]>
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 | |
1abe9b8a | 205 | trace_btrfs_ordered_extent_add(inode, entry); |
206 | ||
49958fd7 | 207 | spin_lock(&tree->lock); |
e6dcd2dc CM |
208 | node = tree_insert(&tree->tree, file_offset, |
209 | &entry->rb_node); | |
d397712b | 210 | BUG_ON(node); |
49958fd7 | 211 | spin_unlock(&tree->lock); |
d397712b | 212 | |
3eaa2885 CM |
213 | spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock); |
214 | list_add_tail(&entry->root_extent_list, | |
215 | &BTRFS_I(inode)->root->fs_info->ordered_extents); | |
216 | spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock); | |
217 | ||
e6dcd2dc | 218 | BUG_ON(node); |
dc17ff8f CM |
219 | return 0; |
220 | } | |
221 | ||
4b46fce2 JB |
222 | int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset, |
223 | u64 start, u64 len, u64 disk_len, int type) | |
224 | { | |
225 | return __btrfs_add_ordered_extent(inode, file_offset, start, len, | |
261507a0 LZ |
226 | disk_len, type, 0, |
227 | BTRFS_COMPRESS_NONE); | |
4b46fce2 JB |
228 | } |
229 | ||
230 | int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset, | |
231 | u64 start, u64 len, u64 disk_len, int type) | |
232 | { | |
233 | return __btrfs_add_ordered_extent(inode, file_offset, start, len, | |
261507a0 LZ |
234 | disk_len, type, 1, |
235 | BTRFS_COMPRESS_NONE); | |
236 | } | |
237 | ||
238 | int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset, | |
239 | u64 start, u64 len, u64 disk_len, | |
240 | int type, int compress_type) | |
241 | { | |
242 | return __btrfs_add_ordered_extent(inode, file_offset, start, len, | |
243 | disk_len, type, 0, | |
244 | compress_type); | |
4b46fce2 JB |
245 | } |
246 | ||
eb84ae03 CM |
247 | /* |
248 | * Add a struct btrfs_ordered_sum into the list of checksums to be inserted | |
3edf7d33 CM |
249 | * when an ordered extent is finished. If the list covers more than one |
250 | * ordered extent, it is split across multiples. | |
eb84ae03 | 251 | */ |
3edf7d33 CM |
252 | int btrfs_add_ordered_sum(struct inode *inode, |
253 | struct btrfs_ordered_extent *entry, | |
254 | struct btrfs_ordered_sum *sum) | |
dc17ff8f | 255 | { |
e6dcd2dc | 256 | struct btrfs_ordered_inode_tree *tree; |
dc17ff8f | 257 | |
e6dcd2dc | 258 | tree = &BTRFS_I(inode)->ordered_tree; |
49958fd7 | 259 | spin_lock(&tree->lock); |
e6dcd2dc | 260 | list_add_tail(&sum->list, &entry->list); |
49958fd7 | 261 | spin_unlock(&tree->lock); |
e6dcd2dc | 262 | return 0; |
dc17ff8f CM |
263 | } |
264 | ||
163cf09c CM |
265 | /* |
266 | * this is used to account for finished IO across a given range | |
267 | * of the file. The IO may span ordered extents. If | |
268 | * a given ordered_extent is completely done, 1 is returned, otherwise | |
269 | * 0. | |
270 | * | |
271 | * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used | |
272 | * to make sure this function only returns 1 once for a given ordered extent. | |
273 | * | |
274 | * file_offset is updated to one byte past the range that is recorded as | |
275 | * complete. This allows you to walk forward in the file. | |
276 | */ | |
277 | int btrfs_dec_test_first_ordered_pending(struct inode *inode, | |
278 | struct btrfs_ordered_extent **cached, | |
279 | u64 *file_offset, u64 io_size) | |
280 | { | |
281 | struct btrfs_ordered_inode_tree *tree; | |
282 | struct rb_node *node; | |
283 | struct btrfs_ordered_extent *entry = NULL; | |
284 | int ret; | |
285 | u64 dec_end; | |
286 | u64 dec_start; | |
287 | u64 to_dec; | |
288 | ||
289 | tree = &BTRFS_I(inode)->ordered_tree; | |
290 | spin_lock(&tree->lock); | |
291 | node = tree_search(tree, *file_offset); | |
292 | if (!node) { | |
293 | ret = 1; | |
294 | goto out; | |
295 | } | |
296 | ||
297 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
298 | if (!offset_in_entry(entry, *file_offset)) { | |
299 | ret = 1; | |
300 | goto out; | |
301 | } | |
302 | ||
303 | dec_start = max(*file_offset, entry->file_offset); | |
304 | dec_end = min(*file_offset + io_size, entry->file_offset + | |
305 | entry->len); | |
306 | *file_offset = dec_end; | |
307 | if (dec_start > dec_end) { | |
308 | printk(KERN_CRIT "bad ordering dec_start %llu end %llu\n", | |
309 | (unsigned long long)dec_start, | |
310 | (unsigned long long)dec_end); | |
311 | } | |
312 | to_dec = dec_end - dec_start; | |
313 | if (to_dec > entry->bytes_left) { | |
314 | printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n", | |
315 | (unsigned long long)entry->bytes_left, | |
316 | (unsigned long long)to_dec); | |
317 | } | |
318 | entry->bytes_left -= to_dec; | |
319 | if (entry->bytes_left == 0) | |
320 | ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags); | |
321 | else | |
322 | ret = 1; | |
323 | out: | |
324 | if (!ret && cached && entry) { | |
325 | *cached = entry; | |
326 | atomic_inc(&entry->refs); | |
327 | } | |
328 | spin_unlock(&tree->lock); | |
329 | return ret == 0; | |
330 | } | |
331 | ||
eb84ae03 CM |
332 | /* |
333 | * this is used to account for finished IO across a given range | |
334 | * of the file. The IO should not span ordered extents. If | |
335 | * a given ordered_extent is completely done, 1 is returned, otherwise | |
336 | * 0. | |
337 | * | |
338 | * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used | |
339 | * to make sure this function only returns 1 once for a given ordered extent. | |
340 | */ | |
e6dcd2dc | 341 | int btrfs_dec_test_ordered_pending(struct inode *inode, |
5a1a3df1 | 342 | struct btrfs_ordered_extent **cached, |
e6dcd2dc | 343 | u64 file_offset, u64 io_size) |
dc17ff8f | 344 | { |
e6dcd2dc | 345 | struct btrfs_ordered_inode_tree *tree; |
dc17ff8f | 346 | struct rb_node *node; |
5a1a3df1 | 347 | struct btrfs_ordered_extent *entry = NULL; |
e6dcd2dc CM |
348 | int ret; |
349 | ||
350 | tree = &BTRFS_I(inode)->ordered_tree; | |
49958fd7 | 351 | spin_lock(&tree->lock); |
e6dcd2dc | 352 | node = tree_search(tree, file_offset); |
dc17ff8f | 353 | if (!node) { |
e6dcd2dc CM |
354 | ret = 1; |
355 | goto out; | |
dc17ff8f CM |
356 | } |
357 | ||
e6dcd2dc CM |
358 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
359 | if (!offset_in_entry(entry, file_offset)) { | |
360 | ret = 1; | |
361 | goto out; | |
dc17ff8f | 362 | } |
e6dcd2dc | 363 | |
8b62b72b CM |
364 | if (io_size > entry->bytes_left) { |
365 | printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n", | |
366 | (unsigned long long)entry->bytes_left, | |
367 | (unsigned long long)io_size); | |
368 | } | |
369 | entry->bytes_left -= io_size; | |
370 | if (entry->bytes_left == 0) | |
e6dcd2dc | 371 | ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags); |
8b62b72b CM |
372 | else |
373 | ret = 1; | |
e6dcd2dc | 374 | out: |
5a1a3df1 JB |
375 | if (!ret && cached && entry) { |
376 | *cached = entry; | |
377 | atomic_inc(&entry->refs); | |
378 | } | |
49958fd7 | 379 | spin_unlock(&tree->lock); |
e6dcd2dc CM |
380 | return ret == 0; |
381 | } | |
dc17ff8f | 382 | |
eb84ae03 CM |
383 | /* |
384 | * used to drop a reference on an ordered extent. This will free | |
385 | * the extent if the last reference is dropped | |
386 | */ | |
e6dcd2dc CM |
387 | int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry) |
388 | { | |
ba1da2f4 CM |
389 | struct list_head *cur; |
390 | struct btrfs_ordered_sum *sum; | |
391 | ||
1abe9b8a | 392 | trace_btrfs_ordered_extent_put(entry->inode, entry); |
393 | ||
ba1da2f4 | 394 | if (atomic_dec_and_test(&entry->refs)) { |
d397712b | 395 | while (!list_empty(&entry->list)) { |
ba1da2f4 CM |
396 | cur = entry->list.next; |
397 | sum = list_entry(cur, struct btrfs_ordered_sum, list); | |
398 | list_del(&sum->list); | |
399 | kfree(sum); | |
400 | } | |
e6dcd2dc | 401 | kfree(entry); |
ba1da2f4 | 402 | } |
e6dcd2dc | 403 | return 0; |
dc17ff8f | 404 | } |
cee36a03 | 405 | |
eb84ae03 CM |
406 | /* |
407 | * remove an ordered extent from the tree. No references are dropped | |
49958fd7 | 408 | * and you must wake_up entry->wait. You must hold the tree lock |
c2167754 | 409 | * while you call this function. |
eb84ae03 | 410 | */ |
c2167754 | 411 | static int __btrfs_remove_ordered_extent(struct inode *inode, |
e6dcd2dc | 412 | struct btrfs_ordered_extent *entry) |
cee36a03 | 413 | { |
e6dcd2dc | 414 | struct btrfs_ordered_inode_tree *tree; |
287a0ab9 | 415 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cee36a03 | 416 | struct rb_node *node; |
cee36a03 | 417 | |
e6dcd2dc | 418 | tree = &BTRFS_I(inode)->ordered_tree; |
e6dcd2dc | 419 | node = &entry->rb_node; |
cee36a03 | 420 | rb_erase(node, &tree->tree); |
e6dcd2dc CM |
421 | tree->last = NULL; |
422 | set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags); | |
3eaa2885 | 423 | |
287a0ab9 | 424 | spin_lock(&root->fs_info->ordered_extent_lock); |
3eaa2885 | 425 | list_del_init(&entry->root_extent_list); |
5a3f23d5 | 426 | |
1abe9b8a | 427 | trace_btrfs_ordered_extent_remove(inode, entry); |
428 | ||
5a3f23d5 CM |
429 | /* |
430 | * we have no more ordered extents for this inode and | |
431 | * no dirty pages. We can safely remove it from the | |
432 | * list of ordered extents | |
433 | */ | |
434 | if (RB_EMPTY_ROOT(&tree->tree) && | |
435 | !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { | |
436 | list_del_init(&BTRFS_I(inode)->ordered_operations); | |
437 | } | |
287a0ab9 | 438 | spin_unlock(&root->fs_info->ordered_extent_lock); |
3eaa2885 | 439 | |
c2167754 YZ |
440 | return 0; |
441 | } | |
442 | ||
443 | /* | |
444 | * remove an ordered extent from the tree. No references are dropped | |
445 | * but any waiters are woken. | |
446 | */ | |
447 | int btrfs_remove_ordered_extent(struct inode *inode, | |
448 | struct btrfs_ordered_extent *entry) | |
449 | { | |
450 | struct btrfs_ordered_inode_tree *tree; | |
451 | int ret; | |
452 | ||
453 | tree = &BTRFS_I(inode)->ordered_tree; | |
49958fd7 | 454 | spin_lock(&tree->lock); |
c2167754 | 455 | ret = __btrfs_remove_ordered_extent(inode, entry); |
49958fd7 | 456 | spin_unlock(&tree->lock); |
e6dcd2dc | 457 | wake_up(&entry->wait); |
c2167754 YZ |
458 | |
459 | return ret; | |
cee36a03 CM |
460 | } |
461 | ||
d352ac68 CM |
462 | /* |
463 | * wait for all the ordered extents in a root. This is done when balancing | |
464 | * space between drives. | |
465 | */ | |
24bbcf04 YZ |
466 | int btrfs_wait_ordered_extents(struct btrfs_root *root, |
467 | int nocow_only, int delay_iput) | |
3eaa2885 CM |
468 | { |
469 | struct list_head splice; | |
470 | struct list_head *cur; | |
471 | struct btrfs_ordered_extent *ordered; | |
472 | struct inode *inode; | |
473 | ||
474 | INIT_LIST_HEAD(&splice); | |
475 | ||
476 | spin_lock(&root->fs_info->ordered_extent_lock); | |
477 | list_splice_init(&root->fs_info->ordered_extents, &splice); | |
5b21f2ed | 478 | while (!list_empty(&splice)) { |
3eaa2885 CM |
479 | cur = splice.next; |
480 | ordered = list_entry(cur, struct btrfs_ordered_extent, | |
481 | root_extent_list); | |
7ea394f1 | 482 | if (nocow_only && |
d899e052 YZ |
483 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags) && |
484 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) { | |
5b21f2ed ZY |
485 | list_move(&ordered->root_extent_list, |
486 | &root->fs_info->ordered_extents); | |
7ea394f1 YZ |
487 | cond_resched_lock(&root->fs_info->ordered_extent_lock); |
488 | continue; | |
489 | } | |
490 | ||
3eaa2885 CM |
491 | list_del_init(&ordered->root_extent_list); |
492 | atomic_inc(&ordered->refs); | |
3eaa2885 CM |
493 | |
494 | /* | |
5b21f2ed | 495 | * the inode may be getting freed (in sys_unlink path). |
3eaa2885 | 496 | */ |
5b21f2ed ZY |
497 | inode = igrab(ordered->inode); |
498 | ||
3eaa2885 CM |
499 | spin_unlock(&root->fs_info->ordered_extent_lock); |
500 | ||
5b21f2ed ZY |
501 | if (inode) { |
502 | btrfs_start_ordered_extent(inode, ordered, 1); | |
503 | btrfs_put_ordered_extent(ordered); | |
24bbcf04 YZ |
504 | if (delay_iput) |
505 | btrfs_add_delayed_iput(inode); | |
506 | else | |
507 | iput(inode); | |
5b21f2ed ZY |
508 | } else { |
509 | btrfs_put_ordered_extent(ordered); | |
510 | } | |
3eaa2885 CM |
511 | |
512 | spin_lock(&root->fs_info->ordered_extent_lock); | |
513 | } | |
514 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
515 | return 0; | |
516 | } | |
517 | ||
5a3f23d5 CM |
518 | /* |
519 | * this is used during transaction commit to write all the inodes | |
520 | * added to the ordered operation list. These files must be fully on | |
521 | * disk before the transaction commits. | |
522 | * | |
523 | * we have two modes here, one is to just start the IO via filemap_flush | |
524 | * and the other is to wait for all the io. When we wait, we have an | |
525 | * extra check to make sure the ordered operation list really is empty | |
526 | * before we return | |
527 | */ | |
528 | int btrfs_run_ordered_operations(struct btrfs_root *root, int wait) | |
529 | { | |
530 | struct btrfs_inode *btrfs_inode; | |
531 | struct inode *inode; | |
532 | struct list_head splice; | |
533 | ||
534 | INIT_LIST_HEAD(&splice); | |
535 | ||
536 | mutex_lock(&root->fs_info->ordered_operations_mutex); | |
537 | spin_lock(&root->fs_info->ordered_extent_lock); | |
538 | again: | |
539 | list_splice_init(&root->fs_info->ordered_operations, &splice); | |
540 | ||
541 | while (!list_empty(&splice)) { | |
542 | btrfs_inode = list_entry(splice.next, struct btrfs_inode, | |
543 | ordered_operations); | |
544 | ||
545 | inode = &btrfs_inode->vfs_inode; | |
546 | ||
547 | list_del_init(&btrfs_inode->ordered_operations); | |
548 | ||
549 | /* | |
550 | * the inode may be getting freed (in sys_unlink path). | |
551 | */ | |
552 | inode = igrab(inode); | |
553 | ||
554 | if (!wait && inode) { | |
555 | list_add_tail(&BTRFS_I(inode)->ordered_operations, | |
556 | &root->fs_info->ordered_operations); | |
557 | } | |
558 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
559 | ||
560 | if (inode) { | |
561 | if (wait) | |
562 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
563 | else | |
564 | filemap_flush(inode->i_mapping); | |
24bbcf04 | 565 | btrfs_add_delayed_iput(inode); |
5a3f23d5 CM |
566 | } |
567 | ||
568 | cond_resched(); | |
569 | spin_lock(&root->fs_info->ordered_extent_lock); | |
570 | } | |
571 | if (wait && !list_empty(&root->fs_info->ordered_operations)) | |
572 | goto again; | |
573 | ||
574 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
575 | mutex_unlock(&root->fs_info->ordered_operations_mutex); | |
576 | ||
577 | return 0; | |
578 | } | |
579 | ||
eb84ae03 CM |
580 | /* |
581 | * Used to start IO or wait for a given ordered extent to finish. | |
582 | * | |
583 | * If wait is one, this effectively waits on page writeback for all the pages | |
584 | * in the extent, and it waits on the io completion code to insert | |
585 | * metadata into the btree corresponding to the extent | |
586 | */ | |
587 | void btrfs_start_ordered_extent(struct inode *inode, | |
588 | struct btrfs_ordered_extent *entry, | |
589 | int wait) | |
e6dcd2dc CM |
590 | { |
591 | u64 start = entry->file_offset; | |
592 | u64 end = start + entry->len - 1; | |
e1b81e67 | 593 | |
1abe9b8a | 594 | trace_btrfs_ordered_extent_start(inode, entry); |
595 | ||
eb84ae03 CM |
596 | /* |
597 | * pages in the range can be dirty, clean or writeback. We | |
598 | * start IO on any dirty ones so the wait doesn't stall waiting | |
599 | * for pdflush to find them | |
600 | */ | |
4b46fce2 JB |
601 | if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags)) |
602 | filemap_fdatawrite_range(inode->i_mapping, start, end); | |
c8b97818 | 603 | if (wait) { |
e6dcd2dc CM |
604 | wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE, |
605 | &entry->flags)); | |
c8b97818 | 606 | } |
e6dcd2dc | 607 | } |
cee36a03 | 608 | |
eb84ae03 CM |
609 | /* |
610 | * Used to wait on ordered extents across a large range of bytes. | |
611 | */ | |
cb843a6f | 612 | int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len) |
e6dcd2dc CM |
613 | { |
614 | u64 end; | |
e5a2217e | 615 | u64 orig_end; |
e6dcd2dc | 616 | struct btrfs_ordered_extent *ordered; |
8b62b72b | 617 | int found; |
e5a2217e CM |
618 | |
619 | if (start + len < start) { | |
f421950f | 620 | orig_end = INT_LIMIT(loff_t); |
e5a2217e CM |
621 | } else { |
622 | orig_end = start + len - 1; | |
f421950f CM |
623 | if (orig_end > INT_LIMIT(loff_t)) |
624 | orig_end = INT_LIMIT(loff_t); | |
e5a2217e | 625 | } |
4a096752 | 626 | again: |
e5a2217e CM |
627 | /* start IO across the range first to instantiate any delalloc |
628 | * extents | |
629 | */ | |
8aa38c31 | 630 | filemap_fdatawrite_range(inode->i_mapping, start, orig_end); |
f421950f | 631 | |
771ed689 CM |
632 | /* The compression code will leave pages locked but return from |
633 | * writepage without setting the page writeback. Starting again | |
634 | * with WB_SYNC_ALL will end up waiting for the IO to actually start. | |
635 | */ | |
8aa38c31 | 636 | filemap_fdatawrite_range(inode->i_mapping, start, orig_end); |
771ed689 | 637 | |
8aa38c31 | 638 | filemap_fdatawait_range(inode->i_mapping, start, orig_end); |
e5a2217e | 639 | |
f421950f | 640 | end = orig_end; |
8b62b72b | 641 | found = 0; |
d397712b | 642 | while (1) { |
e6dcd2dc | 643 | ordered = btrfs_lookup_first_ordered_extent(inode, end); |
d397712b | 644 | if (!ordered) |
e6dcd2dc | 645 | break; |
e5a2217e | 646 | if (ordered->file_offset > orig_end) { |
e6dcd2dc CM |
647 | btrfs_put_ordered_extent(ordered); |
648 | break; | |
649 | } | |
650 | if (ordered->file_offset + ordered->len < start) { | |
651 | btrfs_put_ordered_extent(ordered); | |
652 | break; | |
653 | } | |
8b62b72b | 654 | found++; |
e5a2217e | 655 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
656 | end = ordered->file_offset; |
657 | btrfs_put_ordered_extent(ordered); | |
e5a2217e | 658 | if (end == 0 || end == start) |
e6dcd2dc CM |
659 | break; |
660 | end--; | |
661 | } | |
8b62b72b CM |
662 | if (found || test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end, |
663 | EXTENT_DELALLOC, 0, NULL)) { | |
771ed689 | 664 | schedule_timeout(1); |
4a096752 CM |
665 | goto again; |
666 | } | |
cb843a6f | 667 | return 0; |
cee36a03 CM |
668 | } |
669 | ||
eb84ae03 CM |
670 | /* |
671 | * find an ordered extent corresponding to file_offset. return NULL if | |
672 | * nothing is found, otherwise take a reference on the extent and return it | |
673 | */ | |
e6dcd2dc CM |
674 | struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode, |
675 | u64 file_offset) | |
676 | { | |
677 | struct btrfs_ordered_inode_tree *tree; | |
678 | struct rb_node *node; | |
679 | struct btrfs_ordered_extent *entry = NULL; | |
680 | ||
681 | tree = &BTRFS_I(inode)->ordered_tree; | |
49958fd7 | 682 | spin_lock(&tree->lock); |
e6dcd2dc CM |
683 | node = tree_search(tree, file_offset); |
684 | if (!node) | |
685 | goto out; | |
686 | ||
687 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
688 | if (!offset_in_entry(entry, file_offset)) | |
689 | entry = NULL; | |
690 | if (entry) | |
691 | atomic_inc(&entry->refs); | |
692 | out: | |
49958fd7 | 693 | spin_unlock(&tree->lock); |
e6dcd2dc CM |
694 | return entry; |
695 | } | |
696 | ||
4b46fce2 JB |
697 | /* Since the DIO code tries to lock a wide area we need to look for any ordered |
698 | * extents that exist in the range, rather than just the start of the range. | |
699 | */ | |
700 | struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode, | |
701 | u64 file_offset, | |
702 | u64 len) | |
703 | { | |
704 | struct btrfs_ordered_inode_tree *tree; | |
705 | struct rb_node *node; | |
706 | struct btrfs_ordered_extent *entry = NULL; | |
707 | ||
708 | tree = &BTRFS_I(inode)->ordered_tree; | |
709 | spin_lock(&tree->lock); | |
710 | node = tree_search(tree, file_offset); | |
711 | if (!node) { | |
712 | node = tree_search(tree, file_offset + len); | |
713 | if (!node) | |
714 | goto out; | |
715 | } | |
716 | ||
717 | while (1) { | |
718 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
719 | if (range_overlaps(entry, file_offset, len)) | |
720 | break; | |
721 | ||
722 | if (entry->file_offset >= file_offset + len) { | |
723 | entry = NULL; | |
724 | break; | |
725 | } | |
726 | entry = NULL; | |
727 | node = rb_next(node); | |
728 | if (!node) | |
729 | break; | |
730 | } | |
731 | out: | |
732 | if (entry) | |
733 | atomic_inc(&entry->refs); | |
734 | spin_unlock(&tree->lock); | |
735 | return entry; | |
736 | } | |
737 | ||
eb84ae03 CM |
738 | /* |
739 | * lookup and return any extent before 'file_offset'. NULL is returned | |
740 | * if none is found | |
741 | */ | |
e6dcd2dc | 742 | struct btrfs_ordered_extent * |
d397712b | 743 | btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset) |
e6dcd2dc CM |
744 | { |
745 | struct btrfs_ordered_inode_tree *tree; | |
746 | struct rb_node *node; | |
747 | struct btrfs_ordered_extent *entry = NULL; | |
748 | ||
749 | tree = &BTRFS_I(inode)->ordered_tree; | |
49958fd7 | 750 | spin_lock(&tree->lock); |
e6dcd2dc CM |
751 | node = tree_search(tree, file_offset); |
752 | if (!node) | |
753 | goto out; | |
754 | ||
755 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
756 | atomic_inc(&entry->refs); | |
757 | out: | |
49958fd7 | 758 | spin_unlock(&tree->lock); |
e6dcd2dc | 759 | return entry; |
81d7ed29 | 760 | } |
dbe674a9 | 761 | |
eb84ae03 CM |
762 | /* |
763 | * After an extent is done, call this to conditionally update the on disk | |
764 | * i_size. i_size is updated to cover any fully written part of the file. | |
765 | */ | |
c2167754 | 766 | int btrfs_ordered_update_i_size(struct inode *inode, u64 offset, |
dbe674a9 CM |
767 | struct btrfs_ordered_extent *ordered) |
768 | { | |
769 | struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree; | |
770 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
771 | u64 disk_i_size; | |
772 | u64 new_i_size; | |
773 | u64 i_size_test; | |
c2167754 | 774 | u64 i_size = i_size_read(inode); |
dbe674a9 | 775 | struct rb_node *node; |
c2167754 | 776 | struct rb_node *prev = NULL; |
dbe674a9 | 777 | struct btrfs_ordered_extent *test; |
c2167754 YZ |
778 | int ret = 1; |
779 | ||
780 | if (ordered) | |
781 | offset = entry_end(ordered); | |
a038fab0 YZ |
782 | else |
783 | offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize); | |
dbe674a9 | 784 | |
49958fd7 | 785 | spin_lock(&tree->lock); |
dbe674a9 CM |
786 | disk_i_size = BTRFS_I(inode)->disk_i_size; |
787 | ||
c2167754 YZ |
788 | /* truncate file */ |
789 | if (disk_i_size > i_size) { | |
790 | BTRFS_I(inode)->disk_i_size = i_size; | |
791 | ret = 0; | |
792 | goto out; | |
793 | } | |
794 | ||
dbe674a9 CM |
795 | /* |
796 | * if the disk i_size is already at the inode->i_size, or | |
797 | * this ordered extent is inside the disk i_size, we're done | |
798 | */ | |
c2167754 | 799 | if (disk_i_size == i_size || offset <= disk_i_size) { |
dbe674a9 CM |
800 | goto out; |
801 | } | |
802 | ||
803 | /* | |
804 | * we can't update the disk_isize if there are delalloc bytes | |
805 | * between disk_i_size and this ordered extent | |
806 | */ | |
c2167754 | 807 | if (test_range_bit(io_tree, disk_i_size, offset - 1, |
9655d298 | 808 | EXTENT_DELALLOC, 0, NULL)) { |
dbe674a9 CM |
809 | goto out; |
810 | } | |
811 | /* | |
812 | * walk backward from this ordered extent to disk_i_size. | |
813 | * if we find an ordered extent then we can't update disk i_size | |
814 | * yet | |
815 | */ | |
c2167754 YZ |
816 | if (ordered) { |
817 | node = rb_prev(&ordered->rb_node); | |
818 | } else { | |
819 | prev = tree_search(tree, offset); | |
820 | /* | |
821 | * we insert file extents without involving ordered struct, | |
822 | * so there should be no ordered struct cover this offset | |
823 | */ | |
824 | if (prev) { | |
825 | test = rb_entry(prev, struct btrfs_ordered_extent, | |
826 | rb_node); | |
827 | BUG_ON(offset_in_entry(test, offset)); | |
828 | } | |
829 | node = prev; | |
830 | } | |
831 | while (node) { | |
dbe674a9 CM |
832 | test = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
833 | if (test->file_offset + test->len <= disk_i_size) | |
834 | break; | |
c2167754 | 835 | if (test->file_offset >= i_size) |
dbe674a9 CM |
836 | break; |
837 | if (test->file_offset >= disk_i_size) | |
838 | goto out; | |
c2167754 | 839 | node = rb_prev(node); |
dbe674a9 | 840 | } |
c2167754 | 841 | new_i_size = min_t(u64, offset, i_size); |
dbe674a9 CM |
842 | |
843 | /* | |
844 | * at this point, we know we can safely update i_size to at least | |
845 | * the offset from this ordered extent. But, we need to | |
846 | * walk forward and see if ios from higher up in the file have | |
847 | * finished. | |
848 | */ | |
c2167754 YZ |
849 | if (ordered) { |
850 | node = rb_next(&ordered->rb_node); | |
851 | } else { | |
852 | if (prev) | |
853 | node = rb_next(prev); | |
854 | else | |
855 | node = rb_first(&tree->tree); | |
856 | } | |
dbe674a9 CM |
857 | i_size_test = 0; |
858 | if (node) { | |
859 | /* | |
860 | * do we have an area where IO might have finished | |
861 | * between our ordered extent and the next one. | |
862 | */ | |
863 | test = rb_entry(node, struct btrfs_ordered_extent, rb_node); | |
c2167754 | 864 | if (test->file_offset > offset) |
b48652c1 | 865 | i_size_test = test->file_offset; |
dbe674a9 | 866 | } else { |
c2167754 | 867 | i_size_test = i_size; |
dbe674a9 CM |
868 | } |
869 | ||
870 | /* | |
871 | * i_size_test is the end of a region after this ordered | |
872 | * extent where there are no ordered extents. As long as there | |
873 | * are no delalloc bytes in this area, it is safe to update | |
874 | * disk_i_size to the end of the region. | |
875 | */ | |
c2167754 YZ |
876 | if (i_size_test > offset && |
877 | !test_range_bit(io_tree, offset, i_size_test - 1, | |
878 | EXTENT_DELALLOC, 0, NULL)) { | |
879 | new_i_size = min_t(u64, i_size_test, i_size); | |
dbe674a9 CM |
880 | } |
881 | BTRFS_I(inode)->disk_i_size = new_i_size; | |
c2167754 | 882 | ret = 0; |
dbe674a9 | 883 | out: |
c2167754 YZ |
884 | /* |
885 | * we need to remove the ordered extent with the tree lock held | |
886 | * so that other people calling this function don't find our fully | |
887 | * processed ordered entry and skip updating the i_size | |
888 | */ | |
889 | if (ordered) | |
890 | __btrfs_remove_ordered_extent(inode, ordered); | |
49958fd7 | 891 | spin_unlock(&tree->lock); |
c2167754 YZ |
892 | if (ordered) |
893 | wake_up(&ordered->wait); | |
894 | return ret; | |
dbe674a9 | 895 | } |
ba1da2f4 | 896 | |
eb84ae03 CM |
897 | /* |
898 | * search the ordered extents for one corresponding to 'offset' and | |
899 | * try to find a checksum. This is used because we allow pages to | |
900 | * be reclaimed before their checksum is actually put into the btree | |
901 | */ | |
d20f7043 CM |
902 | int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr, |
903 | u32 *sum) | |
ba1da2f4 CM |
904 | { |
905 | struct btrfs_ordered_sum *ordered_sum; | |
906 | struct btrfs_sector_sum *sector_sums; | |
907 | struct btrfs_ordered_extent *ordered; | |
908 | struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree; | |
3edf7d33 CM |
909 | unsigned long num_sectors; |
910 | unsigned long i; | |
911 | u32 sectorsize = BTRFS_I(inode)->root->sectorsize; | |
ba1da2f4 | 912 | int ret = 1; |
ba1da2f4 CM |
913 | |
914 | ordered = btrfs_lookup_ordered_extent(inode, offset); | |
915 | if (!ordered) | |
916 | return 1; | |
917 | ||
49958fd7 | 918 | spin_lock(&tree->lock); |
c6e30871 | 919 | list_for_each_entry_reverse(ordered_sum, &ordered->list, list) { |
d20f7043 | 920 | if (disk_bytenr >= ordered_sum->bytenr) { |
3edf7d33 | 921 | num_sectors = ordered_sum->len / sectorsize; |
ed98b56a | 922 | sector_sums = ordered_sum->sums; |
3edf7d33 | 923 | for (i = 0; i < num_sectors; i++) { |
d20f7043 | 924 | if (sector_sums[i].bytenr == disk_bytenr) { |
3edf7d33 CM |
925 | *sum = sector_sums[i].sum; |
926 | ret = 0; | |
927 | goto out; | |
928 | } | |
929 | } | |
ba1da2f4 CM |
930 | } |
931 | } | |
932 | out: | |
49958fd7 | 933 | spin_unlock(&tree->lock); |
89642229 | 934 | btrfs_put_ordered_extent(ordered); |
ba1da2f4 CM |
935 | return ret; |
936 | } | |
937 | ||
f421950f | 938 | |
5a3f23d5 CM |
939 | /* |
940 | * add a given inode to the list of inodes that must be fully on | |
941 | * disk before a transaction commit finishes. | |
942 | * | |
943 | * This basically gives us the ext3 style data=ordered mode, and it is mostly | |
944 | * used to make sure renamed files are fully on disk. | |
945 | * | |
946 | * It is a noop if the inode is already fully on disk. | |
947 | * | |
948 | * If trans is not null, we'll do a friendly check for a transaction that | |
949 | * is already flushing things and force the IO down ourselves. | |
950 | */ | |
951 | int btrfs_add_ordered_operation(struct btrfs_trans_handle *trans, | |
952 | struct btrfs_root *root, | |
953 | struct inode *inode) | |
954 | { | |
955 | u64 last_mod; | |
956 | ||
957 | last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans); | |
958 | ||
959 | /* | |
960 | * if this file hasn't been changed since the last transaction | |
961 | * commit, we can safely return without doing anything | |
962 | */ | |
963 | if (last_mod < root->fs_info->last_trans_committed) | |
964 | return 0; | |
965 | ||
966 | /* | |
967 | * the transaction is already committing. Just start the IO and | |
968 | * don't bother with all of this list nonsense | |
969 | */ | |
970 | if (trans && root->fs_info->running_transaction->blocked) { | |
971 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
972 | return 0; | |
973 | } | |
974 | ||
975 | spin_lock(&root->fs_info->ordered_extent_lock); | |
976 | if (list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
977 | list_add_tail(&BTRFS_I(inode)->ordered_operations, | |
978 | &root->fs_info->ordered_operations); | |
979 | } | |
980 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
981 | ||
982 | return 0; | |
983 | } |