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