Commit | Line | Data |
---|---|---|
dc17ff8f 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 | ||
dc17ff8f | 19 | #include <linux/slab.h> |
d6bfde87 | 20 | #include <linux/blkdev.h> |
f421950f CM |
21 | #include <linux/writeback.h> |
22 | #include <linux/pagevec.h> | |
dc17ff8f CM |
23 | #include "ctree.h" |
24 | #include "transaction.h" | |
25 | #include "btrfs_inode.h" | |
e6dcd2dc | 26 | #include "extent_io.h" |
dc17ff8f | 27 | |
e6dcd2dc | 28 | static u64 entry_end(struct btrfs_ordered_extent *entry) |
dc17ff8f | 29 | { |
e6dcd2dc CM |
30 | if (entry->file_offset + entry->len < entry->file_offset) |
31 | return (u64)-1; | |
32 | return entry->file_offset + entry->len; | |
dc17ff8f CM |
33 | } |
34 | ||
d352ac68 CM |
35 | /* returns NULL if the insertion worked, or it returns the node it did find |
36 | * in the tree | |
37 | */ | |
e6dcd2dc CM |
38 | static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset, |
39 | struct rb_node *node) | |
dc17ff8f | 40 | { |
d397712b CM |
41 | struct rb_node **p = &root->rb_node; |
42 | struct rb_node *parent = NULL; | |
e6dcd2dc | 43 | struct btrfs_ordered_extent *entry; |
dc17ff8f | 44 | |
d397712b | 45 | while (*p) { |
dc17ff8f | 46 | parent = *p; |
e6dcd2dc | 47 | entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node); |
dc17ff8f | 48 | |
e6dcd2dc | 49 | if (file_offset < entry->file_offset) |
dc17ff8f | 50 | p = &(*p)->rb_left; |
e6dcd2dc | 51 | else if (file_offset >= entry_end(entry)) |
dc17ff8f CM |
52 | p = &(*p)->rb_right; |
53 | else | |
54 | return parent; | |
55 | } | |
56 | ||
57 | rb_link_node(node, parent, p); | |
58 | rb_insert_color(node, root); | |
59 | return NULL; | |
60 | } | |
61 | ||
d352ac68 CM |
62 | /* |
63 | * look for a given offset in the tree, and if it can't be found return the | |
64 | * first lesser offset | |
65 | */ | |
e6dcd2dc CM |
66 | static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset, |
67 | struct rb_node **prev_ret) | |
dc17ff8f | 68 | { |
d397712b | 69 | struct rb_node *n = root->rb_node; |
dc17ff8f | 70 | struct rb_node *prev = NULL; |
e6dcd2dc CM |
71 | struct rb_node *test; |
72 | struct btrfs_ordered_extent *entry; | |
73 | struct btrfs_ordered_extent *prev_entry = NULL; | |
dc17ff8f | 74 | |
d397712b | 75 | while (n) { |
e6dcd2dc | 76 | entry = rb_entry(n, struct btrfs_ordered_extent, rb_node); |
dc17ff8f CM |
77 | prev = n; |
78 | prev_entry = entry; | |
dc17ff8f | 79 | |
e6dcd2dc | 80 | if (file_offset < entry->file_offset) |
dc17ff8f | 81 | n = n->rb_left; |
e6dcd2dc | 82 | else if (file_offset >= entry_end(entry)) |
dc17ff8f CM |
83 | n = n->rb_right; |
84 | else | |
85 | return n; | |
86 | } | |
87 | if (!prev_ret) | |
88 | return NULL; | |
89 | ||
d397712b | 90 | while (prev && file_offset >= entry_end(prev_entry)) { |
e6dcd2dc CM |
91 | test = rb_next(prev); |
92 | if (!test) | |
93 | break; | |
94 | prev_entry = rb_entry(test, struct btrfs_ordered_extent, | |
95 | rb_node); | |
96 | if (file_offset < entry_end(prev_entry)) | |
97 | break; | |
98 | ||
99 | prev = test; | |
100 | } | |
101 | if (prev) | |
102 | prev_entry = rb_entry(prev, struct btrfs_ordered_extent, | |
103 | rb_node); | |
d397712b | 104 | while (prev && file_offset < entry_end(prev_entry)) { |
e6dcd2dc CM |
105 | test = rb_prev(prev); |
106 | if (!test) | |
107 | break; | |
108 | prev_entry = rb_entry(test, struct btrfs_ordered_extent, | |
109 | rb_node); | |
110 | prev = test; | |
dc17ff8f CM |
111 | } |
112 | *prev_ret = prev; | |
113 | return NULL; | |
114 | } | |
115 | ||
d352ac68 CM |
116 | /* |
117 | * helper to check if a given offset is inside a given entry | |
118 | */ | |
e6dcd2dc CM |
119 | static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset) |
120 | { | |
121 | if (file_offset < entry->file_offset || | |
122 | entry->file_offset + entry->len <= file_offset) | |
123 | return 0; | |
124 | return 1; | |
125 | } | |
126 | ||
4b46fce2 JB |
127 | static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset, |
128 | u64 len) | |
129 | { | |
130 | if (file_offset + len <= entry->file_offset || | |
131 | entry->file_offset + entry->len <= file_offset) | |
132 | return 0; | |
133 | return 1; | |
134 | } | |
135 | ||
d352ac68 CM |
136 | /* |
137 | * look find the first ordered struct that has this offset, otherwise | |
138 | * the first one less than this offset | |
139 | */ | |
e6dcd2dc CM |
140 | static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree, |
141 | u64 file_offset) | |
dc17ff8f | 142 | { |
e6dcd2dc | 143 | struct rb_root *root = &tree->tree; |
c87fb6fd | 144 | struct rb_node *prev = NULL; |
dc17ff8f | 145 | struct rb_node *ret; |
e6dcd2dc CM |
146 | struct btrfs_ordered_extent *entry; |
147 | ||
148 | if (tree->last) { | |
149 | entry = rb_entry(tree->last, struct btrfs_ordered_extent, | |
150 | rb_node); | |
151 | if (offset_in_entry(entry, file_offset)) | |
152 | return tree->last; | |
153 | } | |
154 | ret = __tree_search(root, file_offset, &prev); | |
dc17ff8f | 155 | if (!ret) |
e6dcd2dc CM |
156 | ret = prev; |
157 | if (ret) | |
158 | tree->last = ret; | |
dc17ff8f CM |
159 | return ret; |
160 | } | |
161 | ||
eb84ae03 CM |
162 | /* allocate and add a new ordered_extent into the per-inode tree. |
163 | * file_offset is the logical offset in the file | |
164 | * | |
165 | * start is the disk block number of an extent already reserved in the | |
166 | * extent allocation tree | |
167 | * | |
168 | * len is the length of the extent | |
169 | * | |
eb84ae03 CM |
170 | * The tree is given a single reference on the ordered extent that was |
171 | * inserted. | |
172 | */ | |
4b46fce2 JB |
173 | static int __btrfs_add_ordered_extent(struct inode *inode, u64 file_offset, |
174 | u64 start, u64 len, u64 disk_len, | |
261507a0 | 175 | int type, int dio, int compress_type) |
dc17ff8f | 176 | { |
dc17ff8f | 177 | struct btrfs_ordered_inode_tree *tree; |
e6dcd2dc CM |
178 | struct rb_node *node; |
179 | struct btrfs_ordered_extent *entry; | |
dc17ff8f | 180 | |
e6dcd2dc CM |
181 | tree = &BTRFS_I(inode)->ordered_tree; |
182 | entry = kzalloc(sizeof(*entry), GFP_NOFS); | |
dc17ff8f CM |
183 | if (!entry) |
184 | return -ENOMEM; | |
185 | ||
e6dcd2dc CM |
186 | entry->file_offset = file_offset; |
187 | entry->start = start; | |
188 | entry->len = len; | |
c8b97818 | 189 | entry->disk_len = disk_len; |
8b62b72b | 190 | entry->bytes_left = len; |
3eaa2885 | 191 | entry->inode = inode; |
261507a0 | 192 | entry->compress_type = compress_type; |
d899e052 | 193 | if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE) |
80ff3856 | 194 | set_bit(type, &entry->flags); |
3eaa2885 | 195 | |
4b46fce2 JB |
196 | if (dio) |
197 | set_bit(BTRFS_ORDERED_DIRECT, &entry->flags); | |
198 | ||
e6dcd2dc CM |
199 | /* one ref for the tree */ |
200 | atomic_set(&entry->refs, 1); | |
201 | init_waitqueue_head(&entry->wait); | |
202 | INIT_LIST_HEAD(&entry->list); | |
3eaa2885 | 203 | INIT_LIST_HEAD(&entry->root_extent_list); |
dc17ff8f | 204 | |
49958fd7 | 205 | spin_lock(&tree->lock); |
e6dcd2dc CM |
206 | node = tree_insert(&tree->tree, file_offset, |
207 | &entry->rb_node); | |
d397712b | 208 | BUG_ON(node); |
49958fd7 | 209 | spin_unlock(&tree->lock); |
d397712b | 210 | |
3eaa2885 CM |
211 | spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock); |
212 | list_add_tail(&entry->root_extent_list, | |
213 | &BTRFS_I(inode)->root->fs_info->ordered_extents); | |
214 | spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock); | |
215 | ||
e6dcd2dc | 216 | BUG_ON(node); |
dc17ff8f CM |
217 | return 0; |
218 | } | |
219 | ||
4b46fce2 JB |
220 | int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset, |
221 | u64 start, u64 len, u64 disk_len, int type) | |
222 | { | |
223 | return __btrfs_add_ordered_extent(inode, file_offset, start, len, | |
261507a0 LZ |
224 | disk_len, type, 0, |
225 | BTRFS_COMPRESS_NONE); | |
4b46fce2 JB |
226 | } |
227 | ||
228 | int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset, | |
229 | u64 start, u64 len, u64 disk_len, int type) | |
230 | { | |
231 | return __btrfs_add_ordered_extent(inode, file_offset, start, len, | |
261507a0 LZ |
232 | disk_len, type, 1, |
233 | BTRFS_COMPRESS_NONE); | |
234 | } | |
235 | ||
236 | int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset, | |
237 | u64 start, u64 len, u64 disk_len, | |
238 | int type, int compress_type) | |
239 | { | |
240 | return __btrfs_add_ordered_extent(inode, file_offset, start, len, | |
241 | disk_len, type, 0, | |
242 | compress_type); | |
4b46fce2 JB |
243 | } |
244 | ||
eb84ae03 CM |
245 | /* |
246 | * Add a struct btrfs_ordered_sum into the list of checksums to be inserted | |
3edf7d33 CM |
247 | * when an ordered extent is finished. If the list covers more than one |
248 | * ordered extent, it is split across multiples. | |
eb84ae03 | 249 | */ |
3edf7d33 CM |
250 | int btrfs_add_ordered_sum(struct inode *inode, |
251 | struct btrfs_ordered_extent *entry, | |
252 | struct btrfs_ordered_sum *sum) | |
dc17ff8f | 253 | { |
e6dcd2dc | 254 | struct btrfs_ordered_inode_tree *tree; |
dc17ff8f | 255 | |
e6dcd2dc | 256 | tree = &BTRFS_I(inode)->ordered_tree; |
49958fd7 | 257 | spin_lock(&tree->lock); |
e6dcd2dc | 258 | list_add_tail(&sum->list, &entry->list); |
49958fd7 | 259 | spin_unlock(&tree->lock); |
e6dcd2dc | 260 | return 0; |
dc17ff8f CM |
261 | } |
262 | ||
163cf09c CM |
263 | /* |
264 | * this is used to account for finished IO across a given range | |
265 | * of the file. The IO may span ordered extents. If | |
266 | * a given ordered_extent is completely done, 1 is returned, otherwise | |
267 | * 0. | |
268 | * | |
269 | * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used | |
270 | * to make sure this function only returns 1 once for a given ordered extent. | |
271 | * | |
272 | * file_offset is updated to one byte past the range that is recorded as | |
273 | * complete. This allows you to walk forward in the file. | |
274 | */ | |
275 | int btrfs_dec_test_first_ordered_pending(struct inode *inode, | |
276 | struct btrfs_ordered_extent **cached, | |
277 | u64 *file_offset, u64 io_size) | |
278 | { | |
279 | struct btrfs_ordered_inode_tree *tree; | |
280 | struct rb_node *node; | |
281 | struct btrfs_ordered_extent *entry = NULL; | |
282 | int ret; | |
283 | u64 dec_end; | |
284 | u64 dec_start; | |
285 | u64 to_dec; | |
286 | ||
287 | tree = &BTRFS_I(inode)->ordered_tree; | |
288 | spin_lock(&tree->lock); | |
289 | node = tree_search(tree, *file_offset); | |
290 | if (!node) { | |
291 | ret = 1; | |
292 | goto out; | |
293 | } | |
294 | ||
295 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
296 | if (!offset_in_entry(entry, *file_offset)) { | |
297 | ret = 1; | |
298 | goto out; | |
299 | } | |
300 | ||
301 | dec_start = max(*file_offset, entry->file_offset); | |
302 | dec_end = min(*file_offset + io_size, entry->file_offset + | |
303 | entry->len); | |
304 | *file_offset = dec_end; | |
305 | if (dec_start > dec_end) { | |
306 | printk(KERN_CRIT "bad ordering dec_start %llu end %llu\n", | |
307 | (unsigned long long)dec_start, | |
308 | (unsigned long long)dec_end); | |
309 | } | |
310 | to_dec = dec_end - dec_start; | |
311 | if (to_dec > entry->bytes_left) { | |
312 | printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n", | |
313 | (unsigned long long)entry->bytes_left, | |
314 | (unsigned long long)to_dec); | |
315 | } | |
316 | entry->bytes_left -= to_dec; | |
317 | if (entry->bytes_left == 0) | |
318 | ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags); | |
319 | else | |
320 | ret = 1; | |
321 | out: | |
322 | if (!ret && cached && entry) { | |
323 | *cached = entry; | |
324 | atomic_inc(&entry->refs); | |
325 | } | |
326 | spin_unlock(&tree->lock); | |
327 | return ret == 0; | |
328 | } | |
329 | ||
eb84ae03 CM |
330 | /* |
331 | * this is used to account for finished IO across a given range | |
332 | * of the file. The IO should not span ordered extents. If | |
333 | * a given ordered_extent is completely done, 1 is returned, otherwise | |
334 | * 0. | |
335 | * | |
336 | * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used | |
337 | * to make sure this function only returns 1 once for a given ordered extent. | |
338 | */ | |
e6dcd2dc | 339 | int btrfs_dec_test_ordered_pending(struct inode *inode, |
5a1a3df1 | 340 | struct btrfs_ordered_extent **cached, |
e6dcd2dc | 341 | u64 file_offset, u64 io_size) |
dc17ff8f | 342 | { |
e6dcd2dc | 343 | struct btrfs_ordered_inode_tree *tree; |
dc17ff8f | 344 | struct rb_node *node; |
5a1a3df1 | 345 | struct btrfs_ordered_extent *entry = NULL; |
e6dcd2dc CM |
346 | int ret; |
347 | ||
348 | tree = &BTRFS_I(inode)->ordered_tree; | |
49958fd7 | 349 | spin_lock(&tree->lock); |
e6dcd2dc | 350 | node = tree_search(tree, file_offset); |
dc17ff8f | 351 | if (!node) { |
e6dcd2dc CM |
352 | ret = 1; |
353 | goto out; | |
dc17ff8f CM |
354 | } |
355 | ||
e6dcd2dc CM |
356 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
357 | if (!offset_in_entry(entry, file_offset)) { | |
358 | ret = 1; | |
359 | goto out; | |
dc17ff8f | 360 | } |
e6dcd2dc | 361 | |
8b62b72b CM |
362 | if (io_size > entry->bytes_left) { |
363 | printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n", | |
364 | (unsigned long long)entry->bytes_left, | |
365 | (unsigned long long)io_size); | |
366 | } | |
367 | entry->bytes_left -= io_size; | |
368 | if (entry->bytes_left == 0) | |
e6dcd2dc | 369 | ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags); |
8b62b72b CM |
370 | else |
371 | ret = 1; | |
e6dcd2dc | 372 | out: |
5a1a3df1 JB |
373 | if (!ret && cached && entry) { |
374 | *cached = entry; | |
375 | atomic_inc(&entry->refs); | |
376 | } | |
49958fd7 | 377 | spin_unlock(&tree->lock); |
e6dcd2dc CM |
378 | return ret == 0; |
379 | } | |
dc17ff8f | 380 | |
eb84ae03 CM |
381 | /* |
382 | * used to drop a reference on an ordered extent. This will free | |
383 | * the extent if the last reference is dropped | |
384 | */ | |
e6dcd2dc CM |
385 | int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry) |
386 | { | |
ba1da2f4 CM |
387 | struct list_head *cur; |
388 | struct btrfs_ordered_sum *sum; | |
389 | ||
390 | if (atomic_dec_and_test(&entry->refs)) { | |
d397712b | 391 | while (!list_empty(&entry->list)) { |
ba1da2f4 CM |
392 | cur = entry->list.next; |
393 | sum = list_entry(cur, struct btrfs_ordered_sum, list); | |
394 | list_del(&sum->list); | |
395 | kfree(sum); | |
396 | } | |
e6dcd2dc | 397 | kfree(entry); |
ba1da2f4 | 398 | } |
e6dcd2dc | 399 | return 0; |
dc17ff8f | 400 | } |
cee36a03 | 401 | |
eb84ae03 CM |
402 | /* |
403 | * remove an ordered extent from the tree. No references are dropped | |
49958fd7 | 404 | * and you must wake_up entry->wait. You must hold the tree lock |
c2167754 | 405 | * while you call this function. |
eb84ae03 | 406 | */ |
c2167754 | 407 | static int __btrfs_remove_ordered_extent(struct inode *inode, |
e6dcd2dc | 408 | struct btrfs_ordered_extent *entry) |
cee36a03 | 409 | { |
e6dcd2dc | 410 | struct btrfs_ordered_inode_tree *tree; |
287a0ab9 | 411 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cee36a03 | 412 | struct rb_node *node; |
cee36a03 | 413 | |
e6dcd2dc | 414 | tree = &BTRFS_I(inode)->ordered_tree; |
e6dcd2dc | 415 | node = &entry->rb_node; |
cee36a03 | 416 | rb_erase(node, &tree->tree); |
e6dcd2dc CM |
417 | tree->last = NULL; |
418 | set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags); | |
3eaa2885 | 419 | |
287a0ab9 | 420 | spin_lock(&root->fs_info->ordered_extent_lock); |
3eaa2885 | 421 | list_del_init(&entry->root_extent_list); |
5a3f23d5 CM |
422 | |
423 | /* | |
424 | * we have no more ordered extents for this inode and | |
425 | * no dirty pages. We can safely remove it from the | |
426 | * list of ordered extents | |
427 | */ | |
428 | if (RB_EMPTY_ROOT(&tree->tree) && | |
429 | !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { | |
430 | list_del_init(&BTRFS_I(inode)->ordered_operations); | |
431 | } | |
287a0ab9 | 432 | spin_unlock(&root->fs_info->ordered_extent_lock); |
3eaa2885 | 433 | |
c2167754 YZ |
434 | return 0; |
435 | } | |
436 | ||
437 | /* | |
438 | * remove an ordered extent from the tree. No references are dropped | |
439 | * but any waiters are woken. | |
440 | */ | |
441 | int btrfs_remove_ordered_extent(struct inode *inode, | |
442 | struct btrfs_ordered_extent *entry) | |
443 | { | |
444 | struct btrfs_ordered_inode_tree *tree; | |
445 | int ret; | |
446 | ||
447 | tree = &BTRFS_I(inode)->ordered_tree; | |
49958fd7 | 448 | spin_lock(&tree->lock); |
c2167754 | 449 | ret = __btrfs_remove_ordered_extent(inode, entry); |
49958fd7 | 450 | spin_unlock(&tree->lock); |
e6dcd2dc | 451 | wake_up(&entry->wait); |
c2167754 YZ |
452 | |
453 | return ret; | |
cee36a03 CM |
454 | } |
455 | ||
d352ac68 CM |
456 | /* |
457 | * wait for all the ordered extents in a root. This is done when balancing | |
458 | * space between drives. | |
459 | */ | |
24bbcf04 YZ |
460 | int btrfs_wait_ordered_extents(struct btrfs_root *root, |
461 | int nocow_only, int delay_iput) | |
3eaa2885 CM |
462 | { |
463 | struct list_head splice; | |
464 | struct list_head *cur; | |
465 | struct btrfs_ordered_extent *ordered; | |
466 | struct inode *inode; | |
467 | ||
468 | INIT_LIST_HEAD(&splice); | |
469 | ||
470 | spin_lock(&root->fs_info->ordered_extent_lock); | |
471 | list_splice_init(&root->fs_info->ordered_extents, &splice); | |
5b21f2ed | 472 | while (!list_empty(&splice)) { |
3eaa2885 CM |
473 | cur = splice.next; |
474 | ordered = list_entry(cur, struct btrfs_ordered_extent, | |
475 | root_extent_list); | |
7ea394f1 | 476 | if (nocow_only && |
d899e052 YZ |
477 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags) && |
478 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) { | |
5b21f2ed ZY |
479 | list_move(&ordered->root_extent_list, |
480 | &root->fs_info->ordered_extents); | |
7ea394f1 YZ |
481 | cond_resched_lock(&root->fs_info->ordered_extent_lock); |
482 | continue; | |
483 | } | |
484 | ||
3eaa2885 CM |
485 | list_del_init(&ordered->root_extent_list); |
486 | atomic_inc(&ordered->refs); | |
3eaa2885 CM |
487 | |
488 | /* | |
5b21f2ed | 489 | * the inode may be getting freed (in sys_unlink path). |
3eaa2885 | 490 | */ |
5b21f2ed ZY |
491 | inode = igrab(ordered->inode); |
492 | ||
3eaa2885 CM |
493 | spin_unlock(&root->fs_info->ordered_extent_lock); |
494 | ||
5b21f2ed ZY |
495 | if (inode) { |
496 | btrfs_start_ordered_extent(inode, ordered, 1); | |
497 | btrfs_put_ordered_extent(ordered); | |
24bbcf04 YZ |
498 | if (delay_iput) |
499 | btrfs_add_delayed_iput(inode); | |
500 | else | |
501 | iput(inode); | |
5b21f2ed ZY |
502 | } else { |
503 | btrfs_put_ordered_extent(ordered); | |
504 | } | |
3eaa2885 CM |
505 | |
506 | spin_lock(&root->fs_info->ordered_extent_lock); | |
507 | } | |
508 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
509 | return 0; | |
510 | } | |
511 | ||
5a3f23d5 CM |
512 | /* |
513 | * this is used during transaction commit to write all the inodes | |
514 | * added to the ordered operation list. These files must be fully on | |
515 | * disk before the transaction commits. | |
516 | * | |
517 | * we have two modes here, one is to just start the IO via filemap_flush | |
518 | * and the other is to wait for all the io. When we wait, we have an | |
519 | * extra check to make sure the ordered operation list really is empty | |
520 | * before we return | |
521 | */ | |
522 | int btrfs_run_ordered_operations(struct btrfs_root *root, int wait) | |
523 | { | |
524 | struct btrfs_inode *btrfs_inode; | |
525 | struct inode *inode; | |
526 | struct list_head splice; | |
527 | ||
528 | INIT_LIST_HEAD(&splice); | |
529 | ||
530 | mutex_lock(&root->fs_info->ordered_operations_mutex); | |
531 | spin_lock(&root->fs_info->ordered_extent_lock); | |
532 | again: | |
533 | list_splice_init(&root->fs_info->ordered_operations, &splice); | |
534 | ||
535 | while (!list_empty(&splice)) { | |
536 | btrfs_inode = list_entry(splice.next, struct btrfs_inode, | |
537 | ordered_operations); | |
538 | ||
539 | inode = &btrfs_inode->vfs_inode; | |
540 | ||
541 | list_del_init(&btrfs_inode->ordered_operations); | |
542 | ||
543 | /* | |
544 | * the inode may be getting freed (in sys_unlink path). | |
545 | */ | |
546 | inode = igrab(inode); | |
547 | ||
548 | if (!wait && inode) { | |
549 | list_add_tail(&BTRFS_I(inode)->ordered_operations, | |
550 | &root->fs_info->ordered_operations); | |
551 | } | |
552 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
553 | ||
554 | if (inode) { | |
555 | if (wait) | |
556 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
557 | else | |
558 | filemap_flush(inode->i_mapping); | |
24bbcf04 | 559 | btrfs_add_delayed_iput(inode); |
5a3f23d5 CM |
560 | } |
561 | ||
562 | cond_resched(); | |
563 | spin_lock(&root->fs_info->ordered_extent_lock); | |
564 | } | |
565 | if (wait && !list_empty(&root->fs_info->ordered_operations)) | |
566 | goto again; | |
567 | ||
568 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
569 | mutex_unlock(&root->fs_info->ordered_operations_mutex); | |
570 | ||
571 | return 0; | |
572 | } | |
573 | ||
eb84ae03 CM |
574 | /* |
575 | * Used to start IO or wait for a given ordered extent to finish. | |
576 | * | |
577 | * If wait is one, this effectively waits on page writeback for all the pages | |
578 | * in the extent, and it waits on the io completion code to insert | |
579 | * metadata into the btree corresponding to the extent | |
580 | */ | |
581 | void btrfs_start_ordered_extent(struct inode *inode, | |
582 | struct btrfs_ordered_extent *entry, | |
583 | int wait) | |
e6dcd2dc CM |
584 | { |
585 | u64 start = entry->file_offset; | |
586 | u64 end = start + entry->len - 1; | |
e1b81e67 | 587 | |
eb84ae03 CM |
588 | /* |
589 | * pages in the range can be dirty, clean or writeback. We | |
590 | * start IO on any dirty ones so the wait doesn't stall waiting | |
591 | * for pdflush to find them | |
592 | */ | |
4b46fce2 JB |
593 | if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags)) |
594 | filemap_fdatawrite_range(inode->i_mapping, start, end); | |
c8b97818 | 595 | if (wait) { |
e6dcd2dc CM |
596 | wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE, |
597 | &entry->flags)); | |
c8b97818 | 598 | } |
e6dcd2dc | 599 | } |
cee36a03 | 600 | |
eb84ae03 CM |
601 | /* |
602 | * Used to wait on ordered extents across a large range of bytes. | |
603 | */ | |
cb843a6f | 604 | int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len) |
e6dcd2dc CM |
605 | { |
606 | u64 end; | |
e5a2217e | 607 | u64 orig_end; |
e6dcd2dc | 608 | struct btrfs_ordered_extent *ordered; |
8b62b72b | 609 | int found; |
e5a2217e CM |
610 | |
611 | if (start + len < start) { | |
f421950f | 612 | orig_end = INT_LIMIT(loff_t); |
e5a2217e CM |
613 | } else { |
614 | orig_end = start + len - 1; | |
f421950f CM |
615 | if (orig_end > INT_LIMIT(loff_t)) |
616 | orig_end = INT_LIMIT(loff_t); | |
e5a2217e | 617 | } |
4a096752 | 618 | again: |
e5a2217e CM |
619 | /* start IO across the range first to instantiate any delalloc |
620 | * extents | |
621 | */ | |
8aa38c31 | 622 | filemap_fdatawrite_range(inode->i_mapping, start, orig_end); |
f421950f | 623 | |
771ed689 CM |
624 | /* The compression code will leave pages locked but return from |
625 | * writepage without setting the page writeback. Starting again | |
626 | * with WB_SYNC_ALL will end up waiting for the IO to actually start. | |
627 | */ | |
8aa38c31 | 628 | filemap_fdatawrite_range(inode->i_mapping, start, orig_end); |
771ed689 | 629 | |
8aa38c31 | 630 | filemap_fdatawait_range(inode->i_mapping, start, orig_end); |
e5a2217e | 631 | |
f421950f | 632 | end = orig_end; |
8b62b72b | 633 | found = 0; |
d397712b | 634 | while (1) { |
e6dcd2dc | 635 | ordered = btrfs_lookup_first_ordered_extent(inode, end); |
d397712b | 636 | if (!ordered) |
e6dcd2dc | 637 | break; |
e5a2217e | 638 | if (ordered->file_offset > orig_end) { |
e6dcd2dc CM |
639 | btrfs_put_ordered_extent(ordered); |
640 | break; | |
641 | } | |
642 | if (ordered->file_offset + ordered->len < start) { | |
643 | btrfs_put_ordered_extent(ordered); | |
644 | break; | |
645 | } | |
8b62b72b | 646 | found++; |
e5a2217e | 647 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
648 | end = ordered->file_offset; |
649 | btrfs_put_ordered_extent(ordered); | |
e5a2217e | 650 | if (end == 0 || end == start) |
e6dcd2dc CM |
651 | break; |
652 | end--; | |
653 | } | |
8b62b72b CM |
654 | if (found || test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end, |
655 | EXTENT_DELALLOC, 0, NULL)) { | |
771ed689 | 656 | schedule_timeout(1); |
4a096752 CM |
657 | goto again; |
658 | } | |
cb843a6f | 659 | return 0; |
cee36a03 CM |
660 | } |
661 | ||
eb84ae03 CM |
662 | /* |
663 | * find an ordered extent corresponding to file_offset. return NULL if | |
664 | * nothing is found, otherwise take a reference on the extent and return it | |
665 | */ | |
e6dcd2dc CM |
666 | struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode, |
667 | u64 file_offset) | |
668 | { | |
669 | struct btrfs_ordered_inode_tree *tree; | |
670 | struct rb_node *node; | |
671 | struct btrfs_ordered_extent *entry = NULL; | |
672 | ||
673 | tree = &BTRFS_I(inode)->ordered_tree; | |
49958fd7 | 674 | spin_lock(&tree->lock); |
e6dcd2dc CM |
675 | node = tree_search(tree, file_offset); |
676 | if (!node) | |
677 | goto out; | |
678 | ||
679 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
680 | if (!offset_in_entry(entry, file_offset)) | |
681 | entry = NULL; | |
682 | if (entry) | |
683 | atomic_inc(&entry->refs); | |
684 | out: | |
49958fd7 | 685 | spin_unlock(&tree->lock); |
e6dcd2dc CM |
686 | return entry; |
687 | } | |
688 | ||
4b46fce2 JB |
689 | /* Since the DIO code tries to lock a wide area we need to look for any ordered |
690 | * extents that exist in the range, rather than just the start of the range. | |
691 | */ | |
692 | struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode, | |
693 | u64 file_offset, | |
694 | u64 len) | |
695 | { | |
696 | struct btrfs_ordered_inode_tree *tree; | |
697 | struct rb_node *node; | |
698 | struct btrfs_ordered_extent *entry = NULL; | |
699 | ||
700 | tree = &BTRFS_I(inode)->ordered_tree; | |
701 | spin_lock(&tree->lock); | |
702 | node = tree_search(tree, file_offset); | |
703 | if (!node) { | |
704 | node = tree_search(tree, file_offset + len); | |
705 | if (!node) | |
706 | goto out; | |
707 | } | |
708 | ||
709 | while (1) { | |
710 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
711 | if (range_overlaps(entry, file_offset, len)) | |
712 | break; | |
713 | ||
714 | if (entry->file_offset >= file_offset + len) { | |
715 | entry = NULL; | |
716 | break; | |
717 | } | |
718 | entry = NULL; | |
719 | node = rb_next(node); | |
720 | if (!node) | |
721 | break; | |
722 | } | |
723 | out: | |
724 | if (entry) | |
725 | atomic_inc(&entry->refs); | |
726 | spin_unlock(&tree->lock); | |
727 | return entry; | |
728 | } | |
729 | ||
eb84ae03 CM |
730 | /* |
731 | * lookup and return any extent before 'file_offset'. NULL is returned | |
732 | * if none is found | |
733 | */ | |
e6dcd2dc | 734 | struct btrfs_ordered_extent * |
d397712b | 735 | btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset) |
e6dcd2dc CM |
736 | { |
737 | struct btrfs_ordered_inode_tree *tree; | |
738 | struct rb_node *node; | |
739 | struct btrfs_ordered_extent *entry = NULL; | |
740 | ||
741 | tree = &BTRFS_I(inode)->ordered_tree; | |
49958fd7 | 742 | spin_lock(&tree->lock); |
e6dcd2dc CM |
743 | node = tree_search(tree, file_offset); |
744 | if (!node) | |
745 | goto out; | |
746 | ||
747 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
748 | atomic_inc(&entry->refs); | |
749 | out: | |
49958fd7 | 750 | spin_unlock(&tree->lock); |
e6dcd2dc | 751 | return entry; |
81d7ed29 | 752 | } |
dbe674a9 | 753 | |
eb84ae03 CM |
754 | /* |
755 | * After an extent is done, call this to conditionally update the on disk | |
756 | * i_size. i_size is updated to cover any fully written part of the file. | |
757 | */ | |
c2167754 | 758 | int btrfs_ordered_update_i_size(struct inode *inode, u64 offset, |
dbe674a9 CM |
759 | struct btrfs_ordered_extent *ordered) |
760 | { | |
761 | struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree; | |
762 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
763 | u64 disk_i_size; | |
764 | u64 new_i_size; | |
765 | u64 i_size_test; | |
c2167754 | 766 | u64 i_size = i_size_read(inode); |
dbe674a9 | 767 | struct rb_node *node; |
c2167754 | 768 | struct rb_node *prev = NULL; |
dbe674a9 | 769 | struct btrfs_ordered_extent *test; |
c2167754 YZ |
770 | int ret = 1; |
771 | ||
772 | if (ordered) | |
773 | offset = entry_end(ordered); | |
a038fab0 YZ |
774 | else |
775 | offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize); | |
dbe674a9 | 776 | |
49958fd7 | 777 | spin_lock(&tree->lock); |
dbe674a9 CM |
778 | disk_i_size = BTRFS_I(inode)->disk_i_size; |
779 | ||
c2167754 YZ |
780 | /* truncate file */ |
781 | if (disk_i_size > i_size) { | |
782 | BTRFS_I(inode)->disk_i_size = i_size; | |
783 | ret = 0; | |
784 | goto out; | |
785 | } | |
786 | ||
dbe674a9 CM |
787 | /* |
788 | * if the disk i_size is already at the inode->i_size, or | |
789 | * this ordered extent is inside the disk i_size, we're done | |
790 | */ | |
c2167754 | 791 | if (disk_i_size == i_size || offset <= disk_i_size) { |
dbe674a9 CM |
792 | goto out; |
793 | } | |
794 | ||
795 | /* | |
796 | * we can't update the disk_isize if there are delalloc bytes | |
797 | * between disk_i_size and this ordered extent | |
798 | */ | |
c2167754 | 799 | if (test_range_bit(io_tree, disk_i_size, offset - 1, |
9655d298 | 800 | EXTENT_DELALLOC, 0, NULL)) { |
dbe674a9 CM |
801 | goto out; |
802 | } | |
803 | /* | |
804 | * walk backward from this ordered extent to disk_i_size. | |
805 | * if we find an ordered extent then we can't update disk i_size | |
806 | * yet | |
807 | */ | |
c2167754 YZ |
808 | if (ordered) { |
809 | node = rb_prev(&ordered->rb_node); | |
810 | } else { | |
811 | prev = tree_search(tree, offset); | |
812 | /* | |
813 | * we insert file extents without involving ordered struct, | |
814 | * so there should be no ordered struct cover this offset | |
815 | */ | |
816 | if (prev) { | |
817 | test = rb_entry(prev, struct btrfs_ordered_extent, | |
818 | rb_node); | |
819 | BUG_ON(offset_in_entry(test, offset)); | |
820 | } | |
821 | node = prev; | |
822 | } | |
823 | while (node) { | |
dbe674a9 CM |
824 | test = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
825 | if (test->file_offset + test->len <= disk_i_size) | |
826 | break; | |
c2167754 | 827 | if (test->file_offset >= i_size) |
dbe674a9 CM |
828 | break; |
829 | if (test->file_offset >= disk_i_size) | |
830 | goto out; | |
c2167754 | 831 | node = rb_prev(node); |
dbe674a9 | 832 | } |
c2167754 | 833 | new_i_size = min_t(u64, offset, i_size); |
dbe674a9 CM |
834 | |
835 | /* | |
836 | * at this point, we know we can safely update i_size to at least | |
837 | * the offset from this ordered extent. But, we need to | |
838 | * walk forward and see if ios from higher up in the file have | |
839 | * finished. | |
840 | */ | |
c2167754 YZ |
841 | if (ordered) { |
842 | node = rb_next(&ordered->rb_node); | |
843 | } else { | |
844 | if (prev) | |
845 | node = rb_next(prev); | |
846 | else | |
847 | node = rb_first(&tree->tree); | |
848 | } | |
dbe674a9 CM |
849 | i_size_test = 0; |
850 | if (node) { | |
851 | /* | |
852 | * do we have an area where IO might have finished | |
853 | * between our ordered extent and the next one. | |
854 | */ | |
855 | test = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
c2167754 | 856 | if (test->file_offset > offset) |
b48652c1 | 857 | i_size_test = test->file_offset; |
dbe674a9 | 858 | } else { |
c2167754 | 859 | i_size_test = i_size; |
dbe674a9 CM |
860 | } |
861 | ||
862 | /* | |
863 | * i_size_test is the end of a region after this ordered | |
864 | * extent where there are no ordered extents. As long as there | |
865 | * are no delalloc bytes in this area, it is safe to update | |
866 | * disk_i_size to the end of the region. | |
867 | */ | |
c2167754 YZ |
868 | if (i_size_test > offset && |
869 | !test_range_bit(io_tree, offset, i_size_test - 1, | |
870 | EXTENT_DELALLOC, 0, NULL)) { | |
871 | new_i_size = min_t(u64, i_size_test, i_size); | |
dbe674a9 CM |
872 | } |
873 | BTRFS_I(inode)->disk_i_size = new_i_size; | |
c2167754 | 874 | ret = 0; |
dbe674a9 | 875 | out: |
c2167754 YZ |
876 | /* |
877 | * we need to remove the ordered extent with the tree lock held | |
878 | * so that other people calling this function don't find our fully | |
879 | * processed ordered entry and skip updating the i_size | |
880 | */ | |
881 | if (ordered) | |
882 | __btrfs_remove_ordered_extent(inode, ordered); | |
49958fd7 | 883 | spin_unlock(&tree->lock); |
c2167754 YZ |
884 | if (ordered) |
885 | wake_up(&ordered->wait); | |
886 | return ret; | |
dbe674a9 | 887 | } |
ba1da2f4 | 888 | |
eb84ae03 CM |
889 | /* |
890 | * search the ordered extents for one corresponding to 'offset' and | |
891 | * try to find a checksum. This is used because we allow pages to | |
892 | * be reclaimed before their checksum is actually put into the btree | |
893 | */ | |
d20f7043 CM |
894 | int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr, |
895 | u32 *sum) | |
ba1da2f4 CM |
896 | { |
897 | struct btrfs_ordered_sum *ordered_sum; | |
898 | struct btrfs_sector_sum *sector_sums; | |
899 | struct btrfs_ordered_extent *ordered; | |
900 | struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree; | |
3edf7d33 CM |
901 | unsigned long num_sectors; |
902 | unsigned long i; | |
903 | u32 sectorsize = BTRFS_I(inode)->root->sectorsize; | |
ba1da2f4 | 904 | int ret = 1; |
ba1da2f4 CM |
905 | |
906 | ordered = btrfs_lookup_ordered_extent(inode, offset); | |
907 | if (!ordered) | |
908 | return 1; | |
909 | ||
49958fd7 | 910 | spin_lock(&tree->lock); |
c6e30871 | 911 | list_for_each_entry_reverse(ordered_sum, &ordered->list, list) { |
d20f7043 | 912 | if (disk_bytenr >= ordered_sum->bytenr) { |
3edf7d33 | 913 | num_sectors = ordered_sum->len / sectorsize; |
ed98b56a | 914 | sector_sums = ordered_sum->sums; |
3edf7d33 | 915 | for (i = 0; i < num_sectors; i++) { |
d20f7043 | 916 | if (sector_sums[i].bytenr == disk_bytenr) { |
3edf7d33 CM |
917 | *sum = sector_sums[i].sum; |
918 | ret = 0; | |
919 | goto out; | |
920 | } | |
921 | } | |
ba1da2f4 CM |
922 | } |
923 | } | |
924 | out: | |
49958fd7 | 925 | spin_unlock(&tree->lock); |
89642229 | 926 | btrfs_put_ordered_extent(ordered); |
ba1da2f4 CM |
927 | return ret; |
928 | } | |
929 | ||
f421950f | 930 | |
5a3f23d5 CM |
931 | /* |
932 | * add a given inode to the list of inodes that must be fully on | |
933 | * disk before a transaction commit finishes. | |
934 | * | |
935 | * This basically gives us the ext3 style data=ordered mode, and it is mostly | |
936 | * used to make sure renamed files are fully on disk. | |
937 | * | |
938 | * It is a noop if the inode is already fully on disk. | |
939 | * | |
940 | * If trans is not null, we'll do a friendly check for a transaction that | |
941 | * is already flushing things and force the IO down ourselves. | |
942 | */ | |
943 | int btrfs_add_ordered_operation(struct btrfs_trans_handle *trans, | |
944 | struct btrfs_root *root, | |
945 | struct inode *inode) | |
946 | { | |
947 | u64 last_mod; | |
948 | ||
949 | last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans); | |
950 | ||
951 | /* | |
952 | * if this file hasn't been changed since the last transaction | |
953 | * commit, we can safely return without doing anything | |
954 | */ | |
955 | if (last_mod < root->fs_info->last_trans_committed) | |
956 | return 0; | |
957 | ||
958 | /* | |
959 | * the transaction is already committing. Just start the IO and | |
960 | * don't bother with all of this list nonsense | |
961 | */ | |
962 | if (trans && root->fs_info->running_transaction->blocked) { | |
963 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
964 | return 0; | |
965 | } | |
966 | ||
967 | spin_lock(&root->fs_info->ordered_extent_lock); | |
968 | if (list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
969 | list_add_tail(&BTRFS_I(inode)->ordered_operations, | |
970 | &root->fs_info->ordered_operations); | |
971 | } | |
972 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
973 | ||
974 | return 0; | |
975 | } |