defrag is | |
133 | * enabled | |
134 | */ | |
135 | int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans, | |
6158e1ce | 136 | struct btrfs_inode *inode) |
4cb5300b | 137 | { |
6158e1ce NB |
138 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
139 | struct btrfs_root *root = inode->root; | |
4cb5300b | 140 | struct inode_defrag *defrag; |
4cb5300b | 141 | u64 transid; |
8ddc4734 | 142 | int ret; |
4cb5300b | 143 | |
2ff7e61e | 144 | if (!__need_auto_defrag(fs_info)) |
4cb5300b CM |
145 | return 0; |
146 | ||
6158e1ce | 147 | if (test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags)) |
4cb5300b CM |
148 | return 0; |
149 | ||
150 | if (trans) | |
151 | transid = trans->transid; | |
152 | else | |
6158e1ce | 153 | transid = inode->root->last_trans; |
4cb5300b | 154 | |
9247f317 | 155 | defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS); |
4cb5300b CM |
156 | if (!defrag) |
157 | return -ENOMEM; | |
158 | ||
6158e1ce | 159 | defrag->ino = btrfs_ino(inode); |
4cb5300b CM |
160 | defrag->transid = transid; |
161 | defrag->root = root->root_key.objectid; | |
162 | ||
0b246afa | 163 | spin_lock(&fs_info->defrag_inodes_lock); |
6158e1ce | 164 | if (!test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags)) { |
8ddc4734 MX |
165 | /* |
166 | * If we set IN_DEFRAG flag and evict the inode from memory, | |
167 | * and then re-read this inode, this new inode doesn't have | |
168 | * IN_DEFRAG flag. At the case, we may find the existed defrag. | |
169 | */ | |
170 | ret = __btrfs_add_inode_defrag(inode, defrag); | |
171 | if (ret) | |
172 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
173 | } else { | |
9247f317 | 174 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); |
8ddc4734 | 175 | } |
0b246afa | 176 | spin_unlock(&fs_info->defrag_inodes_lock); |
a0f98dde | 177 | return 0; |
4cb5300b CM |
178 | } |
179 | ||
180 | /* | |
8ddc4734 MX |
181 | * Requeue the defrag object. If there is a defrag object that points to |
182 | * the same inode in the tree, we will merge them together (by | |
183 | * __btrfs_add_inode_defrag()) and free the one that we want to requeue. | |
4cb5300b | 184 | */ |
46e59791 | 185 | static void btrfs_requeue_inode_defrag(struct btrfs_inode *inode, |
48a3b636 | 186 | struct inode_defrag *defrag) |
8ddc4734 | 187 | { |
46e59791 | 188 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
8ddc4734 MX |
189 | int ret; |
190 | ||
2ff7e61e | 191 | if (!__need_auto_defrag(fs_info)) |
8ddc4734 MX |
192 | goto out; |
193 | ||
194 | /* | |
195 | * Here we don't check the IN_DEFRAG flag, because we need merge | |
196 | * them together. | |
197 | */ | |
0b246afa | 198 | spin_lock(&fs_info->defrag_inodes_lock); |
8ddc4734 | 199 | ret = __btrfs_add_inode_defrag(inode, defrag); |
0b246afa | 200 | spin_unlock(&fs_info->defrag_inodes_lock); |
8ddc4734 MX |
201 | if (ret) |
202 | goto out; | |
203 | return; | |
204 | out: | |
205 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
206 | } | |
207 | ||
4cb5300b | 208 | /* |
26176e7c MX |
209 | * pick the defragable inode that we want, if it doesn't exist, we will get |
210 | * the next one. | |
4cb5300b | 211 | */ |
26176e7c MX |
212 | static struct inode_defrag * |
213 | btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino) | |
4cb5300b CM |
214 | { |
215 | struct inode_defrag *entry = NULL; | |
762f2263 | 216 | struct inode_defrag tmp; |
4cb5300b CM |
217 | struct rb_node *p; |
218 | struct rb_node *parent = NULL; | |
762f2263 MX |
219 | int ret; |
220 | ||
221 | tmp.ino = ino; | |
222 | tmp.root = root; | |
4cb5300b | 223 | |
26176e7c MX |
224 | spin_lock(&fs_info->defrag_inodes_lock); |
225 | p = fs_info->defrag_inodes.rb_node; | |
4cb5300b CM |
226 | while (p) { |
227 | parent = p; | |
228 | entry = rb_entry(parent, struct inode_defrag, rb_node); | |
229 | ||
762f2263 MX |
230 | ret = __compare_inode_defrag(&tmp, entry); |
231 | if (ret < 0) | |
4cb5300b | 232 | p = parent->rb_left; |
762f2263 | 233 | else if (ret > 0) |
4cb5300b CM |
234 | p = parent->rb_right; |
235 | else | |
26176e7c | 236 | goto out; |
4cb5300b CM |
237 | } |
238 | ||
26176e7c MX |
239 | if (parent && __compare_inode_defrag(&tmp, entry) > 0) { |
240 | parent = rb_next(parent); | |
241 | if (parent) | |
4cb5300b | 242 | entry = rb_entry(parent, struct inode_defrag, rb_node); |
26176e7c MX |
243 | else |
244 | entry = NULL; | |
4cb5300b | 245 | } |
26176e7c MX |
246 | out: |
247 | if (entry) | |
248 | rb_erase(parent, &fs_info->defrag_inodes); | |
249 | spin_unlock(&fs_info->defrag_inodes_lock); | |
250 | return entry; | |
4cb5300b CM |
251 | } |
252 | ||
26176e7c | 253 | void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info) |
4cb5300b CM |
254 | { |
255 | struct inode_defrag *defrag; | |
26176e7c MX |
256 | struct rb_node *node; |
257 | ||
258 | spin_lock(&fs_info->defrag_inodes_lock); | |
259 | node = rb_first(&fs_info->defrag_inodes); | |
260 | while (node) { | |
261 | rb_erase(node, &fs_info->defrag_inodes); | |
262 | defrag = rb_entry(node, struct inode_defrag, rb_node); | |
263 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
264 | ||
351810c1 | 265 | cond_resched_lock(&fs_info->defrag_inodes_lock); |
26176e7c MX |
266 | |
267 | node = rb_first(&fs_info->defrag_inodes); | |
268 | } | |
269 | spin_unlock(&fs_info->defrag_inodes_lock); | |
270 | } | |
271 | ||
272 | #define BTRFS_DEFRAG_BATCH 1024 | |
273 | ||
274 | static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info, | |
275 | struct inode_defrag *defrag) | |
276 | { | |
4cb5300b CM |
277 | struct btrfs_root *inode_root; |
278 | struct inode *inode; | |
4cb5300b CM |
279 | struct btrfs_key key; |
280 | struct btrfs_ioctl_defrag_range_args range; | |
4cb5300b | 281 | int num_defrag; |
6f1c3605 LB |
282 | int index; |
283 | int ret; | |
4cb5300b | 284 | |
26176e7c MX |
285 | /* get the inode */ |
286 | key.objectid = defrag->root; | |
962a298f | 287 | key.type = BTRFS_ROOT_ITEM_KEY; |
26176e7c | 288 | key.offset = (u64)-1; |
6f1c3605 LB |
289 | |
290 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
291 | ||
26176e7c MX |
292 | inode_root = btrfs_read_fs_root_no_name(fs_info, &key); |
293 | if (IS_ERR(inode_root)) { | |
6f1c3605 LB |
294 | ret = PTR_ERR(inode_root); |
295 | goto cleanup; | |
296 | } | |
26176e7c MX |
297 | |
298 | key.objectid = defrag->ino; | |
962a298f | 299 | key.type = BTRFS_INODE_ITEM_KEY; |
26176e7c MX |
300 | key.offset = 0; |
301 | inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL); | |
302 | if (IS_ERR(inode)) { | |
6f1c3605 LB |
303 | ret = PTR_ERR(inode); |
304 | goto cleanup; | |
26176e7c | 305 | } |
6f1c3605 | 306 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
26176e7c MX |
307 | |
308 | /* do a chunk of defrag */ | |
309 | clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags); | |
4cb5300b CM |
310 | memset(&range, 0, sizeof(range)); |
311 | range.len = (u64)-1; | |
26176e7c | 312 | range.start = defrag->last_offset; |
b66f00da MX |
313 | |
314 | sb_start_write(fs_info->sb); | |
26176e7c MX |
315 | num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid, |
316 | BTRFS_DEFRAG_BATCH); | |
b66f00da | 317 | sb_end_write(fs_info->sb); |
26176e7c MX |
318 | /* |
319 | * if we filled the whole defrag batch, there | |
320 | * must be more work to do. Queue this defrag | |
321 | * again | |
322 | */ | |
323 | if (num_defrag == BTRFS_DEFRAG_BATCH) { | |
324 | defrag->last_offset = range.start; | |
46e59791 | 325 | btrfs_requeue_inode_defrag(BTRFS_I(inode), defrag); |
26176e7c MX |
326 | } else if (defrag->last_offset && !defrag->cycled) { |
327 | /* | |
328 | * we didn't fill our defrag batch, but | |
329 | * we didn't start at zero. Make sure we loop | |
330 | * around to the start of the file. | |
331 | */ | |
332 | defrag->last_offset = 0; | |
333 | defrag->cycled = 1; | |
46e59791 | 334 | btrfs_requeue_inode_defrag(BTRFS_I(inode), defrag); |
26176e7c MX |
335 | } else { |
336 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
337 | } | |
338 | ||
339 | iput(inode); | |
340 | return 0; | |
6f1c3605 LB |
341 | cleanup: |
342 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
343 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
344 | return ret; | |
26176e7c MX |
345 | } |
346 | ||
347 | /* | |
348 | * run through the list of inodes in the FS that need | |
349 | * defragging | |
350 | */ | |
351 | int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info) | |
352 | { | |
353 | struct inode_defrag *defrag; | |
354 | u64 first_ino = 0; | |
355 | u64 root_objectid = 0; | |
4cb5300b CM |
356 | |
357 | atomic_inc(&fs_info->defrag_running); | |
67871254 | 358 | while (1) { |
dc81cdc5 MX |
359 | /* Pause the auto defragger. */ |
360 | if (test_bit(BTRFS_FS_STATE_REMOUNTING, | |
361 | &fs_info->fs_state)) | |
362 | break; | |
363 | ||
2ff7e61e | 364 | if (!__need_auto_defrag(fs_info)) |
26176e7c | 365 | break; |
4cb5300b CM |
366 | |
367 | /* find an inode to defrag */ | |
26176e7c MX |
368 | defrag = btrfs_pick_defrag_inode(fs_info, root_objectid, |
369 | first_ino); | |
4cb5300b | 370 | if (!defrag) { |
26176e7c | 371 | if (root_objectid || first_ino) { |
762f2263 | 372 | root_objectid = 0; |
4cb5300b CM |
373 | first_ino = 0; |
374 | continue; | |
375 | } else { | |
376 | break; | |
377 | } | |
378 | } | |
379 | ||
4cb5300b | 380 | first_ino = defrag->ino + 1; |
762f2263 | 381 | root_objectid = defrag->root; |
4cb5300b | 382 | |
26176e7c | 383 | __btrfs_run_defrag_inode(fs_info, defrag); |
4cb5300b | 384 | } |
4cb5300b CM |
385 | atomic_dec(&fs_info->defrag_running); |
386 | ||
387 | /* | |
388 | * during unmount, we use the transaction_wait queue to | |
389 | * wait for the defragger to stop | |
390 | */ | |
391 | wake_up(&fs_info->transaction_wait); | |
392 | return 0; | |
393 | } | |
39279cc3 | 394 | |
d352ac68 CM |
395 | /* simple helper to fault in pages and copy. This should go away |
396 | * and be replaced with calls into generic code. | |
397 | */ | |
ee22f0c4 | 398 | static noinline int btrfs_copy_from_user(loff_t pos, size_t write_bytes, |
a1b32a59 | 399 | struct page **prepared_pages, |
11c65dcc | 400 | struct iov_iter *i) |
39279cc3 | 401 | { |
914ee295 | 402 | size_t copied = 0; |
d0215f3e | 403 | size_t total_copied = 0; |
11c65dcc | 404 | int pg = 0; |
09cbfeaf | 405 | int offset = pos & (PAGE_SIZE - 1); |
39279cc3 | 406 | |
11c65dcc | 407 | while (write_bytes > 0) { |
39279cc3 | 408 | size_t count = min_t(size_t, |
09cbfeaf | 409 | PAGE_SIZE - offset, write_bytes); |
11c65dcc | 410 | struct page *page = prepared_pages[pg]; |
914ee295 XZ |
411 | /* |
412 | * Copy data from userspace to the current page | |
914ee295 | 413 | */ |
914ee295 | 414 | copied = iov_iter_copy_from_user_atomic(page, i, offset, count); |
11c65dcc | 415 | |
39279cc3 CM |
416 | /* Flush processor's dcache for this page */ |
417 | flush_dcache_page(page); | |
31339acd CM |
418 | |
419 | /* | |
420 | * if we get a partial write, we can end up with | |
421 | * partially up to date pages. These add | |
422 | * a lot of complexity, so make sure they don't | |
423 | * happen by forcing this copy to be retried. | |
424 | * | |
425 | * The rest of the btrfs_file_write code will fall | |
426 | * back to page at a time copies after we return 0. | |
427 | */ | |
428 | if (!PageUptodate(page) && copied < count) | |
429 | copied = 0; | |
430 | ||
11c65dcc JB |
431 | iov_iter_advance(i, copied); |
432 | write_bytes -= copied; | |
914ee295 | 433 | total_copied += copied; |
39279cc3 | 434 | |
b30ac0fc | 435 | /* Return to btrfs_file_write_iter to fault page */ |
9f570b8d | 436 | if (unlikely(copied == 0)) |
914ee295 | 437 | break; |
11c65dcc | 438 | |
09cbfeaf | 439 | if (copied < PAGE_SIZE - offset) { |
11c65dcc JB |
440 | offset += copied; |
441 | } else { | |
442 | pg++; | |
443 | offset = 0; | |
444 | } | |
39279cc3 | 445 | } |
914ee295 | 446 | return total_copied; |
39279cc3 CM |
447 | } |
448 | ||
d352ac68 CM |
449 | /* |
450 | * unlocks pages after btrfs_file_write is done with them | |
451 | */ | |
48a3b636 | 452 | static void btrfs_drop_pages(struct page **pages, size_t num_pages) |
39279cc3 CM |
453 | { |
454 | size_t i; | |
455 | for (i = 0; i < num_pages; i++) { | |
d352ac68 CM |
456 | /* page checked is some magic around finding pages that |
457 | * have been modified without going through btrfs_set_page_dirty | |
2457aec6 MG |
458 | * clear it here. There should be no need to mark the pages |
459 | * accessed as prepare_pages should have marked them accessed | |
460 | * in prepare_pages via find_or_create_page() | |
d352ac68 | 461 | */ |
4a096752 | 462 | ClearPageChecked(pages[i]); |
39279cc3 | 463 | unlock_page(pages[i]); |
09cbfeaf | 464 | put_page(pages[i]); |
39279cc3 CM |
465 | } |
466 | } | |
467 | ||
f48bf66b FM |
468 | static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode, |
469 | const u64 start, | |
470 | const u64 len, | |
471 | struct extent_state **cached_state) | |
472 | { | |
473 | u64 search_start = start; | |
474 | const u64 end = start + len - 1; | |
475 | ||
476 | while (search_start < end) { | |
477 | const u64 search_len = end - search_start + 1; | |
478 | struct extent_map *em; | |
479 | u64 em_len; | |
480 | int ret = 0; | |
481 | ||
482 | em = btrfs_get_extent(inode, NULL, 0, search_start, | |
483 | search_len, 0); | |
484 | if (IS_ERR(em)) | |
485 | return PTR_ERR(em); | |
486 | ||
487 | if (em->block_start != EXTENT_MAP_HOLE) | |
488 | goto next; | |
489 | ||
490 | em_len = em->len; | |
491 | if (em->start < search_start) | |
492 | em_len -= search_start - em->start; | |
493 | if (em_len > search_len) | |
494 | em_len = search_len; | |
495 | ||
496 | ret = set_extent_bit(&inode->io_tree, search_start, | |
497 | search_start + em_len - 1, | |
498 | EXTENT_DELALLOC_NEW, | |
499 | NULL, cached_state, GFP_NOFS); | |
500 | next: | |
501 | search_start = extent_map_end(em); | |
502 | free_extent_map(em); | |
503 | if (ret) | |
504 | return ret; | |
505 | } | |
506 | return 0; | |
507 | } | |
508 | ||
d352ac68 CM |
509 | /* |
510 | * after copy_from_user, pages need to be dirtied and we need to make | |
511 | * sure holes are created between the current EOF and the start of | |
512 | * any next extents (if required). | |
513 | * | |
514 | * this also makes the decision about creating an inline extent vs | |
515 | * doing real data extents, marking pages dirty and delalloc as required. | |
516 | */ | |
2ff7e61e JM |
517 | int btrfs_dirty_pages(struct inode *inode, struct page **pages, |
518 | size_t num_pages, loff_t pos, size_t write_bytes, | |
519 | struct extent_state **cached) | |
39279cc3 | 520 | { |
0b246afa | 521 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 522 | int err = 0; |
a52d9a80 | 523 | int i; |
db94535d | 524 | u64 num_bytes; |
a52d9a80 CM |
525 | u64 start_pos; |
526 | u64 end_of_last_block; | |
527 | u64 end_pos = pos + write_bytes; | |
528 | loff_t isize = i_size_read(inode); | |
e3b8a485 | 529 | unsigned int extra_bits = 0; |
39279cc3 | 530 | |
0b246afa | 531 | start_pos = pos & ~((u64) fs_info->sectorsize - 1); |
da17066c | 532 | num_bytes = round_up(write_bytes + pos - start_pos, |
0b246afa | 533 | fs_info->sectorsize); |
39279cc3 | 534 | |
db94535d | 535 | end_of_last_block = start_pos + num_bytes - 1; |
e3b8a485 FM |
536 | |
537 | if (!btrfs_is_free_space_inode(BTRFS_I(inode))) { | |
538 | if (start_pos >= isize && | |
539 | !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)) { | |
540 | /* | |
541 | * There can't be any extents following eof in this case | |
542 | * so just set the delalloc new bit for the range | |
543 | * directly. | |
544 | */ | |
545 | extra_bits |= EXTENT_DELALLOC_NEW; | |
546 | } else { | |
547 | err = btrfs_find_new_delalloc_bytes(BTRFS_I(inode), | |
548 | start_pos, | |
549 | num_bytes, cached); | |
550 | if (err) | |
551 | return err; | |
552 | } | |
553 | } | |
554 | ||
2ac55d41 | 555 | err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block, |
e3b8a485 | 556 | extra_bits, cached, 0); |
d0215f3e JB |
557 | if (err) |
558 | return err; | |
9ed74f2d | 559 | |
c8b97818 CM |
560 | for (i = 0; i < num_pages; i++) { |
561 | struct page *p = pages[i]; | |
562 | SetPageUptodate(p); | |
563 | ClearPageChecked(p); | |
564 | set_page_dirty(p); | |
a52d9a80 | 565 | } |
9f570b8d JB |
566 | |
567 | /* | |
568 | * we've only changed i_size in ram, and we haven't updated | |
569 | * the disk i_size. There is no need to log the inode | |
570 | * at this time. | |
571 | */ | |
572 | if (end_pos > isize) | |
a52d9a80 | 573 | i_size_write(inode, end_pos); |
a22285a6 | 574 | return 0; |
39279cc3 CM |
575 | } |
576 | ||
d352ac68 CM |
577 | /* |
578 | * this drops all the extents in the cache that intersect the range | |
579 | * [start, end]. Existing extents are split as required. | |
580 | */ | |
dcdbc059 | 581 | void btrfs_drop_extent_cache(struct btrfs_inode *inode, u64 start, u64 end, |
7014cdb4 | 582 | int skip_pinned) |
a52d9a80 CM |
583 | { |
584 | struct extent_map *em; | |
3b951516 CM |
585 | struct extent_map *split = NULL; |
586 | struct extent_map *split2 = NULL; | |
dcdbc059 | 587 | struct extent_map_tree *em_tree = &inode->extent_tree; |
39b5637f | 588 | u64 len = end - start + 1; |
5dc562c5 | 589 | u64 gen; |
3b951516 CM |
590 | int ret; |
591 | int testend = 1; | |
5b21f2ed | 592 | unsigned long flags; |
c8b97818 | 593 | int compressed = 0; |
09a2a8f9 | 594 | bool modified; |
a52d9a80 | 595 | |
e6dcd2dc | 596 | WARN_ON(end < start); |
3b951516 | 597 | if (end == (u64)-1) { |
39b5637f | 598 | len = (u64)-1; |
3b951516 CM |
599 | testend = 0; |
600 | } | |
d397712b | 601 | while (1) { |
7014cdb4 JB |
602 | int no_splits = 0; |
603 | ||
09a2a8f9 | 604 | modified = false; |
3b951516 | 605 | if (!split) |
172ddd60 | 606 | split = alloc_extent_map(); |
3b951516 | 607 | if (!split2) |
172ddd60 | 608 | split2 = alloc_extent_map(); |
7014cdb4 JB |
609 | if (!split || !split2) |
610 | no_splits = 1; | |
3b951516 | 611 | |
890871be | 612 | write_lock(&em_tree->lock); |
39b5637f | 613 | em = lookup_extent_mapping(em_tree, start, len); |
d1310b2e | 614 | if (!em) { |
890871be | 615 | write_unlock(&em_tree->lock); |
a52d9a80 | 616 | break; |
d1310b2e | 617 | } |
5b21f2ed | 618 | flags = em->flags; |
5dc562c5 | 619 | gen = em->generation; |
5b21f2ed | 620 | if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) { |
55ef6899 | 621 | if (testend && em->start + em->len >= start + len) { |
5b21f2ed | 622 | free_extent_map(em); |
a1ed835e | 623 | write_unlock(&em_tree->lock); |
5b21f2ed ZY |
624 | break; |
625 | } | |
55ef6899 YZ |
626 | start = em->start + em->len; |
627 | if (testend) | |
5b21f2ed | 628 | len = start + len - (em->start + em->len); |
5b21f2ed | 629 | free_extent_map(em); |
a1ed835e | 630 | write_unlock(&em_tree->lock); |
5b21f2ed ZY |
631 | continue; |
632 | } | |
c8b97818 | 633 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
3ce7e67a | 634 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
3b277594 | 635 | clear_bit(EXTENT_FLAG_LOGGING, &flags); |
09a2a8f9 | 636 | modified = !list_empty(&em->list); |
7014cdb4 JB |
637 | if (no_splits) |
638 | goto next; | |
3b951516 | 639 | |
ee20a983 | 640 | if (em->start < start) { |
3b951516 CM |
641 | split->start = em->start; |
642 | split->len = start - em->start; | |
ee20a983 JB |
643 | |
644 | if (em->block_start < EXTENT_MAP_LAST_BYTE) { | |
645 | split->orig_start = em->orig_start; | |
646 | split->block_start = em->block_start; | |
647 | ||
648 | if (compressed) | |
649 | split->block_len = em->block_len; | |
650 | else | |
651 | split->block_len = split->len; | |
652 | split->orig_block_len = max(split->block_len, | |
653 | em->orig_block_len); | |
654 | split->ram_bytes = em->ram_bytes; | |
655 | } else { | |
656 | split->orig_start = split->start; | |
657 | split->block_len = 0; | |
658 | split->block_start = em->block_start; | |
659 | split->orig_block_len = 0; | |
660 | split->ram_bytes = split->len; | |
661 | } | |
662 | ||
5dc562c5 | 663 | split->generation = gen; |
3b951516 | 664 | split->bdev = em->bdev; |
5b21f2ed | 665 | split->flags = flags; |
261507a0 | 666 | split->compress_type = em->compress_type; |
176840b3 | 667 | replace_extent_mapping(em_tree, em, split, modified); |
3b951516 CM |
668 | free_extent_map(split); |
669 | split = split2; | |
670 | split2 = NULL; | |
671 | } | |
ee20a983 | 672 | if (testend && em->start + em->len > start + len) { |
3b951516 CM |
673 | u64 diff = start + len - em->start; |
674 | ||
675 | split->start = start + len; | |
676 | split->len = em->start + em->len - (start + len); | |
677 | split->bdev = em->bdev; | |
5b21f2ed | 678 | split->flags = flags; |
261507a0 | 679 | split->compress_type = em->compress_type; |
5dc562c5 | 680 | split->generation = gen; |
ee20a983 JB |
681 | |
682 | if (em->block_start < EXTENT_MAP_LAST_BYTE) { | |
683 | split->orig_block_len = max(em->block_len, | |
b4939680 | 684 | em->orig_block_len); |
3b951516 | 685 | |
ee20a983 JB |
686 | split->ram_bytes = em->ram_bytes; |
687 | if (compressed) { | |
688 | split->block_len = em->block_len; | |
689 | split->block_start = em->block_start; | |
690 | split->orig_start = em->orig_start; | |
691 | } else { | |
692 | split->block_len = split->len; | |
693 | split->block_start = em->block_start | |
694 | + diff; | |
695 | split->orig_start = em->orig_start; | |
696 | } | |
c8b97818 | 697 | } else { |
ee20a983 JB |
698 | split->ram_bytes = split->len; |
699 | split->orig_start = split->start; | |
700 | split->block_len = 0; | |
701 | split->block_start = em->block_start; | |
702 | split->orig_block_len = 0; | |
c8b97818 | 703 | } |
3b951516 | 704 | |
176840b3 FM |
705 | if (extent_map_in_tree(em)) { |
706 | replace_extent_mapping(em_tree, em, split, | |
707 | modified); | |
708 | } else { | |
709 | ret = add_extent_mapping(em_tree, split, | |
710 | modified); | |
711 | ASSERT(ret == 0); /* Logic error */ | |
712 | } | |
3b951516 CM |
713 | free_extent_map(split); |
714 | split = NULL; | |
715 | } | |
7014cdb4 | 716 | next: |
176840b3 FM |
717 | if (extent_map_in_tree(em)) |
718 | remove_extent_mapping(em_tree, em); | |
890871be | 719 | write_unlock(&em_tree->lock); |
d1310b2e | 720 | |
a52d9a80 CM |
721 | /* once for us */ |
722 | free_extent_map(em); | |
723 | /* once for the tree*/ | |
724 | free_extent_map(em); | |
725 | } | |
3b951516 CM |
726 | if (split) |
727 | free_extent_map(split); | |
728 | if (split2) | |
729 | free_extent_map(split2); | |
a52d9a80 CM |
730 | } |
731 | ||
39279cc3 CM |
732 | /* |
733 | * this is very complex, but the basic idea is to drop all extents | |
734 | * in the range start - end. hint_block is filled in with a block number | |
735 | * that would be a good hint to the block allocator for this file. | |
736 | * | |
737 | * If an extent intersects the range but is not entirely inside the range | |
738 | * it is either truncated or split. Anything entirely inside the range | |
739 | * is deleted from the tree. | |
740 | */ | |
5dc562c5 JB |
741 | int __btrfs_drop_extents(struct btrfs_trans_handle *trans, |
742 | struct btrfs_root *root, struct inode *inode, | |
743 | struct btrfs_path *path, u64 start, u64 end, | |
1acae57b FDBM |
744 | u64 *drop_end, int drop_cache, |
745 | int replace_extent, | |
746 | u32 extent_item_size, | |
747 | int *key_inserted) | |
39279cc3 | 748 | { |
0b246afa | 749 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 750 | struct extent_buffer *leaf; |
920bbbfb | 751 | struct btrfs_file_extent_item *fi; |
00f5c795 | 752 | struct btrfs_key key; |
920bbbfb | 753 | struct btrfs_key new_key; |
4a0cc7ca | 754 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
920bbbfb YZ |
755 | u64 search_start = start; |
756 | u64 disk_bytenr = 0; | |
757 | u64 num_bytes = 0; | |
758 | u64 extent_offset = 0; | |
759 | u64 extent_end = 0; | |
62fe51c1 | 760 | u64 last_end = start; |
920bbbfb YZ |
761 | int del_nr = 0; |
762 | int del_slot = 0; | |
763 | int extent_type; | |
ccd467d6 | 764 | int recow; |
00f5c795 | 765 | int ret; |
dc7fdde3 | 766 | int modify_tree = -1; |
27cdeb70 | 767 | int update_refs; |
c3308f84 | 768 | int found = 0; |
1acae57b | 769 | int leafs_visited = 0; |
39279cc3 | 770 | |
a1ed835e | 771 | if (drop_cache) |
dcdbc059 | 772 | btrfs_drop_extent_cache(BTRFS_I(inode), start, end - 1, 0); |
a52d9a80 | 773 | |
d5f37527 | 774 | if (start >= BTRFS_I(inode)->disk_i_size && !replace_extent) |
dc7fdde3 CM |
775 | modify_tree = 0; |
776 | ||
27cdeb70 | 777 | update_refs = (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa | 778 | root == fs_info->tree_root); |
d397712b | 779 | while (1) { |
ccd467d6 | 780 | recow = 0; |
33345d01 | 781 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
dc7fdde3 | 782 | search_start, modify_tree); |
39279cc3 | 783 | if (ret < 0) |
920bbbfb YZ |
784 | break; |
785 | if (ret > 0 && path->slots[0] > 0 && search_start == start) { | |
786 | leaf = path->nodes[0]; | |
787 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1); | |
33345d01 | 788 | if (key.objectid == ino && |
920bbbfb YZ |
789 | key.type == BTRFS_EXTENT_DATA_KEY) |
790 | path->slots[0]--; | |
39279cc3 | 791 | } |
920bbbfb | 792 | ret = 0; |
1acae57b | 793 | leafs_visited++; |
8c2383c3 | 794 | next_slot: |
5f39d397 | 795 | leaf = path->nodes[0]; |
920bbbfb YZ |
796 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { |
797 | BUG_ON(del_nr > 0); | |
798 | ret = btrfs_next_leaf(root, path); | |
799 | if (ret < 0) | |
800 | break; | |
801 | if (ret > 0) { | |
802 | ret = 0; | |
803 | break; | |
8c2383c3 | 804 | } |
1acae57b | 805 | leafs_visited++; |
920bbbfb YZ |
806 | leaf = path->nodes[0]; |
807 | recow = 1; | |
808 | } | |
809 | ||
810 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
aeafbf84 FM |
811 | |
812 | if (key.objectid > ino) | |
813 | break; | |
814 | if (WARN_ON_ONCE(key.objectid < ino) || | |
815 | key.type < BTRFS_EXTENT_DATA_KEY) { | |
816 | ASSERT(del_nr == 0); | |
817 | path->slots[0]++; | |
818 | goto next_slot; | |
819 | } | |
820 | if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end) | |
920bbbfb YZ |
821 | break; |
822 | ||
823 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
824 | struct btrfs_file_extent_item); | |
825 | extent_type = btrfs_file_extent_type(leaf, fi); | |
826 | ||
827 | if (extent_type == BTRFS_FILE_EXTENT_REG || | |
828 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
829 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
830 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
831 | extent_offset = btrfs_file_extent_offset(leaf, fi); | |
832 | extent_end = key.offset + | |
833 | btrfs_file_extent_num_bytes(leaf, fi); | |
834 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
835 | extent_end = key.offset + | |
514ac8ad CM |
836 | btrfs_file_extent_inline_len(leaf, |
837 | path->slots[0], fi); | |
8c2383c3 | 838 | } else { |
aeafbf84 FM |
839 | /* can't happen */ |
840 | BUG(); | |
39279cc3 CM |
841 | } |
842 | ||
fc19c5e7 FM |
843 | /* |
844 | * Don't skip extent items representing 0 byte lengths. They | |
845 | * used to be created (bug) if while punching holes we hit | |
846 | * -ENOSPC condition. So if we find one here, just ensure we | |
847 | * delete it, otherwise we would insert a new file extent item | |
848 | * with the same key (offset) as that 0 bytes length file | |
849 | * extent item in the call to setup_items_for_insert() later | |
850 | * in this function. | |
851 | */ | |
62fe51c1 JB |
852 | if (extent_end == key.offset && extent_end >= search_start) { |
853 | last_end = extent_end; | |
fc19c5e7 | 854 | goto delete_extent_item; |
62fe51c1 | 855 | } |
fc19c5e7 | 856 | |
920bbbfb YZ |
857 | if (extent_end <= search_start) { |
858 | path->slots[0]++; | |
8c2383c3 | 859 | goto next_slot; |
39279cc3 CM |
860 | } |
861 | ||
c3308f84 | 862 | found = 1; |
920bbbfb | 863 | search_start = max(key.offset, start); |
dc7fdde3 CM |
864 | if (recow || !modify_tree) { |
865 | modify_tree = -1; | |
b3b4aa74 | 866 | btrfs_release_path(path); |
920bbbfb | 867 | continue; |
39279cc3 | 868 | } |
6643558d | 869 | |
920bbbfb YZ |
870 | /* |
871 | * | - range to drop - | | |
872 | * | -------- extent -------- | | |
873 | */ | |
874 | if (start > key.offset && end < extent_end) { | |
875 | BUG_ON(del_nr > 0); | |
00fdf13a | 876 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
3f9e3df8 | 877 | ret = -EOPNOTSUPP; |
00fdf13a LB |
878 | break; |
879 | } | |
920bbbfb YZ |
880 | |
881 | memcpy(&new_key, &key, sizeof(new_key)); | |
882 | new_key.offset = start; | |
883 | ret = btrfs_duplicate_item(trans, root, path, | |
884 | &new_key); | |
885 | if (ret == -EAGAIN) { | |
b3b4aa74 | 886 | btrfs_release_path(path); |
920bbbfb | 887 | continue; |
6643558d | 888 | } |
920bbbfb YZ |
889 | if (ret < 0) |
890 | break; | |
891 | ||
892 | leaf = path->nodes[0]; | |
893 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
894 | struct btrfs_file_extent_item); | |
895 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
896 | start - key.offset); | |
897 | ||
898 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
899 | struct btrfs_file_extent_item); | |
900 | ||
901 | extent_offset += start - key.offset; | |
902 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); | |
903 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
904 | extent_end - start); | |
905 | btrfs_mark_buffer_dirty(leaf); | |
906 | ||
5dc562c5 | 907 | if (update_refs && disk_bytenr > 0) { |
84f7d8e6 | 908 | ret = btrfs_inc_extent_ref(trans, root, |
920bbbfb YZ |
909 | disk_bytenr, num_bytes, 0, |
910 | root->root_key.objectid, | |
911 | new_key.objectid, | |
b06c4bf5 | 912 | start - extent_offset); |
79787eaa | 913 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 914 | } |
920bbbfb | 915 | key.offset = start; |
6643558d | 916 | } |
62fe51c1 JB |
917 | /* |
918 | * From here on out we will have actually dropped something, so | |
919 | * last_end can be updated. | |
920 | */ | |
921 | last_end = extent_end; | |
922 | ||
920bbbfb YZ |
923 | /* |
924 | * | ---- range to drop ----- | | |
925 | * | -------- extent -------- | | |
926 | */ | |
927 | if (start <= key.offset && end < extent_end) { | |
00fdf13a | 928 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
3f9e3df8 | 929 | ret = -EOPNOTSUPP; |
00fdf13a LB |
930 | break; |
931 | } | |
6643558d | 932 | |
920bbbfb YZ |
933 | memcpy(&new_key, &key, sizeof(new_key)); |
934 | new_key.offset = end; | |
0b246afa | 935 | btrfs_set_item_key_safe(fs_info, path, &new_key); |
6643558d | 936 | |
920bbbfb YZ |
937 | extent_offset += end - key.offset; |
938 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); | |
939 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
940 | extent_end - end); | |
941 | btrfs_mark_buffer_dirty(leaf); | |
2671485d | 942 | if (update_refs && disk_bytenr > 0) |
920bbbfb | 943 | inode_sub_bytes(inode, end - key.offset); |
920bbbfb | 944 | break; |
39279cc3 | 945 | } |
771ed689 | 946 | |
920bbbfb YZ |
947 | search_start = extent_end; |
948 | /* | |
949 | * | ---- range to drop ----- | | |
950 | * | -------- extent -------- | | |
951 | */ | |
952 | if (start > key.offset && end >= extent_end) { | |
953 | BUG_ON(del_nr > 0); | |
00fdf13a | 954 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
3f9e3df8 | 955 | ret = -EOPNOTSUPP; |
00fdf13a LB |
956 | break; |
957 | } | |
8c2383c3 | 958 | |
920bbbfb YZ |
959 | btrfs_set_file_extent_num_bytes(leaf, fi, |
960 | start - key.offset); | |
961 | btrfs_mark_buffer_dirty(leaf); | |
2671485d | 962 | if (update_refs && disk_bytenr > 0) |
920bbbfb | 963 | inode_sub_bytes(inode, extent_end - start); |
920bbbfb YZ |
964 | if (end == extent_end) |
965 | break; | |
c8b97818 | 966 | |
920bbbfb YZ |
967 | path->slots[0]++; |
968 | goto next_slot; | |
31840ae1 ZY |
969 | } |
970 | ||
920bbbfb YZ |
971 | /* |
972 | * | ---- range to drop ----- | | |
973 | * | ------ extent ------ | | |
974 | */ | |
975 | if (start <= key.offset && end >= extent_end) { | |
fc19c5e7 | 976 | delete_extent_item: |
920bbbfb YZ |
977 | if (del_nr == 0) { |
978 | del_slot = path->slots[0]; | |
979 | del_nr = 1; | |
980 | } else { | |
981 | BUG_ON(del_slot + del_nr != path->slots[0]); | |
982 | del_nr++; | |
983 | } | |
31840ae1 | 984 | |
5dc562c5 JB |
985 | if (update_refs && |
986 | extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
a76a3cd4 | 987 | inode_sub_bytes(inode, |
920bbbfb YZ |
988 | extent_end - key.offset); |
989 | extent_end = ALIGN(extent_end, | |
0b246afa | 990 | fs_info->sectorsize); |
5dc562c5 | 991 | } else if (update_refs && disk_bytenr > 0) { |
84f7d8e6 | 992 | ret = btrfs_free_extent(trans, root, |
920bbbfb YZ |
993 | disk_bytenr, num_bytes, 0, |
994 | root->root_key.objectid, | |
5d4f98a2 | 995 | key.objectid, key.offset - |
b06c4bf5 | 996 | extent_offset); |
79787eaa | 997 | BUG_ON(ret); /* -ENOMEM */ |
920bbbfb YZ |
998 | inode_sub_bytes(inode, |
999 | extent_end - key.offset); | |
31840ae1 | 1000 | } |
31840ae1 | 1001 | |
920bbbfb YZ |
1002 | if (end == extent_end) |
1003 | break; | |
1004 | ||
1005 | if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) { | |
1006 | path->slots[0]++; | |
1007 | goto next_slot; | |
1008 | } | |
1009 | ||
1010 | ret = btrfs_del_items(trans, root, path, del_slot, | |
1011 | del_nr); | |
79787eaa | 1012 | if (ret) { |
66642832 | 1013 | btrfs_abort_transaction(trans, ret); |
5dc562c5 | 1014 | break; |
79787eaa | 1015 | } |
920bbbfb YZ |
1016 | |
1017 | del_nr = 0; | |
1018 | del_slot = 0; | |
1019 | ||
b3b4aa74 | 1020 | btrfs_release_path(path); |
920bbbfb | 1021 | continue; |
39279cc3 | 1022 | } |
920bbbfb YZ |
1023 | |
1024 | BUG_ON(1); | |
39279cc3 | 1025 | } |
920bbbfb | 1026 | |
79787eaa | 1027 | if (!ret && del_nr > 0) { |
1acae57b FDBM |
1028 | /* |
1029 | * Set path->slots[0] to first slot, so that after the delete | |
1030 | * if items are move off from our leaf to its immediate left or | |
1031 | * right neighbor leafs, we end up with a correct and adjusted | |
d5f37527 | 1032 | * path->slots[0] for our insertion (if replace_extent != 0). |
1acae57b FDBM |
1033 | */ |
1034 | path->slots[0] = del_slot; | |
920bbbfb | 1035 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
79787eaa | 1036 | if (ret) |
66642832 | 1037 | btrfs_abort_transaction(trans, ret); |
d5f37527 | 1038 | } |
1acae57b | 1039 | |
d5f37527 FDBM |
1040 | leaf = path->nodes[0]; |
1041 | /* | |
1042 | * If btrfs_del_items() was called, it might have deleted a leaf, in | |
1043 | * which case it unlocked our path, so check path->locks[0] matches a | |
1044 | * write lock. | |
1045 | */ | |
1046 | if (!ret && replace_extent && leafs_visited == 1 && | |
1047 | (path->locks[0] == BTRFS_WRITE_LOCK_BLOCKING || | |
1048 | path->locks[0] == BTRFS_WRITE_LOCK) && | |
2ff7e61e | 1049 | btrfs_leaf_free_space(fs_info, leaf) >= |
d5f37527 FDBM |
1050 | sizeof(struct btrfs_item) + extent_item_size) { |
1051 | ||
1052 | key.objectid = ino; | |
1053 | key.type = BTRFS_EXTENT_DATA_KEY; | |
1054 | key.offset = start; | |
1055 | if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) { | |
1056 | struct btrfs_key slot_key; | |
1057 | ||
1058 | btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]); | |
1059 | if (btrfs_comp_cpu_keys(&key, &slot_key) > 0) | |
1060 | path->slots[0]++; | |
1acae57b | 1061 | } |
d5f37527 FDBM |
1062 | setup_items_for_insert(root, path, &key, |
1063 | &extent_item_size, | |
1064 | extent_item_size, | |
1065 | sizeof(struct btrfs_item) + | |
1066 | extent_item_size, 1); | |
1067 | *key_inserted = 1; | |
6643558d | 1068 | } |
920bbbfb | 1069 | |
1acae57b FDBM |
1070 | if (!replace_extent || !(*key_inserted)) |
1071 | btrfs_release_path(path); | |
2aaa6655 | 1072 | if (drop_end) |
62fe51c1 | 1073 | *drop_end = found ? min(end, last_end) : end; |
5dc562c5 JB |
1074 | return ret; |
1075 | } | |
1076 | ||
1077 | int btrfs_drop_extents(struct btrfs_trans_handle *trans, | |
1078 | struct btrfs_root *root, struct inode *inode, u64 start, | |
2671485d | 1079 | u64 end, int drop_cache) |
5dc562c5 JB |
1080 | { |
1081 | struct btrfs_path *path; | |
1082 | int ret; | |
1083 | ||
1084 | path = btrfs_alloc_path(); | |
1085 | if (!path) | |
1086 | return -ENOMEM; | |
2aaa6655 | 1087 | ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL, |
1acae57b | 1088 | drop_cache, 0, 0, NULL); |
920bbbfb | 1089 | btrfs_free_path(path); |
39279cc3 CM |
1090 | return ret; |
1091 | } | |
1092 | ||
d899e052 | 1093 | static int extent_mergeable(struct extent_buffer *leaf, int slot, |
6c7d54ac YZ |
1094 | u64 objectid, u64 bytenr, u64 orig_offset, |
1095 | u64 *start, u64 *end) | |
d899e052 YZ |
1096 | { |
1097 | struct btrfs_file_extent_item *fi; | |
1098 | struct btrfs_key key; | |
1099 | u64 extent_end; | |
1100 | ||
1101 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) | |
1102 | return 0; | |
1103 | ||
1104 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
1105 | if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY) | |
1106 | return 0; | |
1107 | ||
1108 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
1109 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG || | |
1110 | btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr || | |
6c7d54ac | 1111 | btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset || |
d899e052 YZ |
1112 | btrfs_file_extent_compression(leaf, fi) || |
1113 | btrfs_file_extent_encryption(leaf, fi) || | |
1114 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1115 | return 0; | |
1116 | ||
1117 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
1118 | if ((*start && *start != key.offset) || (*end && *end != extent_end)) | |
1119 | return 0; | |
1120 | ||
1121 | *start = key.offset; | |
1122 | *end = extent_end; | |
1123 | return 1; | |
1124 | } | |
1125 | ||
1126 | /* | |
1127 | * Mark extent in the range start - end as written. | |
1128 | * | |
1129 | * This changes extent type from 'pre-allocated' to 'regular'. If only | |
1130 | * part of extent is marked as written, the extent will be split into | |
1131 | * two or three. | |
1132 | */ | |
1133 | int btrfs_mark_extent_written(struct btrfs_trans_handle *trans, | |
7a6d7067 | 1134 | struct btrfs_inode *inode, u64 start, u64 end) |
d899e052 | 1135 | { |
7a6d7067 NB |
1136 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
1137 | struct btrfs_root *root = inode->root; | |
d899e052 YZ |
1138 | struct extent_buffer *leaf; |
1139 | struct btrfs_path *path; | |
1140 | struct btrfs_file_extent_item *fi; | |
1141 | struct btrfs_key key; | |
920bbbfb | 1142 | struct btrfs_key new_key; |
d899e052 YZ |
1143 | u64 bytenr; |
1144 | u64 num_bytes; | |
1145 | u64 extent_end; | |
5d4f98a2 | 1146 | u64 orig_offset; |
d899e052 YZ |
1147 | u64 other_start; |
1148 | u64 other_end; | |
920bbbfb YZ |
1149 | u64 split; |
1150 | int del_nr = 0; | |
1151 | int del_slot = 0; | |
6c7d54ac | 1152 | int recow; |
d899e052 | 1153 | int ret; |
7a6d7067 | 1154 | u64 ino = btrfs_ino(inode); |
d899e052 | 1155 | |
d899e052 | 1156 | path = btrfs_alloc_path(); |
d8926bb3 MF |
1157 | if (!path) |
1158 | return -ENOMEM; | |
d899e052 | 1159 | again: |
6c7d54ac | 1160 | recow = 0; |
920bbbfb | 1161 | split = start; |
33345d01 | 1162 | key.objectid = ino; |
d899e052 | 1163 | key.type = BTRFS_EXTENT_DATA_KEY; |
920bbbfb | 1164 | key.offset = split; |
d899e052 YZ |
1165 | |
1166 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
41415730 JB |
1167 | if (ret < 0) |
1168 | goto out; | |
d899e052 YZ |
1169 | if (ret > 0 && path->slots[0] > 0) |
1170 | path->slots[0]--; | |
1171 | ||
1172 | leaf = path->nodes[0]; | |
1173 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
9c8e63db JB |
1174 | if (key.objectid != ino || |
1175 | key.type != BTRFS_EXTENT_DATA_KEY) { | |
1176 | ret = -EINVAL; | |
1177 | btrfs_abort_transaction(trans, ret); | |
1178 | goto out; | |
1179 | } | |
d899e052 YZ |
1180 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1181 | struct btrfs_file_extent_item); | |
9c8e63db JB |
1182 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_PREALLOC) { |
1183 | ret = -EINVAL; | |
1184 | btrfs_abort_transaction(trans, ret); | |
1185 | goto out; | |
1186 | } | |
d899e052 | 1187 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
9c8e63db JB |
1188 | if (key.offset > start || extent_end < end) { |
1189 | ret = -EINVAL; | |
1190 | btrfs_abort_transaction(trans, ret); | |
1191 | goto out; | |
1192 | } | |
d899e052 YZ |
1193 | |
1194 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
1195 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
5d4f98a2 | 1196 | orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi); |
6c7d54ac YZ |
1197 | memcpy(&new_key, &key, sizeof(new_key)); |
1198 | ||
1199 | if (start == key.offset && end < extent_end) { | |
1200 | other_start = 0; | |
1201 | other_end = start; | |
1202 | if (extent_mergeable(leaf, path->slots[0] - 1, | |
33345d01 | 1203 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1204 | &other_start, &other_end)) { |
1205 | new_key.offset = end; | |
0b246afa | 1206 | btrfs_set_item_key_safe(fs_info, path, &new_key); |
6c7d54ac YZ |
1207 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1208 | struct btrfs_file_extent_item); | |
224ecce5 JB |
1209 | btrfs_set_file_extent_generation(leaf, fi, |
1210 | trans->transid); | |
6c7d54ac YZ |
1211 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1212 | extent_end - end); | |
1213 | btrfs_set_file_extent_offset(leaf, fi, | |
1214 | end - orig_offset); | |
1215 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
1216 | struct btrfs_file_extent_item); | |
224ecce5 JB |
1217 | btrfs_set_file_extent_generation(leaf, fi, |
1218 | trans->transid); | |
6c7d54ac YZ |
1219 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1220 | end - other_start); | |
1221 | btrfs_mark_buffer_dirty(leaf); | |
1222 | goto out; | |
1223 | } | |
1224 | } | |
1225 | ||
1226 | if (start > key.offset && end == extent_end) { | |
1227 | other_start = end; | |
1228 | other_end = 0; | |
1229 | if (extent_mergeable(leaf, path->slots[0] + 1, | |
33345d01 | 1230 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1231 | &other_start, &other_end)) { |
1232 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1233 | struct btrfs_file_extent_item); | |
1234 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
1235 | start - key.offset); | |
224ecce5 JB |
1236 | btrfs_set_file_extent_generation(leaf, fi, |
1237 | trans->transid); | |
6c7d54ac YZ |
1238 | path->slots[0]++; |
1239 | new_key.offset = start; | |
0b246afa | 1240 | btrfs_set_item_key_safe(fs_info, path, &new_key); |
6c7d54ac YZ |
1241 | |
1242 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1243 | struct btrfs_file_extent_item); | |
224ecce5 JB |
1244 | btrfs_set_file_extent_generation(leaf, fi, |
1245 | trans->transid); | |
6c7d54ac YZ |
1246 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1247 | other_end - start); | |
1248 | btrfs_set_file_extent_offset(leaf, fi, | |
1249 | start - orig_offset); | |
1250 | btrfs_mark_buffer_dirty(leaf); | |
1251 | goto out; | |
1252 | } | |
1253 | } | |
d899e052 | 1254 | |
920bbbfb YZ |
1255 | while (start > key.offset || end < extent_end) { |
1256 | if (key.offset == start) | |
1257 | split = end; | |
1258 | ||
920bbbfb YZ |
1259 | new_key.offset = split; |
1260 | ret = btrfs_duplicate_item(trans, root, path, &new_key); | |
1261 | if (ret == -EAGAIN) { | |
b3b4aa74 | 1262 | btrfs_release_path(path); |
920bbbfb | 1263 | goto again; |
d899e052 | 1264 | } |
79787eaa | 1265 | if (ret < 0) { |
66642832 | 1266 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
1267 | goto out; |
1268 | } | |
d899e052 | 1269 | |
920bbbfb YZ |
1270 | leaf = path->nodes[0]; |
1271 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
d899e052 | 1272 | struct btrfs_file_extent_item); |
224ecce5 | 1273 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
d899e052 | 1274 | btrfs_set_file_extent_num_bytes(leaf, fi, |
920bbbfb YZ |
1275 | split - key.offset); |
1276 | ||
1277 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1278 | struct btrfs_file_extent_item); | |
1279 | ||
224ecce5 | 1280 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
920bbbfb YZ |
1281 | btrfs_set_file_extent_offset(leaf, fi, split - orig_offset); |
1282 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
1283 | extent_end - split); | |
d899e052 YZ |
1284 | btrfs_mark_buffer_dirty(leaf); |
1285 | ||
84f7d8e6 | 1286 | ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, |
2ff7e61e | 1287 | 0, root->root_key.objectid, |
b06c4bf5 | 1288 | ino, orig_offset); |
9c8e63db JB |
1289 | if (ret) { |
1290 | btrfs_abort_transaction(trans, ret); | |
1291 | goto out; | |
1292 | } | |
d899e052 | 1293 | |
920bbbfb YZ |
1294 | if (split == start) { |
1295 | key.offset = start; | |
1296 | } else { | |
9c8e63db JB |
1297 | if (start != key.offset) { |
1298 | ret = -EINVAL; | |
1299 | btrfs_abort_transaction(trans, ret); | |
1300 | goto out; | |
1301 | } | |
d899e052 | 1302 | path->slots[0]--; |
920bbbfb | 1303 | extent_end = end; |
d899e052 | 1304 | } |
6c7d54ac | 1305 | recow = 1; |
d899e052 YZ |
1306 | } |
1307 | ||
920bbbfb YZ |
1308 | other_start = end; |
1309 | other_end = 0; | |
6c7d54ac | 1310 | if (extent_mergeable(leaf, path->slots[0] + 1, |
33345d01 | 1311 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1312 | &other_start, &other_end)) { |
1313 | if (recow) { | |
b3b4aa74 | 1314 | btrfs_release_path(path); |
6c7d54ac YZ |
1315 | goto again; |
1316 | } | |
920bbbfb YZ |
1317 | extent_end = other_end; |
1318 | del_slot = path->slots[0] + 1; | |
1319 | del_nr++; | |
84f7d8e6 | 1320 | ret = btrfs_free_extent(trans, root, bytenr, num_bytes, |
920bbbfb | 1321 | 0, root->root_key.objectid, |
b06c4bf5 | 1322 | ino, orig_offset); |
9c8e63db JB |
1323 | if (ret) { |
1324 | btrfs_abort_transaction(trans, ret); | |
1325 | goto out; | |
1326 | } | |
d899e052 | 1327 | } |
920bbbfb YZ |
1328 | other_start = 0; |
1329 | other_end = start; | |
6c7d54ac | 1330 | if (extent_mergeable(leaf, path->slots[0] - 1, |
33345d01 | 1331 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1332 | &other_start, &other_end)) { |
1333 | if (recow) { | |
b3b4aa74 | 1334 | btrfs_release_path(path); |
6c7d54ac YZ |
1335 | goto again; |
1336 | } | |
920bbbfb YZ |
1337 | key.offset = other_start; |
1338 | del_slot = path->slots[0]; | |
1339 | del_nr++; | |
84f7d8e6 | 1340 | ret = btrfs_free_extent(trans, root, bytenr, num_bytes, |
920bbbfb | 1341 | 0, root->root_key.objectid, |
b06c4bf5 | 1342 | ino, orig_offset); |
9c8e63db JB |
1343 | if (ret) { |
1344 | btrfs_abort_transaction(trans, ret); | |
1345 | goto out; | |
1346 | } | |
920bbbfb YZ |
1347 | } |
1348 | if (del_nr == 0) { | |
3f6fae95 SL |
1349 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1350 | struct btrfs_file_extent_item); | |
920bbbfb YZ |
1351 | btrfs_set_file_extent_type(leaf, fi, |
1352 | BTRFS_FILE_EXTENT_REG); | |
224ecce5 | 1353 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
920bbbfb | 1354 | btrfs_mark_buffer_dirty(leaf); |
6c7d54ac | 1355 | } else { |
3f6fae95 SL |
1356 | fi = btrfs_item_ptr(leaf, del_slot - 1, |
1357 | struct btrfs_file_extent_item); | |
6c7d54ac YZ |
1358 | btrfs_set_file_extent_type(leaf, fi, |
1359 | BTRFS_FILE_EXTENT_REG); | |
224ecce5 | 1360 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
6c7d54ac YZ |
1361 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1362 | extent_end - key.offset); | |
1363 | btrfs_mark_buffer_dirty(leaf); | |
920bbbfb | 1364 | |
6c7d54ac | 1365 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
79787eaa | 1366 | if (ret < 0) { |
66642832 | 1367 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
1368 | goto out; |
1369 | } | |
6c7d54ac | 1370 | } |
920bbbfb | 1371 | out: |
d899e052 YZ |
1372 | btrfs_free_path(path); |
1373 | return 0; | |
1374 | } | |
1375 | ||
b1bf862e CM |
1376 | /* |
1377 | * on error we return an unlocked page and the error value | |
1378 | * on success we return a locked page and 0 | |
1379 | */ | |
bb1591b4 CM |
1380 | static int prepare_uptodate_page(struct inode *inode, |
1381 | struct page *page, u64 pos, | |
b6316429 | 1382 | bool force_uptodate) |
b1bf862e CM |
1383 | { |
1384 | int ret = 0; | |
1385 | ||
09cbfeaf | 1386 | if (((pos & (PAGE_SIZE - 1)) || force_uptodate) && |
b6316429 | 1387 | !PageUptodate(page)) { |
b1bf862e CM |
1388 | ret = btrfs_readpage(NULL, page); |
1389 | if (ret) | |
1390 | return ret; | |
1391 | lock_page(page); | |
1392 | if (!PageUptodate(page)) { | |
1393 | unlock_page(page); | |
1394 | return -EIO; | |
1395 | } | |
bb1591b4 CM |
1396 | if (page->mapping != inode->i_mapping) { |
1397 | unlock_page(page); | |
1398 | return -EAGAIN; | |
1399 | } | |
b1bf862e CM |
1400 | } |
1401 | return 0; | |
1402 | } | |
1403 | ||
39279cc3 | 1404 | /* |
376cc685 | 1405 | * this just gets pages into the page cache and locks them down. |
39279cc3 | 1406 | */ |
b37392ea MX |
1407 | static noinline int prepare_pages(struct inode *inode, struct page **pages, |
1408 | size_t num_pages, loff_t pos, | |
1409 | size_t write_bytes, bool force_uptodate) | |
39279cc3 CM |
1410 | { |
1411 | int i; | |
09cbfeaf | 1412 | unsigned long index = pos >> PAGE_SHIFT; |
3b16a4e3 | 1413 | gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); |
fc28b62d | 1414 | int err = 0; |
376cc685 | 1415 | int faili; |
8c2383c3 | 1416 | |
39279cc3 | 1417 | for (i = 0; i < num_pages; i++) { |
bb1591b4 | 1418 | again: |
a94733d0 | 1419 | pages[i] = find_or_create_page(inode->i_mapping, index + i, |
e3a41a5b | 1420 | mask | __GFP_WRITE); |
39279cc3 | 1421 | if (!pages[i]) { |
b1bf862e CM |
1422 | faili = i - 1; |
1423 | err = -ENOMEM; | |
1424 | goto fail; | |
1425 | } | |
1426 | ||
1427 | if (i == 0) | |
bb1591b4 | 1428 | err = prepare_uptodate_page(inode, pages[i], pos, |
b6316429 | 1429 | force_uptodate); |
bb1591b4 CM |
1430 | if (!err && i == num_pages - 1) |
1431 | err = prepare_uptodate_page(inode, pages[i], | |
b6316429 | 1432 | pos + write_bytes, false); |
b1bf862e | 1433 | if (err) { |
09cbfeaf | 1434 | put_page(pages[i]); |
bb1591b4 CM |
1435 | if (err == -EAGAIN) { |
1436 | err = 0; | |
1437 | goto again; | |
1438 | } | |
b1bf862e CM |
1439 | faili = i - 1; |
1440 | goto fail; | |
39279cc3 | 1441 | } |
ccd467d6 | 1442 | wait_on_page_writeback(pages[i]); |
39279cc3 | 1443 | } |
376cc685 MX |
1444 | |
1445 | return 0; | |
1446 | fail: | |
1447 | while (faili >= 0) { | |
1448 | unlock_page(pages[faili]); | |
09cbfeaf | 1449 | put_page(pages[faili]); |
376cc685 MX |
1450 | faili--; |
1451 | } | |
1452 | return err; | |
1453 | ||
1454 | } | |
1455 | ||
1456 | /* | |
1457 | * This function locks the extent and properly waits for data=ordered extents | |
1458 | * to finish before allowing the pages to be modified if need. | |
1459 | * | |
1460 | * The return value: | |
1461 | * 1 - the extent is locked | |
1462 | * 0 - the extent is not locked, and everything is OK | |
1463 | * -EAGAIN - need re-prepare the pages | |
1464 | * the other < 0 number - Something wrong happens | |
1465 | */ | |
1466 | static noinline int | |
2cff578c | 1467 | lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages, |
376cc685 | 1468 | size_t num_pages, loff_t pos, |
2e78c927 | 1469 | size_t write_bytes, |
376cc685 MX |
1470 | u64 *lockstart, u64 *lockend, |
1471 | struct extent_state **cached_state) | |
1472 | { | |
2cff578c | 1473 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
376cc685 MX |
1474 | u64 start_pos; |
1475 | u64 last_pos; | |
1476 | int i; | |
1477 | int ret = 0; | |
1478 | ||
0b246afa | 1479 | start_pos = round_down(pos, fs_info->sectorsize); |
2e78c927 | 1480 | last_pos = start_pos |
da17066c | 1481 | + round_up(pos + write_bytes - start_pos, |
0b246afa | 1482 | fs_info->sectorsize) - 1; |
376cc685 | 1483 | |
e3b8a485 | 1484 | if (start_pos < inode->vfs_inode.i_size) { |
e6dcd2dc | 1485 | struct btrfs_ordered_extent *ordered; |
a7e3b975 | 1486 | |
2cff578c NB |
1487 | lock_extent_bits(&inode->io_tree, start_pos, last_pos, |
1488 | cached_state); | |
b88935bf MX |
1489 | ordered = btrfs_lookup_ordered_range(inode, start_pos, |
1490 | last_pos - start_pos + 1); | |
e6dcd2dc CM |
1491 | if (ordered && |
1492 | ordered->file_offset + ordered->len > start_pos && | |
376cc685 | 1493 | ordered->file_offset <= last_pos) { |
2cff578c | 1494 | unlock_extent_cached(&inode->io_tree, start_pos, |
e43bbe5e | 1495 | last_pos, cached_state); |
e6dcd2dc CM |
1496 | for (i = 0; i < num_pages; i++) { |
1497 | unlock_page(pages[i]); | |
09cbfeaf | 1498 | put_page(pages[i]); |
e6dcd2dc | 1499 | } |
2cff578c NB |
1500 | btrfs_start_ordered_extent(&inode->vfs_inode, |
1501 | ordered, 1); | |
b88935bf MX |
1502 | btrfs_put_ordered_extent(ordered); |
1503 | return -EAGAIN; | |
e6dcd2dc CM |
1504 | } |
1505 | if (ordered) | |
1506 | btrfs_put_ordered_extent(ordered); | |
e3b8a485 FM |
1507 | clear_extent_bit(&inode->io_tree, start_pos, last_pos, |
1508 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
1509 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
ae0f1625 | 1510 | 0, 0, cached_state); |
376cc685 MX |
1511 | *lockstart = start_pos; |
1512 | *lockend = last_pos; | |
1513 | ret = 1; | |
0762704b | 1514 | } |
376cc685 | 1515 | |
e6dcd2dc | 1516 | for (i = 0; i < num_pages; i++) { |
32c7f202 WF |
1517 | if (clear_page_dirty_for_io(pages[i])) |
1518 | account_page_redirty(pages[i]); | |
e6dcd2dc CM |
1519 | set_page_extent_mapped(pages[i]); |
1520 | WARN_ON(!PageLocked(pages[i])); | |
1521 | } | |
b1bf862e | 1522 | |
376cc685 | 1523 | return ret; |
39279cc3 CM |
1524 | } |
1525 | ||
85b7ab67 | 1526 | static noinline int check_can_nocow(struct btrfs_inode *inode, loff_t pos, |
7ee9e440 JB |
1527 | size_t *write_bytes) |
1528 | { | |
85b7ab67 NB |
1529 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
1530 | struct btrfs_root *root = inode->root; | |
7ee9e440 JB |
1531 | struct btrfs_ordered_extent *ordered; |
1532 | u64 lockstart, lockend; | |
1533 | u64 num_bytes; | |
1534 | int ret; | |
1535 | ||
ea14b57f | 1536 | ret = btrfs_start_write_no_snapshotting(root); |
8257b2dc MX |
1537 | if (!ret) |
1538 | return -ENOSPC; | |
1539 | ||
0b246afa | 1540 | lockstart = round_down(pos, fs_info->sectorsize); |
da17066c | 1541 | lockend = round_up(pos + *write_bytes, |
0b246afa | 1542 | fs_info->sectorsize) - 1; |
7ee9e440 JB |
1543 | |
1544 | while (1) { | |
85b7ab67 | 1545 | lock_extent(&inode->io_tree, lockstart, lockend); |
7ee9e440 JB |
1546 | ordered = btrfs_lookup_ordered_range(inode, lockstart, |
1547 | lockend - lockstart + 1); | |
1548 | if (!ordered) { | |
1549 | break; | |
1550 | } | |
85b7ab67 NB |
1551 | unlock_extent(&inode->io_tree, lockstart, lockend); |
1552 | btrfs_start_ordered_extent(&inode->vfs_inode, ordered, 1); | |
7ee9e440 JB |
1553 | btrfs_put_ordered_extent(ordered); |
1554 | } | |
1555 | ||
7ee9e440 | 1556 | num_bytes = lockend - lockstart + 1; |
85b7ab67 NB |
1557 | ret = can_nocow_extent(&inode->vfs_inode, lockstart, &num_bytes, |
1558 | NULL, NULL, NULL); | |
7ee9e440 JB |
1559 | if (ret <= 0) { |
1560 | ret = 0; | |
ea14b57f | 1561 | btrfs_end_write_no_snapshotting(root); |
7ee9e440 | 1562 | } else { |
c933956d MX |
1563 | *write_bytes = min_t(size_t, *write_bytes , |
1564 | num_bytes - pos + lockstart); | |
7ee9e440 JB |
1565 | } |
1566 | ||
85b7ab67 | 1567 | unlock_extent(&inode->io_tree, lockstart, lockend); |
7ee9e440 JB |
1568 | |
1569 | return ret; | |
1570 | } | |
1571 | ||
d0215f3e JB |
1572 | static noinline ssize_t __btrfs_buffered_write(struct file *file, |
1573 | struct iov_iter *i, | |
1574 | loff_t pos) | |
4b46fce2 | 1575 | { |
496ad9aa | 1576 | struct inode *inode = file_inode(file); |
0b246afa | 1577 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
11c65dcc | 1578 | struct btrfs_root *root = BTRFS_I(inode)->root; |
11c65dcc | 1579 | struct page **pages = NULL; |
376cc685 | 1580 | struct extent_state *cached_state = NULL; |
364ecf36 | 1581 | struct extent_changeset *data_reserved = NULL; |
7ee9e440 | 1582 | u64 release_bytes = 0; |
376cc685 MX |
1583 | u64 lockstart; |
1584 | u64 lockend; | |
d0215f3e JB |
1585 | size_t num_written = 0; |
1586 | int nrptrs; | |
c9149235 | 1587 | int ret = 0; |
7ee9e440 | 1588 | bool only_release_metadata = false; |
b6316429 | 1589 | bool force_page_uptodate = false; |
4b46fce2 | 1590 | |
09cbfeaf KS |
1591 | nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_SIZE), |
1592 | PAGE_SIZE / (sizeof(struct page *))); | |
142349f5 WF |
1593 | nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied); |
1594 | nrptrs = max(nrptrs, 8); | |
31e818fe | 1595 | pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL); |
d0215f3e JB |
1596 | if (!pages) |
1597 | return -ENOMEM; | |
ab93dbec | 1598 | |
d0215f3e | 1599 | while (iov_iter_count(i) > 0) { |
09cbfeaf | 1600 | size_t offset = pos & (PAGE_SIZE - 1); |
2e78c927 | 1601 | size_t sector_offset; |
d0215f3e | 1602 | size_t write_bytes = min(iov_iter_count(i), |
09cbfeaf | 1603 | nrptrs * (size_t)PAGE_SIZE - |
8c2383c3 | 1604 | offset); |
ed6078f7 | 1605 | size_t num_pages = DIV_ROUND_UP(write_bytes + offset, |
09cbfeaf | 1606 | PAGE_SIZE); |
7ee9e440 | 1607 | size_t reserve_bytes; |
d0215f3e JB |
1608 | size_t dirty_pages; |
1609 | size_t copied; | |
2e78c927 CR |
1610 | size_t dirty_sectors; |
1611 | size_t num_sectors; | |
79f015f2 | 1612 | int extents_locked; |
39279cc3 | 1613 | |
8c2383c3 | 1614 | WARN_ON(num_pages > nrptrs); |
1832a6d5 | 1615 | |
914ee295 XZ |
1616 | /* |
1617 | * Fault pages before locking them in prepare_pages | |
1618 | * to avoid recursive lock | |
1619 | */ | |
d0215f3e | 1620 | if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) { |
914ee295 | 1621 | ret = -EFAULT; |
d0215f3e | 1622 | break; |
914ee295 XZ |
1623 | } |
1624 | ||
da17066c | 1625 | sector_offset = pos & (fs_info->sectorsize - 1); |
2e78c927 | 1626 | reserve_bytes = round_up(write_bytes + sector_offset, |
da17066c | 1627 | fs_info->sectorsize); |
d9d8b2a5 | 1628 | |
364ecf36 QW |
1629 | extent_changeset_release(data_reserved); |
1630 | ret = btrfs_check_data_free_space(inode, &data_reserved, pos, | |
1631 | write_bytes); | |
c6887cd1 JB |
1632 | if (ret < 0) { |
1633 | if ((BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW | | |
1634 | BTRFS_INODE_PREALLOC)) && | |
85b7ab67 NB |
1635 | check_can_nocow(BTRFS_I(inode), pos, |
1636 | &write_bytes) > 0) { | |
c6887cd1 JB |
1637 | /* |
1638 | * For nodata cow case, no need to reserve | |
1639 | * data space. | |
1640 | */ | |
1641 | only_release_metadata = true; | |
1642 | /* | |
1643 | * our prealloc extent may be smaller than | |
1644 | * write_bytes, so scale down. | |
1645 | */ | |
1646 | num_pages = DIV_ROUND_UP(write_bytes + offset, | |
1647 | PAGE_SIZE); | |
1648 | reserve_bytes = round_up(write_bytes + | |
1649 | sector_offset, | |
da17066c | 1650 | fs_info->sectorsize); |
c6887cd1 JB |
1651 | } else { |
1652 | break; | |
1653 | } | |
1654 | } | |
1832a6d5 | 1655 | |
8b62f87b | 1656 | WARN_ON(reserve_bytes == 0); |
9f3db423 NB |
1657 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), |
1658 | reserve_bytes); | |
7ee9e440 JB |
1659 | if (ret) { |
1660 | if (!only_release_metadata) | |
bc42bda2 QW |
1661 | btrfs_free_reserved_data_space(inode, |
1662 | data_reserved, pos, | |
1663 | write_bytes); | |
8257b2dc | 1664 | else |
ea14b57f | 1665 | btrfs_end_write_no_snapshotting(root); |
7ee9e440 JB |
1666 | break; |
1667 | } | |
1668 | ||
1669 | release_bytes = reserve_bytes; | |
376cc685 | 1670 | again: |
4a64001f JB |
1671 | /* |
1672 | * This is going to setup the pages array with the number of | |
1673 | * pages we want, so we don't really need to worry about the | |
1674 | * contents of pages from loop to loop | |
1675 | */ | |
b37392ea MX |
1676 | ret = prepare_pages(inode, pages, num_pages, |
1677 | pos, write_bytes, | |
b6316429 | 1678 | force_page_uptodate); |
8b62f87b JB |
1679 | if (ret) { |
1680 | btrfs_delalloc_release_extents(BTRFS_I(inode), | |
43b18595 | 1681 | reserve_bytes, true); |
d0215f3e | 1682 | break; |
8b62f87b | 1683 | } |
39279cc3 | 1684 | |
79f015f2 GR |
1685 | extents_locked = lock_and_cleanup_extent_if_need( |
1686 | BTRFS_I(inode), pages, | |
2cff578c NB |
1687 | num_pages, pos, write_bytes, &lockstart, |
1688 | &lockend, &cached_state); | |
79f015f2 GR |
1689 | if (extents_locked < 0) { |
1690 | if (extents_locked == -EAGAIN) | |
376cc685 | 1691 | goto again; |
8b62f87b | 1692 | btrfs_delalloc_release_extents(BTRFS_I(inode), |
43b18595 | 1693 | reserve_bytes, true); |
79f015f2 | 1694 | ret = extents_locked; |
376cc685 | 1695 | break; |
376cc685 MX |
1696 | } |
1697 | ||
ee22f0c4 | 1698 | copied = btrfs_copy_from_user(pos, write_bytes, pages, i); |
b1bf862e | 1699 | |
0b246afa | 1700 | num_sectors = BTRFS_BYTES_TO_BLKS(fs_info, reserve_bytes); |
56244ef1 | 1701 | dirty_sectors = round_up(copied + sector_offset, |
0b246afa JM |
1702 | fs_info->sectorsize); |
1703 | dirty_sectors = BTRFS_BYTES_TO_BLKS(fs_info, dirty_sectors); | |
56244ef1 | 1704 | |
b1bf862e CM |
1705 | /* |
1706 | * if we have trouble faulting in the pages, fall | |
1707 | * back to one page at a time | |
1708 | */ | |
1709 | if (copied < write_bytes) | |
1710 | nrptrs = 1; | |
1711 | ||
b6316429 JB |
1712 | if (copied == 0) { |
1713 | force_page_uptodate = true; | |
56244ef1 | 1714 | dirty_sectors = 0; |
b1bf862e | 1715 | dirty_pages = 0; |
b6316429 JB |
1716 | } else { |
1717 | force_page_uptodate = false; | |
ed6078f7 | 1718 | dirty_pages = DIV_ROUND_UP(copied + offset, |
09cbfeaf | 1719 | PAGE_SIZE); |
b6316429 | 1720 | } |
914ee295 | 1721 | |
2e78c927 | 1722 | if (num_sectors > dirty_sectors) { |
8b8b08cb CM |
1723 | /* release everything except the sectors we dirtied */ |
1724 | release_bytes -= dirty_sectors << | |
0b246afa | 1725 | fs_info->sb->s_blocksize_bits; |
485290a7 | 1726 | if (only_release_metadata) { |
691fa059 | 1727 | btrfs_delalloc_release_metadata(BTRFS_I(inode), |
43b18595 | 1728 | release_bytes, true); |
485290a7 QW |
1729 | } else { |
1730 | u64 __pos; | |
1731 | ||
da17066c | 1732 | __pos = round_down(pos, |
0b246afa | 1733 | fs_info->sectorsize) + |
09cbfeaf | 1734 | (dirty_pages << PAGE_SHIFT); |
bc42bda2 QW |
1735 | btrfs_delalloc_release_space(inode, |
1736 | data_reserved, __pos, | |
43b18595 | 1737 | release_bytes, true); |
485290a7 | 1738 | } |
914ee295 XZ |
1739 | } |
1740 | ||
2e78c927 | 1741 | release_bytes = round_up(copied + sector_offset, |
0b246afa | 1742 | fs_info->sectorsize); |
376cc685 MX |
1743 | |
1744 | if (copied > 0) | |
2ff7e61e | 1745 | ret = btrfs_dirty_pages(inode, pages, dirty_pages, |
94f45071 | 1746 | pos, copied, &cached_state); |
79f015f2 | 1747 | if (extents_locked) |
376cc685 | 1748 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, |
e43bbe5e | 1749 | lockstart, lockend, &cached_state); |
43b18595 | 1750 | btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes, |
336a8bb8 | 1751 | true); |
f1de9683 MX |
1752 | if (ret) { |
1753 | btrfs_drop_pages(pages, num_pages); | |
376cc685 | 1754 | break; |
f1de9683 | 1755 | } |
39279cc3 | 1756 | |
376cc685 | 1757 | release_bytes = 0; |
8257b2dc | 1758 | if (only_release_metadata) |
ea14b57f | 1759 | btrfs_end_write_no_snapshotting(root); |
8257b2dc | 1760 | |
7ee9e440 | 1761 | if (only_release_metadata && copied > 0) { |
da17066c | 1762 | lockstart = round_down(pos, |
0b246afa | 1763 | fs_info->sectorsize); |
da17066c | 1764 | lockend = round_up(pos + copied, |
0b246afa | 1765 | fs_info->sectorsize) - 1; |
7ee9e440 JB |
1766 | |
1767 | set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
1768 | lockend, EXTENT_NORESERVE, NULL, | |
1769 | NULL, GFP_NOFS); | |
1770 | only_release_metadata = false; | |
1771 | } | |
1772 | ||
f1de9683 MX |
1773 | btrfs_drop_pages(pages, num_pages); |
1774 | ||
d0215f3e JB |
1775 | cond_resched(); |
1776 | ||
d0e1d66b | 1777 | balance_dirty_pages_ratelimited(inode->i_mapping); |
0b246afa | 1778 | if (dirty_pages < (fs_info->nodesize >> PAGE_SHIFT) + 1) |
2ff7e61e | 1779 | btrfs_btree_balance_dirty(fs_info); |
cb843a6f | 1780 | |
914ee295 XZ |
1781 | pos += copied; |
1782 | num_written += copied; | |
d0215f3e | 1783 | } |
39279cc3 | 1784 | |
d0215f3e JB |
1785 | kfree(pages); |
1786 | ||
7ee9e440 | 1787 | if (release_bytes) { |
8257b2dc | 1788 | if (only_release_metadata) { |
ea14b57f | 1789 | btrfs_end_write_no_snapshotting(root); |
691fa059 | 1790 | btrfs_delalloc_release_metadata(BTRFS_I(inode), |
43b18595 | 1791 | release_bytes, true); |
8257b2dc | 1792 | } else { |
bc42bda2 QW |
1793 | btrfs_delalloc_release_space(inode, data_reserved, |
1794 | round_down(pos, fs_info->sectorsize), | |
43b18595 | 1795 | release_bytes, true); |
8257b2dc | 1796 | } |
7ee9e440 JB |
1797 | } |
1798 | ||
364ecf36 | 1799 | extent_changeset_free(data_reserved); |
d0215f3e JB |
1800 | return num_written ? num_written : ret; |
1801 | } | |
1802 | ||
1af5bb49 | 1803 | static ssize_t __btrfs_direct_write(struct kiocb *iocb, struct iov_iter *from) |
d0215f3e JB |
1804 | { |
1805 | struct file *file = iocb->ki_filp; | |
728404da | 1806 | struct inode *inode = file_inode(file); |
1af5bb49 | 1807 | loff_t pos = iocb->ki_pos; |
d0215f3e JB |
1808 | ssize_t written; |
1809 | ssize_t written_buffered; | |
1810 | loff_t endbyte; | |
1811 | int err; | |
1812 | ||
1af5bb49 | 1813 | written = generic_file_direct_write(iocb, from); |
d0215f3e | 1814 | |
0c949334 | 1815 | if (written < 0 || !iov_iter_count(from)) |
d0215f3e JB |
1816 | return written; |
1817 | ||
1818 | pos += written; | |
0ae5e4d3 | 1819 | written_buffered = __btrfs_buffered_write(file, from, pos); |
d0215f3e JB |
1820 | if (written_buffered < 0) { |
1821 | err = written_buffered; | |
1822 | goto out; | |
39279cc3 | 1823 | } |
075bdbdb FM |
1824 | /* |
1825 | * Ensure all data is persisted. We want the next direct IO read to be | |
1826 | * able to read what was just written. | |
1827 | */ | |
d0215f3e | 1828 | endbyte = pos + written_buffered - 1; |
728404da | 1829 | err = btrfs_fdatawrite_range(inode, pos, endbyte); |
075bdbdb FM |
1830 | if (err) |
1831 | goto out; | |
728404da | 1832 | err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte); |
d0215f3e JB |
1833 | if (err) |
1834 | goto out; | |
1835 | written += written_buffered; | |
867c4f93 | 1836 | iocb->ki_pos = pos + written_buffered; |
09cbfeaf KS |
1837 | invalidate_mapping_pages(file->f_mapping, pos >> PAGE_SHIFT, |
1838 | endbyte >> PAGE_SHIFT); | |
39279cc3 | 1839 | out: |
d0215f3e JB |
1840 | return written ? written : err; |
1841 | } | |
5b92ee72 | 1842 | |
6c760c07 JB |
1843 | static void update_time_for_write(struct inode *inode) |
1844 | { | |
95582b00 | 1845 | struct timespec64 now; |
6c760c07 JB |
1846 | |
1847 | if (IS_NOCMTIME(inode)) | |
1848 | return; | |
1849 | ||
c2050a45 | 1850 | now = current_time(inode); |
95582b00 | 1851 | if (!timespec64_equal(&inode->i_mtime, &now)) |
6c760c07 JB |
1852 | inode->i_mtime = now; |
1853 | ||
95582b00 | 1854 | if (!timespec64_equal(&inode->i_ctime, &now)) |
6c760c07 JB |
1855 | inode->i_ctime = now; |
1856 | ||
1857 | if (IS_I_VERSION(inode)) | |
1858 | inode_inc_iversion(inode); | |
1859 | } | |
1860 | ||
b30ac0fc AV |
1861 | static ssize_t btrfs_file_write_iter(struct kiocb *iocb, |
1862 | struct iov_iter *from) | |
d0215f3e JB |
1863 | { |
1864 | struct file *file = iocb->ki_filp; | |
496ad9aa | 1865 | struct inode *inode = file_inode(file); |
0b246afa | 1866 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
d0215f3e | 1867 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0c1a98c8 | 1868 | u64 start_pos; |
3ac0d7b9 | 1869 | u64 end_pos; |
d0215f3e | 1870 | ssize_t num_written = 0; |
b812ce28 | 1871 | bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host); |
3309dd04 | 1872 | ssize_t err; |
ff0fa732 | 1873 | loff_t pos; |
edf064e7 | 1874 | size_t count = iov_iter_count(from); |
27772b68 CR |
1875 | loff_t oldsize; |
1876 | int clean_page = 0; | |
d0215f3e | 1877 | |
91f9943e CH |
1878 | if (!(iocb->ki_flags & IOCB_DIRECT) && |
1879 | (iocb->ki_flags & IOCB_NOWAIT)) | |
1880 | return -EOPNOTSUPP; | |
1881 | ||
ff0fa732 GR |
1882 | if (!inode_trylock(inode)) { |
1883 | if (iocb->ki_flags & IOCB_NOWAIT) | |
edf064e7 | 1884 | return -EAGAIN; |
ff0fa732 GR |
1885 | inode_lock(inode); |
1886 | } | |
1887 | ||
1888 | err = generic_write_checks(iocb, from); | |
1889 | if (err <= 0) { | |
1890 | inode_unlock(inode); | |
1891 | return err; | |
1892 | } | |
1893 | ||
1894 | pos = iocb->ki_pos; | |
1895 | if (iocb->ki_flags & IOCB_NOWAIT) { | |
edf064e7 GR |
1896 | /* |
1897 | * We will allocate space in case nodatacow is not set, | |
1898 | * so bail | |
1899 | */ | |
1900 | if (!(BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW | | |
1901 | BTRFS_INODE_PREALLOC)) || | |
1902 | check_can_nocow(BTRFS_I(inode), pos, &count) <= 0) { | |
1903 | inode_unlock(inode); | |
1904 | return -EAGAIN; | |
1905 | } | |
d0215f3e JB |
1906 | } |
1907 | ||
3309dd04 | 1908 | current->backing_dev_info = inode_to_bdi(inode); |
5fa8e0a1 | 1909 | err = file_remove_privs(file); |
d0215f3e | 1910 | if (err) { |
5955102c | 1911 | inode_unlock(inode); |
d0215f3e JB |
1912 | goto out; |
1913 | } | |
1914 | ||
1915 | /* | |
1916 | * If BTRFS flips readonly due to some impossible error | |
1917 | * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR), | |
1918 | * although we have opened a file as writable, we have | |
1919 | * to stop this write operation to ensure FS consistency. | |
1920 | */ | |
0b246afa | 1921 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
5955102c | 1922 | inode_unlock(inode); |
d0215f3e JB |
1923 | err = -EROFS; |
1924 | goto out; | |
1925 | } | |
1926 | ||
6c760c07 JB |
1927 | /* |
1928 | * We reserve space for updating the inode when we reserve space for the | |
1929 | * extent we are going to write, so we will enospc out there. We don't | |
1930 | * need to start yet another transaction to update the inode as we will | |
1931 | * update the inode when we finish writing whatever data we write. | |
1932 | */ | |
1933 | update_time_for_write(inode); | |
d0215f3e | 1934 | |
0b246afa | 1935 | start_pos = round_down(pos, fs_info->sectorsize); |
27772b68 CR |
1936 | oldsize = i_size_read(inode); |
1937 | if (start_pos > oldsize) { | |
3ac0d7b9 | 1938 | /* Expand hole size to cover write data, preventing empty gap */ |
da17066c | 1939 | end_pos = round_up(pos + count, |
0b246afa | 1940 | fs_info->sectorsize); |
27772b68 | 1941 | err = btrfs_cont_expand(inode, oldsize, end_pos); |
0c1a98c8 | 1942 | if (err) { |
5955102c | 1943 | inode_unlock(inode); |
0c1a98c8 MX |
1944 | goto out; |
1945 | } | |
0b246afa | 1946 | if (start_pos > round_up(oldsize, fs_info->sectorsize)) |
27772b68 | 1947 | clean_page = 1; |
0c1a98c8 MX |
1948 | } |
1949 | ||
b812ce28 JB |
1950 | if (sync) |
1951 | atomic_inc(&BTRFS_I(inode)->sync_writers); | |
1952 | ||
2ba48ce5 | 1953 | if (iocb->ki_flags & IOCB_DIRECT) { |
1af5bb49 | 1954 | num_written = __btrfs_direct_write(iocb, from); |
d0215f3e | 1955 | } else { |
b30ac0fc | 1956 | num_written = __btrfs_buffered_write(file, from, pos); |
d0215f3e | 1957 | if (num_written > 0) |
867c4f93 | 1958 | iocb->ki_pos = pos + num_written; |
27772b68 CR |
1959 | if (clean_page) |
1960 | pagecache_isize_extended(inode, oldsize, | |
1961 | i_size_read(inode)); | |
d0215f3e JB |
1962 | } |
1963 | ||
5955102c | 1964 | inode_unlock(inode); |
2ff3e9b6 | 1965 | |
5a3f23d5 | 1966 | /* |
6c760c07 JB |
1967 | * We also have to set last_sub_trans to the current log transid, |
1968 | * otherwise subsequent syncs to a file that's been synced in this | |
bb7ab3b9 | 1969 | * transaction will appear to have already occurred. |
5a3f23d5 | 1970 | */ |
2f2ff0ee | 1971 | spin_lock(&BTRFS_I(inode)->lock); |
6c760c07 | 1972 | BTRFS_I(inode)->last_sub_trans = root->log_transid; |
2f2ff0ee | 1973 | spin_unlock(&BTRFS_I(inode)->lock); |
e2592217 CH |
1974 | if (num_written > 0) |
1975 | num_written = generic_write_sync(iocb, num_written); | |
0a3404dc | 1976 | |
b812ce28 JB |
1977 | if (sync) |
1978 | atomic_dec(&BTRFS_I(inode)->sync_writers); | |
0a3404dc | 1979 | out: |
39279cc3 | 1980 | current->backing_dev_info = NULL; |
39279cc3 CM |
1981 | return num_written ? num_written : err; |
1982 | } | |
1983 | ||
d397712b | 1984 | int btrfs_release_file(struct inode *inode, struct file *filp) |
e1b81e67 | 1985 | { |
23b5ec74 JB |
1986 | struct btrfs_file_private *private = filp->private_data; |
1987 | ||
23b5ec74 JB |
1988 | if (private && private->filldir_buf) |
1989 | kfree(private->filldir_buf); | |
1990 | kfree(private); | |
1991 | filp->private_data = NULL; | |
1992 | ||
f6dc45c7 CM |
1993 | /* |
1994 | * ordered_data_close is set by settattr when we are about to truncate | |
1995 | * a file from a non-zero size to a zero size. This tries to | |
1996 | * flush down new bytes that may have been written if the | |
1997 | * application were using truncate to replace a file in place. | |
1998 | */ | |
1999 | if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, | |
2000 | &BTRFS_I(inode)->runtime_flags)) | |
2001 | filemap_flush(inode->i_mapping); | |
e1b81e67 M |
2002 | return 0; |
2003 | } | |
2004 | ||
669249ee FM |
2005 | static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end) |
2006 | { | |
2007 | int ret; | |
343e4fc1 | 2008 | struct blk_plug plug; |
669249ee | 2009 | |
343e4fc1 LB |
2010 | /* |
2011 | * This is only called in fsync, which would do synchronous writes, so | |
2012 | * a plug can merge adjacent IOs as much as possible. Esp. in case of | |
2013 | * multiple disks using raid profile, a large IO can be split to | |
2014 | * several segments of stripe length (currently 64K). | |
2015 | */ | |
2016 | blk_start_plug(&plug); | |
669249ee | 2017 | atomic_inc(&BTRFS_I(inode)->sync_writers); |
728404da | 2018 | ret = btrfs_fdatawrite_range(inode, start, end); |
669249ee | 2019 | atomic_dec(&BTRFS_I(inode)->sync_writers); |
343e4fc1 | 2020 | blk_finish_plug(&plug); |
669249ee FM |
2021 | |
2022 | return ret; | |
2023 | } | |
2024 | ||
d352ac68 CM |
2025 | /* |
2026 | * fsync call for both files and directories. This logs the inode into | |
2027 | * the tree log instead of forcing full commits whenever possible. | |
2028 | * | |
2029 | * It needs to call filemap_fdatawait so that all ordered extent updates are | |
2030 | * in the metadata btree are up to date for copying to the log. | |
2031 | * | |
2032 | * It drops the inode mutex before doing the tree log commit. This is an | |
2033 | * important optimization for directories because holding the mutex prevents | |
2034 | * new operations on the dir while we write to disk. | |
2035 | */ | |
02c24a82 | 2036 | int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync) |
39279cc3 | 2037 | { |
de17e793 | 2038 | struct dentry *dentry = file_dentry(file); |
2b0143b5 | 2039 | struct inode *inode = d_inode(dentry); |
0b246afa | 2040 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 2041 | struct btrfs_root *root = BTRFS_I(inode)->root; |
39279cc3 | 2042 | struct btrfs_trans_handle *trans; |
8b050d35 | 2043 | struct btrfs_log_ctx ctx; |
333427a5 | 2044 | int ret = 0, err; |
897ca819 | 2045 | bool full_sync = false; |
9dcbeed4 | 2046 | u64 len; |
39279cc3 | 2047 | |
9dcbeed4 DS |
2048 | /* |
2049 | * The range length can be represented by u64, we have to do the typecasts | |
2050 | * to avoid signed overflow if it's [0, LLONG_MAX] eg. from fsync() | |
2051 | */ | |
2052 | len = (u64)end - (u64)start + 1; | |
1abe9b8a | 2053 | trace_btrfs_sync_file(file, datasync); |
257c62e1 | 2054 | |
ebb70442 LB |
2055 | btrfs_init_log_ctx(&ctx, inode); |
2056 | ||
90abccf2 MX |
2057 | /* |
2058 | * We write the dirty pages in the range and wait until they complete | |
2059 | * out of the ->i_mutex. If so, we can flush the dirty pages by | |
2ab28f32 JB |
2060 | * multi-task, and make the performance up. See |
2061 | * btrfs_wait_ordered_range for an explanation of the ASYNC check. | |
90abccf2 | 2062 | */ |
669249ee | 2063 | ret = start_ordered_ops(inode, start, end); |
90abccf2 | 2064 | if (ret) |
333427a5 | 2065 | goto out; |
90abccf2 | 2066 | |
5955102c | 2067 | inode_lock(inode); |
2ecb7923 | 2068 | atomic_inc(&root->log_batch); |
2ab28f32 JB |
2069 | full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
2070 | &BTRFS_I(inode)->runtime_flags); | |
669249ee FM |
2071 | /* |
2072 | * We might have have had more pages made dirty after calling | |
2073 | * start_ordered_ops and before acquiring the inode's i_mutex. | |
2074 | */ | |
0ef8b726 | 2075 | if (full_sync) { |
669249ee FM |
2076 | /* |
2077 | * For a full sync, we need to make sure any ordered operations | |
2078 | * start and finish before we start logging the inode, so that | |
2079 | * all extents are persisted and the respective file extent | |
2080 | * items are in the fs/subvol btree. | |
2081 | */ | |
b659ef02 | 2082 | ret = btrfs_wait_ordered_range(inode, start, len); |
669249ee FM |
2083 | } else { |
2084 | /* | |
2085 | * Start any new ordered operations before starting to log the | |
2086 | * inode. We will wait for them to finish in btrfs_sync_log(). | |
2087 | * | |
2088 | * Right before acquiring the inode's mutex, we might have new | |
2089 | * writes dirtying pages, which won't immediately start the | |
2090 | * respective ordered operations - that is done through the | |
2091 | * fill_delalloc callbacks invoked from the writepage and | |
2092 | * writepages address space operations. So make sure we start | |
2093 | * all ordered operations before starting to log our inode. Not | |
2094 | * doing this means that while logging the inode, writeback | |
2095 | * could start and invoke writepage/writepages, which would call | |
2096 | * the fill_delalloc callbacks (cow_file_range, | |
2097 | * submit_compressed_extents). These callbacks add first an | |
2098 | * extent map to the modified list of extents and then create | |
2099 | * the respective ordered operation, which means in | |
2100 | * tree-log.c:btrfs_log_inode() we might capture all existing | |
2101 | * ordered operations (with btrfs_get_logged_extents()) before | |
2102 | * the fill_delalloc callback adds its ordered operation, and by | |
2103 | * the time we visit the modified list of extent maps (with | |
2104 | * btrfs_log_changed_extents()), we see and process the extent | |
2105 | * map they created. We then use the extent map to construct a | |
2106 | * file extent item for logging without waiting for the | |
2107 | * respective ordered operation to finish - this file extent | |
2108 | * item points to a disk location that might not have yet been | |
2109 | * written to, containing random data - so after a crash a log | |
2110 | * replay will make our inode have file extent items that point | |
2111 | * to disk locations containing invalid data, as we returned | |
2112 | * success to userspace without waiting for the respective | |
2113 | * ordered operation to finish, because it wasn't captured by | |
2114 | * btrfs_get_logged_extents(). | |
2115 | */ | |
2116 | ret = start_ordered_ops(inode, start, end); | |
2117 | } | |
2118 | if (ret) { | |
5955102c | 2119 | inode_unlock(inode); |
669249ee | 2120 | goto out; |
0ef8b726 | 2121 | } |
2ecb7923 | 2122 | atomic_inc(&root->log_batch); |
257c62e1 | 2123 | |
39279cc3 | 2124 | /* |
3a8b36f3 FM |
2125 | * If the last transaction that changed this file was before the current |
2126 | * transaction and we have the full sync flag set in our inode, we can | |
2127 | * bail out now without any syncing. | |
2128 | * | |
2129 | * Note that we can't bail out if the full sync flag isn't set. This is | |
2130 | * because when the full sync flag is set we start all ordered extents | |
2131 | * and wait for them to fully complete - when they complete they update | |
2132 | * the inode's last_trans field through: | |
2133 | * | |
2134 | * btrfs_finish_ordered_io() -> | |
2135 | * btrfs_update_inode_fallback() -> | |
2136 | * btrfs_update_inode() -> | |
2137 | * btrfs_set_inode_last_trans() | |
2138 | * | |
2139 | * So we are sure that last_trans is up to date and can do this check to | |
2140 | * bail out safely. For the fast path, when the full sync flag is not | |
2141 | * set in our inode, we can not do it because we start only our ordered | |
2142 | * extents and don't wait for them to complete (that is when | |
2143 | * btrfs_finish_ordered_io runs), so here at this point their last_trans | |
2144 | * value might be less than or equals to fs_info->last_trans_committed, | |
2145 | * and setting a speculative last_trans for an inode when a buffered | |
2146 | * write is made (such as fs_info->generation + 1 for example) would not | |
2147 | * be reliable since after setting the value and before fsync is called | |
2148 | * any number of transactions can start and commit (transaction kthread | |
2149 | * commits the current transaction periodically), and a transaction | |
2150 | * commit does not start nor waits for ordered extents to complete. | |
257c62e1 | 2151 | */ |
a4abeea4 | 2152 | smp_mb(); |
0f8939b8 | 2153 | if (btrfs_inode_in_log(BTRFS_I(inode), fs_info->generation) || |
affc0ff9 | 2154 | (full_sync && BTRFS_I(inode)->last_trans <= |
0b246afa | 2155 | fs_info->last_trans_committed) || |
affc0ff9 FM |
2156 | (!btrfs_have_ordered_extents_in_range(inode, start, len) && |
2157 | BTRFS_I(inode)->last_trans | |
0b246afa | 2158 | <= fs_info->last_trans_committed)) { |
5dc562c5 | 2159 | /* |
01327610 | 2160 | * We've had everything committed since the last time we were |
5dc562c5 JB |
2161 | * modified so clear this flag in case it was set for whatever |
2162 | * reason, it's no longer relevant. | |
2163 | */ | |
2164 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
2165 | &BTRFS_I(inode)->runtime_flags); | |
0596a904 FM |
2166 | /* |
2167 | * An ordered extent might have started before and completed | |
2168 | * already with io errors, in which case the inode was not | |
2169 | * updated and we end up here. So check the inode's mapping | |
333427a5 JL |
2170 | * for any errors that might have happened since we last |
2171 | * checked called fsync. | |
0596a904 | 2172 | */ |
333427a5 | 2173 | ret = filemap_check_wb_err(inode->i_mapping, file->f_wb_err); |
5955102c | 2174 | inode_unlock(inode); |
15ee9bc7 JB |
2175 | goto out; |
2176 | } | |
15ee9bc7 | 2177 | |
5039eddc JB |
2178 | /* |
2179 | * We use start here because we will need to wait on the IO to complete | |
2180 | * in btrfs_sync_log, which could require joining a transaction (for | |
2181 | * example checking cross references in the nocow path). If we use join | |
2182 | * here we could get into a situation where we're waiting on IO to | |
2183 | * happen that is blocked on a transaction trying to commit. With start | |
2184 | * we inc the extwriter counter, so we wait for all extwriters to exit | |
2185 | * before we start blocking join'ers. This comment is to keep somebody | |
2186 | * from thinking they are super smart and changing this to | |
2187 | * btrfs_join_transaction *cough*Josef*cough*. | |
2188 | */ | |
a22285a6 YZ |
2189 | trans = btrfs_start_transaction(root, 0); |
2190 | if (IS_ERR(trans)) { | |
2191 | ret = PTR_ERR(trans); | |
5955102c | 2192 | inode_unlock(inode); |
39279cc3 CM |
2193 | goto out; |
2194 | } | |
5039eddc | 2195 | trans->sync = true; |
e02119d5 | 2196 | |
e5b84f7a | 2197 | ret = btrfs_log_dentry_safe(trans, dentry, start, end, &ctx); |
02c24a82 | 2198 | if (ret < 0) { |
a0634be5 FDBM |
2199 | /* Fallthrough and commit/free transaction. */ |
2200 | ret = 1; | |
02c24a82 | 2201 | } |
49eb7e46 CM |
2202 | |
2203 | /* we've logged all the items and now have a consistent | |
2204 | * version of the file in the log. It is possible that | |
2205 | * someone will come in and modify the file, but that's | |
2206 | * fine because the log is consistent on disk, and we | |
2207 | * have references to all of the file's extents | |
2208 | * | |
2209 | * It is possible that someone will come in and log the | |
2210 | * file again, but that will end up using the synchronization | |
2211 | * inside btrfs_sync_log to keep things safe. | |
2212 | */ | |
5955102c | 2213 | inode_unlock(inode); |
49eb7e46 | 2214 | |
8407f553 FM |
2215 | /* |
2216 | * If any of the ordered extents had an error, just return it to user | |
2217 | * space, so that the application knows some writes didn't succeed and | |
2218 | * can take proper action (retry for e.g.). Blindly committing the | |
2219 | * transaction in this case, would fool userspace that everything was | |
2220 | * successful. And we also want to make sure our log doesn't contain | |
2221 | * file extent items pointing to extents that weren't fully written to - | |
2222 | * just like in the non fast fsync path, where we check for the ordered | |
2223 | * operation's error flag before writing to the log tree and return -EIO | |
2224 | * if any of them had this flag set (btrfs_wait_ordered_range) - | |
2225 | * therefore we need to check for errors in the ordered operations, | |
2226 | * which are indicated by ctx.io_err. | |
2227 | */ | |
2228 | if (ctx.io_err) { | |
3a45bb20 | 2229 | btrfs_end_transaction(trans); |
8407f553 FM |
2230 | ret = ctx.io_err; |
2231 | goto out; | |
2232 | } | |
2233 | ||
257c62e1 | 2234 | if (ret != BTRFS_NO_LOG_SYNC) { |
0ef8b726 | 2235 | if (!ret) { |
8b050d35 | 2236 | ret = btrfs_sync_log(trans, root, &ctx); |
0ef8b726 | 2237 | if (!ret) { |
3a45bb20 | 2238 | ret = btrfs_end_transaction(trans); |
0ef8b726 | 2239 | goto out; |
2ab28f32 | 2240 | } |
257c62e1 | 2241 | } |
0ef8b726 | 2242 | if (!full_sync) { |
9dcbeed4 | 2243 | ret = btrfs_wait_ordered_range(inode, start, len); |
b05fd874 | 2244 | if (ret) { |
3a45bb20 | 2245 | btrfs_end_transaction(trans); |
0ef8b726 | 2246 | goto out; |
b05fd874 | 2247 | } |
0ef8b726 | 2248 | } |
3a45bb20 | 2249 | ret = btrfs_commit_transaction(trans); |
257c62e1 | 2250 | } else { |
3a45bb20 | 2251 | ret = btrfs_end_transaction(trans); |
e02119d5 | 2252 | } |
39279cc3 | 2253 | out: |
ebb70442 | 2254 | ASSERT(list_empty(&ctx.list)); |
333427a5 JL |
2255 | err = file_check_and_advance_wb_err(file); |
2256 | if (!ret) | |
2257 | ret = err; | |
014e4ac4 | 2258 | return ret > 0 ? -EIO : ret; |
39279cc3 CM |
2259 | } |
2260 | ||
f0f37e2f | 2261 | static const struct vm_operations_struct btrfs_file_vm_ops = { |
92fee66d | 2262 | .fault = filemap_fault, |
f1820361 | 2263 | .map_pages = filemap_map_pages, |
9ebefb18 CM |
2264 | .page_mkwrite = btrfs_page_mkwrite, |
2265 | }; | |
2266 | ||
2267 | static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma) | |
2268 | { | |
058a457e MX |
2269 | struct address_space *mapping = filp->f_mapping; |
2270 | ||
2271 | if (!mapping->a_ops->readpage) | |
2272 | return -ENOEXEC; | |
2273 | ||
9ebefb18 | 2274 | file_accessed(filp); |
058a457e | 2275 | vma->vm_ops = &btrfs_file_vm_ops; |
058a457e | 2276 | |
9ebefb18 CM |
2277 | return 0; |
2278 | } | |
2279 | ||
35339c24 | 2280 | static int hole_mergeable(struct btrfs_inode *inode, struct extent_buffer *leaf, |
2aaa6655 JB |
2281 | int slot, u64 start, u64 end) |
2282 | { | |
2283 | struct btrfs_file_extent_item *fi; | |
2284 | struct btrfs_key key; | |
2285 | ||
2286 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) | |
2287 | return 0; | |
2288 | ||
2289 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
35339c24 | 2290 | if (key.objectid != btrfs_ino(inode) || |
2aaa6655 JB |
2291 | key.type != BTRFS_EXTENT_DATA_KEY) |
2292 | return 0; | |
2293 | ||
2294 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
2295 | ||
2296 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2297 | return 0; | |
2298 | ||
2299 | if (btrfs_file_extent_disk_bytenr(leaf, fi)) | |
2300 | return 0; | |
2301 | ||
2302 | if (key.offset == end) | |
2303 | return 1; | |
2304 | if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start) | |
2305 | return 1; | |
2306 | return 0; | |
2307 | } | |
2308 | ||
a012a74e NB |
2309 | static int fill_holes(struct btrfs_trans_handle *trans, |
2310 | struct btrfs_inode *inode, | |
2311 | struct btrfs_path *path, u64 offset, u64 end) | |
2aaa6655 | 2312 | { |
a012a74e NB |
2313 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
2314 | struct btrfs_root *root = inode->root; | |
2aaa6655 JB |
2315 | struct extent_buffer *leaf; |
2316 | struct btrfs_file_extent_item *fi; | |
2317 | struct extent_map *hole_em; | |
a012a74e | 2318 | struct extent_map_tree *em_tree = &inode->extent_tree; |
2aaa6655 JB |
2319 | struct btrfs_key key; |
2320 | int ret; | |
2321 | ||
0b246afa | 2322 | if (btrfs_fs_incompat(fs_info, NO_HOLES)) |
16e7549f JB |
2323 | goto out; |
2324 | ||
a012a74e | 2325 | key.objectid = btrfs_ino(inode); |
2aaa6655 JB |
2326 | key.type = BTRFS_EXTENT_DATA_KEY; |
2327 | key.offset = offset; | |
2328 | ||
2aaa6655 | 2329 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
f94480bd JB |
2330 | if (ret <= 0) { |
2331 | /* | |
2332 | * We should have dropped this offset, so if we find it then | |
2333 | * something has gone horribly wrong. | |
2334 | */ | |
2335 | if (ret == 0) | |
2336 | ret = -EINVAL; | |
2aaa6655 | 2337 | return ret; |
f94480bd | 2338 | } |
2aaa6655 JB |
2339 | |
2340 | leaf = path->nodes[0]; | |
a012a74e | 2341 | if (hole_mergeable(inode, leaf, path->slots[0] - 1, offset, end)) { |
2aaa6655 JB |
2342 | u64 num_bytes; |
2343 | ||
2344 | path->slots[0]--; | |
2345 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2346 | struct btrfs_file_extent_item); | |
2347 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + | |
2348 | end - offset; | |
2349 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2350 | btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes); | |
2351 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
2352 | btrfs_mark_buffer_dirty(leaf); | |
2353 | goto out; | |
2354 | } | |
2355 | ||
1707e26d | 2356 | if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) { |
2aaa6655 JB |
2357 | u64 num_bytes; |
2358 | ||
2aaa6655 | 2359 | key.offset = offset; |
0b246afa | 2360 | btrfs_set_item_key_safe(fs_info, path, &key); |
2aaa6655 JB |
2361 | fi = btrfs_item_ptr(leaf, path->slots[0], |
2362 | struct btrfs_file_extent_item); | |
2363 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end - | |
2364 | offset; | |
2365 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2366 | btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes); | |
2367 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
2368 | btrfs_mark_buffer_dirty(leaf); | |
2369 | goto out; | |
2370 | } | |
2371 | btrfs_release_path(path); | |
2372 | ||
a012a74e | 2373 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), |
f85b7379 | 2374 | offset, 0, 0, end - offset, 0, end - offset, 0, 0, 0); |
2aaa6655 JB |
2375 | if (ret) |
2376 | return ret; | |
2377 | ||
2378 | out: | |
2379 | btrfs_release_path(path); | |
2380 | ||
2381 | hole_em = alloc_extent_map(); | |
2382 | if (!hole_em) { | |
2383 | btrfs_drop_extent_cache(inode, offset, end - 1, 0); | |
a012a74e | 2384 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags); |
2aaa6655 JB |
2385 | } else { |
2386 | hole_em->start = offset; | |
2387 | hole_em->len = end - offset; | |
cc95bef6 | 2388 | hole_em->ram_bytes = hole_em->len; |
2aaa6655 JB |
2389 | hole_em->orig_start = offset; |
2390 | ||
2391 | hole_em->block_start = EXTENT_MAP_HOLE; | |
2392 | hole_em->block_len = 0; | |
b4939680 | 2393 | hole_em->orig_block_len = 0; |
0b246afa | 2394 | hole_em->bdev = fs_info->fs_devices->latest_bdev; |
2aaa6655 JB |
2395 | hole_em->compress_type = BTRFS_COMPRESS_NONE; |
2396 | hole_em->generation = trans->transid; | |
2397 | ||
2398 | do { | |
2399 | btrfs_drop_extent_cache(inode, offset, end - 1, 0); | |
2400 | write_lock(&em_tree->lock); | |
09a2a8f9 | 2401 | ret = add_extent_mapping(em_tree, hole_em, 1); |
2aaa6655 JB |
2402 | write_unlock(&em_tree->lock); |
2403 | } while (ret == -EEXIST); | |
2404 | free_extent_map(hole_em); | |
2405 | if (ret) | |
2406 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
a012a74e | 2407 | &inode->runtime_flags); |
2aaa6655 JB |
2408 | } |
2409 | ||
2410 | return 0; | |
2411 | } | |
2412 | ||
d7781546 QW |
2413 | /* |
2414 | * Find a hole extent on given inode and change start/len to the end of hole | |
2415 | * extent.(hole/vacuum extent whose em->start <= start && | |
2416 | * em->start + em->len > start) | |
2417 | * When a hole extent is found, return 1 and modify start/len. | |
2418 | */ | |
2419 | static int find_first_non_hole(struct inode *inode, u64 *start, u64 *len) | |
2420 | { | |
609805d8 | 2421 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
d7781546 QW |
2422 | struct extent_map *em; |
2423 | int ret = 0; | |
2424 | ||
609805d8 FM |
2425 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, |
2426 | round_down(*start, fs_info->sectorsize), | |
2427 | round_up(*len, fs_info->sectorsize), 0); | |
9986277e DC |
2428 | if (IS_ERR(em)) |
2429 | return PTR_ERR(em); | |
d7781546 QW |
2430 | |
2431 | /* Hole or vacuum extent(only exists in no-hole mode) */ | |
2432 | if (em->block_start == EXTENT_MAP_HOLE) { | |
2433 | ret = 1; | |
2434 | *len = em->start + em->len > *start + *len ? | |
2435 | 0 : *start + *len - em->start - em->len; | |
2436 | *start = em->start + em->len; | |
2437 | } | |
2438 | free_extent_map(em); | |
2439 | return ret; | |
2440 | } | |
2441 | ||
f27451f2 FM |
2442 | static int btrfs_punch_hole_lock_range(struct inode *inode, |
2443 | const u64 lockstart, | |
2444 | const u64 lockend, | |
2445 | struct extent_state **cached_state) | |
2446 | { | |
2447 | while (1) { | |
2448 | struct btrfs_ordered_extent *ordered; | |
2449 | int ret; | |
2450 | ||
2451 | truncate_pagecache_range(inode, lockstart, lockend); | |
2452 | ||
2453 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
2454 | cached_state); | |
2455 | ordered = btrfs_lookup_first_ordered_extent(inode, lockend); | |
2456 | ||
2457 | /* | |
2458 | * We need to make sure we have no ordered extents in this range | |
2459 | * and nobody raced in and read a page in this range, if we did | |
2460 | * we need to try again. | |
2461 | */ | |
2462 | if ((!ordered || | |
2463 | (ordered->file_offset + ordered->len <= lockstart || | |
2464 | ordered->file_offset > lockend)) && | |
051c98eb DS |
2465 | !filemap_range_has_page(inode->i_mapping, |
2466 | lockstart, lockend)) { | |
f27451f2 FM |
2467 | if (ordered) |
2468 | btrfs_put_ordered_extent(ordered); | |
2469 | break; | |
2470 | } | |
2471 | if (ordered) | |
2472 | btrfs_put_ordered_extent(ordered); | |
2473 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, | |
2474 | lockend, cached_state); | |
2475 | ret = btrfs_wait_ordered_range(inode, lockstart, | |
2476 | lockend - lockstart + 1); | |
2477 | if (ret) | |
2478 | return ret; | |
2479 | } | |
2480 | return 0; | |
2481 | } | |
2482 | ||
2aaa6655 JB |
2483 | static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len) |
2484 | { | |
0b246afa | 2485 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2aaa6655 JB |
2486 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2487 | struct extent_state *cached_state = NULL; | |
2488 | struct btrfs_path *path; | |
2489 | struct btrfs_block_rsv *rsv; | |
2490 | struct btrfs_trans_handle *trans; | |
d7781546 QW |
2491 | u64 lockstart; |
2492 | u64 lockend; | |
2493 | u64 tail_start; | |
2494 | u64 tail_len; | |
2495 | u64 orig_start = offset; | |
2496 | u64 cur_offset; | |
5f52a2c5 | 2497 | u64 min_size = btrfs_calc_trans_metadata_size(fs_info, 1); |
2aaa6655 | 2498 | u64 drop_end; |
2aaa6655 JB |
2499 | int ret = 0; |
2500 | int err = 0; | |
6e4d6fa1 | 2501 | unsigned int rsv_count; |
9703fefe | 2502 | bool same_block; |
0b246afa | 2503 | bool no_holes = btrfs_fs_incompat(fs_info, NO_HOLES); |
a1a50f60 | 2504 | u64 ino_size; |
9703fefe | 2505 | bool truncated_block = false; |
e8c1c76e | 2506 | bool updated_inode = false; |
2aaa6655 | 2507 | |
0ef8b726 JB |
2508 | ret = btrfs_wait_ordered_range(inode, offset, len); |
2509 | if (ret) | |
2510 | return ret; | |
2aaa6655 | 2511 | |
5955102c | 2512 | inode_lock(inode); |
0b246afa | 2513 | ino_size = round_up(inode->i_size, fs_info->sectorsize); |
d7781546 QW |
2514 | ret = find_first_non_hole(inode, &offset, &len); |
2515 | if (ret < 0) | |
2516 | goto out_only_mutex; | |
2517 | if (ret && !len) { | |
2518 | /* Already in a large hole */ | |
2519 | ret = 0; | |
2520 | goto out_only_mutex; | |
2521 | } | |
2522 | ||
da17066c | 2523 | lockstart = round_up(offset, btrfs_inode_sectorsize(inode)); |
d7781546 | 2524 | lockend = round_down(offset + len, |
da17066c | 2525 | btrfs_inode_sectorsize(inode)) - 1; |
0b246afa JM |
2526 | same_block = (BTRFS_BYTES_TO_BLKS(fs_info, offset)) |
2527 | == (BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)); | |
7426cc04 | 2528 | /* |
9703fefe | 2529 | * We needn't truncate any block which is beyond the end of the file |
7426cc04 MX |
2530 | * because we are sure there is no data there. |
2531 | */ | |
2aaa6655 | 2532 | /* |
9703fefe CR |
2533 | * Only do this if we are in the same block and we aren't doing the |
2534 | * entire block. | |
2aaa6655 | 2535 | */ |
0b246afa | 2536 | if (same_block && len < fs_info->sectorsize) { |
e8c1c76e | 2537 | if (offset < ino_size) { |
9703fefe CR |
2538 | truncated_block = true; |
2539 | ret = btrfs_truncate_block(inode, offset, len, 0); | |
e8c1c76e FM |
2540 | } else { |
2541 | ret = 0; | |
2542 | } | |
d7781546 | 2543 | goto out_only_mutex; |
2aaa6655 JB |
2544 | } |
2545 | ||
9703fefe | 2546 | /* zero back part of the first block */ |
12870f1c | 2547 | if (offset < ino_size) { |
9703fefe CR |
2548 | truncated_block = true; |
2549 | ret = btrfs_truncate_block(inode, offset, 0, 0); | |
7426cc04 | 2550 | if (ret) { |
5955102c | 2551 | inode_unlock(inode); |
7426cc04 MX |
2552 | return ret; |
2553 | } | |
2aaa6655 JB |
2554 | } |
2555 | ||
d7781546 QW |
2556 | /* Check the aligned pages after the first unaligned page, |
2557 | * if offset != orig_start, which means the first unaligned page | |
01327610 | 2558 | * including several following pages are already in holes, |
d7781546 QW |
2559 | * the extra check can be skipped */ |
2560 | if (offset == orig_start) { | |
2561 | /* after truncate page, check hole again */ | |
2562 | len = offset + len - lockstart; | |
2563 | offset = lockstart; | |
2564 | ret = find_first_non_hole(inode, &offset, &len); | |
2565 | if (ret < 0) | |
2566 | goto out_only_mutex; | |
2567 | if (ret && !len) { | |
2568 | ret = 0; | |
2569 | goto out_only_mutex; | |
2570 | } | |
2571 | lockstart = offset; | |
2572 | } | |
2573 | ||
2574 | /* Check the tail unaligned part is in a hole */ | |
2575 | tail_start = lockend + 1; | |
2576 | tail_len = offset + len - tail_start; | |
2577 | if (tail_len) { | |
2578 | ret = find_first_non_hole(inode, &tail_start, &tail_len); | |
2579 | if (unlikely(ret < 0)) | |
2580 | goto out_only_mutex; | |
2581 | if (!ret) { | |
2582 | /* zero the front end of the last page */ | |
2583 | if (tail_start + tail_len < ino_size) { | |
9703fefe CR |
2584 | truncated_block = true; |
2585 | ret = btrfs_truncate_block(inode, | |
2586 | tail_start + tail_len, | |
2587 | 0, 1); | |
d7781546 QW |
2588 | if (ret) |
2589 | goto out_only_mutex; | |
51f395ad | 2590 | } |
0061280d | 2591 | } |
2aaa6655 JB |
2592 | } |
2593 | ||
2594 | if (lockend < lockstart) { | |
e8c1c76e FM |
2595 | ret = 0; |
2596 | goto out_only_mutex; | |
2aaa6655 JB |
2597 | } |
2598 | ||
f27451f2 FM |
2599 | ret = btrfs_punch_hole_lock_range(inode, lockstart, lockend, |
2600 | &cached_state); | |
2601 | if (ret) { | |
2602 | inode_unlock(inode); | |
2603 | goto out_only_mutex; | |
2aaa6655 JB |
2604 | } |
2605 | ||
2606 | path = btrfs_alloc_path(); | |
2607 | if (!path) { | |
2608 | ret = -ENOMEM; | |
2609 | goto out; | |
2610 | } | |
2611 | ||
2ff7e61e | 2612 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
2aaa6655 JB |
2613 | if (!rsv) { |
2614 | ret = -ENOMEM; | |
2615 | goto out_free; | |
2616 | } | |
5f52a2c5 | 2617 | rsv->size = btrfs_calc_trans_metadata_size(fs_info, 1); |
2aaa6655 JB |
2618 | rsv->failfast = 1; |
2619 | ||
2620 | /* | |
2621 | * 1 - update the inode | |
2622 | * 1 - removing the extents in the range | |
16e7549f | 2623 | * 1 - adding the hole extent if no_holes isn't set |
2aaa6655 | 2624 | */ |
16e7549f JB |
2625 | rsv_count = no_holes ? 2 : 3; |
2626 | trans = btrfs_start_transaction(root, rsv_count); | |
2aaa6655 JB |
2627 | if (IS_ERR(trans)) { |
2628 | err = PTR_ERR(trans); | |
2629 | goto out_free; | |
2630 | } | |
2631 | ||
0b246afa | 2632 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv, |
25d609f8 | 2633 | min_size, 0); |
2aaa6655 JB |
2634 | BUG_ON(ret); |
2635 | trans->block_rsv = rsv; | |
2636 | ||
d7781546 QW |
2637 | cur_offset = lockstart; |
2638 | len = lockend - cur_offset; | |
2aaa6655 JB |
2639 | while (cur_offset < lockend) { |
2640 | ret = __btrfs_drop_extents(trans, root, inode, path, | |
2641 | cur_offset, lockend + 1, | |
1acae57b | 2642 | &drop_end, 1, 0, 0, NULL); |
2aaa6655 JB |
2643 | if (ret != -ENOSPC) |
2644 | break; | |
2645 | ||
0b246afa | 2646 | trans->block_rsv = &fs_info->trans_block_rsv; |
2aaa6655 | 2647 | |
62fe51c1 | 2648 | if (cur_offset < drop_end && cur_offset < ino_size) { |
a012a74e NB |
2649 | ret = fill_holes(trans, BTRFS_I(inode), path, |
2650 | cur_offset, drop_end); | |
12870f1c | 2651 | if (ret) { |
f94480bd JB |
2652 | /* |
2653 | * If we failed then we didn't insert our hole | |
2654 | * entries for the area we dropped, so now the | |
2655 | * fs is corrupted, so we must abort the | |
2656 | * transaction. | |
2657 | */ | |
2658 | btrfs_abort_transaction(trans, ret); | |
12870f1c FM |
2659 | err = ret; |
2660 | break; | |
2661 | } | |
2aaa6655 JB |
2662 | } |
2663 | ||
2664 | cur_offset = drop_end; | |
2665 | ||
2666 | ret = btrfs_update_inode(trans, root, inode); | |
2667 | if (ret) { | |
2668 | err = ret; | |
2669 | break; | |
2670 | } | |
2671 | ||
3a45bb20 | 2672 | btrfs_end_transaction(trans); |
2ff7e61e | 2673 | btrfs_btree_balance_dirty(fs_info); |
2aaa6655 | 2674 | |
16e7549f | 2675 | trans = btrfs_start_transaction(root, rsv_count); |
2aaa6655 JB |
2676 | if (IS_ERR(trans)) { |
2677 | ret = PTR_ERR(trans); | |
2678 | trans = NULL; | |
2679 | break; | |
2680 | } | |
2681 | ||
0b246afa | 2682 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, |
25d609f8 | 2683 | rsv, min_size, 0); |
2aaa6655 JB |
2684 | BUG_ON(ret); /* shouldn't happen */ |
2685 | trans->block_rsv = rsv; | |
d7781546 QW |
2686 | |
2687 | ret = find_first_non_hole(inode, &cur_offset, &len); | |
2688 | if (unlikely(ret < 0)) | |
2689 | break; | |
2690 | if (ret && !len) { | |
2691 | ret = 0; | |
2692 | break; | |
2693 | } | |
2aaa6655 JB |
2694 | } |
2695 | ||
2696 | if (ret) { | |
2697 | err = ret; | |
2698 | goto out_trans; | |
2699 | } | |
2700 | ||
0b246afa | 2701 | trans->block_rsv = &fs_info->trans_block_rsv; |
2959a32a FM |
2702 | /* |
2703 | * If we are using the NO_HOLES feature we might have had already an | |
2704 | * hole that overlaps a part of the region [lockstart, lockend] and | |
2705 | * ends at (or beyond) lockend. Since we have no file extent items to | |
2706 | * represent holes, drop_end can be less than lockend and so we must | |
2707 | * make sure we have an extent map representing the existing hole (the | |
2708 | * call to __btrfs_drop_extents() might have dropped the existing extent | |
2709 | * map representing the existing hole), otherwise the fast fsync path | |
2710 | * will not record the existence of the hole region | |
2711 | * [existing_hole_start, lockend]. | |
2712 | */ | |
2713 | if (drop_end <= lockend) | |
2714 | drop_end = lockend + 1; | |
fc19c5e7 FM |
2715 | /* |
2716 | * Don't insert file hole extent item if it's for a range beyond eof | |
2717 | * (because it's useless) or if it represents a 0 bytes range (when | |
2718 | * cur_offset == drop_end). | |
2719 | */ | |
2720 | if (cur_offset < ino_size && cur_offset < drop_end) { | |
a012a74e NB |
2721 | ret = fill_holes(trans, BTRFS_I(inode), path, |
2722 | cur_offset, drop_end); | |
12870f1c | 2723 | if (ret) { |
f94480bd JB |
2724 | /* Same comment as above. */ |
2725 | btrfs_abort_transaction(trans, ret); | |
12870f1c FM |
2726 | err = ret; |
2727 | goto out_trans; | |
2728 | } | |
2aaa6655 JB |
2729 | } |
2730 | ||
2731 | out_trans: | |
2732 | if (!trans) | |
2733 | goto out_free; | |
2734 | ||
e1f5790e | 2735 | inode_inc_iversion(inode); |
c2050a45 | 2736 | inode->i_mtime = inode->i_ctime = current_time(inode); |
e1f5790e | 2737 | |
0b246afa | 2738 | trans->block_rsv = &fs_info->trans_block_rsv; |
2aaa6655 | 2739 | ret = btrfs_update_inode(trans, root, inode); |
e8c1c76e | 2740 | updated_inode = true; |
3a45bb20 | 2741 | btrfs_end_transaction(trans); |
2ff7e61e | 2742 | btrfs_btree_balance_dirty(fs_info); |
2aaa6655 JB |
2743 | out_free: |
2744 | btrfs_free_path(path); | |
2ff7e61e | 2745 | btrfs_free_block_rsv(fs_info, rsv); |
2aaa6655 JB |
2746 | out: |
2747 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
e43bbe5e | 2748 | &cached_state); |
d7781546 | 2749 | out_only_mutex: |
9703fefe | 2750 | if (!updated_inode && truncated_block && !ret && !err) { |
e8c1c76e FM |
2751 | /* |
2752 | * If we only end up zeroing part of a page, we still need to | |
2753 | * update the inode item, so that all the time fields are | |
2754 | * updated as well as the necessary btrfs inode in memory fields | |
2755 | * for detecting, at fsync time, if the inode isn't yet in the | |
2756 | * log tree or it's there but not up to date. | |
2757 | */ | |
2758 | trans = btrfs_start_transaction(root, 1); | |
2759 | if (IS_ERR(trans)) { | |
2760 | err = PTR_ERR(trans); | |
2761 | } else { | |
2762 | err = btrfs_update_inode(trans, root, inode); | |
3a45bb20 | 2763 | ret = btrfs_end_transaction(trans); |
e8c1c76e FM |
2764 | } |
2765 | } | |
5955102c | 2766 | inode_unlock(inode); |
2aaa6655 JB |
2767 | if (ret && !err) |
2768 | err = ret; | |
2769 | return err; | |
2770 | } | |
2771 | ||
14524a84 QW |
2772 | /* Helper structure to record which range is already reserved */ |
2773 | struct falloc_range { | |
2774 | struct list_head list; | |
2775 | u64 start; | |
2776 | u64 len; | |
2777 | }; | |
2778 | ||
2779 | /* | |
2780 | * Helper function to add falloc range | |
2781 | * | |
2782 | * Caller should have locked the larger range of extent containing | |
2783 | * [start, len) | |
2784 | */ | |
2785 | static int add_falloc_range(struct list_head *head, u64 start, u64 len) | |
2786 | { | |
2787 | struct falloc_range *prev = NULL; | |
2788 | struct falloc_range *range = NULL; | |
2789 | ||
2790 | if (list_empty(head)) | |
2791 | goto insert; | |
2792 | ||
2793 | /* | |
2794 | * As fallocate iterate by bytenr order, we only need to check | |
2795 | * the last range. | |
2796 | */ | |
2797 | prev = list_entry(head->prev, struct falloc_range, list); | |
2798 | if (prev->start + prev->len == start) { | |
2799 | prev->len += len; | |
2800 | return 0; | |
2801 | } | |
2802 | insert: | |
32fc932e | 2803 | range = kmalloc(sizeof(*range), GFP_KERNEL); |
14524a84 QW |
2804 | if (!range) |
2805 | return -ENOMEM; | |
2806 | range->start = start; | |
2807 | range->len = len; | |
2808 | list_add_tail(&range->list, head); | |
2809 | return 0; | |
2810 | } | |
2811 | ||
f27451f2 FM |
2812 | static int btrfs_fallocate_update_isize(struct inode *inode, |
2813 | const u64 end, | |
2814 | const int mode) | |
2815 | { | |
2816 | struct btrfs_trans_handle *trans; | |
2817 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2818 | int ret; | |
2819 | int ret2; | |
2820 | ||
2821 | if (mode & FALLOC_FL_KEEP_SIZE || end <= i_size_read(inode)) | |
2822 | return 0; | |
2823 | ||
2824 | trans = btrfs_start_transaction(root, 1); | |
2825 | if (IS_ERR(trans)) | |
2826 | return PTR_ERR(trans); | |
2827 | ||
2828 | inode->i_ctime = current_time(inode); | |
2829 | i_size_write(inode, end); | |
2830 | btrfs_ordered_update_i_size(inode, end, NULL); | |
2831 | ret = btrfs_update_inode(trans, root, inode); | |
2832 | ret2 = btrfs_end_transaction(trans); | |
2833 | ||
2834 | return ret ? ret : ret2; | |
2835 | } | |
2836 | ||
81fdf638 FM |
2837 | enum { |
2838 | RANGE_BOUNDARY_WRITTEN_EXTENT = 0, | |
2839 | RANGE_BOUNDARY_PREALLOC_EXTENT = 1, | |
2840 | RANGE_BOUNDARY_HOLE = 2, | |
2841 | }; | |
2842 | ||
f27451f2 FM |
2843 | static int btrfs_zero_range_check_range_boundary(struct inode *inode, |
2844 | u64 offset) | |
2845 | { | |
2846 | const u64 sectorsize = btrfs_inode_sectorsize(inode); | |
2847 | struct extent_map *em; | |
81fdf638 | 2848 | int ret; |
f27451f2 FM |
2849 | |
2850 | offset = round_down(offset, sectorsize); | |
2851 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, offset, sectorsize, 0); | |
2852 | if (IS_ERR(em)) | |
2853 | return PTR_ERR(em); | |
2854 | ||
2855 | if (em->block_start == EXTENT_MAP_HOLE) | |
81fdf638 FM |
2856 | ret = RANGE_BOUNDARY_HOLE; |
2857 | else if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
2858 | ret = RANGE_BOUNDARY_PREALLOC_EXTENT; | |
2859 | else | |
2860 | ret = RANGE_BOUNDARY_WRITTEN_EXTENT; | |
f27451f2 FM |
2861 | |
2862 | free_extent_map(em); | |
2863 | return ret; | |
2864 | } | |
2865 | ||
2866 | static int btrfs_zero_range(struct inode *inode, | |
2867 | loff_t offset, | |
2868 | loff_t len, | |
2869 | const int mode) | |
2870 | { | |
2871 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2872 | struct extent_map *em; | |
2873 | struct extent_changeset *data_reserved = NULL; | |
2874 | int ret; | |
2875 | u64 alloc_hint = 0; | |
2876 | const u64 sectorsize = btrfs_inode_sectorsize(inode); | |
2877 | u64 alloc_start = round_down(offset, sectorsize); | |
2878 | u64 alloc_end = round_up(offset + len, sectorsize); | |
2879 | u64 bytes_to_reserve = 0; | |
2880 | bool space_reserved = false; | |
2881 | ||
2882 | inode_dio_wait(inode); | |
2883 | ||
2884 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, | |
2885 | alloc_start, alloc_end - alloc_start, 0); | |
2886 | if (IS_ERR(em)) { | |
2887 | ret = PTR_ERR(em); | |
2888 | goto out; | |
2889 | } | |
2890 | ||
2891 | /* | |
2892 | * Avoid hole punching and extent allocation for some cases. More cases | |
2893 | * could be considered, but these are unlikely common and we keep things | |
2894 | * as simple as possible for now. Also, intentionally, if the target | |
2895 | * range contains one or more prealloc extents together with regular | |
2896 | * extents and holes, we drop all the existing extents and allocate a | |
2897 | * new prealloc extent, so that we get a larger contiguous disk extent. | |
2898 | */ | |
2899 | if (em->start <= alloc_start && | |
2900 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { | |
2901 | const u64 em_end = em->start + em->len; | |
2902 | ||
2903 | if (em_end >= offset + len) { | |
2904 | /* | |
2905 | * The whole range is already a prealloc extent, | |
2906 | * do nothing except updating the inode's i_size if | |
2907 | * needed. | |
2908 | */ | |
2909 | free_extent_map(em); | |
2910 | ret = btrfs_fallocate_update_isize(inode, offset + len, | |
2911 | mode); | |
2912 | goto out; | |
2913 | } | |
2914 | /* | |
2915 | * Part of the range is already a prealloc extent, so operate | |
2916 | * only on the remaining part of the range. | |
2917 | */ | |
2918 | alloc_start = em_end; | |
2919 | ASSERT(IS_ALIGNED(alloc_start, sectorsize)); | |
2920 | len = offset + len - alloc_start; | |
2921 | offset = alloc_start; | |
2922 | alloc_hint = em->block_start + em->len; | |
2923 | } | |
2924 | free_extent_map(em); | |
2925 | ||
2926 | if (BTRFS_BYTES_TO_BLKS(fs_info, offset) == | |
2927 | BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)) { | |
2928 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, | |
2929 | alloc_start, sectorsize, 0); | |
2930 | if (IS_ERR(em)) { | |
2931 | ret = PTR_ERR(em); | |
2932 | goto out; | |
2933 | } | |
2934 | ||
2935 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { | |
2936 | free_extent_map(em); | |
2937 | ret = btrfs_fallocate_update_isize(inode, offset + len, | |
2938 | mode); | |
2939 | goto out; | |
2940 | } | |
2941 | if (len < sectorsize && em->block_start != EXTENT_MAP_HOLE) { | |
2942 | free_extent_map(em); | |
2943 | ret = btrfs_truncate_block(inode, offset, len, 0); | |
2944 | if (!ret) | |
2945 | ret = btrfs_fallocate_update_isize(inode, | |
2946 | offset + len, | |
2947 | mode); | |
2948 | return ret; | |
2949 | } | |
2950 | free_extent_map(em); | |
2951 | alloc_start = round_down(offset, sectorsize); | |
2952 | alloc_end = alloc_start + sectorsize; | |
2953 | goto reserve_space; | |
2954 | } | |
2955 | ||
2956 | alloc_start = round_up(offset, sectorsize); | |
2957 | alloc_end = round_down(offset + len, sectorsize); | |
2958 | ||
2959 | /* | |
2960 | * For unaligned ranges, check the pages at the boundaries, they might | |
2961 | * map to an extent, in which case we need to partially zero them, or | |
2962 | * they might map to a hole, in which case we need our allocation range | |
2963 | * to cover them. | |
2964 | */ | |
2965 | if (!IS_ALIGNED(offset, sectorsize)) { | |
2966 | ret = btrfs_zero_range_check_range_boundary(inode, offset); | |
2967 | if (ret < 0) | |
2968 | goto out; | |
81fdf638 | 2969 | if (ret == RANGE_BOUNDARY_HOLE) { |
f27451f2 FM |
2970 | alloc_start = round_down(offset, sectorsize); |
2971 | ret = 0; | |
81fdf638 | 2972 | } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) { |
f27451f2 FM |
2973 | ret = btrfs_truncate_block(inode, offset, 0, 0); |
2974 | if (ret) | |
2975 | goto out; | |
81fdf638 FM |
2976 | } else { |
2977 | ret = 0; | |
f27451f2 FM |
2978 | } |
2979 | } | |
2980 | ||
2981 | if (!IS_ALIGNED(offset + len, sectorsize)) { | |
2982 | ret = btrfs_zero_range_check_range_boundary(inode, | |
2983 | offset + len); | |
2984 | if (ret < 0) | |
2985 | goto out; | |
81fdf638 | 2986 | if (ret == RANGE_BOUNDARY_HOLE) { |
f27451f2 FM |
2987 | alloc_end = round_up(offset + len, sectorsize); |
2988 | ret = 0; | |
81fdf638 | 2989 | } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) { |
f27451f2 FM |
2990 | ret = btrfs_truncate_block(inode, offset + len, 0, 1); |
2991 | if (ret) | |
2992 | goto out; | |
81fdf638 FM |
2993 | } else { |
2994 | ret = 0; | |
f27451f2 FM |
2995 | } |
2996 | } | |
2997 | ||
2998 | reserve_space: | |
2999 | if (alloc_start < alloc_end) { | |
3000 | struct extent_state *cached_state = NULL; | |
3001 | const u64 lockstart = alloc_start; | |
3002 | const u64 lockend = alloc_end - 1; | |
3003 | ||
3004 | bytes_to_reserve = alloc_end - alloc_start; | |
3005 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), | |
3006 | bytes_to_reserve); | |
3007 | if (ret < 0) | |
3008 | goto out; | |
3009 | space_reserved = true; | |
3010 | ret = btrfs_qgroup_reserve_data(inode, &data_reserved, | |
3011 | alloc_start, bytes_to_reserve); | |
3012 | if (ret) | |
3013 | goto out; | |
3014 | ret = btrfs_punch_hole_lock_range(inode, lockstart, lockend, | |
3015 | &cached_state); | |
3016 | if (ret) | |
3017 | goto out; | |
3018 | ret = btrfs_prealloc_file_range(inode, mode, alloc_start, | |
3019 | alloc_end - alloc_start, | |
3020 | i_blocksize(inode), | |
3021 | offset + len, &alloc_hint); | |
3022 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, | |
3023 | lockend, &cached_state); | |
3024 | /* btrfs_prealloc_file_range releases reserved space on error */ | |
9f13ce74 | 3025 | if (ret) { |
f27451f2 | 3026 | space_reserved = false; |
9f13ce74 FM |
3027 | goto out; |
3028 | } | |
f27451f2 | 3029 | } |
9f13ce74 | 3030 | ret = btrfs_fallocate_update_isize(inode, offset + len, mode); |
f27451f2 FM |
3031 | out: |
3032 | if (ret && space_reserved) | |
3033 | btrfs_free_reserved_data_space(inode, data_reserved, | |
3034 | alloc_start, bytes_to_reserve); | |
3035 | extent_changeset_free(data_reserved); | |
3036 | ||
3037 | return ret; | |
3038 | } | |
3039 | ||
2fe17c10 CH |
3040 | static long btrfs_fallocate(struct file *file, int mode, |
3041 | loff_t offset, loff_t len) | |
3042 | { | |
496ad9aa | 3043 | struct inode *inode = file_inode(file); |
2fe17c10 | 3044 | struct extent_state *cached_state = NULL; |
364ecf36 | 3045 | struct extent_changeset *data_reserved = NULL; |
14524a84 QW |
3046 | struct falloc_range *range; |
3047 | struct falloc_range *tmp; | |
3048 | struct list_head reserve_list; | |
2fe17c10 CH |
3049 | u64 cur_offset; |
3050 | u64 last_byte; | |
3051 | u64 alloc_start; | |
3052 | u64 alloc_end; | |
3053 | u64 alloc_hint = 0; | |
3054 | u64 locked_end; | |
14524a84 | 3055 | u64 actual_end = 0; |
2fe17c10 | 3056 | struct extent_map *em; |
da17066c | 3057 | int blocksize = btrfs_inode_sectorsize(inode); |
2fe17c10 CH |
3058 | int ret; |
3059 | ||
797f4277 MX |
3060 | alloc_start = round_down(offset, blocksize); |
3061 | alloc_end = round_up(offset + len, blocksize); | |
18513091 | 3062 | cur_offset = alloc_start; |
2fe17c10 | 3063 | |
2aaa6655 | 3064 | /* Make sure we aren't being give some crap mode */ |
f27451f2 FM |
3065 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | |
3066 | FALLOC_FL_ZERO_RANGE)) | |
2fe17c10 CH |
3067 | return -EOPNOTSUPP; |
3068 | ||
2aaa6655 JB |
3069 | if (mode & FALLOC_FL_PUNCH_HOLE) |
3070 | return btrfs_punch_hole(inode, offset, len); | |
3071 | ||
d98456fc | 3072 | /* |
14524a84 QW |
3073 | * Only trigger disk allocation, don't trigger qgroup reserve |
3074 | * | |
3075 | * For qgroup space, it will be checked later. | |
d98456fc | 3076 | */ |
f27451f2 FM |
3077 | if (!(mode & FALLOC_FL_ZERO_RANGE)) { |
3078 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), | |
3079 | alloc_end - alloc_start); | |
3080 | if (ret < 0) | |
3081 | return ret; | |
3082 | } | |
d98456fc | 3083 | |
5955102c | 3084 | inode_lock(inode); |
2a162ce9 DI |
3085 | |
3086 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) { | |
3087 | ret = inode_newsize_ok(inode, offset + len); | |
3088 | if (ret) | |
3089 | goto out; | |
3090 | } | |
2fe17c10 | 3091 | |
14524a84 QW |
3092 | /* |
3093 | * TODO: Move these two operations after we have checked | |
3094 | * accurate reserved space, or fallocate can still fail but | |
3095 | * with page truncated or size expanded. | |
3096 | * | |
3097 | * But that's a minor problem and won't do much harm BTW. | |
3098 | */ | |
2fe17c10 | 3099 | if (alloc_start > inode->i_size) { |
a41ad394 JB |
3100 | ret = btrfs_cont_expand(inode, i_size_read(inode), |
3101 | alloc_start); | |
2fe17c10 CH |
3102 | if (ret) |
3103 | goto out; | |
0f6925fa | 3104 | } else if (offset + len > inode->i_size) { |
a71754fc JB |
3105 | /* |
3106 | * If we are fallocating from the end of the file onward we | |
9703fefe CR |
3107 | * need to zero out the end of the block if i_size lands in the |
3108 | * middle of a block. | |
a71754fc | 3109 | */ |
9703fefe | 3110 | ret = btrfs_truncate_block(inode, inode->i_size, 0, 0); |
a71754fc JB |
3111 | if (ret) |
3112 | goto out; | |
2fe17c10 CH |
3113 | } |
3114 | ||
a71754fc JB |
3115 | /* |
3116 | * wait for ordered IO before we have any locks. We'll loop again | |
3117 | * below with the locks held. | |
3118 | */ | |
0ef8b726 JB |
3119 | ret = btrfs_wait_ordered_range(inode, alloc_start, |
3120 | alloc_end - alloc_start); | |
3121 | if (ret) | |
3122 | goto out; | |
a71754fc | 3123 | |
f27451f2 FM |
3124 | if (mode & FALLOC_FL_ZERO_RANGE) { |
3125 | ret = btrfs_zero_range(inode, offset, len, mode); | |
3126 | inode_unlock(inode); | |
3127 | return ret; | |
3128 | } | |
3129 | ||
2fe17c10 CH |
3130 | locked_end = alloc_end - 1; |
3131 | while (1) { | |
3132 | struct btrfs_ordered_extent *ordered; | |
3133 | ||
3134 | /* the extent lock is ordered inside the running | |
3135 | * transaction | |
3136 | */ | |
3137 | lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start, | |
ff13db41 | 3138 | locked_end, &cached_state); |
96b09dde NB |
3139 | ordered = btrfs_lookup_first_ordered_extent(inode, locked_end); |
3140 | ||
2fe17c10 CH |
3141 | if (ordered && |
3142 | ordered->file_offset + ordered->len > alloc_start && | |
3143 | ordered->file_offset < alloc_end) { | |
3144 | btrfs_put_ordered_extent(ordered); | |
3145 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, | |
3146 | alloc_start, locked_end, | |
e43bbe5e | 3147 | &cached_state); |
2fe17c10 CH |
3148 | /* |
3149 | * we can't wait on the range with the transaction | |
3150 | * running or with the extent lock held | |
3151 | */ | |
0ef8b726 JB |
3152 | ret = btrfs_wait_ordered_range(inode, alloc_start, |
3153 | alloc_end - alloc_start); | |
3154 | if (ret) | |
3155 | goto out; | |
2fe17c10 CH |
3156 | } else { |
3157 | if (ordered) | |
3158 | btrfs_put_ordered_extent(ordered); | |
3159 | break; | |
3160 | } | |
3161 | } | |
3162 | ||
14524a84 QW |
3163 | /* First, check if we exceed the qgroup limit */ |
3164 | INIT_LIST_HEAD(&reserve_list); | |
6b7d6e93 | 3165 | while (cur_offset < alloc_end) { |
fc4f21b1 | 3166 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset, |
2fe17c10 | 3167 | alloc_end - cur_offset, 0); |
9986277e DC |
3168 | if (IS_ERR(em)) { |
3169 | ret = PTR_ERR(em); | |
79787eaa JM |
3170 | break; |
3171 | } | |
2fe17c10 | 3172 | last_byte = min(extent_map_end(em), alloc_end); |
f1e490a7 | 3173 | actual_end = min_t(u64, extent_map_end(em), offset + len); |
797f4277 | 3174 | last_byte = ALIGN(last_byte, blocksize); |
2fe17c10 CH |
3175 | if (em->block_start == EXTENT_MAP_HOLE || |
3176 | (cur_offset >= inode->i_size && | |
3177 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
14524a84 QW |
3178 | ret = add_falloc_range(&reserve_list, cur_offset, |
3179 | last_byte - cur_offset); | |
3180 | if (ret < 0) { | |
3181 | free_extent_map(em); | |
3182 | break; | |
3d850dd4 | 3183 | } |
364ecf36 QW |
3184 | ret = btrfs_qgroup_reserve_data(inode, &data_reserved, |
3185 | cur_offset, last_byte - cur_offset); | |
be2d253c FM |
3186 | if (ret < 0) { |
3187 | free_extent_map(em); | |
14524a84 | 3188 | break; |
be2d253c | 3189 | } |
18513091 WX |
3190 | } else { |
3191 | /* | |
3192 | * Do not need to reserve unwritten extent for this | |
3193 | * range, free reserved data space first, otherwise | |
3194 | * it'll result in false ENOSPC error. | |
3195 | */ | |
bc42bda2 QW |
3196 | btrfs_free_reserved_data_space(inode, data_reserved, |
3197 | cur_offset, last_byte - cur_offset); | |
2fe17c10 CH |
3198 | } |
3199 | free_extent_map(em); | |
2fe17c10 | 3200 | cur_offset = last_byte; |
14524a84 QW |
3201 | } |
3202 | ||
3203 | /* | |
3204 | * If ret is still 0, means we're OK to fallocate. | |
3205 | * Or just cleanup the list and exit. | |
3206 | */ | |
3207 | list_for_each_entry_safe(range, tmp, &reserve_list, list) { | |
3208 | if (!ret) | |
3209 | ret = btrfs_prealloc_file_range(inode, mode, | |
3210 | range->start, | |
93407472 | 3211 | range->len, i_blocksize(inode), |
14524a84 | 3212 | offset + len, &alloc_hint); |
18513091 | 3213 | else |
bc42bda2 QW |
3214 | btrfs_free_reserved_data_space(inode, |
3215 | data_reserved, range->start, | |
3216 | range->len); | |
14524a84 QW |
3217 | list_del(&range->list); |
3218 | kfree(range); | |
3219 | } | |
3220 | if (ret < 0) | |
3221 | goto out_unlock; | |
3222 | ||
f27451f2 FM |
3223 | /* |
3224 | * We didn't need to allocate any more space, but we still extended the | |
3225 | * size of the file so we need to update i_size and the inode item. | |
3226 | */ | |
3227 | ret = btrfs_fallocate_update_isize(inode, actual_end, mode); | |
14524a84 | 3228 | out_unlock: |
2fe17c10 | 3229 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
e43bbe5e | 3230 | &cached_state); |
2fe17c10 | 3231 | out: |
5955102c | 3232 | inode_unlock(inode); |
d98456fc | 3233 | /* Let go of our reservation. */ |
f27451f2 | 3234 | if (ret != 0 && !(mode & FALLOC_FL_ZERO_RANGE)) |
bc42bda2 QW |
3235 | btrfs_free_reserved_data_space(inode, data_reserved, |
3236 | alloc_start, alloc_end - cur_offset); | |
364ecf36 | 3237 | extent_changeset_free(data_reserved); |
2fe17c10 CH |
3238 | return ret; |
3239 | } | |
3240 | ||
965c8e59 | 3241 | static int find_desired_extent(struct inode *inode, loff_t *offset, int whence) |
b2675157 | 3242 | { |
0b246afa | 3243 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
7f4ca37c | 3244 | struct extent_map *em = NULL; |
b2675157 | 3245 | struct extent_state *cached_state = NULL; |
4d1a40c6 LB |
3246 | u64 lockstart; |
3247 | u64 lockend; | |
3248 | u64 start; | |
3249 | u64 len; | |
b2675157 JB |
3250 | int ret = 0; |
3251 | ||
4d1a40c6 LB |
3252 | if (inode->i_size == 0) |
3253 | return -ENXIO; | |
3254 | ||
3255 | /* | |
3256 | * *offset can be negative, in this case we start finding DATA/HOLE from | |
3257 | * the very start of the file. | |
3258 | */ | |
3259 | start = max_t(loff_t, 0, *offset); | |
3260 | ||
0b246afa | 3261 | lockstart = round_down(start, fs_info->sectorsize); |
da17066c | 3262 | lockend = round_up(i_size_read(inode), |
0b246afa | 3263 | fs_info->sectorsize); |
b2675157 | 3264 | if (lockend <= lockstart) |
0b246afa | 3265 | lockend = lockstart + fs_info->sectorsize; |
1214b53f | 3266 | lockend--; |
b2675157 JB |
3267 | len = lockend - lockstart + 1; |
3268 | ||
ff13db41 | 3269 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
d0082371 | 3270 | &cached_state); |
b2675157 | 3271 | |
7f4ca37c | 3272 | while (start < inode->i_size) { |
fc4f21b1 NB |
3273 | em = btrfs_get_extent_fiemap(BTRFS_I(inode), NULL, 0, |
3274 | start, len, 0); | |
b2675157 | 3275 | if (IS_ERR(em)) { |
6af021d8 | 3276 | ret = PTR_ERR(em); |
7f4ca37c | 3277 | em = NULL; |
b2675157 JB |
3278 | break; |
3279 | } | |
3280 | ||
7f4ca37c JB |
3281 | if (whence == SEEK_HOLE && |
3282 | (em->block_start == EXTENT_MAP_HOLE || | |
3283 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) | |
3284 | break; | |
3285 | else if (whence == SEEK_DATA && | |
3286 | (em->block_start != EXTENT_MAP_HOLE && | |
3287 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) | |
3288 | break; | |
b2675157 JB |
3289 | |
3290 | start = em->start + em->len; | |
b2675157 | 3291 | free_extent_map(em); |
7f4ca37c | 3292 | em = NULL; |
b2675157 JB |
3293 | cond_resched(); |
3294 | } | |
7f4ca37c JB |
3295 | free_extent_map(em); |
3296 | if (!ret) { | |
3297 | if (whence == SEEK_DATA && start >= inode->i_size) | |
3298 | ret = -ENXIO; | |
3299 | else | |
3300 | *offset = min_t(loff_t, start, inode->i_size); | |
3301 | } | |
b2675157 | 3302 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
e43bbe5e | 3303 | &cached_state); |
b2675157 JB |
3304 | return ret; |
3305 | } | |
3306 | ||
965c8e59 | 3307 | static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence) |
b2675157 JB |
3308 | { |
3309 | struct inode *inode = file->f_mapping->host; | |
3310 | int ret; | |
3311 | ||
5955102c | 3312 | inode_lock(inode); |
965c8e59 | 3313 | switch (whence) { |
b2675157 JB |
3314 | case SEEK_END: |
3315 | case SEEK_CUR: | |
965c8e59 | 3316 | offset = generic_file_llseek(file, offset, whence); |
b2675157 JB |
3317 | goto out; |
3318 | case SEEK_DATA: | |
3319 | case SEEK_HOLE: | |
48802c8a | 3320 | if (offset >= i_size_read(inode)) { |
5955102c | 3321 | inode_unlock(inode); |
48802c8a JL |
3322 | return -ENXIO; |
3323 | } | |
3324 | ||
965c8e59 | 3325 | ret = find_desired_extent(inode, &offset, whence); |
b2675157 | 3326 | if (ret) { |
5955102c | 3327 | inode_unlock(inode); |
b2675157 JB |
3328 | return ret; |
3329 | } | |
3330 | } | |
3331 | ||
46a1c2c7 | 3332 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
b2675157 | 3333 | out: |
5955102c | 3334 | inode_unlock(inode); |
b2675157 JB |
3335 | return offset; |
3336 | } | |
3337 | ||
edf064e7 GR |
3338 | static int btrfs_file_open(struct inode *inode, struct file *filp) |
3339 | { | |
91f9943e | 3340 | filp->f_mode |= FMODE_NOWAIT; |
edf064e7 GR |
3341 | return generic_file_open(inode, filp); |
3342 | } | |
3343 | ||
828c0950 | 3344 | const struct file_operations btrfs_file_operations = { |
b2675157 | 3345 | .llseek = btrfs_file_llseek, |
aad4f8bb | 3346 | .read_iter = generic_file_read_iter, |
e9906a98 | 3347 | .splice_read = generic_file_splice_read, |
b30ac0fc | 3348 | .write_iter = btrfs_file_write_iter, |
9ebefb18 | 3349 | .mmap = btrfs_file_mmap, |
edf064e7 | 3350 | .open = btrfs_file_open, |
e1b81e67 | 3351 | .release = btrfs_release_file, |
39279cc3 | 3352 | .fsync = btrfs_sync_file, |
2fe17c10 | 3353 | .fallocate = btrfs_fallocate, |
34287aa3 | 3354 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 3355 | #ifdef CONFIG_COMPAT |
4c63c245 | 3356 | .compat_ioctl = btrfs_compat_ioctl, |
39279cc3 | 3357 | #endif |
04b38d60 | 3358 | .clone_file_range = btrfs_clone_file_range, |
2b3909f8 | 3359 | .dedupe_file_range = btrfs_dedupe_file_range, |
39279cc3 | 3360 | }; |
9247f317 | 3361 | |
e67c718b | 3362 | void __cold btrfs_auto_defrag_exit(void) |
9247f317 | 3363 | { |
5598e900 | 3364 | kmem_cache_destroy(btrfs_inode_defrag_cachep); |
9247f317 MX |
3365 | } |
3366 | ||
f5c29bd9 | 3367 | int __init btrfs_auto_defrag_init(void) |
9247f317 MX |
3368 | { |
3369 | btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag", | |
3370 | sizeof(struct inode_defrag), 0, | |
fba4b697 | 3371 | SLAB_MEM_SPREAD, |
9247f317 MX |
3372 | NULL); |
3373 | if (!btrfs_inode_defrag_cachep) | |
3374 | return -ENOMEM; | |
3375 | ||
3376 | return 0; | |
3377 | } | |
728404da FM |
3378 | |
3379 | int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end) | |
3380 | { | |
3381 | int ret; | |
3382 | ||
3383 | /* | |
3384 | * So with compression we will find and lock a dirty page and clear the | |
3385 | * first one as dirty, setup an async extent, and immediately return | |
3386 | * with the entire range locked but with nobody actually marked with | |
3387 | * writeback. So we can't just filemap_write_and_wait_range() and | |
3388 | * expect it to work since it will just kick off a thread to do the | |
3389 | * actual work. So we need to call filemap_fdatawrite_range _again_ | |
3390 | * since it will wait on the page lock, which won't be unlocked until | |
3391 | * after the pages have been marked as writeback and so we're good to go | |
3392 | * from there. We have to do this otherwise we'll miss the ordered | |
3393 | * extents and that results in badness. Please Josef, do not think you | |
3394 | * know better and pull this out at some point in the future, it is | |
3395 | * right and you are wrong. | |
3396 | */ | |
3397 | ret = filemap_fdatawrite_range(inode->i_mapping, start, end); | |
3398 | if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
3399 | &BTRFS_I(inode)->runtime_flags)) | |
3400 | ret = filemap_fdatawrite_range(inode->i_mapping, start, end); | |
3401 | ||
3402 | return ret; | |
3403 | } |