Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
39279cc3 CM |
19 | #include <linux/fs.h> |
20 | #include <linux/pagemap.h> | |
21 | #include <linux/highmem.h> | |
22 | #include <linux/time.h> | |
23 | #include <linux/init.h> | |
24 | #include <linux/string.h> | |
39279cc3 CM |
25 | #include <linux/backing-dev.h> |
26 | #include <linux/mpage.h> | |
2fe17c10 | 27 | #include <linux/falloc.h> |
39279cc3 CM |
28 | #include <linux/swap.h> |
29 | #include <linux/writeback.h> | |
39279cc3 | 30 | #include <linux/compat.h> |
5a0e3ad6 | 31 | #include <linux/slab.h> |
55e301fd | 32 | #include <linux/btrfs.h> |
e2e40f2c | 33 | #include <linux/uio.h> |
39279cc3 CM |
34 | #include "ctree.h" |
35 | #include "disk-io.h" | |
36 | #include "transaction.h" | |
37 | #include "btrfs_inode.h" | |
39279cc3 | 38 | #include "print-tree.h" |
e02119d5 CM |
39 | #include "tree-log.h" |
40 | #include "locking.h" | |
2aaa6655 | 41 | #include "volumes.h" |
fcebe456 | 42 | #include "qgroup.h" |
ebb8765b | 43 | #include "compression.h" |
39279cc3 | 44 | |
9247f317 | 45 | static struct kmem_cache *btrfs_inode_defrag_cachep; |
4cb5300b CM |
46 | /* |
47 | * when auto defrag is enabled we | |
48 | * queue up these defrag structs to remember which | |
49 | * inodes need defragging passes | |
50 | */ | |
51 | struct inode_defrag { | |
52 | struct rb_node rb_node; | |
53 | /* objectid */ | |
54 | u64 ino; | |
55 | /* | |
56 | * transid where the defrag was added, we search for | |
57 | * extents newer than this | |
58 | */ | |
59 | u64 transid; | |
60 | ||
61 | /* root objectid */ | |
62 | u64 root; | |
63 | ||
64 | /* last offset we were able to defrag */ | |
65 | u64 last_offset; | |
66 | ||
67 | /* if we've wrapped around back to zero once already */ | |
68 | int cycled; | |
69 | }; | |
70 | ||
762f2263 MX |
71 | static int __compare_inode_defrag(struct inode_defrag *defrag1, |
72 | struct inode_defrag *defrag2) | |
73 | { | |
74 | if (defrag1->root > defrag2->root) | |
75 | return 1; | |
76 | else if (defrag1->root < defrag2->root) | |
77 | return -1; | |
78 | else if (defrag1->ino > defrag2->ino) | |
79 | return 1; | |
80 | else if (defrag1->ino < defrag2->ino) | |
81 | return -1; | |
82 | else | |
83 | return 0; | |
84 | } | |
85 | ||
4cb5300b CM |
86 | /* pop a record for an inode into the defrag tree. The lock |
87 | * must be held already | |
88 | * | |
89 | * If you're inserting a record for an older transid than an | |
90 | * existing record, the transid already in the tree is lowered | |
91 | * | |
92 | * If an existing record is found the defrag item you | |
93 | * pass in is freed | |
94 | */ | |
6158e1ce | 95 | static int __btrfs_add_inode_defrag(struct btrfs_inode *inode, |
4cb5300b CM |
96 | struct inode_defrag *defrag) |
97 | { | |
6158e1ce | 98 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
4cb5300b CM |
99 | struct inode_defrag *entry; |
100 | struct rb_node **p; | |
101 | struct rb_node *parent = NULL; | |
762f2263 | 102 | int ret; |
4cb5300b | 103 | |
0b246afa | 104 | p = &fs_info->defrag_inodes.rb_node; |
4cb5300b CM |
105 | while (*p) { |
106 | parent = *p; | |
107 | entry = rb_entry(parent, struct inode_defrag, rb_node); | |
108 | ||
762f2263 MX |
109 | ret = __compare_inode_defrag(defrag, entry); |
110 | if (ret < 0) | |
4cb5300b | 111 | p = &parent->rb_left; |
762f2263 | 112 | else if (ret > 0) |
4cb5300b CM |
113 | p = &parent->rb_right; |
114 | else { | |
115 | /* if we're reinserting an entry for | |
116 | * an old defrag run, make sure to | |
117 | * lower the transid of our existing record | |
118 | */ | |
119 | if (defrag->transid < entry->transid) | |
120 | entry->transid = defrag->transid; | |
121 | if (defrag->last_offset > entry->last_offset) | |
122 | entry->last_offset = defrag->last_offset; | |
8ddc4734 | 123 | return -EEXIST; |
4cb5300b CM |
124 | } |
125 | } | |
6158e1ce | 126 | set_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags); |
4cb5300b | 127 | rb_link_node(&defrag->rb_node, parent, p); |
0b246afa | 128 | rb_insert_color(&defrag->rb_node, &fs_info->defrag_inodes); |
8ddc4734 MX |
129 | return 0; |
130 | } | |
4cb5300b | 131 | |
2ff7e61e | 132 | static inline int __need_auto_defrag(struct btrfs_fs_info *fs_info) |
8ddc4734 | 133 | { |
0b246afa | 134 | if (!btrfs_test_opt(fs_info, AUTO_DEFRAG)) |
8ddc4734 MX |
135 | return 0; |
136 | ||
0b246afa | 137 | if (btrfs_fs_closing(fs_info)) |
8ddc4734 | 138 | return 0; |
4cb5300b | 139 | |
8ddc4734 | 140 | return 1; |
4cb5300b CM |
141 | } |
142 | ||
143 | /* | |
144 | * insert a defrag record for this inode if auto defrag is | |
145 | * enabled | |
146 | */ | |
147 | int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans, | |
6158e1ce | 148 | struct btrfs_inode *inode) |
4cb5300b | 149 | { |
6158e1ce NB |
150 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
151 | struct btrfs_root *root = inode->root; | |
4cb5300b | 152 | struct inode_defrag *defrag; |
4cb5300b | 153 | u64 transid; |
8ddc4734 | 154 | int ret; |
4cb5300b | 155 | |
2ff7e61e | 156 | if (!__need_auto_defrag(fs_info)) |
4cb5300b CM |
157 | return 0; |
158 | ||
6158e1ce | 159 | if (test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags)) |
4cb5300b CM |
160 | return 0; |
161 | ||
162 | if (trans) | |
163 | transid = trans->transid; | |
164 | else | |
6158e1ce | 165 | transid = inode->root->last_trans; |
4cb5300b | 166 | |
9247f317 | 167 | defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS); |
4cb5300b CM |
168 | if (!defrag) |
169 | return -ENOMEM; | |
170 | ||
6158e1ce | 171 | defrag->ino = btrfs_ino(inode); |
4cb5300b CM |
172 | defrag->transid = transid; |
173 | defrag->root = root->root_key.objectid; | |
174 | ||
0b246afa | 175 | spin_lock(&fs_info->defrag_inodes_lock); |
6158e1ce | 176 | if (!test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags)) { |
8ddc4734 MX |
177 | /* |
178 | * If we set IN_DEFRAG flag and evict the inode from memory, | |
179 | * and then re-read this inode, this new inode doesn't have | |
180 | * IN_DEFRAG flag. At the case, we may find the existed defrag. | |
181 | */ | |
182 | ret = __btrfs_add_inode_defrag(inode, defrag); | |
183 | if (ret) | |
184 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
185 | } else { | |
9247f317 | 186 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); |
8ddc4734 | 187 | } |
0b246afa | 188 | spin_unlock(&fs_info->defrag_inodes_lock); |
a0f98dde | 189 | return 0; |
4cb5300b CM |
190 | } |
191 | ||
192 | /* | |
8ddc4734 MX |
193 | * Requeue the defrag object. If there is a defrag object that points to |
194 | * the same inode in the tree, we will merge them together (by | |
195 | * __btrfs_add_inode_defrag()) and free the one that we want to requeue. | |
4cb5300b | 196 | */ |
46e59791 | 197 | static void btrfs_requeue_inode_defrag(struct btrfs_inode *inode, |
48a3b636 | 198 | struct inode_defrag *defrag) |
8ddc4734 | 199 | { |
46e59791 | 200 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
8ddc4734 MX |
201 | int ret; |
202 | ||
2ff7e61e | 203 | if (!__need_auto_defrag(fs_info)) |
8ddc4734 MX |
204 | goto out; |
205 | ||
206 | /* | |
207 | * Here we don't check the IN_DEFRAG flag, because we need merge | |
208 | * them together. | |
209 | */ | |
0b246afa | 210 | spin_lock(&fs_info->defrag_inodes_lock); |
8ddc4734 | 211 | ret = __btrfs_add_inode_defrag(inode, defrag); |
0b246afa | 212 | spin_unlock(&fs_info->defrag_inodes_lock); |
8ddc4734 MX |
213 | if (ret) |
214 | goto out; | |
215 | return; | |
216 | out: | |
217 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
218 | } | |
219 | ||
4cb5300b | 220 | /* |
26176e7c MX |
221 | * pick the defragable inode that we want, if it doesn't exist, we will get |
222 | * the next one. | |
4cb5300b | 223 | */ |
26176e7c MX |
224 | static struct inode_defrag * |
225 | btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino) | |
4cb5300b CM |
226 | { |
227 | struct inode_defrag *entry = NULL; | |
762f2263 | 228 | struct inode_defrag tmp; |
4cb5300b CM |
229 | struct rb_node *p; |
230 | struct rb_node *parent = NULL; | |
762f2263 MX |
231 | int ret; |
232 | ||
233 | tmp.ino = ino; | |
234 | tmp.root = root; | |
4cb5300b | 235 | |
26176e7c MX |
236 | spin_lock(&fs_info->defrag_inodes_lock); |
237 | p = fs_info->defrag_inodes.rb_node; | |
4cb5300b CM |
238 | while (p) { |
239 | parent = p; | |
240 | entry = rb_entry(parent, struct inode_defrag, rb_node); | |
241 | ||
762f2263 MX |
242 | ret = __compare_inode_defrag(&tmp, entry); |
243 | if (ret < 0) | |
4cb5300b | 244 | p = parent->rb_left; |
762f2263 | 245 | else if (ret > 0) |
4cb5300b CM |
246 | p = parent->rb_right; |
247 | else | |
26176e7c | 248 | goto out; |
4cb5300b CM |
249 | } |
250 | ||
26176e7c MX |
251 | if (parent && __compare_inode_defrag(&tmp, entry) > 0) { |
252 | parent = rb_next(parent); | |
253 | if (parent) | |
4cb5300b | 254 | entry = rb_entry(parent, struct inode_defrag, rb_node); |
26176e7c MX |
255 | else |
256 | entry = NULL; | |
4cb5300b | 257 | } |
26176e7c MX |
258 | out: |
259 | if (entry) | |
260 | rb_erase(parent, &fs_info->defrag_inodes); | |
261 | spin_unlock(&fs_info->defrag_inodes_lock); | |
262 | return entry; | |
4cb5300b CM |
263 | } |
264 | ||
26176e7c | 265 | void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info) |
4cb5300b CM |
266 | { |
267 | struct inode_defrag *defrag; | |
26176e7c MX |
268 | struct rb_node *node; |
269 | ||
270 | spin_lock(&fs_info->defrag_inodes_lock); | |
271 | node = rb_first(&fs_info->defrag_inodes); | |
272 | while (node) { | |
273 | rb_erase(node, &fs_info->defrag_inodes); | |
274 | defrag = rb_entry(node, struct inode_defrag, rb_node); | |
275 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
276 | ||
351810c1 | 277 | cond_resched_lock(&fs_info->defrag_inodes_lock); |
26176e7c MX |
278 | |
279 | node = rb_first(&fs_info->defrag_inodes); | |
280 | } | |
281 | spin_unlock(&fs_info->defrag_inodes_lock); | |
282 | } | |
283 | ||
284 | #define BTRFS_DEFRAG_BATCH 1024 | |
285 | ||
286 | static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info, | |
287 | struct inode_defrag *defrag) | |
288 | { | |
4cb5300b CM |
289 | struct btrfs_root *inode_root; |
290 | struct inode *inode; | |
4cb5300b CM |
291 | struct btrfs_key key; |
292 | struct btrfs_ioctl_defrag_range_args range; | |
4cb5300b | 293 | int num_defrag; |
6f1c3605 LB |
294 | int index; |
295 | int ret; | |
4cb5300b | 296 | |
26176e7c MX |
297 | /* get the inode */ |
298 | key.objectid = defrag->root; | |
962a298f | 299 | key.type = BTRFS_ROOT_ITEM_KEY; |
26176e7c | 300 | key.offset = (u64)-1; |
6f1c3605 LB |
301 | |
302 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
303 | ||
26176e7c MX |
304 | inode_root = btrfs_read_fs_root_no_name(fs_info, &key); |
305 | if (IS_ERR(inode_root)) { | |
6f1c3605 LB |
306 | ret = PTR_ERR(inode_root); |
307 | goto cleanup; | |
308 | } | |
26176e7c MX |
309 | |
310 | key.objectid = defrag->ino; | |
962a298f | 311 | key.type = BTRFS_INODE_ITEM_KEY; |
26176e7c MX |
312 | key.offset = 0; |
313 | inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL); | |
314 | if (IS_ERR(inode)) { | |
6f1c3605 LB |
315 | ret = PTR_ERR(inode); |
316 | goto cleanup; | |
26176e7c | 317 | } |
6f1c3605 | 318 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
26176e7c MX |
319 | |
320 | /* do a chunk of defrag */ | |
321 | clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags); | |
4cb5300b CM |
322 | memset(&range, 0, sizeof(range)); |
323 | range.len = (u64)-1; | |
26176e7c | 324 | range.start = defrag->last_offset; |
b66f00da MX |
325 | |
326 | sb_start_write(fs_info->sb); | |
26176e7c MX |
327 | num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid, |
328 | BTRFS_DEFRAG_BATCH); | |
b66f00da | 329 | sb_end_write(fs_info->sb); |
26176e7c MX |
330 | /* |
331 | * if we filled the whole defrag batch, there | |
332 | * must be more work to do. Queue this defrag | |
333 | * again | |
334 | */ | |
335 | if (num_defrag == BTRFS_DEFRAG_BATCH) { | |
336 | defrag->last_offset = range.start; | |
46e59791 | 337 | btrfs_requeue_inode_defrag(BTRFS_I(inode), defrag); |
26176e7c MX |
338 | } else if (defrag->last_offset && !defrag->cycled) { |
339 | /* | |
340 | * we didn't fill our defrag batch, but | |
341 | * we didn't start at zero. Make sure we loop | |
342 | * around to the start of the file. | |
343 | */ | |
344 | defrag->last_offset = 0; | |
345 | defrag->cycled = 1; | |
46e59791 | 346 | btrfs_requeue_inode_defrag(BTRFS_I(inode), defrag); |
26176e7c MX |
347 | } else { |
348 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
349 | } | |
350 | ||
351 | iput(inode); | |
352 | return 0; | |
6f1c3605 LB |
353 | cleanup: |
354 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
355 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); | |
356 | return ret; | |
26176e7c MX |
357 | } |
358 | ||
359 | /* | |
360 | * run through the list of inodes in the FS that need | |
361 | * defragging | |
362 | */ | |
363 | int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info) | |
364 | { | |
365 | struct inode_defrag *defrag; | |
366 | u64 first_ino = 0; | |
367 | u64 root_objectid = 0; | |
4cb5300b CM |
368 | |
369 | atomic_inc(&fs_info->defrag_running); | |
67871254 | 370 | while (1) { |
dc81cdc5 MX |
371 | /* Pause the auto defragger. */ |
372 | if (test_bit(BTRFS_FS_STATE_REMOUNTING, | |
373 | &fs_info->fs_state)) | |
374 | break; | |
375 | ||
2ff7e61e | 376 | if (!__need_auto_defrag(fs_info)) |
26176e7c | 377 | break; |
4cb5300b CM |
378 | |
379 | /* find an inode to defrag */ | |
26176e7c MX |
380 | defrag = btrfs_pick_defrag_inode(fs_info, root_objectid, |
381 | first_ino); | |
4cb5300b | 382 | if (!defrag) { |
26176e7c | 383 | if (root_objectid || first_ino) { |
762f2263 | 384 | root_objectid = 0; |
4cb5300b CM |
385 | first_ino = 0; |
386 | continue; | |
387 | } else { | |
388 | break; | |
389 | } | |
390 | } | |
391 | ||
4cb5300b | 392 | first_ino = defrag->ino + 1; |
762f2263 | 393 | root_objectid = defrag->root; |
4cb5300b | 394 | |
26176e7c | 395 | __btrfs_run_defrag_inode(fs_info, defrag); |
4cb5300b | 396 | } |
4cb5300b CM |
397 | atomic_dec(&fs_info->defrag_running); |
398 | ||
399 | /* | |
400 | * during unmount, we use the transaction_wait queue to | |
401 | * wait for the defragger to stop | |
402 | */ | |
403 | wake_up(&fs_info->transaction_wait); | |
404 | return 0; | |
405 | } | |
39279cc3 | 406 | |
d352ac68 CM |
407 | /* simple helper to fault in pages and copy. This should go away |
408 | * and be replaced with calls into generic code. | |
409 | */ | |
ee22f0c4 | 410 | static noinline int btrfs_copy_from_user(loff_t pos, size_t write_bytes, |
a1b32a59 | 411 | struct page **prepared_pages, |
11c65dcc | 412 | struct iov_iter *i) |
39279cc3 | 413 | { |
914ee295 | 414 | size_t copied = 0; |
d0215f3e | 415 | size_t total_copied = 0; |
11c65dcc | 416 | int pg = 0; |
09cbfeaf | 417 | int offset = pos & (PAGE_SIZE - 1); |
39279cc3 | 418 | |
11c65dcc | 419 | while (write_bytes > 0) { |
39279cc3 | 420 | size_t count = min_t(size_t, |
09cbfeaf | 421 | PAGE_SIZE - offset, write_bytes); |
11c65dcc | 422 | struct page *page = prepared_pages[pg]; |
914ee295 XZ |
423 | /* |
424 | * Copy data from userspace to the current page | |
914ee295 | 425 | */ |
914ee295 | 426 | copied = iov_iter_copy_from_user_atomic(page, i, offset, count); |
11c65dcc | 427 | |
39279cc3 CM |
428 | /* Flush processor's dcache for this page */ |
429 | flush_dcache_page(page); | |
31339acd CM |
430 | |
431 | /* | |
432 | * if we get a partial write, we can end up with | |
433 | * partially up to date pages. These add | |
434 | * a lot of complexity, so make sure they don't | |
435 | * happen by forcing this copy to be retried. | |
436 | * | |
437 | * The rest of the btrfs_file_write code will fall | |
438 | * back to page at a time copies after we return 0. | |
439 | */ | |
440 | if (!PageUptodate(page) && copied < count) | |
441 | copied = 0; | |
442 | ||
11c65dcc JB |
443 | iov_iter_advance(i, copied); |
444 | write_bytes -= copied; | |
914ee295 | 445 | total_copied += copied; |
39279cc3 | 446 | |
b30ac0fc | 447 | /* Return to btrfs_file_write_iter to fault page */ |
9f570b8d | 448 | if (unlikely(copied == 0)) |
914ee295 | 449 | break; |
11c65dcc | 450 | |
09cbfeaf | 451 | if (copied < PAGE_SIZE - offset) { |
11c65dcc JB |
452 | offset += copied; |
453 | } else { | |
454 | pg++; | |
455 | offset = 0; | |
456 | } | |
39279cc3 | 457 | } |
914ee295 | 458 | return total_copied; |
39279cc3 CM |
459 | } |
460 | ||
d352ac68 CM |
461 | /* |
462 | * unlocks pages after btrfs_file_write is done with them | |
463 | */ | |
48a3b636 | 464 | static void btrfs_drop_pages(struct page **pages, size_t num_pages) |
39279cc3 CM |
465 | { |
466 | size_t i; | |
467 | for (i = 0; i < num_pages; i++) { | |
d352ac68 CM |
468 | /* page checked is some magic around finding pages that |
469 | * have been modified without going through btrfs_set_page_dirty | |
2457aec6 MG |
470 | * clear it here. There should be no need to mark the pages |
471 | * accessed as prepare_pages should have marked them accessed | |
472 | * in prepare_pages via find_or_create_page() | |
d352ac68 | 473 | */ |
4a096752 | 474 | ClearPageChecked(pages[i]); |
39279cc3 | 475 | unlock_page(pages[i]); |
09cbfeaf | 476 | put_page(pages[i]); |
39279cc3 CM |
477 | } |
478 | } | |
479 | ||
f48bf66b FM |
480 | static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode, |
481 | const u64 start, | |
482 | const u64 len, | |
483 | struct extent_state **cached_state) | |
484 | { | |
485 | u64 search_start = start; | |
486 | const u64 end = start + len - 1; | |
487 | ||
488 | while (search_start < end) { | |
489 | const u64 search_len = end - search_start + 1; | |
490 | struct extent_map *em; | |
491 | u64 em_len; | |
492 | int ret = 0; | |
493 | ||
494 | em = btrfs_get_extent(inode, NULL, 0, search_start, | |
495 | search_len, 0); | |
496 | if (IS_ERR(em)) | |
497 | return PTR_ERR(em); | |
498 | ||
499 | if (em->block_start != EXTENT_MAP_HOLE) | |
500 | goto next; | |
501 | ||
502 | em_len = em->len; | |
503 | if (em->start < search_start) | |
504 | em_len -= search_start - em->start; | |
505 | if (em_len > search_len) | |
506 | em_len = search_len; | |
507 | ||
508 | ret = set_extent_bit(&inode->io_tree, search_start, | |
509 | search_start + em_len - 1, | |
510 | EXTENT_DELALLOC_NEW, | |
511 | NULL, cached_state, GFP_NOFS); | |
512 | next: | |
513 | search_start = extent_map_end(em); | |
514 | free_extent_map(em); | |
515 | if (ret) | |
516 | return ret; | |
517 | } | |
518 | return 0; | |
519 | } | |
520 | ||
d352ac68 CM |
521 | /* |
522 | * after copy_from_user, pages need to be dirtied and we need to make | |
523 | * sure holes are created between the current EOF and the start of | |
524 | * any next extents (if required). | |
525 | * | |
526 | * this also makes the decision about creating an inline extent vs | |
527 | * doing real data extents, marking pages dirty and delalloc as required. | |
528 | */ | |
2ff7e61e JM |
529 | int btrfs_dirty_pages(struct inode *inode, struct page **pages, |
530 | size_t num_pages, loff_t pos, size_t write_bytes, | |
531 | struct extent_state **cached) | |
39279cc3 | 532 | { |
0b246afa | 533 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 534 | int err = 0; |
a52d9a80 | 535 | int i; |
db94535d | 536 | u64 num_bytes; |
a52d9a80 CM |
537 | u64 start_pos; |
538 | u64 end_of_last_block; | |
539 | u64 end_pos = pos + write_bytes; | |
540 | loff_t isize = i_size_read(inode); | |
e3b8a485 | 541 | unsigned int extra_bits = 0; |
39279cc3 | 542 | |
0b246afa | 543 | start_pos = pos & ~((u64) fs_info->sectorsize - 1); |
da17066c | 544 | num_bytes = round_up(write_bytes + pos - start_pos, |
0b246afa | 545 | fs_info->sectorsize); |
39279cc3 | 546 | |
db94535d | 547 | end_of_last_block = start_pos + num_bytes - 1; |
e3b8a485 FM |
548 | |
549 | if (!btrfs_is_free_space_inode(BTRFS_I(inode))) { | |
550 | if (start_pos >= isize && | |
551 | !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)) { | |
552 | /* | |
553 | * There can't be any extents following eof in this case | |
554 | * so just set the delalloc new bit for the range | |
555 | * directly. | |
556 | */ | |
557 | extra_bits |= EXTENT_DELALLOC_NEW; | |
558 | } else { | |
559 | err = btrfs_find_new_delalloc_bytes(BTRFS_I(inode), | |
560 | start_pos, | |
561 | num_bytes, cached); | |
562 | if (err) | |
563 | return err; | |
564 | } | |
565 | } | |
566 | ||
2ac55d41 | 567 | err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block, |
e3b8a485 | 568 | extra_bits, cached, 0); |
d0215f3e JB |
569 | if (err) |
570 | return err; | |
9ed74f2d | 571 | |
c8b97818 CM |
572 | for (i = 0; i < num_pages; i++) { |
573 | struct page *p = pages[i]; | |
574 | SetPageUptodate(p); | |
575 | ClearPageChecked(p); | |
576 | set_page_dirty(p); | |
a52d9a80 | 577 | } |
9f570b8d JB |
578 | |
579 | /* | |
580 | * we've only changed i_size in ram, and we haven't updated | |
581 | * the disk i_size. There is no need to log the inode | |
582 | * at this time. | |
583 | */ | |
584 | if (end_pos > isize) | |
a52d9a80 | 585 | i_size_write(inode, end_pos); |
a22285a6 | 586 | return 0; |
39279cc3 CM |
587 | } |
588 | ||
d352ac68 CM |
589 | /* |
590 | * this drops all the extents in the cache that intersect the range | |
591 | * [start, end]. Existing extents are split as required. | |
592 | */ | |
dcdbc059 | 593 | void btrfs_drop_extent_cache(struct btrfs_inode *inode, u64 start, u64 end, |
7014cdb4 | 594 | int skip_pinned) |
a52d9a80 CM |
595 | { |
596 | struct extent_map *em; | |
3b951516 CM |
597 | struct extent_map *split = NULL; |
598 | struct extent_map *split2 = NULL; | |
dcdbc059 | 599 | struct extent_map_tree *em_tree = &inode->extent_tree; |
39b5637f | 600 | u64 len = end - start + 1; |
5dc562c5 | 601 | u64 gen; |
3b951516 CM |
602 | int ret; |
603 | int testend = 1; | |
5b21f2ed | 604 | unsigned long flags; |
c8b97818 | 605 | int compressed = 0; |
09a2a8f9 | 606 | bool modified; |
a52d9a80 | 607 | |
e6dcd2dc | 608 | WARN_ON(end < start); |
3b951516 | 609 | if (end == (u64)-1) { |
39b5637f | 610 | len = (u64)-1; |
3b951516 CM |
611 | testend = 0; |
612 | } | |
d397712b | 613 | while (1) { |
7014cdb4 JB |
614 | int no_splits = 0; |
615 | ||
09a2a8f9 | 616 | modified = false; |
3b951516 | 617 | if (!split) |
172ddd60 | 618 | split = alloc_extent_map(); |
3b951516 | 619 | if (!split2) |
172ddd60 | 620 | split2 = alloc_extent_map(); |
7014cdb4 JB |
621 | if (!split || !split2) |
622 | no_splits = 1; | |
3b951516 | 623 | |
890871be | 624 | write_lock(&em_tree->lock); |
39b5637f | 625 | em = lookup_extent_mapping(em_tree, start, len); |
d1310b2e | 626 | if (!em) { |
890871be | 627 | write_unlock(&em_tree->lock); |
a52d9a80 | 628 | break; |
d1310b2e | 629 | } |
5b21f2ed | 630 | flags = em->flags; |
5dc562c5 | 631 | gen = em->generation; |
5b21f2ed | 632 | if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) { |
55ef6899 | 633 | if (testend && em->start + em->len >= start + len) { |
5b21f2ed | 634 | free_extent_map(em); |
a1ed835e | 635 | write_unlock(&em_tree->lock); |
5b21f2ed ZY |
636 | break; |
637 | } | |
55ef6899 YZ |
638 | start = em->start + em->len; |
639 | if (testend) | |
5b21f2ed | 640 | len = start + len - (em->start + em->len); |
5b21f2ed | 641 | free_extent_map(em); |
a1ed835e | 642 | write_unlock(&em_tree->lock); |
5b21f2ed ZY |
643 | continue; |
644 | } | |
c8b97818 | 645 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
3ce7e67a | 646 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
3b277594 | 647 | clear_bit(EXTENT_FLAG_LOGGING, &flags); |
09a2a8f9 | 648 | modified = !list_empty(&em->list); |
7014cdb4 JB |
649 | if (no_splits) |
650 | goto next; | |
3b951516 | 651 | |
ee20a983 | 652 | if (em->start < start) { |
3b951516 CM |
653 | split->start = em->start; |
654 | split->len = start - em->start; | |
ee20a983 JB |
655 | |
656 | if (em->block_start < EXTENT_MAP_LAST_BYTE) { | |
657 | split->orig_start = em->orig_start; | |
658 | split->block_start = em->block_start; | |
659 | ||
660 | if (compressed) | |
661 | split->block_len = em->block_len; | |
662 | else | |
663 | split->block_len = split->len; | |
664 | split->orig_block_len = max(split->block_len, | |
665 | em->orig_block_len); | |
666 | split->ram_bytes = em->ram_bytes; | |
667 | } else { | |
668 | split->orig_start = split->start; | |
669 | split->block_len = 0; | |
670 | split->block_start = em->block_start; | |
671 | split->orig_block_len = 0; | |
672 | split->ram_bytes = split->len; | |
673 | } | |
674 | ||
5dc562c5 | 675 | split->generation = gen; |
3b951516 | 676 | split->bdev = em->bdev; |
5b21f2ed | 677 | split->flags = flags; |
261507a0 | 678 | split->compress_type = em->compress_type; |
176840b3 | 679 | replace_extent_mapping(em_tree, em, split, modified); |
3b951516 CM |
680 | free_extent_map(split); |
681 | split = split2; | |
682 | split2 = NULL; | |
683 | } | |
ee20a983 | 684 | if (testend && em->start + em->len > start + len) { |
3b951516 CM |
685 | u64 diff = start + len - em->start; |
686 | ||
687 | split->start = start + len; | |
688 | split->len = em->start + em->len - (start + len); | |
689 | split->bdev = em->bdev; | |
5b21f2ed | 690 | split->flags = flags; |
261507a0 | 691 | split->compress_type = em->compress_type; |
5dc562c5 | 692 | split->generation = gen; |
ee20a983 JB |
693 | |
694 | if (em->block_start < EXTENT_MAP_LAST_BYTE) { | |
695 | split->orig_block_len = max(em->block_len, | |
b4939680 | 696 | em->orig_block_len); |
3b951516 | 697 | |
ee20a983 JB |
698 | split->ram_bytes = em->ram_bytes; |
699 | if (compressed) { | |
700 | split->block_len = em->block_len; | |
701 | split->block_start = em->block_start; | |
702 | split->orig_start = em->orig_start; | |
703 | } else { | |
704 | split->block_len = split->len; | |
705 | split->block_start = em->block_start | |
706 | + diff; | |
707 | split->orig_start = em->orig_start; | |
708 | } | |
c8b97818 | 709 | } else { |
ee20a983 JB |
710 | split->ram_bytes = split->len; |
711 | split->orig_start = split->start; | |
712 | split->block_len = 0; | |
713 | split->block_start = em->block_start; | |
714 | split->orig_block_len = 0; | |
c8b97818 | 715 | } |
3b951516 | 716 | |
176840b3 FM |
717 | if (extent_map_in_tree(em)) { |
718 | replace_extent_mapping(em_tree, em, split, | |
719 | modified); | |
720 | } else { | |
721 | ret = add_extent_mapping(em_tree, split, | |
722 | modified); | |
723 | ASSERT(ret == 0); /* Logic error */ | |
724 | } | |
3b951516 CM |
725 | free_extent_map(split); |
726 | split = NULL; | |
727 | } | |
7014cdb4 | 728 | next: |
176840b3 FM |
729 | if (extent_map_in_tree(em)) |
730 | remove_extent_mapping(em_tree, em); | |
890871be | 731 | write_unlock(&em_tree->lock); |
d1310b2e | 732 | |
a52d9a80 CM |
733 | /* once for us */ |
734 | free_extent_map(em); | |
735 | /* once for the tree*/ | |
736 | free_extent_map(em); | |
737 | } | |
3b951516 CM |
738 | if (split) |
739 | free_extent_map(split); | |
740 | if (split2) | |
741 | free_extent_map(split2); | |
a52d9a80 CM |
742 | } |
743 | ||
39279cc3 CM |
744 | /* |
745 | * this is very complex, but the basic idea is to drop all extents | |
746 | * in the range start - end. hint_block is filled in with a block number | |
747 | * that would be a good hint to the block allocator for this file. | |
748 | * | |
749 | * If an extent intersects the range but is not entirely inside the range | |
750 | * it is either truncated or split. Anything entirely inside the range | |
751 | * is deleted from the tree. | |
752 | */ | |
5dc562c5 JB |
753 | int __btrfs_drop_extents(struct btrfs_trans_handle *trans, |
754 | struct btrfs_root *root, struct inode *inode, | |
755 | struct btrfs_path *path, u64 start, u64 end, | |
1acae57b FDBM |
756 | u64 *drop_end, int drop_cache, |
757 | int replace_extent, | |
758 | u32 extent_item_size, | |
759 | int *key_inserted) | |
39279cc3 | 760 | { |
0b246afa | 761 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 762 | struct extent_buffer *leaf; |
920bbbfb | 763 | struct btrfs_file_extent_item *fi; |
00f5c795 | 764 | struct btrfs_key key; |
920bbbfb | 765 | struct btrfs_key new_key; |
4a0cc7ca | 766 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
920bbbfb YZ |
767 | u64 search_start = start; |
768 | u64 disk_bytenr = 0; | |
769 | u64 num_bytes = 0; | |
770 | u64 extent_offset = 0; | |
771 | u64 extent_end = 0; | |
62fe51c1 | 772 | u64 last_end = start; |
920bbbfb YZ |
773 | int del_nr = 0; |
774 | int del_slot = 0; | |
775 | int extent_type; | |
ccd467d6 | 776 | int recow; |
00f5c795 | 777 | int ret; |
dc7fdde3 | 778 | int modify_tree = -1; |
27cdeb70 | 779 | int update_refs; |
c3308f84 | 780 | int found = 0; |
1acae57b | 781 | int leafs_visited = 0; |
39279cc3 | 782 | |
a1ed835e | 783 | if (drop_cache) |
dcdbc059 | 784 | btrfs_drop_extent_cache(BTRFS_I(inode), start, end - 1, 0); |
a52d9a80 | 785 | |
d5f37527 | 786 | if (start >= BTRFS_I(inode)->disk_i_size && !replace_extent) |
dc7fdde3 CM |
787 | modify_tree = 0; |
788 | ||
27cdeb70 | 789 | update_refs = (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa | 790 | root == fs_info->tree_root); |
d397712b | 791 | while (1) { |
ccd467d6 | 792 | recow = 0; |
33345d01 | 793 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
dc7fdde3 | 794 | search_start, modify_tree); |
39279cc3 | 795 | if (ret < 0) |
920bbbfb YZ |
796 | break; |
797 | if (ret > 0 && path->slots[0] > 0 && search_start == start) { | |
798 | leaf = path->nodes[0]; | |
799 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1); | |
33345d01 | 800 | if (key.objectid == ino && |
920bbbfb YZ |
801 | key.type == BTRFS_EXTENT_DATA_KEY) |
802 | path->slots[0]--; | |
39279cc3 | 803 | } |
920bbbfb | 804 | ret = 0; |
1acae57b | 805 | leafs_visited++; |
8c2383c3 | 806 | next_slot: |
5f39d397 | 807 | leaf = path->nodes[0]; |
920bbbfb YZ |
808 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { |
809 | BUG_ON(del_nr > 0); | |
810 | ret = btrfs_next_leaf(root, path); | |
811 | if (ret < 0) | |
812 | break; | |
813 | if (ret > 0) { | |
814 | ret = 0; | |
815 | break; | |
8c2383c3 | 816 | } |
1acae57b | 817 | leafs_visited++; |
920bbbfb YZ |
818 | leaf = path->nodes[0]; |
819 | recow = 1; | |
820 | } | |
821 | ||
822 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
aeafbf84 FM |
823 | |
824 | if (key.objectid > ino) | |
825 | break; | |
826 | if (WARN_ON_ONCE(key.objectid < ino) || | |
827 | key.type < BTRFS_EXTENT_DATA_KEY) { | |
828 | ASSERT(del_nr == 0); | |
829 | path->slots[0]++; | |
830 | goto next_slot; | |
831 | } | |
832 | if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end) | |
920bbbfb YZ |
833 | break; |
834 | ||
835 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
836 | struct btrfs_file_extent_item); | |
837 | extent_type = btrfs_file_extent_type(leaf, fi); | |
838 | ||
839 | if (extent_type == BTRFS_FILE_EXTENT_REG || | |
840 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
841 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
842 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
843 | extent_offset = btrfs_file_extent_offset(leaf, fi); | |
844 | extent_end = key.offset + | |
845 | btrfs_file_extent_num_bytes(leaf, fi); | |
846 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
847 | extent_end = key.offset + | |
514ac8ad CM |
848 | btrfs_file_extent_inline_len(leaf, |
849 | path->slots[0], fi); | |
8c2383c3 | 850 | } else { |
aeafbf84 FM |
851 | /* can't happen */ |
852 | BUG(); | |
39279cc3 CM |
853 | } |
854 | ||
fc19c5e7 FM |
855 | /* |
856 | * Don't skip extent items representing 0 byte lengths. They | |
857 | * used to be created (bug) if while punching holes we hit | |
858 | * -ENOSPC condition. So if we find one here, just ensure we | |
859 | * delete it, otherwise we would insert a new file extent item | |
860 | * with the same key (offset) as that 0 bytes length file | |
861 | * extent item in the call to setup_items_for_insert() later | |
862 | * in this function. | |
863 | */ | |
62fe51c1 JB |
864 | if (extent_end == key.offset && extent_end >= search_start) { |
865 | last_end = extent_end; | |
fc19c5e7 | 866 | goto delete_extent_item; |
62fe51c1 | 867 | } |
fc19c5e7 | 868 | |
920bbbfb YZ |
869 | if (extent_end <= search_start) { |
870 | path->slots[0]++; | |
8c2383c3 | 871 | goto next_slot; |
39279cc3 CM |
872 | } |
873 | ||
c3308f84 | 874 | found = 1; |
920bbbfb | 875 | search_start = max(key.offset, start); |
dc7fdde3 CM |
876 | if (recow || !modify_tree) { |
877 | modify_tree = -1; | |
b3b4aa74 | 878 | btrfs_release_path(path); |
920bbbfb | 879 | continue; |
39279cc3 | 880 | } |
6643558d | 881 | |
920bbbfb YZ |
882 | /* |
883 | * | - range to drop - | | |
884 | * | -------- extent -------- | | |
885 | */ | |
886 | if (start > key.offset && end < extent_end) { | |
887 | BUG_ON(del_nr > 0); | |
00fdf13a | 888 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
3f9e3df8 | 889 | ret = -EOPNOTSUPP; |
00fdf13a LB |
890 | break; |
891 | } | |
920bbbfb YZ |
892 | |
893 | memcpy(&new_key, &key, sizeof(new_key)); | |
894 | new_key.offset = start; | |
895 | ret = btrfs_duplicate_item(trans, root, path, | |
896 | &new_key); | |
897 | if (ret == -EAGAIN) { | |
b3b4aa74 | 898 | btrfs_release_path(path); |
920bbbfb | 899 | continue; |
6643558d | 900 | } |
920bbbfb YZ |
901 | if (ret < 0) |
902 | break; | |
903 | ||
904 | leaf = path->nodes[0]; | |
905 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
906 | struct btrfs_file_extent_item); | |
907 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
908 | start - key.offset); | |
909 | ||
910 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
911 | struct btrfs_file_extent_item); | |
912 | ||
913 | extent_offset += start - key.offset; | |
914 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); | |
915 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
916 | extent_end - start); | |
917 | btrfs_mark_buffer_dirty(leaf); | |
918 | ||
5dc562c5 | 919 | if (update_refs && disk_bytenr > 0) { |
84f7d8e6 | 920 | ret = btrfs_inc_extent_ref(trans, root, |
920bbbfb YZ |
921 | disk_bytenr, num_bytes, 0, |
922 | root->root_key.objectid, | |
923 | new_key.objectid, | |
b06c4bf5 | 924 | start - extent_offset); |
79787eaa | 925 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 926 | } |
920bbbfb | 927 | key.offset = start; |
6643558d | 928 | } |
62fe51c1 JB |
929 | /* |
930 | * From here on out we will have actually dropped something, so | |
931 | * last_end can be updated. | |
932 | */ | |
933 | last_end = extent_end; | |
934 | ||
920bbbfb YZ |
935 | /* |
936 | * | ---- range to drop ----- | | |
937 | * | -------- extent -------- | | |
938 | */ | |
939 | if (start <= key.offset && end < extent_end) { | |
00fdf13a | 940 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
3f9e3df8 | 941 | ret = -EOPNOTSUPP; |
00fdf13a LB |
942 | break; |
943 | } | |
6643558d | 944 | |
920bbbfb YZ |
945 | memcpy(&new_key, &key, sizeof(new_key)); |
946 | new_key.offset = end; | |
0b246afa | 947 | btrfs_set_item_key_safe(fs_info, path, &new_key); |
6643558d | 948 | |
920bbbfb YZ |
949 | extent_offset += end - key.offset; |
950 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); | |
951 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
952 | extent_end - end); | |
953 | btrfs_mark_buffer_dirty(leaf); | |
2671485d | 954 | if (update_refs && disk_bytenr > 0) |
920bbbfb | 955 | inode_sub_bytes(inode, end - key.offset); |
920bbbfb | 956 | break; |
39279cc3 | 957 | } |
771ed689 | 958 | |
920bbbfb YZ |
959 | search_start = extent_end; |
960 | /* | |
961 | * | ---- range to drop ----- | | |
962 | * | -------- extent -------- | | |
963 | */ | |
964 | if (start > key.offset && end >= extent_end) { | |
965 | BUG_ON(del_nr > 0); | |
00fdf13a | 966 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
3f9e3df8 | 967 | ret = -EOPNOTSUPP; |
00fdf13a LB |
968 | break; |
969 | } | |
8c2383c3 | 970 | |
920bbbfb YZ |
971 | btrfs_set_file_extent_num_bytes(leaf, fi, |
972 | start - key.offset); | |
973 | btrfs_mark_buffer_dirty(leaf); | |
2671485d | 974 | if (update_refs && disk_bytenr > 0) |
920bbbfb | 975 | inode_sub_bytes(inode, extent_end - start); |
920bbbfb YZ |
976 | if (end == extent_end) |
977 | break; | |
c8b97818 | 978 | |
920bbbfb YZ |
979 | path->slots[0]++; |
980 | goto next_slot; | |
31840ae1 ZY |
981 | } |
982 | ||
920bbbfb YZ |
983 | /* |
984 | * | ---- range to drop ----- | | |
985 | * | ------ extent ------ | | |
986 | */ | |
987 | if (start <= key.offset && end >= extent_end) { | |
fc19c5e7 | 988 | delete_extent_item: |
920bbbfb YZ |
989 | if (del_nr == 0) { |
990 | del_slot = path->slots[0]; | |
991 | del_nr = 1; | |
992 | } else { | |
993 | BUG_ON(del_slot + del_nr != path->slots[0]); | |
994 | del_nr++; | |
995 | } | |
31840ae1 | 996 | |
5dc562c5 JB |
997 | if (update_refs && |
998 | extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
a76a3cd4 | 999 | inode_sub_bytes(inode, |
920bbbfb YZ |
1000 | extent_end - key.offset); |
1001 | extent_end = ALIGN(extent_end, | |
0b246afa | 1002 | fs_info->sectorsize); |
5dc562c5 | 1003 | } else if (update_refs && disk_bytenr > 0) { |
84f7d8e6 | 1004 | ret = btrfs_free_extent(trans, root, |
920bbbfb YZ |
1005 | disk_bytenr, num_bytes, 0, |
1006 | root->root_key.objectid, | |
5d4f98a2 | 1007 | key.objectid, key.offset - |
b06c4bf5 | 1008 | extent_offset); |
79787eaa | 1009 | BUG_ON(ret); /* -ENOMEM */ |
920bbbfb YZ |
1010 | inode_sub_bytes(inode, |
1011 | extent_end - key.offset); | |
31840ae1 | 1012 | } |
31840ae1 | 1013 | |
920bbbfb YZ |
1014 | if (end == extent_end) |
1015 | break; | |
1016 | ||
1017 | if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) { | |
1018 | path->slots[0]++; | |
1019 | goto next_slot; | |
1020 | } | |
1021 | ||
1022 | ret = btrfs_del_items(trans, root, path, del_slot, | |
1023 | del_nr); | |
79787eaa | 1024 | if (ret) { |
66642832 | 1025 | btrfs_abort_transaction(trans, ret); |
5dc562c5 | 1026 | break; |
79787eaa | 1027 | } |
920bbbfb YZ |
1028 | |
1029 | del_nr = 0; | |
1030 | del_slot = 0; | |
1031 | ||
b3b4aa74 | 1032 | btrfs_release_path(path); |
920bbbfb | 1033 | continue; |
39279cc3 | 1034 | } |
920bbbfb YZ |
1035 | |
1036 | BUG_ON(1); | |
39279cc3 | 1037 | } |
920bbbfb | 1038 | |
79787eaa | 1039 | if (!ret && del_nr > 0) { |
1acae57b FDBM |
1040 | /* |
1041 | * Set path->slots[0] to first slot, so that after the delete | |
1042 | * if items are move off from our leaf to its immediate left or | |
1043 | * right neighbor leafs, we end up with a correct and adjusted | |
d5f37527 | 1044 | * path->slots[0] for our insertion (if replace_extent != 0). |
1acae57b FDBM |
1045 | */ |
1046 | path->slots[0] = del_slot; | |
920bbbfb | 1047 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
79787eaa | 1048 | if (ret) |
66642832 | 1049 | btrfs_abort_transaction(trans, ret); |
d5f37527 | 1050 | } |
1acae57b | 1051 | |
d5f37527 FDBM |
1052 | leaf = path->nodes[0]; |
1053 | /* | |
1054 | * If btrfs_del_items() was called, it might have deleted a leaf, in | |
1055 | * which case it unlocked our path, so check path->locks[0] matches a | |
1056 | * write lock. | |
1057 | */ | |
1058 | if (!ret && replace_extent && leafs_visited == 1 && | |
1059 | (path->locks[0] == BTRFS_WRITE_LOCK_BLOCKING || | |
1060 | path->locks[0] == BTRFS_WRITE_LOCK) && | |
2ff7e61e | 1061 | btrfs_leaf_free_space(fs_info, leaf) >= |
d5f37527 FDBM |
1062 | sizeof(struct btrfs_item) + extent_item_size) { |
1063 | ||
1064 | key.objectid = ino; | |
1065 | key.type = BTRFS_EXTENT_DATA_KEY; | |
1066 | key.offset = start; | |
1067 | if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) { | |
1068 | struct btrfs_key slot_key; | |
1069 | ||
1070 | btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]); | |
1071 | if (btrfs_comp_cpu_keys(&key, &slot_key) > 0) | |
1072 | path->slots[0]++; | |
1acae57b | 1073 | } |
d5f37527 FDBM |
1074 | setup_items_for_insert(root, path, &key, |
1075 | &extent_item_size, | |
1076 | extent_item_size, | |
1077 | sizeof(struct btrfs_item) + | |
1078 | extent_item_size, 1); | |
1079 | *key_inserted = 1; | |
6643558d | 1080 | } |
920bbbfb | 1081 | |
1acae57b FDBM |
1082 | if (!replace_extent || !(*key_inserted)) |
1083 | btrfs_release_path(path); | |
2aaa6655 | 1084 | if (drop_end) |
62fe51c1 | 1085 | *drop_end = found ? min(end, last_end) : end; |
5dc562c5 JB |
1086 | return ret; |
1087 | } | |
1088 | ||
1089 | int btrfs_drop_extents(struct btrfs_trans_handle *trans, | |
1090 | struct btrfs_root *root, struct inode *inode, u64 start, | |
2671485d | 1091 | u64 end, int drop_cache) |
5dc562c5 JB |
1092 | { |
1093 | struct btrfs_path *path; | |
1094 | int ret; | |
1095 | ||
1096 | path = btrfs_alloc_path(); | |
1097 | if (!path) | |
1098 | return -ENOMEM; | |
2aaa6655 | 1099 | ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL, |
1acae57b | 1100 | drop_cache, 0, 0, NULL); |
920bbbfb | 1101 | btrfs_free_path(path); |
39279cc3 CM |
1102 | return ret; |
1103 | } | |
1104 | ||
d899e052 | 1105 | static int extent_mergeable(struct extent_buffer *leaf, int slot, |
6c7d54ac YZ |
1106 | u64 objectid, u64 bytenr, u64 orig_offset, |
1107 | u64 *start, u64 *end) | |
d899e052 YZ |
1108 | { |
1109 | struct btrfs_file_extent_item *fi; | |
1110 | struct btrfs_key key; | |
1111 | u64 extent_end; | |
1112 | ||
1113 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) | |
1114 | return 0; | |
1115 | ||
1116 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
1117 | if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY) | |
1118 | return 0; | |
1119 | ||
1120 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
1121 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG || | |
1122 | btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr || | |
6c7d54ac | 1123 | btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset || |
d899e052 YZ |
1124 | btrfs_file_extent_compression(leaf, fi) || |
1125 | btrfs_file_extent_encryption(leaf, fi) || | |
1126 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1127 | return 0; | |
1128 | ||
1129 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
1130 | if ((*start && *start != key.offset) || (*end && *end != extent_end)) | |
1131 | return 0; | |
1132 | ||
1133 | *start = key.offset; | |
1134 | *end = extent_end; | |
1135 | return 1; | |
1136 | } | |
1137 | ||
1138 | /* | |
1139 | * Mark extent in the range start - end as written. | |
1140 | * | |
1141 | * This changes extent type from 'pre-allocated' to 'regular'. If only | |
1142 | * part of extent is marked as written, the extent will be split into | |
1143 | * two or three. | |
1144 | */ | |
1145 | int btrfs_mark_extent_written(struct btrfs_trans_handle *trans, | |
7a6d7067 | 1146 | struct btrfs_inode *inode, u64 start, u64 end) |
d899e052 | 1147 | { |
7a6d7067 NB |
1148 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
1149 | struct btrfs_root *root = inode->root; | |
d899e052 YZ |
1150 | struct extent_buffer *leaf; |
1151 | struct btrfs_path *path; | |
1152 | struct btrfs_file_extent_item *fi; | |
1153 | struct btrfs_key key; | |
920bbbfb | 1154 | struct btrfs_key new_key; |
d899e052 YZ |
1155 | u64 bytenr; |
1156 | u64 num_bytes; | |
1157 | u64 extent_end; | |
5d4f98a2 | 1158 | u64 orig_offset; |
d899e052 YZ |
1159 | u64 other_start; |
1160 | u64 other_end; | |
920bbbfb YZ |
1161 | u64 split; |
1162 | int del_nr = 0; | |
1163 | int del_slot = 0; | |
6c7d54ac | 1164 | int recow; |
d899e052 | 1165 | int ret; |
7a6d7067 | 1166 | u64 ino = btrfs_ino(inode); |
d899e052 | 1167 | |
d899e052 | 1168 | path = btrfs_alloc_path(); |
d8926bb3 MF |
1169 | if (!path) |
1170 | return -ENOMEM; | |
d899e052 | 1171 | again: |
6c7d54ac | 1172 | recow = 0; |
920bbbfb | 1173 | split = start; |
33345d01 | 1174 | key.objectid = ino; |
d899e052 | 1175 | key.type = BTRFS_EXTENT_DATA_KEY; |
920bbbfb | 1176 | key.offset = split; |
d899e052 YZ |
1177 | |
1178 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
41415730 JB |
1179 | if (ret < 0) |
1180 | goto out; | |
d899e052 YZ |
1181 | if (ret > 0 && path->slots[0] > 0) |
1182 | path->slots[0]--; | |
1183 | ||
1184 | leaf = path->nodes[0]; | |
1185 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
9c8e63db JB |
1186 | if (key.objectid != ino || |
1187 | key.type != BTRFS_EXTENT_DATA_KEY) { | |
1188 | ret = -EINVAL; | |
1189 | btrfs_abort_transaction(trans, ret); | |
1190 | goto out; | |
1191 | } | |
d899e052 YZ |
1192 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1193 | struct btrfs_file_extent_item); | |
9c8e63db JB |
1194 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_PREALLOC) { |
1195 | ret = -EINVAL; | |
1196 | btrfs_abort_transaction(trans, ret); | |
1197 | goto out; | |
1198 | } | |
d899e052 | 1199 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
9c8e63db JB |
1200 | if (key.offset > start || extent_end < end) { |
1201 | ret = -EINVAL; | |
1202 | btrfs_abort_transaction(trans, ret); | |
1203 | goto out; | |
1204 | } | |
d899e052 YZ |
1205 | |
1206 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
1207 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
5d4f98a2 | 1208 | orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi); |
6c7d54ac YZ |
1209 | memcpy(&new_key, &key, sizeof(new_key)); |
1210 | ||
1211 | if (start == key.offset && end < extent_end) { | |
1212 | other_start = 0; | |
1213 | other_end = start; | |
1214 | if (extent_mergeable(leaf, path->slots[0] - 1, | |
33345d01 | 1215 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1216 | &other_start, &other_end)) { |
1217 | new_key.offset = end; | |
0b246afa | 1218 | btrfs_set_item_key_safe(fs_info, path, &new_key); |
6c7d54ac YZ |
1219 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1220 | struct btrfs_file_extent_item); | |
224ecce5 JB |
1221 | btrfs_set_file_extent_generation(leaf, fi, |
1222 | trans->transid); | |
6c7d54ac YZ |
1223 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1224 | extent_end - end); | |
1225 | btrfs_set_file_extent_offset(leaf, fi, | |
1226 | end - orig_offset); | |
1227 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
1228 | struct btrfs_file_extent_item); | |
224ecce5 JB |
1229 | btrfs_set_file_extent_generation(leaf, fi, |
1230 | trans->transid); | |
6c7d54ac YZ |
1231 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1232 | end - other_start); | |
1233 | btrfs_mark_buffer_dirty(leaf); | |
1234 | goto out; | |
1235 | } | |
1236 | } | |
1237 | ||
1238 | if (start > key.offset && end == extent_end) { | |
1239 | other_start = end; | |
1240 | other_end = 0; | |
1241 | if (extent_mergeable(leaf, path->slots[0] + 1, | |
33345d01 | 1242 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1243 | &other_start, &other_end)) { |
1244 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1245 | struct btrfs_file_extent_item); | |
1246 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
1247 | start - key.offset); | |
224ecce5 JB |
1248 | btrfs_set_file_extent_generation(leaf, fi, |
1249 | trans->transid); | |
6c7d54ac YZ |
1250 | path->slots[0]++; |
1251 | new_key.offset = start; | |
0b246afa | 1252 | btrfs_set_item_key_safe(fs_info, path, &new_key); |
6c7d54ac YZ |
1253 | |
1254 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1255 | struct btrfs_file_extent_item); | |
224ecce5 JB |
1256 | btrfs_set_file_extent_generation(leaf, fi, |
1257 | trans->transid); | |
6c7d54ac YZ |
1258 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1259 | other_end - start); | |
1260 | btrfs_set_file_extent_offset(leaf, fi, | |
1261 | start - orig_offset); | |
1262 | btrfs_mark_buffer_dirty(leaf); | |
1263 | goto out; | |
1264 | } | |
1265 | } | |
d899e052 | 1266 | |
920bbbfb YZ |
1267 | while (start > key.offset || end < extent_end) { |
1268 | if (key.offset == start) | |
1269 | split = end; | |
1270 | ||
920bbbfb YZ |
1271 | new_key.offset = split; |
1272 | ret = btrfs_duplicate_item(trans, root, path, &new_key); | |
1273 | if (ret == -EAGAIN) { | |
b3b4aa74 | 1274 | btrfs_release_path(path); |
920bbbfb | 1275 | goto again; |
d899e052 | 1276 | } |
79787eaa | 1277 | if (ret < 0) { |
66642832 | 1278 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
1279 | goto out; |
1280 | } | |
d899e052 | 1281 | |
920bbbfb YZ |
1282 | leaf = path->nodes[0]; |
1283 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, | |
d899e052 | 1284 | struct btrfs_file_extent_item); |
224ecce5 | 1285 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
d899e052 | 1286 | btrfs_set_file_extent_num_bytes(leaf, fi, |
920bbbfb YZ |
1287 | split - key.offset); |
1288 | ||
1289 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1290 | struct btrfs_file_extent_item); | |
1291 | ||
224ecce5 | 1292 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
920bbbfb YZ |
1293 | btrfs_set_file_extent_offset(leaf, fi, split - orig_offset); |
1294 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
1295 | extent_end - split); | |
d899e052 YZ |
1296 | btrfs_mark_buffer_dirty(leaf); |
1297 | ||
84f7d8e6 | 1298 | ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, |
2ff7e61e | 1299 | 0, root->root_key.objectid, |
b06c4bf5 | 1300 | ino, orig_offset); |
9c8e63db JB |
1301 | if (ret) { |
1302 | btrfs_abort_transaction(trans, ret); | |
1303 | goto out; | |
1304 | } | |
d899e052 | 1305 | |
920bbbfb YZ |
1306 | if (split == start) { |
1307 | key.offset = start; | |
1308 | } else { | |
9c8e63db JB |
1309 | if (start != key.offset) { |
1310 | ret = -EINVAL; | |
1311 | btrfs_abort_transaction(trans, ret); | |
1312 | goto out; | |
1313 | } | |
d899e052 | 1314 | path->slots[0]--; |
920bbbfb | 1315 | extent_end = end; |
d899e052 | 1316 | } |
6c7d54ac | 1317 | recow = 1; |
d899e052 YZ |
1318 | } |
1319 | ||
920bbbfb YZ |
1320 | other_start = end; |
1321 | other_end = 0; | |
6c7d54ac | 1322 | if (extent_mergeable(leaf, path->slots[0] + 1, |
33345d01 | 1323 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1324 | &other_start, &other_end)) { |
1325 | if (recow) { | |
b3b4aa74 | 1326 | btrfs_release_path(path); |
6c7d54ac YZ |
1327 | goto again; |
1328 | } | |
920bbbfb YZ |
1329 | extent_end = other_end; |
1330 | del_slot = path->slots[0] + 1; | |
1331 | del_nr++; | |
84f7d8e6 | 1332 | ret = btrfs_free_extent(trans, root, bytenr, num_bytes, |
920bbbfb | 1333 | 0, root->root_key.objectid, |
b06c4bf5 | 1334 | ino, orig_offset); |
9c8e63db JB |
1335 | if (ret) { |
1336 | btrfs_abort_transaction(trans, ret); | |
1337 | goto out; | |
1338 | } | |
d899e052 | 1339 | } |
920bbbfb YZ |
1340 | other_start = 0; |
1341 | other_end = start; | |
6c7d54ac | 1342 | if (extent_mergeable(leaf, path->slots[0] - 1, |
33345d01 | 1343 | ino, bytenr, orig_offset, |
6c7d54ac YZ |
1344 | &other_start, &other_end)) { |
1345 | if (recow) { | |
b3b4aa74 | 1346 | btrfs_release_path(path); |
6c7d54ac YZ |
1347 | goto again; |
1348 | } | |
920bbbfb YZ |
1349 | key.offset = other_start; |
1350 | del_slot = path->slots[0]; | |
1351 | del_nr++; | |
84f7d8e6 | 1352 | ret = btrfs_free_extent(trans, root, bytenr, num_bytes, |
920bbbfb | 1353 | 0, root->root_key.objectid, |
b06c4bf5 | 1354 | ino, orig_offset); |
9c8e63db JB |
1355 | if (ret) { |
1356 | btrfs_abort_transaction(trans, ret); | |
1357 | goto out; | |
1358 | } | |
920bbbfb YZ |
1359 | } |
1360 | if (del_nr == 0) { | |
3f6fae95 SL |
1361 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1362 | struct btrfs_file_extent_item); | |
920bbbfb YZ |
1363 | btrfs_set_file_extent_type(leaf, fi, |
1364 | BTRFS_FILE_EXTENT_REG); | |
224ecce5 | 1365 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
920bbbfb | 1366 | btrfs_mark_buffer_dirty(leaf); |
6c7d54ac | 1367 | } else { |
3f6fae95 SL |
1368 | fi = btrfs_item_ptr(leaf, del_slot - 1, |
1369 | struct btrfs_file_extent_item); | |
6c7d54ac YZ |
1370 | btrfs_set_file_extent_type(leaf, fi, |
1371 | BTRFS_FILE_EXTENT_REG); | |
224ecce5 | 1372 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
6c7d54ac YZ |
1373 | btrfs_set_file_extent_num_bytes(leaf, fi, |
1374 | extent_end - key.offset); | |
1375 | btrfs_mark_buffer_dirty(leaf); | |
920bbbfb | 1376 | |
6c7d54ac | 1377 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
79787eaa | 1378 | if (ret < 0) { |
66642832 | 1379 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
1380 | goto out; |
1381 | } | |
6c7d54ac | 1382 | } |
920bbbfb | 1383 | out: |
d899e052 YZ |
1384 | btrfs_free_path(path); |
1385 | return 0; | |
1386 | } | |
1387 | ||
b1bf862e CM |
1388 | /* |
1389 | * on error we return an unlocked page and the error value | |
1390 | * on success we return a locked page and 0 | |
1391 | */ | |
bb1591b4 CM |
1392 | static int prepare_uptodate_page(struct inode *inode, |
1393 | struct page *page, u64 pos, | |
b6316429 | 1394 | bool force_uptodate) |
b1bf862e CM |
1395 | { |
1396 | int ret = 0; | |
1397 | ||
09cbfeaf | 1398 | if (((pos & (PAGE_SIZE - 1)) || force_uptodate) && |
b6316429 | 1399 | !PageUptodate(page)) { |
b1bf862e CM |
1400 | ret = btrfs_readpage(NULL, page); |
1401 | if (ret) | |
1402 | return ret; | |
1403 | lock_page(page); | |
1404 | if (!PageUptodate(page)) { | |
1405 | unlock_page(page); | |
1406 | return -EIO; | |
1407 | } | |
bb1591b4 CM |
1408 | if (page->mapping != inode->i_mapping) { |
1409 | unlock_page(page); | |
1410 | return -EAGAIN; | |
1411 | } | |
b1bf862e CM |
1412 | } |
1413 | return 0; | |
1414 | } | |
1415 | ||
39279cc3 | 1416 | /* |
376cc685 | 1417 | * this just gets pages into the page cache and locks them down. |
39279cc3 | 1418 | */ |
b37392ea MX |
1419 | static noinline int prepare_pages(struct inode *inode, struct page **pages, |
1420 | size_t num_pages, loff_t pos, | |
1421 | size_t write_bytes, bool force_uptodate) | |
39279cc3 CM |
1422 | { |
1423 | int i; | |
09cbfeaf | 1424 | unsigned long index = pos >> PAGE_SHIFT; |
3b16a4e3 | 1425 | gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); |
fc28b62d | 1426 | int err = 0; |
376cc685 | 1427 | int faili; |
8c2383c3 | 1428 | |
39279cc3 | 1429 | for (i = 0; i < num_pages; i++) { |
bb1591b4 | 1430 | again: |
a94733d0 | 1431 | pages[i] = find_or_create_page(inode->i_mapping, index + i, |
e3a41a5b | 1432 | mask | __GFP_WRITE); |
39279cc3 | 1433 | if (!pages[i]) { |
b1bf862e CM |
1434 | faili = i - 1; |
1435 | err = -ENOMEM; | |
1436 | goto fail; | |
1437 | } | |
1438 | ||
1439 | if (i == 0) | |
bb1591b4 | 1440 | err = prepare_uptodate_page(inode, pages[i], pos, |
b6316429 | 1441 | force_uptodate); |
bb1591b4 CM |
1442 | if (!err && i == num_pages - 1) |
1443 | err = prepare_uptodate_page(inode, pages[i], | |
b6316429 | 1444 | pos + write_bytes, false); |
b1bf862e | 1445 | if (err) { |
09cbfeaf | 1446 | put_page(pages[i]); |
bb1591b4 CM |
1447 | if (err == -EAGAIN) { |
1448 | err = 0; | |
1449 | goto again; | |
1450 | } | |
b1bf862e CM |
1451 | faili = i - 1; |
1452 | goto fail; | |
39279cc3 | 1453 | } |
ccd467d6 | 1454 | wait_on_page_writeback(pages[i]); |
39279cc3 | 1455 | } |
376cc685 MX |
1456 | |
1457 | return 0; | |
1458 | fail: | |
1459 | while (faili >= 0) { | |
1460 | unlock_page(pages[faili]); | |
09cbfeaf | 1461 | put_page(pages[faili]); |
376cc685 MX |
1462 | faili--; |
1463 | } | |
1464 | return err; | |
1465 | ||
1466 | } | |
1467 | ||
1468 | /* | |
1469 | * This function locks the extent and properly waits for data=ordered extents | |
1470 | * to finish before allowing the pages to be modified if need. | |
1471 | * | |
1472 | * The return value: | |
1473 | * 1 - the extent is locked | |
1474 | * 0 - the extent is not locked, and everything is OK | |
1475 | * -EAGAIN - need re-prepare the pages | |
1476 | * the other < 0 number - Something wrong happens | |
1477 | */ | |
1478 | static noinline int | |
2cff578c | 1479 | lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages, |
376cc685 | 1480 | size_t num_pages, loff_t pos, |
2e78c927 | 1481 | size_t write_bytes, |
376cc685 MX |
1482 | u64 *lockstart, u64 *lockend, |
1483 | struct extent_state **cached_state) | |
1484 | { | |
2cff578c | 1485 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
376cc685 MX |
1486 | u64 start_pos; |
1487 | u64 last_pos; | |
1488 | int i; | |
1489 | int ret = 0; | |
1490 | ||
0b246afa | 1491 | start_pos = round_down(pos, fs_info->sectorsize); |
2e78c927 | 1492 | last_pos = start_pos |
da17066c | 1493 | + round_up(pos + write_bytes - start_pos, |
0b246afa | 1494 | fs_info->sectorsize) - 1; |
376cc685 | 1495 | |
e3b8a485 | 1496 | if (start_pos < inode->vfs_inode.i_size) { |
e6dcd2dc | 1497 | struct btrfs_ordered_extent *ordered; |
a7e3b975 | 1498 | |
2cff578c NB |
1499 | lock_extent_bits(&inode->io_tree, start_pos, last_pos, |
1500 | cached_state); | |
b88935bf MX |
1501 | ordered = btrfs_lookup_ordered_range(inode, start_pos, |
1502 | last_pos - start_pos + 1); | |
e6dcd2dc CM |
1503 | if (ordered && |
1504 | ordered->file_offset + ordered->len > start_pos && | |
376cc685 | 1505 | ordered->file_offset <= last_pos) { |
2cff578c | 1506 | unlock_extent_cached(&inode->io_tree, start_pos, |
e43bbe5e | 1507 | last_pos, cached_state); |
e6dcd2dc CM |
1508 | for (i = 0; i < num_pages; i++) { |
1509 | unlock_page(pages[i]); | |
09cbfeaf | 1510 | put_page(pages[i]); |
e6dcd2dc | 1511 | } |
2cff578c NB |
1512 | btrfs_start_ordered_extent(&inode->vfs_inode, |
1513 | ordered, 1); | |
b88935bf MX |
1514 | btrfs_put_ordered_extent(ordered); |
1515 | return -EAGAIN; | |
e6dcd2dc CM |
1516 | } |
1517 | if (ordered) | |
1518 | btrfs_put_ordered_extent(ordered); | |
e3b8a485 FM |
1519 | clear_extent_bit(&inode->io_tree, start_pos, last_pos, |
1520 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
1521 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
ae0f1625 | 1522 | 0, 0, cached_state); |
376cc685 MX |
1523 | *lockstart = start_pos; |
1524 | *lockend = last_pos; | |
1525 | ret = 1; | |
0762704b | 1526 | } |
376cc685 | 1527 | |
e6dcd2dc | 1528 | for (i = 0; i < num_pages; i++) { |
32c7f202 WF |
1529 | if (clear_page_dirty_for_io(pages[i])) |
1530 | account_page_redirty(pages[i]); | |
e6dcd2dc CM |
1531 | set_page_extent_mapped(pages[i]); |
1532 | WARN_ON(!PageLocked(pages[i])); | |
1533 | } | |
b1bf862e | 1534 | |
376cc685 | 1535 | return ret; |
39279cc3 CM |
1536 | } |
1537 | ||
85b7ab67 | 1538 | static noinline int check_can_nocow(struct btrfs_inode *inode, loff_t pos, |
7ee9e440 JB |
1539 | size_t *write_bytes) |
1540 | { | |
85b7ab67 NB |
1541 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
1542 | struct btrfs_root *root = inode->root; | |
7ee9e440 JB |
1543 | struct btrfs_ordered_extent *ordered; |
1544 | u64 lockstart, lockend; | |
1545 | u64 num_bytes; | |
1546 | int ret; | |
1547 | ||
ea14b57f | 1548 | ret = btrfs_start_write_no_snapshotting(root); |
8257b2dc MX |
1549 | if (!ret) |
1550 | return -ENOSPC; | |
1551 | ||
0b246afa | 1552 | lockstart = round_down(pos, fs_info->sectorsize); |
da17066c | 1553 | lockend = round_up(pos + *write_bytes, |
0b246afa | 1554 | fs_info->sectorsize) - 1; |
7ee9e440 JB |
1555 | |
1556 | while (1) { | |
85b7ab67 | 1557 | lock_extent(&inode->io_tree, lockstart, lockend); |
7ee9e440 JB |
1558 | ordered = btrfs_lookup_ordered_range(inode, lockstart, |
1559 | lockend - lockstart + 1); | |
1560 | if (!ordered) { | |
1561 | break; | |
1562 | } | |
85b7ab67 NB |
1563 | unlock_extent(&inode->io_tree, lockstart, lockend); |
1564 | btrfs_start_ordered_extent(&inode->vfs_inode, ordered, 1); | |
7ee9e440 JB |
1565 | btrfs_put_ordered_extent(ordered); |
1566 | } | |
1567 | ||
7ee9e440 | 1568 | num_bytes = lockend - lockstart + 1; |
85b7ab67 NB |
1569 | ret = can_nocow_extent(&inode->vfs_inode, lockstart, &num_bytes, |
1570 | NULL, NULL, NULL); | |
7ee9e440 JB |
1571 | if (ret <= 0) { |
1572 | ret = 0; | |
ea14b57f | 1573 | btrfs_end_write_no_snapshotting(root); |
7ee9e440 | 1574 | } else { |
c933956d MX |
1575 | *write_bytes = min_t(size_t, *write_bytes , |
1576 | num_bytes - pos + lockstart); | |
7ee9e440 JB |
1577 | } |
1578 | ||
85b7ab67 | 1579 | unlock_extent(&inode->io_tree, lockstart, lockend); |
7ee9e440 JB |
1580 | |
1581 | return ret; | |
1582 | } | |
1583 | ||
d0215f3e JB |
1584 | static noinline ssize_t __btrfs_buffered_write(struct file *file, |
1585 | struct iov_iter *i, | |
1586 | loff_t pos) | |
4b46fce2 | 1587 | { |
496ad9aa | 1588 | struct inode *inode = file_inode(file); |
0b246afa | 1589 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
11c65dcc | 1590 | struct btrfs_root *root = BTRFS_I(inode)->root; |
11c65dcc | 1591 | struct page **pages = NULL; |
376cc685 | 1592 | struct extent_state *cached_state = NULL; |
364ecf36 | 1593 | struct extent_changeset *data_reserved = NULL; |
7ee9e440 | 1594 | u64 release_bytes = 0; |
376cc685 MX |
1595 | u64 lockstart; |
1596 | u64 lockend; | |
d0215f3e JB |
1597 | size_t num_written = 0; |
1598 | int nrptrs; | |
c9149235 | 1599 | int ret = 0; |
7ee9e440 | 1600 | bool only_release_metadata = false; |
b6316429 | 1601 | bool force_page_uptodate = false; |
4b46fce2 | 1602 | |
09cbfeaf KS |
1603 | nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_SIZE), |
1604 | PAGE_SIZE / (sizeof(struct page *))); | |
142349f5 WF |
1605 | nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied); |
1606 | nrptrs = max(nrptrs, 8); | |
31e818fe | 1607 | pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL); |
d0215f3e JB |
1608 | if (!pages) |
1609 | return -ENOMEM; | |
ab93dbec | 1610 | |
d0215f3e | 1611 | while (iov_iter_count(i) > 0) { |
09cbfeaf | 1612 | size_t offset = pos & (PAGE_SIZE - 1); |
2e78c927 | 1613 | size_t sector_offset; |
d0215f3e | 1614 | size_t write_bytes = min(iov_iter_count(i), |
09cbfeaf | 1615 | nrptrs * (size_t)PAGE_SIZE - |
8c2383c3 | 1616 | offset); |
ed6078f7 | 1617 | size_t num_pages = DIV_ROUND_UP(write_bytes + offset, |
09cbfeaf | 1618 | PAGE_SIZE); |
7ee9e440 | 1619 | size_t reserve_bytes; |
d0215f3e JB |
1620 | size_t dirty_pages; |
1621 | size_t copied; | |
2e78c927 CR |
1622 | size_t dirty_sectors; |
1623 | size_t num_sectors; | |
79f015f2 | 1624 | int extents_locked; |
39279cc3 | 1625 | |
8c2383c3 | 1626 | WARN_ON(num_pages > nrptrs); |
1832a6d5 | 1627 | |
914ee295 XZ |
1628 | /* |
1629 | * Fault pages before locking them in prepare_pages | |
1630 | * to avoid recursive lock | |
1631 | */ | |
d0215f3e | 1632 | if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) { |
914ee295 | 1633 | ret = -EFAULT; |
d0215f3e | 1634 | break; |
914ee295 XZ |
1635 | } |
1636 | ||
da17066c | 1637 | sector_offset = pos & (fs_info->sectorsize - 1); |
2e78c927 | 1638 | reserve_bytes = round_up(write_bytes + sector_offset, |
da17066c | 1639 | fs_info->sectorsize); |
d9d8b2a5 | 1640 | |
364ecf36 QW |
1641 | extent_changeset_release(data_reserved); |
1642 | ret = btrfs_check_data_free_space(inode, &data_reserved, pos, | |
1643 | write_bytes); | |
c6887cd1 JB |
1644 | if (ret < 0) { |
1645 | if ((BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW | | |
1646 | BTRFS_INODE_PREALLOC)) && | |
85b7ab67 NB |
1647 | check_can_nocow(BTRFS_I(inode), pos, |
1648 | &write_bytes) > 0) { | |
c6887cd1 JB |
1649 | /* |
1650 | * For nodata cow case, no need to reserve | |
1651 | * data space. | |
1652 | */ | |
1653 | only_release_metadata = true; | |
1654 | /* | |
1655 | * our prealloc extent may be smaller than | |
1656 | * write_bytes, so scale down. | |
1657 | */ | |
1658 | num_pages = DIV_ROUND_UP(write_bytes + offset, | |
1659 | PAGE_SIZE); | |
1660 | reserve_bytes = round_up(write_bytes + | |
1661 | sector_offset, | |
da17066c | 1662 | fs_info->sectorsize); |
c6887cd1 JB |
1663 | } else { |
1664 | break; | |
1665 | } | |
1666 | } | |
1832a6d5 | 1667 | |
8b62f87b | 1668 | WARN_ON(reserve_bytes == 0); |
9f3db423 NB |
1669 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), |
1670 | reserve_bytes); | |
7ee9e440 JB |
1671 | if (ret) { |
1672 | if (!only_release_metadata) | |
bc42bda2 QW |
1673 | btrfs_free_reserved_data_space(inode, |
1674 | data_reserved, pos, | |
1675 | write_bytes); | |
8257b2dc | 1676 | else |
ea14b57f | 1677 | btrfs_end_write_no_snapshotting(root); |
7ee9e440 JB |
1678 | break; |
1679 | } | |
1680 | ||
1681 | release_bytes = reserve_bytes; | |
376cc685 | 1682 | again: |
4a64001f JB |
1683 | /* |
1684 | * This is going to setup the pages array with the number of | |
1685 | * pages we want, so we don't really need to worry about the | |
1686 | * contents of pages from loop to loop | |
1687 | */ | |
b37392ea MX |
1688 | ret = prepare_pages(inode, pages, num_pages, |
1689 | pos, write_bytes, | |
b6316429 | 1690 | force_page_uptodate); |
8b62f87b JB |
1691 | if (ret) { |
1692 | btrfs_delalloc_release_extents(BTRFS_I(inode), | |
1693 | reserve_bytes); | |
d0215f3e | 1694 | break; |
8b62f87b | 1695 | } |
39279cc3 | 1696 | |
79f015f2 GR |
1697 | extents_locked = lock_and_cleanup_extent_if_need( |
1698 | BTRFS_I(inode), pages, | |
2cff578c NB |
1699 | num_pages, pos, write_bytes, &lockstart, |
1700 | &lockend, &cached_state); | |
79f015f2 GR |
1701 | if (extents_locked < 0) { |
1702 | if (extents_locked == -EAGAIN) | |
376cc685 | 1703 | goto again; |
8b62f87b JB |
1704 | btrfs_delalloc_release_extents(BTRFS_I(inode), |
1705 | reserve_bytes); | |
79f015f2 | 1706 | ret = extents_locked; |
376cc685 | 1707 | break; |
376cc685 MX |
1708 | } |
1709 | ||
ee22f0c4 | 1710 | copied = btrfs_copy_from_user(pos, write_bytes, pages, i); |
b1bf862e | 1711 | |
0b246afa | 1712 | num_sectors = BTRFS_BYTES_TO_BLKS(fs_info, reserve_bytes); |
56244ef1 | 1713 | dirty_sectors = round_up(copied + sector_offset, |
0b246afa JM |
1714 | fs_info->sectorsize); |
1715 | dirty_sectors = BTRFS_BYTES_TO_BLKS(fs_info, dirty_sectors); | |
56244ef1 | 1716 | |
b1bf862e CM |
1717 | /* |
1718 | * if we have trouble faulting in the pages, fall | |
1719 | * back to one page at a time | |
1720 | */ | |
1721 | if (copied < write_bytes) | |
1722 | nrptrs = 1; | |
1723 | ||
b6316429 JB |
1724 | if (copied == 0) { |
1725 | force_page_uptodate = true; | |
56244ef1 | 1726 | dirty_sectors = 0; |
b1bf862e | 1727 | dirty_pages = 0; |
b6316429 JB |
1728 | } else { |
1729 | force_page_uptodate = false; | |
ed6078f7 | 1730 | dirty_pages = DIV_ROUND_UP(copied + offset, |
09cbfeaf | 1731 | PAGE_SIZE); |
b6316429 | 1732 | } |
914ee295 | 1733 | |
2e78c927 | 1734 | if (num_sectors > dirty_sectors) { |
8b8b08cb CM |
1735 | /* release everything except the sectors we dirtied */ |
1736 | release_bytes -= dirty_sectors << | |
0b246afa | 1737 | fs_info->sb->s_blocksize_bits; |
485290a7 | 1738 | if (only_release_metadata) { |
691fa059 | 1739 | btrfs_delalloc_release_metadata(BTRFS_I(inode), |
7ee9e440 | 1740 | release_bytes); |
485290a7 QW |
1741 | } else { |
1742 | u64 __pos; | |
1743 | ||
da17066c | 1744 | __pos = round_down(pos, |
0b246afa | 1745 | fs_info->sectorsize) + |
09cbfeaf | 1746 | (dirty_pages << PAGE_SHIFT); |
bc42bda2 QW |
1747 | btrfs_delalloc_release_space(inode, |
1748 | data_reserved, __pos, | |
1749 | release_bytes); | |
485290a7 | 1750 | } |
914ee295 XZ |
1751 | } |
1752 | ||
2e78c927 | 1753 | release_bytes = round_up(copied + sector_offset, |
0b246afa | 1754 | fs_info->sectorsize); |
376cc685 MX |
1755 | |
1756 | if (copied > 0) | |
2ff7e61e JM |
1757 | ret = btrfs_dirty_pages(inode, pages, dirty_pages, |
1758 | pos, copied, NULL); | |
79f015f2 | 1759 | if (extents_locked) |
376cc685 | 1760 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, |
e43bbe5e | 1761 | lockstart, lockend, &cached_state); |
8b62f87b | 1762 | btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes); |
f1de9683 MX |
1763 | if (ret) { |
1764 | btrfs_drop_pages(pages, num_pages); | |
376cc685 | 1765 | break; |
f1de9683 | 1766 | } |
39279cc3 | 1767 | |
376cc685 | 1768 | release_bytes = 0; |
8257b2dc | 1769 | if (only_release_metadata) |
ea14b57f | 1770 | btrfs_end_write_no_snapshotting(root); |
8257b2dc | 1771 | |
7ee9e440 | 1772 | if (only_release_metadata && copied > 0) { |
da17066c | 1773 | lockstart = round_down(pos, |
0b246afa | 1774 | fs_info->sectorsize); |
da17066c | 1775 | lockend = round_up(pos + copied, |
0b246afa | 1776 | fs_info->sectorsize) - 1; |
7ee9e440 JB |
1777 | |
1778 | set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
1779 | lockend, EXTENT_NORESERVE, NULL, | |
1780 | NULL, GFP_NOFS); | |
1781 | only_release_metadata = false; | |
1782 | } | |
1783 | ||
f1de9683 MX |
1784 | btrfs_drop_pages(pages, num_pages); |
1785 | ||
d0215f3e JB |
1786 | cond_resched(); |
1787 | ||
d0e1d66b | 1788 | balance_dirty_pages_ratelimited(inode->i_mapping); |
0b246afa | 1789 | if (dirty_pages < (fs_info->nodesize >> PAGE_SHIFT) + 1) |
2ff7e61e | 1790 | btrfs_btree_balance_dirty(fs_info); |
cb843a6f | 1791 | |
914ee295 XZ |
1792 | pos += copied; |
1793 | num_written += copied; | |
d0215f3e | 1794 | } |
39279cc3 | 1795 | |
d0215f3e JB |
1796 | kfree(pages); |
1797 | ||
7ee9e440 | 1798 | if (release_bytes) { |
8257b2dc | 1799 | if (only_release_metadata) { |
ea14b57f | 1800 | btrfs_end_write_no_snapshotting(root); |
691fa059 NB |
1801 | btrfs_delalloc_release_metadata(BTRFS_I(inode), |
1802 | release_bytes); | |
8257b2dc | 1803 | } else { |
bc42bda2 QW |
1804 | btrfs_delalloc_release_space(inode, data_reserved, |
1805 | round_down(pos, fs_info->sectorsize), | |
1806 | release_bytes); | |
8257b2dc | 1807 | } |
7ee9e440 JB |
1808 | } |
1809 | ||
364ecf36 | 1810 | extent_changeset_free(data_reserved); |
d0215f3e JB |
1811 | return num_written ? num_written : ret; |
1812 | } | |
1813 | ||
1af5bb49 | 1814 | static ssize_t __btrfs_direct_write(struct kiocb *iocb, struct iov_iter *from) |
d0215f3e JB |
1815 | { |
1816 | struct file *file = iocb->ki_filp; | |
728404da | 1817 | struct inode *inode = file_inode(file); |
1af5bb49 | 1818 | loff_t pos = iocb->ki_pos; |
d0215f3e JB |
1819 | ssize_t written; |
1820 | ssize_t written_buffered; | |
1821 | loff_t endbyte; | |
1822 | int err; | |
1823 | ||
1af5bb49 | 1824 | written = generic_file_direct_write(iocb, from); |
d0215f3e | 1825 | |
0c949334 | 1826 | if (written < 0 || !iov_iter_count(from)) |
d0215f3e JB |
1827 | return written; |
1828 | ||
1829 | pos += written; | |
0ae5e4d3 | 1830 | written_buffered = __btrfs_buffered_write(file, from, pos); |
d0215f3e JB |
1831 | if (written_buffered < 0) { |
1832 | err = written_buffered; | |
1833 | goto out; | |
39279cc3 | 1834 | } |
075bdbdb FM |
1835 | /* |
1836 | * Ensure all data is persisted. We want the next direct IO read to be | |
1837 | * able to read what was just written. | |
1838 | */ | |
d0215f3e | 1839 | endbyte = pos + written_buffered - 1; |
728404da | 1840 | err = btrfs_fdatawrite_range(inode, pos, endbyte); |
075bdbdb FM |
1841 | if (err) |
1842 | goto out; | |
728404da | 1843 | err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte); |
d0215f3e JB |
1844 | if (err) |
1845 | goto out; | |
1846 | written += written_buffered; | |
867c4f93 | 1847 | iocb->ki_pos = pos + written_buffered; |
09cbfeaf KS |
1848 | invalidate_mapping_pages(file->f_mapping, pos >> PAGE_SHIFT, |
1849 | endbyte >> PAGE_SHIFT); | |
39279cc3 | 1850 | out: |
d0215f3e JB |
1851 | return written ? written : err; |
1852 | } | |
5b92ee72 | 1853 | |
6c760c07 JB |
1854 | static void update_time_for_write(struct inode *inode) |
1855 | { | |
1856 | struct timespec now; | |
1857 | ||
1858 | if (IS_NOCMTIME(inode)) | |
1859 | return; | |
1860 | ||
c2050a45 | 1861 | now = current_time(inode); |
6c760c07 JB |
1862 | if (!timespec_equal(&inode->i_mtime, &now)) |
1863 | inode->i_mtime = now; | |
1864 | ||
1865 | if (!timespec_equal(&inode->i_ctime, &now)) | |
1866 | inode->i_ctime = now; | |
1867 | ||
1868 | if (IS_I_VERSION(inode)) | |
1869 | inode_inc_iversion(inode); | |
1870 | } | |
1871 | ||
b30ac0fc AV |
1872 | static ssize_t btrfs_file_write_iter(struct kiocb *iocb, |
1873 | struct iov_iter *from) | |
d0215f3e JB |
1874 | { |
1875 | struct file *file = iocb->ki_filp; | |
496ad9aa | 1876 | struct inode *inode = file_inode(file); |
0b246afa | 1877 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
d0215f3e | 1878 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0c1a98c8 | 1879 | u64 start_pos; |
3ac0d7b9 | 1880 | u64 end_pos; |
d0215f3e | 1881 | ssize_t num_written = 0; |
b812ce28 | 1882 | bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host); |
3309dd04 | 1883 | ssize_t err; |
ff0fa732 | 1884 | loff_t pos; |
edf064e7 | 1885 | size_t count = iov_iter_count(from); |
27772b68 CR |
1886 | loff_t oldsize; |
1887 | int clean_page = 0; | |
d0215f3e | 1888 | |
91f9943e CH |
1889 | if (!(iocb->ki_flags & IOCB_DIRECT) && |
1890 | (iocb->ki_flags & IOCB_NOWAIT)) | |
1891 | return -EOPNOTSUPP; | |
1892 | ||
ff0fa732 GR |
1893 | if (!inode_trylock(inode)) { |
1894 | if (iocb->ki_flags & IOCB_NOWAIT) | |
edf064e7 | 1895 | return -EAGAIN; |
ff0fa732 GR |
1896 | inode_lock(inode); |
1897 | } | |
1898 | ||
1899 | err = generic_write_checks(iocb, from); | |
1900 | if (err <= 0) { | |
1901 | inode_unlock(inode); | |
1902 | return err; | |
1903 | } | |
1904 | ||
1905 | pos = iocb->ki_pos; | |
1906 | if (iocb->ki_flags & IOCB_NOWAIT) { | |
edf064e7 GR |
1907 | /* |
1908 | * We will allocate space in case nodatacow is not set, | |
1909 | * so bail | |
1910 | */ | |
1911 | if (!(BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW | | |
1912 | BTRFS_INODE_PREALLOC)) || | |
1913 | check_can_nocow(BTRFS_I(inode), pos, &count) <= 0) { | |
1914 | inode_unlock(inode); | |
1915 | return -EAGAIN; | |
1916 | } | |
d0215f3e JB |
1917 | } |
1918 | ||
3309dd04 | 1919 | current->backing_dev_info = inode_to_bdi(inode); |
5fa8e0a1 | 1920 | err = file_remove_privs(file); |
d0215f3e | 1921 | if (err) { |
5955102c | 1922 | inode_unlock(inode); |
d0215f3e JB |
1923 | goto out; |
1924 | } | |
1925 | ||
1926 | /* | |
1927 | * If BTRFS flips readonly due to some impossible error | |
1928 | * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR), | |
1929 | * although we have opened a file as writable, we have | |
1930 | * to stop this write operation to ensure FS consistency. | |
1931 | */ | |
0b246afa | 1932 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
5955102c | 1933 | inode_unlock(inode); |
d0215f3e JB |
1934 | err = -EROFS; |
1935 | goto out; | |
1936 | } | |
1937 | ||
6c760c07 JB |
1938 | /* |
1939 | * We reserve space for updating the inode when we reserve space for the | |
1940 | * extent we are going to write, so we will enospc out there. We don't | |
1941 | * need to start yet another transaction to update the inode as we will | |
1942 | * update the inode when we finish writing whatever data we write. | |
1943 | */ | |
1944 | update_time_for_write(inode); | |
d0215f3e | 1945 | |
0b246afa | 1946 | start_pos = round_down(pos, fs_info->sectorsize); |
27772b68 CR |
1947 | oldsize = i_size_read(inode); |
1948 | if (start_pos > oldsize) { | |
3ac0d7b9 | 1949 | /* Expand hole size to cover write data, preventing empty gap */ |
da17066c | 1950 | end_pos = round_up(pos + count, |
0b246afa | 1951 | fs_info->sectorsize); |
27772b68 | 1952 | err = btrfs_cont_expand(inode, oldsize, end_pos); |
0c1a98c8 | 1953 | if (err) { |
5955102c | 1954 | inode_unlock(inode); |
0c1a98c8 MX |
1955 | goto out; |
1956 | } | |
0b246afa | 1957 | if (start_pos > round_up(oldsize, fs_info->sectorsize)) |
27772b68 | 1958 | clean_page = 1; |
0c1a98c8 MX |
1959 | } |
1960 | ||
b812ce28 JB |
1961 | if (sync) |
1962 | atomic_inc(&BTRFS_I(inode)->sync_writers); | |
1963 | ||
2ba48ce5 | 1964 | if (iocb->ki_flags & IOCB_DIRECT) { |
1af5bb49 | 1965 | num_written = __btrfs_direct_write(iocb, from); |
d0215f3e | 1966 | } else { |
b30ac0fc | 1967 | num_written = __btrfs_buffered_write(file, from, pos); |
d0215f3e | 1968 | if (num_written > 0) |
867c4f93 | 1969 | iocb->ki_pos = pos + num_written; |
27772b68 CR |
1970 | if (clean_page) |
1971 | pagecache_isize_extended(inode, oldsize, | |
1972 | i_size_read(inode)); | |
d0215f3e JB |
1973 | } |
1974 | ||
5955102c | 1975 | inode_unlock(inode); |
2ff3e9b6 | 1976 | |
5a3f23d5 | 1977 | /* |
6c760c07 JB |
1978 | * We also have to set last_sub_trans to the current log transid, |
1979 | * otherwise subsequent syncs to a file that's been synced in this | |
bb7ab3b9 | 1980 | * transaction will appear to have already occurred. |
5a3f23d5 | 1981 | */ |
2f2ff0ee | 1982 | spin_lock(&BTRFS_I(inode)->lock); |
6c760c07 | 1983 | BTRFS_I(inode)->last_sub_trans = root->log_transid; |
2f2ff0ee | 1984 | spin_unlock(&BTRFS_I(inode)->lock); |
e2592217 CH |
1985 | if (num_written > 0) |
1986 | num_written = generic_write_sync(iocb, num_written); | |
0a3404dc | 1987 | |
b812ce28 JB |
1988 | if (sync) |
1989 | atomic_dec(&BTRFS_I(inode)->sync_writers); | |
0a3404dc | 1990 | out: |
39279cc3 | 1991 | current->backing_dev_info = NULL; |
39279cc3 CM |
1992 | return num_written ? num_written : err; |
1993 | } | |
1994 | ||
d397712b | 1995 | int btrfs_release_file(struct inode *inode, struct file *filp) |
e1b81e67 | 1996 | { |
23b5ec74 JB |
1997 | struct btrfs_file_private *private = filp->private_data; |
1998 | ||
1999 | if (private && private->trans) | |
6bf13c0c | 2000 | btrfs_ioctl_trans_end(filp); |
23b5ec74 JB |
2001 | if (private && private->filldir_buf) |
2002 | kfree(private->filldir_buf); | |
2003 | kfree(private); | |
2004 | filp->private_data = NULL; | |
2005 | ||
f6dc45c7 CM |
2006 | /* |
2007 | * ordered_data_close is set by settattr when we are about to truncate | |
2008 | * a file from a non-zero size to a zero size. This tries to | |
2009 | * flush down new bytes that may have been written if the | |
2010 | * application were using truncate to replace a file in place. | |
2011 | */ | |
2012 | if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, | |
2013 | &BTRFS_I(inode)->runtime_flags)) | |
2014 | filemap_flush(inode->i_mapping); | |
e1b81e67 M |
2015 | return 0; |
2016 | } | |
2017 | ||
669249ee FM |
2018 | static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end) |
2019 | { | |
2020 | int ret; | |
343e4fc1 | 2021 | struct blk_plug plug; |
669249ee | 2022 | |
343e4fc1 LB |
2023 | /* |
2024 | * This is only called in fsync, which would do synchronous writes, so | |
2025 | * a plug can merge adjacent IOs as much as possible. Esp. in case of | |
2026 | * multiple disks using raid profile, a large IO can be split to | |
2027 | * several segments of stripe length (currently 64K). | |
2028 | */ | |
2029 | blk_start_plug(&plug); | |
669249ee | 2030 | atomic_inc(&BTRFS_I(inode)->sync_writers); |
728404da | 2031 | ret = btrfs_fdatawrite_range(inode, start, end); |
669249ee | 2032 | atomic_dec(&BTRFS_I(inode)->sync_writers); |
343e4fc1 | 2033 | blk_finish_plug(&plug); |
669249ee FM |
2034 | |
2035 | return ret; | |
2036 | } | |
2037 | ||
d352ac68 CM |
2038 | /* |
2039 | * fsync call for both files and directories. This logs the inode into | |
2040 | * the tree log instead of forcing full commits whenever possible. | |
2041 | * | |
2042 | * It needs to call filemap_fdatawait so that all ordered extent updates are | |
2043 | * in the metadata btree are up to date for copying to the log. | |
2044 | * | |
2045 | * It drops the inode mutex before doing the tree log commit. This is an | |
2046 | * important optimization for directories because holding the mutex prevents | |
2047 | * new operations on the dir while we write to disk. | |
2048 | */ | |
02c24a82 | 2049 | int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync) |
39279cc3 | 2050 | { |
de17e793 | 2051 | struct dentry *dentry = file_dentry(file); |
2b0143b5 | 2052 | struct inode *inode = d_inode(dentry); |
0b246afa | 2053 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 2054 | struct btrfs_root *root = BTRFS_I(inode)->root; |
39279cc3 | 2055 | struct btrfs_trans_handle *trans; |
8b050d35 | 2056 | struct btrfs_log_ctx ctx; |
333427a5 | 2057 | int ret = 0, err; |
897ca819 | 2058 | bool full_sync = false; |
9dcbeed4 | 2059 | u64 len; |
39279cc3 | 2060 | |
9dcbeed4 DS |
2061 | /* |
2062 | * The range length can be represented by u64, we have to do the typecasts | |
2063 | * to avoid signed overflow if it's [0, LLONG_MAX] eg. from fsync() | |
2064 | */ | |
2065 | len = (u64)end - (u64)start + 1; | |
1abe9b8a | 2066 | trace_btrfs_sync_file(file, datasync); |
257c62e1 | 2067 | |
ebb70442 LB |
2068 | btrfs_init_log_ctx(&ctx, inode); |
2069 | ||
90abccf2 MX |
2070 | /* |
2071 | * We write the dirty pages in the range and wait until they complete | |
2072 | * out of the ->i_mutex. If so, we can flush the dirty pages by | |
2ab28f32 JB |
2073 | * multi-task, and make the performance up. See |
2074 | * btrfs_wait_ordered_range for an explanation of the ASYNC check. | |
90abccf2 | 2075 | */ |
669249ee | 2076 | ret = start_ordered_ops(inode, start, end); |
90abccf2 | 2077 | if (ret) |
333427a5 | 2078 | goto out; |
90abccf2 | 2079 | |
5955102c | 2080 | inode_lock(inode); |
2ecb7923 | 2081 | atomic_inc(&root->log_batch); |
2ab28f32 JB |
2082 | full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
2083 | &BTRFS_I(inode)->runtime_flags); | |
669249ee FM |
2084 | /* |
2085 | * We might have have had more pages made dirty after calling | |
2086 | * start_ordered_ops and before acquiring the inode's i_mutex. | |
2087 | */ | |
0ef8b726 | 2088 | if (full_sync) { |
669249ee FM |
2089 | /* |
2090 | * For a full sync, we need to make sure any ordered operations | |
2091 | * start and finish before we start logging the inode, so that | |
2092 | * all extents are persisted and the respective file extent | |
2093 | * items are in the fs/subvol btree. | |
2094 | */ | |
b659ef02 | 2095 | ret = btrfs_wait_ordered_range(inode, start, len); |
669249ee FM |
2096 | } else { |
2097 | /* | |
2098 | * Start any new ordered operations before starting to log the | |
2099 | * inode. We will wait for them to finish in btrfs_sync_log(). | |
2100 | * | |
2101 | * Right before acquiring the inode's mutex, we might have new | |
2102 | * writes dirtying pages, which won't immediately start the | |
2103 | * respective ordered operations - that is done through the | |
2104 | * fill_delalloc callbacks invoked from the writepage and | |
2105 | * writepages address space operations. So make sure we start | |
2106 | * all ordered operations before starting to log our inode. Not | |
2107 | * doing this means that while logging the inode, writeback | |
2108 | * could start and invoke writepage/writepages, which would call | |
2109 | * the fill_delalloc callbacks (cow_file_range, | |
2110 | * submit_compressed_extents). These callbacks add first an | |
2111 | * extent map to the modified list of extents and then create | |
2112 | * the respective ordered operation, which means in | |
2113 | * tree-log.c:btrfs_log_inode() we might capture all existing | |
2114 | * ordered operations (with btrfs_get_logged_extents()) before | |
2115 | * the fill_delalloc callback adds its ordered operation, and by | |
2116 | * the time we visit the modified list of extent maps (with | |
2117 | * btrfs_log_changed_extents()), we see and process the extent | |
2118 | * map they created. We then use the extent map to construct a | |
2119 | * file extent item for logging without waiting for the | |
2120 | * respective ordered operation to finish - this file extent | |
2121 | * item points to a disk location that might not have yet been | |
2122 | * written to, containing random data - so after a crash a log | |
2123 | * replay will make our inode have file extent items that point | |
2124 | * to disk locations containing invalid data, as we returned | |
2125 | * success to userspace without waiting for the respective | |
2126 | * ordered operation to finish, because it wasn't captured by | |
2127 | * btrfs_get_logged_extents(). | |
2128 | */ | |
2129 | ret = start_ordered_ops(inode, start, end); | |
2130 | } | |
2131 | if (ret) { | |
5955102c | 2132 | inode_unlock(inode); |
669249ee | 2133 | goto out; |
0ef8b726 | 2134 | } |
2ecb7923 | 2135 | atomic_inc(&root->log_batch); |
257c62e1 | 2136 | |
39279cc3 | 2137 | /* |
3a8b36f3 FM |
2138 | * If the last transaction that changed this file was before the current |
2139 | * transaction and we have the full sync flag set in our inode, we can | |
2140 | * bail out now without any syncing. | |
2141 | * | |
2142 | * Note that we can't bail out if the full sync flag isn't set. This is | |
2143 | * because when the full sync flag is set we start all ordered extents | |
2144 | * and wait for them to fully complete - when they complete they update | |
2145 | * the inode's last_trans field through: | |
2146 | * | |
2147 | * btrfs_finish_ordered_io() -> | |
2148 | * btrfs_update_inode_fallback() -> | |
2149 | * btrfs_update_inode() -> | |
2150 | * btrfs_set_inode_last_trans() | |
2151 | * | |
2152 | * So we are sure that last_trans is up to date and can do this check to | |
2153 | * bail out safely. For the fast path, when the full sync flag is not | |
2154 | * set in our inode, we can not do it because we start only our ordered | |
2155 | * extents and don't wait for them to complete (that is when | |
2156 | * btrfs_finish_ordered_io runs), so here at this point their last_trans | |
2157 | * value might be less than or equals to fs_info->last_trans_committed, | |
2158 | * and setting a speculative last_trans for an inode when a buffered | |
2159 | * write is made (such as fs_info->generation + 1 for example) would not | |
2160 | * be reliable since after setting the value and before fsync is called | |
2161 | * any number of transactions can start and commit (transaction kthread | |
2162 | * commits the current transaction periodically), and a transaction | |
2163 | * commit does not start nor waits for ordered extents to complete. | |
257c62e1 | 2164 | */ |
a4abeea4 | 2165 | smp_mb(); |
0f8939b8 | 2166 | if (btrfs_inode_in_log(BTRFS_I(inode), fs_info->generation) || |
affc0ff9 | 2167 | (full_sync && BTRFS_I(inode)->last_trans <= |
0b246afa | 2168 | fs_info->last_trans_committed) || |
affc0ff9 FM |
2169 | (!btrfs_have_ordered_extents_in_range(inode, start, len) && |
2170 | BTRFS_I(inode)->last_trans | |
0b246afa | 2171 | <= fs_info->last_trans_committed)) { |
5dc562c5 | 2172 | /* |
01327610 | 2173 | * We've had everything committed since the last time we were |
5dc562c5 JB |
2174 | * modified so clear this flag in case it was set for whatever |
2175 | * reason, it's no longer relevant. | |
2176 | */ | |
2177 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
2178 | &BTRFS_I(inode)->runtime_flags); | |
0596a904 FM |
2179 | /* |
2180 | * An ordered extent might have started before and completed | |
2181 | * already with io errors, in which case the inode was not | |
2182 | * updated and we end up here. So check the inode's mapping | |
333427a5 JL |
2183 | * for any errors that might have happened since we last |
2184 | * checked called fsync. | |
0596a904 | 2185 | */ |
333427a5 | 2186 | ret = filemap_check_wb_err(inode->i_mapping, file->f_wb_err); |
5955102c | 2187 | inode_unlock(inode); |
15ee9bc7 JB |
2188 | goto out; |
2189 | } | |
15ee9bc7 JB |
2190 | |
2191 | /* | |
a52d9a80 CM |
2192 | * ok we haven't committed the transaction yet, lets do a commit |
2193 | */ | |
6f902af4 | 2194 | if (file->private_data) |
6bf13c0c SW |
2195 | btrfs_ioctl_trans_end(file); |
2196 | ||
5039eddc JB |
2197 | /* |
2198 | * We use start here because we will need to wait on the IO to complete | |
2199 | * in btrfs_sync_log, which could require joining a transaction (for | |
2200 | * example checking cross references in the nocow path). If we use join | |
2201 | * here we could get into a situation where we're waiting on IO to | |
2202 | * happen that is blocked on a transaction trying to commit. With start | |
2203 | * we inc the extwriter counter, so we wait for all extwriters to exit | |
2204 | * before we start blocking join'ers. This comment is to keep somebody | |
2205 | * from thinking they are super smart and changing this to | |
2206 | * btrfs_join_transaction *cough*Josef*cough*. | |
2207 | */ | |
a22285a6 YZ |
2208 | trans = btrfs_start_transaction(root, 0); |
2209 | if (IS_ERR(trans)) { | |
2210 | ret = PTR_ERR(trans); | |
5955102c | 2211 | inode_unlock(inode); |
39279cc3 CM |
2212 | goto out; |
2213 | } | |
5039eddc | 2214 | trans->sync = true; |
e02119d5 | 2215 | |
49dae1bc | 2216 | ret = btrfs_log_dentry_safe(trans, root, dentry, start, end, &ctx); |
02c24a82 | 2217 | if (ret < 0) { |
a0634be5 FDBM |
2218 | /* Fallthrough and commit/free transaction. */ |
2219 | ret = 1; | |
02c24a82 | 2220 | } |
49eb7e46 CM |
2221 | |
2222 | /* we've logged all the items and now have a consistent | |
2223 | * version of the file in the log. It is possible that | |
2224 | * someone will come in and modify the file, but that's | |
2225 | * fine because the log is consistent on disk, and we | |
2226 | * have references to all of the file's extents | |
2227 | * | |
2228 | * It is possible that someone will come in and log the | |
2229 | * file again, but that will end up using the synchronization | |
2230 | * inside btrfs_sync_log to keep things safe. | |
2231 | */ | |
5955102c | 2232 | inode_unlock(inode); |
49eb7e46 | 2233 | |
8407f553 FM |
2234 | /* |
2235 | * If any of the ordered extents had an error, just return it to user | |
2236 | * space, so that the application knows some writes didn't succeed and | |
2237 | * can take proper action (retry for e.g.). Blindly committing the | |
2238 | * transaction in this case, would fool userspace that everything was | |
2239 | * successful. And we also want to make sure our log doesn't contain | |
2240 | * file extent items pointing to extents that weren't fully written to - | |
2241 | * just like in the non fast fsync path, where we check for the ordered | |
2242 | * operation's error flag before writing to the log tree and return -EIO | |
2243 | * if any of them had this flag set (btrfs_wait_ordered_range) - | |
2244 | * therefore we need to check for errors in the ordered operations, | |
2245 | * which are indicated by ctx.io_err. | |
2246 | */ | |
2247 | if (ctx.io_err) { | |
3a45bb20 | 2248 | btrfs_end_transaction(trans); |
8407f553 FM |
2249 | ret = ctx.io_err; |
2250 | goto out; | |
2251 | } | |
2252 | ||
257c62e1 | 2253 | if (ret != BTRFS_NO_LOG_SYNC) { |
0ef8b726 | 2254 | if (!ret) { |
8b050d35 | 2255 | ret = btrfs_sync_log(trans, root, &ctx); |
0ef8b726 | 2256 | if (!ret) { |
3a45bb20 | 2257 | ret = btrfs_end_transaction(trans); |
0ef8b726 | 2258 | goto out; |
2ab28f32 | 2259 | } |
257c62e1 | 2260 | } |
0ef8b726 | 2261 | if (!full_sync) { |
9dcbeed4 | 2262 | ret = btrfs_wait_ordered_range(inode, start, len); |
b05fd874 | 2263 | if (ret) { |
3a45bb20 | 2264 | btrfs_end_transaction(trans); |
0ef8b726 | 2265 | goto out; |
b05fd874 | 2266 | } |
0ef8b726 | 2267 | } |
3a45bb20 | 2268 | ret = btrfs_commit_transaction(trans); |
257c62e1 | 2269 | } else { |
3a45bb20 | 2270 | ret = btrfs_end_transaction(trans); |
e02119d5 | 2271 | } |
39279cc3 | 2272 | out: |
ebb70442 | 2273 | ASSERT(list_empty(&ctx.list)); |
333427a5 JL |
2274 | err = file_check_and_advance_wb_err(file); |
2275 | if (!ret) | |
2276 | ret = err; | |
014e4ac4 | 2277 | return ret > 0 ? -EIO : ret; |
39279cc3 CM |
2278 | } |
2279 | ||
f0f37e2f | 2280 | static const struct vm_operations_struct btrfs_file_vm_ops = { |
92fee66d | 2281 | .fault = filemap_fault, |
f1820361 | 2282 | .map_pages = filemap_map_pages, |
9ebefb18 CM |
2283 | .page_mkwrite = btrfs_page_mkwrite, |
2284 | }; | |
2285 | ||
2286 | static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma) | |
2287 | { | |
058a457e MX |
2288 | struct address_space *mapping = filp->f_mapping; |
2289 | ||
2290 | if (!mapping->a_ops->readpage) | |
2291 | return -ENOEXEC; | |
2292 | ||
9ebefb18 | 2293 | file_accessed(filp); |
058a457e | 2294 | vma->vm_ops = &btrfs_file_vm_ops; |
058a457e | 2295 | |
9ebefb18 CM |
2296 | return 0; |
2297 | } | |
2298 | ||
35339c24 | 2299 | static int hole_mergeable(struct btrfs_inode *inode, struct extent_buffer *leaf, |
2aaa6655 JB |
2300 | int slot, u64 start, u64 end) |
2301 | { | |
2302 | struct btrfs_file_extent_item *fi; | |
2303 | struct btrfs_key key; | |
2304 | ||
2305 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) | |
2306 | return 0; | |
2307 | ||
2308 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
35339c24 | 2309 | if (key.objectid != btrfs_ino(inode) || |
2aaa6655 JB |
2310 | key.type != BTRFS_EXTENT_DATA_KEY) |
2311 | return 0; | |
2312 | ||
2313 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
2314 | ||
2315 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2316 | return 0; | |
2317 | ||
2318 | if (btrfs_file_extent_disk_bytenr(leaf, fi)) | |
2319 | return 0; | |
2320 | ||
2321 | if (key.offset == end) | |
2322 | return 1; | |
2323 | if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start) | |
2324 | return 1; | |
2325 | return 0; | |
2326 | } | |
2327 | ||
a012a74e NB |
2328 | static int fill_holes(struct btrfs_trans_handle *trans, |
2329 | struct btrfs_inode *inode, | |
2330 | struct btrfs_path *path, u64 offset, u64 end) | |
2aaa6655 | 2331 | { |
a012a74e NB |
2332 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
2333 | struct btrfs_root *root = inode->root; | |
2aaa6655 JB |
2334 | struct extent_buffer *leaf; |
2335 | struct btrfs_file_extent_item *fi; | |
2336 | struct extent_map *hole_em; | |
a012a74e | 2337 | struct extent_map_tree *em_tree = &inode->extent_tree; |
2aaa6655 JB |
2338 | struct btrfs_key key; |
2339 | int ret; | |
2340 | ||
0b246afa | 2341 | if (btrfs_fs_incompat(fs_info, NO_HOLES)) |
16e7549f JB |
2342 | goto out; |
2343 | ||
a012a74e | 2344 | key.objectid = btrfs_ino(inode); |
2aaa6655 JB |
2345 | key.type = BTRFS_EXTENT_DATA_KEY; |
2346 | key.offset = offset; | |
2347 | ||
2aaa6655 | 2348 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
f94480bd JB |
2349 | if (ret <= 0) { |
2350 | /* | |
2351 | * We should have dropped this offset, so if we find it then | |
2352 | * something has gone horribly wrong. | |
2353 | */ | |
2354 | if (ret == 0) | |
2355 | ret = -EINVAL; | |
2aaa6655 | 2356 | return ret; |
f94480bd | 2357 | } |
2aaa6655 JB |
2358 | |
2359 | leaf = path->nodes[0]; | |
a012a74e | 2360 | if (hole_mergeable(inode, leaf, path->slots[0] - 1, offset, end)) { |
2aaa6655 JB |
2361 | u64 num_bytes; |
2362 | ||
2363 | path->slots[0]--; | |
2364 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2365 | struct btrfs_file_extent_item); | |
2366 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + | |
2367 | end - offset; | |
2368 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2369 | btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes); | |
2370 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
2371 | btrfs_mark_buffer_dirty(leaf); | |
2372 | goto out; | |
2373 | } | |
2374 | ||
1707e26d | 2375 | if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) { |
2aaa6655 JB |
2376 | u64 num_bytes; |
2377 | ||
2aaa6655 | 2378 | key.offset = offset; |
0b246afa | 2379 | btrfs_set_item_key_safe(fs_info, path, &key); |
2aaa6655 JB |
2380 | fi = btrfs_item_ptr(leaf, path->slots[0], |
2381 | struct btrfs_file_extent_item); | |
2382 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end - | |
2383 | offset; | |
2384 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2385 | btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes); | |
2386 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
2387 | btrfs_mark_buffer_dirty(leaf); | |
2388 | goto out; | |
2389 | } | |
2390 | btrfs_release_path(path); | |
2391 | ||
a012a74e | 2392 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), |
f85b7379 | 2393 | offset, 0, 0, end - offset, 0, end - offset, 0, 0, 0); |
2aaa6655 JB |
2394 | if (ret) |
2395 | return ret; | |
2396 | ||
2397 | out: | |
2398 | btrfs_release_path(path); | |
2399 | ||
2400 | hole_em = alloc_extent_map(); | |
2401 | if (!hole_em) { | |
2402 | btrfs_drop_extent_cache(inode, offset, end - 1, 0); | |
a012a74e | 2403 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags); |
2aaa6655 JB |
2404 | } else { |
2405 | hole_em->start = offset; | |
2406 | hole_em->len = end - offset; | |
cc95bef6 | 2407 | hole_em->ram_bytes = hole_em->len; |
2aaa6655 JB |
2408 | hole_em->orig_start = offset; |
2409 | ||
2410 | hole_em->block_start = EXTENT_MAP_HOLE; | |
2411 | hole_em->block_len = 0; | |
b4939680 | 2412 | hole_em->orig_block_len = 0; |
0b246afa | 2413 | hole_em->bdev = fs_info->fs_devices->latest_bdev; |
2aaa6655 JB |
2414 | hole_em->compress_type = BTRFS_COMPRESS_NONE; |
2415 | hole_em->generation = trans->transid; | |
2416 | ||
2417 | do { | |
2418 | btrfs_drop_extent_cache(inode, offset, end - 1, 0); | |
2419 | write_lock(&em_tree->lock); | |
09a2a8f9 | 2420 | ret = add_extent_mapping(em_tree, hole_em, 1); |
2aaa6655 JB |
2421 | write_unlock(&em_tree->lock); |
2422 | } while (ret == -EEXIST); | |
2423 | free_extent_map(hole_em); | |
2424 | if (ret) | |
2425 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
a012a74e | 2426 | &inode->runtime_flags); |
2aaa6655 JB |
2427 | } |
2428 | ||
2429 | return 0; | |
2430 | } | |
2431 | ||
d7781546 QW |
2432 | /* |
2433 | * Find a hole extent on given inode and change start/len to the end of hole | |
2434 | * extent.(hole/vacuum extent whose em->start <= start && | |
2435 | * em->start + em->len > start) | |
2436 | * When a hole extent is found, return 1 and modify start/len. | |
2437 | */ | |
2438 | static int find_first_non_hole(struct inode *inode, u64 *start, u64 *len) | |
2439 | { | |
609805d8 | 2440 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
d7781546 QW |
2441 | struct extent_map *em; |
2442 | int ret = 0; | |
2443 | ||
609805d8 FM |
2444 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, |
2445 | round_down(*start, fs_info->sectorsize), | |
2446 | round_up(*len, fs_info->sectorsize), 0); | |
9986277e DC |
2447 | if (IS_ERR(em)) |
2448 | return PTR_ERR(em); | |
d7781546 QW |
2449 | |
2450 | /* Hole or vacuum extent(only exists in no-hole mode) */ | |
2451 | if (em->block_start == EXTENT_MAP_HOLE) { | |
2452 | ret = 1; | |
2453 | *len = em->start + em->len > *start + *len ? | |
2454 | 0 : *start + *len - em->start - em->len; | |
2455 | *start = em->start + em->len; | |
2456 | } | |
2457 | free_extent_map(em); | |
2458 | return ret; | |
2459 | } | |
2460 | ||
2aaa6655 JB |
2461 | static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len) |
2462 | { | |
0b246afa | 2463 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2aaa6655 JB |
2464 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2465 | struct extent_state *cached_state = NULL; | |
2466 | struct btrfs_path *path; | |
2467 | struct btrfs_block_rsv *rsv; | |
2468 | struct btrfs_trans_handle *trans; | |
d7781546 QW |
2469 | u64 lockstart; |
2470 | u64 lockend; | |
2471 | u64 tail_start; | |
2472 | u64 tail_len; | |
2473 | u64 orig_start = offset; | |
2474 | u64 cur_offset; | |
5f52a2c5 | 2475 | u64 min_size = btrfs_calc_trans_metadata_size(fs_info, 1); |
2aaa6655 | 2476 | u64 drop_end; |
2aaa6655 JB |
2477 | int ret = 0; |
2478 | int err = 0; | |
6e4d6fa1 | 2479 | unsigned int rsv_count; |
9703fefe | 2480 | bool same_block; |
0b246afa | 2481 | bool no_holes = btrfs_fs_incompat(fs_info, NO_HOLES); |
a1a50f60 | 2482 | u64 ino_size; |
9703fefe | 2483 | bool truncated_block = false; |
e8c1c76e | 2484 | bool updated_inode = false; |
2aaa6655 | 2485 | |
0ef8b726 JB |
2486 | ret = btrfs_wait_ordered_range(inode, offset, len); |
2487 | if (ret) | |
2488 | return ret; | |
2aaa6655 | 2489 | |
5955102c | 2490 | inode_lock(inode); |
0b246afa | 2491 | ino_size = round_up(inode->i_size, fs_info->sectorsize); |
d7781546 QW |
2492 | ret = find_first_non_hole(inode, &offset, &len); |
2493 | if (ret < 0) | |
2494 | goto out_only_mutex; | |
2495 | if (ret && !len) { | |
2496 | /* Already in a large hole */ | |
2497 | ret = 0; | |
2498 | goto out_only_mutex; | |
2499 | } | |
2500 | ||
da17066c | 2501 | lockstart = round_up(offset, btrfs_inode_sectorsize(inode)); |
d7781546 | 2502 | lockend = round_down(offset + len, |
da17066c | 2503 | btrfs_inode_sectorsize(inode)) - 1; |
0b246afa JM |
2504 | same_block = (BTRFS_BYTES_TO_BLKS(fs_info, offset)) |
2505 | == (BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)); | |
7426cc04 | 2506 | /* |
9703fefe | 2507 | * We needn't truncate any block which is beyond the end of the file |
7426cc04 MX |
2508 | * because we are sure there is no data there. |
2509 | */ | |
2aaa6655 | 2510 | /* |
9703fefe CR |
2511 | * Only do this if we are in the same block and we aren't doing the |
2512 | * entire block. | |
2aaa6655 | 2513 | */ |
0b246afa | 2514 | if (same_block && len < fs_info->sectorsize) { |
e8c1c76e | 2515 | if (offset < ino_size) { |
9703fefe CR |
2516 | truncated_block = true; |
2517 | ret = btrfs_truncate_block(inode, offset, len, 0); | |
e8c1c76e FM |
2518 | } else { |
2519 | ret = 0; | |
2520 | } | |
d7781546 | 2521 | goto out_only_mutex; |
2aaa6655 JB |
2522 | } |
2523 | ||
9703fefe | 2524 | /* zero back part of the first block */ |
12870f1c | 2525 | if (offset < ino_size) { |
9703fefe CR |
2526 | truncated_block = true; |
2527 | ret = btrfs_truncate_block(inode, offset, 0, 0); | |
7426cc04 | 2528 | if (ret) { |
5955102c | 2529 | inode_unlock(inode); |
7426cc04 MX |
2530 | return ret; |
2531 | } | |
2aaa6655 JB |
2532 | } |
2533 | ||
d7781546 QW |
2534 | /* Check the aligned pages after the first unaligned page, |
2535 | * if offset != orig_start, which means the first unaligned page | |
01327610 | 2536 | * including several following pages are already in holes, |
d7781546 QW |
2537 | * the extra check can be skipped */ |
2538 | if (offset == orig_start) { | |
2539 | /* after truncate page, check hole again */ | |
2540 | len = offset + len - lockstart; | |
2541 | offset = lockstart; | |
2542 | ret = find_first_non_hole(inode, &offset, &len); | |
2543 | if (ret < 0) | |
2544 | goto out_only_mutex; | |
2545 | if (ret && !len) { | |
2546 | ret = 0; | |
2547 | goto out_only_mutex; | |
2548 | } | |
2549 | lockstart = offset; | |
2550 | } | |
2551 | ||
2552 | /* Check the tail unaligned part is in a hole */ | |
2553 | tail_start = lockend + 1; | |
2554 | tail_len = offset + len - tail_start; | |
2555 | if (tail_len) { | |
2556 | ret = find_first_non_hole(inode, &tail_start, &tail_len); | |
2557 | if (unlikely(ret < 0)) | |
2558 | goto out_only_mutex; | |
2559 | if (!ret) { | |
2560 | /* zero the front end of the last page */ | |
2561 | if (tail_start + tail_len < ino_size) { | |
9703fefe CR |
2562 | truncated_block = true; |
2563 | ret = btrfs_truncate_block(inode, | |
2564 | tail_start + tail_len, | |
2565 | 0, 1); | |
d7781546 QW |
2566 | if (ret) |
2567 | goto out_only_mutex; | |
51f395ad | 2568 | } |
0061280d | 2569 | } |
2aaa6655 JB |
2570 | } |
2571 | ||
2572 | if (lockend < lockstart) { | |
e8c1c76e FM |
2573 | ret = 0; |
2574 | goto out_only_mutex; | |
2aaa6655 JB |
2575 | } |
2576 | ||
2577 | while (1) { | |
2578 | struct btrfs_ordered_extent *ordered; | |
2579 | ||
2580 | truncate_pagecache_range(inode, lockstart, lockend); | |
2581 | ||
2582 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
ff13db41 | 2583 | &cached_state); |
2aaa6655 JB |
2584 | ordered = btrfs_lookup_first_ordered_extent(inode, lockend); |
2585 | ||
2586 | /* | |
2587 | * We need to make sure we have no ordered extents in this range | |
2588 | * and nobody raced in and read a page in this range, if we did | |
2589 | * we need to try again. | |
2590 | */ | |
2591 | if ((!ordered || | |
6126e3ca | 2592 | (ordered->file_offset + ordered->len <= lockstart || |
2aaa6655 | 2593 | ordered->file_offset > lockend)) && |
fc4adbff | 2594 | !btrfs_page_exists_in_range(inode, lockstart, lockend)) { |
2aaa6655 JB |
2595 | if (ordered) |
2596 | btrfs_put_ordered_extent(ordered); | |
2597 | break; | |
2598 | } | |
2599 | if (ordered) | |
2600 | btrfs_put_ordered_extent(ordered); | |
2601 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, | |
e43bbe5e | 2602 | lockend, &cached_state); |
0ef8b726 JB |
2603 | ret = btrfs_wait_ordered_range(inode, lockstart, |
2604 | lockend - lockstart + 1); | |
2605 | if (ret) { | |
5955102c | 2606 | inode_unlock(inode); |
0ef8b726 JB |
2607 | return ret; |
2608 | } | |
2aaa6655 JB |
2609 | } |
2610 | ||
2611 | path = btrfs_alloc_path(); | |
2612 | if (!path) { | |
2613 | ret = -ENOMEM; | |
2614 | goto out; | |
2615 | } | |
2616 | ||
2ff7e61e | 2617 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
2aaa6655 JB |
2618 | if (!rsv) { |
2619 | ret = -ENOMEM; | |
2620 | goto out_free; | |
2621 | } | |
5f52a2c5 | 2622 | rsv->size = btrfs_calc_trans_metadata_size(fs_info, 1); |
2aaa6655 JB |
2623 | rsv->failfast = 1; |
2624 | ||
2625 | /* | |
2626 | * 1 - update the inode | |
2627 | * 1 - removing the extents in the range | |
16e7549f | 2628 | * 1 - adding the hole extent if no_holes isn't set |
2aaa6655 | 2629 | */ |
16e7549f JB |
2630 | rsv_count = no_holes ? 2 : 3; |
2631 | trans = btrfs_start_transaction(root, rsv_count); | |
2aaa6655 JB |
2632 | if (IS_ERR(trans)) { |
2633 | err = PTR_ERR(trans); | |
2634 | goto out_free; | |
2635 | } | |
2636 | ||
0b246afa | 2637 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv, |
25d609f8 | 2638 | min_size, 0); |
2aaa6655 JB |
2639 | BUG_ON(ret); |
2640 | trans->block_rsv = rsv; | |
2641 | ||
d7781546 QW |
2642 | cur_offset = lockstart; |
2643 | len = lockend - cur_offset; | |
2aaa6655 JB |
2644 | while (cur_offset < lockend) { |
2645 | ret = __btrfs_drop_extents(trans, root, inode, path, | |
2646 | cur_offset, lockend + 1, | |
1acae57b | 2647 | &drop_end, 1, 0, 0, NULL); |
2aaa6655 JB |
2648 | if (ret != -ENOSPC) |
2649 | break; | |
2650 | ||
0b246afa | 2651 | trans->block_rsv = &fs_info->trans_block_rsv; |
2aaa6655 | 2652 | |
62fe51c1 | 2653 | if (cur_offset < drop_end && cur_offset < ino_size) { |
a012a74e NB |
2654 | ret = fill_holes(trans, BTRFS_I(inode), path, |
2655 | cur_offset, drop_end); | |
12870f1c | 2656 | if (ret) { |
f94480bd JB |
2657 | /* |
2658 | * If we failed then we didn't insert our hole | |
2659 | * entries for the area we dropped, so now the | |
2660 | * fs is corrupted, so we must abort the | |
2661 | * transaction. | |
2662 | */ | |
2663 | btrfs_abort_transaction(trans, ret); | |
12870f1c FM |
2664 | err = ret; |
2665 | break; | |
2666 | } | |
2aaa6655 JB |
2667 | } |
2668 | ||
2669 | cur_offset = drop_end; | |
2670 | ||
2671 | ret = btrfs_update_inode(trans, root, inode); | |
2672 | if (ret) { | |
2673 | err = ret; | |
2674 | break; | |
2675 | } | |
2676 | ||
3a45bb20 | 2677 | btrfs_end_transaction(trans); |
2ff7e61e | 2678 | btrfs_btree_balance_dirty(fs_info); |
2aaa6655 | 2679 | |
16e7549f | 2680 | trans = btrfs_start_transaction(root, rsv_count); |
2aaa6655 JB |
2681 | if (IS_ERR(trans)) { |
2682 | ret = PTR_ERR(trans); | |
2683 | trans = NULL; | |
2684 | break; | |
2685 | } | |
2686 | ||
0b246afa | 2687 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, |
25d609f8 | 2688 | rsv, min_size, 0); |
2aaa6655 JB |
2689 | BUG_ON(ret); /* shouldn't happen */ |
2690 | trans->block_rsv = rsv; | |
d7781546 QW |
2691 | |
2692 | ret = find_first_non_hole(inode, &cur_offset, &len); | |
2693 | if (unlikely(ret < 0)) | |
2694 | break; | |
2695 | if (ret && !len) { | |
2696 | ret = 0; | |
2697 | break; | |
2698 | } | |
2aaa6655 JB |
2699 | } |
2700 | ||
2701 | if (ret) { | |
2702 | err = ret; | |
2703 | goto out_trans; | |
2704 | } | |
2705 | ||
0b246afa | 2706 | trans->block_rsv = &fs_info->trans_block_rsv; |
2959a32a FM |
2707 | /* |
2708 | * If we are using the NO_HOLES feature we might have had already an | |
2709 | * hole that overlaps a part of the region [lockstart, lockend] and | |
2710 | * ends at (or beyond) lockend. Since we have no file extent items to | |
2711 | * represent holes, drop_end can be less than lockend and so we must | |
2712 | * make sure we have an extent map representing the existing hole (the | |
2713 | * call to __btrfs_drop_extents() might have dropped the existing extent | |
2714 | * map representing the existing hole), otherwise the fast fsync path | |
2715 | * will not record the existence of the hole region | |
2716 | * [existing_hole_start, lockend]. | |
2717 | */ | |
2718 | if (drop_end <= lockend) | |
2719 | drop_end = lockend + 1; | |
fc19c5e7 FM |
2720 | /* |
2721 | * Don't insert file hole extent item if it's for a range beyond eof | |
2722 | * (because it's useless) or if it represents a 0 bytes range (when | |
2723 | * cur_offset == drop_end). | |
2724 | */ | |
2725 | if (cur_offset < ino_size && cur_offset < drop_end) { | |
a012a74e NB |
2726 | ret = fill_holes(trans, BTRFS_I(inode), path, |
2727 | cur_offset, drop_end); | |
12870f1c | 2728 | if (ret) { |
f94480bd JB |
2729 | /* Same comment as above. */ |
2730 | btrfs_abort_transaction(trans, ret); | |
12870f1c FM |
2731 | err = ret; |
2732 | goto out_trans; | |
2733 | } | |
2aaa6655 JB |
2734 | } |
2735 | ||
2736 | out_trans: | |
2737 | if (!trans) | |
2738 | goto out_free; | |
2739 | ||
e1f5790e | 2740 | inode_inc_iversion(inode); |
c2050a45 | 2741 | inode->i_mtime = inode->i_ctime = current_time(inode); |
e1f5790e | 2742 | |
0b246afa | 2743 | trans->block_rsv = &fs_info->trans_block_rsv; |
2aaa6655 | 2744 | ret = btrfs_update_inode(trans, root, inode); |
e8c1c76e | 2745 | updated_inode = true; |
3a45bb20 | 2746 | btrfs_end_transaction(trans); |
2ff7e61e | 2747 | btrfs_btree_balance_dirty(fs_info); |
2aaa6655 JB |
2748 | out_free: |
2749 | btrfs_free_path(path); | |
2ff7e61e | 2750 | btrfs_free_block_rsv(fs_info, rsv); |
2aaa6655 JB |
2751 | out: |
2752 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
e43bbe5e | 2753 | &cached_state); |
d7781546 | 2754 | out_only_mutex: |
9703fefe | 2755 | if (!updated_inode && truncated_block && !ret && !err) { |
e8c1c76e FM |
2756 | /* |
2757 | * If we only end up zeroing part of a page, we still need to | |
2758 | * update the inode item, so that all the time fields are | |
2759 | * updated as well as the necessary btrfs inode in memory fields | |
2760 | * for detecting, at fsync time, if the inode isn't yet in the | |
2761 | * log tree or it's there but not up to date. | |
2762 | */ | |
2763 | trans = btrfs_start_transaction(root, 1); | |
2764 | if (IS_ERR(trans)) { | |
2765 | err = PTR_ERR(trans); | |
2766 | } else { | |
2767 | err = btrfs_update_inode(trans, root, inode); | |
3a45bb20 | 2768 | ret = btrfs_end_transaction(trans); |
e8c1c76e FM |
2769 | } |
2770 | } | |
5955102c | 2771 | inode_unlock(inode); |
2aaa6655 JB |
2772 | if (ret && !err) |
2773 | err = ret; | |
2774 | return err; | |
2775 | } | |
2776 | ||
14524a84 QW |
2777 | /* Helper structure to record which range is already reserved */ |
2778 | struct falloc_range { | |
2779 | struct list_head list; | |
2780 | u64 start; | |
2781 | u64 len; | |
2782 | }; | |
2783 | ||
2784 | /* | |
2785 | * Helper function to add falloc range | |
2786 | * | |
2787 | * Caller should have locked the larger range of extent containing | |
2788 | * [start, len) | |
2789 | */ | |
2790 | static int add_falloc_range(struct list_head *head, u64 start, u64 len) | |
2791 | { | |
2792 | struct falloc_range *prev = NULL; | |
2793 | struct falloc_range *range = NULL; | |
2794 | ||
2795 | if (list_empty(head)) | |
2796 | goto insert; | |
2797 | ||
2798 | /* | |
2799 | * As fallocate iterate by bytenr order, we only need to check | |
2800 | * the last range. | |
2801 | */ | |
2802 | prev = list_entry(head->prev, struct falloc_range, list); | |
2803 | if (prev->start + prev->len == start) { | |
2804 | prev->len += len; | |
2805 | return 0; | |
2806 | } | |
2807 | insert: | |
32fc932e | 2808 | range = kmalloc(sizeof(*range), GFP_KERNEL); |
14524a84 QW |
2809 | if (!range) |
2810 | return -ENOMEM; | |
2811 | range->start = start; | |
2812 | range->len = len; | |
2813 | list_add_tail(&range->list, head); | |
2814 | return 0; | |
2815 | } | |
2816 | ||
2fe17c10 CH |
2817 | static long btrfs_fallocate(struct file *file, int mode, |
2818 | loff_t offset, loff_t len) | |
2819 | { | |
496ad9aa | 2820 | struct inode *inode = file_inode(file); |
2fe17c10 | 2821 | struct extent_state *cached_state = NULL; |
364ecf36 | 2822 | struct extent_changeset *data_reserved = NULL; |
14524a84 QW |
2823 | struct falloc_range *range; |
2824 | struct falloc_range *tmp; | |
2825 | struct list_head reserve_list; | |
2fe17c10 CH |
2826 | u64 cur_offset; |
2827 | u64 last_byte; | |
2828 | u64 alloc_start; | |
2829 | u64 alloc_end; | |
2830 | u64 alloc_hint = 0; | |
2831 | u64 locked_end; | |
14524a84 | 2832 | u64 actual_end = 0; |
2fe17c10 | 2833 | struct extent_map *em; |
da17066c | 2834 | int blocksize = btrfs_inode_sectorsize(inode); |
2fe17c10 CH |
2835 | int ret; |
2836 | ||
797f4277 MX |
2837 | alloc_start = round_down(offset, blocksize); |
2838 | alloc_end = round_up(offset + len, blocksize); | |
18513091 | 2839 | cur_offset = alloc_start; |
2fe17c10 | 2840 | |
2aaa6655 JB |
2841 | /* Make sure we aren't being give some crap mode */ |
2842 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
2fe17c10 CH |
2843 | return -EOPNOTSUPP; |
2844 | ||
2aaa6655 JB |
2845 | if (mode & FALLOC_FL_PUNCH_HOLE) |
2846 | return btrfs_punch_hole(inode, offset, len); | |
2847 | ||
d98456fc | 2848 | /* |
14524a84 QW |
2849 | * Only trigger disk allocation, don't trigger qgroup reserve |
2850 | * | |
2851 | * For qgroup space, it will be checked later. | |
d98456fc | 2852 | */ |
04f4f916 NB |
2853 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), |
2854 | alloc_end - alloc_start); | |
14524a84 | 2855 | if (ret < 0) |
d98456fc CM |
2856 | return ret; |
2857 | ||
5955102c | 2858 | inode_lock(inode); |
2a162ce9 DI |
2859 | |
2860 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) { | |
2861 | ret = inode_newsize_ok(inode, offset + len); | |
2862 | if (ret) | |
2863 | goto out; | |
2864 | } | |
2fe17c10 | 2865 | |
14524a84 QW |
2866 | /* |
2867 | * TODO: Move these two operations after we have checked | |
2868 | * accurate reserved space, or fallocate can still fail but | |
2869 | * with page truncated or size expanded. | |
2870 | * | |
2871 | * But that's a minor problem and won't do much harm BTW. | |
2872 | */ | |
2fe17c10 | 2873 | if (alloc_start > inode->i_size) { |
a41ad394 JB |
2874 | ret = btrfs_cont_expand(inode, i_size_read(inode), |
2875 | alloc_start); | |
2fe17c10 CH |
2876 | if (ret) |
2877 | goto out; | |
0f6925fa | 2878 | } else if (offset + len > inode->i_size) { |
a71754fc JB |
2879 | /* |
2880 | * If we are fallocating from the end of the file onward we | |
9703fefe CR |
2881 | * need to zero out the end of the block if i_size lands in the |
2882 | * middle of a block. | |
a71754fc | 2883 | */ |
9703fefe | 2884 | ret = btrfs_truncate_block(inode, inode->i_size, 0, 0); |
a71754fc JB |
2885 | if (ret) |
2886 | goto out; | |
2fe17c10 CH |
2887 | } |
2888 | ||
a71754fc JB |
2889 | /* |
2890 | * wait for ordered IO before we have any locks. We'll loop again | |
2891 | * below with the locks held. | |
2892 | */ | |
0ef8b726 JB |
2893 | ret = btrfs_wait_ordered_range(inode, alloc_start, |
2894 | alloc_end - alloc_start); | |
2895 | if (ret) | |
2896 | goto out; | |
a71754fc | 2897 | |
2fe17c10 CH |
2898 | locked_end = alloc_end - 1; |
2899 | while (1) { | |
2900 | struct btrfs_ordered_extent *ordered; | |
2901 | ||
2902 | /* the extent lock is ordered inside the running | |
2903 | * transaction | |
2904 | */ | |
2905 | lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start, | |
ff13db41 | 2906 | locked_end, &cached_state); |
96b09dde NB |
2907 | ordered = btrfs_lookup_first_ordered_extent(inode, locked_end); |
2908 | ||
2fe17c10 CH |
2909 | if (ordered && |
2910 | ordered->file_offset + ordered->len > alloc_start && | |
2911 | ordered->file_offset < alloc_end) { | |
2912 | btrfs_put_ordered_extent(ordered); | |
2913 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, | |
2914 | alloc_start, locked_end, | |
e43bbe5e | 2915 | &cached_state); |
2fe17c10 CH |
2916 | /* |
2917 | * we can't wait on the range with the transaction | |
2918 | * running or with the extent lock held | |
2919 | */ | |
0ef8b726 JB |
2920 | ret = btrfs_wait_ordered_range(inode, alloc_start, |
2921 | alloc_end - alloc_start); | |
2922 | if (ret) | |
2923 | goto out; | |
2fe17c10 CH |
2924 | } else { |
2925 | if (ordered) | |
2926 | btrfs_put_ordered_extent(ordered); | |
2927 | break; | |
2928 | } | |
2929 | } | |
2930 | ||
14524a84 QW |
2931 | /* First, check if we exceed the qgroup limit */ |
2932 | INIT_LIST_HEAD(&reserve_list); | |
6b7d6e93 | 2933 | while (cur_offset < alloc_end) { |
fc4f21b1 | 2934 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset, |
2fe17c10 | 2935 | alloc_end - cur_offset, 0); |
9986277e DC |
2936 | if (IS_ERR(em)) { |
2937 | ret = PTR_ERR(em); | |
79787eaa JM |
2938 | break; |
2939 | } | |
2fe17c10 | 2940 | last_byte = min(extent_map_end(em), alloc_end); |
f1e490a7 | 2941 | actual_end = min_t(u64, extent_map_end(em), offset + len); |
797f4277 | 2942 | last_byte = ALIGN(last_byte, blocksize); |
2fe17c10 CH |
2943 | if (em->block_start == EXTENT_MAP_HOLE || |
2944 | (cur_offset >= inode->i_size && | |
2945 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
14524a84 QW |
2946 | ret = add_falloc_range(&reserve_list, cur_offset, |
2947 | last_byte - cur_offset); | |
2948 | if (ret < 0) { | |
2949 | free_extent_map(em); | |
2950 | break; | |
3d850dd4 | 2951 | } |
364ecf36 QW |
2952 | ret = btrfs_qgroup_reserve_data(inode, &data_reserved, |
2953 | cur_offset, last_byte - cur_offset); | |
be2d253c FM |
2954 | if (ret < 0) { |
2955 | free_extent_map(em); | |
14524a84 | 2956 | break; |
be2d253c | 2957 | } |
18513091 WX |
2958 | } else { |
2959 | /* | |
2960 | * Do not need to reserve unwritten extent for this | |
2961 | * range, free reserved data space first, otherwise | |
2962 | * it'll result in false ENOSPC error. | |
2963 | */ | |
bc42bda2 QW |
2964 | btrfs_free_reserved_data_space(inode, data_reserved, |
2965 | cur_offset, last_byte - cur_offset); | |
2fe17c10 CH |
2966 | } |
2967 | free_extent_map(em); | |
2fe17c10 | 2968 | cur_offset = last_byte; |
14524a84 QW |
2969 | } |
2970 | ||
2971 | /* | |
2972 | * If ret is still 0, means we're OK to fallocate. | |
2973 | * Or just cleanup the list and exit. | |
2974 | */ | |
2975 | list_for_each_entry_safe(range, tmp, &reserve_list, list) { | |
2976 | if (!ret) | |
2977 | ret = btrfs_prealloc_file_range(inode, mode, | |
2978 | range->start, | |
93407472 | 2979 | range->len, i_blocksize(inode), |
14524a84 | 2980 | offset + len, &alloc_hint); |
18513091 | 2981 | else |
bc42bda2 QW |
2982 | btrfs_free_reserved_data_space(inode, |
2983 | data_reserved, range->start, | |
2984 | range->len); | |
14524a84 QW |
2985 | list_del(&range->list); |
2986 | kfree(range); | |
2987 | } | |
2988 | if (ret < 0) | |
2989 | goto out_unlock; | |
2990 | ||
2991 | if (actual_end > inode->i_size && | |
2992 | !(mode & FALLOC_FL_KEEP_SIZE)) { | |
2993 | struct btrfs_trans_handle *trans; | |
2994 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2995 | ||
2996 | /* | |
2997 | * We didn't need to allocate any more space, but we | |
2998 | * still extended the size of the file so we need to | |
2999 | * update i_size and the inode item. | |
3000 | */ | |
3001 | trans = btrfs_start_transaction(root, 1); | |
3002 | if (IS_ERR(trans)) { | |
3003 | ret = PTR_ERR(trans); | |
3004 | } else { | |
c2050a45 | 3005 | inode->i_ctime = current_time(inode); |
14524a84 QW |
3006 | i_size_write(inode, actual_end); |
3007 | btrfs_ordered_update_i_size(inode, actual_end, NULL); | |
3008 | ret = btrfs_update_inode(trans, root, inode); | |
3009 | if (ret) | |
3a45bb20 | 3010 | btrfs_end_transaction(trans); |
14524a84 | 3011 | else |
3a45bb20 | 3012 | ret = btrfs_end_transaction(trans); |
2fe17c10 CH |
3013 | } |
3014 | } | |
14524a84 | 3015 | out_unlock: |
2fe17c10 | 3016 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
e43bbe5e | 3017 | &cached_state); |
2fe17c10 | 3018 | out: |
5955102c | 3019 | inode_unlock(inode); |
d98456fc | 3020 | /* Let go of our reservation. */ |
18513091 | 3021 | if (ret != 0) |
bc42bda2 QW |
3022 | btrfs_free_reserved_data_space(inode, data_reserved, |
3023 | alloc_start, alloc_end - cur_offset); | |
364ecf36 | 3024 | extent_changeset_free(data_reserved); |
2fe17c10 CH |
3025 | return ret; |
3026 | } | |
3027 | ||
965c8e59 | 3028 | static int find_desired_extent(struct inode *inode, loff_t *offset, int whence) |
b2675157 | 3029 | { |
0b246afa | 3030 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
7f4ca37c | 3031 | struct extent_map *em = NULL; |
b2675157 | 3032 | struct extent_state *cached_state = NULL; |
4d1a40c6 LB |
3033 | u64 lockstart; |
3034 | u64 lockend; | |
3035 | u64 start; | |
3036 | u64 len; | |
b2675157 JB |
3037 | int ret = 0; |
3038 | ||
4d1a40c6 LB |
3039 | if (inode->i_size == 0) |
3040 | return -ENXIO; | |
3041 | ||
3042 | /* | |
3043 | * *offset can be negative, in this case we start finding DATA/HOLE from | |
3044 | * the very start of the file. | |
3045 | */ | |
3046 | start = max_t(loff_t, 0, *offset); | |
3047 | ||
0b246afa | 3048 | lockstart = round_down(start, fs_info->sectorsize); |
da17066c | 3049 | lockend = round_up(i_size_read(inode), |
0b246afa | 3050 | fs_info->sectorsize); |
b2675157 | 3051 | if (lockend <= lockstart) |
0b246afa | 3052 | lockend = lockstart + fs_info->sectorsize; |
1214b53f | 3053 | lockend--; |
b2675157 JB |
3054 | len = lockend - lockstart + 1; |
3055 | ||
ff13db41 | 3056 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
d0082371 | 3057 | &cached_state); |
b2675157 | 3058 | |
7f4ca37c | 3059 | while (start < inode->i_size) { |
fc4f21b1 NB |
3060 | em = btrfs_get_extent_fiemap(BTRFS_I(inode), NULL, 0, |
3061 | start, len, 0); | |
b2675157 | 3062 | if (IS_ERR(em)) { |
6af021d8 | 3063 | ret = PTR_ERR(em); |
7f4ca37c | 3064 | em = NULL; |
b2675157 JB |
3065 | break; |
3066 | } | |
3067 | ||
7f4ca37c JB |
3068 | if (whence == SEEK_HOLE && |
3069 | (em->block_start == EXTENT_MAP_HOLE || | |
3070 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) | |
3071 | break; | |
3072 | else if (whence == SEEK_DATA && | |
3073 | (em->block_start != EXTENT_MAP_HOLE && | |
3074 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) | |
3075 | break; | |
b2675157 JB |
3076 | |
3077 | start = em->start + em->len; | |
b2675157 | 3078 | free_extent_map(em); |
7f4ca37c | 3079 | em = NULL; |
b2675157 JB |
3080 | cond_resched(); |
3081 | } | |
7f4ca37c JB |
3082 | free_extent_map(em); |
3083 | if (!ret) { | |
3084 | if (whence == SEEK_DATA && start >= inode->i_size) | |
3085 | ret = -ENXIO; | |
3086 | else | |
3087 | *offset = min_t(loff_t, start, inode->i_size); | |
3088 | } | |
b2675157 | 3089 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
e43bbe5e | 3090 | &cached_state); |
b2675157 JB |
3091 | return ret; |
3092 | } | |
3093 | ||
965c8e59 | 3094 | static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence) |
b2675157 JB |
3095 | { |
3096 | struct inode *inode = file->f_mapping->host; | |
3097 | int ret; | |
3098 | ||
5955102c | 3099 | inode_lock(inode); |
965c8e59 | 3100 | switch (whence) { |
b2675157 JB |
3101 | case SEEK_END: |
3102 | case SEEK_CUR: | |
965c8e59 | 3103 | offset = generic_file_llseek(file, offset, whence); |
b2675157 JB |
3104 | goto out; |
3105 | case SEEK_DATA: | |
3106 | case SEEK_HOLE: | |
48802c8a | 3107 | if (offset >= i_size_read(inode)) { |
5955102c | 3108 | inode_unlock(inode); |
48802c8a JL |
3109 | return -ENXIO; |
3110 | } | |
3111 | ||
965c8e59 | 3112 | ret = find_desired_extent(inode, &offset, whence); |
b2675157 | 3113 | if (ret) { |
5955102c | 3114 | inode_unlock(inode); |
b2675157 JB |
3115 | return ret; |
3116 | } | |
3117 | } | |
3118 | ||
46a1c2c7 | 3119 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
b2675157 | 3120 | out: |
5955102c | 3121 | inode_unlock(inode); |
b2675157 JB |
3122 | return offset; |
3123 | } | |
3124 | ||
edf064e7 GR |
3125 | static int btrfs_file_open(struct inode *inode, struct file *filp) |
3126 | { | |
91f9943e | 3127 | filp->f_mode |= FMODE_NOWAIT; |
edf064e7 GR |
3128 | return generic_file_open(inode, filp); |
3129 | } | |
3130 | ||
828c0950 | 3131 | const struct file_operations btrfs_file_operations = { |
b2675157 | 3132 | .llseek = btrfs_file_llseek, |
aad4f8bb | 3133 | .read_iter = generic_file_read_iter, |
e9906a98 | 3134 | .splice_read = generic_file_splice_read, |
b30ac0fc | 3135 | .write_iter = btrfs_file_write_iter, |
9ebefb18 | 3136 | .mmap = btrfs_file_mmap, |
edf064e7 | 3137 | .open = btrfs_file_open, |
e1b81e67 | 3138 | .release = btrfs_release_file, |
39279cc3 | 3139 | .fsync = btrfs_sync_file, |
2fe17c10 | 3140 | .fallocate = btrfs_fallocate, |
34287aa3 | 3141 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 3142 | #ifdef CONFIG_COMPAT |
4c63c245 | 3143 | .compat_ioctl = btrfs_compat_ioctl, |
39279cc3 | 3144 | #endif |
04b38d60 | 3145 | .clone_file_range = btrfs_clone_file_range, |
2b3909f8 | 3146 | .dedupe_file_range = btrfs_dedupe_file_range, |
39279cc3 | 3147 | }; |
9247f317 MX |
3148 | |
3149 | void btrfs_auto_defrag_exit(void) | |
3150 | { | |
5598e900 | 3151 | kmem_cache_destroy(btrfs_inode_defrag_cachep); |
9247f317 MX |
3152 | } |
3153 | ||
f5c29bd9 | 3154 | int __init btrfs_auto_defrag_init(void) |
9247f317 MX |
3155 | { |
3156 | btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag", | |
3157 | sizeof(struct inode_defrag), 0, | |
fba4b697 | 3158 | SLAB_MEM_SPREAD, |
9247f317 MX |
3159 | NULL); |
3160 | if (!btrfs_inode_defrag_cachep) | |
3161 | return -ENOMEM; | |
3162 | ||
3163 | return 0; | |
3164 | } | |
728404da FM |
3165 | |
3166 | int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end) | |
3167 | { | |
3168 | int ret; | |
3169 | ||
3170 | /* | |
3171 | * So with compression we will find and lock a dirty page and clear the | |
3172 | * first one as dirty, setup an async extent, and immediately return | |
3173 | * with the entire range locked but with nobody actually marked with | |
3174 | * writeback. So we can't just filemap_write_and_wait_range() and | |
3175 | * expect it to work since it will just kick off a thread to do the | |
3176 | * actual work. So we need to call filemap_fdatawrite_range _again_ | |
3177 | * since it will wait on the page lock, which won't be unlocked until | |
3178 | * after the pages have been marked as writeback and so we're good to go | |
3179 | * from there. We have to do this otherwise we'll miss the ordered | |
3180 | * extents and that results in badness. Please Josef, do not think you | |
3181 | * know better and pull this out at some point in the future, it is | |
3182 | * right and you are wrong. | |
3183 | */ | |
3184 | ret = filemap_fdatawrite_range(inode->i_mapping, start, end); | |
3185 | if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
3186 | &BTRFS_I(inode)->runtime_flags)) | |
3187 | ret = filemap_fdatawrite_range(inode->i_mapping, start, end); | |
3188 | ||
3189 | return ret; | |
3190 | } |